cpumask: Use optimized CPUMASK_ALLOC macros in the centrino_target
[linux-2.6/mini2440.git] / net / unix / af_unix.c
blob783317dacd3090f7d602b3c56e82859f3cc22a46
1 /*
2 * NET4: Implementation of BSD Unix domain sockets.
4 * Authors: Alan Cox, <alan.cox@linux.org>
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.
11 * Version: $Id: af_unix.c,v 1.133 2002/02/08 03:57:19 davem Exp $
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+)
55 * Known differences from reference BSD that was tested:
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)
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.
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.
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>
120 static struct hlist_head unix_socket_table[UNIX_HASH_SIZE + 1];
121 static DEFINE_SPINLOCK(unix_table_lock);
122 static atomic_t unix_nr_socks = ATOMIC_INIT(0);
124 #define unix_sockets_unbound (&unix_socket_table[UNIX_HASH_SIZE])
126 #define UNIX_ABSTRACT(sk) (unix_sk(sk)->addr->hash != UNIX_HASH_SIZE)
128 #ifdef CONFIG_SECURITY_NETWORK
129 static void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb)
131 memcpy(UNIXSID(skb), &scm->secid, sizeof(u32));
134 static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb)
136 scm->secid = *UNIXSID(skb);
138 #else
139 static inline void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb)
142 static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb)
144 #endif /* CONFIG_SECURITY_NETWORK */
147 * SMP locking strategy:
148 * hash table is protected with spinlock unix_table_lock
149 * each socket state is protected by separate rwlock.
152 static inline unsigned unix_hash_fold(__wsum n)
154 unsigned hash = (__force unsigned)n;
155 hash ^= hash>>16;
156 hash ^= hash>>8;
157 return hash&(UNIX_HASH_SIZE-1);
160 #define unix_peer(sk) (unix_sk(sk)->peer)
162 static inline int unix_our_peer(struct sock *sk, struct sock *osk)
164 return unix_peer(osk) == sk;
167 static inline int unix_may_send(struct sock *sk, struct sock *osk)
169 return (unix_peer(osk) == NULL || unix_our_peer(sk, osk));
172 static inline int unix_recvq_full(struct sock const *sk)
174 return skb_queue_len(&sk->sk_receive_queue) > sk->sk_max_ack_backlog;
177 static struct sock *unix_peer_get(struct sock *s)
179 struct sock *peer;
181 unix_state_lock(s);
182 peer = unix_peer(s);
183 if (peer)
184 sock_hold(peer);
185 unix_state_unlock(s);
186 return peer;
189 static inline void unix_release_addr(struct unix_address *addr)
191 if (atomic_dec_and_test(&addr->refcnt))
192 kfree(addr);
196 * Check unix socket name:
197 * - should be not zero length.
198 * - if started by not zero, should be NULL terminated (FS object)
199 * - if started by zero, it is abstract name.
202 static int unix_mkname(struct sockaddr_un * sunaddr, int len, unsigned *hashp)
204 if (len <= sizeof(short) || len > sizeof(*sunaddr))
205 return -EINVAL;
206 if (!sunaddr || sunaddr->sun_family != AF_UNIX)
207 return -EINVAL;
208 if (sunaddr->sun_path[0]) {
210 * This may look like an off by one error but it is a bit more
211 * subtle. 108 is the longest valid AF_UNIX path for a binding.
212 * sun_path[108] doesnt as such exist. However in kernel space
213 * we are guaranteed that it is a valid memory location in our
214 * kernel address buffer.
216 ((char *)sunaddr)[len]=0;
217 len = strlen(sunaddr->sun_path)+1+sizeof(short);
218 return len;
221 *hashp = unix_hash_fold(csum_partial((char*)sunaddr, len, 0));
222 return len;
225 static void __unix_remove_socket(struct sock *sk)
227 sk_del_node_init(sk);
230 static void __unix_insert_socket(struct hlist_head *list, struct sock *sk)
232 BUG_TRAP(sk_unhashed(sk));
233 sk_add_node(sk, list);
236 static inline void unix_remove_socket(struct sock *sk)
238 spin_lock(&unix_table_lock);
239 __unix_remove_socket(sk);
240 spin_unlock(&unix_table_lock);
243 static inline void unix_insert_socket(struct hlist_head *list, struct sock *sk)
245 spin_lock(&unix_table_lock);
246 __unix_insert_socket(list, sk);
247 spin_unlock(&unix_table_lock);
250 static struct sock *__unix_find_socket_byname(struct net *net,
251 struct sockaddr_un *sunname,
252 int len, int type, unsigned hash)
254 struct sock *s;
255 struct hlist_node *node;
257 sk_for_each(s, node, &unix_socket_table[hash ^ type]) {
258 struct unix_sock *u = unix_sk(s);
260 if (!net_eq(sock_net(s), net))
261 continue;
263 if (u->addr->len == len &&
264 !memcmp(u->addr->name, sunname, len))
265 goto found;
267 s = NULL;
268 found:
269 return s;
272 static inline struct sock *unix_find_socket_byname(struct net *net,
273 struct sockaddr_un *sunname,
274 int len, int type,
275 unsigned hash)
277 struct sock *s;
279 spin_lock(&unix_table_lock);
280 s = __unix_find_socket_byname(net, sunname, len, type, hash);
281 if (s)
282 sock_hold(s);
283 spin_unlock(&unix_table_lock);
284 return s;
287 static struct sock *unix_find_socket_byinode(struct net *net, struct inode *i)
289 struct sock *s;
290 struct hlist_node *node;
292 spin_lock(&unix_table_lock);
293 sk_for_each(s, node,
294 &unix_socket_table[i->i_ino & (UNIX_HASH_SIZE - 1)]) {
295 struct dentry *dentry = unix_sk(s)->dentry;
297 if (!net_eq(sock_net(s), net))
298 continue;
300 if(dentry && dentry->d_inode == i)
302 sock_hold(s);
303 goto found;
306 s = NULL;
307 found:
308 spin_unlock(&unix_table_lock);
309 return s;
312 static inline int unix_writable(struct sock *sk)
314 return (atomic_read(&sk->sk_wmem_alloc) << 2) <= sk->sk_sndbuf;
317 static void unix_write_space(struct sock *sk)
319 read_lock(&sk->sk_callback_lock);
320 if (unix_writable(sk)) {
321 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
322 wake_up_interruptible_sync(sk->sk_sleep);
323 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
325 read_unlock(&sk->sk_callback_lock);
328 /* When dgram socket disconnects (or changes its peer), we clear its receive
329 * queue of packets arrived from previous peer. First, it allows to do
330 * flow control based only on wmem_alloc; second, sk connected to peer
331 * may receive messages only from that peer. */
332 static void unix_dgram_disconnected(struct sock *sk, struct sock *other)
334 if (!skb_queue_empty(&sk->sk_receive_queue)) {
335 skb_queue_purge(&sk->sk_receive_queue);
336 wake_up_interruptible_all(&unix_sk(sk)->peer_wait);
338 /* If one link of bidirectional dgram pipe is disconnected,
339 * we signal error. Messages are lost. Do not make this,
340 * when peer was not connected to us.
342 if (!sock_flag(other, SOCK_DEAD) && unix_peer(other) == sk) {
343 other->sk_err = ECONNRESET;
344 other->sk_error_report(other);
349 static void unix_sock_destructor(struct sock *sk)
351 struct unix_sock *u = unix_sk(sk);
353 skb_queue_purge(&sk->sk_receive_queue);
355 BUG_TRAP(!atomic_read(&sk->sk_wmem_alloc));
356 BUG_TRAP(sk_unhashed(sk));
357 BUG_TRAP(!sk->sk_socket);
358 if (!sock_flag(sk, SOCK_DEAD)) {
359 printk("Attempt to release alive unix socket: %p\n", sk);
360 return;
363 if (u->addr)
364 unix_release_addr(u->addr);
366 atomic_dec(&unix_nr_socks);
367 #ifdef UNIX_REFCNT_DEBUG
368 printk(KERN_DEBUG "UNIX %p is destroyed, %d are still alive.\n", sk, atomic_read(&unix_nr_socks));
369 #endif
372 static int unix_release_sock (struct sock *sk, int embrion)
374 struct unix_sock *u = unix_sk(sk);
375 struct dentry *dentry;
376 struct vfsmount *mnt;
377 struct sock *skpair;
378 struct sk_buff *skb;
379 int state;
381 unix_remove_socket(sk);
383 /* Clear state */
384 unix_state_lock(sk);
385 sock_orphan(sk);
386 sk->sk_shutdown = SHUTDOWN_MASK;
387 dentry = u->dentry;
388 u->dentry = NULL;
389 mnt = u->mnt;
390 u->mnt = NULL;
391 state = sk->sk_state;
392 sk->sk_state = TCP_CLOSE;
393 unix_state_unlock(sk);
395 wake_up_interruptible_all(&u->peer_wait);
397 skpair=unix_peer(sk);
399 if (skpair!=NULL) {
400 if (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) {
401 unix_state_lock(skpair);
402 /* No more writes */
403 skpair->sk_shutdown = SHUTDOWN_MASK;
404 if (!skb_queue_empty(&sk->sk_receive_queue) || embrion)
405 skpair->sk_err = ECONNRESET;
406 unix_state_unlock(skpair);
407 skpair->sk_state_change(skpair);
408 read_lock(&skpair->sk_callback_lock);
409 sk_wake_async(skpair, SOCK_WAKE_WAITD, POLL_HUP);
410 read_unlock(&skpair->sk_callback_lock);
412 sock_put(skpair); /* It may now die */
413 unix_peer(sk) = NULL;
416 /* Try to flush out this socket. Throw out buffers at least */
418 while ((skb = skb_dequeue(&sk->sk_receive_queue)) != NULL) {
419 if (state==TCP_LISTEN)
420 unix_release_sock(skb->sk, 1);
421 /* passed fds are erased in the kfree_skb hook */
422 kfree_skb(skb);
425 if (dentry) {
426 dput(dentry);
427 mntput(mnt);
430 sock_put(sk);
432 /* ---- Socket is dead now and most probably destroyed ---- */
435 * Fixme: BSD difference: In BSD all sockets connected to use get
436 * ECONNRESET and we die on the spot. In Linux we behave
437 * like files and pipes do and wait for the last
438 * dereference.
440 * Can't we simply set sock->err?
442 * What the above comment does talk about? --ANK(980817)
445 if (unix_tot_inflight)
446 unix_gc(); /* Garbage collect fds */
448 return 0;
451 static int unix_listen(struct socket *sock, int backlog)
453 int err;
454 struct sock *sk = sock->sk;
455 struct unix_sock *u = unix_sk(sk);
457 err = -EOPNOTSUPP;
458 if (sock->type!=SOCK_STREAM && sock->type!=SOCK_SEQPACKET)
459 goto out; /* Only stream/seqpacket sockets accept */
460 err = -EINVAL;
461 if (!u->addr)
462 goto out; /* No listens on an unbound socket */
463 unix_state_lock(sk);
464 if (sk->sk_state != TCP_CLOSE && sk->sk_state != TCP_LISTEN)
465 goto out_unlock;
466 if (backlog > sk->sk_max_ack_backlog)
467 wake_up_interruptible_all(&u->peer_wait);
468 sk->sk_max_ack_backlog = backlog;
469 sk->sk_state = TCP_LISTEN;
470 /* set credentials so connect can copy them */
471 sk->sk_peercred.pid = task_tgid_vnr(current);
472 sk->sk_peercred.uid = current->euid;
473 sk->sk_peercred.gid = current->egid;
474 err = 0;
476 out_unlock:
477 unix_state_unlock(sk);
478 out:
479 return err;
482 static int unix_release(struct socket *);
483 static int unix_bind(struct socket *, struct sockaddr *, int);
484 static int unix_stream_connect(struct socket *, struct sockaddr *,
485 int addr_len, int flags);
486 static int unix_socketpair(struct socket *, struct socket *);
487 static int unix_accept(struct socket *, struct socket *, int);
488 static int unix_getname(struct socket *, struct sockaddr *, int *, int);
489 static unsigned int unix_poll(struct file *, struct socket *, poll_table *);
490 static unsigned int unix_dgram_poll(struct file *, struct socket *,
491 poll_table *);
492 static int unix_ioctl(struct socket *, unsigned int, unsigned long);
493 static int unix_shutdown(struct socket *, int);
494 static int unix_stream_sendmsg(struct kiocb *, struct socket *,
495 struct msghdr *, size_t);
496 static int unix_stream_recvmsg(struct kiocb *, struct socket *,
497 struct msghdr *, size_t, int);
498 static int unix_dgram_sendmsg(struct kiocb *, struct socket *,
499 struct msghdr *, size_t);
500 static int unix_dgram_recvmsg(struct kiocb *, struct socket *,
501 struct msghdr *, size_t, int);
502 static int unix_dgram_connect(struct socket *, struct sockaddr *,
503 int, int);
504 static int unix_seqpacket_sendmsg(struct kiocb *, struct socket *,
505 struct msghdr *, size_t);
507 static const struct proto_ops unix_stream_ops = {
508 .family = PF_UNIX,
509 .owner = THIS_MODULE,
510 .release = unix_release,
511 .bind = unix_bind,
512 .connect = unix_stream_connect,
513 .socketpair = unix_socketpair,
514 .accept = unix_accept,
515 .getname = unix_getname,
516 .poll = unix_poll,
517 .ioctl = unix_ioctl,
518 .listen = unix_listen,
519 .shutdown = unix_shutdown,
520 .setsockopt = sock_no_setsockopt,
521 .getsockopt = sock_no_getsockopt,
522 .sendmsg = unix_stream_sendmsg,
523 .recvmsg = unix_stream_recvmsg,
524 .mmap = sock_no_mmap,
525 .sendpage = sock_no_sendpage,
528 static const struct proto_ops unix_dgram_ops = {
529 .family = PF_UNIX,
530 .owner = THIS_MODULE,
531 .release = unix_release,
532 .bind = unix_bind,
533 .connect = unix_dgram_connect,
534 .socketpair = unix_socketpair,
535 .accept = sock_no_accept,
536 .getname = unix_getname,
537 .poll = unix_dgram_poll,
538 .ioctl = unix_ioctl,
539 .listen = sock_no_listen,
540 .shutdown = unix_shutdown,
541 .setsockopt = sock_no_setsockopt,
542 .getsockopt = sock_no_getsockopt,
543 .sendmsg = unix_dgram_sendmsg,
544 .recvmsg = unix_dgram_recvmsg,
545 .mmap = sock_no_mmap,
546 .sendpage = sock_no_sendpage,
549 static const struct proto_ops unix_seqpacket_ops = {
550 .family = PF_UNIX,
551 .owner = THIS_MODULE,
552 .release = unix_release,
553 .bind = unix_bind,
554 .connect = unix_stream_connect,
555 .socketpair = unix_socketpair,
556 .accept = unix_accept,
557 .getname = unix_getname,
558 .poll = unix_dgram_poll,
559 .ioctl = unix_ioctl,
560 .listen = unix_listen,
561 .shutdown = unix_shutdown,
562 .setsockopt = sock_no_setsockopt,
563 .getsockopt = sock_no_getsockopt,
564 .sendmsg = unix_seqpacket_sendmsg,
565 .recvmsg = unix_dgram_recvmsg,
566 .mmap = sock_no_mmap,
567 .sendpage = sock_no_sendpage,
570 static struct proto unix_proto = {
571 .name = "UNIX",
572 .owner = THIS_MODULE,
573 .obj_size = sizeof(struct unix_sock),
577 * AF_UNIX sockets do not interact with hardware, hence they
578 * dont trigger interrupts - so it's safe for them to have
579 * bh-unsafe locking for their sk_receive_queue.lock. Split off
580 * this special lock-class by reinitializing the spinlock key:
582 static struct lock_class_key af_unix_sk_receive_queue_lock_key;
584 static struct sock * unix_create1(struct net *net, struct socket *sock)
586 struct sock *sk = NULL;
587 struct unix_sock *u;
589 atomic_inc(&unix_nr_socks);
590 if (atomic_read(&unix_nr_socks) > 2 * get_max_files())
591 goto out;
593 sk = sk_alloc(net, PF_UNIX, GFP_KERNEL, &unix_proto);
594 if (!sk)
595 goto out;
597 sock_init_data(sock,sk);
598 lockdep_set_class(&sk->sk_receive_queue.lock,
599 &af_unix_sk_receive_queue_lock_key);
601 sk->sk_write_space = unix_write_space;
602 sk->sk_max_ack_backlog = net->unx.sysctl_max_dgram_qlen;
603 sk->sk_destruct = unix_sock_destructor;
604 u = unix_sk(sk);
605 u->dentry = NULL;
606 u->mnt = NULL;
607 spin_lock_init(&u->lock);
608 atomic_set(&u->inflight, 0);
609 INIT_LIST_HEAD(&u->link);
610 mutex_init(&u->readlock); /* single task reading lock */
611 init_waitqueue_head(&u->peer_wait);
612 unix_insert_socket(unix_sockets_unbound, sk);
613 out:
614 if (sk == NULL)
615 atomic_dec(&unix_nr_socks);
616 return sk;
619 static int unix_create(struct net *net, struct socket *sock, int protocol)
621 if (protocol && protocol != PF_UNIX)
622 return -EPROTONOSUPPORT;
624 sock->state = SS_UNCONNECTED;
626 switch (sock->type) {
627 case SOCK_STREAM:
628 sock->ops = &unix_stream_ops;
629 break;
631 * Believe it or not BSD has AF_UNIX, SOCK_RAW though
632 * nothing uses it.
634 case SOCK_RAW:
635 sock->type=SOCK_DGRAM;
636 case SOCK_DGRAM:
637 sock->ops = &unix_dgram_ops;
638 break;
639 case SOCK_SEQPACKET:
640 sock->ops = &unix_seqpacket_ops;
641 break;
642 default:
643 return -ESOCKTNOSUPPORT;
646 return unix_create1(net, sock) ? 0 : -ENOMEM;
649 static int unix_release(struct socket *sock)
651 struct sock *sk = sock->sk;
653 if (!sk)
654 return 0;
656 sock->sk = NULL;
658 return unix_release_sock (sk, 0);
661 static int unix_autobind(struct socket *sock)
663 struct sock *sk = sock->sk;
664 struct net *net = sock_net(sk);
665 struct unix_sock *u = unix_sk(sk);
666 static u32 ordernum = 1;
667 struct unix_address * addr;
668 int err;
670 mutex_lock(&u->readlock);
672 err = 0;
673 if (u->addr)
674 goto out;
676 err = -ENOMEM;
677 addr = kzalloc(sizeof(*addr) + sizeof(short) + 16, GFP_KERNEL);
678 if (!addr)
679 goto out;
681 addr->name->sun_family = AF_UNIX;
682 atomic_set(&addr->refcnt, 1);
684 retry:
685 addr->len = sprintf(addr->name->sun_path+1, "%05x", ordernum) + 1 + sizeof(short);
686 addr->hash = unix_hash_fold(csum_partial((void*)addr->name, addr->len, 0));
688 spin_lock(&unix_table_lock);
689 ordernum = (ordernum+1)&0xFFFFF;
691 if (__unix_find_socket_byname(net, addr->name, addr->len, sock->type,
692 addr->hash)) {
693 spin_unlock(&unix_table_lock);
694 /* Sanity yield. It is unusual case, but yet... */
695 if (!(ordernum&0xFF))
696 yield();
697 goto retry;
699 addr->hash ^= sk->sk_type;
701 __unix_remove_socket(sk);
702 u->addr = addr;
703 __unix_insert_socket(&unix_socket_table[addr->hash], sk);
704 spin_unlock(&unix_table_lock);
705 err = 0;
707 out: mutex_unlock(&u->readlock);
708 return err;
711 static struct sock *unix_find_other(struct net *net,
712 struct sockaddr_un *sunname, int len,
713 int type, unsigned hash, int *error)
715 struct sock *u;
716 struct nameidata nd;
717 int err = 0;
719 if (sunname->sun_path[0]) {
720 err = path_lookup(sunname->sun_path, LOOKUP_FOLLOW, &nd);
721 if (err)
722 goto fail;
723 err = vfs_permission(&nd, MAY_WRITE);
724 if (err)
725 goto put_fail;
727 err = -ECONNREFUSED;
728 if (!S_ISSOCK(nd.path.dentry->d_inode->i_mode))
729 goto put_fail;
730 u = unix_find_socket_byinode(net, nd.path.dentry->d_inode);
731 if (!u)
732 goto put_fail;
734 if (u->sk_type == type)
735 touch_atime(nd.path.mnt, nd.path.dentry);
737 path_put(&nd.path);
739 err=-EPROTOTYPE;
740 if (u->sk_type != type) {
741 sock_put(u);
742 goto fail;
744 } else {
745 err = -ECONNREFUSED;
746 u=unix_find_socket_byname(net, sunname, len, type, hash);
747 if (u) {
748 struct dentry *dentry;
749 dentry = unix_sk(u)->dentry;
750 if (dentry)
751 touch_atime(unix_sk(u)->mnt, dentry);
752 } else
753 goto fail;
755 return u;
757 put_fail:
758 path_put(&nd.path);
759 fail:
760 *error=err;
761 return NULL;
765 static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
767 struct sock *sk = sock->sk;
768 struct net *net = sock_net(sk);
769 struct unix_sock *u = unix_sk(sk);
770 struct sockaddr_un *sunaddr=(struct sockaddr_un *)uaddr;
771 struct dentry * dentry = NULL;
772 struct nameidata nd;
773 int err;
774 unsigned hash;
775 struct unix_address *addr;
776 struct hlist_head *list;
778 err = -EINVAL;
779 if (sunaddr->sun_family != AF_UNIX)
780 goto out;
782 if (addr_len==sizeof(short)) {
783 err = unix_autobind(sock);
784 goto out;
787 err = unix_mkname(sunaddr, addr_len, &hash);
788 if (err < 0)
789 goto out;
790 addr_len = err;
792 mutex_lock(&u->readlock);
794 err = -EINVAL;
795 if (u->addr)
796 goto out_up;
798 err = -ENOMEM;
799 addr = kmalloc(sizeof(*addr)+addr_len, GFP_KERNEL);
800 if (!addr)
801 goto out_up;
803 memcpy(addr->name, sunaddr, addr_len);
804 addr->len = addr_len;
805 addr->hash = hash ^ sk->sk_type;
806 atomic_set(&addr->refcnt, 1);
808 if (sunaddr->sun_path[0]) {
809 unsigned int mode;
810 err = 0;
812 * Get the parent directory, calculate the hash for last
813 * component.
815 err = path_lookup(sunaddr->sun_path, LOOKUP_PARENT, &nd);
816 if (err)
817 goto out_mknod_parent;
819 dentry = lookup_create(&nd, 0);
820 err = PTR_ERR(dentry);
821 if (IS_ERR(dentry))
822 goto out_mknod_unlock;
825 * All right, let's create it.
827 mode = S_IFSOCK |
828 (SOCK_INODE(sock)->i_mode & ~current->fs->umask);
829 err = mnt_want_write(nd.path.mnt);
830 if (err)
831 goto out_mknod_dput;
832 err = vfs_mknod(nd.path.dentry->d_inode, dentry, mode, 0);
833 mnt_drop_write(nd.path.mnt);
834 if (err)
835 goto out_mknod_dput;
836 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
837 dput(nd.path.dentry);
838 nd.path.dentry = dentry;
840 addr->hash = UNIX_HASH_SIZE;
843 spin_lock(&unix_table_lock);
845 if (!sunaddr->sun_path[0]) {
846 err = -EADDRINUSE;
847 if (__unix_find_socket_byname(net, sunaddr, addr_len,
848 sk->sk_type, hash)) {
849 unix_release_addr(addr);
850 goto out_unlock;
853 list = &unix_socket_table[addr->hash];
854 } else {
855 list = &unix_socket_table[dentry->d_inode->i_ino & (UNIX_HASH_SIZE-1)];
856 u->dentry = nd.path.dentry;
857 u->mnt = nd.path.mnt;
860 err = 0;
861 __unix_remove_socket(sk);
862 u->addr = addr;
863 __unix_insert_socket(list, sk);
865 out_unlock:
866 spin_unlock(&unix_table_lock);
867 out_up:
868 mutex_unlock(&u->readlock);
869 out:
870 return err;
872 out_mknod_dput:
873 dput(dentry);
874 out_mknod_unlock:
875 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
876 path_put(&nd.path);
877 out_mknod_parent:
878 if (err==-EEXIST)
879 err=-EADDRINUSE;
880 unix_release_addr(addr);
881 goto out_up;
884 static void unix_state_double_lock(struct sock *sk1, struct sock *sk2)
886 if (unlikely(sk1 == sk2) || !sk2) {
887 unix_state_lock(sk1);
888 return;
890 if (sk1 < sk2) {
891 unix_state_lock(sk1);
892 unix_state_lock_nested(sk2);
893 } else {
894 unix_state_lock(sk2);
895 unix_state_lock_nested(sk1);
899 static void unix_state_double_unlock(struct sock *sk1, struct sock *sk2)
901 if (unlikely(sk1 == sk2) || !sk2) {
902 unix_state_unlock(sk1);
903 return;
905 unix_state_unlock(sk1);
906 unix_state_unlock(sk2);
909 static int unix_dgram_connect(struct socket *sock, struct sockaddr *addr,
910 int alen, int flags)
912 struct sock *sk = sock->sk;
913 struct net *net = sock_net(sk);
914 struct sockaddr_un *sunaddr=(struct sockaddr_un*)addr;
915 struct sock *other;
916 unsigned hash;
917 int err;
919 if (addr->sa_family != AF_UNSPEC) {
920 err = unix_mkname(sunaddr, alen, &hash);
921 if (err < 0)
922 goto out;
923 alen = err;
925 if (test_bit(SOCK_PASSCRED, &sock->flags) &&
926 !unix_sk(sk)->addr && (err = unix_autobind(sock)) != 0)
927 goto out;
929 restart:
930 other=unix_find_other(net, sunaddr, alen, sock->type, hash, &err);
931 if (!other)
932 goto out;
934 unix_state_double_lock(sk, other);
936 /* Apparently VFS overslept socket death. Retry. */
937 if (sock_flag(other, SOCK_DEAD)) {
938 unix_state_double_unlock(sk, other);
939 sock_put(other);
940 goto restart;
943 err = -EPERM;
944 if (!unix_may_send(sk, other))
945 goto out_unlock;
947 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
948 if (err)
949 goto out_unlock;
951 } else {
953 * 1003.1g breaking connected state with AF_UNSPEC
955 other = NULL;
956 unix_state_double_lock(sk, other);
960 * If it was connected, reconnect.
962 if (unix_peer(sk)) {
963 struct sock *old_peer = unix_peer(sk);
964 unix_peer(sk)=other;
965 unix_state_double_unlock(sk, other);
967 if (other != old_peer)
968 unix_dgram_disconnected(sk, old_peer);
969 sock_put(old_peer);
970 } else {
971 unix_peer(sk)=other;
972 unix_state_double_unlock(sk, other);
974 return 0;
976 out_unlock:
977 unix_state_double_unlock(sk, other);
978 sock_put(other);
979 out:
980 return err;
983 static long unix_wait_for_peer(struct sock *other, long timeo)
985 struct unix_sock *u = unix_sk(other);
986 int sched;
987 DEFINE_WAIT(wait);
989 prepare_to_wait_exclusive(&u->peer_wait, &wait, TASK_INTERRUPTIBLE);
991 sched = !sock_flag(other, SOCK_DEAD) &&
992 !(other->sk_shutdown & RCV_SHUTDOWN) &&
993 unix_recvq_full(other);
995 unix_state_unlock(other);
997 if (sched)
998 timeo = schedule_timeout(timeo);
1000 finish_wait(&u->peer_wait, &wait);
1001 return timeo;
1004 static int unix_stream_connect(struct socket *sock, struct sockaddr *uaddr,
1005 int addr_len, int flags)
1007 struct sockaddr_un *sunaddr=(struct sockaddr_un *)uaddr;
1008 struct sock *sk = sock->sk;
1009 struct net *net = sock_net(sk);
1010 struct unix_sock *u = unix_sk(sk), *newu, *otheru;
1011 struct sock *newsk = NULL;
1012 struct sock *other = NULL;
1013 struct sk_buff *skb = NULL;
1014 unsigned hash;
1015 int st;
1016 int err;
1017 long timeo;
1019 err = unix_mkname(sunaddr, addr_len, &hash);
1020 if (err < 0)
1021 goto out;
1022 addr_len = err;
1024 if (test_bit(SOCK_PASSCRED, &sock->flags)
1025 && !u->addr && (err = unix_autobind(sock)) != 0)
1026 goto out;
1028 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
1030 /* First of all allocate resources.
1031 If we will make it after state is locked,
1032 we will have to recheck all again in any case.
1035 err = -ENOMEM;
1037 /* create new sock for complete connection */
1038 newsk = unix_create1(sock_net(sk), NULL);
1039 if (newsk == NULL)
1040 goto out;
1042 /* Allocate skb for sending to listening sock */
1043 skb = sock_wmalloc(newsk, 1, 0, GFP_KERNEL);
1044 if (skb == NULL)
1045 goto out;
1047 restart:
1048 /* Find listening sock. */
1049 other = unix_find_other(net, sunaddr, addr_len, sk->sk_type, hash, &err);
1050 if (!other)
1051 goto out;
1053 /* Latch state of peer */
1054 unix_state_lock(other);
1056 /* Apparently VFS overslept socket death. Retry. */
1057 if (sock_flag(other, SOCK_DEAD)) {
1058 unix_state_unlock(other);
1059 sock_put(other);
1060 goto restart;
1063 err = -ECONNREFUSED;
1064 if (other->sk_state != TCP_LISTEN)
1065 goto out_unlock;
1067 if (unix_recvq_full(other)) {
1068 err = -EAGAIN;
1069 if (!timeo)
1070 goto out_unlock;
1072 timeo = unix_wait_for_peer(other, timeo);
1074 err = sock_intr_errno(timeo);
1075 if (signal_pending(current))
1076 goto out;
1077 sock_put(other);
1078 goto restart;
1081 /* Latch our state.
1083 It is tricky place. We need to grab write lock and cannot
1084 drop lock on peer. It is dangerous because deadlock is
1085 possible. Connect to self case and simultaneous
1086 attempt to connect are eliminated by checking socket
1087 state. other is TCP_LISTEN, if sk is TCP_LISTEN we
1088 check this before attempt to grab lock.
1090 Well, and we have to recheck the state after socket locked.
1092 st = sk->sk_state;
1094 switch (st) {
1095 case TCP_CLOSE:
1096 /* This is ok... continue with connect */
1097 break;
1098 case TCP_ESTABLISHED:
1099 /* Socket is already connected */
1100 err = -EISCONN;
1101 goto out_unlock;
1102 default:
1103 err = -EINVAL;
1104 goto out_unlock;
1107 unix_state_lock_nested(sk);
1109 if (sk->sk_state != st) {
1110 unix_state_unlock(sk);
1111 unix_state_unlock(other);
1112 sock_put(other);
1113 goto restart;
1116 err = security_unix_stream_connect(sock, other->sk_socket, newsk);
1117 if (err) {
1118 unix_state_unlock(sk);
1119 goto out_unlock;
1122 /* The way is open! Fastly set all the necessary fields... */
1124 sock_hold(sk);
1125 unix_peer(newsk) = sk;
1126 newsk->sk_state = TCP_ESTABLISHED;
1127 newsk->sk_type = sk->sk_type;
1128 newsk->sk_peercred.pid = task_tgid_vnr(current);
1129 newsk->sk_peercred.uid = current->euid;
1130 newsk->sk_peercred.gid = current->egid;
1131 newu = unix_sk(newsk);
1132 newsk->sk_sleep = &newu->peer_wait;
1133 otheru = unix_sk(other);
1135 /* copy address information from listening to new sock*/
1136 if (otheru->addr) {
1137 atomic_inc(&otheru->addr->refcnt);
1138 newu->addr = otheru->addr;
1140 if (otheru->dentry) {
1141 newu->dentry = dget(otheru->dentry);
1142 newu->mnt = mntget(otheru->mnt);
1145 /* Set credentials */
1146 sk->sk_peercred = other->sk_peercred;
1148 sock->state = SS_CONNECTED;
1149 sk->sk_state = TCP_ESTABLISHED;
1150 sock_hold(newsk);
1152 smp_mb__after_atomic_inc(); /* sock_hold() does an atomic_inc() */
1153 unix_peer(sk) = newsk;
1155 unix_state_unlock(sk);
1157 /* take ten and and send info to listening sock */
1158 spin_lock(&other->sk_receive_queue.lock);
1159 __skb_queue_tail(&other->sk_receive_queue, skb);
1160 spin_unlock(&other->sk_receive_queue.lock);
1161 unix_state_unlock(other);
1162 other->sk_data_ready(other, 0);
1163 sock_put(other);
1164 return 0;
1166 out_unlock:
1167 if (other)
1168 unix_state_unlock(other);
1170 out:
1171 if (skb)
1172 kfree_skb(skb);
1173 if (newsk)
1174 unix_release_sock(newsk, 0);
1175 if (other)
1176 sock_put(other);
1177 return err;
1180 static int unix_socketpair(struct socket *socka, struct socket *sockb)
1182 struct sock *ska=socka->sk, *skb = sockb->sk;
1184 /* Join our sockets back to back */
1185 sock_hold(ska);
1186 sock_hold(skb);
1187 unix_peer(ska)=skb;
1188 unix_peer(skb)=ska;
1189 ska->sk_peercred.pid = skb->sk_peercred.pid = task_tgid_vnr(current);
1190 ska->sk_peercred.uid = skb->sk_peercred.uid = current->euid;
1191 ska->sk_peercred.gid = skb->sk_peercred.gid = current->egid;
1193 if (ska->sk_type != SOCK_DGRAM) {
1194 ska->sk_state = TCP_ESTABLISHED;
1195 skb->sk_state = TCP_ESTABLISHED;
1196 socka->state = SS_CONNECTED;
1197 sockb->state = SS_CONNECTED;
1199 return 0;
1202 static int unix_accept(struct socket *sock, struct socket *newsock, int flags)
1204 struct sock *sk = sock->sk;
1205 struct sock *tsk;
1206 struct sk_buff *skb;
1207 int err;
1209 err = -EOPNOTSUPP;
1210 if (sock->type!=SOCK_STREAM && sock->type!=SOCK_SEQPACKET)
1211 goto out;
1213 err = -EINVAL;
1214 if (sk->sk_state != TCP_LISTEN)
1215 goto out;
1217 /* If socket state is TCP_LISTEN it cannot change (for now...),
1218 * so that no locks are necessary.
1221 skb = skb_recv_datagram(sk, 0, flags&O_NONBLOCK, &err);
1222 if (!skb) {
1223 /* This means receive shutdown. */
1224 if (err == 0)
1225 err = -EINVAL;
1226 goto out;
1229 tsk = skb->sk;
1230 skb_free_datagram(sk, skb);
1231 wake_up_interruptible(&unix_sk(sk)->peer_wait);
1233 /* attach accepted sock to socket */
1234 unix_state_lock(tsk);
1235 newsock->state = SS_CONNECTED;
1236 sock_graft(tsk, newsock);
1237 unix_state_unlock(tsk);
1238 return 0;
1240 out:
1241 return err;
1245 static int unix_getname(struct socket *sock, struct sockaddr *uaddr, int *uaddr_len, int peer)
1247 struct sock *sk = sock->sk;
1248 struct unix_sock *u;
1249 struct sockaddr_un *sunaddr=(struct sockaddr_un *)uaddr;
1250 int err = 0;
1252 if (peer) {
1253 sk = unix_peer_get(sk);
1255 err = -ENOTCONN;
1256 if (!sk)
1257 goto out;
1258 err = 0;
1259 } else {
1260 sock_hold(sk);
1263 u = unix_sk(sk);
1264 unix_state_lock(sk);
1265 if (!u->addr) {
1266 sunaddr->sun_family = AF_UNIX;
1267 sunaddr->sun_path[0] = 0;
1268 *uaddr_len = sizeof(short);
1269 } else {
1270 struct unix_address *addr = u->addr;
1272 *uaddr_len = addr->len;
1273 memcpy(sunaddr, addr->name, *uaddr_len);
1275 unix_state_unlock(sk);
1276 sock_put(sk);
1277 out:
1278 return err;
1281 static void unix_detach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1283 int i;
1285 scm->fp = UNIXCB(skb).fp;
1286 skb->destructor = sock_wfree;
1287 UNIXCB(skb).fp = NULL;
1289 for (i=scm->fp->count-1; i>=0; i--)
1290 unix_notinflight(scm->fp->fp[i]);
1293 static void unix_destruct_fds(struct sk_buff *skb)
1295 struct scm_cookie scm;
1296 memset(&scm, 0, sizeof(scm));
1297 unix_detach_fds(&scm, skb);
1299 /* Alas, it calls VFS */
1300 /* So fscking what? fput() had been SMP-safe since the last Summer */
1301 scm_destroy(&scm);
1302 sock_wfree(skb);
1305 static void unix_attach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1307 int i;
1308 for (i=scm->fp->count-1; i>=0; i--)
1309 unix_inflight(scm->fp->fp[i]);
1310 UNIXCB(skb).fp = scm->fp;
1311 skb->destructor = unix_destruct_fds;
1312 scm->fp = NULL;
1316 * Send AF_UNIX data.
1319 static int unix_dgram_sendmsg(struct kiocb *kiocb, struct socket *sock,
1320 struct msghdr *msg, size_t len)
1322 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1323 struct sock *sk = sock->sk;
1324 struct net *net = sock_net(sk);
1325 struct unix_sock *u = unix_sk(sk);
1326 struct sockaddr_un *sunaddr=msg->msg_name;
1327 struct sock *other = NULL;
1328 int namelen = 0; /* fake GCC */
1329 int err;
1330 unsigned hash;
1331 struct sk_buff *skb;
1332 long timeo;
1333 struct scm_cookie tmp_scm;
1335 if (NULL == siocb->scm)
1336 siocb->scm = &tmp_scm;
1337 err = scm_send(sock, msg, siocb->scm);
1338 if (err < 0)
1339 return err;
1341 err = -EOPNOTSUPP;
1342 if (msg->msg_flags&MSG_OOB)
1343 goto out;
1345 if (msg->msg_namelen) {
1346 err = unix_mkname(sunaddr, msg->msg_namelen, &hash);
1347 if (err < 0)
1348 goto out;
1349 namelen = err;
1350 } else {
1351 sunaddr = NULL;
1352 err = -ENOTCONN;
1353 other = unix_peer_get(sk);
1354 if (!other)
1355 goto out;
1358 if (test_bit(SOCK_PASSCRED, &sock->flags)
1359 && !u->addr && (err = unix_autobind(sock)) != 0)
1360 goto out;
1362 err = -EMSGSIZE;
1363 if (len > sk->sk_sndbuf - 32)
1364 goto out;
1366 skb = sock_alloc_send_skb(sk, len, msg->msg_flags&MSG_DONTWAIT, &err);
1367 if (skb==NULL)
1368 goto out;
1370 memcpy(UNIXCREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1371 if (siocb->scm->fp)
1372 unix_attach_fds(siocb->scm, skb);
1373 unix_get_secdata(siocb->scm, skb);
1375 skb_reset_transport_header(skb);
1376 err = memcpy_fromiovec(skb_put(skb,len), msg->msg_iov, len);
1377 if (err)
1378 goto out_free;
1380 timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1382 restart:
1383 if (!other) {
1384 err = -ECONNRESET;
1385 if (sunaddr == NULL)
1386 goto out_free;
1388 other = unix_find_other(net, sunaddr, namelen, sk->sk_type,
1389 hash, &err);
1390 if (other==NULL)
1391 goto out_free;
1394 unix_state_lock(other);
1395 err = -EPERM;
1396 if (!unix_may_send(sk, other))
1397 goto out_unlock;
1399 if (sock_flag(other, SOCK_DEAD)) {
1401 * Check with 1003.1g - what should
1402 * datagram error
1404 unix_state_unlock(other);
1405 sock_put(other);
1407 err = 0;
1408 unix_state_lock(sk);
1409 if (unix_peer(sk) == other) {
1410 unix_peer(sk)=NULL;
1411 unix_state_unlock(sk);
1413 unix_dgram_disconnected(sk, other);
1414 sock_put(other);
1415 err = -ECONNREFUSED;
1416 } else {
1417 unix_state_unlock(sk);
1420 other = NULL;
1421 if (err)
1422 goto out_free;
1423 goto restart;
1426 err = -EPIPE;
1427 if (other->sk_shutdown & RCV_SHUTDOWN)
1428 goto out_unlock;
1430 if (sk->sk_type != SOCK_SEQPACKET) {
1431 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1432 if (err)
1433 goto out_unlock;
1436 if (unix_peer(other) != sk && unix_recvq_full(other)) {
1437 if (!timeo) {
1438 err = -EAGAIN;
1439 goto out_unlock;
1442 timeo = unix_wait_for_peer(other, timeo);
1444 err = sock_intr_errno(timeo);
1445 if (signal_pending(current))
1446 goto out_free;
1448 goto restart;
1451 skb_queue_tail(&other->sk_receive_queue, skb);
1452 unix_state_unlock(other);
1453 other->sk_data_ready(other, len);
1454 sock_put(other);
1455 scm_destroy(siocb->scm);
1456 return len;
1458 out_unlock:
1459 unix_state_unlock(other);
1460 out_free:
1461 kfree_skb(skb);
1462 out:
1463 if (other)
1464 sock_put(other);
1465 scm_destroy(siocb->scm);
1466 return err;
1470 static int unix_stream_sendmsg(struct kiocb *kiocb, struct socket *sock,
1471 struct msghdr *msg, size_t len)
1473 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1474 struct sock *sk = sock->sk;
1475 struct sock *other = NULL;
1476 struct sockaddr_un *sunaddr=msg->msg_name;
1477 int err,size;
1478 struct sk_buff *skb;
1479 int sent=0;
1480 struct scm_cookie tmp_scm;
1482 if (NULL == siocb->scm)
1483 siocb->scm = &tmp_scm;
1484 err = scm_send(sock, msg, siocb->scm);
1485 if (err < 0)
1486 return err;
1488 err = -EOPNOTSUPP;
1489 if (msg->msg_flags&MSG_OOB)
1490 goto out_err;
1492 if (msg->msg_namelen) {
1493 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
1494 goto out_err;
1495 } else {
1496 sunaddr = NULL;
1497 err = -ENOTCONN;
1498 other = unix_peer(sk);
1499 if (!other)
1500 goto out_err;
1503 if (sk->sk_shutdown & SEND_SHUTDOWN)
1504 goto pipe_err;
1506 while(sent < len)
1509 * Optimisation for the fact that under 0.01% of X
1510 * messages typically need breaking up.
1513 size = len-sent;
1515 /* Keep two messages in the pipe so it schedules better */
1516 if (size > ((sk->sk_sndbuf >> 1) - 64))
1517 size = (sk->sk_sndbuf >> 1) - 64;
1519 if (size > SKB_MAX_ALLOC)
1520 size = SKB_MAX_ALLOC;
1523 * Grab a buffer
1526 skb=sock_alloc_send_skb(sk,size,msg->msg_flags&MSG_DONTWAIT, &err);
1528 if (skb==NULL)
1529 goto out_err;
1532 * If you pass two values to the sock_alloc_send_skb
1533 * it tries to grab the large buffer with GFP_NOFS
1534 * (which can fail easily), and if it fails grab the
1535 * fallback size buffer which is under a page and will
1536 * succeed. [Alan]
1538 size = min_t(int, size, skb_tailroom(skb));
1540 memcpy(UNIXCREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1541 if (siocb->scm->fp)
1542 unix_attach_fds(siocb->scm, skb);
1544 if ((err = memcpy_fromiovec(skb_put(skb,size), msg->msg_iov, size)) != 0) {
1545 kfree_skb(skb);
1546 goto out_err;
1549 unix_state_lock(other);
1551 if (sock_flag(other, SOCK_DEAD) ||
1552 (other->sk_shutdown & RCV_SHUTDOWN))
1553 goto pipe_err_free;
1555 skb_queue_tail(&other->sk_receive_queue, skb);
1556 unix_state_unlock(other);
1557 other->sk_data_ready(other, size);
1558 sent+=size;
1561 scm_destroy(siocb->scm);
1562 siocb->scm = NULL;
1564 return sent;
1566 pipe_err_free:
1567 unix_state_unlock(other);
1568 kfree_skb(skb);
1569 pipe_err:
1570 if (sent==0 && !(msg->msg_flags&MSG_NOSIGNAL))
1571 send_sig(SIGPIPE,current,0);
1572 err = -EPIPE;
1573 out_err:
1574 scm_destroy(siocb->scm);
1575 siocb->scm = NULL;
1576 return sent ? : err;
1579 static int unix_seqpacket_sendmsg(struct kiocb *kiocb, struct socket *sock,
1580 struct msghdr *msg, size_t len)
1582 int err;
1583 struct sock *sk = sock->sk;
1585 err = sock_error(sk);
1586 if (err)
1587 return err;
1589 if (sk->sk_state != TCP_ESTABLISHED)
1590 return -ENOTCONN;
1592 if (msg->msg_namelen)
1593 msg->msg_namelen = 0;
1595 return unix_dgram_sendmsg(kiocb, sock, msg, len);
1598 static void unix_copy_addr(struct msghdr *msg, struct sock *sk)
1600 struct unix_sock *u = unix_sk(sk);
1602 msg->msg_namelen = 0;
1603 if (u->addr) {
1604 msg->msg_namelen = u->addr->len;
1605 memcpy(msg->msg_name, u->addr->name, u->addr->len);
1609 static int unix_dgram_recvmsg(struct kiocb *iocb, struct socket *sock,
1610 struct msghdr *msg, size_t size,
1611 int flags)
1613 struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1614 struct scm_cookie tmp_scm;
1615 struct sock *sk = sock->sk;
1616 struct unix_sock *u = unix_sk(sk);
1617 int noblock = flags & MSG_DONTWAIT;
1618 struct sk_buff *skb;
1619 int err;
1621 err = -EOPNOTSUPP;
1622 if (flags&MSG_OOB)
1623 goto out;
1625 msg->msg_namelen = 0;
1627 mutex_lock(&u->readlock);
1629 skb = skb_recv_datagram(sk, flags, noblock, &err);
1630 if (!skb) {
1631 unix_state_lock(sk);
1632 /* Signal EOF on disconnected non-blocking SEQPACKET socket. */
1633 if (sk->sk_type == SOCK_SEQPACKET && err == -EAGAIN &&
1634 (sk->sk_shutdown & RCV_SHUTDOWN))
1635 err = 0;
1636 unix_state_unlock(sk);
1637 goto out_unlock;
1640 wake_up_interruptible_sync(&u->peer_wait);
1642 if (msg->msg_name)
1643 unix_copy_addr(msg, skb->sk);
1645 if (size > skb->len)
1646 size = skb->len;
1647 else if (size < skb->len)
1648 msg->msg_flags |= MSG_TRUNC;
1650 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, size);
1651 if (err)
1652 goto out_free;
1654 if (!siocb->scm) {
1655 siocb->scm = &tmp_scm;
1656 memset(&tmp_scm, 0, sizeof(tmp_scm));
1658 siocb->scm->creds = *UNIXCREDS(skb);
1659 unix_set_secdata(siocb->scm, skb);
1661 if (!(flags & MSG_PEEK))
1663 if (UNIXCB(skb).fp)
1664 unix_detach_fds(siocb->scm, skb);
1666 else
1668 /* It is questionable: on PEEK we could:
1669 - do not return fds - good, but too simple 8)
1670 - return fds, and do not return them on read (old strategy,
1671 apparently wrong)
1672 - clone fds (I chose it for now, it is the most universal
1673 solution)
1675 POSIX 1003.1g does not actually define this clearly
1676 at all. POSIX 1003.1g doesn't define a lot of things
1677 clearly however!
1680 if (UNIXCB(skb).fp)
1681 siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1683 err = size;
1685 scm_recv(sock, msg, siocb->scm, flags);
1687 out_free:
1688 skb_free_datagram(sk,skb);
1689 out_unlock:
1690 mutex_unlock(&u->readlock);
1691 out:
1692 return err;
1696 * Sleep until data has arrive. But check for races..
1699 static long unix_stream_data_wait(struct sock * sk, long timeo)
1701 DEFINE_WAIT(wait);
1703 unix_state_lock(sk);
1705 for (;;) {
1706 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1708 if (!skb_queue_empty(&sk->sk_receive_queue) ||
1709 sk->sk_err ||
1710 (sk->sk_shutdown & RCV_SHUTDOWN) ||
1711 signal_pending(current) ||
1712 !timeo)
1713 break;
1715 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1716 unix_state_unlock(sk);
1717 timeo = schedule_timeout(timeo);
1718 unix_state_lock(sk);
1719 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1722 finish_wait(sk->sk_sleep, &wait);
1723 unix_state_unlock(sk);
1724 return timeo;
1729 static int unix_stream_recvmsg(struct kiocb *iocb, struct socket *sock,
1730 struct msghdr *msg, size_t size,
1731 int flags)
1733 struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1734 struct scm_cookie tmp_scm;
1735 struct sock *sk = sock->sk;
1736 struct unix_sock *u = unix_sk(sk);
1737 struct sockaddr_un *sunaddr=msg->msg_name;
1738 int copied = 0;
1739 int check_creds = 0;
1740 int target;
1741 int err = 0;
1742 long timeo;
1744 err = -EINVAL;
1745 if (sk->sk_state != TCP_ESTABLISHED)
1746 goto out;
1748 err = -EOPNOTSUPP;
1749 if (flags&MSG_OOB)
1750 goto out;
1752 target = sock_rcvlowat(sk, flags&MSG_WAITALL, size);
1753 timeo = sock_rcvtimeo(sk, flags&MSG_DONTWAIT);
1755 msg->msg_namelen = 0;
1757 /* Lock the socket to prevent queue disordering
1758 * while sleeps in memcpy_tomsg
1761 if (!siocb->scm) {
1762 siocb->scm = &tmp_scm;
1763 memset(&tmp_scm, 0, sizeof(tmp_scm));
1766 mutex_lock(&u->readlock);
1770 int chunk;
1771 struct sk_buff *skb;
1773 unix_state_lock(sk);
1774 skb = skb_dequeue(&sk->sk_receive_queue);
1775 if (skb==NULL)
1777 if (copied >= target)
1778 goto unlock;
1781 * POSIX 1003.1g mandates this order.
1784 if ((err = sock_error(sk)) != 0)
1785 goto unlock;
1786 if (sk->sk_shutdown & RCV_SHUTDOWN)
1787 goto unlock;
1789 unix_state_unlock(sk);
1790 err = -EAGAIN;
1791 if (!timeo)
1792 break;
1793 mutex_unlock(&u->readlock);
1795 timeo = unix_stream_data_wait(sk, timeo);
1797 if (signal_pending(current)) {
1798 err = sock_intr_errno(timeo);
1799 goto out;
1801 mutex_lock(&u->readlock);
1802 continue;
1803 unlock:
1804 unix_state_unlock(sk);
1805 break;
1807 unix_state_unlock(sk);
1809 if (check_creds) {
1810 /* Never glue messages from different writers */
1811 if (memcmp(UNIXCREDS(skb), &siocb->scm->creds, sizeof(siocb->scm->creds)) != 0) {
1812 skb_queue_head(&sk->sk_receive_queue, skb);
1813 break;
1815 } else {
1816 /* Copy credentials */
1817 siocb->scm->creds = *UNIXCREDS(skb);
1818 check_creds = 1;
1821 /* Copy address just once */
1822 if (sunaddr)
1824 unix_copy_addr(msg, skb->sk);
1825 sunaddr = NULL;
1828 chunk = min_t(unsigned int, skb->len, size);
1829 if (memcpy_toiovec(msg->msg_iov, skb->data, chunk)) {
1830 skb_queue_head(&sk->sk_receive_queue, skb);
1831 if (copied == 0)
1832 copied = -EFAULT;
1833 break;
1835 copied += chunk;
1836 size -= chunk;
1838 /* Mark read part of skb as used */
1839 if (!(flags & MSG_PEEK))
1841 skb_pull(skb, chunk);
1843 if (UNIXCB(skb).fp)
1844 unix_detach_fds(siocb->scm, skb);
1846 /* put the skb back if we didn't use it up.. */
1847 if (skb->len)
1849 skb_queue_head(&sk->sk_receive_queue, skb);
1850 break;
1853 kfree_skb(skb);
1855 if (siocb->scm->fp)
1856 break;
1858 else
1860 /* It is questionable, see note in unix_dgram_recvmsg.
1862 if (UNIXCB(skb).fp)
1863 siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1865 /* put message back and return */
1866 skb_queue_head(&sk->sk_receive_queue, skb);
1867 break;
1869 } while (size);
1871 mutex_unlock(&u->readlock);
1872 scm_recv(sock, msg, siocb->scm, flags);
1873 out:
1874 return copied ? : err;
1877 static int unix_shutdown(struct socket *sock, int mode)
1879 struct sock *sk = sock->sk;
1880 struct sock *other;
1882 mode = (mode+1)&(RCV_SHUTDOWN|SEND_SHUTDOWN);
1884 if (mode) {
1885 unix_state_lock(sk);
1886 sk->sk_shutdown |= mode;
1887 other=unix_peer(sk);
1888 if (other)
1889 sock_hold(other);
1890 unix_state_unlock(sk);
1891 sk->sk_state_change(sk);
1893 if (other &&
1894 (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET)) {
1896 int peer_mode = 0;
1898 if (mode&RCV_SHUTDOWN)
1899 peer_mode |= SEND_SHUTDOWN;
1900 if (mode&SEND_SHUTDOWN)
1901 peer_mode |= RCV_SHUTDOWN;
1902 unix_state_lock(other);
1903 other->sk_shutdown |= peer_mode;
1904 unix_state_unlock(other);
1905 other->sk_state_change(other);
1906 read_lock(&other->sk_callback_lock);
1907 if (peer_mode == SHUTDOWN_MASK)
1908 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_HUP);
1909 else if (peer_mode & RCV_SHUTDOWN)
1910 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_IN);
1911 read_unlock(&other->sk_callback_lock);
1913 if (other)
1914 sock_put(other);
1916 return 0;
1919 static int unix_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1921 struct sock *sk = sock->sk;
1922 long amount=0;
1923 int err;
1925 switch(cmd)
1927 case SIOCOUTQ:
1928 amount = atomic_read(&sk->sk_wmem_alloc);
1929 err = put_user(amount, (int __user *)arg);
1930 break;
1931 case SIOCINQ:
1933 struct sk_buff *skb;
1935 if (sk->sk_state == TCP_LISTEN) {
1936 err = -EINVAL;
1937 break;
1940 spin_lock(&sk->sk_receive_queue.lock);
1941 if (sk->sk_type == SOCK_STREAM ||
1942 sk->sk_type == SOCK_SEQPACKET) {
1943 skb_queue_walk(&sk->sk_receive_queue, skb)
1944 amount += skb->len;
1945 } else {
1946 skb = skb_peek(&sk->sk_receive_queue);
1947 if (skb)
1948 amount=skb->len;
1950 spin_unlock(&sk->sk_receive_queue.lock);
1951 err = put_user(amount, (int __user *)arg);
1952 break;
1955 default:
1956 err = -ENOIOCTLCMD;
1957 break;
1959 return err;
1962 static unsigned int unix_poll(struct file * file, struct socket *sock, poll_table *wait)
1964 struct sock *sk = sock->sk;
1965 unsigned int mask;
1967 poll_wait(file, sk->sk_sleep, wait);
1968 mask = 0;
1970 /* exceptional events? */
1971 if (sk->sk_err)
1972 mask |= POLLERR;
1973 if (sk->sk_shutdown == SHUTDOWN_MASK)
1974 mask |= POLLHUP;
1975 if (sk->sk_shutdown & RCV_SHUTDOWN)
1976 mask |= POLLRDHUP;
1978 /* readable? */
1979 if (!skb_queue_empty(&sk->sk_receive_queue) ||
1980 (sk->sk_shutdown & RCV_SHUTDOWN))
1981 mask |= POLLIN | POLLRDNORM;
1983 /* Connection-based need to check for termination and startup */
1984 if ((sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) && sk->sk_state == TCP_CLOSE)
1985 mask |= POLLHUP;
1988 * we set writable also when the other side has shut down the
1989 * connection. This prevents stuck sockets.
1991 if (unix_writable(sk))
1992 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
1994 return mask;
1997 static unsigned int unix_dgram_poll(struct file *file, struct socket *sock,
1998 poll_table *wait)
2000 struct sock *sk = sock->sk, *other;
2001 unsigned int mask, writable;
2003 poll_wait(file, sk->sk_sleep, wait);
2004 mask = 0;
2006 /* exceptional events? */
2007 if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue))
2008 mask |= POLLERR;
2009 if (sk->sk_shutdown & RCV_SHUTDOWN)
2010 mask |= POLLRDHUP;
2011 if (sk->sk_shutdown == SHUTDOWN_MASK)
2012 mask |= POLLHUP;
2014 /* readable? */
2015 if (!skb_queue_empty(&sk->sk_receive_queue) ||
2016 (sk->sk_shutdown & RCV_SHUTDOWN))
2017 mask |= POLLIN | POLLRDNORM;
2019 /* Connection-based need to check for termination and startup */
2020 if (sk->sk_type == SOCK_SEQPACKET) {
2021 if (sk->sk_state == TCP_CLOSE)
2022 mask |= POLLHUP;
2023 /* connection hasn't started yet? */
2024 if (sk->sk_state == TCP_SYN_SENT)
2025 return mask;
2028 /* writable? */
2029 writable = unix_writable(sk);
2030 if (writable) {
2031 other = unix_peer_get(sk);
2032 if (other) {
2033 if (unix_peer(other) != sk) {
2034 poll_wait(file, &unix_sk(other)->peer_wait,
2035 wait);
2036 if (unix_recvq_full(other))
2037 writable = 0;
2040 sock_put(other);
2044 if (writable)
2045 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2046 else
2047 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
2049 return mask;
2052 #ifdef CONFIG_PROC_FS
2053 static struct sock *first_unix_socket(int *i)
2055 for (*i = 0; *i <= UNIX_HASH_SIZE; (*i)++) {
2056 if (!hlist_empty(&unix_socket_table[*i]))
2057 return __sk_head(&unix_socket_table[*i]);
2059 return NULL;
2062 static struct sock *next_unix_socket(int *i, struct sock *s)
2064 struct sock *next = sk_next(s);
2065 /* More in this chain? */
2066 if (next)
2067 return next;
2068 /* Look for next non-empty chain. */
2069 for ((*i)++; *i <= UNIX_HASH_SIZE; (*i)++) {
2070 if (!hlist_empty(&unix_socket_table[*i]))
2071 return __sk_head(&unix_socket_table[*i]);
2073 return NULL;
2076 struct unix_iter_state {
2077 struct seq_net_private p;
2078 int i;
2080 static struct sock *unix_seq_idx(struct seq_file *seq, loff_t pos)
2082 struct unix_iter_state *iter = seq->private;
2083 loff_t off = 0;
2084 struct sock *s;
2086 for (s = first_unix_socket(&iter->i); s; s = next_unix_socket(&iter->i, s)) {
2087 if (sock_net(s) != seq_file_net(seq))
2088 continue;
2089 if (off == pos)
2090 return s;
2091 ++off;
2093 return NULL;
2097 static void *unix_seq_start(struct seq_file *seq, loff_t *pos)
2098 __acquires(unix_table_lock)
2100 spin_lock(&unix_table_lock);
2101 return *pos ? unix_seq_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2104 static void *unix_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2106 struct unix_iter_state *iter = seq->private;
2107 struct sock *sk = v;
2108 ++*pos;
2110 if (v == SEQ_START_TOKEN)
2111 sk = first_unix_socket(&iter->i);
2112 else
2113 sk = next_unix_socket(&iter->i, sk);
2114 while (sk && (sock_net(sk) != seq_file_net(seq)))
2115 sk = next_unix_socket(&iter->i, sk);
2116 return sk;
2119 static void unix_seq_stop(struct seq_file *seq, void *v)
2120 __releases(unix_table_lock)
2122 spin_unlock(&unix_table_lock);
2125 static int unix_seq_show(struct seq_file *seq, void *v)
2128 if (v == SEQ_START_TOKEN)
2129 seq_puts(seq, "Num RefCount Protocol Flags Type St "
2130 "Inode Path\n");
2131 else {
2132 struct sock *s = v;
2133 struct unix_sock *u = unix_sk(s);
2134 unix_state_lock(s);
2136 seq_printf(seq, "%p: %08X %08X %08X %04X %02X %5lu",
2138 atomic_read(&s->sk_refcnt),
2140 s->sk_state == TCP_LISTEN ? __SO_ACCEPTCON : 0,
2141 s->sk_type,
2142 s->sk_socket ?
2143 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTED : SS_UNCONNECTED) :
2144 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTING : SS_DISCONNECTING),
2145 sock_i_ino(s));
2147 if (u->addr) {
2148 int i, len;
2149 seq_putc(seq, ' ');
2151 i = 0;
2152 len = u->addr->len - sizeof(short);
2153 if (!UNIX_ABSTRACT(s))
2154 len--;
2155 else {
2156 seq_putc(seq, '@');
2157 i++;
2159 for ( ; i < len; i++)
2160 seq_putc(seq, u->addr->name->sun_path[i]);
2162 unix_state_unlock(s);
2163 seq_putc(seq, '\n');
2166 return 0;
2169 static const struct seq_operations unix_seq_ops = {
2170 .start = unix_seq_start,
2171 .next = unix_seq_next,
2172 .stop = unix_seq_stop,
2173 .show = unix_seq_show,
2177 static int unix_seq_open(struct inode *inode, struct file *file)
2179 return seq_open_net(inode, file, &unix_seq_ops,
2180 sizeof(struct unix_iter_state));
2183 static const struct file_operations unix_seq_fops = {
2184 .owner = THIS_MODULE,
2185 .open = unix_seq_open,
2186 .read = seq_read,
2187 .llseek = seq_lseek,
2188 .release = seq_release_net,
2191 #endif
2193 static struct net_proto_family unix_family_ops = {
2194 .family = PF_UNIX,
2195 .create = unix_create,
2196 .owner = THIS_MODULE,
2200 static int unix_net_init(struct net *net)
2202 int error = -ENOMEM;
2204 net->unx.sysctl_max_dgram_qlen = 10;
2205 if (unix_sysctl_register(net))
2206 goto out;
2208 #ifdef CONFIG_PROC_FS
2209 if (!proc_net_fops_create(net, "unix", 0, &unix_seq_fops)) {
2210 unix_sysctl_unregister(net);
2211 goto out;
2213 #endif
2214 error = 0;
2215 out:
2216 return 0;
2219 static void unix_net_exit(struct net *net)
2221 unix_sysctl_unregister(net);
2222 proc_net_remove(net, "unix");
2225 static struct pernet_operations unix_net_ops = {
2226 .init = unix_net_init,
2227 .exit = unix_net_exit,
2230 static int __init af_unix_init(void)
2232 int rc = -1;
2233 struct sk_buff *dummy_skb;
2235 BUILD_BUG_ON(sizeof(struct unix_skb_parms) > sizeof(dummy_skb->cb));
2237 rc = proto_register(&unix_proto, 1);
2238 if (rc != 0) {
2239 printk(KERN_CRIT "%s: Cannot create unix_sock SLAB cache!\n",
2240 __func__);
2241 goto out;
2244 sock_register(&unix_family_ops);
2245 register_pernet_subsys(&unix_net_ops);
2246 out:
2247 return rc;
2250 static void __exit af_unix_exit(void)
2252 sock_unregister(PF_UNIX);
2253 proto_unregister(&unix_proto);
2254 unregister_pernet_subsys(&unix_net_ops);
2257 /* Earlier than device_initcall() so that other drivers invoking
2258 request_module() don't end up in a loop when modprobe tries
2259 to use a UNIX socket. But later than subsys_initcall() because
2260 we depend on stuff initialised there */
2261 fs_initcall(af_unix_init);
2262 module_exit(af_unix_exit);
2264 MODULE_LICENSE("GPL");
2265 MODULE_ALIAS_NETPROTO(PF_UNIX);