[UNIX]: Extend unix_sysctl_(un)register prototypes
[linux-2.6/openmoko-kernel/knife-kernel.git] / net / unix / af_unix.c
bloba0aa6d3c2c084ad713979da5dd3778a233d63a64
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 int sysctl_unix_max_dgram_qlen __read_mostly = 10;
122 static struct hlist_head unix_socket_table[UNIX_HASH_SIZE + 1];
123 static DEFINE_SPINLOCK(unix_table_lock);
124 static atomic_t unix_nr_socks = ATOMIC_INIT(0);
126 #define unix_sockets_unbound (&unix_socket_table[UNIX_HASH_SIZE])
128 #define UNIX_ABSTRACT(sk) (unix_sk(sk)->addr->hash != UNIX_HASH_SIZE)
130 #ifdef CONFIG_SECURITY_NETWORK
131 static void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb)
133 memcpy(UNIXSID(skb), &scm->secid, sizeof(u32));
136 static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb)
138 scm->secid = *UNIXSID(skb);
140 #else
141 static inline void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb)
144 static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb)
146 #endif /* CONFIG_SECURITY_NETWORK */
149 * SMP locking strategy:
150 * hash table is protected with spinlock unix_table_lock
151 * each socket state is protected by separate rwlock.
154 static inline unsigned unix_hash_fold(__wsum n)
156 unsigned hash = (__force unsigned)n;
157 hash ^= hash>>16;
158 hash ^= hash>>8;
159 return hash&(UNIX_HASH_SIZE-1);
162 #define unix_peer(sk) (unix_sk(sk)->peer)
164 static inline int unix_our_peer(struct sock *sk, struct sock *osk)
166 return unix_peer(osk) == sk;
169 static inline int unix_may_send(struct sock *sk, struct sock *osk)
171 return (unix_peer(osk) == NULL || unix_our_peer(sk, osk));
174 static struct sock *unix_peer_get(struct sock *s)
176 struct sock *peer;
178 unix_state_lock(s);
179 peer = unix_peer(s);
180 if (peer)
181 sock_hold(peer);
182 unix_state_unlock(s);
183 return peer;
186 static inline void unix_release_addr(struct unix_address *addr)
188 if (atomic_dec_and_test(&addr->refcnt))
189 kfree(addr);
193 * Check unix socket name:
194 * - should be not zero length.
195 * - if started by not zero, should be NULL terminated (FS object)
196 * - if started by zero, it is abstract name.
199 static int unix_mkname(struct sockaddr_un * sunaddr, int len, unsigned *hashp)
201 if (len <= sizeof(short) || len > sizeof(*sunaddr))
202 return -EINVAL;
203 if (!sunaddr || sunaddr->sun_family != AF_UNIX)
204 return -EINVAL;
205 if (sunaddr->sun_path[0]) {
207 * This may look like an off by one error but it is a bit more
208 * subtle. 108 is the longest valid AF_UNIX path for a binding.
209 * sun_path[108] doesnt as such exist. However in kernel space
210 * we are guaranteed that it is a valid memory location in our
211 * kernel address buffer.
213 ((char *)sunaddr)[len]=0;
214 len = strlen(sunaddr->sun_path)+1+sizeof(short);
215 return len;
218 *hashp = unix_hash_fold(csum_partial((char*)sunaddr, len, 0));
219 return len;
222 static void __unix_remove_socket(struct sock *sk)
224 sk_del_node_init(sk);
227 static void __unix_insert_socket(struct hlist_head *list, struct sock *sk)
229 BUG_TRAP(sk_unhashed(sk));
230 sk_add_node(sk, list);
233 static inline void unix_remove_socket(struct sock *sk)
235 spin_lock(&unix_table_lock);
236 __unix_remove_socket(sk);
237 spin_unlock(&unix_table_lock);
240 static inline void unix_insert_socket(struct hlist_head *list, struct sock *sk)
242 spin_lock(&unix_table_lock);
243 __unix_insert_socket(list, sk);
244 spin_unlock(&unix_table_lock);
247 static struct sock *__unix_find_socket_byname(struct net *net,
248 struct sockaddr_un *sunname,
249 int len, int type, unsigned hash)
251 struct sock *s;
252 struct hlist_node *node;
254 sk_for_each(s, node, &unix_socket_table[hash ^ type]) {
255 struct unix_sock *u = unix_sk(s);
257 if (s->sk_net != net)
258 continue;
260 if (u->addr->len == len &&
261 !memcmp(u->addr->name, sunname, len))
262 goto found;
264 s = NULL;
265 found:
266 return s;
269 static inline struct sock *unix_find_socket_byname(struct net *net,
270 struct sockaddr_un *sunname,
271 int len, int type,
272 unsigned hash)
274 struct sock *s;
276 spin_lock(&unix_table_lock);
277 s = __unix_find_socket_byname(net, sunname, len, type, hash);
278 if (s)
279 sock_hold(s);
280 spin_unlock(&unix_table_lock);
281 return s;
284 static struct sock *unix_find_socket_byinode(struct net *net, struct inode *i)
286 struct sock *s;
287 struct hlist_node *node;
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;
294 if (s->sk_net != net)
295 continue;
297 if(dentry && dentry->d_inode == i)
299 sock_hold(s);
300 goto found;
303 s = NULL;
304 found:
305 spin_unlock(&unix_table_lock);
306 return s;
309 static inline int unix_writable(struct sock *sk)
311 return (atomic_read(&sk->sk_wmem_alloc) << 2) <= sk->sk_sndbuf;
314 static void unix_write_space(struct sock *sk)
316 read_lock(&sk->sk_callback_lock);
317 if (unix_writable(sk)) {
318 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
319 wake_up_interruptible_sync(sk->sk_sleep);
320 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
322 read_unlock(&sk->sk_callback_lock);
325 /* When dgram socket disconnects (or changes its peer), we clear its receive
326 * queue of packets arrived from previous peer. First, it allows to do
327 * flow control based only on wmem_alloc; second, sk connected to peer
328 * may receive messages only from that peer. */
329 static void unix_dgram_disconnected(struct sock *sk, struct sock *other)
331 if (!skb_queue_empty(&sk->sk_receive_queue)) {
332 skb_queue_purge(&sk->sk_receive_queue);
333 wake_up_interruptible_all(&unix_sk(sk)->peer_wait);
335 /* If one link of bidirectional dgram pipe is disconnected,
336 * we signal error. Messages are lost. Do not make this,
337 * when peer was not connected to us.
339 if (!sock_flag(other, SOCK_DEAD) && unix_peer(other) == sk) {
340 other->sk_err = ECONNRESET;
341 other->sk_error_report(other);
346 static void unix_sock_destructor(struct sock *sk)
348 struct unix_sock *u = unix_sk(sk);
350 skb_queue_purge(&sk->sk_receive_queue);
352 BUG_TRAP(!atomic_read(&sk->sk_wmem_alloc));
353 BUG_TRAP(sk_unhashed(sk));
354 BUG_TRAP(!sk->sk_socket);
355 if (!sock_flag(sk, SOCK_DEAD)) {
356 printk("Attempt to release alive unix socket: %p\n", sk);
357 return;
360 if (u->addr)
361 unix_release_addr(u->addr);
363 atomic_dec(&unix_nr_socks);
364 #ifdef UNIX_REFCNT_DEBUG
365 printk(KERN_DEBUG "UNIX %p is destroyed, %d are still alive.\n", sk, atomic_read(&unix_nr_socks));
366 #endif
369 static int unix_release_sock (struct sock *sk, int embrion)
371 struct unix_sock *u = unix_sk(sk);
372 struct dentry *dentry;
373 struct vfsmount *mnt;
374 struct sock *skpair;
375 struct sk_buff *skb;
376 int state;
378 unix_remove_socket(sk);
380 /* Clear state */
381 unix_state_lock(sk);
382 sock_orphan(sk);
383 sk->sk_shutdown = SHUTDOWN_MASK;
384 dentry = u->dentry;
385 u->dentry = NULL;
386 mnt = u->mnt;
387 u->mnt = NULL;
388 state = sk->sk_state;
389 sk->sk_state = TCP_CLOSE;
390 unix_state_unlock(sk);
392 wake_up_interruptible_all(&u->peer_wait);
394 skpair=unix_peer(sk);
396 if (skpair!=NULL) {
397 if (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) {
398 unix_state_lock(skpair);
399 /* No more writes */
400 skpair->sk_shutdown = SHUTDOWN_MASK;
401 if (!skb_queue_empty(&sk->sk_receive_queue) || embrion)
402 skpair->sk_err = ECONNRESET;
403 unix_state_unlock(skpair);
404 skpair->sk_state_change(skpair);
405 read_lock(&skpair->sk_callback_lock);
406 sk_wake_async(skpair, SOCK_WAKE_WAITD, POLL_HUP);
407 read_unlock(&skpair->sk_callback_lock);
409 sock_put(skpair); /* It may now die */
410 unix_peer(sk) = NULL;
413 /* Try to flush out this socket. Throw out buffers at least */
415 while ((skb = skb_dequeue(&sk->sk_receive_queue)) != NULL) {
416 if (state==TCP_LISTEN)
417 unix_release_sock(skb->sk, 1);
418 /* passed fds are erased in the kfree_skb hook */
419 kfree_skb(skb);
422 if (dentry) {
423 dput(dentry);
424 mntput(mnt);
427 sock_put(sk);
429 /* ---- Socket is dead now and most probably destroyed ---- */
432 * Fixme: BSD difference: In BSD all sockets connected to use get
433 * ECONNRESET and we die on the spot. In Linux we behave
434 * like files and pipes do and wait for the last
435 * dereference.
437 * Can't we simply set sock->err?
439 * What the above comment does talk about? --ANK(980817)
442 if (unix_tot_inflight)
443 unix_gc(); /* Garbage collect fds */
445 return 0;
448 static int unix_listen(struct socket *sock, int backlog)
450 int err;
451 struct sock *sk = sock->sk;
452 struct unix_sock *u = unix_sk(sk);
454 err = -EOPNOTSUPP;
455 if (sock->type!=SOCK_STREAM && sock->type!=SOCK_SEQPACKET)
456 goto out; /* Only stream/seqpacket sockets accept */
457 err = -EINVAL;
458 if (!u->addr)
459 goto out; /* No listens on an unbound socket */
460 unix_state_lock(sk);
461 if (sk->sk_state != TCP_CLOSE && sk->sk_state != TCP_LISTEN)
462 goto out_unlock;
463 if (backlog > sk->sk_max_ack_backlog)
464 wake_up_interruptible_all(&u->peer_wait);
465 sk->sk_max_ack_backlog = backlog;
466 sk->sk_state = TCP_LISTEN;
467 /* set credentials so connect can copy them */
468 sk->sk_peercred.pid = task_tgid_vnr(current);
469 sk->sk_peercred.uid = current->euid;
470 sk->sk_peercred.gid = current->egid;
471 err = 0;
473 out_unlock:
474 unix_state_unlock(sk);
475 out:
476 return err;
479 static int unix_release(struct socket *);
480 static int unix_bind(struct socket *, struct sockaddr *, int);
481 static int unix_stream_connect(struct socket *, struct sockaddr *,
482 int addr_len, int flags);
483 static int unix_socketpair(struct socket *, struct socket *);
484 static int unix_accept(struct socket *, struct socket *, int);
485 static int unix_getname(struct socket *, struct sockaddr *, int *, int);
486 static unsigned int unix_poll(struct file *, struct socket *, poll_table *);
487 static int unix_ioctl(struct socket *, unsigned int, unsigned long);
488 static int unix_shutdown(struct socket *, int);
489 static int unix_stream_sendmsg(struct kiocb *, struct socket *,
490 struct msghdr *, size_t);
491 static int unix_stream_recvmsg(struct kiocb *, struct socket *,
492 struct msghdr *, size_t, int);
493 static int unix_dgram_sendmsg(struct kiocb *, struct socket *,
494 struct msghdr *, size_t);
495 static int unix_dgram_recvmsg(struct kiocb *, struct socket *,
496 struct msghdr *, size_t, int);
497 static int unix_dgram_connect(struct socket *, struct sockaddr *,
498 int, int);
499 static int unix_seqpacket_sendmsg(struct kiocb *, struct socket *,
500 struct msghdr *, size_t);
502 static const struct proto_ops unix_stream_ops = {
503 .family = PF_UNIX,
504 .owner = THIS_MODULE,
505 .release = unix_release,
506 .bind = unix_bind,
507 .connect = unix_stream_connect,
508 .socketpair = unix_socketpair,
509 .accept = unix_accept,
510 .getname = unix_getname,
511 .poll = unix_poll,
512 .ioctl = unix_ioctl,
513 .listen = unix_listen,
514 .shutdown = unix_shutdown,
515 .setsockopt = sock_no_setsockopt,
516 .getsockopt = sock_no_getsockopt,
517 .sendmsg = unix_stream_sendmsg,
518 .recvmsg = unix_stream_recvmsg,
519 .mmap = sock_no_mmap,
520 .sendpage = sock_no_sendpage,
523 static const struct proto_ops unix_dgram_ops = {
524 .family = PF_UNIX,
525 .owner = THIS_MODULE,
526 .release = unix_release,
527 .bind = unix_bind,
528 .connect = unix_dgram_connect,
529 .socketpair = unix_socketpair,
530 .accept = sock_no_accept,
531 .getname = unix_getname,
532 .poll = datagram_poll,
533 .ioctl = unix_ioctl,
534 .listen = sock_no_listen,
535 .shutdown = unix_shutdown,
536 .setsockopt = sock_no_setsockopt,
537 .getsockopt = sock_no_getsockopt,
538 .sendmsg = unix_dgram_sendmsg,
539 .recvmsg = unix_dgram_recvmsg,
540 .mmap = sock_no_mmap,
541 .sendpage = sock_no_sendpage,
544 static const struct proto_ops unix_seqpacket_ops = {
545 .family = PF_UNIX,
546 .owner = THIS_MODULE,
547 .release = unix_release,
548 .bind = unix_bind,
549 .connect = unix_stream_connect,
550 .socketpair = unix_socketpair,
551 .accept = unix_accept,
552 .getname = unix_getname,
553 .poll = datagram_poll,
554 .ioctl = unix_ioctl,
555 .listen = unix_listen,
556 .shutdown = unix_shutdown,
557 .setsockopt = sock_no_setsockopt,
558 .getsockopt = sock_no_getsockopt,
559 .sendmsg = unix_seqpacket_sendmsg,
560 .recvmsg = unix_dgram_recvmsg,
561 .mmap = sock_no_mmap,
562 .sendpage = sock_no_sendpage,
565 static struct proto unix_proto = {
566 .name = "UNIX",
567 .owner = THIS_MODULE,
568 .obj_size = sizeof(struct unix_sock),
572 * AF_UNIX sockets do not interact with hardware, hence they
573 * dont trigger interrupts - so it's safe for them to have
574 * bh-unsafe locking for their sk_receive_queue.lock. Split off
575 * this special lock-class by reinitializing the spinlock key:
577 static struct lock_class_key af_unix_sk_receive_queue_lock_key;
579 static struct sock * unix_create1(struct net *net, struct socket *sock)
581 struct sock *sk = NULL;
582 struct unix_sock *u;
584 atomic_inc(&unix_nr_socks);
585 if (atomic_read(&unix_nr_socks) > 2 * get_max_files())
586 goto out;
588 sk = sk_alloc(net, PF_UNIX, GFP_KERNEL, &unix_proto);
589 if (!sk)
590 goto out;
592 sock_init_data(sock,sk);
593 lockdep_set_class(&sk->sk_receive_queue.lock,
594 &af_unix_sk_receive_queue_lock_key);
596 sk->sk_write_space = unix_write_space;
597 sk->sk_max_ack_backlog = sysctl_unix_max_dgram_qlen;
598 sk->sk_destruct = unix_sock_destructor;
599 u = unix_sk(sk);
600 u->dentry = NULL;
601 u->mnt = NULL;
602 spin_lock_init(&u->lock);
603 atomic_set(&u->inflight, 0);
604 INIT_LIST_HEAD(&u->link);
605 mutex_init(&u->readlock); /* single task reading lock */
606 init_waitqueue_head(&u->peer_wait);
607 unix_insert_socket(unix_sockets_unbound, sk);
608 out:
609 if (sk == NULL)
610 atomic_dec(&unix_nr_socks);
611 return sk;
614 static int unix_create(struct net *net, struct socket *sock, int protocol)
616 if (protocol && protocol != PF_UNIX)
617 return -EPROTONOSUPPORT;
619 sock->state = SS_UNCONNECTED;
621 switch (sock->type) {
622 case SOCK_STREAM:
623 sock->ops = &unix_stream_ops;
624 break;
626 * Believe it or not BSD has AF_UNIX, SOCK_RAW though
627 * nothing uses it.
629 case SOCK_RAW:
630 sock->type=SOCK_DGRAM;
631 case SOCK_DGRAM:
632 sock->ops = &unix_dgram_ops;
633 break;
634 case SOCK_SEQPACKET:
635 sock->ops = &unix_seqpacket_ops;
636 break;
637 default:
638 return -ESOCKTNOSUPPORT;
641 return unix_create1(net, sock) ? 0 : -ENOMEM;
644 static int unix_release(struct socket *sock)
646 struct sock *sk = sock->sk;
648 if (!sk)
649 return 0;
651 sock->sk = NULL;
653 return unix_release_sock (sk, 0);
656 static int unix_autobind(struct socket *sock)
658 struct sock *sk = sock->sk;
659 struct net *net = sk->sk_net;
660 struct unix_sock *u = unix_sk(sk);
661 static u32 ordernum = 1;
662 struct unix_address * addr;
663 int err;
665 mutex_lock(&u->readlock);
667 err = 0;
668 if (u->addr)
669 goto out;
671 err = -ENOMEM;
672 addr = kzalloc(sizeof(*addr) + sizeof(short) + 16, GFP_KERNEL);
673 if (!addr)
674 goto out;
676 addr->name->sun_family = AF_UNIX;
677 atomic_set(&addr->refcnt, 1);
679 retry:
680 addr->len = sprintf(addr->name->sun_path+1, "%05x", ordernum) + 1 + sizeof(short);
681 addr->hash = unix_hash_fold(csum_partial((void*)addr->name, addr->len, 0));
683 spin_lock(&unix_table_lock);
684 ordernum = (ordernum+1)&0xFFFFF;
686 if (__unix_find_socket_byname(net, addr->name, addr->len, sock->type,
687 addr->hash)) {
688 spin_unlock(&unix_table_lock);
689 /* Sanity yield. It is unusual case, but yet... */
690 if (!(ordernum&0xFF))
691 yield();
692 goto retry;
694 addr->hash ^= sk->sk_type;
696 __unix_remove_socket(sk);
697 u->addr = addr;
698 __unix_insert_socket(&unix_socket_table[addr->hash], sk);
699 spin_unlock(&unix_table_lock);
700 err = 0;
702 out: mutex_unlock(&u->readlock);
703 return err;
706 static struct sock *unix_find_other(struct net *net,
707 struct sockaddr_un *sunname, int len,
708 int type, unsigned hash, int *error)
710 struct sock *u;
711 struct nameidata nd;
712 int err = 0;
714 if (sunname->sun_path[0]) {
715 err = path_lookup(sunname->sun_path, LOOKUP_FOLLOW, &nd);
716 if (err)
717 goto fail;
718 err = vfs_permission(&nd, MAY_WRITE);
719 if (err)
720 goto put_fail;
722 err = -ECONNREFUSED;
723 if (!S_ISSOCK(nd.dentry->d_inode->i_mode))
724 goto put_fail;
725 u=unix_find_socket_byinode(net, nd.dentry->d_inode);
726 if (!u)
727 goto put_fail;
729 if (u->sk_type == type)
730 touch_atime(nd.mnt, nd.dentry);
732 path_release(&nd);
734 err=-EPROTOTYPE;
735 if (u->sk_type != type) {
736 sock_put(u);
737 goto fail;
739 } else {
740 err = -ECONNREFUSED;
741 u=unix_find_socket_byname(net, sunname, len, type, hash);
742 if (u) {
743 struct dentry *dentry;
744 dentry = unix_sk(u)->dentry;
745 if (dentry)
746 touch_atime(unix_sk(u)->mnt, dentry);
747 } else
748 goto fail;
750 return u;
752 put_fail:
753 path_release(&nd);
754 fail:
755 *error=err;
756 return NULL;
760 static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
762 struct sock *sk = sock->sk;
763 struct net *net = sk->sk_net;
764 struct unix_sock *u = unix_sk(sk);
765 struct sockaddr_un *sunaddr=(struct sockaddr_un *)uaddr;
766 struct dentry * dentry = NULL;
767 struct nameidata nd;
768 int err;
769 unsigned hash;
770 struct unix_address *addr;
771 struct hlist_head *list;
773 err = -EINVAL;
774 if (sunaddr->sun_family != AF_UNIX)
775 goto out;
777 if (addr_len==sizeof(short)) {
778 err = unix_autobind(sock);
779 goto out;
782 err = unix_mkname(sunaddr, addr_len, &hash);
783 if (err < 0)
784 goto out;
785 addr_len = err;
787 mutex_lock(&u->readlock);
789 err = -EINVAL;
790 if (u->addr)
791 goto out_up;
793 err = -ENOMEM;
794 addr = kmalloc(sizeof(*addr)+addr_len, GFP_KERNEL);
795 if (!addr)
796 goto out_up;
798 memcpy(addr->name, sunaddr, addr_len);
799 addr->len = addr_len;
800 addr->hash = hash ^ sk->sk_type;
801 atomic_set(&addr->refcnt, 1);
803 if (sunaddr->sun_path[0]) {
804 unsigned int mode;
805 err = 0;
807 * Get the parent directory, calculate the hash for last
808 * component.
810 err = path_lookup(sunaddr->sun_path, LOOKUP_PARENT, &nd);
811 if (err)
812 goto out_mknod_parent;
814 dentry = lookup_create(&nd, 0);
815 err = PTR_ERR(dentry);
816 if (IS_ERR(dentry))
817 goto out_mknod_unlock;
820 * All right, let's create it.
822 mode = S_IFSOCK |
823 (SOCK_INODE(sock)->i_mode & ~current->fs->umask);
824 err = vfs_mknod(nd.dentry->d_inode, dentry, mode, 0);
825 if (err)
826 goto out_mknod_dput;
827 mutex_unlock(&nd.dentry->d_inode->i_mutex);
828 dput(nd.dentry);
829 nd.dentry = dentry;
831 addr->hash = UNIX_HASH_SIZE;
834 spin_lock(&unix_table_lock);
836 if (!sunaddr->sun_path[0]) {
837 err = -EADDRINUSE;
838 if (__unix_find_socket_byname(net, sunaddr, addr_len,
839 sk->sk_type, hash)) {
840 unix_release_addr(addr);
841 goto out_unlock;
844 list = &unix_socket_table[addr->hash];
845 } else {
846 list = &unix_socket_table[dentry->d_inode->i_ino & (UNIX_HASH_SIZE-1)];
847 u->dentry = nd.dentry;
848 u->mnt = nd.mnt;
851 err = 0;
852 __unix_remove_socket(sk);
853 u->addr = addr;
854 __unix_insert_socket(list, sk);
856 out_unlock:
857 spin_unlock(&unix_table_lock);
858 out_up:
859 mutex_unlock(&u->readlock);
860 out:
861 return err;
863 out_mknod_dput:
864 dput(dentry);
865 out_mknod_unlock:
866 mutex_unlock(&nd.dentry->d_inode->i_mutex);
867 path_release(&nd);
868 out_mknod_parent:
869 if (err==-EEXIST)
870 err=-EADDRINUSE;
871 unix_release_addr(addr);
872 goto out_up;
875 static void unix_state_double_lock(struct sock *sk1, struct sock *sk2)
877 if (unlikely(sk1 == sk2) || !sk2) {
878 unix_state_lock(sk1);
879 return;
881 if (sk1 < sk2) {
882 unix_state_lock(sk1);
883 unix_state_lock_nested(sk2);
884 } else {
885 unix_state_lock(sk2);
886 unix_state_lock_nested(sk1);
890 static void unix_state_double_unlock(struct sock *sk1, struct sock *sk2)
892 if (unlikely(sk1 == sk2) || !sk2) {
893 unix_state_unlock(sk1);
894 return;
896 unix_state_unlock(sk1);
897 unix_state_unlock(sk2);
900 static int unix_dgram_connect(struct socket *sock, struct sockaddr *addr,
901 int alen, int flags)
903 struct sock *sk = sock->sk;
904 struct net *net = sk->sk_net;
905 struct sockaddr_un *sunaddr=(struct sockaddr_un*)addr;
906 struct sock *other;
907 unsigned hash;
908 int err;
910 if (addr->sa_family != AF_UNSPEC) {
911 err = unix_mkname(sunaddr, alen, &hash);
912 if (err < 0)
913 goto out;
914 alen = err;
916 if (test_bit(SOCK_PASSCRED, &sock->flags) &&
917 !unix_sk(sk)->addr && (err = unix_autobind(sock)) != 0)
918 goto out;
920 restart:
921 other=unix_find_other(net, sunaddr, alen, sock->type, hash, &err);
922 if (!other)
923 goto out;
925 unix_state_double_lock(sk, other);
927 /* Apparently VFS overslept socket death. Retry. */
928 if (sock_flag(other, SOCK_DEAD)) {
929 unix_state_double_unlock(sk, other);
930 sock_put(other);
931 goto restart;
934 err = -EPERM;
935 if (!unix_may_send(sk, other))
936 goto out_unlock;
938 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
939 if (err)
940 goto out_unlock;
942 } else {
944 * 1003.1g breaking connected state with AF_UNSPEC
946 other = NULL;
947 unix_state_double_lock(sk, other);
951 * If it was connected, reconnect.
953 if (unix_peer(sk)) {
954 struct sock *old_peer = unix_peer(sk);
955 unix_peer(sk)=other;
956 unix_state_double_unlock(sk, other);
958 if (other != old_peer)
959 unix_dgram_disconnected(sk, old_peer);
960 sock_put(old_peer);
961 } else {
962 unix_peer(sk)=other;
963 unix_state_double_unlock(sk, other);
965 return 0;
967 out_unlock:
968 unix_state_double_unlock(sk, other);
969 sock_put(other);
970 out:
971 return err;
974 static long unix_wait_for_peer(struct sock *other, long timeo)
976 struct unix_sock *u = unix_sk(other);
977 int sched;
978 DEFINE_WAIT(wait);
980 prepare_to_wait_exclusive(&u->peer_wait, &wait, TASK_INTERRUPTIBLE);
982 sched = !sock_flag(other, SOCK_DEAD) &&
983 !(other->sk_shutdown & RCV_SHUTDOWN) &&
984 (skb_queue_len(&other->sk_receive_queue) >
985 other->sk_max_ack_backlog);
987 unix_state_unlock(other);
989 if (sched)
990 timeo = schedule_timeout(timeo);
992 finish_wait(&u->peer_wait, &wait);
993 return timeo;
996 static int unix_stream_connect(struct socket *sock, struct sockaddr *uaddr,
997 int addr_len, int flags)
999 struct sockaddr_un *sunaddr=(struct sockaddr_un *)uaddr;
1000 struct sock *sk = sock->sk;
1001 struct net *net = sk->sk_net;
1002 struct unix_sock *u = unix_sk(sk), *newu, *otheru;
1003 struct sock *newsk = NULL;
1004 struct sock *other = NULL;
1005 struct sk_buff *skb = NULL;
1006 unsigned hash;
1007 int st;
1008 int err;
1009 long timeo;
1011 err = unix_mkname(sunaddr, addr_len, &hash);
1012 if (err < 0)
1013 goto out;
1014 addr_len = err;
1016 if (test_bit(SOCK_PASSCRED, &sock->flags)
1017 && !u->addr && (err = unix_autobind(sock)) != 0)
1018 goto out;
1020 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
1022 /* First of all allocate resources.
1023 If we will make it after state is locked,
1024 we will have to recheck all again in any case.
1027 err = -ENOMEM;
1029 /* create new sock for complete connection */
1030 newsk = unix_create1(sk->sk_net, NULL);
1031 if (newsk == NULL)
1032 goto out;
1034 /* Allocate skb for sending to listening sock */
1035 skb = sock_wmalloc(newsk, 1, 0, GFP_KERNEL);
1036 if (skb == NULL)
1037 goto out;
1039 restart:
1040 /* Find listening sock. */
1041 other = unix_find_other(net, sunaddr, addr_len, sk->sk_type, hash, &err);
1042 if (!other)
1043 goto out;
1045 /* Latch state of peer */
1046 unix_state_lock(other);
1048 /* Apparently VFS overslept socket death. Retry. */
1049 if (sock_flag(other, SOCK_DEAD)) {
1050 unix_state_unlock(other);
1051 sock_put(other);
1052 goto restart;
1055 err = -ECONNREFUSED;
1056 if (other->sk_state != TCP_LISTEN)
1057 goto out_unlock;
1059 if (skb_queue_len(&other->sk_receive_queue) >
1060 other->sk_max_ack_backlog) {
1061 err = -EAGAIN;
1062 if (!timeo)
1063 goto out_unlock;
1065 timeo = unix_wait_for_peer(other, timeo);
1067 err = sock_intr_errno(timeo);
1068 if (signal_pending(current))
1069 goto out;
1070 sock_put(other);
1071 goto restart;
1074 /* Latch our state.
1076 It is tricky place. We need to grab write lock and cannot
1077 drop lock on peer. It is dangerous because deadlock is
1078 possible. Connect to self case and simultaneous
1079 attempt to connect are eliminated by checking socket
1080 state. other is TCP_LISTEN, if sk is TCP_LISTEN we
1081 check this before attempt to grab lock.
1083 Well, and we have to recheck the state after socket locked.
1085 st = sk->sk_state;
1087 switch (st) {
1088 case TCP_CLOSE:
1089 /* This is ok... continue with connect */
1090 break;
1091 case TCP_ESTABLISHED:
1092 /* Socket is already connected */
1093 err = -EISCONN;
1094 goto out_unlock;
1095 default:
1096 err = -EINVAL;
1097 goto out_unlock;
1100 unix_state_lock_nested(sk);
1102 if (sk->sk_state != st) {
1103 unix_state_unlock(sk);
1104 unix_state_unlock(other);
1105 sock_put(other);
1106 goto restart;
1109 err = security_unix_stream_connect(sock, other->sk_socket, newsk);
1110 if (err) {
1111 unix_state_unlock(sk);
1112 goto out_unlock;
1115 /* The way is open! Fastly set all the necessary fields... */
1117 sock_hold(sk);
1118 unix_peer(newsk) = sk;
1119 newsk->sk_state = TCP_ESTABLISHED;
1120 newsk->sk_type = sk->sk_type;
1121 newsk->sk_peercred.pid = task_tgid_vnr(current);
1122 newsk->sk_peercred.uid = current->euid;
1123 newsk->sk_peercred.gid = current->egid;
1124 newu = unix_sk(newsk);
1125 newsk->sk_sleep = &newu->peer_wait;
1126 otheru = unix_sk(other);
1128 /* copy address information from listening to new sock*/
1129 if (otheru->addr) {
1130 atomic_inc(&otheru->addr->refcnt);
1131 newu->addr = otheru->addr;
1133 if (otheru->dentry) {
1134 newu->dentry = dget(otheru->dentry);
1135 newu->mnt = mntget(otheru->mnt);
1138 /* Set credentials */
1139 sk->sk_peercred = other->sk_peercred;
1141 sock->state = SS_CONNECTED;
1142 sk->sk_state = TCP_ESTABLISHED;
1143 sock_hold(newsk);
1145 smp_mb__after_atomic_inc(); /* sock_hold() does an atomic_inc() */
1146 unix_peer(sk) = newsk;
1148 unix_state_unlock(sk);
1150 /* take ten and and send info to listening sock */
1151 spin_lock(&other->sk_receive_queue.lock);
1152 __skb_queue_tail(&other->sk_receive_queue, skb);
1153 spin_unlock(&other->sk_receive_queue.lock);
1154 unix_state_unlock(other);
1155 other->sk_data_ready(other, 0);
1156 sock_put(other);
1157 return 0;
1159 out_unlock:
1160 if (other)
1161 unix_state_unlock(other);
1163 out:
1164 if (skb)
1165 kfree_skb(skb);
1166 if (newsk)
1167 unix_release_sock(newsk, 0);
1168 if (other)
1169 sock_put(other);
1170 return err;
1173 static int unix_socketpair(struct socket *socka, struct socket *sockb)
1175 struct sock *ska=socka->sk, *skb = sockb->sk;
1177 /* Join our sockets back to back */
1178 sock_hold(ska);
1179 sock_hold(skb);
1180 unix_peer(ska)=skb;
1181 unix_peer(skb)=ska;
1182 ska->sk_peercred.pid = skb->sk_peercred.pid = task_tgid_vnr(current);
1183 ska->sk_peercred.uid = skb->sk_peercred.uid = current->euid;
1184 ska->sk_peercred.gid = skb->sk_peercred.gid = current->egid;
1186 if (ska->sk_type != SOCK_DGRAM) {
1187 ska->sk_state = TCP_ESTABLISHED;
1188 skb->sk_state = TCP_ESTABLISHED;
1189 socka->state = SS_CONNECTED;
1190 sockb->state = SS_CONNECTED;
1192 return 0;
1195 static int unix_accept(struct socket *sock, struct socket *newsock, int flags)
1197 struct sock *sk = sock->sk;
1198 struct sock *tsk;
1199 struct sk_buff *skb;
1200 int err;
1202 err = -EOPNOTSUPP;
1203 if (sock->type!=SOCK_STREAM && sock->type!=SOCK_SEQPACKET)
1204 goto out;
1206 err = -EINVAL;
1207 if (sk->sk_state != TCP_LISTEN)
1208 goto out;
1210 /* If socket state is TCP_LISTEN it cannot change (for now...),
1211 * so that no locks are necessary.
1214 skb = skb_recv_datagram(sk, 0, flags&O_NONBLOCK, &err);
1215 if (!skb) {
1216 /* This means receive shutdown. */
1217 if (err == 0)
1218 err = -EINVAL;
1219 goto out;
1222 tsk = skb->sk;
1223 skb_free_datagram(sk, skb);
1224 wake_up_interruptible(&unix_sk(sk)->peer_wait);
1226 /* attach accepted sock to socket */
1227 unix_state_lock(tsk);
1228 newsock->state = SS_CONNECTED;
1229 sock_graft(tsk, newsock);
1230 unix_state_unlock(tsk);
1231 return 0;
1233 out:
1234 return err;
1238 static int unix_getname(struct socket *sock, struct sockaddr *uaddr, int *uaddr_len, int peer)
1240 struct sock *sk = sock->sk;
1241 struct unix_sock *u;
1242 struct sockaddr_un *sunaddr=(struct sockaddr_un *)uaddr;
1243 int err = 0;
1245 if (peer) {
1246 sk = unix_peer_get(sk);
1248 err = -ENOTCONN;
1249 if (!sk)
1250 goto out;
1251 err = 0;
1252 } else {
1253 sock_hold(sk);
1256 u = unix_sk(sk);
1257 unix_state_lock(sk);
1258 if (!u->addr) {
1259 sunaddr->sun_family = AF_UNIX;
1260 sunaddr->sun_path[0] = 0;
1261 *uaddr_len = sizeof(short);
1262 } else {
1263 struct unix_address *addr = u->addr;
1265 *uaddr_len = addr->len;
1266 memcpy(sunaddr, addr->name, *uaddr_len);
1268 unix_state_unlock(sk);
1269 sock_put(sk);
1270 out:
1271 return err;
1274 static void unix_detach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1276 int i;
1278 scm->fp = UNIXCB(skb).fp;
1279 skb->destructor = sock_wfree;
1280 UNIXCB(skb).fp = NULL;
1282 for (i=scm->fp->count-1; i>=0; i--)
1283 unix_notinflight(scm->fp->fp[i]);
1286 static void unix_destruct_fds(struct sk_buff *skb)
1288 struct scm_cookie scm;
1289 memset(&scm, 0, sizeof(scm));
1290 unix_detach_fds(&scm, skb);
1292 /* Alas, it calls VFS */
1293 /* So fscking what? fput() had been SMP-safe since the last Summer */
1294 scm_destroy(&scm);
1295 sock_wfree(skb);
1298 static void unix_attach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1300 int i;
1301 for (i=scm->fp->count-1; i>=0; i--)
1302 unix_inflight(scm->fp->fp[i]);
1303 UNIXCB(skb).fp = scm->fp;
1304 skb->destructor = unix_destruct_fds;
1305 scm->fp = NULL;
1309 * Send AF_UNIX data.
1312 static int unix_dgram_sendmsg(struct kiocb *kiocb, struct socket *sock,
1313 struct msghdr *msg, size_t len)
1315 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1316 struct sock *sk = sock->sk;
1317 struct net *net = sk->sk_net;
1318 struct unix_sock *u = unix_sk(sk);
1319 struct sockaddr_un *sunaddr=msg->msg_name;
1320 struct sock *other = NULL;
1321 int namelen = 0; /* fake GCC */
1322 int err;
1323 unsigned hash;
1324 struct sk_buff *skb;
1325 long timeo;
1326 struct scm_cookie tmp_scm;
1328 if (NULL == siocb->scm)
1329 siocb->scm = &tmp_scm;
1330 err = scm_send(sock, msg, siocb->scm);
1331 if (err < 0)
1332 return err;
1334 err = -EOPNOTSUPP;
1335 if (msg->msg_flags&MSG_OOB)
1336 goto out;
1338 if (msg->msg_namelen) {
1339 err = unix_mkname(sunaddr, msg->msg_namelen, &hash);
1340 if (err < 0)
1341 goto out;
1342 namelen = err;
1343 } else {
1344 sunaddr = NULL;
1345 err = -ENOTCONN;
1346 other = unix_peer_get(sk);
1347 if (!other)
1348 goto out;
1351 if (test_bit(SOCK_PASSCRED, &sock->flags)
1352 && !u->addr && (err = unix_autobind(sock)) != 0)
1353 goto out;
1355 err = -EMSGSIZE;
1356 if (len > sk->sk_sndbuf - 32)
1357 goto out;
1359 skb = sock_alloc_send_skb(sk, len, msg->msg_flags&MSG_DONTWAIT, &err);
1360 if (skb==NULL)
1361 goto out;
1363 memcpy(UNIXCREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1364 if (siocb->scm->fp)
1365 unix_attach_fds(siocb->scm, skb);
1366 unix_get_secdata(siocb->scm, skb);
1368 skb_reset_transport_header(skb);
1369 err = memcpy_fromiovec(skb_put(skb,len), msg->msg_iov, len);
1370 if (err)
1371 goto out_free;
1373 timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1375 restart:
1376 if (!other) {
1377 err = -ECONNRESET;
1378 if (sunaddr == NULL)
1379 goto out_free;
1381 other = unix_find_other(net, sunaddr, namelen, sk->sk_type,
1382 hash, &err);
1383 if (other==NULL)
1384 goto out_free;
1387 unix_state_lock(other);
1388 err = -EPERM;
1389 if (!unix_may_send(sk, other))
1390 goto out_unlock;
1392 if (sock_flag(other, SOCK_DEAD)) {
1394 * Check with 1003.1g - what should
1395 * datagram error
1397 unix_state_unlock(other);
1398 sock_put(other);
1400 err = 0;
1401 unix_state_lock(sk);
1402 if (unix_peer(sk) == other) {
1403 unix_peer(sk)=NULL;
1404 unix_state_unlock(sk);
1406 unix_dgram_disconnected(sk, other);
1407 sock_put(other);
1408 err = -ECONNREFUSED;
1409 } else {
1410 unix_state_unlock(sk);
1413 other = NULL;
1414 if (err)
1415 goto out_free;
1416 goto restart;
1419 err = -EPIPE;
1420 if (other->sk_shutdown & RCV_SHUTDOWN)
1421 goto out_unlock;
1423 if (sk->sk_type != SOCK_SEQPACKET) {
1424 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1425 if (err)
1426 goto out_unlock;
1429 if (unix_peer(other) != sk &&
1430 (skb_queue_len(&other->sk_receive_queue) >
1431 other->sk_max_ack_backlog)) {
1432 if (!timeo) {
1433 err = -EAGAIN;
1434 goto out_unlock;
1437 timeo = unix_wait_for_peer(other, timeo);
1439 err = sock_intr_errno(timeo);
1440 if (signal_pending(current))
1441 goto out_free;
1443 goto restart;
1446 skb_queue_tail(&other->sk_receive_queue, skb);
1447 unix_state_unlock(other);
1448 other->sk_data_ready(other, len);
1449 sock_put(other);
1450 scm_destroy(siocb->scm);
1451 return len;
1453 out_unlock:
1454 unix_state_unlock(other);
1455 out_free:
1456 kfree_skb(skb);
1457 out:
1458 if (other)
1459 sock_put(other);
1460 scm_destroy(siocb->scm);
1461 return err;
1465 static int unix_stream_sendmsg(struct kiocb *kiocb, struct socket *sock,
1466 struct msghdr *msg, size_t len)
1468 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1469 struct sock *sk = sock->sk;
1470 struct sock *other = NULL;
1471 struct sockaddr_un *sunaddr=msg->msg_name;
1472 int err,size;
1473 struct sk_buff *skb;
1474 int sent=0;
1475 struct scm_cookie tmp_scm;
1477 if (NULL == siocb->scm)
1478 siocb->scm = &tmp_scm;
1479 err = scm_send(sock, msg, siocb->scm);
1480 if (err < 0)
1481 return err;
1483 err = -EOPNOTSUPP;
1484 if (msg->msg_flags&MSG_OOB)
1485 goto out_err;
1487 if (msg->msg_namelen) {
1488 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
1489 goto out_err;
1490 } else {
1491 sunaddr = NULL;
1492 err = -ENOTCONN;
1493 other = unix_peer(sk);
1494 if (!other)
1495 goto out_err;
1498 if (sk->sk_shutdown & SEND_SHUTDOWN)
1499 goto pipe_err;
1501 while(sent < len)
1504 * Optimisation for the fact that under 0.01% of X
1505 * messages typically need breaking up.
1508 size = len-sent;
1510 /* Keep two messages in the pipe so it schedules better */
1511 if (size > ((sk->sk_sndbuf >> 1) - 64))
1512 size = (sk->sk_sndbuf >> 1) - 64;
1514 if (size > SKB_MAX_ALLOC)
1515 size = SKB_MAX_ALLOC;
1518 * Grab a buffer
1521 skb=sock_alloc_send_skb(sk,size,msg->msg_flags&MSG_DONTWAIT, &err);
1523 if (skb==NULL)
1524 goto out_err;
1527 * If you pass two values to the sock_alloc_send_skb
1528 * it tries to grab the large buffer with GFP_NOFS
1529 * (which can fail easily), and if it fails grab the
1530 * fallback size buffer which is under a page and will
1531 * succeed. [Alan]
1533 size = min_t(int, size, skb_tailroom(skb));
1535 memcpy(UNIXCREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1536 if (siocb->scm->fp)
1537 unix_attach_fds(siocb->scm, skb);
1539 if ((err = memcpy_fromiovec(skb_put(skb,size), msg->msg_iov, size)) != 0) {
1540 kfree_skb(skb);
1541 goto out_err;
1544 unix_state_lock(other);
1546 if (sock_flag(other, SOCK_DEAD) ||
1547 (other->sk_shutdown & RCV_SHUTDOWN))
1548 goto pipe_err_free;
1550 skb_queue_tail(&other->sk_receive_queue, skb);
1551 unix_state_unlock(other);
1552 other->sk_data_ready(other, size);
1553 sent+=size;
1556 scm_destroy(siocb->scm);
1557 siocb->scm = NULL;
1559 return sent;
1561 pipe_err_free:
1562 unix_state_unlock(other);
1563 kfree_skb(skb);
1564 pipe_err:
1565 if (sent==0 && !(msg->msg_flags&MSG_NOSIGNAL))
1566 send_sig(SIGPIPE,current,0);
1567 err = -EPIPE;
1568 out_err:
1569 scm_destroy(siocb->scm);
1570 siocb->scm = NULL;
1571 return sent ? : err;
1574 static int unix_seqpacket_sendmsg(struct kiocb *kiocb, struct socket *sock,
1575 struct msghdr *msg, size_t len)
1577 int err;
1578 struct sock *sk = sock->sk;
1580 err = sock_error(sk);
1581 if (err)
1582 return err;
1584 if (sk->sk_state != TCP_ESTABLISHED)
1585 return -ENOTCONN;
1587 if (msg->msg_namelen)
1588 msg->msg_namelen = 0;
1590 return unix_dgram_sendmsg(kiocb, sock, msg, len);
1593 static void unix_copy_addr(struct msghdr *msg, struct sock *sk)
1595 struct unix_sock *u = unix_sk(sk);
1597 msg->msg_namelen = 0;
1598 if (u->addr) {
1599 msg->msg_namelen = u->addr->len;
1600 memcpy(msg->msg_name, u->addr->name, u->addr->len);
1604 static int unix_dgram_recvmsg(struct kiocb *iocb, struct socket *sock,
1605 struct msghdr *msg, size_t size,
1606 int flags)
1608 struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1609 struct scm_cookie tmp_scm;
1610 struct sock *sk = sock->sk;
1611 struct unix_sock *u = unix_sk(sk);
1612 int noblock = flags & MSG_DONTWAIT;
1613 struct sk_buff *skb;
1614 int err;
1616 err = -EOPNOTSUPP;
1617 if (flags&MSG_OOB)
1618 goto out;
1620 msg->msg_namelen = 0;
1622 mutex_lock(&u->readlock);
1624 skb = skb_recv_datagram(sk, flags, noblock, &err);
1625 if (!skb) {
1626 unix_state_lock(sk);
1627 /* Signal EOF on disconnected non-blocking SEQPACKET socket. */
1628 if (sk->sk_type == SOCK_SEQPACKET && err == -EAGAIN &&
1629 (sk->sk_shutdown & RCV_SHUTDOWN))
1630 err = 0;
1631 unix_state_unlock(sk);
1632 goto out_unlock;
1635 wake_up_interruptible_sync(&u->peer_wait);
1637 if (msg->msg_name)
1638 unix_copy_addr(msg, skb->sk);
1640 if (size > skb->len)
1641 size = skb->len;
1642 else if (size < skb->len)
1643 msg->msg_flags |= MSG_TRUNC;
1645 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, size);
1646 if (err)
1647 goto out_free;
1649 if (!siocb->scm) {
1650 siocb->scm = &tmp_scm;
1651 memset(&tmp_scm, 0, sizeof(tmp_scm));
1653 siocb->scm->creds = *UNIXCREDS(skb);
1654 unix_set_secdata(siocb->scm, skb);
1656 if (!(flags & MSG_PEEK))
1658 if (UNIXCB(skb).fp)
1659 unix_detach_fds(siocb->scm, skb);
1661 else
1663 /* It is questionable: on PEEK we could:
1664 - do not return fds - good, but too simple 8)
1665 - return fds, and do not return them on read (old strategy,
1666 apparently wrong)
1667 - clone fds (I chose it for now, it is the most universal
1668 solution)
1670 POSIX 1003.1g does not actually define this clearly
1671 at all. POSIX 1003.1g doesn't define a lot of things
1672 clearly however!
1675 if (UNIXCB(skb).fp)
1676 siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1678 err = size;
1680 scm_recv(sock, msg, siocb->scm, flags);
1682 out_free:
1683 skb_free_datagram(sk,skb);
1684 out_unlock:
1685 mutex_unlock(&u->readlock);
1686 out:
1687 return err;
1691 * Sleep until data has arrive. But check for races..
1694 static long unix_stream_data_wait(struct sock * sk, long timeo)
1696 DEFINE_WAIT(wait);
1698 unix_state_lock(sk);
1700 for (;;) {
1701 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1703 if (!skb_queue_empty(&sk->sk_receive_queue) ||
1704 sk->sk_err ||
1705 (sk->sk_shutdown & RCV_SHUTDOWN) ||
1706 signal_pending(current) ||
1707 !timeo)
1708 break;
1710 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1711 unix_state_unlock(sk);
1712 timeo = schedule_timeout(timeo);
1713 unix_state_lock(sk);
1714 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1717 finish_wait(sk->sk_sleep, &wait);
1718 unix_state_unlock(sk);
1719 return timeo;
1724 static int unix_stream_recvmsg(struct kiocb *iocb, struct socket *sock,
1725 struct msghdr *msg, size_t size,
1726 int flags)
1728 struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1729 struct scm_cookie tmp_scm;
1730 struct sock *sk = sock->sk;
1731 struct unix_sock *u = unix_sk(sk);
1732 struct sockaddr_un *sunaddr=msg->msg_name;
1733 int copied = 0;
1734 int check_creds = 0;
1735 int target;
1736 int err = 0;
1737 long timeo;
1739 err = -EINVAL;
1740 if (sk->sk_state != TCP_ESTABLISHED)
1741 goto out;
1743 err = -EOPNOTSUPP;
1744 if (flags&MSG_OOB)
1745 goto out;
1747 target = sock_rcvlowat(sk, flags&MSG_WAITALL, size);
1748 timeo = sock_rcvtimeo(sk, flags&MSG_DONTWAIT);
1750 msg->msg_namelen = 0;
1752 /* Lock the socket to prevent queue disordering
1753 * while sleeps in memcpy_tomsg
1756 if (!siocb->scm) {
1757 siocb->scm = &tmp_scm;
1758 memset(&tmp_scm, 0, sizeof(tmp_scm));
1761 mutex_lock(&u->readlock);
1765 int chunk;
1766 struct sk_buff *skb;
1768 unix_state_lock(sk);
1769 skb = skb_dequeue(&sk->sk_receive_queue);
1770 if (skb==NULL)
1772 if (copied >= target)
1773 goto unlock;
1776 * POSIX 1003.1g mandates this order.
1779 if ((err = sock_error(sk)) != 0)
1780 goto unlock;
1781 if (sk->sk_shutdown & RCV_SHUTDOWN)
1782 goto unlock;
1784 unix_state_unlock(sk);
1785 err = -EAGAIN;
1786 if (!timeo)
1787 break;
1788 mutex_unlock(&u->readlock);
1790 timeo = unix_stream_data_wait(sk, timeo);
1792 if (signal_pending(current)) {
1793 err = sock_intr_errno(timeo);
1794 goto out;
1796 mutex_lock(&u->readlock);
1797 continue;
1798 unlock:
1799 unix_state_unlock(sk);
1800 break;
1802 unix_state_unlock(sk);
1804 if (check_creds) {
1805 /* Never glue messages from different writers */
1806 if (memcmp(UNIXCREDS(skb), &siocb->scm->creds, sizeof(siocb->scm->creds)) != 0) {
1807 skb_queue_head(&sk->sk_receive_queue, skb);
1808 break;
1810 } else {
1811 /* Copy credentials */
1812 siocb->scm->creds = *UNIXCREDS(skb);
1813 check_creds = 1;
1816 /* Copy address just once */
1817 if (sunaddr)
1819 unix_copy_addr(msg, skb->sk);
1820 sunaddr = NULL;
1823 chunk = min_t(unsigned int, skb->len, size);
1824 if (memcpy_toiovec(msg->msg_iov, skb->data, chunk)) {
1825 skb_queue_head(&sk->sk_receive_queue, skb);
1826 if (copied == 0)
1827 copied = -EFAULT;
1828 break;
1830 copied += chunk;
1831 size -= chunk;
1833 /* Mark read part of skb as used */
1834 if (!(flags & MSG_PEEK))
1836 skb_pull(skb, chunk);
1838 if (UNIXCB(skb).fp)
1839 unix_detach_fds(siocb->scm, skb);
1841 /* put the skb back if we didn't use it up.. */
1842 if (skb->len)
1844 skb_queue_head(&sk->sk_receive_queue, skb);
1845 break;
1848 kfree_skb(skb);
1850 if (siocb->scm->fp)
1851 break;
1853 else
1855 /* It is questionable, see note in unix_dgram_recvmsg.
1857 if (UNIXCB(skb).fp)
1858 siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1860 /* put message back and return */
1861 skb_queue_head(&sk->sk_receive_queue, skb);
1862 break;
1864 } while (size);
1866 mutex_unlock(&u->readlock);
1867 scm_recv(sock, msg, siocb->scm, flags);
1868 out:
1869 return copied ? : err;
1872 static int unix_shutdown(struct socket *sock, int mode)
1874 struct sock *sk = sock->sk;
1875 struct sock *other;
1877 mode = (mode+1)&(RCV_SHUTDOWN|SEND_SHUTDOWN);
1879 if (mode) {
1880 unix_state_lock(sk);
1881 sk->sk_shutdown |= mode;
1882 other=unix_peer(sk);
1883 if (other)
1884 sock_hold(other);
1885 unix_state_unlock(sk);
1886 sk->sk_state_change(sk);
1888 if (other &&
1889 (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET)) {
1891 int peer_mode = 0;
1893 if (mode&RCV_SHUTDOWN)
1894 peer_mode |= SEND_SHUTDOWN;
1895 if (mode&SEND_SHUTDOWN)
1896 peer_mode |= RCV_SHUTDOWN;
1897 unix_state_lock(other);
1898 other->sk_shutdown |= peer_mode;
1899 unix_state_unlock(other);
1900 other->sk_state_change(other);
1901 read_lock(&other->sk_callback_lock);
1902 if (peer_mode == SHUTDOWN_MASK)
1903 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_HUP);
1904 else if (peer_mode & RCV_SHUTDOWN)
1905 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_IN);
1906 read_unlock(&other->sk_callback_lock);
1908 if (other)
1909 sock_put(other);
1911 return 0;
1914 static int unix_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1916 struct sock *sk = sock->sk;
1917 long amount=0;
1918 int err;
1920 switch(cmd)
1922 case SIOCOUTQ:
1923 amount = atomic_read(&sk->sk_wmem_alloc);
1924 err = put_user(amount, (int __user *)arg);
1925 break;
1926 case SIOCINQ:
1928 struct sk_buff *skb;
1930 if (sk->sk_state == TCP_LISTEN) {
1931 err = -EINVAL;
1932 break;
1935 spin_lock(&sk->sk_receive_queue.lock);
1936 if (sk->sk_type == SOCK_STREAM ||
1937 sk->sk_type == SOCK_SEQPACKET) {
1938 skb_queue_walk(&sk->sk_receive_queue, skb)
1939 amount += skb->len;
1940 } else {
1941 skb = skb_peek(&sk->sk_receive_queue);
1942 if (skb)
1943 amount=skb->len;
1945 spin_unlock(&sk->sk_receive_queue.lock);
1946 err = put_user(amount, (int __user *)arg);
1947 break;
1950 default:
1951 err = -ENOIOCTLCMD;
1952 break;
1954 return err;
1957 static unsigned int unix_poll(struct file * file, struct socket *sock, poll_table *wait)
1959 struct sock *sk = sock->sk;
1960 unsigned int mask;
1962 poll_wait(file, sk->sk_sleep, wait);
1963 mask = 0;
1965 /* exceptional events? */
1966 if (sk->sk_err)
1967 mask |= POLLERR;
1968 if (sk->sk_shutdown == SHUTDOWN_MASK)
1969 mask |= POLLHUP;
1970 if (sk->sk_shutdown & RCV_SHUTDOWN)
1971 mask |= POLLRDHUP;
1973 /* readable? */
1974 if (!skb_queue_empty(&sk->sk_receive_queue) ||
1975 (sk->sk_shutdown & RCV_SHUTDOWN))
1976 mask |= POLLIN | POLLRDNORM;
1978 /* Connection-based need to check for termination and startup */
1979 if ((sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) && sk->sk_state == TCP_CLOSE)
1980 mask |= POLLHUP;
1983 * we set writable also when the other side has shut down the
1984 * connection. This prevents stuck sockets.
1986 if (unix_writable(sk))
1987 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
1989 return mask;
1993 #ifdef CONFIG_PROC_FS
1994 static struct sock *first_unix_socket(int *i)
1996 for (*i = 0; *i <= UNIX_HASH_SIZE; (*i)++) {
1997 if (!hlist_empty(&unix_socket_table[*i]))
1998 return __sk_head(&unix_socket_table[*i]);
2000 return NULL;
2003 static struct sock *next_unix_socket(int *i, struct sock *s)
2005 struct sock *next = sk_next(s);
2006 /* More in this chain? */
2007 if (next)
2008 return next;
2009 /* Look for next non-empty chain. */
2010 for ((*i)++; *i <= UNIX_HASH_SIZE; (*i)++) {
2011 if (!hlist_empty(&unix_socket_table[*i]))
2012 return __sk_head(&unix_socket_table[*i]);
2014 return NULL;
2017 struct unix_iter_state {
2018 struct seq_net_private p;
2019 int i;
2021 static struct sock *unix_seq_idx(struct unix_iter_state *iter, loff_t pos)
2023 loff_t off = 0;
2024 struct sock *s;
2026 for (s = first_unix_socket(&iter->i); s; s = next_unix_socket(&iter->i, s)) {
2027 if (s->sk_net != iter->p.net)
2028 continue;
2029 if (off == pos)
2030 return s;
2031 ++off;
2033 return NULL;
2037 static void *unix_seq_start(struct seq_file *seq, loff_t *pos)
2039 struct unix_iter_state *iter = seq->private;
2040 spin_lock(&unix_table_lock);
2041 return *pos ? unix_seq_idx(iter, *pos - 1) : ((void *) 1);
2044 static void *unix_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2046 struct unix_iter_state *iter = seq->private;
2047 struct sock *sk = v;
2048 ++*pos;
2050 if (v == (void *)1)
2051 sk = first_unix_socket(&iter->i);
2052 else
2053 sk = next_unix_socket(&iter->i, sk);
2054 while (sk && (sk->sk_net != iter->p.net))
2055 sk = next_unix_socket(&iter->i, sk);
2056 return sk;
2059 static void unix_seq_stop(struct seq_file *seq, void *v)
2061 spin_unlock(&unix_table_lock);
2064 static int unix_seq_show(struct seq_file *seq, void *v)
2067 if (v == (void *)1)
2068 seq_puts(seq, "Num RefCount Protocol Flags Type St "
2069 "Inode Path\n");
2070 else {
2071 struct sock *s = v;
2072 struct unix_sock *u = unix_sk(s);
2073 unix_state_lock(s);
2075 seq_printf(seq, "%p: %08X %08X %08X %04X %02X %5lu",
2077 atomic_read(&s->sk_refcnt),
2079 s->sk_state == TCP_LISTEN ? __SO_ACCEPTCON : 0,
2080 s->sk_type,
2081 s->sk_socket ?
2082 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTED : SS_UNCONNECTED) :
2083 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTING : SS_DISCONNECTING),
2084 sock_i_ino(s));
2086 if (u->addr) {
2087 int i, len;
2088 seq_putc(seq, ' ');
2090 i = 0;
2091 len = u->addr->len - sizeof(short);
2092 if (!UNIX_ABSTRACT(s))
2093 len--;
2094 else {
2095 seq_putc(seq, '@');
2096 i++;
2098 for ( ; i < len; i++)
2099 seq_putc(seq, u->addr->name->sun_path[i]);
2101 unix_state_unlock(s);
2102 seq_putc(seq, '\n');
2105 return 0;
2108 static const struct seq_operations unix_seq_ops = {
2109 .start = unix_seq_start,
2110 .next = unix_seq_next,
2111 .stop = unix_seq_stop,
2112 .show = unix_seq_show,
2116 static int unix_seq_open(struct inode *inode, struct file *file)
2118 return seq_open_net(inode, file, &unix_seq_ops,
2119 sizeof(struct unix_iter_state));
2122 static const struct file_operations unix_seq_fops = {
2123 .owner = THIS_MODULE,
2124 .open = unix_seq_open,
2125 .read = seq_read,
2126 .llseek = seq_lseek,
2127 .release = seq_release_net,
2130 #endif
2132 static struct net_proto_family unix_family_ops = {
2133 .family = PF_UNIX,
2134 .create = unix_create,
2135 .owner = THIS_MODULE,
2139 static int unix_net_init(struct net *net)
2141 int error = -ENOMEM;
2143 #ifdef CONFIG_PROC_FS
2144 if (!proc_net_fops_create(net, "unix", 0, &unix_seq_fops))
2145 goto out;
2146 #endif
2147 error = 0;
2148 out:
2149 return 0;
2152 static void unix_net_exit(struct net *net)
2154 proc_net_remove(net, "unix");
2157 static struct pernet_operations unix_net_ops = {
2158 .init = unix_net_init,
2159 .exit = unix_net_exit,
2162 static int __init af_unix_init(void)
2164 int rc = -1;
2165 struct sk_buff *dummy_skb;
2167 BUILD_BUG_ON(sizeof(struct unix_skb_parms) > sizeof(dummy_skb->cb));
2169 rc = proto_register(&unix_proto, 1);
2170 if (rc != 0) {
2171 printk(KERN_CRIT "%s: Cannot create unix_sock SLAB cache!\n",
2172 __FUNCTION__);
2173 goto out;
2176 sock_register(&unix_family_ops);
2177 register_pernet_subsys(&unix_net_ops);
2178 unix_sysctl_register(&init_net);
2179 out:
2180 return rc;
2183 static void __exit af_unix_exit(void)
2185 sock_unregister(PF_UNIX);
2186 unix_sysctl_unregister(&init_net);
2187 proto_unregister(&unix_proto);
2188 unregister_pernet_subsys(&unix_net_ops);
2191 module_init(af_unix_init);
2192 module_exit(af_unix_exit);
2194 MODULE_LICENSE("GPL");
2195 MODULE_ALIAS_NETPROTO(PF_UNIX);