projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge branch 'upstream' of git://electric-eye.fr.zoreil.com/home/romieu/linux-2.6...
[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/sock.h>
107 #include <net/tcp_states.h>
108 #include <net/af_unix.h>
109 #include <linux/proc_fs.h>
110 #include <linux/seq_file.h>
111 #include <net/scm.h>
112 #include <linux/init.h>
113 #include <linux/poll.h>
114 #include <linux/smp_lock.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 = 10;
121
122 struct hlist_head unix_socket_table[UNIX_HASH_SIZE + 1];
123 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 #ifdef CONFIG_SECURITY_NETWORK
131 static void unix_get_peersec_dgram(struct sk_buff *skb)
132 {
133 int err;
134
135 err = security_socket_getpeersec_dgram(skb, UNIXSECDATA(skb),
136 UNIXSECLEN(skb));
137 if (err)
138 *(UNIXSECDATA(skb)) = NULL;
139 }
140
141 static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb)
142 {
143 scm->secdata = *UNIXSECDATA(skb);
144 scm->seclen = *UNIXSECLEN(skb);
145 }
146 #else
147 static inline void unix_get_peersec_dgram(struct sk_buff *skb)
148 { }
149
150 static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb)
151 { }
152 #endif /* CONFIG_SECURITY_NETWORK */
153
154 /*
155 * SMP locking strategy:
156 * hash table is protected with spinlock unix_table_lock
157 * each socket state is protected by separate rwlock.
158 */
159
160 static inline unsigned unix_hash_fold(unsigned hash)
161 {
162 hash ^= hash>>16;
163 hash ^= hash>>8;
164 return hash&(UNIX_HASH_SIZE-1);
165 }
166
167 #define unix_peer(sk) (unix_sk(sk)->peer)
168
169 static inline int unix_our_peer(struct sock *sk, struct sock *osk)
170 {
171 return unix_peer(osk) == sk;
172 }
173
174 static inline int unix_may_send(struct sock *sk, struct sock *osk)
175 {
176 return (unix_peer(osk) == NULL || unix_our_peer(sk, osk));
177 }
178
179 static struct sock *unix_peer_get(struct sock *s)
180 {
181 struct sock *peer;
182
183 unix_state_rlock(s);
184 peer = unix_peer(s);
185 if (peer)
186 sock_hold(peer);
187 unix_state_runlock(s);
188 return peer;
189 }
190
191 static inline void unix_release_addr(struct unix_address *addr)
192 {
193 if (atomic_dec_and_test(&addr->refcnt))
194 kfree(addr);
195 }
196
197 /*
198 * Check unix socket name:
199 * - should be not zero length.
200 * - if started by not zero, should be NULL terminated (FS object)
201 * - if started by zero, it is abstract name.
202 */
203
204 static int unix_mkname(struct sockaddr_un * sunaddr, int len, unsigned *hashp)
205 {
206 if (len <= sizeof(short) || len > sizeof(*sunaddr))
207 return -EINVAL;
208 if (!sunaddr || sunaddr->sun_family != AF_UNIX)
209 return -EINVAL;
210 if (sunaddr->sun_path[0]) {
211 /*
212 * This may look like an off by one error but it is a bit more
213 * subtle. 108 is the longest valid AF_UNIX path for a binding.
214 * sun_path[108] doesnt as such exist. However in kernel space
215 * we are guaranteed that it is a valid memory location in our
216 * kernel address buffer.
217 */
218 ((char *)sunaddr)[len]=0;
219 len = strlen(sunaddr->sun_path)+1+sizeof(short);
220 return len;
221 }
222
223 *hashp = unix_hash_fold(csum_partial((char*)sunaddr, len, 0));
224 return len;
225 }
226
227 static void __unix_remove_socket(struct sock *sk)
228 {
229 sk_del_node_init(sk);
230 }
231
232 static void __unix_insert_socket(struct hlist_head *list, struct sock *sk)
233 {
234 BUG_TRAP(sk_unhashed(sk));
235 sk_add_node(sk, list);
236 }
237
238 static inline void unix_remove_socket(struct sock *sk)
239 {
240 spin_lock(&unix_table_lock);
241 __unix_remove_socket(sk);
242 spin_unlock(&unix_table_lock);
243 }
244
245 static inline void unix_insert_socket(struct hlist_head *list, struct sock *sk)
246 {
247 spin_lock(&unix_table_lock);
248 __unix_insert_socket(list, sk);
249 spin_unlock(&unix_table_lock);
250 }
251
252 static struct sock *__unix_find_socket_byname(struct sockaddr_un *sunname,
253 int len, int type, unsigned hash)
254 {
255 struct sock *s;
256 struct hlist_node *node;
257
258 sk_for_each(s, node, &unix_socket_table[hash ^ type]) {
259 struct unix_sock *u = unix_sk(s);
260
261 if (u->addr->len == len &&
262 !memcmp(u->addr->name, sunname, len))
263 goto found;
264 }
265 s = NULL;
266 found:
267 return s;
268 }
269
270 static inline struct sock *unix_find_socket_byname(struct sockaddr_un *sunname,
271 int len, int type,
272 unsigned hash)
273 {
274 struct sock *s;
275
276 spin_lock(&unix_table_lock);
277 s = __unix_find_socket_byname(sunname, len, type, hash);
278 if (s)
279 sock_hold(s);
280 spin_unlock(&unix_table_lock);
281 return s;
282 }
283
284 static struct sock *unix_find_socket_byinode(struct inode *i)
285 {
286 struct sock *s;
287 struct hlist_node *node;
288
289 spin_lock(&unix_table_lock);
290 sk_for_each(s, node,
291 &unix_socket_table[i->i_ino & (UNIX_HASH_SIZE - 1)]) {
292 struct dentry *dentry = unix_sk(s)->dentry;
293
294 if(dentry && dentry->d_inode == i)
295 {
296 sock_hold(s);
297 goto found;
298 }
299 }
300 s = NULL;
301 found:
302 spin_unlock(&unix_table_lock);
303 return s;
304 }
305
306 static inline int unix_writable(struct sock *sk)
307 {
308 return (atomic_read(&sk->sk_wmem_alloc) << 2) <= sk->sk_sndbuf;
309 }
310
311 static void unix_write_space(struct sock *sk)
312 {
313 read_lock(&sk->sk_callback_lock);
314 if (unix_writable(sk)) {
315 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
316 wake_up_interruptible(sk->sk_sleep);
317 sk_wake_async(sk, 2, POLL_OUT);
318 }
319 read_unlock(&sk->sk_callback_lock);
320 }
321
322 /* When dgram socket disconnects (or changes its peer), we clear its receive
323 * queue of packets arrived from previous peer. First, it allows to do
324 * flow control based only on wmem_alloc; second, sk connected to peer
325 * may receive messages only from that peer. */
326 static void unix_dgram_disconnected(struct sock *sk, struct sock *other)
327 {
328 if (!skb_queue_empty(&sk->sk_receive_queue)) {
329 skb_queue_purge(&sk->sk_receive_queue);
330 wake_up_interruptible_all(&unix_sk(sk)->peer_wait);
331
332 /* If one link of bidirectional dgram pipe is disconnected,
333 * we signal error. Messages are lost. Do not make this,
334 * when peer was not connected to us.
335 */
336 if (!sock_flag(other, SOCK_DEAD) && unix_peer(other) == sk) {
337 other->sk_err = ECONNRESET;
338 other->sk_error_report(other);
339 }
340 }
341 }
342
343 static void unix_sock_destructor(struct sock *sk)
344 {
345 struct unix_sock *u = unix_sk(sk);
346
347 skb_queue_purge(&sk->sk_receive_queue);
348
349 BUG_TRAP(!atomic_read(&sk->sk_wmem_alloc));
350 BUG_TRAP(sk_unhashed(sk));
351 BUG_TRAP(!sk->sk_socket);
352 if (!sock_flag(sk, SOCK_DEAD)) {
353 printk("Attempt to release alive unix socket: %p\n", sk);
354 return;
355 }
356
357 if (u->addr)
358 unix_release_addr(u->addr);
359
360 atomic_dec(&unix_nr_socks);
361 #ifdef UNIX_REFCNT_DEBUG
362 printk(KERN_DEBUG "UNIX %p is destroyed, %d are still alive.\n", sk, atomic_read(&unix_nr_socks));
363 #endif
364 }
365
366 static int unix_release_sock (struct sock *sk, int embrion)
367 {
368 struct unix_sock *u = unix_sk(sk);
369 struct dentry *dentry;
370 struct vfsmount *mnt;
371 struct sock *skpair;
372 struct sk_buff *skb;
373 int state;
374
375 unix_remove_socket(sk);
376
377 /* Clear state */
378 unix_state_wlock(sk);
379 sock_orphan(sk);
380 sk->sk_shutdown = SHUTDOWN_MASK;
381 dentry = u->dentry;
382 u->dentry = NULL;
383 mnt = u->mnt;
384 u->mnt = NULL;
385 state = sk->sk_state;
386 sk->sk_state = TCP_CLOSE;
387 unix_state_wunlock(sk);
388
389 wake_up_interruptible_all(&u->peer_wait);
390
391 skpair=unix_peer(sk);
392
393 if (skpair!=NULL) {
394 if (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) {
395 unix_state_wlock(skpair);
396 /* No more writes */
397 skpair->sk_shutdown = SHUTDOWN_MASK;
398 if (!skb_queue_empty(&sk->sk_receive_queue) || embrion)
399 skpair->sk_err = ECONNRESET;
400 unix_state_wunlock(skpair);
401 skpair->sk_state_change(skpair);
402 read_lock(&skpair->sk_callback_lock);
403 sk_wake_async(skpair,1,POLL_HUP);
404 read_unlock(&skpair->sk_callback_lock);
405 }
406 sock_put(skpair); /* It may now die */
407 unix_peer(sk) = NULL;
408 }
409
410 /* Try to flush out this socket. Throw out buffers at least */
411
412 while ((skb = skb_dequeue(&sk->sk_receive_queue)) != NULL) {
413 if (state==TCP_LISTEN)
414 unix_release_sock(skb->sk, 1);
415 /* passed fds are erased in the kfree_skb hook */
416 kfree_skb(skb);
417 }
418
419 if (dentry) {
420 dput(dentry);
421 mntput(mnt);
422 }
423
424 sock_put(sk);
425
426 /* ---- Socket is dead now and most probably destroyed ---- */
427
428 /*
429 * Fixme: BSD difference: In BSD all sockets connected to use get
430 * ECONNRESET and we die on the spot. In Linux we behave
431 * like files and pipes do and wait for the last
432 * dereference.
433 *
434 * Can't we simply set sock->err?
435 *
436 * What the above comment does talk about? --ANK(980817)
437 */
438
439 if (atomic_read(&unix_tot_inflight))
440 unix_gc(); /* Garbage collect fds */
441
442 return 0;
443 }
444
445 static int unix_listen(struct socket *sock, int backlog)
446 {
447 int err;
448 struct sock *sk = sock->sk;
449 struct unix_sock *u = unix_sk(sk);
450
451 err = -EOPNOTSUPP;
452 if (sock->type!=SOCK_STREAM && sock->type!=SOCK_SEQPACKET)
453 goto out; /* Only stream/seqpacket sockets accept */
454 err = -EINVAL;
455 if (!u->addr)
456 goto out; /* No listens on an unbound socket */
457 unix_state_wlock(sk);
458 if (sk->sk_state != TCP_CLOSE && sk->sk_state != TCP_LISTEN)
459 goto out_unlock;
460 if (backlog > sk->sk_max_ack_backlog)
461 wake_up_interruptible_all(&u->peer_wait);
462 sk->sk_max_ack_backlog = backlog;
463 sk->sk_state = TCP_LISTEN;
464 /* set credentials so connect can copy them */
465 sk->sk_peercred.pid = current->tgid;
466 sk->sk_peercred.uid = current->euid;
467 sk->sk_peercred.gid = current->egid;
468 err = 0;
469
470 out_unlock:
471 unix_state_wunlock(sk);
472 out:
473 return err;
474 }
475
476 static int unix_release(struct socket *);
477 static int unix_bind(struct socket *, struct sockaddr *, int);
478 static int unix_stream_connect(struct socket *, struct sockaddr *,
479 int addr_len, int flags);
480 static int unix_socketpair(struct socket *, struct socket *);
481 static int unix_accept(struct socket *, struct socket *, int);
482 static int unix_getname(struct socket *, struct sockaddr *, int *, int);
483 static unsigned int unix_poll(struct file *, struct socket *, poll_table *);
484 static int unix_ioctl(struct socket *, unsigned int, unsigned long);
485 static int unix_shutdown(struct socket *, int);
486 static int unix_stream_sendmsg(struct kiocb *, struct socket *,
487 struct msghdr *, size_t);
488 static int unix_stream_recvmsg(struct kiocb *, struct socket *,
489 struct msghdr *, size_t, int);
490 static int unix_dgram_sendmsg(struct kiocb *, struct socket *,
491 struct msghdr *, size_t);
492 static int unix_dgram_recvmsg(struct kiocb *, struct socket *,
493 struct msghdr *, size_t, int);
494 static int unix_dgram_connect(struct socket *, struct sockaddr *,
495 int, int);
496 static int unix_seqpacket_sendmsg(struct kiocb *, struct socket *,
497 struct msghdr *, size_t);
498
499 static const struct proto_ops unix_stream_ops = {
500 .family = PF_UNIX,
501 .owner = THIS_MODULE,
502 .release = unix_release,
503 .bind = unix_bind,
504 .connect = unix_stream_connect,
505 .socketpair = unix_socketpair,
506 .accept = unix_accept,
507 .getname = unix_getname,
508 .poll = unix_poll,
509 .ioctl = unix_ioctl,
510 .listen = unix_listen,
511 .shutdown = unix_shutdown,
512 .setsockopt = sock_no_setsockopt,
513 .getsockopt = sock_no_getsockopt,
514 .sendmsg = unix_stream_sendmsg,
515 .recvmsg = unix_stream_recvmsg,
516 .mmap = sock_no_mmap,
517 .sendpage = sock_no_sendpage,
518 };
519
520 static const struct proto_ops unix_dgram_ops = {
521 .family = PF_UNIX,
522 .owner = THIS_MODULE,
523 .release = unix_release,
524 .bind = unix_bind,
525 .connect = unix_dgram_connect,
526 .socketpair = unix_socketpair,
527 .accept = sock_no_accept,
528 .getname = unix_getname,
529 .poll = datagram_poll,
530 .ioctl = unix_ioctl,
531 .listen = sock_no_listen,
532 .shutdown = unix_shutdown,
533 .setsockopt = sock_no_setsockopt,
534 .getsockopt = sock_no_getsockopt,
535 .sendmsg = unix_dgram_sendmsg,
536 .recvmsg = unix_dgram_recvmsg,
537 .mmap = sock_no_mmap,
538 .sendpage = sock_no_sendpage,
539 };
540
541 static const struct proto_ops unix_seqpacket_ops = {
542 .family = PF_UNIX,
543 .owner = THIS_MODULE,
544 .release = unix_release,
545 .bind = unix_bind,
546 .connect = unix_stream_connect,
547 .socketpair = unix_socketpair,
548 .accept = unix_accept,
549 .getname = unix_getname,
550 .poll = datagram_poll,
551 .ioctl = unix_ioctl,
552 .listen = unix_listen,
553 .shutdown = unix_shutdown,
554 .setsockopt = sock_no_setsockopt,
555 .getsockopt = sock_no_getsockopt,
556 .sendmsg = unix_seqpacket_sendmsg,
557 .recvmsg = unix_dgram_recvmsg,
558 .mmap = sock_no_mmap,
559 .sendpage = sock_no_sendpage,
560 };
561
562 static struct proto unix_proto = {
563 .name = "UNIX",
564 .owner = THIS_MODULE,
565 .obj_size = sizeof(struct unix_sock),
566 };
567
568 /*
569 * AF_UNIX sockets do not interact with hardware, hence they
570 * dont trigger interrupts - so it's safe for them to have
571 * bh-unsafe locking for their sk_receive_queue.lock. Split off
572 * this special lock-class by reinitializing the spinlock key:
573 */
574 static struct lock_class_key af_unix_sk_receive_queue_lock_key;
575
576 static struct sock * unix_create1(struct socket *sock)
577 {
578 struct sock *sk = NULL;
579 struct unix_sock *u;
580
581 if (atomic_read(&unix_nr_socks) >= 2*get_max_files())
582 goto out;
583
584 sk = sk_alloc(PF_UNIX, GFP_KERNEL, &unix_proto, 1);
585 if (!sk)
586 goto out;
587
588 atomic_inc(&unix_nr_socks);
589
590 sock_init_data(sock,sk);
591 lockdep_set_class(&sk->sk_receive_queue.lock,
592 &af_unix_sk_receive_queue_lock_key);
593
594 sk->sk_write_space = unix_write_space;
595 sk->sk_max_ack_backlog = sysctl_unix_max_dgram_qlen;
596 sk->sk_destruct = unix_sock_destructor;
597 u = unix_sk(sk);
598 u->dentry = NULL;
599 u->mnt = NULL;
600 spin_lock_init(&u->lock);
601 atomic_set(&u->inflight, sock ? 0 : -1);
602 mutex_init(&u->readlock); /* single task reading lock */
603 init_waitqueue_head(&u->peer_wait);
604 unix_insert_socket(unix_sockets_unbound, sk);
605 out:
606 return sk;
607 }
608
609 static int unix_create(struct socket *sock, int protocol)
610 {
611 if (protocol && protocol != PF_UNIX)
612 return -EPROTONOSUPPORT;
613
614 sock->state = SS_UNCONNECTED;
615
616 switch (sock->type) {
617 case SOCK_STREAM:
618 sock->ops = &unix_stream_ops;
619 break;
620 /*
621 * Believe it or not BSD has AF_UNIX, SOCK_RAW though
622 * nothing uses it.
623 */
624 case SOCK_RAW:
625 sock->type=SOCK_DGRAM;
626 case SOCK_DGRAM:
627 sock->ops = &unix_dgram_ops;
628 break;
629 case SOCK_SEQPACKET:
630 sock->ops = &unix_seqpacket_ops;
631 break;
632 default:
633 return -ESOCKTNOSUPPORT;
634 }
635
636 return unix_create1(sock) ? 0 : -ENOMEM;
637 }
638
639 static int unix_release(struct socket *sock)
640 {
641 struct sock *sk = sock->sk;
642
643 if (!sk)
644 return 0;
645
646 sock->sk = NULL;
647
648 return unix_release_sock (sk, 0);
649 }
650
651 static int unix_autobind(struct socket *sock)
652 {
653 struct sock *sk = sock->sk;
654 struct unix_sock *u = unix_sk(sk);
655 static u32 ordernum = 1;
656 struct unix_address * addr;
657 int err;
658
659 mutex_lock(&u->readlock);
660
661 err = 0;
662 if (u->addr)
663 goto out;
664
665 err = -ENOMEM;
666 addr = kzalloc(sizeof(*addr) + sizeof(short) + 16, GFP_KERNEL);
667 if (!addr)
668 goto out;
669
670 addr->name->sun_family = AF_UNIX;
671 atomic_set(&addr->refcnt, 1);
672
673 retry:
674 addr->len = sprintf(addr->name->sun_path+1, "%05x", ordernum) + 1 + sizeof(short);
675 addr->hash = unix_hash_fold(csum_partial((void*)addr->name, addr->len, 0));
676
677 spin_lock(&unix_table_lock);
678 ordernum = (ordernum+1)&0xFFFFF;
679
680 if (__unix_find_socket_byname(addr->name, addr->len, sock->type,
681 addr->hash)) {
682 spin_unlock(&unix_table_lock);
683 /* Sanity yield. It is unusual case, but yet... */
684 if (!(ordernum&0xFF))
685 yield();
686 goto retry;
687 }
688 addr->hash ^= sk->sk_type;
689
690 __unix_remove_socket(sk);
691 u->addr = addr;
692 __unix_insert_socket(&unix_socket_table[addr->hash], sk);
693 spin_unlock(&unix_table_lock);
694 err = 0;
695
696 out: mutex_unlock(&u->readlock);
697 return err;
698 }
699
700 static struct sock *unix_find_other(struct sockaddr_un *sunname, int len,
701 int type, unsigned hash, int *error)
702 {
703 struct sock *u;
704 struct nameidata nd;
705 int err = 0;
706
707 if (sunname->sun_path[0]) {
708 err = path_lookup(sunname->sun_path, LOOKUP_FOLLOW, &nd);
709 if (err)
710 goto fail;
711 err = vfs_permission(&nd, MAY_WRITE);
712 if (err)
713 goto put_fail;
714
715 err = -ECONNREFUSED;
716 if (!S_ISSOCK(nd.dentry->d_inode->i_mode))
717 goto put_fail;
718 u=unix_find_socket_byinode(nd.dentry->d_inode);
719 if (!u)
720 goto put_fail;
721
722 if (u->sk_type == type)
723 touch_atime(nd.mnt, nd.dentry);
724
725 path_release(&nd);
726
727 err=-EPROTOTYPE;
728 if (u->sk_type != type) {
729 sock_put(u);
730 goto fail;
731 }
732 } else {
733 err = -ECONNREFUSED;
734 u=unix_find_socket_byname(sunname, len, type, hash);
735 if (u) {
736 struct dentry *dentry;
737 dentry = unix_sk(u)->dentry;
738 if (dentry)
739 touch_atime(unix_sk(u)->mnt, dentry);
740 } else
741 goto fail;
742 }
743 return u;
744
745 put_fail:
746 path_release(&nd);
747 fail:
748 *error=err;
749 return NULL;
750 }
751
752
753 static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
754 {
755 struct sock *sk = sock->sk;
756 struct unix_sock *u = unix_sk(sk);
757 struct sockaddr_un *sunaddr=(struct sockaddr_un *)uaddr;
758 struct dentry * dentry = NULL;
759 struct nameidata nd;
760 int err;
761 unsigned hash;
762 struct unix_address *addr;
763 struct hlist_head *list;
764
765 err = -EINVAL;
766 if (sunaddr->sun_family != AF_UNIX)
767 goto out;
768
769 if (addr_len==sizeof(short)) {
770 err = unix_autobind(sock);
771 goto out;
772 }
773
774 err = unix_mkname(sunaddr, addr_len, &hash);
775 if (err < 0)
776 goto out;
777 addr_len = err;
778
779 mutex_lock(&u->readlock);
780
781 err = -EINVAL;
782 if (u->addr)
783 goto out_up;
784
785 err = -ENOMEM;
786 addr = kmalloc(sizeof(*addr)+addr_len, GFP_KERNEL);
787 if (!addr)
788 goto out_up;
789
790 memcpy(addr->name, sunaddr, addr_len);
791 addr->len = addr_len;
792 addr->hash = hash ^ sk->sk_type;
793 atomic_set(&addr->refcnt, 1);
794
795 if (sunaddr->sun_path[0]) {
796 unsigned int mode;
797 err = 0;
798 /*
799 * Get the parent directory, calculate the hash for last
800 * component.
801 */
802 err = path_lookup(sunaddr->sun_path, LOOKUP_PARENT, &nd);
803 if (err)
804 goto out_mknod_parent;
805
806 dentry = lookup_create(&nd, 0);
807 err = PTR_ERR(dentry);
808 if (IS_ERR(dentry))
809 goto out_mknod_unlock;
810
811 /*
812 * All right, let's create it.
813 */
814 mode = S_IFSOCK |
815 (SOCK_INODE(sock)->i_mode & ~current->fs->umask);
816 err = vfs_mknod(nd.dentry->d_inode, dentry, mode, 0);
817 if (err)
818 goto out_mknod_dput;
819 mutex_unlock(&nd.dentry->d_inode->i_mutex);
820 dput(nd.dentry);
821 nd.dentry = dentry;
822
823 addr->hash = UNIX_HASH_SIZE;
824 }
825
826 spin_lock(&unix_table_lock);
827
828 if (!sunaddr->sun_path[0]) {
829 err = -EADDRINUSE;
830 if (__unix_find_socket_byname(sunaddr, addr_len,
831 sk->sk_type, hash)) {
832 unix_release_addr(addr);
833 goto out_unlock;
834 }
835
836 list = &unix_socket_table[addr->hash];
837 } else {
838 list = &unix_socket_table[dentry->d_inode->i_ino & (UNIX_HASH_SIZE-1)];
839 u->dentry = nd.dentry;
840 u->mnt = nd.mnt;
841 }
842
843 err = 0;
844 __unix_remove_socket(sk);
845 u->addr = addr;
846 __unix_insert_socket(list, sk);
847
848 out_unlock:
849 spin_unlock(&unix_table_lock);
850 out_up:
851 mutex_unlock(&u->readlock);
852 out:
853 return err;
854
855 out_mknod_dput:
856 dput(dentry);
857 out_mknod_unlock:
858 mutex_unlock(&nd.dentry->d_inode->i_mutex);
859 path_release(&nd);
860 out_mknod_parent:
861 if (err==-EEXIST)
862 err=-EADDRINUSE;
863 unix_release_addr(addr);
864 goto out_up;
865 }
866
867 static int unix_dgram_connect(struct socket *sock, struct sockaddr *addr,
868 int alen, int flags)
869 {
870 struct sock *sk = sock->sk;
871 struct sockaddr_un *sunaddr=(struct sockaddr_un*)addr;
872 struct sock *other;
873 unsigned hash;
874 int err;
875
876 if (addr->sa_family != AF_UNSPEC) {
877 err = unix_mkname(sunaddr, alen, &hash);
878 if (err < 0)
879 goto out;
880 alen = err;
881
882 if (test_bit(SOCK_PASSCRED, &sock->flags) &&
883 !unix_sk(sk)->addr && (err = unix_autobind(sock)) != 0)
884 goto out;
885
886 other=unix_find_other(sunaddr, alen, sock->type, hash, &err);
887 if (!other)
888 goto out;
889
890 unix_state_wlock(sk);
891
892 err = -EPERM;
893 if (!unix_may_send(sk, other))
894 goto out_unlock;
895
896 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
897 if (err)
898 goto out_unlock;
899
900 } else {
901 /*
902 * 1003.1g breaking connected state with AF_UNSPEC
903 */
904 other = NULL;
905 unix_state_wlock(sk);
906 }
907
908 /*
909 * If it was connected, reconnect.
910 */
911 if (unix_peer(sk)) {
912 struct sock *old_peer = unix_peer(sk);
913 unix_peer(sk)=other;
914 unix_state_wunlock(sk);
915
916 if (other != old_peer)
917 unix_dgram_disconnected(sk, old_peer);
918 sock_put(old_peer);
919 } else {
920 unix_peer(sk)=other;
921 unix_state_wunlock(sk);
922 }
923 return 0;
924
925 out_unlock:
926 unix_state_wunlock(sk);
927 sock_put(other);
928 out:
929 return err;
930 }
931
932 static long unix_wait_for_peer(struct sock *other, long timeo)
933 {
934 struct unix_sock *u = unix_sk(other);
935 int sched;
936 DEFINE_WAIT(wait);
937
938 prepare_to_wait_exclusive(&u->peer_wait, &wait, TASK_INTERRUPTIBLE);
939
940 sched = !sock_flag(other, SOCK_DEAD) &&
941 !(other->sk_shutdown & RCV_SHUTDOWN) &&
942 (skb_queue_len(&other->sk_receive_queue) >
943 other->sk_max_ack_backlog);
944
945 unix_state_runlock(other);
946
947 if (sched)
948 timeo = schedule_timeout(timeo);
949
950 finish_wait(&u->peer_wait, &wait);
951 return timeo;
952 }
953
954 static int unix_stream_connect(struct socket *sock, struct sockaddr *uaddr,
955 int addr_len, int flags)
956 {
957 struct sockaddr_un *sunaddr=(struct sockaddr_un *)uaddr;
958 struct sock *sk = sock->sk;
959 struct unix_sock *u = unix_sk(sk), *newu, *otheru;
960 struct sock *newsk = NULL;
961 struct sock *other = NULL;
962 struct sk_buff *skb = NULL;
963 unsigned hash;
964 int st;
965 int err;
966 long timeo;
967
968 err = unix_mkname(sunaddr, addr_len, &hash);
969 if (err < 0)
970 goto out;
971 addr_len = err;
972
973 if (test_bit(SOCK_PASSCRED, &sock->flags)
974 && !u->addr && (err = unix_autobind(sock)) != 0)
975 goto out;
976
977 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
978
979 /* First of all allocate resources.
980 If we will make it after state is locked,
981 we will have to recheck all again in any case.
982 */
983
984 err = -ENOMEM;
985
986 /* create new sock for complete connection */
987 newsk = unix_create1(NULL);
988 if (newsk == NULL)
989 goto out;
990
991 /* Allocate skb for sending to listening sock */
992 skb = sock_wmalloc(newsk, 1, 0, GFP_KERNEL);
993 if (skb == NULL)
994 goto out;
995
996 restart:
997 /* Find listening sock. */
998 other = unix_find_other(sunaddr, addr_len, sk->sk_type, hash, &err);
999 if (!other)
1000 goto out;
1001
1002 /* Latch state of peer */
1003 unix_state_rlock(other);
1004
1005 /* Apparently VFS overslept socket death. Retry. */
1006 if (sock_flag(other, SOCK_DEAD)) {
1007 unix_state_runlock(other);
1008 sock_put(other);
1009 goto restart;
1010 }
1011
1012 err = -ECONNREFUSED;
1013 if (other->sk_state != TCP_LISTEN)
1014 goto out_unlock;
1015
1016 if (skb_queue_len(&other->sk_receive_queue) >
1017 other->sk_max_ack_backlog) {
1018 err = -EAGAIN;
1019 if (!timeo)
1020 goto out_unlock;
1021
1022 timeo = unix_wait_for_peer(other, timeo);
1023
1024 err = sock_intr_errno(timeo);
1025 if (signal_pending(current))
1026 goto out;
1027 sock_put(other);
1028 goto restart;
1029 }
1030
1031 /* Latch our state.
1032
1033 It is tricky place. We need to grab write lock and cannot
1034 drop lock on peer. It is dangerous because deadlock is
1035 possible. Connect to self case and simultaneous
1036 attempt to connect are eliminated by checking socket
1037 state. other is TCP_LISTEN, if sk is TCP_LISTEN we
1038 check this before attempt to grab lock.
1039
1040 Well, and we have to recheck the state after socket locked.
1041 */
1042 st = sk->sk_state;
1043
1044 switch (st) {
1045 case TCP_CLOSE:
1046 /* This is ok... continue with connect */
1047 break;
1048 case TCP_ESTABLISHED:
1049 /* Socket is already connected */
1050 err = -EISCONN;
1051 goto out_unlock;
1052 default:
1053 err = -EINVAL;
1054 goto out_unlock;
1055 }
1056
1057 unix_state_wlock_nested(sk);
1058
1059 if (sk->sk_state != st) {
1060 unix_state_wunlock(sk);
1061 unix_state_runlock(other);
1062 sock_put(other);
1063 goto restart;
1064 }
1065
1066 err = security_unix_stream_connect(sock, other->sk_socket, newsk);
1067 if (err) {
1068 unix_state_wunlock(sk);
1069 goto out_unlock;
1070 }
1071
1072 /* The way is open! Fastly set all the necessary fields... */
1073
1074 sock_hold(sk);
1075 unix_peer(newsk) = sk;
1076 newsk->sk_state = TCP_ESTABLISHED;
1077 newsk->sk_type = sk->sk_type;
1078 newsk->sk_peercred.pid = current->tgid;
1079 newsk->sk_peercred.uid = current->euid;
1080 newsk->sk_peercred.gid = current->egid;
1081 newu = unix_sk(newsk);
1082 newsk->sk_sleep = &newu->peer_wait;
1083 otheru = unix_sk(other);
1084
1085 /* copy address information from listening to new sock*/
1086 if (otheru->addr) {
1087 atomic_inc(&otheru->addr->refcnt);
1088 newu->addr = otheru->addr;
1089 }
1090 if (otheru->dentry) {
1091 newu->dentry = dget(otheru->dentry);
1092 newu->mnt = mntget(otheru->mnt);
1093 }
1094
1095 /* Set credentials */
1096 sk->sk_peercred = other->sk_peercred;
1097
1098 sock->state = SS_CONNECTED;
1099 sk->sk_state = TCP_ESTABLISHED;
1100 sock_hold(newsk);
1101
1102 smp_mb__after_atomic_inc(); /* sock_hold() does an atomic_inc() */
1103 unix_peer(sk) = newsk;
1104
1105 unix_state_wunlock(sk);
1106
1107 /* take ten and and send info to listening sock */
1108 spin_lock(&other->sk_receive_queue.lock);
1109 __skb_queue_tail(&other->sk_receive_queue, skb);
1110 /* Undo artificially decreased inflight after embrion
1111 * is installed to listening socket. */
1112 atomic_inc(&newu->inflight);
1113 spin_unlock(&other->sk_receive_queue.lock);
1114 unix_state_runlock(other);
1115 other->sk_data_ready(other, 0);
1116 sock_put(other);
1117 return 0;
1118
1119 out_unlock:
1120 if (other)
1121 unix_state_runlock(other);
1122
1123 out:
1124 if (skb)
1125 kfree_skb(skb);
1126 if (newsk)
1127 unix_release_sock(newsk, 0);
1128 if (other)
1129 sock_put(other);
1130 return err;
1131 }
1132
1133 static int unix_socketpair(struct socket *socka, struct socket *sockb)
1134 {
1135 struct sock *ska=socka->sk, *skb = sockb->sk;
1136
1137 /* Join our sockets back to back */
1138 sock_hold(ska);
1139 sock_hold(skb);
1140 unix_peer(ska)=skb;
1141 unix_peer(skb)=ska;
1142 ska->sk_peercred.pid = skb->sk_peercred.pid = current->tgid;
1143 ska->sk_peercred.uid = skb->sk_peercred.uid = current->euid;
1144 ska->sk_peercred.gid = skb->sk_peercred.gid = current->egid;
1145
1146 if (ska->sk_type != SOCK_DGRAM) {
1147 ska->sk_state = TCP_ESTABLISHED;
1148 skb->sk_state = TCP_ESTABLISHED;
1149 socka->state = SS_CONNECTED;
1150 sockb->state = SS_CONNECTED;
1151 }
1152 return 0;
1153 }
1154
1155 static int unix_accept(struct socket *sock, struct socket *newsock, int flags)
1156 {
1157 struct sock *sk = sock->sk;
1158 struct sock *tsk;
1159 struct sk_buff *skb;
1160 int err;
1161
1162 err = -EOPNOTSUPP;
1163 if (sock->type!=SOCK_STREAM && sock->type!=SOCK_SEQPACKET)
1164 goto out;
1165
1166 err = -EINVAL;
1167 if (sk->sk_state != TCP_LISTEN)
1168 goto out;
1169
1170 /* If socket state is TCP_LISTEN it cannot change (for now...),
1171 * so that no locks are necessary.
1172 */
1173
1174 skb = skb_recv_datagram(sk, 0, flags&O_NONBLOCK, &err);
1175 if (!skb) {
1176 /* This means receive shutdown. */
1177 if (err == 0)
1178 err = -EINVAL;
1179 goto out;
1180 }
1181
1182 tsk = skb->sk;
1183 skb_free_datagram(sk, skb);
1184 wake_up_interruptible(&unix_sk(sk)->peer_wait);
1185
1186 /* attach accepted sock to socket */
1187 unix_state_wlock(tsk);
1188 newsock->state = SS_CONNECTED;
1189 sock_graft(tsk, newsock);
1190 unix_state_wunlock(tsk);
1191 return 0;
1192
1193 out:
1194 return err;
1195 }
1196
1197
1198 static int unix_getname(struct socket *sock, struct sockaddr *uaddr, int *uaddr_len, int peer)
1199 {
1200 struct sock *sk = sock->sk;
1201 struct unix_sock *u;
1202 struct sockaddr_un *sunaddr=(struct sockaddr_un *)uaddr;
1203 int err = 0;
1204
1205 if (peer) {
1206 sk = unix_peer_get(sk);
1207
1208 err = -ENOTCONN;
1209 if (!sk)
1210 goto out;
1211 err = 0;
1212 } else {
1213 sock_hold(sk);
1214 }
1215
1216 u = unix_sk(sk);
1217 unix_state_rlock(sk);
1218 if (!u->addr) {
1219 sunaddr->sun_family = AF_UNIX;
1220 sunaddr->sun_path[0] = 0;
1221 *uaddr_len = sizeof(short);
1222 } else {
1223 struct unix_address *addr = u->addr;
1224
1225 *uaddr_len = addr->len;
1226 memcpy(sunaddr, addr->name, *uaddr_len);
1227 }
1228 unix_state_runlock(sk);
1229 sock_put(sk);
1230 out:
1231 return err;
1232 }
1233
1234 static void unix_detach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1235 {
1236 int i;
1237
1238 scm->fp = UNIXCB(skb).fp;
1239 skb->destructor = sock_wfree;
1240 UNIXCB(skb).fp = NULL;
1241
1242 for (i=scm->fp->count-1; i>=0; i--)
1243 unix_notinflight(scm->fp->fp[i]);
1244 }
1245
1246 static void unix_destruct_fds(struct sk_buff *skb)
1247 {
1248 struct scm_cookie scm;
1249 memset(&scm, 0, sizeof(scm));
1250 unix_detach_fds(&scm, skb);
1251
1252 /* Alas, it calls VFS */
1253 /* So fscking what? fput() had been SMP-safe since the last Summer */
1254 scm_destroy(&scm);
1255 sock_wfree(skb);
1256 }
1257
1258 static void unix_attach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1259 {
1260 int i;
1261 for (i=scm->fp->count-1; i>=0; i--)
1262 unix_inflight(scm->fp->fp[i]);
1263 UNIXCB(skb).fp = scm->fp;
1264 skb->destructor = unix_destruct_fds;
1265 scm->fp = NULL;
1266 }
1267
1268 /*
1269 * Send AF_UNIX data.
1270 */
1271
1272 static int unix_dgram_sendmsg(struct kiocb *kiocb, struct socket *sock,
1273 struct msghdr *msg, size_t len)
1274 {
1275 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1276 struct sock *sk = sock->sk;
1277 struct unix_sock *u = unix_sk(sk);
1278 struct sockaddr_un *sunaddr=msg->msg_name;
1279 struct sock *other = NULL;
1280 int namelen = 0; /* fake GCC */
1281 int err;
1282 unsigned hash;
1283 struct sk_buff *skb;
1284 long timeo;
1285 struct scm_cookie tmp_scm;
1286
1287 if (NULL == siocb->scm)
1288 siocb->scm = &tmp_scm;
1289 err = scm_send(sock, msg, siocb->scm);
1290 if (err < 0)
1291 return err;
1292
1293 err = -EOPNOTSUPP;
1294 if (msg->msg_flags&MSG_OOB)
1295 goto out;
1296
1297 if (msg->msg_namelen) {
1298 err = unix_mkname(sunaddr, msg->msg_namelen, &hash);
1299 if (err < 0)
1300 goto out;
1301 namelen = err;
1302 } else {
1303 sunaddr = NULL;
1304 err = -ENOTCONN;
1305 other = unix_peer_get(sk);
1306 if (!other)
1307 goto out;
1308 }
1309
1310 if (test_bit(SOCK_PASSCRED, &sock->flags)
1311 && !u->addr && (err = unix_autobind(sock)) != 0)
1312 goto out;
1313
1314 err = -EMSGSIZE;
1315 if (len > sk->sk_sndbuf - 32)
1316 goto out;
1317
1318 skb = sock_alloc_send_skb(sk, len, msg->msg_flags&MSG_DONTWAIT, &err);
1319 if (skb==NULL)
1320 goto out;
1321
1322 memcpy(UNIXCREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1323 if (siocb->scm->fp)
1324 unix_attach_fds(siocb->scm, skb);
1325
1326 unix_get_peersec_dgram(skb);
1327
1328 skb->h.raw = skb->data;
1329 err = memcpy_fromiovec(skb_put(skb,len), msg->msg_iov, len);
1330 if (err)
1331 goto out_free;
1332
1333 timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1334
1335 restart:
1336 if (!other) {
1337 err = -ECONNRESET;
1338 if (sunaddr == NULL)
1339 goto out_free;
1340
1341 other = unix_find_other(sunaddr, namelen, sk->sk_type,
1342 hash, &err);
1343 if (other==NULL)
1344 goto out_free;
1345 }
1346
1347 unix_state_rlock(other);
1348 err = -EPERM;
1349 if (!unix_may_send(sk, other))
1350 goto out_unlock;
1351
1352 if (sock_flag(other, SOCK_DEAD)) {
1353 /*
1354 * Check with 1003.1g - what should
1355 * datagram error
1356 */
1357 unix_state_runlock(other);
1358 sock_put(other);
1359
1360 err = 0;
1361 unix_state_wlock(sk);
1362 if (unix_peer(sk) == other) {
1363 unix_peer(sk)=NULL;
1364 unix_state_wunlock(sk);
1365
1366 unix_dgram_disconnected(sk, other);
1367 sock_put(other);
1368 err = -ECONNREFUSED;
1369 } else {
1370 unix_state_wunlock(sk);
1371 }
1372
1373 other = NULL;
1374 if (err)
1375 goto out_free;
1376 goto restart;
1377 }
1378
1379 err = -EPIPE;
1380 if (other->sk_shutdown & RCV_SHUTDOWN)
1381 goto out_unlock;
1382
1383 if (sk->sk_type != SOCK_SEQPACKET) {
1384 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1385 if (err)
1386 goto out_unlock;
1387 }
1388
1389 if (unix_peer(other) != sk &&
1390 (skb_queue_len(&other->sk_receive_queue) >
1391 other->sk_max_ack_backlog)) {
1392 if (!timeo) {
1393 err = -EAGAIN;
1394 goto out_unlock;
1395 }
1396
1397 timeo = unix_wait_for_peer(other, timeo);
1398
1399 err = sock_intr_errno(timeo);
1400 if (signal_pending(current))
1401 goto out_free;
1402
1403 goto restart;
1404 }
1405
1406 skb_queue_tail(&other->sk_receive_queue, skb);
1407 unix_state_runlock(other);
1408 other->sk_data_ready(other, len);
1409 sock_put(other);
1410 scm_destroy(siocb->scm);
1411 return len;
1412
1413 out_unlock:
1414 unix_state_runlock(other);
1415 out_free:
1416 kfree_skb(skb);
1417 out:
1418 if (other)
1419 sock_put(other);
1420 scm_destroy(siocb->scm);
1421 return err;
1422 }
1423
1424
1425 static int unix_stream_sendmsg(struct kiocb *kiocb, struct socket *sock,
1426 struct msghdr *msg, size_t len)
1427 {
1428 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1429 struct sock *sk = sock->sk;
1430 struct sock *other = NULL;
1431 struct sockaddr_un *sunaddr=msg->msg_name;
1432 int err,size;
1433 struct sk_buff *skb;
1434 int sent=0;
1435 struct scm_cookie tmp_scm;
1436
1437 if (NULL == siocb->scm)
1438 siocb->scm = &tmp_scm;
1439 err = scm_send(sock, msg, siocb->scm);
1440 if (err < 0)
1441 return err;
1442
1443 err = -EOPNOTSUPP;
1444 if (msg->msg_flags&MSG_OOB)
1445 goto out_err;
1446
1447 if (msg->msg_namelen) {
1448 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
1449 goto out_err;
1450 } else {
1451 sunaddr = NULL;
1452 err = -ENOTCONN;
1453 other = unix_peer(sk);
1454 if (!other)
1455 goto out_err;
1456 }
1457
1458 if (sk->sk_shutdown & SEND_SHUTDOWN)
1459 goto pipe_err;
1460
1461 while(sent < len)
1462 {
1463 /*
1464 * Optimisation for the fact that under 0.01% of X
1465 * messages typically need breaking up.
1466 */
1467
1468 size = len-sent;
1469
1470 /* Keep two messages in the pipe so it schedules better */
1471 if (size > ((sk->sk_sndbuf >> 1) - 64))
1472 size = (sk->sk_sndbuf >> 1) - 64;
1473
1474 if (size > SKB_MAX_ALLOC)
1475 size = SKB_MAX_ALLOC;
1476
1477 /*
1478 * Grab a buffer
1479 */
1480
1481 skb=sock_alloc_send_skb(sk,size,msg->msg_flags&MSG_DONTWAIT, &err);
1482
1483 if (skb==NULL)
1484 goto out_err;
1485
1486 /*
1487 * If you pass two values to the sock_alloc_send_skb
1488 * it tries to grab the large buffer with GFP_NOFS
1489 * (which can fail easily), and if it fails grab the
1490 * fallback size buffer which is under a page and will
1491 * succeed. [Alan]
1492 */
1493 size = min_t(int, size, skb_tailroom(skb));
1494
1495 memcpy(UNIXCREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1496 if (siocb->scm->fp)
1497 unix_attach_fds(siocb->scm, skb);
1498
1499 if ((err = memcpy_fromiovec(skb_put(skb,size), msg->msg_iov, size)) != 0) {
1500 kfree_skb(skb);
1501 goto out_err;
1502 }
1503
1504 unix_state_rlock(other);
1505
1506 if (sock_flag(other, SOCK_DEAD) ||
1507 (other->sk_shutdown & RCV_SHUTDOWN))
1508 goto pipe_err_free;
1509
1510 skb_queue_tail(&other->sk_receive_queue, skb);
1511 unix_state_runlock(other);
1512 other->sk_data_ready(other, size);
1513 sent+=size;
1514 }
1515
1516 scm_destroy(siocb->scm);
1517 siocb->scm = NULL;
1518
1519 return sent;
1520
1521 pipe_err_free:
1522 unix_state_runlock(other);
1523 kfree_skb(skb);
1524 pipe_err:
1525 if (sent==0 && !(msg->msg_flags&MSG_NOSIGNAL))
1526 send_sig(SIGPIPE,current,0);
1527 err = -EPIPE;
1528 out_err:
1529 scm_destroy(siocb->scm);
1530 siocb->scm = NULL;
1531 return sent ? : err;
1532 }
1533
1534 static int unix_seqpacket_sendmsg(struct kiocb *kiocb, struct socket *sock,
1535 struct msghdr *msg, size_t len)
1536 {
1537 int err;
1538 struct sock *sk = sock->sk;
1539
1540 err = sock_error(sk);
1541 if (err)
1542 return err;
1543
1544 if (sk->sk_state != TCP_ESTABLISHED)
1545 return -ENOTCONN;
1546
1547 if (msg->msg_namelen)
1548 msg->msg_namelen = 0;
1549
1550 return unix_dgram_sendmsg(kiocb, sock, msg, len);
1551 }
1552
1553 static void unix_copy_addr(struct msghdr *msg, struct sock *sk)
1554 {
1555 struct unix_sock *u = unix_sk(sk);
1556
1557 msg->msg_namelen = 0;
1558 if (u->addr) {
1559 msg->msg_namelen = u->addr->len;
1560 memcpy(msg->msg_name, u->addr->name, u->addr->len);
1561 }
1562 }
1563
1564 static int unix_dgram_recvmsg(struct kiocb *iocb, struct socket *sock,
1565 struct msghdr *msg, size_t size,
1566 int flags)
1567 {
1568 struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1569 struct scm_cookie tmp_scm;
1570 struct sock *sk = sock->sk;
1571 struct unix_sock *u = unix_sk(sk);
1572 int noblock = flags & MSG_DONTWAIT;
1573 struct sk_buff *skb;
1574 int err;
1575
1576 err = -EOPNOTSUPP;
1577 if (flags&MSG_OOB)
1578 goto out;
1579
1580 msg->msg_namelen = 0;
1581
1582 mutex_lock(&u->readlock);
1583
1584 skb = skb_recv_datagram(sk, flags, noblock, &err);
1585 if (!skb)
1586 goto out_unlock;
1587
1588 wake_up_interruptible(&u->peer_wait);
1589
1590 if (msg->msg_name)
1591 unix_copy_addr(msg, skb->sk);
1592
1593 if (size > skb->len)
1594 size = skb->len;
1595 else if (size < skb->len)
1596 msg->msg_flags |= MSG_TRUNC;
1597
1598 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, size);
1599 if (err)
1600 goto out_free;
1601
1602 if (!siocb->scm) {
1603 siocb->scm = &tmp_scm;
1604 memset(&tmp_scm, 0, sizeof(tmp_scm));
1605 }
1606 siocb->scm->creds = *UNIXCREDS(skb);
1607 unix_set_secdata(siocb->scm, skb);
1608
1609 if (!(flags & MSG_PEEK))
1610 {
1611 if (UNIXCB(skb).fp)
1612 unix_detach_fds(siocb->scm, skb);
1613 }
1614 else
1615 {
1616 /* It is questionable: on PEEK we could:
1617 - do not return fds - good, but too simple 8)
1618 - return fds, and do not return them on read (old strategy,
1619 apparently wrong)
1620 - clone fds (I chose it for now, it is the most universal
1621 solution)
1622
1623 POSIX 1003.1g does not actually define this clearly
1624 at all. POSIX 1003.1g doesn't define a lot of things
1625 clearly however!
1626
1627 */
1628 if (UNIXCB(skb).fp)
1629 siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1630 }
1631 err = size;
1632
1633 scm_recv(sock, msg, siocb->scm, flags);
1634
1635 out_free:
1636 skb_free_datagram(sk,skb);
1637 out_unlock:
1638 mutex_unlock(&u->readlock);
1639 out:
1640 return err;
1641 }
1642
1643 /*
1644 * Sleep until data has arrive. But check for races..
1645 */
1646
1647 static long unix_stream_data_wait(struct sock * sk, long timeo)
1648 {
1649 DEFINE_WAIT(wait);
1650
1651 unix_state_rlock(sk);
1652
1653 for (;;) {
1654 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1655
1656 if (!skb_queue_empty(&sk->sk_receive_queue) ||
1657 sk->sk_err ||
1658 (sk->sk_shutdown & RCV_SHUTDOWN) ||
1659 signal_pending(current) ||
1660 !timeo)
1661 break;
1662
1663 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1664 unix_state_runlock(sk);
1665 timeo = schedule_timeout(timeo);
1666 unix_state_rlock(sk);
1667 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1668 }
1669
1670 finish_wait(sk->sk_sleep, &wait);
1671 unix_state_runlock(sk);
1672 return timeo;
1673 }
1674
1675
1676
1677 static int unix_stream_recvmsg(struct kiocb *iocb, struct socket *sock,
1678 struct msghdr *msg, size_t size,
1679 int flags)
1680 {
1681 struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1682 struct scm_cookie tmp_scm;
1683 struct sock *sk = sock->sk;
1684 struct unix_sock *u = unix_sk(sk);
1685 struct sockaddr_un *sunaddr=msg->msg_name;
1686 int copied = 0;
1687 int check_creds = 0;
1688 int target;
1689 int err = 0;
1690 long timeo;
1691
1692 err = -EINVAL;
1693 if (sk->sk_state != TCP_ESTABLISHED)
1694 goto out;
1695
1696 err = -EOPNOTSUPP;
1697 if (flags&MSG_OOB)
1698 goto out;
1699
1700 target = sock_rcvlowat(sk, flags&MSG_WAITALL, size);
1701 timeo = sock_rcvtimeo(sk, flags&MSG_DONTWAIT);
1702
1703 msg->msg_namelen = 0;
1704
1705 /* Lock the socket to prevent queue disordering
1706 * while sleeps in memcpy_tomsg
1707 */
1708
1709 if (!siocb->scm) {
1710 siocb->scm = &tmp_scm;
1711 memset(&tmp_scm, 0, sizeof(tmp_scm));
1712 }
1713
1714 mutex_lock(&u->readlock);
1715
1716 do
1717 {
1718 int chunk;
1719 struct sk_buff *skb;
1720
1721 skb = skb_dequeue(&sk->sk_receive_queue);
1722 if (skb==NULL)
1723 {
1724 if (copied >= target)
1725 break;
1726
1727 /*
1728 * POSIX 1003.1g mandates this order.
1729 */
1730
1731 if ((err = sock_error(sk)) != 0)
1732 break;
1733 if (sk->sk_shutdown & RCV_SHUTDOWN)
1734 break;
1735 err = -EAGAIN;
1736 if (!timeo)
1737 break;
1738 mutex_unlock(&u->readlock);
1739
1740 timeo = unix_stream_data_wait(sk, timeo);
1741
1742 if (signal_pending(current)) {
1743 err = sock_intr_errno(timeo);
1744 goto out;
1745 }
1746 mutex_lock(&u->readlock);
1747 continue;
1748 }
1749
1750 if (check_creds) {
1751 /* Never glue messages from different writers */
1752 if (memcmp(UNIXCREDS(skb), &siocb->scm->creds, sizeof(siocb->scm->creds)) != 0) {
1753 skb_queue_head(&sk->sk_receive_queue, skb);
1754 break;
1755 }
1756 } else {
1757 /* Copy credentials */
1758 siocb->scm->creds = *UNIXCREDS(skb);
1759 check_creds = 1;
1760 }
1761
1762 /* Copy address just once */
1763 if (sunaddr)
1764 {
1765 unix_copy_addr(msg, skb->sk);
1766 sunaddr = NULL;
1767 }
1768
1769 chunk = min_t(unsigned int, skb->len, size);
1770 if (memcpy_toiovec(msg->msg_iov, skb->data, chunk)) {
1771 skb_queue_head(&sk->sk_receive_queue, skb);
1772 if (copied == 0)
1773 copied = -EFAULT;
1774 break;
1775 }
1776 copied += chunk;
1777 size -= chunk;
1778
1779 /* Mark read part of skb as used */
1780 if (!(flags & MSG_PEEK))
1781 {
1782 skb_pull(skb, chunk);
1783
1784 if (UNIXCB(skb).fp)
1785 unix_detach_fds(siocb->scm, skb);
1786
1787 /* put the skb back if we didn't use it up.. */
1788 if (skb->len)
1789 {
1790 skb_queue_head(&sk->sk_receive_queue, skb);
1791 break;
1792 }
1793
1794 kfree_skb(skb);
1795
1796 if (siocb->scm->fp)
1797 break;
1798 }
1799 else
1800 {
1801 /* It is questionable, see note in unix_dgram_recvmsg.
1802 */
1803 if (UNIXCB(skb).fp)
1804 siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1805
1806 /* put message back and return */
1807 skb_queue_head(&sk->sk_receive_queue, skb);
1808 break;
1809 }
1810 } while (size);
1811
1812 mutex_unlock(&u->readlock);
1813 scm_recv(sock, msg, siocb->scm, flags);
1814 out:
1815 return copied ? : err;
1816 }
1817
1818 static int unix_shutdown(struct socket *sock, int mode)
1819 {
1820 struct sock *sk = sock->sk;
1821 struct sock *other;
1822
1823 mode = (mode+1)&(RCV_SHUTDOWN|SEND_SHUTDOWN);
1824
1825 if (mode) {
1826 unix_state_wlock(sk);
1827 sk->sk_shutdown |= mode;
1828 other=unix_peer(sk);
1829 if (other)
1830 sock_hold(other);
1831 unix_state_wunlock(sk);
1832 sk->sk_state_change(sk);
1833
1834 if (other &&
1835 (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET)) {
1836
1837 int peer_mode = 0;
1838
1839 if (mode&RCV_SHUTDOWN)
1840 peer_mode |= SEND_SHUTDOWN;
1841 if (mode&SEND_SHUTDOWN)
1842 peer_mode |= RCV_SHUTDOWN;
1843 unix_state_wlock(other);
1844 other->sk_shutdown |= peer_mode;
1845 unix_state_wunlock(other);
1846 other->sk_state_change(other);
1847 read_lock(&other->sk_callback_lock);
1848 if (peer_mode == SHUTDOWN_MASK)
1849 sk_wake_async(other,1,POLL_HUP);
1850 else if (peer_mode & RCV_SHUTDOWN)
1851 sk_wake_async(other,1,POLL_IN);
1852 read_unlock(&other->sk_callback_lock);
1853 }
1854 if (other)
1855 sock_put(other);
1856 }
1857 return 0;
1858 }
1859
1860 static int unix_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1861 {
1862 struct sock *sk = sock->sk;
1863 long amount=0;
1864 int err;
1865
1866 switch(cmd)
1867 {
1868 case SIOCOUTQ:
1869 amount = atomic_read(&sk->sk_wmem_alloc);
1870 err = put_user(amount, (int __user *)arg);
1871 break;
1872 case SIOCINQ:
1873 {
1874 struct sk_buff *skb;
1875
1876 if (sk->sk_state == TCP_LISTEN) {
1877 err = -EINVAL;
1878 break;
1879 }
1880
1881 spin_lock(&sk->sk_receive_queue.lock);
1882 if (sk->sk_type == SOCK_STREAM ||
1883 sk->sk_type == SOCK_SEQPACKET) {
1884 skb_queue_walk(&sk->sk_receive_queue, skb)
1885 amount += skb->len;
1886 } else {
1887 skb = skb_peek(&sk->sk_receive_queue);
1888 if (skb)
1889 amount=skb->len;
1890 }
1891 spin_unlock(&sk->sk_receive_queue.lock);
1892 err = put_user(amount, (int __user *)arg);
1893 break;
1894 }
1895
1896 default:
1897 err = -ENOIOCTLCMD;
1898 break;
1899 }
1900 return err;
1901 }
1902
1903 static unsigned int unix_poll(struct file * file, struct socket *sock, poll_table *wait)
1904 {
1905 struct sock *sk = sock->sk;
1906 unsigned int mask;
1907
1908 poll_wait(file, sk->sk_sleep, wait);
1909 mask = 0;
1910
1911 /* exceptional events? */
1912 if (sk->sk_err)
1913 mask |= POLLERR;
1914 if (sk->sk_shutdown == SHUTDOWN_MASK)
1915 mask |= POLLHUP;
1916 if (sk->sk_shutdown & RCV_SHUTDOWN)
1917 mask |= POLLRDHUP;
1918
1919 /* readable? */
1920 if (!skb_queue_empty(&sk->sk_receive_queue) ||
1921 (sk->sk_shutdown & RCV_SHUTDOWN))
1922 mask |= POLLIN | POLLRDNORM;
1923
1924 /* Connection-based need to check for termination and startup */
1925 if ((sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) && sk->sk_state == TCP_CLOSE)
1926 mask |= POLLHUP;
1927
1928 /*
1929 * we set writable also when the other side has shut down the
1930 * connection. This prevents stuck sockets.
1931 */
1932 if (unix_writable(sk))
1933 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
1934
1935 return mask;
1936 }
1937
1938
1939 #ifdef CONFIG_PROC_FS
1940 static struct sock *unix_seq_idx(int *iter, loff_t pos)
1941 {
1942 loff_t off = 0;
1943 struct sock *s;
1944
1945 for (s = first_unix_socket(iter); s; s = next_unix_socket(iter, s)) {
1946 if (off == pos)
1947 return s;
1948 ++off;
1949 }
1950 return NULL;
1951 }
1952
1953
1954 static void *unix_seq_start(struct seq_file *seq, loff_t *pos)
1955 {
1956 spin_lock(&unix_table_lock);
1957 return *pos ? unix_seq_idx(seq->private, *pos - 1) : ((void *) 1);
1958 }
1959
1960 static void *unix_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1961 {
1962 ++*pos;
1963
1964 if (v == (void *)1)
1965 return first_unix_socket(seq->private);
1966 return next_unix_socket(seq->private, v);
1967 }
1968
1969 static void unix_seq_stop(struct seq_file *seq, void *v)
1970 {
1971 spin_unlock(&unix_table_lock);
1972 }
1973
1974 static int unix_seq_show(struct seq_file *seq, void *v)
1975 {
1976
1977 if (v == (void *)1)
1978 seq_puts(seq, "Num RefCount Protocol Flags Type St "
1979 "Inode Path\n");
1980 else {
1981 struct sock *s = v;
1982 struct unix_sock *u = unix_sk(s);
1983 unix_state_rlock(s);
1984
1985 seq_printf(seq, "%p: %08X %08X %08X %04X %02X %5lu",
1986 s,
1987 atomic_read(&s->sk_refcnt),
1988 0,
1989 s->sk_state == TCP_LISTEN ? __SO_ACCEPTCON : 0,
1990 s->sk_type,
1991 s->sk_socket ?
1992 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTED : SS_UNCONNECTED) :
1993 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTING : SS_DISCONNECTING),
1994 sock_i_ino(s));
1995
1996 if (u->addr) {
1997 int i, len;
1998 seq_putc(seq, ' ');
1999
2000 i = 0;
2001 len = u->addr->len - sizeof(short);
2002 if (!UNIX_ABSTRACT(s))
2003 len--;
2004 else {
2005 seq_putc(seq, '@');
2006 i++;
2007 }
2008 for ( ; i < len; i++)
2009 seq_putc(seq, u->addr->name->sun_path[i]);
2010 }
2011 unix_state_runlock(s);
2012 seq_putc(seq, '\n');
2013 }
2014
2015 return 0;
2016 }
2017
2018 static struct seq_operations unix_seq_ops = {
2019 .start = unix_seq_start,
2020 .next = unix_seq_next,
2021 .stop = unix_seq_stop,
2022 .show = unix_seq_show,
2023 };
2024
2025
2026 static int unix_seq_open(struct inode *inode, struct file *file)
2027 {
2028 struct seq_file *seq;
2029 int rc = -ENOMEM;
2030 int *iter = kmalloc(sizeof(int), GFP_KERNEL);
2031
2032 if (!iter)
2033 goto out;
2034
2035 rc = seq_open(file, &unix_seq_ops);
2036 if (rc)
2037 goto out_kfree;
2038
2039 seq = file->private_data;
2040 seq->private = iter;
2041 *iter = 0;
2042 out:
2043 return rc;
2044 out_kfree:
2045 kfree(iter);
2046 goto out;
2047 }
2048
2049 static struct file_operations unix_seq_fops = {
2050 .owner = THIS_MODULE,
2051 .open = unix_seq_open,
2052 .read = seq_read,
2053 .llseek = seq_lseek,
2054 .release = seq_release_private,
2055 };
2056
2057 #endif
2058
2059 static struct net_proto_family unix_family_ops = {
2060 .family = PF_UNIX,
2061 .create = unix_create,
2062 .owner = THIS_MODULE,
2063 };
2064
2065 static int __init af_unix_init(void)
2066 {
2067 int rc = -1;
2068 struct sk_buff *dummy_skb;
2069
2070 if (sizeof(struct unix_skb_parms) > sizeof(dummy_skb->cb)) {
2071 printk(KERN_CRIT "%s: panic\n", __FUNCTION__);
2072 goto out;
2073 }
2074
2075 rc = proto_register(&unix_proto, 1);
2076 if (rc != 0) {
2077 printk(KERN_CRIT "%s: Cannot create unix_sock SLAB cache!\n",
2078 __FUNCTION__);
2079 goto out;
2080 }
2081
2082 sock_register(&unix_family_ops);
2083 #ifdef CONFIG_PROC_FS
2084 proc_net_fops_create("unix", 0, &unix_seq_fops);
2085 #endif
2086 unix_sysctl_register();
2087 out:
2088 return rc;
2089 }
2090
2091 static void __exit af_unix_exit(void)
2092 {
2093 sock_unregister(PF_UNIX);
2094 unix_sysctl_unregister();
2095 proc_net_remove("unix");
2096 proto_unregister(&unix_proto);
2097 }
2098
2099 module_init(af_unix_init);
2100 module_exit(af_unix_exit);
2101
2102 MODULE_LICENSE("GPL");
2103 MODULE_ALIAS_NETPROTO(PF_UNIX);
Linux kernel - Deliverables
This page took 0.077563 seconds and 5 git commands to generate.