x86: 64-bit, add the new split_large_page() function
[wrt350n-kernel.git] / net / unix / af_unix.c
blobeea75888805e151d2d26ffae2518971ee330349d
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 struct sock *unix_peer_get(struct sock *s)
174 struct sock *peer;
176 unix_state_lock(s);
177 peer = unix_peer(s);
178 if (peer)
179 sock_hold(peer);
180 unix_state_unlock(s);
181 return peer;
184 static inline void unix_release_addr(struct unix_address *addr)
186 if (atomic_dec_and_test(&addr->refcnt))
187 kfree(addr);
191 * Check unix socket name:
192 * - should be not zero length.
193 * - if started by not zero, should be NULL terminated (FS object)
194 * - if started by zero, it is abstract name.
197 static int unix_mkname(struct sockaddr_un * sunaddr, int len, unsigned *hashp)
199 if (len <= sizeof(short) || len > sizeof(*sunaddr))
200 return -EINVAL;
201 if (!sunaddr || sunaddr->sun_family != AF_UNIX)
202 return -EINVAL;
203 if (sunaddr->sun_path[0]) {
205 * This may look like an off by one error but it is a bit more
206 * subtle. 108 is the longest valid AF_UNIX path for a binding.
207 * sun_path[108] doesnt as such exist. However in kernel space
208 * we are guaranteed that it is a valid memory location in our
209 * kernel address buffer.
211 ((char *)sunaddr)[len]=0;
212 len = strlen(sunaddr->sun_path)+1+sizeof(short);
213 return len;
216 *hashp = unix_hash_fold(csum_partial((char*)sunaddr, len, 0));
217 return len;
220 static void __unix_remove_socket(struct sock *sk)
222 sk_del_node_init(sk);
225 static void __unix_insert_socket(struct hlist_head *list, struct sock *sk)
227 BUG_TRAP(sk_unhashed(sk));
228 sk_add_node(sk, list);
231 static inline void unix_remove_socket(struct sock *sk)
233 spin_lock(&unix_table_lock);
234 __unix_remove_socket(sk);
235 spin_unlock(&unix_table_lock);
238 static inline void unix_insert_socket(struct hlist_head *list, struct sock *sk)
240 spin_lock(&unix_table_lock);
241 __unix_insert_socket(list, sk);
242 spin_unlock(&unix_table_lock);
245 static struct sock *__unix_find_socket_byname(struct net *net,
246 struct sockaddr_un *sunname,
247 int len, int type, unsigned hash)
249 struct sock *s;
250 struct hlist_node *node;
252 sk_for_each(s, node, &unix_socket_table[hash ^ type]) {
253 struct unix_sock *u = unix_sk(s);
255 if (s->sk_net != net)
256 continue;
258 if (u->addr->len == len &&
259 !memcmp(u->addr->name, sunname, len))
260 goto found;
262 s = NULL;
263 found:
264 return s;
267 static inline struct sock *unix_find_socket_byname(struct net *net,
268 struct sockaddr_un *sunname,
269 int len, int type,
270 unsigned hash)
272 struct sock *s;
274 spin_lock(&unix_table_lock);
275 s = __unix_find_socket_byname(net, sunname, len, type, hash);
276 if (s)
277 sock_hold(s);
278 spin_unlock(&unix_table_lock);
279 return s;
282 static struct sock *unix_find_socket_byinode(struct net *net, struct inode *i)
284 struct sock *s;
285 struct hlist_node *node;
287 spin_lock(&unix_table_lock);
288 sk_for_each(s, node,
289 &unix_socket_table[i->i_ino & (UNIX_HASH_SIZE - 1)]) {
290 struct dentry *dentry = unix_sk(s)->dentry;
292 if (s->sk_net != net)
293 continue;
295 if(dentry && dentry->d_inode == i)
297 sock_hold(s);
298 goto found;
301 s = NULL;
302 found:
303 spin_unlock(&unix_table_lock);
304 return s;
307 static inline int unix_writable(struct sock *sk)
309 return (atomic_read(&sk->sk_wmem_alloc) << 2) <= sk->sk_sndbuf;
312 static void unix_write_space(struct sock *sk)
314 read_lock(&sk->sk_callback_lock);
315 if (unix_writable(sk)) {
316 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
317 wake_up_interruptible_sync(sk->sk_sleep);
318 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
320 read_unlock(&sk->sk_callback_lock);
323 /* When dgram socket disconnects (or changes its peer), we clear its receive
324 * queue of packets arrived from previous peer. First, it allows to do
325 * flow control based only on wmem_alloc; second, sk connected to peer
326 * may receive messages only from that peer. */
327 static void unix_dgram_disconnected(struct sock *sk, struct sock *other)
329 if (!skb_queue_empty(&sk->sk_receive_queue)) {
330 skb_queue_purge(&sk->sk_receive_queue);
331 wake_up_interruptible_all(&unix_sk(sk)->peer_wait);
333 /* If one link of bidirectional dgram pipe is disconnected,
334 * we signal error. Messages are lost. Do not make this,
335 * when peer was not connected to us.
337 if (!sock_flag(other, SOCK_DEAD) && unix_peer(other) == sk) {
338 other->sk_err = ECONNRESET;
339 other->sk_error_report(other);
344 static void unix_sock_destructor(struct sock *sk)
346 struct unix_sock *u = unix_sk(sk);
348 skb_queue_purge(&sk->sk_receive_queue);
350 BUG_TRAP(!atomic_read(&sk->sk_wmem_alloc));
351 BUG_TRAP(sk_unhashed(sk));
352 BUG_TRAP(!sk->sk_socket);
353 if (!sock_flag(sk, SOCK_DEAD)) {
354 printk("Attempt to release alive unix socket: %p\n", sk);
355 return;
358 if (u->addr)
359 unix_release_addr(u->addr);
361 atomic_dec(&unix_nr_socks);
362 #ifdef UNIX_REFCNT_DEBUG
363 printk(KERN_DEBUG "UNIX %p is destroyed, %d are still alive.\n", sk, atomic_read(&unix_nr_socks));
364 #endif
367 static int unix_release_sock (struct sock *sk, int embrion)
369 struct unix_sock *u = unix_sk(sk);
370 struct dentry *dentry;
371 struct vfsmount *mnt;
372 struct sock *skpair;
373 struct sk_buff *skb;
374 int state;
376 unix_remove_socket(sk);
378 /* Clear state */
379 unix_state_lock(sk);
380 sock_orphan(sk);
381 sk->sk_shutdown = SHUTDOWN_MASK;
382 dentry = u->dentry;
383 u->dentry = NULL;
384 mnt = u->mnt;
385 u->mnt = NULL;
386 state = sk->sk_state;
387 sk->sk_state = TCP_CLOSE;
388 unix_state_unlock(sk);
390 wake_up_interruptible_all(&u->peer_wait);
392 skpair=unix_peer(sk);
394 if (skpair!=NULL) {
395 if (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) {
396 unix_state_lock(skpair);
397 /* No more writes */
398 skpair->sk_shutdown = SHUTDOWN_MASK;
399 if (!skb_queue_empty(&sk->sk_receive_queue) || embrion)
400 skpair->sk_err = ECONNRESET;
401 unix_state_unlock(skpair);
402 skpair->sk_state_change(skpair);
403 read_lock(&skpair->sk_callback_lock);
404 sk_wake_async(skpair, SOCK_WAKE_WAITD, POLL_HUP);
405 read_unlock(&skpair->sk_callback_lock);
407 sock_put(skpair); /* It may now die */
408 unix_peer(sk) = NULL;
411 /* Try to flush out this socket. Throw out buffers at least */
413 while ((skb = skb_dequeue(&sk->sk_receive_queue)) != NULL) {
414 if (state==TCP_LISTEN)
415 unix_release_sock(skb->sk, 1);
416 /* passed fds are erased in the kfree_skb hook */
417 kfree_skb(skb);
420 if (dentry) {
421 dput(dentry);
422 mntput(mnt);
425 sock_put(sk);
427 /* ---- Socket is dead now and most probably destroyed ---- */
430 * Fixme: BSD difference: In BSD all sockets connected to use get
431 * ECONNRESET and we die on the spot. In Linux we behave
432 * like files and pipes do and wait for the last
433 * dereference.
435 * Can't we simply set sock->err?
437 * What the above comment does talk about? --ANK(980817)
440 if (unix_tot_inflight)
441 unix_gc(); /* Garbage collect fds */
443 return 0;
446 static int unix_listen(struct socket *sock, int backlog)
448 int err;
449 struct sock *sk = sock->sk;
450 struct unix_sock *u = unix_sk(sk);
452 err = -EOPNOTSUPP;
453 if (sock->type!=SOCK_STREAM && sock->type!=SOCK_SEQPACKET)
454 goto out; /* Only stream/seqpacket sockets accept */
455 err = -EINVAL;
456 if (!u->addr)
457 goto out; /* No listens on an unbound socket */
458 unix_state_lock(sk);
459 if (sk->sk_state != TCP_CLOSE && sk->sk_state != TCP_LISTEN)
460 goto out_unlock;
461 if (backlog > sk->sk_max_ack_backlog)
462 wake_up_interruptible_all(&u->peer_wait);
463 sk->sk_max_ack_backlog = backlog;
464 sk->sk_state = TCP_LISTEN;
465 /* set credentials so connect can copy them */
466 sk->sk_peercred.pid = task_tgid_vnr(current);
467 sk->sk_peercred.uid = current->euid;
468 sk->sk_peercred.gid = current->egid;
469 err = 0;
471 out_unlock:
472 unix_state_unlock(sk);
473 out:
474 return err;
477 static int unix_release(struct socket *);
478 static int unix_bind(struct socket *, struct sockaddr *, int);
479 static int unix_stream_connect(struct socket *, struct sockaddr *,
480 int addr_len, int flags);
481 static int unix_socketpair(struct socket *, struct socket *);
482 static int unix_accept(struct socket *, struct socket *, int);
483 static int unix_getname(struct socket *, struct sockaddr *, int *, int);
484 static unsigned int unix_poll(struct file *, struct socket *, poll_table *);
485 static int unix_ioctl(struct socket *, unsigned int, unsigned long);
486 static int unix_shutdown(struct socket *, int);
487 static int unix_stream_sendmsg(struct kiocb *, struct socket *,
488 struct msghdr *, size_t);
489 static int unix_stream_recvmsg(struct kiocb *, struct socket *,
490 struct msghdr *, size_t, int);
491 static int unix_dgram_sendmsg(struct kiocb *, struct socket *,
492 struct msghdr *, size_t);
493 static int unix_dgram_recvmsg(struct kiocb *, struct socket *,
494 struct msghdr *, size_t, int);
495 static int unix_dgram_connect(struct socket *, struct sockaddr *,
496 int, int);
497 static int unix_seqpacket_sendmsg(struct kiocb *, struct socket *,
498 struct msghdr *, size_t);
500 static const struct proto_ops unix_stream_ops = {
501 .family = PF_UNIX,
502 .owner = THIS_MODULE,
503 .release = unix_release,
504 .bind = unix_bind,
505 .connect = unix_stream_connect,
506 .socketpair = unix_socketpair,
507 .accept = unix_accept,
508 .getname = unix_getname,
509 .poll = unix_poll,
510 .ioctl = unix_ioctl,
511 .listen = unix_listen,
512 .shutdown = unix_shutdown,
513 .setsockopt = sock_no_setsockopt,
514 .getsockopt = sock_no_getsockopt,
515 .sendmsg = unix_stream_sendmsg,
516 .recvmsg = unix_stream_recvmsg,
517 .mmap = sock_no_mmap,
518 .sendpage = sock_no_sendpage,
521 static const struct proto_ops unix_dgram_ops = {
522 .family = PF_UNIX,
523 .owner = THIS_MODULE,
524 .release = unix_release,
525 .bind = unix_bind,
526 .connect = unix_dgram_connect,
527 .socketpair = unix_socketpair,
528 .accept = sock_no_accept,
529 .getname = unix_getname,
530 .poll = datagram_poll,
531 .ioctl = unix_ioctl,
532 .listen = sock_no_listen,
533 .shutdown = unix_shutdown,
534 .setsockopt = sock_no_setsockopt,
535 .getsockopt = sock_no_getsockopt,
536 .sendmsg = unix_dgram_sendmsg,
537 .recvmsg = unix_dgram_recvmsg,
538 .mmap = sock_no_mmap,
539 .sendpage = sock_no_sendpage,
542 static const struct proto_ops unix_seqpacket_ops = {
543 .family = PF_UNIX,
544 .owner = THIS_MODULE,
545 .release = unix_release,
546 .bind = unix_bind,
547 .connect = unix_stream_connect,
548 .socketpair = unix_socketpair,
549 .accept = unix_accept,
550 .getname = unix_getname,
551 .poll = datagram_poll,
552 .ioctl = unix_ioctl,
553 .listen = unix_listen,
554 .shutdown = unix_shutdown,
555 .setsockopt = sock_no_setsockopt,
556 .getsockopt = sock_no_getsockopt,
557 .sendmsg = unix_seqpacket_sendmsg,
558 .recvmsg = unix_dgram_recvmsg,
559 .mmap = sock_no_mmap,
560 .sendpage = sock_no_sendpage,
563 static struct proto unix_proto = {
564 .name = "UNIX",
565 .owner = THIS_MODULE,
566 .obj_size = sizeof(struct unix_sock),
570 * AF_UNIX sockets do not interact with hardware, hence they
571 * dont trigger interrupts - so it's safe for them to have
572 * bh-unsafe locking for their sk_receive_queue.lock. Split off
573 * this special lock-class by reinitializing the spinlock key:
575 static struct lock_class_key af_unix_sk_receive_queue_lock_key;
577 static struct sock * unix_create1(struct net *net, struct socket *sock)
579 struct sock *sk = NULL;
580 struct unix_sock *u;
582 atomic_inc(&unix_nr_socks);
583 if (atomic_read(&unix_nr_socks) > 2 * get_max_files())
584 goto out;
586 sk = sk_alloc(net, PF_UNIX, GFP_KERNEL, &unix_proto);
587 if (!sk)
588 goto out;
590 sock_init_data(sock,sk);
591 lockdep_set_class(&sk->sk_receive_queue.lock,
592 &af_unix_sk_receive_queue_lock_key);
594 sk->sk_write_space = unix_write_space;
595 sk->sk_max_ack_backlog = net->unx.sysctl_max_dgram_qlen;
596 sk->sk_destruct = unix_sock_destructor;
597 u = unix_sk(sk);
598 u->dentry = NULL;
599 u->mnt = NULL;
600 spin_lock_init(&u->lock);
601 atomic_set(&u->inflight, 0);
602 INIT_LIST_HEAD(&u->link);
603 mutex_init(&u->readlock); /* single task reading lock */
604 init_waitqueue_head(&u->peer_wait);
605 unix_insert_socket(unix_sockets_unbound, sk);
606 out:
607 if (sk == NULL)
608 atomic_dec(&unix_nr_socks);
609 return sk;
612 static int unix_create(struct net *net, struct socket *sock, int protocol)
614 if (protocol && protocol != PF_UNIX)
615 return -EPROTONOSUPPORT;
617 sock->state = SS_UNCONNECTED;
619 switch (sock->type) {
620 case SOCK_STREAM:
621 sock->ops = &unix_stream_ops;
622 break;
624 * Believe it or not BSD has AF_UNIX, SOCK_RAW though
625 * nothing uses it.
627 case SOCK_RAW:
628 sock->type=SOCK_DGRAM;
629 case SOCK_DGRAM:
630 sock->ops = &unix_dgram_ops;
631 break;
632 case SOCK_SEQPACKET:
633 sock->ops = &unix_seqpacket_ops;
634 break;
635 default:
636 return -ESOCKTNOSUPPORT;
639 return unix_create1(net, sock) ? 0 : -ENOMEM;
642 static int unix_release(struct socket *sock)
644 struct sock *sk = sock->sk;
646 if (!sk)
647 return 0;
649 sock->sk = NULL;
651 return unix_release_sock (sk, 0);
654 static int unix_autobind(struct socket *sock)
656 struct sock *sk = sock->sk;
657 struct net *net = sk->sk_net;
658 struct unix_sock *u = unix_sk(sk);
659 static u32 ordernum = 1;
660 struct unix_address * addr;
661 int err;
663 mutex_lock(&u->readlock);
665 err = 0;
666 if (u->addr)
667 goto out;
669 err = -ENOMEM;
670 addr = kzalloc(sizeof(*addr) + sizeof(short) + 16, GFP_KERNEL);
671 if (!addr)
672 goto out;
674 addr->name->sun_family = AF_UNIX;
675 atomic_set(&addr->refcnt, 1);
677 retry:
678 addr->len = sprintf(addr->name->sun_path+1, "%05x", ordernum) + 1 + sizeof(short);
679 addr->hash = unix_hash_fold(csum_partial((void*)addr->name, addr->len, 0));
681 spin_lock(&unix_table_lock);
682 ordernum = (ordernum+1)&0xFFFFF;
684 if (__unix_find_socket_byname(net, addr->name, addr->len, sock->type,
685 addr->hash)) {
686 spin_unlock(&unix_table_lock);
687 /* Sanity yield. It is unusual case, but yet... */
688 if (!(ordernum&0xFF))
689 yield();
690 goto retry;
692 addr->hash ^= sk->sk_type;
694 __unix_remove_socket(sk);
695 u->addr = addr;
696 __unix_insert_socket(&unix_socket_table[addr->hash], sk);
697 spin_unlock(&unix_table_lock);
698 err = 0;
700 out: mutex_unlock(&u->readlock);
701 return err;
704 static struct sock *unix_find_other(struct net *net,
705 struct sockaddr_un *sunname, int len,
706 int type, unsigned hash, int *error)
708 struct sock *u;
709 struct nameidata nd;
710 int err = 0;
712 if (sunname->sun_path[0]) {
713 err = path_lookup(sunname->sun_path, LOOKUP_FOLLOW, &nd);
714 if (err)
715 goto fail;
716 err = vfs_permission(&nd, MAY_WRITE);
717 if (err)
718 goto put_fail;
720 err = -ECONNREFUSED;
721 if (!S_ISSOCK(nd.dentry->d_inode->i_mode))
722 goto put_fail;
723 u=unix_find_socket_byinode(net, nd.dentry->d_inode);
724 if (!u)
725 goto put_fail;
727 if (u->sk_type == type)
728 touch_atime(nd.mnt, nd.dentry);
730 path_release(&nd);
732 err=-EPROTOTYPE;
733 if (u->sk_type != type) {
734 sock_put(u);
735 goto fail;
737 } else {
738 err = -ECONNREFUSED;
739 u=unix_find_socket_byname(net, sunname, len, type, hash);
740 if (u) {
741 struct dentry *dentry;
742 dentry = unix_sk(u)->dentry;
743 if (dentry)
744 touch_atime(unix_sk(u)->mnt, dentry);
745 } else
746 goto fail;
748 return u;
750 put_fail:
751 path_release(&nd);
752 fail:
753 *error=err;
754 return NULL;
758 static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
760 struct sock *sk = sock->sk;
761 struct net *net = sk->sk_net;
762 struct unix_sock *u = unix_sk(sk);
763 struct sockaddr_un *sunaddr=(struct sockaddr_un *)uaddr;
764 struct dentry * dentry = NULL;
765 struct nameidata nd;
766 int err;
767 unsigned hash;
768 struct unix_address *addr;
769 struct hlist_head *list;
771 err = -EINVAL;
772 if (sunaddr->sun_family != AF_UNIX)
773 goto out;
775 if (addr_len==sizeof(short)) {
776 err = unix_autobind(sock);
777 goto out;
780 err = unix_mkname(sunaddr, addr_len, &hash);
781 if (err < 0)
782 goto out;
783 addr_len = err;
785 mutex_lock(&u->readlock);
787 err = -EINVAL;
788 if (u->addr)
789 goto out_up;
791 err = -ENOMEM;
792 addr = kmalloc(sizeof(*addr)+addr_len, GFP_KERNEL);
793 if (!addr)
794 goto out_up;
796 memcpy(addr->name, sunaddr, addr_len);
797 addr->len = addr_len;
798 addr->hash = hash ^ sk->sk_type;
799 atomic_set(&addr->refcnt, 1);
801 if (sunaddr->sun_path[0]) {
802 unsigned int mode;
803 err = 0;
805 * Get the parent directory, calculate the hash for last
806 * component.
808 err = path_lookup(sunaddr->sun_path, LOOKUP_PARENT, &nd);
809 if (err)
810 goto out_mknod_parent;
812 dentry = lookup_create(&nd, 0);
813 err = PTR_ERR(dentry);
814 if (IS_ERR(dentry))
815 goto out_mknod_unlock;
818 * All right, let's create it.
820 mode = S_IFSOCK |
821 (SOCK_INODE(sock)->i_mode & ~current->fs->umask);
822 err = vfs_mknod(nd.dentry->d_inode, dentry, mode, 0);
823 if (err)
824 goto out_mknod_dput;
825 mutex_unlock(&nd.dentry->d_inode->i_mutex);
826 dput(nd.dentry);
827 nd.dentry = dentry;
829 addr->hash = UNIX_HASH_SIZE;
832 spin_lock(&unix_table_lock);
834 if (!sunaddr->sun_path[0]) {
835 err = -EADDRINUSE;
836 if (__unix_find_socket_byname(net, sunaddr, addr_len,
837 sk->sk_type, hash)) {
838 unix_release_addr(addr);
839 goto out_unlock;
842 list = &unix_socket_table[addr->hash];
843 } else {
844 list = &unix_socket_table[dentry->d_inode->i_ino & (UNIX_HASH_SIZE-1)];
845 u->dentry = nd.dentry;
846 u->mnt = nd.mnt;
849 err = 0;
850 __unix_remove_socket(sk);
851 u->addr = addr;
852 __unix_insert_socket(list, sk);
854 out_unlock:
855 spin_unlock(&unix_table_lock);
856 out_up:
857 mutex_unlock(&u->readlock);
858 out:
859 return err;
861 out_mknod_dput:
862 dput(dentry);
863 out_mknod_unlock:
864 mutex_unlock(&nd.dentry->d_inode->i_mutex);
865 path_release(&nd);
866 out_mknod_parent:
867 if (err==-EEXIST)
868 err=-EADDRINUSE;
869 unix_release_addr(addr);
870 goto out_up;
873 static void unix_state_double_lock(struct sock *sk1, struct sock *sk2)
875 if (unlikely(sk1 == sk2) || !sk2) {
876 unix_state_lock(sk1);
877 return;
879 if (sk1 < sk2) {
880 unix_state_lock(sk1);
881 unix_state_lock_nested(sk2);
882 } else {
883 unix_state_lock(sk2);
884 unix_state_lock_nested(sk1);
888 static void unix_state_double_unlock(struct sock *sk1, struct sock *sk2)
890 if (unlikely(sk1 == sk2) || !sk2) {
891 unix_state_unlock(sk1);
892 return;
894 unix_state_unlock(sk1);
895 unix_state_unlock(sk2);
898 static int unix_dgram_connect(struct socket *sock, struct sockaddr *addr,
899 int alen, int flags)
901 struct sock *sk = sock->sk;
902 struct net *net = sk->sk_net;
903 struct sockaddr_un *sunaddr=(struct sockaddr_un*)addr;
904 struct sock *other;
905 unsigned hash;
906 int err;
908 if (addr->sa_family != AF_UNSPEC) {
909 err = unix_mkname(sunaddr, alen, &hash);
910 if (err < 0)
911 goto out;
912 alen = err;
914 if (test_bit(SOCK_PASSCRED, &sock->flags) &&
915 !unix_sk(sk)->addr && (err = unix_autobind(sock)) != 0)
916 goto out;
918 restart:
919 other=unix_find_other(net, sunaddr, alen, sock->type, hash, &err);
920 if (!other)
921 goto out;
923 unix_state_double_lock(sk, other);
925 /* Apparently VFS overslept socket death. Retry. */
926 if (sock_flag(other, SOCK_DEAD)) {
927 unix_state_double_unlock(sk, other);
928 sock_put(other);
929 goto restart;
932 err = -EPERM;
933 if (!unix_may_send(sk, other))
934 goto out_unlock;
936 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
937 if (err)
938 goto out_unlock;
940 } else {
942 * 1003.1g breaking connected state with AF_UNSPEC
944 other = NULL;
945 unix_state_double_lock(sk, other);
949 * If it was connected, reconnect.
951 if (unix_peer(sk)) {
952 struct sock *old_peer = unix_peer(sk);
953 unix_peer(sk)=other;
954 unix_state_double_unlock(sk, other);
956 if (other != old_peer)
957 unix_dgram_disconnected(sk, old_peer);
958 sock_put(old_peer);
959 } else {
960 unix_peer(sk)=other;
961 unix_state_double_unlock(sk, other);
963 return 0;
965 out_unlock:
966 unix_state_double_unlock(sk, other);
967 sock_put(other);
968 out:
969 return err;
972 static long unix_wait_for_peer(struct sock *other, long timeo)
974 struct unix_sock *u = unix_sk(other);
975 int sched;
976 DEFINE_WAIT(wait);
978 prepare_to_wait_exclusive(&u->peer_wait, &wait, TASK_INTERRUPTIBLE);
980 sched = !sock_flag(other, SOCK_DEAD) &&
981 !(other->sk_shutdown & RCV_SHUTDOWN) &&
982 (skb_queue_len(&other->sk_receive_queue) >
983 other->sk_max_ack_backlog);
985 unix_state_unlock(other);
987 if (sched)
988 timeo = schedule_timeout(timeo);
990 finish_wait(&u->peer_wait, &wait);
991 return timeo;
994 static int unix_stream_connect(struct socket *sock, struct sockaddr *uaddr,
995 int addr_len, int flags)
997 struct sockaddr_un *sunaddr=(struct sockaddr_un *)uaddr;
998 struct sock *sk = sock->sk;
999 struct net *net = sk->sk_net;
1000 struct unix_sock *u = unix_sk(sk), *newu, *otheru;
1001 struct sock *newsk = NULL;
1002 struct sock *other = NULL;
1003 struct sk_buff *skb = NULL;
1004 unsigned hash;
1005 int st;
1006 int err;
1007 long timeo;
1009 err = unix_mkname(sunaddr, addr_len, &hash);
1010 if (err < 0)
1011 goto out;
1012 addr_len = err;
1014 if (test_bit(SOCK_PASSCRED, &sock->flags)
1015 && !u->addr && (err = unix_autobind(sock)) != 0)
1016 goto out;
1018 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
1020 /* First of all allocate resources.
1021 If we will make it after state is locked,
1022 we will have to recheck all again in any case.
1025 err = -ENOMEM;
1027 /* create new sock for complete connection */
1028 newsk = unix_create1(sk->sk_net, NULL);
1029 if (newsk == NULL)
1030 goto out;
1032 /* Allocate skb for sending to listening sock */
1033 skb = sock_wmalloc(newsk, 1, 0, GFP_KERNEL);
1034 if (skb == NULL)
1035 goto out;
1037 restart:
1038 /* Find listening sock. */
1039 other = unix_find_other(net, sunaddr, addr_len, sk->sk_type, hash, &err);
1040 if (!other)
1041 goto out;
1043 /* Latch state of peer */
1044 unix_state_lock(other);
1046 /* Apparently VFS overslept socket death. Retry. */
1047 if (sock_flag(other, SOCK_DEAD)) {
1048 unix_state_unlock(other);
1049 sock_put(other);
1050 goto restart;
1053 err = -ECONNREFUSED;
1054 if (other->sk_state != TCP_LISTEN)
1055 goto out_unlock;
1057 if (skb_queue_len(&other->sk_receive_queue) >
1058 other->sk_max_ack_backlog) {
1059 err = -EAGAIN;
1060 if (!timeo)
1061 goto out_unlock;
1063 timeo = unix_wait_for_peer(other, timeo);
1065 err = sock_intr_errno(timeo);
1066 if (signal_pending(current))
1067 goto out;
1068 sock_put(other);
1069 goto restart;
1072 /* Latch our state.
1074 It is tricky place. We need to grab write lock and cannot
1075 drop lock on peer. It is dangerous because deadlock is
1076 possible. Connect to self case and simultaneous
1077 attempt to connect are eliminated by checking socket
1078 state. other is TCP_LISTEN, if sk is TCP_LISTEN we
1079 check this before attempt to grab lock.
1081 Well, and we have to recheck the state after socket locked.
1083 st = sk->sk_state;
1085 switch (st) {
1086 case TCP_CLOSE:
1087 /* This is ok... continue with connect */
1088 break;
1089 case TCP_ESTABLISHED:
1090 /* Socket is already connected */
1091 err = -EISCONN;
1092 goto out_unlock;
1093 default:
1094 err = -EINVAL;
1095 goto out_unlock;
1098 unix_state_lock_nested(sk);
1100 if (sk->sk_state != st) {
1101 unix_state_unlock(sk);
1102 unix_state_unlock(other);
1103 sock_put(other);
1104 goto restart;
1107 err = security_unix_stream_connect(sock, other->sk_socket, newsk);
1108 if (err) {
1109 unix_state_unlock(sk);
1110 goto out_unlock;
1113 /* The way is open! Fastly set all the necessary fields... */
1115 sock_hold(sk);
1116 unix_peer(newsk) = sk;
1117 newsk->sk_state = TCP_ESTABLISHED;
1118 newsk->sk_type = sk->sk_type;
1119 newsk->sk_peercred.pid = task_tgid_vnr(current);
1120 newsk->sk_peercred.uid = current->euid;
1121 newsk->sk_peercred.gid = current->egid;
1122 newu = unix_sk(newsk);
1123 newsk->sk_sleep = &newu->peer_wait;
1124 otheru = unix_sk(other);
1126 /* copy address information from listening to new sock*/
1127 if (otheru->addr) {
1128 atomic_inc(&otheru->addr->refcnt);
1129 newu->addr = otheru->addr;
1131 if (otheru->dentry) {
1132 newu->dentry = dget(otheru->dentry);
1133 newu->mnt = mntget(otheru->mnt);
1136 /* Set credentials */
1137 sk->sk_peercred = other->sk_peercred;
1139 sock->state = SS_CONNECTED;
1140 sk->sk_state = TCP_ESTABLISHED;
1141 sock_hold(newsk);
1143 smp_mb__after_atomic_inc(); /* sock_hold() does an atomic_inc() */
1144 unix_peer(sk) = newsk;
1146 unix_state_unlock(sk);
1148 /* take ten and and send info to listening sock */
1149 spin_lock(&other->sk_receive_queue.lock);
1150 __skb_queue_tail(&other->sk_receive_queue, skb);
1151 spin_unlock(&other->sk_receive_queue.lock);
1152 unix_state_unlock(other);
1153 other->sk_data_ready(other, 0);
1154 sock_put(other);
1155 return 0;
1157 out_unlock:
1158 if (other)
1159 unix_state_unlock(other);
1161 out:
1162 if (skb)
1163 kfree_skb(skb);
1164 if (newsk)
1165 unix_release_sock(newsk, 0);
1166 if (other)
1167 sock_put(other);
1168 return err;
1171 static int unix_socketpair(struct socket *socka, struct socket *sockb)
1173 struct sock *ska=socka->sk, *skb = sockb->sk;
1175 /* Join our sockets back to back */
1176 sock_hold(ska);
1177 sock_hold(skb);
1178 unix_peer(ska)=skb;
1179 unix_peer(skb)=ska;
1180 ska->sk_peercred.pid = skb->sk_peercred.pid = task_tgid_vnr(current);
1181 ska->sk_peercred.uid = skb->sk_peercred.uid = current->euid;
1182 ska->sk_peercred.gid = skb->sk_peercred.gid = current->egid;
1184 if (ska->sk_type != SOCK_DGRAM) {
1185 ska->sk_state = TCP_ESTABLISHED;
1186 skb->sk_state = TCP_ESTABLISHED;
1187 socka->state = SS_CONNECTED;
1188 sockb->state = SS_CONNECTED;
1190 return 0;
1193 static int unix_accept(struct socket *sock, struct socket *newsock, int flags)
1195 struct sock *sk = sock->sk;
1196 struct sock *tsk;
1197 struct sk_buff *skb;
1198 int err;
1200 err = -EOPNOTSUPP;
1201 if (sock->type!=SOCK_STREAM && sock->type!=SOCK_SEQPACKET)
1202 goto out;
1204 err = -EINVAL;
1205 if (sk->sk_state != TCP_LISTEN)
1206 goto out;
1208 /* If socket state is TCP_LISTEN it cannot change (for now...),
1209 * so that no locks are necessary.
1212 skb = skb_recv_datagram(sk, 0, flags&O_NONBLOCK, &err);
1213 if (!skb) {
1214 /* This means receive shutdown. */
1215 if (err == 0)
1216 err = -EINVAL;
1217 goto out;
1220 tsk = skb->sk;
1221 skb_free_datagram(sk, skb);
1222 wake_up_interruptible(&unix_sk(sk)->peer_wait);
1224 /* attach accepted sock to socket */
1225 unix_state_lock(tsk);
1226 newsock->state = SS_CONNECTED;
1227 sock_graft(tsk, newsock);
1228 unix_state_unlock(tsk);
1229 return 0;
1231 out:
1232 return err;
1236 static int unix_getname(struct socket *sock, struct sockaddr *uaddr, int *uaddr_len, int peer)
1238 struct sock *sk = sock->sk;
1239 struct unix_sock *u;
1240 struct sockaddr_un *sunaddr=(struct sockaddr_un *)uaddr;
1241 int err = 0;
1243 if (peer) {
1244 sk = unix_peer_get(sk);
1246 err = -ENOTCONN;
1247 if (!sk)
1248 goto out;
1249 err = 0;
1250 } else {
1251 sock_hold(sk);
1254 u = unix_sk(sk);
1255 unix_state_lock(sk);
1256 if (!u->addr) {
1257 sunaddr->sun_family = AF_UNIX;
1258 sunaddr->sun_path[0] = 0;
1259 *uaddr_len = sizeof(short);
1260 } else {
1261 struct unix_address *addr = u->addr;
1263 *uaddr_len = addr->len;
1264 memcpy(sunaddr, addr->name, *uaddr_len);
1266 unix_state_unlock(sk);
1267 sock_put(sk);
1268 out:
1269 return err;
1272 static void unix_detach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1274 int i;
1276 scm->fp = UNIXCB(skb).fp;
1277 skb->destructor = sock_wfree;
1278 UNIXCB(skb).fp = NULL;
1280 for (i=scm->fp->count-1; i>=0; i--)
1281 unix_notinflight(scm->fp->fp[i]);
1284 static void unix_destruct_fds(struct sk_buff *skb)
1286 struct scm_cookie scm;
1287 memset(&scm, 0, sizeof(scm));
1288 unix_detach_fds(&scm, skb);
1290 /* Alas, it calls VFS */
1291 /* So fscking what? fput() had been SMP-safe since the last Summer */
1292 scm_destroy(&scm);
1293 sock_wfree(skb);
1296 static void unix_attach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1298 int i;
1299 for (i=scm->fp->count-1; i>=0; i--)
1300 unix_inflight(scm->fp->fp[i]);
1301 UNIXCB(skb).fp = scm->fp;
1302 skb->destructor = unix_destruct_fds;
1303 scm->fp = NULL;
1307 * Send AF_UNIX data.
1310 static int unix_dgram_sendmsg(struct kiocb *kiocb, struct socket *sock,
1311 struct msghdr *msg, size_t len)
1313 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1314 struct sock *sk = sock->sk;
1315 struct net *net = sk->sk_net;
1316 struct unix_sock *u = unix_sk(sk);
1317 struct sockaddr_un *sunaddr=msg->msg_name;
1318 struct sock *other = NULL;
1319 int namelen = 0; /* fake GCC */
1320 int err;
1321 unsigned hash;
1322 struct sk_buff *skb;
1323 long timeo;
1324 struct scm_cookie tmp_scm;
1326 if (NULL == siocb->scm)
1327 siocb->scm = &tmp_scm;
1328 err = scm_send(sock, msg, siocb->scm);
1329 if (err < 0)
1330 return err;
1332 err = -EOPNOTSUPP;
1333 if (msg->msg_flags&MSG_OOB)
1334 goto out;
1336 if (msg->msg_namelen) {
1337 err = unix_mkname(sunaddr, msg->msg_namelen, &hash);
1338 if (err < 0)
1339 goto out;
1340 namelen = err;
1341 } else {
1342 sunaddr = NULL;
1343 err = -ENOTCONN;
1344 other = unix_peer_get(sk);
1345 if (!other)
1346 goto out;
1349 if (test_bit(SOCK_PASSCRED, &sock->flags)
1350 && !u->addr && (err = unix_autobind(sock)) != 0)
1351 goto out;
1353 err = -EMSGSIZE;
1354 if (len > sk->sk_sndbuf - 32)
1355 goto out;
1357 skb = sock_alloc_send_skb(sk, len, msg->msg_flags&MSG_DONTWAIT, &err);
1358 if (skb==NULL)
1359 goto out;
1361 memcpy(UNIXCREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1362 if (siocb->scm->fp)
1363 unix_attach_fds(siocb->scm, skb);
1364 unix_get_secdata(siocb->scm, skb);
1366 skb_reset_transport_header(skb);
1367 err = memcpy_fromiovec(skb_put(skb,len), msg->msg_iov, len);
1368 if (err)
1369 goto out_free;
1371 timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1373 restart:
1374 if (!other) {
1375 err = -ECONNRESET;
1376 if (sunaddr == NULL)
1377 goto out_free;
1379 other = unix_find_other(net, sunaddr, namelen, sk->sk_type,
1380 hash, &err);
1381 if (other==NULL)
1382 goto out_free;
1385 unix_state_lock(other);
1386 err = -EPERM;
1387 if (!unix_may_send(sk, other))
1388 goto out_unlock;
1390 if (sock_flag(other, SOCK_DEAD)) {
1392 * Check with 1003.1g - what should
1393 * datagram error
1395 unix_state_unlock(other);
1396 sock_put(other);
1398 err = 0;
1399 unix_state_lock(sk);
1400 if (unix_peer(sk) == other) {
1401 unix_peer(sk)=NULL;
1402 unix_state_unlock(sk);
1404 unix_dgram_disconnected(sk, other);
1405 sock_put(other);
1406 err = -ECONNREFUSED;
1407 } else {
1408 unix_state_unlock(sk);
1411 other = NULL;
1412 if (err)
1413 goto out_free;
1414 goto restart;
1417 err = -EPIPE;
1418 if (other->sk_shutdown & RCV_SHUTDOWN)
1419 goto out_unlock;
1421 if (sk->sk_type != SOCK_SEQPACKET) {
1422 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1423 if (err)
1424 goto out_unlock;
1427 if (unix_peer(other) != sk &&
1428 (skb_queue_len(&other->sk_receive_queue) >
1429 other->sk_max_ack_backlog)) {
1430 if (!timeo) {
1431 err = -EAGAIN;
1432 goto out_unlock;
1435 timeo = unix_wait_for_peer(other, timeo);
1437 err = sock_intr_errno(timeo);
1438 if (signal_pending(current))
1439 goto out_free;
1441 goto restart;
1444 skb_queue_tail(&other->sk_receive_queue, skb);
1445 unix_state_unlock(other);
1446 other->sk_data_ready(other, len);
1447 sock_put(other);
1448 scm_destroy(siocb->scm);
1449 return len;
1451 out_unlock:
1452 unix_state_unlock(other);
1453 out_free:
1454 kfree_skb(skb);
1455 out:
1456 if (other)
1457 sock_put(other);
1458 scm_destroy(siocb->scm);
1459 return err;
1463 static int unix_stream_sendmsg(struct kiocb *kiocb, struct socket *sock,
1464 struct msghdr *msg, size_t len)
1466 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1467 struct sock *sk = sock->sk;
1468 struct sock *other = NULL;
1469 struct sockaddr_un *sunaddr=msg->msg_name;
1470 int err,size;
1471 struct sk_buff *skb;
1472 int sent=0;
1473 struct scm_cookie tmp_scm;
1475 if (NULL == siocb->scm)
1476 siocb->scm = &tmp_scm;
1477 err = scm_send(sock, msg, siocb->scm);
1478 if (err < 0)
1479 return err;
1481 err = -EOPNOTSUPP;
1482 if (msg->msg_flags&MSG_OOB)
1483 goto out_err;
1485 if (msg->msg_namelen) {
1486 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
1487 goto out_err;
1488 } else {
1489 sunaddr = NULL;
1490 err = -ENOTCONN;
1491 other = unix_peer(sk);
1492 if (!other)
1493 goto out_err;
1496 if (sk->sk_shutdown & SEND_SHUTDOWN)
1497 goto pipe_err;
1499 while(sent < len)
1502 * Optimisation for the fact that under 0.01% of X
1503 * messages typically need breaking up.
1506 size = len-sent;
1508 /* Keep two messages in the pipe so it schedules better */
1509 if (size > ((sk->sk_sndbuf >> 1) - 64))
1510 size = (sk->sk_sndbuf >> 1) - 64;
1512 if (size > SKB_MAX_ALLOC)
1513 size = SKB_MAX_ALLOC;
1516 * Grab a buffer
1519 skb=sock_alloc_send_skb(sk,size,msg->msg_flags&MSG_DONTWAIT, &err);
1521 if (skb==NULL)
1522 goto out_err;
1525 * If you pass two values to the sock_alloc_send_skb
1526 * it tries to grab the large buffer with GFP_NOFS
1527 * (which can fail easily), and if it fails grab the
1528 * fallback size buffer which is under a page and will
1529 * succeed. [Alan]
1531 size = min_t(int, size, skb_tailroom(skb));
1533 memcpy(UNIXCREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1534 if (siocb->scm->fp)
1535 unix_attach_fds(siocb->scm, skb);
1537 if ((err = memcpy_fromiovec(skb_put(skb,size), msg->msg_iov, size)) != 0) {
1538 kfree_skb(skb);
1539 goto out_err;
1542 unix_state_lock(other);
1544 if (sock_flag(other, SOCK_DEAD) ||
1545 (other->sk_shutdown & RCV_SHUTDOWN))
1546 goto pipe_err_free;
1548 skb_queue_tail(&other->sk_receive_queue, skb);
1549 unix_state_unlock(other);
1550 other->sk_data_ready(other, size);
1551 sent+=size;
1554 scm_destroy(siocb->scm);
1555 siocb->scm = NULL;
1557 return sent;
1559 pipe_err_free:
1560 unix_state_unlock(other);
1561 kfree_skb(skb);
1562 pipe_err:
1563 if (sent==0 && !(msg->msg_flags&MSG_NOSIGNAL))
1564 send_sig(SIGPIPE,current,0);
1565 err = -EPIPE;
1566 out_err:
1567 scm_destroy(siocb->scm);
1568 siocb->scm = NULL;
1569 return sent ? : err;
1572 static int unix_seqpacket_sendmsg(struct kiocb *kiocb, struct socket *sock,
1573 struct msghdr *msg, size_t len)
1575 int err;
1576 struct sock *sk = sock->sk;
1578 err = sock_error(sk);
1579 if (err)
1580 return err;
1582 if (sk->sk_state != TCP_ESTABLISHED)
1583 return -ENOTCONN;
1585 if (msg->msg_namelen)
1586 msg->msg_namelen = 0;
1588 return unix_dgram_sendmsg(kiocb, sock, msg, len);
1591 static void unix_copy_addr(struct msghdr *msg, struct sock *sk)
1593 struct unix_sock *u = unix_sk(sk);
1595 msg->msg_namelen = 0;
1596 if (u->addr) {
1597 msg->msg_namelen = u->addr->len;
1598 memcpy(msg->msg_name, u->addr->name, u->addr->len);
1602 static int unix_dgram_recvmsg(struct kiocb *iocb, struct socket *sock,
1603 struct msghdr *msg, size_t size,
1604 int flags)
1606 struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1607 struct scm_cookie tmp_scm;
1608 struct sock *sk = sock->sk;
1609 struct unix_sock *u = unix_sk(sk);
1610 int noblock = flags & MSG_DONTWAIT;
1611 struct sk_buff *skb;
1612 int err;
1614 err = -EOPNOTSUPP;
1615 if (flags&MSG_OOB)
1616 goto out;
1618 msg->msg_namelen = 0;
1620 mutex_lock(&u->readlock);
1622 skb = skb_recv_datagram(sk, flags, noblock, &err);
1623 if (!skb) {
1624 unix_state_lock(sk);
1625 /* Signal EOF on disconnected non-blocking SEQPACKET socket. */
1626 if (sk->sk_type == SOCK_SEQPACKET && err == -EAGAIN &&
1627 (sk->sk_shutdown & RCV_SHUTDOWN))
1628 err = 0;
1629 unix_state_unlock(sk);
1630 goto out_unlock;
1633 wake_up_interruptible_sync(&u->peer_wait);
1635 if (msg->msg_name)
1636 unix_copy_addr(msg, skb->sk);
1638 if (size > skb->len)
1639 size = skb->len;
1640 else if (size < skb->len)
1641 msg->msg_flags |= MSG_TRUNC;
1643 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, size);
1644 if (err)
1645 goto out_free;
1647 if (!siocb->scm) {
1648 siocb->scm = &tmp_scm;
1649 memset(&tmp_scm, 0, sizeof(tmp_scm));
1651 siocb->scm->creds = *UNIXCREDS(skb);
1652 unix_set_secdata(siocb->scm, skb);
1654 if (!(flags & MSG_PEEK))
1656 if (UNIXCB(skb).fp)
1657 unix_detach_fds(siocb->scm, skb);
1659 else
1661 /* It is questionable: on PEEK we could:
1662 - do not return fds - good, but too simple 8)
1663 - return fds, and do not return them on read (old strategy,
1664 apparently wrong)
1665 - clone fds (I chose it for now, it is the most universal
1666 solution)
1668 POSIX 1003.1g does not actually define this clearly
1669 at all. POSIX 1003.1g doesn't define a lot of things
1670 clearly however!
1673 if (UNIXCB(skb).fp)
1674 siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1676 err = size;
1678 scm_recv(sock, msg, siocb->scm, flags);
1680 out_free:
1681 skb_free_datagram(sk,skb);
1682 out_unlock:
1683 mutex_unlock(&u->readlock);
1684 out:
1685 return err;
1689 * Sleep until data has arrive. But check for races..
1692 static long unix_stream_data_wait(struct sock * sk, long timeo)
1694 DEFINE_WAIT(wait);
1696 unix_state_lock(sk);
1698 for (;;) {
1699 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1701 if (!skb_queue_empty(&sk->sk_receive_queue) ||
1702 sk->sk_err ||
1703 (sk->sk_shutdown & RCV_SHUTDOWN) ||
1704 signal_pending(current) ||
1705 !timeo)
1706 break;
1708 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1709 unix_state_unlock(sk);
1710 timeo = schedule_timeout(timeo);
1711 unix_state_lock(sk);
1712 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1715 finish_wait(sk->sk_sleep, &wait);
1716 unix_state_unlock(sk);
1717 return timeo;
1722 static int unix_stream_recvmsg(struct kiocb *iocb, struct socket *sock,
1723 struct msghdr *msg, size_t size,
1724 int flags)
1726 struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1727 struct scm_cookie tmp_scm;
1728 struct sock *sk = sock->sk;
1729 struct unix_sock *u = unix_sk(sk);
1730 struct sockaddr_un *sunaddr=msg->msg_name;
1731 int copied = 0;
1732 int check_creds = 0;
1733 int target;
1734 int err = 0;
1735 long timeo;
1737 err = -EINVAL;
1738 if (sk->sk_state != TCP_ESTABLISHED)
1739 goto out;
1741 err = -EOPNOTSUPP;
1742 if (flags&MSG_OOB)
1743 goto out;
1745 target = sock_rcvlowat(sk, flags&MSG_WAITALL, size);
1746 timeo = sock_rcvtimeo(sk, flags&MSG_DONTWAIT);
1748 msg->msg_namelen = 0;
1750 /* Lock the socket to prevent queue disordering
1751 * while sleeps in memcpy_tomsg
1754 if (!siocb->scm) {
1755 siocb->scm = &tmp_scm;
1756 memset(&tmp_scm, 0, sizeof(tmp_scm));
1759 mutex_lock(&u->readlock);
1763 int chunk;
1764 struct sk_buff *skb;
1766 unix_state_lock(sk);
1767 skb = skb_dequeue(&sk->sk_receive_queue);
1768 if (skb==NULL)
1770 if (copied >= target)
1771 goto unlock;
1774 * POSIX 1003.1g mandates this order.
1777 if ((err = sock_error(sk)) != 0)
1778 goto unlock;
1779 if (sk->sk_shutdown & RCV_SHUTDOWN)
1780 goto unlock;
1782 unix_state_unlock(sk);
1783 err = -EAGAIN;
1784 if (!timeo)
1785 break;
1786 mutex_unlock(&u->readlock);
1788 timeo = unix_stream_data_wait(sk, timeo);
1790 if (signal_pending(current)) {
1791 err = sock_intr_errno(timeo);
1792 goto out;
1794 mutex_lock(&u->readlock);
1795 continue;
1796 unlock:
1797 unix_state_unlock(sk);
1798 break;
1800 unix_state_unlock(sk);
1802 if (check_creds) {
1803 /* Never glue messages from different writers */
1804 if (memcmp(UNIXCREDS(skb), &siocb->scm->creds, sizeof(siocb->scm->creds)) != 0) {
1805 skb_queue_head(&sk->sk_receive_queue, skb);
1806 break;
1808 } else {
1809 /* Copy credentials */
1810 siocb->scm->creds = *UNIXCREDS(skb);
1811 check_creds = 1;
1814 /* Copy address just once */
1815 if (sunaddr)
1817 unix_copy_addr(msg, skb->sk);
1818 sunaddr = NULL;
1821 chunk = min_t(unsigned int, skb->len, size);
1822 if (memcpy_toiovec(msg->msg_iov, skb->data, chunk)) {
1823 skb_queue_head(&sk->sk_receive_queue, skb);
1824 if (copied == 0)
1825 copied = -EFAULT;
1826 break;
1828 copied += chunk;
1829 size -= chunk;
1831 /* Mark read part of skb as used */
1832 if (!(flags & MSG_PEEK))
1834 skb_pull(skb, chunk);
1836 if (UNIXCB(skb).fp)
1837 unix_detach_fds(siocb->scm, skb);
1839 /* put the skb back if we didn't use it up.. */
1840 if (skb->len)
1842 skb_queue_head(&sk->sk_receive_queue, skb);
1843 break;
1846 kfree_skb(skb);
1848 if (siocb->scm->fp)
1849 break;
1851 else
1853 /* It is questionable, see note in unix_dgram_recvmsg.
1855 if (UNIXCB(skb).fp)
1856 siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1858 /* put message back and return */
1859 skb_queue_head(&sk->sk_receive_queue, skb);
1860 break;
1862 } while (size);
1864 mutex_unlock(&u->readlock);
1865 scm_recv(sock, msg, siocb->scm, flags);
1866 out:
1867 return copied ? : err;
1870 static int unix_shutdown(struct socket *sock, int mode)
1872 struct sock *sk = sock->sk;
1873 struct sock *other;
1875 mode = (mode+1)&(RCV_SHUTDOWN|SEND_SHUTDOWN);
1877 if (mode) {
1878 unix_state_lock(sk);
1879 sk->sk_shutdown |= mode;
1880 other=unix_peer(sk);
1881 if (other)
1882 sock_hold(other);
1883 unix_state_unlock(sk);
1884 sk->sk_state_change(sk);
1886 if (other &&
1887 (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET)) {
1889 int peer_mode = 0;
1891 if (mode&RCV_SHUTDOWN)
1892 peer_mode |= SEND_SHUTDOWN;
1893 if (mode&SEND_SHUTDOWN)
1894 peer_mode |= RCV_SHUTDOWN;
1895 unix_state_lock(other);
1896 other->sk_shutdown |= peer_mode;
1897 unix_state_unlock(other);
1898 other->sk_state_change(other);
1899 read_lock(&other->sk_callback_lock);
1900 if (peer_mode == SHUTDOWN_MASK)
1901 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_HUP);
1902 else if (peer_mode & RCV_SHUTDOWN)
1903 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_IN);
1904 read_unlock(&other->sk_callback_lock);
1906 if (other)
1907 sock_put(other);
1909 return 0;
1912 static int unix_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1914 struct sock *sk = sock->sk;
1915 long amount=0;
1916 int err;
1918 switch(cmd)
1920 case SIOCOUTQ:
1921 amount = atomic_read(&sk->sk_wmem_alloc);
1922 err = put_user(amount, (int __user *)arg);
1923 break;
1924 case SIOCINQ:
1926 struct sk_buff *skb;
1928 if (sk->sk_state == TCP_LISTEN) {
1929 err = -EINVAL;
1930 break;
1933 spin_lock(&sk->sk_receive_queue.lock);
1934 if (sk->sk_type == SOCK_STREAM ||
1935 sk->sk_type == SOCK_SEQPACKET) {
1936 skb_queue_walk(&sk->sk_receive_queue, skb)
1937 amount += skb->len;
1938 } else {
1939 skb = skb_peek(&sk->sk_receive_queue);
1940 if (skb)
1941 amount=skb->len;
1943 spin_unlock(&sk->sk_receive_queue.lock);
1944 err = put_user(amount, (int __user *)arg);
1945 break;
1948 default:
1949 err = -ENOIOCTLCMD;
1950 break;
1952 return err;
1955 static unsigned int unix_poll(struct file * file, struct socket *sock, poll_table *wait)
1957 struct sock *sk = sock->sk;
1958 unsigned int mask;
1960 poll_wait(file, sk->sk_sleep, wait);
1961 mask = 0;
1963 /* exceptional events? */
1964 if (sk->sk_err)
1965 mask |= POLLERR;
1966 if (sk->sk_shutdown == SHUTDOWN_MASK)
1967 mask |= POLLHUP;
1968 if (sk->sk_shutdown & RCV_SHUTDOWN)
1969 mask |= POLLRDHUP;
1971 /* readable? */
1972 if (!skb_queue_empty(&sk->sk_receive_queue) ||
1973 (sk->sk_shutdown & RCV_SHUTDOWN))
1974 mask |= POLLIN | POLLRDNORM;
1976 /* Connection-based need to check for termination and startup */
1977 if ((sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) && sk->sk_state == TCP_CLOSE)
1978 mask |= POLLHUP;
1981 * we set writable also when the other side has shut down the
1982 * connection. This prevents stuck sockets.
1984 if (unix_writable(sk))
1985 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
1987 return mask;
1991 #ifdef CONFIG_PROC_FS
1992 static struct sock *first_unix_socket(int *i)
1994 for (*i = 0; *i <= UNIX_HASH_SIZE; (*i)++) {
1995 if (!hlist_empty(&unix_socket_table[*i]))
1996 return __sk_head(&unix_socket_table[*i]);
1998 return NULL;
2001 static struct sock *next_unix_socket(int *i, struct sock *s)
2003 struct sock *next = sk_next(s);
2004 /* More in this chain? */
2005 if (next)
2006 return next;
2007 /* Look for next non-empty chain. */
2008 for ((*i)++; *i <= UNIX_HASH_SIZE; (*i)++) {
2009 if (!hlist_empty(&unix_socket_table[*i]))
2010 return __sk_head(&unix_socket_table[*i]);
2012 return NULL;
2015 struct unix_iter_state {
2016 struct seq_net_private p;
2017 int i;
2019 static struct sock *unix_seq_idx(struct unix_iter_state *iter, loff_t pos)
2021 loff_t off = 0;
2022 struct sock *s;
2024 for (s = first_unix_socket(&iter->i); s; s = next_unix_socket(&iter->i, s)) {
2025 if (s->sk_net != iter->p.net)
2026 continue;
2027 if (off == pos)
2028 return s;
2029 ++off;
2031 return NULL;
2035 static void *unix_seq_start(struct seq_file *seq, loff_t *pos)
2036 __acquires(unix_table_lock)
2038 struct unix_iter_state *iter = seq->private;
2039 spin_lock(&unix_table_lock);
2040 return *pos ? unix_seq_idx(iter, *pos - 1) : ((void *) 1);
2043 static void *unix_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2045 struct unix_iter_state *iter = seq->private;
2046 struct sock *sk = v;
2047 ++*pos;
2049 if (v == (void *)1)
2050 sk = first_unix_socket(&iter->i);
2051 else
2052 sk = next_unix_socket(&iter->i, sk);
2053 while (sk && (sk->sk_net != iter->p.net))
2054 sk = next_unix_socket(&iter->i, sk);
2055 return sk;
2058 static void unix_seq_stop(struct seq_file *seq, void *v)
2059 __releases(unix_table_lock)
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 net->unx.sysctl_max_dgram_qlen = 10;
2144 if (unix_sysctl_register(net))
2145 goto out;
2147 #ifdef CONFIG_PROC_FS
2148 if (!proc_net_fops_create(net, "unix", 0, &unix_seq_fops)) {
2149 unix_sysctl_unregister(net);
2150 goto out;
2152 #endif
2153 error = 0;
2154 out:
2155 return 0;
2158 static void unix_net_exit(struct net *net)
2160 unix_sysctl_unregister(net);
2161 proc_net_remove(net, "unix");
2164 static struct pernet_operations unix_net_ops = {
2165 .init = unix_net_init,
2166 .exit = unix_net_exit,
2169 static int __init af_unix_init(void)
2171 int rc = -1;
2172 struct sk_buff *dummy_skb;
2174 BUILD_BUG_ON(sizeof(struct unix_skb_parms) > sizeof(dummy_skb->cb));
2176 rc = proto_register(&unix_proto, 1);
2177 if (rc != 0) {
2178 printk(KERN_CRIT "%s: Cannot create unix_sock SLAB cache!\n",
2179 __FUNCTION__);
2180 goto out;
2183 sock_register(&unix_family_ops);
2184 register_pernet_subsys(&unix_net_ops);
2185 out:
2186 return rc;
2189 static void __exit af_unix_exit(void)
2191 sock_unregister(PF_UNIX);
2192 proto_unregister(&unix_proto);
2193 unregister_pernet_subsys(&unix_net_ops);
2196 module_init(af_unix_init);
2197 module_exit(af_unix_exit);
2199 MODULE_LICENSE("GPL");
2200 MODULE_ALIAS_NETPROTO(PF_UNIX);