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 $
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
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
31 * Kirk Petersen : Made this a module
32 * Christoph Rohland : Elegant non-blocking accept/connect algorithm.
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
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:
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).
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
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>
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>
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 static struct sock *first_unix_socket(int *i)
132 for (*i = 0; *i <= UNIX_HASH_SIZE; (*i)++) {
133 if (!hlist_empty(&unix_socket_table[*i]))
134 return __sk_head(&unix_socket_table[*i]);
139 static struct sock *next_unix_socket(int *i, struct sock *s)
141 struct sock *next = sk_next(s);
142 /* More in this chain? */
145 /* Look for next non-empty chain. */
146 for ((*i)++; *i <= UNIX_HASH_SIZE; (*i)++) {
147 if (!hlist_empty(&unix_socket_table[*i]))
148 return __sk_head(&unix_socket_table[*i]);
153 #define forall_unix_sockets(i, s) \
154 for (s = first_unix_socket(&(i)); s; s = next_unix_socket(&(i),(s)))
156 #ifdef CONFIG_SECURITY_NETWORK
157 static void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb)
159 memcpy(UNIXSID(skb), &scm->secid, sizeof(u32));
162 static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb)
164 scm->secid = *UNIXSID(skb);
167 static inline void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb)
170 static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb)
172 #endif /* CONFIG_SECURITY_NETWORK */
175 * SMP locking strategy:
176 * hash table is protected with spinlock unix_table_lock
177 * each socket state is protected by separate rwlock.
180 static inline unsigned unix_hash_fold(__wsum n)
182 unsigned hash = (__force unsigned)n;
185 return hash&(UNIX_HASH_SIZE-1);
188 #define unix_peer(sk) (unix_sk(sk)->peer)
190 static inline int unix_our_peer(struct sock *sk, struct sock *osk)
192 return unix_peer(osk) == sk;
195 static inline int unix_may_send(struct sock *sk, struct sock *osk)
197 return (unix_peer(osk) == NULL || unix_our_peer(sk, osk));
200 static struct sock *unix_peer_get(struct sock *s)
208 unix_state_unlock(s);
212 static inline void unix_release_addr(struct unix_address *addr)
214 if (atomic_dec_and_test(&addr->refcnt))
219 * Check unix socket name:
220 * - should be not zero length.
221 * - if started by not zero, should be NULL terminated (FS object)
222 * - if started by zero, it is abstract name.
225 static int unix_mkname(struct sockaddr_un * sunaddr, int len, unsigned *hashp)
227 if (len <= sizeof(short) || len > sizeof(*sunaddr))
229 if (!sunaddr || sunaddr->sun_family != AF_UNIX)
231 if (sunaddr->sun_path[0]) {
233 * This may look like an off by one error but it is a bit more
234 * subtle. 108 is the longest valid AF_UNIX path for a binding.
235 * sun_path[108] doesnt as such exist. However in kernel space
236 * we are guaranteed that it is a valid memory location in our
237 * kernel address buffer.
239 ((char *)sunaddr)[len]=0;
240 len = strlen(sunaddr->sun_path)+1+sizeof(short);
244 *hashp = unix_hash_fold(csum_partial((char*)sunaddr, len, 0));
248 static void __unix_remove_socket(struct sock *sk)
250 sk_del_node_init(sk);
253 static void __unix_insert_socket(struct hlist_head *list, struct sock *sk)
255 BUG_TRAP(sk_unhashed(sk));
256 sk_add_node(sk, list);
259 static inline void unix_remove_socket(struct sock *sk)
261 spin_lock(&unix_table_lock);
262 __unix_remove_socket(sk);
263 spin_unlock(&unix_table_lock);
266 static inline void unix_insert_socket(struct hlist_head *list, struct sock *sk)
268 spin_lock(&unix_table_lock);
269 __unix_insert_socket(list, sk);
270 spin_unlock(&unix_table_lock);
273 static struct sock *__unix_find_socket_byname(struct net *net,
274 struct sockaddr_un *sunname,
275 int len, int type, unsigned hash)
278 struct hlist_node *node;
280 sk_for_each(s, node, &unix_socket_table[hash ^ type]) {
281 struct unix_sock *u = unix_sk(s);
283 if (s->sk_net != net)
286 if (u->addr->len == len &&
287 !memcmp(u->addr->name, sunname, len))
295 static inline struct sock *unix_find_socket_byname(struct net *net,
296 struct sockaddr_un *sunname,
302 spin_lock(&unix_table_lock);
303 s = __unix_find_socket_byname(net, sunname, len, type, hash);
306 spin_unlock(&unix_table_lock);
310 static struct sock *unix_find_socket_byinode(struct net *net, struct inode *i)
313 struct hlist_node *node;
315 spin_lock(&unix_table_lock);
317 &unix_socket_table[i->i_ino & (UNIX_HASH_SIZE - 1)]) {
318 struct dentry *dentry = unix_sk(s)->dentry;
320 if (s->sk_net != net)
323 if(dentry && dentry->d_inode == i)
331 spin_unlock(&unix_table_lock);
335 static inline int unix_writable(struct sock *sk)
337 return (atomic_read(&sk->sk_wmem_alloc) << 2) <= sk->sk_sndbuf;
340 static void unix_write_space(struct sock *sk)
342 read_lock(&sk->sk_callback_lock);
343 if (unix_writable(sk)) {
344 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
345 wake_up_interruptible_sync(sk->sk_sleep);
346 sk_wake_async(sk, 2, POLL_OUT);
348 read_unlock(&sk->sk_callback_lock);
351 /* When dgram socket disconnects (or changes its peer), we clear its receive
352 * queue of packets arrived from previous peer. First, it allows to do
353 * flow control based only on wmem_alloc; second, sk connected to peer
354 * may receive messages only from that peer. */
355 static void unix_dgram_disconnected(struct sock *sk, struct sock *other)
357 if (!skb_queue_empty(&sk->sk_receive_queue)) {
358 skb_queue_purge(&sk->sk_receive_queue);
359 wake_up_interruptible_all(&unix_sk(sk)->peer_wait);
361 /* If one link of bidirectional dgram pipe is disconnected,
362 * we signal error. Messages are lost. Do not make this,
363 * when peer was not connected to us.
365 if (!sock_flag(other, SOCK_DEAD) && unix_peer(other) == sk) {
366 other->sk_err = ECONNRESET;
367 other->sk_error_report(other);
372 static void unix_sock_destructor(struct sock *sk)
374 struct unix_sock *u = unix_sk(sk);
376 skb_queue_purge(&sk->sk_receive_queue);
378 BUG_TRAP(!atomic_read(&sk->sk_wmem_alloc));
379 BUG_TRAP(sk_unhashed(sk));
380 BUG_TRAP(!sk->sk_socket);
381 if (!sock_flag(sk, SOCK_DEAD)) {
382 printk("Attempt to release alive unix socket: %p\n", sk);
387 unix_release_addr(u->addr);
389 atomic_dec(&unix_nr_socks);
390 #ifdef UNIX_REFCNT_DEBUG
391 printk(KERN_DEBUG "UNIX %p is destroyed, %d are still alive.\n", sk, atomic_read(&unix_nr_socks));
395 static int unix_release_sock (struct sock *sk, int embrion)
397 struct unix_sock *u = unix_sk(sk);
398 struct dentry *dentry;
399 struct vfsmount *mnt;
404 unix_remove_socket(sk);
409 sk->sk_shutdown = SHUTDOWN_MASK;
414 state = sk->sk_state;
415 sk->sk_state = TCP_CLOSE;
416 unix_state_unlock(sk);
418 wake_up_interruptible_all(&u->peer_wait);
420 skpair=unix_peer(sk);
423 if (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) {
424 unix_state_lock(skpair);
426 skpair->sk_shutdown = SHUTDOWN_MASK;
427 if (!skb_queue_empty(&sk->sk_receive_queue) || embrion)
428 skpair->sk_err = ECONNRESET;
429 unix_state_unlock(skpair);
430 skpair->sk_state_change(skpair);
431 read_lock(&skpair->sk_callback_lock);
432 sk_wake_async(skpair,1,POLL_HUP);
433 read_unlock(&skpair->sk_callback_lock);
435 sock_put(skpair); /* It may now die */
436 unix_peer(sk) = NULL;
439 /* Try to flush out this socket. Throw out buffers at least */
441 while ((skb = skb_dequeue(&sk->sk_receive_queue)) != NULL) {
442 if (state==TCP_LISTEN)
443 unix_release_sock(skb->sk, 1);
444 /* passed fds are erased in the kfree_skb hook */
455 /* ---- Socket is dead now and most probably destroyed ---- */
458 * Fixme: BSD difference: In BSD all sockets connected to use get
459 * ECONNRESET and we die on the spot. In Linux we behave
460 * like files and pipes do and wait for the last
463 * Can't we simply set sock->err?
465 * What the above comment does talk about? --ANK(980817)
468 if (unix_tot_inflight)
469 unix_gc(); /* Garbage collect fds */
474 static int unix_listen(struct socket *sock, int backlog)
477 struct sock *sk = sock->sk;
478 struct unix_sock *u = unix_sk(sk);
481 if (sock->type!=SOCK_STREAM && sock->type!=SOCK_SEQPACKET)
482 goto out; /* Only stream/seqpacket sockets accept */
485 goto out; /* No listens on an unbound socket */
487 if (sk->sk_state != TCP_CLOSE && sk->sk_state != TCP_LISTEN)
489 if (backlog > sk->sk_max_ack_backlog)
490 wake_up_interruptible_all(&u->peer_wait);
491 sk->sk_max_ack_backlog = backlog;
492 sk->sk_state = TCP_LISTEN;
493 /* set credentials so connect can copy them */
494 sk->sk_peercred.pid = task_tgid_vnr(current);
495 sk->sk_peercred.uid = current->euid;
496 sk->sk_peercred.gid = current->egid;
500 unix_state_unlock(sk);
505 static int unix_release(struct socket *);
506 static int unix_bind(struct socket *, struct sockaddr *, int);
507 static int unix_stream_connect(struct socket *, struct sockaddr *,
508 int addr_len, int flags);
509 static int unix_socketpair(struct socket *, struct socket *);
510 static int unix_accept(struct socket *, struct socket *, int);
511 static int unix_getname(struct socket *, struct sockaddr *, int *, int);
512 static unsigned int unix_poll(struct file *, struct socket *, poll_table *);
513 static int unix_ioctl(struct socket *, unsigned int, unsigned long);
514 static int unix_shutdown(struct socket *, int);
515 static int unix_stream_sendmsg(struct kiocb *, struct socket *,
516 struct msghdr *, size_t);
517 static int unix_stream_recvmsg(struct kiocb *, struct socket *,
518 struct msghdr *, size_t, int);
519 static int unix_dgram_sendmsg(struct kiocb *, struct socket *,
520 struct msghdr *, size_t);
521 static int unix_dgram_recvmsg(struct kiocb *, struct socket *,
522 struct msghdr *, size_t, int);
523 static int unix_dgram_connect(struct socket *, struct sockaddr *,
525 static int unix_seqpacket_sendmsg(struct kiocb *, struct socket *,
526 struct msghdr *, size_t);
528 static const struct proto_ops unix_stream_ops = {
530 .owner = THIS_MODULE,
531 .release = unix_release,
533 .connect = unix_stream_connect,
534 .socketpair = unix_socketpair,
535 .accept = unix_accept,
536 .getname = unix_getname,
539 .listen = unix_listen,
540 .shutdown = unix_shutdown,
541 .setsockopt = sock_no_setsockopt,
542 .getsockopt = sock_no_getsockopt,
543 .sendmsg = unix_stream_sendmsg,
544 .recvmsg = unix_stream_recvmsg,
545 .mmap = sock_no_mmap,
546 .sendpage = sock_no_sendpage,
549 static const struct proto_ops unix_dgram_ops = {
551 .owner = THIS_MODULE,
552 .release = unix_release,
554 .connect = unix_dgram_connect,
555 .socketpair = unix_socketpair,
556 .accept = sock_no_accept,
557 .getname = unix_getname,
558 .poll = datagram_poll,
560 .listen = sock_no_listen,
561 .shutdown = unix_shutdown,
562 .setsockopt = sock_no_setsockopt,
563 .getsockopt = sock_no_getsockopt,
564 .sendmsg = unix_dgram_sendmsg,
565 .recvmsg = unix_dgram_recvmsg,
566 .mmap = sock_no_mmap,
567 .sendpage = sock_no_sendpage,
570 static const struct proto_ops unix_seqpacket_ops = {
572 .owner = THIS_MODULE,
573 .release = unix_release,
575 .connect = unix_stream_connect,
576 .socketpair = unix_socketpair,
577 .accept = unix_accept,
578 .getname = unix_getname,
579 .poll = datagram_poll,
581 .listen = unix_listen,
582 .shutdown = unix_shutdown,
583 .setsockopt = sock_no_setsockopt,
584 .getsockopt = sock_no_getsockopt,
585 .sendmsg = unix_seqpacket_sendmsg,
586 .recvmsg = unix_dgram_recvmsg,
587 .mmap = sock_no_mmap,
588 .sendpage = sock_no_sendpage,
591 static struct proto unix_proto = {
593 .owner = THIS_MODULE,
594 .obj_size = sizeof(struct unix_sock),
598 * AF_UNIX sockets do not interact with hardware, hence they
599 * dont trigger interrupts - so it's safe for them to have
600 * bh-unsafe locking for their sk_receive_queue.lock. Split off
601 * this special lock-class by reinitializing the spinlock key:
603 static struct lock_class_key af_unix_sk_receive_queue_lock_key;
605 static struct sock * unix_create1(struct net *net, struct socket *sock)
607 struct sock *sk = NULL;
610 atomic_inc(&unix_nr_socks);
611 if (atomic_read(&unix_nr_socks) > 2 * get_max_files())
614 sk = sk_alloc(net, PF_UNIX, GFP_KERNEL, &unix_proto);
618 sock_init_data(sock,sk);
619 lockdep_set_class(&sk->sk_receive_queue.lock,
620 &af_unix_sk_receive_queue_lock_key);
622 sk->sk_write_space = unix_write_space;
623 sk->sk_max_ack_backlog = sysctl_unix_max_dgram_qlen;
624 sk->sk_destruct = unix_sock_destructor;
628 spin_lock_init(&u->lock);
629 atomic_set(&u->inflight, 0);
630 INIT_LIST_HEAD(&u->link);
631 mutex_init(&u->readlock); /* single task reading lock */
632 init_waitqueue_head(&u->peer_wait);
633 unix_insert_socket(unix_sockets_unbound, sk);
636 atomic_dec(&unix_nr_socks);
640 static int unix_create(struct net *net, struct socket *sock, int protocol)
642 if (protocol && protocol != PF_UNIX)
643 return -EPROTONOSUPPORT;
645 sock->state = SS_UNCONNECTED;
647 switch (sock->type) {
649 sock->ops = &unix_stream_ops;
652 * Believe it or not BSD has AF_UNIX, SOCK_RAW though
656 sock->type=SOCK_DGRAM;
658 sock->ops = &unix_dgram_ops;
661 sock->ops = &unix_seqpacket_ops;
664 return -ESOCKTNOSUPPORT;
667 return unix_create1(net, sock) ? 0 : -ENOMEM;
670 static int unix_release(struct socket *sock)
672 struct sock *sk = sock->sk;
679 return unix_release_sock (sk, 0);
682 static int unix_autobind(struct socket *sock)
684 struct sock *sk = sock->sk;
685 struct net *net = sk->sk_net;
686 struct unix_sock *u = unix_sk(sk);
687 static u32 ordernum = 1;
688 struct unix_address * addr;
691 mutex_lock(&u->readlock);
698 addr = kzalloc(sizeof(*addr) + sizeof(short) + 16, GFP_KERNEL);
702 addr->name->sun_family = AF_UNIX;
703 atomic_set(&addr->refcnt, 1);
706 addr->len = sprintf(addr->name->sun_path+1, "%05x", ordernum) + 1 + sizeof(short);
707 addr->hash = unix_hash_fold(csum_partial((void*)addr->name, addr->len, 0));
709 spin_lock(&unix_table_lock);
710 ordernum = (ordernum+1)&0xFFFFF;
712 if (__unix_find_socket_byname(net, addr->name, addr->len, sock->type,
714 spin_unlock(&unix_table_lock);
715 /* Sanity yield. It is unusual case, but yet... */
716 if (!(ordernum&0xFF))
720 addr->hash ^= sk->sk_type;
722 __unix_remove_socket(sk);
724 __unix_insert_socket(&unix_socket_table[addr->hash], sk);
725 spin_unlock(&unix_table_lock);
728 out: mutex_unlock(&u->readlock);
732 static struct sock *unix_find_other(struct net *net,
733 struct sockaddr_un *sunname, int len,
734 int type, unsigned hash, int *error)
740 if (sunname->sun_path[0]) {
741 err = path_lookup(sunname->sun_path, LOOKUP_FOLLOW, &nd);
744 err = vfs_permission(&nd, MAY_WRITE);
749 if (!S_ISSOCK(nd.dentry->d_inode->i_mode))
751 u=unix_find_socket_byinode(net, nd.dentry->d_inode);
755 if (u->sk_type == type)
756 touch_atime(nd.mnt, nd.dentry);
761 if (u->sk_type != type) {
767 u=unix_find_socket_byname(net, sunname, len, type, hash);
769 struct dentry *dentry;
770 dentry = unix_sk(u)->dentry;
772 touch_atime(unix_sk(u)->mnt, dentry);
786 static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
788 struct sock *sk = sock->sk;
789 struct net *net = sk->sk_net;
790 struct unix_sock *u = unix_sk(sk);
791 struct sockaddr_un *sunaddr=(struct sockaddr_un *)uaddr;
792 struct dentry * dentry = NULL;
796 struct unix_address *addr;
797 struct hlist_head *list;
800 if (sunaddr->sun_family != AF_UNIX)
803 if (addr_len==sizeof(short)) {
804 err = unix_autobind(sock);
808 err = unix_mkname(sunaddr, addr_len, &hash);
813 mutex_lock(&u->readlock);
820 addr = kmalloc(sizeof(*addr)+addr_len, GFP_KERNEL);
824 memcpy(addr->name, sunaddr, addr_len);
825 addr->len = addr_len;
826 addr->hash = hash ^ sk->sk_type;
827 atomic_set(&addr->refcnt, 1);
829 if (sunaddr->sun_path[0]) {
833 * Get the parent directory, calculate the hash for last
836 err = path_lookup(sunaddr->sun_path, LOOKUP_PARENT, &nd);
838 goto out_mknod_parent;
840 dentry = lookup_create(&nd, 0);
841 err = PTR_ERR(dentry);
843 goto out_mknod_unlock;
846 * All right, let's create it.
849 (SOCK_INODE(sock)->i_mode & ~current->fs->umask);
850 err = vfs_mknod(nd.dentry->d_inode, dentry, mode, 0);
853 mutex_unlock(&nd.dentry->d_inode->i_mutex);
857 addr->hash = UNIX_HASH_SIZE;
860 spin_lock(&unix_table_lock);
862 if (!sunaddr->sun_path[0]) {
864 if (__unix_find_socket_byname(net, sunaddr, addr_len,
865 sk->sk_type, hash)) {
866 unix_release_addr(addr);
870 list = &unix_socket_table[addr->hash];
872 list = &unix_socket_table[dentry->d_inode->i_ino & (UNIX_HASH_SIZE-1)];
873 u->dentry = nd.dentry;
878 __unix_remove_socket(sk);
880 __unix_insert_socket(list, sk);
883 spin_unlock(&unix_table_lock);
885 mutex_unlock(&u->readlock);
892 mutex_unlock(&nd.dentry->d_inode->i_mutex);
897 unix_release_addr(addr);
901 static void unix_state_double_lock(struct sock *sk1, struct sock *sk2)
903 if (unlikely(sk1 == sk2) || !sk2) {
904 unix_state_lock(sk1);
908 unix_state_lock(sk1);
909 unix_state_lock_nested(sk2);
911 unix_state_lock(sk2);
912 unix_state_lock_nested(sk1);
916 static void unix_state_double_unlock(struct sock *sk1, struct sock *sk2)
918 if (unlikely(sk1 == sk2) || !sk2) {
919 unix_state_unlock(sk1);
922 unix_state_unlock(sk1);
923 unix_state_unlock(sk2);
926 static int unix_dgram_connect(struct socket *sock, struct sockaddr *addr,
929 struct sock *sk = sock->sk;
930 struct net *net = sk->sk_net;
931 struct sockaddr_un *sunaddr=(struct sockaddr_un*)addr;
936 if (addr->sa_family != AF_UNSPEC) {
937 err = unix_mkname(sunaddr, alen, &hash);
942 if (test_bit(SOCK_PASSCRED, &sock->flags) &&
943 !unix_sk(sk)->addr && (err = unix_autobind(sock)) != 0)
947 other=unix_find_other(net, sunaddr, alen, sock->type, hash, &err);
951 unix_state_double_lock(sk, other);
953 /* Apparently VFS overslept socket death. Retry. */
954 if (sock_flag(other, SOCK_DEAD)) {
955 unix_state_double_unlock(sk, other);
961 if (!unix_may_send(sk, other))
964 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
970 * 1003.1g breaking connected state with AF_UNSPEC
973 unix_state_double_lock(sk, other);
977 * If it was connected, reconnect.
980 struct sock *old_peer = unix_peer(sk);
982 unix_state_double_unlock(sk, other);
984 if (other != old_peer)
985 unix_dgram_disconnected(sk, old_peer);
989 unix_state_double_unlock(sk, other);
994 unix_state_double_unlock(sk, other);
1000 static long unix_wait_for_peer(struct sock *other, long timeo)
1002 struct unix_sock *u = unix_sk(other);
1006 prepare_to_wait_exclusive(&u->peer_wait, &wait, TASK_INTERRUPTIBLE);
1008 sched = !sock_flag(other, SOCK_DEAD) &&
1009 !(other->sk_shutdown & RCV_SHUTDOWN) &&
1010 (skb_queue_len(&other->sk_receive_queue) >
1011 other->sk_max_ack_backlog);
1013 unix_state_unlock(other);
1016 timeo = schedule_timeout(timeo);
1018 finish_wait(&u->peer_wait, &wait);
1022 static int unix_stream_connect(struct socket *sock, struct sockaddr *uaddr,
1023 int addr_len, int flags)
1025 struct sockaddr_un *sunaddr=(struct sockaddr_un *)uaddr;
1026 struct sock *sk = sock->sk;
1027 struct net *net = sk->sk_net;
1028 struct unix_sock *u = unix_sk(sk), *newu, *otheru;
1029 struct sock *newsk = NULL;
1030 struct sock *other = NULL;
1031 struct sk_buff *skb = NULL;
1037 err = unix_mkname(sunaddr, addr_len, &hash);
1042 if (test_bit(SOCK_PASSCRED, &sock->flags)
1043 && !u->addr && (err = unix_autobind(sock)) != 0)
1046 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
1048 /* First of all allocate resources.
1049 If we will make it after state is locked,
1050 we will have to recheck all again in any case.
1055 /* create new sock for complete connection */
1056 newsk = unix_create1(sk->sk_net, NULL);
1060 /* Allocate skb for sending to listening sock */
1061 skb = sock_wmalloc(newsk, 1, 0, GFP_KERNEL);
1066 /* Find listening sock. */
1067 other = unix_find_other(net, sunaddr, addr_len, sk->sk_type, hash, &err);
1071 /* Latch state of peer */
1072 unix_state_lock(other);
1074 /* Apparently VFS overslept socket death. Retry. */
1075 if (sock_flag(other, SOCK_DEAD)) {
1076 unix_state_unlock(other);
1081 err = -ECONNREFUSED;
1082 if (other->sk_state != TCP_LISTEN)
1085 if (skb_queue_len(&other->sk_receive_queue) >
1086 other->sk_max_ack_backlog) {
1091 timeo = unix_wait_for_peer(other, timeo);
1093 err = sock_intr_errno(timeo);
1094 if (signal_pending(current))
1102 It is tricky place. We need to grab write lock and cannot
1103 drop lock on peer. It is dangerous because deadlock is
1104 possible. Connect to self case and simultaneous
1105 attempt to connect are eliminated by checking socket
1106 state. other is TCP_LISTEN, if sk is TCP_LISTEN we
1107 check this before attempt to grab lock.
1109 Well, and we have to recheck the state after socket locked.
1115 /* This is ok... continue with connect */
1117 case TCP_ESTABLISHED:
1118 /* Socket is already connected */
1126 unix_state_lock_nested(sk);
1128 if (sk->sk_state != st) {
1129 unix_state_unlock(sk);
1130 unix_state_unlock(other);
1135 err = security_unix_stream_connect(sock, other->sk_socket, newsk);
1137 unix_state_unlock(sk);
1141 /* The way is open! Fastly set all the necessary fields... */
1144 unix_peer(newsk) = sk;
1145 newsk->sk_state = TCP_ESTABLISHED;
1146 newsk->sk_type = sk->sk_type;
1147 newsk->sk_peercred.pid = task_tgid_vnr(current);
1148 newsk->sk_peercred.uid = current->euid;
1149 newsk->sk_peercred.gid = current->egid;
1150 newu = unix_sk(newsk);
1151 newsk->sk_sleep = &newu->peer_wait;
1152 otheru = unix_sk(other);
1154 /* copy address information from listening to new sock*/
1156 atomic_inc(&otheru->addr->refcnt);
1157 newu->addr = otheru->addr;
1159 if (otheru->dentry) {
1160 newu->dentry = dget(otheru->dentry);
1161 newu->mnt = mntget(otheru->mnt);
1164 /* Set credentials */
1165 sk->sk_peercred = other->sk_peercred;
1167 sock->state = SS_CONNECTED;
1168 sk->sk_state = TCP_ESTABLISHED;
1171 smp_mb__after_atomic_inc(); /* sock_hold() does an atomic_inc() */
1172 unix_peer(sk) = newsk;
1174 unix_state_unlock(sk);
1176 /* take ten and and send info to listening sock */
1177 spin_lock(&other->sk_receive_queue.lock);
1178 __skb_queue_tail(&other->sk_receive_queue, skb);
1179 spin_unlock(&other->sk_receive_queue.lock);
1180 unix_state_unlock(other);
1181 other->sk_data_ready(other, 0);
1187 unix_state_unlock(other);
1193 unix_release_sock(newsk, 0);
1199 static int unix_socketpair(struct socket *socka, struct socket *sockb)
1201 struct sock *ska=socka->sk, *skb = sockb->sk;
1203 /* Join our sockets back to back */
1208 ska->sk_peercred.pid = skb->sk_peercred.pid = task_tgid_vnr(current);
1209 ska->sk_peercred.uid = skb->sk_peercred.uid = current->euid;
1210 ska->sk_peercred.gid = skb->sk_peercred.gid = current->egid;
1212 if (ska->sk_type != SOCK_DGRAM) {
1213 ska->sk_state = TCP_ESTABLISHED;
1214 skb->sk_state = TCP_ESTABLISHED;
1215 socka->state = SS_CONNECTED;
1216 sockb->state = SS_CONNECTED;
1221 static int unix_accept(struct socket *sock, struct socket *newsock, int flags)
1223 struct sock *sk = sock->sk;
1225 struct sk_buff *skb;
1229 if (sock->type!=SOCK_STREAM && sock->type!=SOCK_SEQPACKET)
1233 if (sk->sk_state != TCP_LISTEN)
1236 /* If socket state is TCP_LISTEN it cannot change (for now...),
1237 * so that no locks are necessary.
1240 skb = skb_recv_datagram(sk, 0, flags&O_NONBLOCK, &err);
1242 /* This means receive shutdown. */
1249 skb_free_datagram(sk, skb);
1250 wake_up_interruptible(&unix_sk(sk)->peer_wait);
1252 /* attach accepted sock to socket */
1253 unix_state_lock(tsk);
1254 newsock->state = SS_CONNECTED;
1255 sock_graft(tsk, newsock);
1256 unix_state_unlock(tsk);
1264 static int unix_getname(struct socket *sock, struct sockaddr *uaddr, int *uaddr_len, int peer)
1266 struct sock *sk = sock->sk;
1267 struct unix_sock *u;
1268 struct sockaddr_un *sunaddr=(struct sockaddr_un *)uaddr;
1272 sk = unix_peer_get(sk);
1283 unix_state_lock(sk);
1285 sunaddr->sun_family = AF_UNIX;
1286 sunaddr->sun_path[0] = 0;
1287 *uaddr_len = sizeof(short);
1289 struct unix_address *addr = u->addr;
1291 *uaddr_len = addr->len;
1292 memcpy(sunaddr, addr->name, *uaddr_len);
1294 unix_state_unlock(sk);
1300 static void unix_detach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1304 scm->fp = UNIXCB(skb).fp;
1305 skb->destructor = sock_wfree;
1306 UNIXCB(skb).fp = NULL;
1308 for (i=scm->fp->count-1; i>=0; i--)
1309 unix_notinflight(scm->fp->fp[i]);
1312 static void unix_destruct_fds(struct sk_buff *skb)
1314 struct scm_cookie scm;
1315 memset(&scm, 0, sizeof(scm));
1316 unix_detach_fds(&scm, skb);
1318 /* Alas, it calls VFS */
1319 /* So fscking what? fput() had been SMP-safe since the last Summer */
1324 static void unix_attach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1327 for (i=scm->fp->count-1; i>=0; i--)
1328 unix_inflight(scm->fp->fp[i]);
1329 UNIXCB(skb).fp = scm->fp;
1330 skb->destructor = unix_destruct_fds;
1335 * Send AF_UNIX data.
1338 static int unix_dgram_sendmsg(struct kiocb *kiocb, struct socket *sock,
1339 struct msghdr *msg, size_t len)
1341 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1342 struct sock *sk = sock->sk;
1343 struct net *net = sk->sk_net;
1344 struct unix_sock *u = unix_sk(sk);
1345 struct sockaddr_un *sunaddr=msg->msg_name;
1346 struct sock *other = NULL;
1347 int namelen = 0; /* fake GCC */
1350 struct sk_buff *skb;
1352 struct scm_cookie tmp_scm;
1354 if (NULL == siocb->scm)
1355 siocb->scm = &tmp_scm;
1356 err = scm_send(sock, msg, siocb->scm);
1361 if (msg->msg_flags&MSG_OOB)
1364 if (msg->msg_namelen) {
1365 err = unix_mkname(sunaddr, msg->msg_namelen, &hash);
1372 other = unix_peer_get(sk);
1377 if (test_bit(SOCK_PASSCRED, &sock->flags)
1378 && !u->addr && (err = unix_autobind(sock)) != 0)
1382 if (len > sk->sk_sndbuf - 32)
1385 skb = sock_alloc_send_skb(sk, len, msg->msg_flags&MSG_DONTWAIT, &err);
1389 memcpy(UNIXCREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1391 unix_attach_fds(siocb->scm, skb);
1392 unix_get_secdata(siocb->scm, skb);
1394 skb_reset_transport_header(skb);
1395 err = memcpy_fromiovec(skb_put(skb,len), msg->msg_iov, len);
1399 timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1404 if (sunaddr == NULL)
1407 other = unix_find_other(net, sunaddr, namelen, sk->sk_type,
1413 unix_state_lock(other);
1415 if (!unix_may_send(sk, other))
1418 if (sock_flag(other, SOCK_DEAD)) {
1420 * Check with 1003.1g - what should
1423 unix_state_unlock(other);
1427 unix_state_lock(sk);
1428 if (unix_peer(sk) == other) {
1430 unix_state_unlock(sk);
1432 unix_dgram_disconnected(sk, other);
1434 err = -ECONNREFUSED;
1436 unix_state_unlock(sk);
1446 if (other->sk_shutdown & RCV_SHUTDOWN)
1449 if (sk->sk_type != SOCK_SEQPACKET) {
1450 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1455 if (unix_peer(other) != sk &&
1456 (skb_queue_len(&other->sk_receive_queue) >
1457 other->sk_max_ack_backlog)) {
1463 timeo = unix_wait_for_peer(other, timeo);
1465 err = sock_intr_errno(timeo);
1466 if (signal_pending(current))
1472 skb_queue_tail(&other->sk_receive_queue, skb);
1473 unix_state_unlock(other);
1474 other->sk_data_ready(other, len);
1476 scm_destroy(siocb->scm);
1480 unix_state_unlock(other);
1486 scm_destroy(siocb->scm);
1491 static int unix_stream_sendmsg(struct kiocb *kiocb, struct socket *sock,
1492 struct msghdr *msg, size_t len)
1494 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1495 struct sock *sk = sock->sk;
1496 struct sock *other = NULL;
1497 struct sockaddr_un *sunaddr=msg->msg_name;
1499 struct sk_buff *skb;
1501 struct scm_cookie tmp_scm;
1503 if (NULL == siocb->scm)
1504 siocb->scm = &tmp_scm;
1505 err = scm_send(sock, msg, siocb->scm);
1510 if (msg->msg_flags&MSG_OOB)
1513 if (msg->msg_namelen) {
1514 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
1519 other = unix_peer(sk);
1524 if (sk->sk_shutdown & SEND_SHUTDOWN)
1530 * Optimisation for the fact that under 0.01% of X
1531 * messages typically need breaking up.
1536 /* Keep two messages in the pipe so it schedules better */
1537 if (size > ((sk->sk_sndbuf >> 1) - 64))
1538 size = (sk->sk_sndbuf >> 1) - 64;
1540 if (size > SKB_MAX_ALLOC)
1541 size = SKB_MAX_ALLOC;
1547 skb=sock_alloc_send_skb(sk,size,msg->msg_flags&MSG_DONTWAIT, &err);
1553 * If you pass two values to the sock_alloc_send_skb
1554 * it tries to grab the large buffer with GFP_NOFS
1555 * (which can fail easily), and if it fails grab the
1556 * fallback size buffer which is under a page and will
1559 size = min_t(int, size, skb_tailroom(skb));
1561 memcpy(UNIXCREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1563 unix_attach_fds(siocb->scm, skb);
1565 if ((err = memcpy_fromiovec(skb_put(skb,size), msg->msg_iov, size)) != 0) {
1570 unix_state_lock(other);
1572 if (sock_flag(other, SOCK_DEAD) ||
1573 (other->sk_shutdown & RCV_SHUTDOWN))
1576 skb_queue_tail(&other->sk_receive_queue, skb);
1577 unix_state_unlock(other);
1578 other->sk_data_ready(other, size);
1582 scm_destroy(siocb->scm);
1588 unix_state_unlock(other);
1591 if (sent==0 && !(msg->msg_flags&MSG_NOSIGNAL))
1592 send_sig(SIGPIPE,current,0);
1595 scm_destroy(siocb->scm);
1597 return sent ? : err;
1600 static int unix_seqpacket_sendmsg(struct kiocb *kiocb, struct socket *sock,
1601 struct msghdr *msg, size_t len)
1604 struct sock *sk = sock->sk;
1606 err = sock_error(sk);
1610 if (sk->sk_state != TCP_ESTABLISHED)
1613 if (msg->msg_namelen)
1614 msg->msg_namelen = 0;
1616 return unix_dgram_sendmsg(kiocb, sock, msg, len);
1619 static void unix_copy_addr(struct msghdr *msg, struct sock *sk)
1621 struct unix_sock *u = unix_sk(sk);
1623 msg->msg_namelen = 0;
1625 msg->msg_namelen = u->addr->len;
1626 memcpy(msg->msg_name, u->addr->name, u->addr->len);
1630 static int unix_dgram_recvmsg(struct kiocb *iocb, struct socket *sock,
1631 struct msghdr *msg, size_t size,
1634 struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1635 struct scm_cookie tmp_scm;
1636 struct sock *sk = sock->sk;
1637 struct unix_sock *u = unix_sk(sk);
1638 int noblock = flags & MSG_DONTWAIT;
1639 struct sk_buff *skb;
1646 msg->msg_namelen = 0;
1648 mutex_lock(&u->readlock);
1650 skb = skb_recv_datagram(sk, flags, noblock, &err);
1652 unix_state_lock(sk);
1653 /* Signal EOF on disconnected non-blocking SEQPACKET socket. */
1654 if (sk->sk_type == SOCK_SEQPACKET && err == -EAGAIN &&
1655 (sk->sk_shutdown & RCV_SHUTDOWN))
1657 unix_state_unlock(sk);
1661 wake_up_interruptible_sync(&u->peer_wait);
1664 unix_copy_addr(msg, skb->sk);
1666 if (size > skb->len)
1668 else if (size < skb->len)
1669 msg->msg_flags |= MSG_TRUNC;
1671 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, size);
1676 siocb->scm = &tmp_scm;
1677 memset(&tmp_scm, 0, sizeof(tmp_scm));
1679 siocb->scm->creds = *UNIXCREDS(skb);
1680 unix_set_secdata(siocb->scm, skb);
1682 if (!(flags & MSG_PEEK))
1685 unix_detach_fds(siocb->scm, skb);
1689 /* It is questionable: on PEEK we could:
1690 - do not return fds - good, but too simple 8)
1691 - return fds, and do not return them on read (old strategy,
1693 - clone fds (I chose it for now, it is the most universal
1696 POSIX 1003.1g does not actually define this clearly
1697 at all. POSIX 1003.1g doesn't define a lot of things
1702 siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1706 scm_recv(sock, msg, siocb->scm, flags);
1709 skb_free_datagram(sk,skb);
1711 mutex_unlock(&u->readlock);
1717 * Sleep until data has arrive. But check for races..
1720 static long unix_stream_data_wait(struct sock * sk, long timeo)
1724 unix_state_lock(sk);
1727 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1729 if (!skb_queue_empty(&sk->sk_receive_queue) ||
1731 (sk->sk_shutdown & RCV_SHUTDOWN) ||
1732 signal_pending(current) ||
1736 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1737 unix_state_unlock(sk);
1738 timeo = schedule_timeout(timeo);
1739 unix_state_lock(sk);
1740 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1743 finish_wait(sk->sk_sleep, &wait);
1744 unix_state_unlock(sk);
1750 static int unix_stream_recvmsg(struct kiocb *iocb, struct socket *sock,
1751 struct msghdr *msg, size_t size,
1754 struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1755 struct scm_cookie tmp_scm;
1756 struct sock *sk = sock->sk;
1757 struct unix_sock *u = unix_sk(sk);
1758 struct sockaddr_un *sunaddr=msg->msg_name;
1760 int check_creds = 0;
1766 if (sk->sk_state != TCP_ESTABLISHED)
1773 target = sock_rcvlowat(sk, flags&MSG_WAITALL, size);
1774 timeo = sock_rcvtimeo(sk, flags&MSG_DONTWAIT);
1776 msg->msg_namelen = 0;
1778 /* Lock the socket to prevent queue disordering
1779 * while sleeps in memcpy_tomsg
1783 siocb->scm = &tmp_scm;
1784 memset(&tmp_scm, 0, sizeof(tmp_scm));
1787 mutex_lock(&u->readlock);
1792 struct sk_buff *skb;
1794 unix_state_lock(sk);
1795 skb = skb_dequeue(&sk->sk_receive_queue);
1798 if (copied >= target)
1802 * POSIX 1003.1g mandates this order.
1805 if ((err = sock_error(sk)) != 0)
1807 if (sk->sk_shutdown & RCV_SHUTDOWN)
1810 unix_state_unlock(sk);
1814 mutex_unlock(&u->readlock);
1816 timeo = unix_stream_data_wait(sk, timeo);
1818 if (signal_pending(current)) {
1819 err = sock_intr_errno(timeo);
1822 mutex_lock(&u->readlock);
1825 unix_state_unlock(sk);
1828 unix_state_unlock(sk);
1831 /* Never glue messages from different writers */
1832 if (memcmp(UNIXCREDS(skb), &siocb->scm->creds, sizeof(siocb->scm->creds)) != 0) {
1833 skb_queue_head(&sk->sk_receive_queue, skb);
1837 /* Copy credentials */
1838 siocb->scm->creds = *UNIXCREDS(skb);
1842 /* Copy address just once */
1845 unix_copy_addr(msg, skb->sk);
1849 chunk = min_t(unsigned int, skb->len, size);
1850 if (memcpy_toiovec(msg->msg_iov, skb->data, chunk)) {
1851 skb_queue_head(&sk->sk_receive_queue, skb);
1859 /* Mark read part of skb as used */
1860 if (!(flags & MSG_PEEK))
1862 skb_pull(skb, chunk);
1865 unix_detach_fds(siocb->scm, skb);
1867 /* put the skb back if we didn't use it up.. */
1870 skb_queue_head(&sk->sk_receive_queue, skb);
1881 /* It is questionable, see note in unix_dgram_recvmsg.
1884 siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1886 /* put message back and return */
1887 skb_queue_head(&sk->sk_receive_queue, skb);
1892 mutex_unlock(&u->readlock);
1893 scm_recv(sock, msg, siocb->scm, flags);
1895 return copied ? : err;
1898 static int unix_shutdown(struct socket *sock, int mode)
1900 struct sock *sk = sock->sk;
1903 mode = (mode+1)&(RCV_SHUTDOWN|SEND_SHUTDOWN);
1906 unix_state_lock(sk);
1907 sk->sk_shutdown |= mode;
1908 other=unix_peer(sk);
1911 unix_state_unlock(sk);
1912 sk->sk_state_change(sk);
1915 (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET)) {
1919 if (mode&RCV_SHUTDOWN)
1920 peer_mode |= SEND_SHUTDOWN;
1921 if (mode&SEND_SHUTDOWN)
1922 peer_mode |= RCV_SHUTDOWN;
1923 unix_state_lock(other);
1924 other->sk_shutdown |= peer_mode;
1925 unix_state_unlock(other);
1926 other->sk_state_change(other);
1927 read_lock(&other->sk_callback_lock);
1928 if (peer_mode == SHUTDOWN_MASK)
1929 sk_wake_async(other,1,POLL_HUP);
1930 else if (peer_mode & RCV_SHUTDOWN)
1931 sk_wake_async(other,1,POLL_IN);
1932 read_unlock(&other->sk_callback_lock);
1940 static int unix_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1942 struct sock *sk = sock->sk;
1949 amount = atomic_read(&sk->sk_wmem_alloc);
1950 err = put_user(amount, (int __user *)arg);
1954 struct sk_buff *skb;
1956 if (sk->sk_state == TCP_LISTEN) {
1961 spin_lock(&sk->sk_receive_queue.lock);
1962 if (sk->sk_type == SOCK_STREAM ||
1963 sk->sk_type == SOCK_SEQPACKET) {
1964 skb_queue_walk(&sk->sk_receive_queue, skb)
1967 skb = skb_peek(&sk->sk_receive_queue);
1971 spin_unlock(&sk->sk_receive_queue.lock);
1972 err = put_user(amount, (int __user *)arg);
1983 static unsigned int unix_poll(struct file * file, struct socket *sock, poll_table *wait)
1985 struct sock *sk = sock->sk;
1988 poll_wait(file, sk->sk_sleep, wait);
1991 /* exceptional events? */
1994 if (sk->sk_shutdown == SHUTDOWN_MASK)
1996 if (sk->sk_shutdown & RCV_SHUTDOWN)
2000 if (!skb_queue_empty(&sk->sk_receive_queue) ||
2001 (sk->sk_shutdown & RCV_SHUTDOWN))
2002 mask |= POLLIN | POLLRDNORM;
2004 /* Connection-based need to check for termination and startup */
2005 if ((sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) && sk->sk_state == TCP_CLOSE)
2009 * we set writable also when the other side has shut down the
2010 * connection. This prevents stuck sockets.
2012 if (unix_writable(sk))
2013 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2019 #ifdef CONFIG_PROC_FS
2020 struct unix_iter_state {
2021 struct seq_net_private p;
2024 static struct sock *unix_seq_idx(struct unix_iter_state *iter, loff_t pos)
2029 for (s = first_unix_socket(&iter->i); s; s = next_unix_socket(&iter->i, s)) {
2030 if (s->sk_net != iter->p.net)
2040 static void *unix_seq_start(struct seq_file *seq, loff_t *pos)
2042 struct unix_iter_state *iter = seq->private;
2043 spin_lock(&unix_table_lock);
2044 return *pos ? unix_seq_idx(iter, *pos - 1) : ((void *) 1);
2047 static void *unix_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2049 struct unix_iter_state *iter = seq->private;
2050 struct sock *sk = v;
2054 sk = first_unix_socket(&iter->i);
2056 sk = next_unix_socket(&iter->i, sk);
2057 while (sk && (sk->sk_net != iter->p.net))
2058 sk = next_unix_socket(&iter->i, sk);
2062 static void unix_seq_stop(struct seq_file *seq, void *v)
2064 spin_unlock(&unix_table_lock);
2067 static int unix_seq_show(struct seq_file *seq, void *v)
2071 seq_puts(seq, "Num RefCount Protocol Flags Type St "
2075 struct unix_sock *u = unix_sk(s);
2078 seq_printf(seq, "%p: %08X %08X %08X %04X %02X %5lu",
2080 atomic_read(&s->sk_refcnt),
2082 s->sk_state == TCP_LISTEN ? __SO_ACCEPTCON : 0,
2085 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTED : SS_UNCONNECTED) :
2086 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTING : SS_DISCONNECTING),
2094 len = u->addr->len - sizeof(short);
2095 if (!UNIX_ABSTRACT(s))
2101 for ( ; i < len; i++)
2102 seq_putc(seq, u->addr->name->sun_path[i]);
2104 unix_state_unlock(s);
2105 seq_putc(seq, '\n');
2111 static const struct seq_operations unix_seq_ops = {
2112 .start = unix_seq_start,
2113 .next = unix_seq_next,
2114 .stop = unix_seq_stop,
2115 .show = unix_seq_show,
2119 static int unix_seq_open(struct inode *inode, struct file *file)
2121 return seq_open_net(inode, file, &unix_seq_ops,
2122 sizeof(struct unix_iter_state));
2125 static const struct file_operations unix_seq_fops = {
2126 .owner = THIS_MODULE,
2127 .open = unix_seq_open,
2129 .llseek = seq_lseek,
2130 .release = seq_release_net,
2135 static struct net_proto_family unix_family_ops = {
2137 .create = unix_create,
2138 .owner = THIS_MODULE,
2142 static int unix_net_init(struct net *net)
2144 int error = -ENOMEM;
2146 #ifdef CONFIG_PROC_FS
2147 if (!proc_net_fops_create(net, "unix", 0, &unix_seq_fops))
2155 static void unix_net_exit(struct net *net)
2157 proc_net_remove(net, "unix");
2160 static struct pernet_operations unix_net_ops = {
2161 .init = unix_net_init,
2162 .exit = unix_net_exit,
2165 static int __init af_unix_init(void)
2168 struct sk_buff *dummy_skb;
2170 BUILD_BUG_ON(sizeof(struct unix_skb_parms) > sizeof(dummy_skb->cb));
2172 rc = proto_register(&unix_proto, 1);
2174 printk(KERN_CRIT "%s: Cannot create unix_sock SLAB cache!\n",
2179 sock_register(&unix_family_ops);
2180 register_pernet_subsys(&unix_net_ops);
2181 unix_sysctl_register();
2186 static void __exit af_unix_exit(void)
2188 sock_unregister(PF_UNIX);
2189 unix_sysctl_unregister();
2190 proto_unregister(&unix_proto);
2191 unregister_pernet_subsys(&unix_net_ops);
2194 module_init(af_unix_init);
2195 module_exit(af_unix_exit);
2197 MODULE_LICENSE("GPL");
2198 MODULE_ALIAS_NETPROTO(PF_UNIX);