2 * NET4: Implementation of BSD Unix domain sockets.
4 * Authors: Alan Cox, <alan@lxorguk.ukuu.org.uk>
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.
12 * Linus Torvalds : Assorted bug cures.
13 * Niibe Yutaka : async I/O support.
14 * Carsten Paeth : PF_UNIX check, address fixes.
15 * Alan Cox : Limit size of allocated blocks.
16 * Alan Cox : Fixed the stupid socketpair bug.
17 * Alan Cox : BSD compatibility fine tuning.
18 * Alan Cox : Fixed a bug in connect when interrupted.
19 * Alan Cox : Sorted out a proper draft version of
20 * file descriptor passing hacked up from
22 * Marty Leisner : Fixes to fd passing
23 * Nick Nevin : recvmsg bugfix.
24 * Alan Cox : Started proper garbage collector
25 * Heiko EiBfeldt : Missing verify_area check
26 * Alan Cox : Started POSIXisms
27 * Andreas Schwab : Replace inode by dentry for proper
29 * Kirk Petersen : Made this a module
30 * Christoph Rohland : Elegant non-blocking accept/connect algorithm.
32 * Alexey Kuznetosv : Repaired (I hope) bugs introduces
33 * by above two patches.
34 * Andrea Arcangeli : If possible we block in connect(2)
35 * if the max backlog of the listen socket
36 * is been reached. This won't break
37 * old apps and it will avoid huge amount
38 * of socks hashed (this for unix_gc()
39 * performances reasons).
40 * Security fix that limits the max
41 * number of socks to 2*max_files and
42 * the number of skb queueable in the
44 * Artur Skawina : Hash function optimizations
45 * Alexey Kuznetsov : Full scale SMP. Lot of bugs are introduced 8)
46 * Malcolm Beattie : Set peercred for socketpair
47 * Michal Ostrowski : Module initialization cleanup.
48 * Arnaldo C. Melo : Remove MOD_{INC,DEC}_USE_COUNT,
49 * the core infrastructure is doing that
50 * for all net proto families now (2.5.69+)
53 * Known differences from reference BSD that was tested:
56 * ECONNREFUSED is not returned from one end of a connected() socket to the
57 * other the moment one end closes.
58 * fstat() doesn't return st_dev=0, and give the blksize as high water mark
59 * and a fake inode identifier (nor the BSD first socket fstat twice bug).
61 * accept() returns a path name even if the connecting socket has closed
62 * in the meantime (BSD loses the path and gives up).
63 * accept() returns 0 length path for an unbound connector. BSD returns 16
64 * and a null first byte in the path (but not for gethost/peername - BSD bug ??)
65 * socketpair(...SOCK_RAW..) doesn't panic the kernel.
66 * BSD af_unix apparently has connect forgetting to block properly.
67 * (need to check this with the POSIX spec in detail)
69 * Differences from 2.0.0-11-... (ANK)
70 * Bug fixes and improvements.
71 * - client shutdown killed server socket.
72 * - removed all useless cli/sti pairs.
74 * Semantic changes/extensions.
75 * - generic control message passing.
76 * - SCM_CREDENTIALS control message.
77 * - "Abstract" (not FS based) socket bindings.
78 * Abstract names are sequences of bytes (not zero terminated)
79 * started by 0, so that this name space does not intersect
83 #include <linux/module.h>
84 #include <linux/kernel.h>
85 #include <linux/signal.h>
86 #include <linux/sched.h>
87 #include <linux/errno.h>
88 #include <linux/string.h>
89 #include <linux/stat.h>
90 #include <linux/dcache.h>
91 #include <linux/namei.h>
92 #include <linux/socket.h>
94 #include <linux/fcntl.h>
95 #include <linux/termios.h>
96 #include <linux/sockios.h>
97 #include <linux/net.h>
100 #include <linux/slab.h>
101 #include <asm/uaccess.h>
102 #include <linux/skbuff.h>
103 #include <linux/netdevice.h>
104 #include <net/net_namespace.h>
105 #include <net/sock.h>
106 #include <net/tcp_states.h>
107 #include <net/af_unix.h>
108 #include <linux/proc_fs.h>
109 #include <linux/seq_file.h>
111 #include <linux/init.h>
112 #include <linux/poll.h>
113 #include <linux/rtnetlink.h>
114 #include <linux/mount.h>
115 #include <net/checksum.h>
116 #include <linux/security.h>
118 static struct hlist_head unix_socket_table[UNIX_HASH_SIZE + 1];
119 static DEFINE_SPINLOCK(unix_table_lock);
120 static atomic_t unix_nr_socks = ATOMIC_INIT(0);
122 #define unix_sockets_unbound (&unix_socket_table[UNIX_HASH_SIZE])
124 #define UNIX_ABSTRACT(sk) (unix_sk(sk)->addr->hash != UNIX_HASH_SIZE)
126 #ifdef CONFIG_SECURITY_NETWORK
127 static void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb)
129 memcpy(UNIXSID(skb), &scm->secid, sizeof(u32));
132 static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb)
134 scm->secid = *UNIXSID(skb);
137 static inline void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb)
140 static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb)
142 #endif /* CONFIG_SECURITY_NETWORK */
145 * SMP locking strategy:
146 * hash table is protected with spinlock unix_table_lock
147 * each socket state is protected by separate rwlock.
150 static inline unsigned unix_hash_fold(__wsum n)
152 unsigned hash = (__force unsigned)n;
155 return hash&(UNIX_HASH_SIZE-1);
158 #define unix_peer(sk) (unix_sk(sk)->peer)
160 static inline int unix_our_peer(struct sock *sk, struct sock *osk)
162 return unix_peer(osk) == sk;
165 static inline int unix_may_send(struct sock *sk, struct sock *osk)
167 return unix_peer(osk) == NULL || unix_our_peer(sk, osk);
170 static inline int unix_recvq_full(struct sock const *sk)
172 return skb_queue_len(&sk->sk_receive_queue) > sk->sk_max_ack_backlog;
175 static struct sock *unix_peer_get(struct sock *s)
183 unix_state_unlock(s);
187 static inline void unix_release_addr(struct unix_address *addr)
189 if (atomic_dec_and_test(&addr->refcnt))
194 * Check unix socket name:
195 * - should be not zero length.
196 * - if started by not zero, should be NULL terminated (FS object)
197 * - if started by zero, it is abstract name.
200 static int unix_mkname(struct sockaddr_un *sunaddr, int len, unsigned *hashp)
202 if (len <= sizeof(short) || len > sizeof(*sunaddr))
204 if (!sunaddr || sunaddr->sun_family != AF_UNIX)
206 if (sunaddr->sun_path[0]) {
208 * This may look like an off by one error but it is a bit more
209 * subtle. 108 is the longest valid AF_UNIX path for a binding.
210 * sun_path[108] doesnt as such exist. However in kernel space
211 * we are guaranteed that it is a valid memory location in our
212 * kernel address buffer.
214 ((char *)sunaddr)[len] = 0;
215 len = strlen(sunaddr->sun_path)+1+sizeof(short);
219 *hashp = unix_hash_fold(csum_partial((char *)sunaddr, len, 0));
223 static void __unix_remove_socket(struct sock *sk)
225 sk_del_node_init(sk);
228 static void __unix_insert_socket(struct hlist_head *list, struct sock *sk)
230 WARN_ON(!sk_unhashed(sk));
231 sk_add_node(sk, list);
234 static inline void unix_remove_socket(struct sock *sk)
236 spin_lock(&unix_table_lock);
237 __unix_remove_socket(sk);
238 spin_unlock(&unix_table_lock);
241 static inline void unix_insert_socket(struct hlist_head *list, struct sock *sk)
243 spin_lock(&unix_table_lock);
244 __unix_insert_socket(list, sk);
245 spin_unlock(&unix_table_lock);
248 static struct sock *__unix_find_socket_byname(struct net *net,
249 struct sockaddr_un *sunname,
250 int len, int type, unsigned hash)
253 struct hlist_node *node;
255 sk_for_each(s, node, &unix_socket_table[hash ^ type]) {
256 struct unix_sock *u = unix_sk(s);
258 if (!net_eq(sock_net(s), net))
261 if (u->addr->len == len &&
262 !memcmp(u->addr->name, sunname, len))
270 static inline struct sock *unix_find_socket_byname(struct net *net,
271 struct sockaddr_un *sunname,
277 spin_lock(&unix_table_lock);
278 s = __unix_find_socket_byname(net, sunname, len, type, hash);
281 spin_unlock(&unix_table_lock);
285 static struct sock *unix_find_socket_byinode(struct net *net, struct inode *i)
288 struct hlist_node *node;
290 spin_lock(&unix_table_lock);
292 &unix_socket_table[i->i_ino & (UNIX_HASH_SIZE - 1)]) {
293 struct dentry *dentry = unix_sk(s)->dentry;
295 if (!net_eq(sock_net(s), net))
298 if (dentry && dentry->d_inode == i) {
305 spin_unlock(&unix_table_lock);
309 static inline int unix_writable(struct sock *sk)
311 return (atomic_read(&sk->sk_wmem_alloc) << 2) <= sk->sk_sndbuf;
314 static void unix_write_space(struct sock *sk)
316 read_lock(&sk->sk_callback_lock);
317 if (unix_writable(sk)) {
318 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
319 wake_up_interruptible_sync(sk->sk_sleep);
320 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
322 read_unlock(&sk->sk_callback_lock);
325 /* When dgram socket disconnects (or changes its peer), we clear its receive
326 * queue of packets arrived from previous peer. First, it allows to do
327 * flow control based only on wmem_alloc; second, sk connected to peer
328 * may receive messages only from that peer. */
329 static void unix_dgram_disconnected(struct sock *sk, struct sock *other)
331 if (!skb_queue_empty(&sk->sk_receive_queue)) {
332 skb_queue_purge(&sk->sk_receive_queue);
333 wake_up_interruptible_all(&unix_sk(sk)->peer_wait);
335 /* If one link of bidirectional dgram pipe is disconnected,
336 * we signal error. Messages are lost. Do not make this,
337 * when peer was not connected to us.
339 if (!sock_flag(other, SOCK_DEAD) && unix_peer(other) == sk) {
340 other->sk_err = ECONNRESET;
341 other->sk_error_report(other);
346 static void unix_sock_destructor(struct sock *sk)
348 struct unix_sock *u = unix_sk(sk);
350 skb_queue_purge(&sk->sk_receive_queue);
352 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
353 WARN_ON(!sk_unhashed(sk));
354 WARN_ON(sk->sk_socket);
355 if (!sock_flag(sk, SOCK_DEAD)) {
356 printk(KERN_DEBUG "Attempt to release alive unix socket: %p\n", sk);
361 unix_release_addr(u->addr);
363 atomic_dec(&unix_nr_socks);
364 #ifdef UNIX_REFCNT_DEBUG
365 printk(KERN_DEBUG "UNIX %p is destroyed, %d are still alive.\n", sk,
366 atomic_read(&unix_nr_socks));
370 static int unix_release_sock(struct sock *sk, int embrion)
372 struct unix_sock *u = unix_sk(sk);
373 struct dentry *dentry;
374 struct vfsmount *mnt;
379 unix_remove_socket(sk);
384 sk->sk_shutdown = SHUTDOWN_MASK;
389 state = sk->sk_state;
390 sk->sk_state = TCP_CLOSE;
391 unix_state_unlock(sk);
393 wake_up_interruptible_all(&u->peer_wait);
395 skpair = unix_peer(sk);
397 if (skpair != NULL) {
398 if (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) {
399 unix_state_lock(skpair);
401 skpair->sk_shutdown = SHUTDOWN_MASK;
402 if (!skb_queue_empty(&sk->sk_receive_queue) || embrion)
403 skpair->sk_err = ECONNRESET;
404 unix_state_unlock(skpair);
405 skpair->sk_state_change(skpair);
406 read_lock(&skpair->sk_callback_lock);
407 sk_wake_async(skpair, SOCK_WAKE_WAITD, POLL_HUP);
408 read_unlock(&skpair->sk_callback_lock);
410 sock_put(skpair); /* It may now die */
411 unix_peer(sk) = NULL;
414 /* Try to flush out this socket. Throw out buffers at least */
416 while ((skb = skb_dequeue(&sk->sk_receive_queue)) != NULL) {
417 if (state == TCP_LISTEN)
418 unix_release_sock(skb->sk, 1);
419 /* passed fds are erased in the kfree_skb hook */
430 /* ---- Socket is dead now and most probably destroyed ---- */
433 * Fixme: BSD difference: In BSD all sockets connected to use get
434 * ECONNRESET and we die on the spot. In Linux we behave
435 * like files and pipes do and wait for the last
438 * Can't we simply set sock->err?
440 * What the above comment does talk about? --ANK(980817)
443 if (unix_tot_inflight)
444 unix_gc(); /* Garbage collect fds */
449 static int unix_listen(struct socket *sock, int backlog)
452 struct sock *sk = sock->sk;
453 struct unix_sock *u = unix_sk(sk);
456 if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
457 goto out; /* Only stream/seqpacket sockets accept */
460 goto out; /* No listens on an unbound socket */
462 if (sk->sk_state != TCP_CLOSE && sk->sk_state != TCP_LISTEN)
464 if (backlog > sk->sk_max_ack_backlog)
465 wake_up_interruptible_all(&u->peer_wait);
466 sk->sk_max_ack_backlog = backlog;
467 sk->sk_state = TCP_LISTEN;
468 /* set credentials so connect can copy them */
469 sk->sk_peercred.pid = task_tgid_vnr(current);
470 sk->sk_peercred.uid = current->euid;
471 sk->sk_peercred.gid = current->egid;
475 unix_state_unlock(sk);
480 static int unix_release(struct socket *);
481 static int unix_bind(struct socket *, struct sockaddr *, int);
482 static int unix_stream_connect(struct socket *, struct sockaddr *,
483 int addr_len, int flags);
484 static int unix_socketpair(struct socket *, struct socket *);
485 static int unix_accept(struct socket *, struct socket *, int);
486 static int unix_getname(struct socket *, struct sockaddr *, int *, int);
487 static unsigned int unix_poll(struct file *, struct socket *, poll_table *);
488 static unsigned int unix_dgram_poll(struct file *, struct socket *,
490 static int unix_ioctl(struct socket *, unsigned int, unsigned long);
491 static int unix_shutdown(struct socket *, int);
492 static int unix_stream_sendmsg(struct kiocb *, struct socket *,
493 struct msghdr *, size_t);
494 static int unix_stream_recvmsg(struct kiocb *, struct socket *,
495 struct msghdr *, size_t, int);
496 static int unix_dgram_sendmsg(struct kiocb *, struct socket *,
497 struct msghdr *, size_t);
498 static int unix_dgram_recvmsg(struct kiocb *, struct socket *,
499 struct msghdr *, size_t, int);
500 static int unix_dgram_connect(struct socket *, struct sockaddr *,
502 static int unix_seqpacket_sendmsg(struct kiocb *, struct socket *,
503 struct msghdr *, size_t);
505 static const struct proto_ops unix_stream_ops = {
507 .owner = THIS_MODULE,
508 .release = unix_release,
510 .connect = unix_stream_connect,
511 .socketpair = unix_socketpair,
512 .accept = unix_accept,
513 .getname = unix_getname,
516 .listen = unix_listen,
517 .shutdown = unix_shutdown,
518 .setsockopt = sock_no_setsockopt,
519 .getsockopt = sock_no_getsockopt,
520 .sendmsg = unix_stream_sendmsg,
521 .recvmsg = unix_stream_recvmsg,
522 .mmap = sock_no_mmap,
523 .sendpage = sock_no_sendpage,
526 static const struct proto_ops unix_dgram_ops = {
528 .owner = THIS_MODULE,
529 .release = unix_release,
531 .connect = unix_dgram_connect,
532 .socketpair = unix_socketpair,
533 .accept = sock_no_accept,
534 .getname = unix_getname,
535 .poll = unix_dgram_poll,
537 .listen = sock_no_listen,
538 .shutdown = unix_shutdown,
539 .setsockopt = sock_no_setsockopt,
540 .getsockopt = sock_no_getsockopt,
541 .sendmsg = unix_dgram_sendmsg,
542 .recvmsg = unix_dgram_recvmsg,
543 .mmap = sock_no_mmap,
544 .sendpage = sock_no_sendpage,
547 static const struct proto_ops unix_seqpacket_ops = {
549 .owner = THIS_MODULE,
550 .release = unix_release,
552 .connect = unix_stream_connect,
553 .socketpair = unix_socketpair,
554 .accept = unix_accept,
555 .getname = unix_getname,
556 .poll = unix_dgram_poll,
558 .listen = unix_listen,
559 .shutdown = unix_shutdown,
560 .setsockopt = sock_no_setsockopt,
561 .getsockopt = sock_no_getsockopt,
562 .sendmsg = unix_seqpacket_sendmsg,
563 .recvmsg = unix_dgram_recvmsg,
564 .mmap = sock_no_mmap,
565 .sendpage = sock_no_sendpage,
568 static struct proto unix_proto = {
570 .owner = THIS_MODULE,
571 .obj_size = sizeof(struct unix_sock),
575 * AF_UNIX sockets do not interact with hardware, hence they
576 * dont trigger interrupts - so it's safe for them to have
577 * bh-unsafe locking for their sk_receive_queue.lock. Split off
578 * this special lock-class by reinitializing the spinlock key:
580 static struct lock_class_key af_unix_sk_receive_queue_lock_key;
582 static struct sock *unix_create1(struct net *net, struct socket *sock)
584 struct sock *sk = NULL;
587 atomic_inc(&unix_nr_socks);
588 if (atomic_read(&unix_nr_socks) > 2 * get_max_files())
591 sk = sk_alloc(net, PF_UNIX, GFP_KERNEL, &unix_proto);
595 sock_init_data(sock, sk);
596 lockdep_set_class(&sk->sk_receive_queue.lock,
597 &af_unix_sk_receive_queue_lock_key);
599 sk->sk_write_space = unix_write_space;
600 sk->sk_max_ack_backlog = net->unx.sysctl_max_dgram_qlen;
601 sk->sk_destruct = unix_sock_destructor;
605 spin_lock_init(&u->lock);
606 atomic_long_set(&u->inflight, 0);
607 INIT_LIST_HEAD(&u->link);
608 mutex_init(&u->readlock); /* single task reading lock */
609 init_waitqueue_head(&u->peer_wait);
610 unix_insert_socket(unix_sockets_unbound, sk);
613 atomic_dec(&unix_nr_socks);
617 static int unix_create(struct net *net, struct socket *sock, int protocol)
619 if (protocol && protocol != PF_UNIX)
620 return -EPROTONOSUPPORT;
622 sock->state = SS_UNCONNECTED;
624 switch (sock->type) {
626 sock->ops = &unix_stream_ops;
629 * Believe it or not BSD has AF_UNIX, SOCK_RAW though
633 sock->type = SOCK_DGRAM;
635 sock->ops = &unix_dgram_ops;
638 sock->ops = &unix_seqpacket_ops;
641 return -ESOCKTNOSUPPORT;
644 return unix_create1(net, sock) ? 0 : -ENOMEM;
647 static int unix_release(struct socket *sock)
649 struct sock *sk = sock->sk;
656 return unix_release_sock(sk, 0);
659 static int unix_autobind(struct socket *sock)
661 struct sock *sk = sock->sk;
662 struct net *net = sock_net(sk);
663 struct unix_sock *u = unix_sk(sk);
664 static u32 ordernum = 1;
665 struct unix_address *addr;
668 mutex_lock(&u->readlock);
675 addr = kzalloc(sizeof(*addr) + sizeof(short) + 16, GFP_KERNEL);
679 addr->name->sun_family = AF_UNIX;
680 atomic_set(&addr->refcnt, 1);
683 addr->len = sprintf(addr->name->sun_path+1, "%05x", ordernum) + 1 + sizeof(short);
684 addr->hash = unix_hash_fold(csum_partial((void *)addr->name, addr->len, 0));
686 spin_lock(&unix_table_lock);
687 ordernum = (ordernum+1)&0xFFFFF;
689 if (__unix_find_socket_byname(net, addr->name, addr->len, sock->type,
691 spin_unlock(&unix_table_lock);
692 /* Sanity yield. It is unusual case, but yet... */
693 if (!(ordernum&0xFF))
697 addr->hash ^= sk->sk_type;
699 __unix_remove_socket(sk);
701 __unix_insert_socket(&unix_socket_table[addr->hash], sk);
702 spin_unlock(&unix_table_lock);
705 out: mutex_unlock(&u->readlock);
709 static struct sock *unix_find_other(struct net *net,
710 struct sockaddr_un *sunname, int len,
711 int type, unsigned hash, int *error)
717 if (sunname->sun_path[0]) {
719 err = kern_path(sunname->sun_path, LOOKUP_FOLLOW, &path);
722 inode = path.dentry->d_inode;
723 err = inode_permission(inode, MAY_WRITE);
728 if (!S_ISSOCK(inode->i_mode))
730 u = unix_find_socket_byinode(net, inode);
734 if (u->sk_type == type)
735 touch_atime(path.mnt, path.dentry);
740 if (u->sk_type != type) {
746 u = unix_find_socket_byname(net, sunname, len, type, hash);
748 struct dentry *dentry;
749 dentry = unix_sk(u)->dentry;
751 touch_atime(unix_sk(u)->mnt, dentry);
765 static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
767 struct sock *sk = sock->sk;
768 struct net *net = sock_net(sk);
769 struct unix_sock *u = unix_sk(sk);
770 struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
771 struct dentry *dentry = NULL;
775 struct unix_address *addr;
776 struct hlist_head *list;
779 if (sunaddr->sun_family != AF_UNIX)
782 if (addr_len == sizeof(short)) {
783 err = unix_autobind(sock);
787 err = unix_mkname(sunaddr, addr_len, &hash);
792 mutex_lock(&u->readlock);
799 addr = kmalloc(sizeof(*addr)+addr_len, GFP_KERNEL);
803 memcpy(addr->name, sunaddr, addr_len);
804 addr->len = addr_len;
805 addr->hash = hash ^ sk->sk_type;
806 atomic_set(&addr->refcnt, 1);
808 if (sunaddr->sun_path[0]) {
812 * Get the parent directory, calculate the hash for last
815 err = path_lookup(sunaddr->sun_path, LOOKUP_PARENT, &nd);
817 goto out_mknod_parent;
819 dentry = lookup_create(&nd, 0);
820 err = PTR_ERR(dentry);
822 goto out_mknod_unlock;
825 * All right, let's create it.
828 (SOCK_INODE(sock)->i_mode & ~current->fs->umask);
829 err = mnt_want_write(nd.path.mnt);
832 err = vfs_mknod(nd.path.dentry->d_inode, dentry, mode, 0);
833 mnt_drop_write(nd.path.mnt);
836 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
837 dput(nd.path.dentry);
838 nd.path.dentry = dentry;
840 addr->hash = UNIX_HASH_SIZE;
843 spin_lock(&unix_table_lock);
845 if (!sunaddr->sun_path[0]) {
847 if (__unix_find_socket_byname(net, sunaddr, addr_len,
848 sk->sk_type, hash)) {
849 unix_release_addr(addr);
853 list = &unix_socket_table[addr->hash];
855 list = &unix_socket_table[dentry->d_inode->i_ino & (UNIX_HASH_SIZE-1)];
856 u->dentry = nd.path.dentry;
857 u->mnt = nd.path.mnt;
861 __unix_remove_socket(sk);
863 __unix_insert_socket(list, sk);
866 spin_unlock(&unix_table_lock);
868 mutex_unlock(&u->readlock);
875 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
880 unix_release_addr(addr);
884 static void unix_state_double_lock(struct sock *sk1, struct sock *sk2)
886 if (unlikely(sk1 == sk2) || !sk2) {
887 unix_state_lock(sk1);
891 unix_state_lock(sk1);
892 unix_state_lock_nested(sk2);
894 unix_state_lock(sk2);
895 unix_state_lock_nested(sk1);
899 static void unix_state_double_unlock(struct sock *sk1, struct sock *sk2)
901 if (unlikely(sk1 == sk2) || !sk2) {
902 unix_state_unlock(sk1);
905 unix_state_unlock(sk1);
906 unix_state_unlock(sk2);
909 static int unix_dgram_connect(struct socket *sock, struct sockaddr *addr,
912 struct sock *sk = sock->sk;
913 struct net *net = sock_net(sk);
914 struct sockaddr_un *sunaddr = (struct sockaddr_un *)addr;
919 if (addr->sa_family != AF_UNSPEC) {
920 err = unix_mkname(sunaddr, alen, &hash);
925 if (test_bit(SOCK_PASSCRED, &sock->flags) &&
926 !unix_sk(sk)->addr && (err = unix_autobind(sock)) != 0)
930 other = unix_find_other(net, sunaddr, alen, sock->type, hash, &err);
934 unix_state_double_lock(sk, other);
936 /* Apparently VFS overslept socket death. Retry. */
937 if (sock_flag(other, SOCK_DEAD)) {
938 unix_state_double_unlock(sk, other);
944 if (!unix_may_send(sk, other))
947 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
953 * 1003.1g breaking connected state with AF_UNSPEC
956 unix_state_double_lock(sk, other);
960 * If it was connected, reconnect.
963 struct sock *old_peer = unix_peer(sk);
964 unix_peer(sk) = other;
965 unix_state_double_unlock(sk, other);
967 if (other != old_peer)
968 unix_dgram_disconnected(sk, old_peer);
971 unix_peer(sk) = other;
972 unix_state_double_unlock(sk, other);
977 unix_state_double_unlock(sk, other);
983 static long unix_wait_for_peer(struct sock *other, long timeo)
985 struct unix_sock *u = unix_sk(other);
989 prepare_to_wait_exclusive(&u->peer_wait, &wait, TASK_INTERRUPTIBLE);
991 sched = !sock_flag(other, SOCK_DEAD) &&
992 !(other->sk_shutdown & RCV_SHUTDOWN) &&
993 unix_recvq_full(other);
995 unix_state_unlock(other);
998 timeo = schedule_timeout(timeo);
1000 finish_wait(&u->peer_wait, &wait);
1004 static int unix_stream_connect(struct socket *sock, struct sockaddr *uaddr,
1005 int addr_len, int flags)
1007 struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
1008 struct sock *sk = sock->sk;
1009 struct net *net = sock_net(sk);
1010 struct unix_sock *u = unix_sk(sk), *newu, *otheru;
1011 struct sock *newsk = NULL;
1012 struct sock *other = NULL;
1013 struct sk_buff *skb = NULL;
1019 err = unix_mkname(sunaddr, addr_len, &hash);
1024 if (test_bit(SOCK_PASSCRED, &sock->flags)
1025 && !u->addr && (err = unix_autobind(sock)) != 0)
1028 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
1030 /* First of all allocate resources.
1031 If we will make it after state is locked,
1032 we will have to recheck all again in any case.
1037 /* create new sock for complete connection */
1038 newsk = unix_create1(sock_net(sk), NULL);
1042 /* Allocate skb for sending to listening sock */
1043 skb = sock_wmalloc(newsk, 1, 0, GFP_KERNEL);
1048 /* Find listening sock. */
1049 other = unix_find_other(net, sunaddr, addr_len, sk->sk_type, hash, &err);
1053 /* Latch state of peer */
1054 unix_state_lock(other);
1056 /* Apparently VFS overslept socket death. Retry. */
1057 if (sock_flag(other, SOCK_DEAD)) {
1058 unix_state_unlock(other);
1063 err = -ECONNREFUSED;
1064 if (other->sk_state != TCP_LISTEN)
1067 if (unix_recvq_full(other)) {
1072 timeo = unix_wait_for_peer(other, timeo);
1074 err = sock_intr_errno(timeo);
1075 if (signal_pending(current))
1083 It is tricky place. We need to grab write lock and cannot
1084 drop lock on peer. It is dangerous because deadlock is
1085 possible. Connect to self case and simultaneous
1086 attempt to connect are eliminated by checking socket
1087 state. other is TCP_LISTEN, if sk is TCP_LISTEN we
1088 check this before attempt to grab lock.
1090 Well, and we have to recheck the state after socket locked.
1096 /* This is ok... continue with connect */
1098 case TCP_ESTABLISHED:
1099 /* Socket is already connected */
1107 unix_state_lock_nested(sk);
1109 if (sk->sk_state != st) {
1110 unix_state_unlock(sk);
1111 unix_state_unlock(other);
1116 err = security_unix_stream_connect(sock, other->sk_socket, newsk);
1118 unix_state_unlock(sk);
1122 /* The way is open! Fastly set all the necessary fields... */
1125 unix_peer(newsk) = sk;
1126 newsk->sk_state = TCP_ESTABLISHED;
1127 newsk->sk_type = sk->sk_type;
1128 newsk->sk_peercred.pid = task_tgid_vnr(current);
1129 newsk->sk_peercred.uid = current->euid;
1130 newsk->sk_peercred.gid = current->egid;
1131 newu = unix_sk(newsk);
1132 newsk->sk_sleep = &newu->peer_wait;
1133 otheru = unix_sk(other);
1135 /* copy address information from listening to new sock*/
1137 atomic_inc(&otheru->addr->refcnt);
1138 newu->addr = otheru->addr;
1140 if (otheru->dentry) {
1141 newu->dentry = dget(otheru->dentry);
1142 newu->mnt = mntget(otheru->mnt);
1145 /* Set credentials */
1146 sk->sk_peercred = other->sk_peercred;
1148 sock->state = SS_CONNECTED;
1149 sk->sk_state = TCP_ESTABLISHED;
1152 smp_mb__after_atomic_inc(); /* sock_hold() does an atomic_inc() */
1153 unix_peer(sk) = newsk;
1155 unix_state_unlock(sk);
1157 /* take ten and and send info to listening sock */
1158 spin_lock(&other->sk_receive_queue.lock);
1159 __skb_queue_tail(&other->sk_receive_queue, skb);
1160 spin_unlock(&other->sk_receive_queue.lock);
1161 unix_state_unlock(other);
1162 other->sk_data_ready(other, 0);
1168 unix_state_unlock(other);
1174 unix_release_sock(newsk, 0);
1180 static int unix_socketpair(struct socket *socka, struct socket *sockb)
1182 struct sock *ska = socka->sk, *skb = sockb->sk;
1184 /* Join our sockets back to back */
1187 unix_peer(ska) = skb;
1188 unix_peer(skb) = ska;
1189 ska->sk_peercred.pid = skb->sk_peercred.pid = task_tgid_vnr(current);
1190 ska->sk_peercred.uid = skb->sk_peercred.uid = current->euid;
1191 ska->sk_peercred.gid = skb->sk_peercred.gid = current->egid;
1193 if (ska->sk_type != SOCK_DGRAM) {
1194 ska->sk_state = TCP_ESTABLISHED;
1195 skb->sk_state = TCP_ESTABLISHED;
1196 socka->state = SS_CONNECTED;
1197 sockb->state = SS_CONNECTED;
1202 static int unix_accept(struct socket *sock, struct socket *newsock, int flags)
1204 struct sock *sk = sock->sk;
1206 struct sk_buff *skb;
1210 if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
1214 if (sk->sk_state != TCP_LISTEN)
1217 /* If socket state is TCP_LISTEN it cannot change (for now...),
1218 * so that no locks are necessary.
1221 skb = skb_recv_datagram(sk, 0, flags&O_NONBLOCK, &err);
1223 /* This means receive shutdown. */
1230 skb_free_datagram(sk, skb);
1231 wake_up_interruptible(&unix_sk(sk)->peer_wait);
1233 /* attach accepted sock to socket */
1234 unix_state_lock(tsk);
1235 newsock->state = SS_CONNECTED;
1236 sock_graft(tsk, newsock);
1237 unix_state_unlock(tsk);
1245 static int unix_getname(struct socket *sock, struct sockaddr *uaddr, int *uaddr_len, int peer)
1247 struct sock *sk = sock->sk;
1248 struct unix_sock *u;
1249 struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
1253 sk = unix_peer_get(sk);
1264 unix_state_lock(sk);
1266 sunaddr->sun_family = AF_UNIX;
1267 sunaddr->sun_path[0] = 0;
1268 *uaddr_len = sizeof(short);
1270 struct unix_address *addr = u->addr;
1272 *uaddr_len = addr->len;
1273 memcpy(sunaddr, addr->name, *uaddr_len);
1275 unix_state_unlock(sk);
1281 static void unix_detach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1285 scm->fp = UNIXCB(skb).fp;
1286 skb->destructor = sock_wfree;
1287 UNIXCB(skb).fp = NULL;
1289 for (i = scm->fp->count-1; i >= 0; i--)
1290 unix_notinflight(scm->fp->fp[i]);
1293 static void unix_destruct_fds(struct sk_buff *skb)
1295 struct scm_cookie scm;
1296 memset(&scm, 0, sizeof(scm));
1297 unix_detach_fds(&scm, skb);
1299 /* Alas, it calls VFS */
1300 /* So fscking what? fput() had been SMP-safe since the last Summer */
1305 static int unix_attach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1310 * Need to duplicate file references for the sake of garbage
1311 * collection. Otherwise a socket in the fps might become a
1312 * candidate for GC while the skb is not yet queued.
1314 UNIXCB(skb).fp = scm_fp_dup(scm->fp);
1315 if (!UNIXCB(skb).fp)
1318 for (i = scm->fp->count-1; i >= 0; i--)
1319 unix_inflight(scm->fp->fp[i]);
1320 skb->destructor = unix_destruct_fds;
1325 * Send AF_UNIX data.
1328 static int unix_dgram_sendmsg(struct kiocb *kiocb, struct socket *sock,
1329 struct msghdr *msg, size_t len)
1331 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1332 struct sock *sk = sock->sk;
1333 struct net *net = sock_net(sk);
1334 struct unix_sock *u = unix_sk(sk);
1335 struct sockaddr_un *sunaddr = msg->msg_name;
1336 struct sock *other = NULL;
1337 int namelen = 0; /* fake GCC */
1340 struct sk_buff *skb;
1342 struct scm_cookie tmp_scm;
1344 if (NULL == siocb->scm)
1345 siocb->scm = &tmp_scm;
1346 err = scm_send(sock, msg, siocb->scm);
1351 if (msg->msg_flags&MSG_OOB)
1354 if (msg->msg_namelen) {
1355 err = unix_mkname(sunaddr, msg->msg_namelen, &hash);
1362 other = unix_peer_get(sk);
1367 if (test_bit(SOCK_PASSCRED, &sock->flags)
1368 && !u->addr && (err = unix_autobind(sock)) != 0)
1372 if (len > sk->sk_sndbuf - 32)
1375 skb = sock_alloc_send_skb(sk, len, msg->msg_flags&MSG_DONTWAIT, &err);
1379 memcpy(UNIXCREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1380 if (siocb->scm->fp) {
1381 err = unix_attach_fds(siocb->scm, skb);
1385 unix_get_secdata(siocb->scm, skb);
1387 skb_reset_transport_header(skb);
1388 err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
1392 timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1397 if (sunaddr == NULL)
1400 other = unix_find_other(net, sunaddr, namelen, sk->sk_type,
1406 unix_state_lock(other);
1408 if (!unix_may_send(sk, other))
1411 if (sock_flag(other, SOCK_DEAD)) {
1413 * Check with 1003.1g - what should
1416 unix_state_unlock(other);
1420 unix_state_lock(sk);
1421 if (unix_peer(sk) == other) {
1422 unix_peer(sk) = NULL;
1423 unix_state_unlock(sk);
1425 unix_dgram_disconnected(sk, other);
1427 err = -ECONNREFUSED;
1429 unix_state_unlock(sk);
1439 if (other->sk_shutdown & RCV_SHUTDOWN)
1442 if (sk->sk_type != SOCK_SEQPACKET) {
1443 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1448 if (unix_peer(other) != sk && unix_recvq_full(other)) {
1454 timeo = unix_wait_for_peer(other, timeo);
1456 err = sock_intr_errno(timeo);
1457 if (signal_pending(current))
1463 skb_queue_tail(&other->sk_receive_queue, skb);
1464 unix_state_unlock(other);
1465 other->sk_data_ready(other, len);
1467 scm_destroy(siocb->scm);
1471 unix_state_unlock(other);
1477 scm_destroy(siocb->scm);
1482 static int unix_stream_sendmsg(struct kiocb *kiocb, struct socket *sock,
1483 struct msghdr *msg, size_t len)
1485 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1486 struct sock *sk = sock->sk;
1487 struct sock *other = NULL;
1488 struct sockaddr_un *sunaddr = msg->msg_name;
1490 struct sk_buff *skb;
1492 struct scm_cookie tmp_scm;
1494 if (NULL == siocb->scm)
1495 siocb->scm = &tmp_scm;
1496 err = scm_send(sock, msg, siocb->scm);
1501 if (msg->msg_flags&MSG_OOB)
1504 if (msg->msg_namelen) {
1505 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
1510 other = unix_peer(sk);
1515 if (sk->sk_shutdown & SEND_SHUTDOWN)
1518 while (sent < len) {
1520 * Optimisation for the fact that under 0.01% of X
1521 * messages typically need breaking up.
1526 /* Keep two messages in the pipe so it schedules better */
1527 if (size > ((sk->sk_sndbuf >> 1) - 64))
1528 size = (sk->sk_sndbuf >> 1) - 64;
1530 if (size > SKB_MAX_ALLOC)
1531 size = SKB_MAX_ALLOC;
1537 skb = sock_alloc_send_skb(sk, size, msg->msg_flags&MSG_DONTWAIT,
1544 * If you pass two values to the sock_alloc_send_skb
1545 * it tries to grab the large buffer with GFP_NOFS
1546 * (which can fail easily), and if it fails grab the
1547 * fallback size buffer which is under a page and will
1550 size = min_t(int, size, skb_tailroom(skb));
1552 memcpy(UNIXCREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1553 if (siocb->scm->fp) {
1554 err = unix_attach_fds(siocb->scm, skb);
1561 err = memcpy_fromiovec(skb_put(skb, size), msg->msg_iov, size);
1567 unix_state_lock(other);
1569 if (sock_flag(other, SOCK_DEAD) ||
1570 (other->sk_shutdown & RCV_SHUTDOWN))
1573 skb_queue_tail(&other->sk_receive_queue, skb);
1574 unix_state_unlock(other);
1575 other->sk_data_ready(other, size);
1579 scm_destroy(siocb->scm);
1585 unix_state_unlock(other);
1588 if (sent == 0 && !(msg->msg_flags&MSG_NOSIGNAL))
1589 send_sig(SIGPIPE, current, 0);
1592 scm_destroy(siocb->scm);
1594 return sent ? : err;
1597 static int unix_seqpacket_sendmsg(struct kiocb *kiocb, struct socket *sock,
1598 struct msghdr *msg, size_t len)
1601 struct sock *sk = sock->sk;
1603 err = sock_error(sk);
1607 if (sk->sk_state != TCP_ESTABLISHED)
1610 if (msg->msg_namelen)
1611 msg->msg_namelen = 0;
1613 return unix_dgram_sendmsg(kiocb, sock, msg, len);
1616 static void unix_copy_addr(struct msghdr *msg, struct sock *sk)
1618 struct unix_sock *u = unix_sk(sk);
1620 msg->msg_namelen = 0;
1622 msg->msg_namelen = u->addr->len;
1623 memcpy(msg->msg_name, u->addr->name, u->addr->len);
1627 static int unix_dgram_recvmsg(struct kiocb *iocb, struct socket *sock,
1628 struct msghdr *msg, size_t size,
1631 struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1632 struct scm_cookie tmp_scm;
1633 struct sock *sk = sock->sk;
1634 struct unix_sock *u = unix_sk(sk);
1635 int noblock = flags & MSG_DONTWAIT;
1636 struct sk_buff *skb;
1643 msg->msg_namelen = 0;
1645 mutex_lock(&u->readlock);
1647 skb = skb_recv_datagram(sk, flags, noblock, &err);
1649 unix_state_lock(sk);
1650 /* Signal EOF on disconnected non-blocking SEQPACKET socket. */
1651 if (sk->sk_type == SOCK_SEQPACKET && err == -EAGAIN &&
1652 (sk->sk_shutdown & RCV_SHUTDOWN))
1654 unix_state_unlock(sk);
1658 wake_up_interruptible_sync(&u->peer_wait);
1661 unix_copy_addr(msg, skb->sk);
1663 if (size > skb->len)
1665 else if (size < skb->len)
1666 msg->msg_flags |= MSG_TRUNC;
1668 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, size);
1673 siocb->scm = &tmp_scm;
1674 memset(&tmp_scm, 0, sizeof(tmp_scm));
1676 siocb->scm->creds = *UNIXCREDS(skb);
1677 unix_set_secdata(siocb->scm, skb);
1679 if (!(flags & MSG_PEEK)) {
1681 unix_detach_fds(siocb->scm, skb);
1683 /* It is questionable: on PEEK we could:
1684 - do not return fds - good, but too simple 8)
1685 - return fds, and do not return them on read (old strategy,
1687 - clone fds (I chose it for now, it is the most universal
1690 POSIX 1003.1g does not actually define this clearly
1691 at all. POSIX 1003.1g doesn't define a lot of things
1696 siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1700 scm_recv(sock, msg, siocb->scm, flags);
1703 skb_free_datagram(sk, skb);
1705 mutex_unlock(&u->readlock);
1711 * Sleep until data has arrive. But check for races..
1714 static long unix_stream_data_wait(struct sock *sk, long timeo)
1718 unix_state_lock(sk);
1721 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1723 if (!skb_queue_empty(&sk->sk_receive_queue) ||
1725 (sk->sk_shutdown & RCV_SHUTDOWN) ||
1726 signal_pending(current) ||
1730 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1731 unix_state_unlock(sk);
1732 timeo = schedule_timeout(timeo);
1733 unix_state_lock(sk);
1734 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1737 finish_wait(sk->sk_sleep, &wait);
1738 unix_state_unlock(sk);
1744 static int unix_stream_recvmsg(struct kiocb *iocb, struct socket *sock,
1745 struct msghdr *msg, size_t size,
1748 struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1749 struct scm_cookie tmp_scm;
1750 struct sock *sk = sock->sk;
1751 struct unix_sock *u = unix_sk(sk);
1752 struct sockaddr_un *sunaddr = msg->msg_name;
1754 int check_creds = 0;
1760 if (sk->sk_state != TCP_ESTABLISHED)
1767 target = sock_rcvlowat(sk, flags&MSG_WAITALL, size);
1768 timeo = sock_rcvtimeo(sk, flags&MSG_DONTWAIT);
1770 msg->msg_namelen = 0;
1772 /* Lock the socket to prevent queue disordering
1773 * while sleeps in memcpy_tomsg
1777 siocb->scm = &tmp_scm;
1778 memset(&tmp_scm, 0, sizeof(tmp_scm));
1781 mutex_lock(&u->readlock);
1785 struct sk_buff *skb;
1787 unix_state_lock(sk);
1788 skb = skb_dequeue(&sk->sk_receive_queue);
1790 if (copied >= target)
1794 * POSIX 1003.1g mandates this order.
1797 err = sock_error(sk);
1800 if (sk->sk_shutdown & RCV_SHUTDOWN)
1803 unix_state_unlock(sk);
1807 mutex_unlock(&u->readlock);
1809 timeo = unix_stream_data_wait(sk, timeo);
1811 if (signal_pending(current)) {
1812 err = sock_intr_errno(timeo);
1815 mutex_lock(&u->readlock);
1818 unix_state_unlock(sk);
1821 unix_state_unlock(sk);
1824 /* Never glue messages from different writers */
1825 if (memcmp(UNIXCREDS(skb), &siocb->scm->creds,
1826 sizeof(siocb->scm->creds)) != 0) {
1827 skb_queue_head(&sk->sk_receive_queue, skb);
1831 /* Copy credentials */
1832 siocb->scm->creds = *UNIXCREDS(skb);
1836 /* Copy address just once */
1838 unix_copy_addr(msg, skb->sk);
1842 chunk = min_t(unsigned int, skb->len, size);
1843 if (memcpy_toiovec(msg->msg_iov, skb->data, chunk)) {
1844 skb_queue_head(&sk->sk_receive_queue, skb);
1852 /* Mark read part of skb as used */
1853 if (!(flags & MSG_PEEK)) {
1854 skb_pull(skb, chunk);
1857 unix_detach_fds(siocb->scm, skb);
1859 /* put the skb back if we didn't use it up.. */
1861 skb_queue_head(&sk->sk_receive_queue, skb);
1870 /* It is questionable, see note in unix_dgram_recvmsg.
1873 siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1875 /* put message back and return */
1876 skb_queue_head(&sk->sk_receive_queue, skb);
1881 mutex_unlock(&u->readlock);
1882 scm_recv(sock, msg, siocb->scm, flags);
1884 return copied ? : err;
1887 static int unix_shutdown(struct socket *sock, int mode)
1889 struct sock *sk = sock->sk;
1892 mode = (mode+1)&(RCV_SHUTDOWN|SEND_SHUTDOWN);
1895 unix_state_lock(sk);
1896 sk->sk_shutdown |= mode;
1897 other = unix_peer(sk);
1900 unix_state_unlock(sk);
1901 sk->sk_state_change(sk);
1904 (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET)) {
1908 if (mode&RCV_SHUTDOWN)
1909 peer_mode |= SEND_SHUTDOWN;
1910 if (mode&SEND_SHUTDOWN)
1911 peer_mode |= RCV_SHUTDOWN;
1912 unix_state_lock(other);
1913 other->sk_shutdown |= peer_mode;
1914 unix_state_unlock(other);
1915 other->sk_state_change(other);
1916 read_lock(&other->sk_callback_lock);
1917 if (peer_mode == SHUTDOWN_MASK)
1918 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_HUP);
1919 else if (peer_mode & RCV_SHUTDOWN)
1920 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_IN);
1921 read_unlock(&other->sk_callback_lock);
1929 static int unix_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1931 struct sock *sk = sock->sk;
1937 amount = atomic_read(&sk->sk_wmem_alloc);
1938 err = put_user(amount, (int __user *)arg);
1942 struct sk_buff *skb;
1944 if (sk->sk_state == TCP_LISTEN) {
1949 spin_lock(&sk->sk_receive_queue.lock);
1950 if (sk->sk_type == SOCK_STREAM ||
1951 sk->sk_type == SOCK_SEQPACKET) {
1952 skb_queue_walk(&sk->sk_receive_queue, skb)
1955 skb = skb_peek(&sk->sk_receive_queue);
1959 spin_unlock(&sk->sk_receive_queue.lock);
1960 err = put_user(amount, (int __user *)arg);
1971 static unsigned int unix_poll(struct file *file, struct socket *sock, poll_table *wait)
1973 struct sock *sk = sock->sk;
1976 poll_wait(file, sk->sk_sleep, wait);
1979 /* exceptional events? */
1982 if (sk->sk_shutdown == SHUTDOWN_MASK)
1984 if (sk->sk_shutdown & RCV_SHUTDOWN)
1988 if (!skb_queue_empty(&sk->sk_receive_queue) ||
1989 (sk->sk_shutdown & RCV_SHUTDOWN))
1990 mask |= POLLIN | POLLRDNORM;
1992 /* Connection-based need to check for termination and startup */
1993 if ((sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) &&
1994 sk->sk_state == TCP_CLOSE)
1998 * we set writable also when the other side has shut down the
1999 * connection. This prevents stuck sockets.
2001 if (unix_writable(sk))
2002 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2007 static unsigned int unix_dgram_poll(struct file *file, struct socket *sock,
2010 struct sock *sk = sock->sk, *other;
2011 unsigned int mask, writable;
2013 poll_wait(file, sk->sk_sleep, wait);
2016 /* exceptional events? */
2017 if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue))
2019 if (sk->sk_shutdown & RCV_SHUTDOWN)
2021 if (sk->sk_shutdown == SHUTDOWN_MASK)
2025 if (!skb_queue_empty(&sk->sk_receive_queue) ||
2026 (sk->sk_shutdown & RCV_SHUTDOWN))
2027 mask |= POLLIN | POLLRDNORM;
2029 /* Connection-based need to check for termination and startup */
2030 if (sk->sk_type == SOCK_SEQPACKET) {
2031 if (sk->sk_state == TCP_CLOSE)
2033 /* connection hasn't started yet? */
2034 if (sk->sk_state == TCP_SYN_SENT)
2039 writable = unix_writable(sk);
2041 other = unix_peer_get(sk);
2043 if (unix_peer(other) != sk) {
2044 poll_wait(file, &unix_sk(other)->peer_wait,
2046 if (unix_recvq_full(other))
2055 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2057 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
2062 #ifdef CONFIG_PROC_FS
2063 static struct sock *first_unix_socket(int *i)
2065 for (*i = 0; *i <= UNIX_HASH_SIZE; (*i)++) {
2066 if (!hlist_empty(&unix_socket_table[*i]))
2067 return __sk_head(&unix_socket_table[*i]);
2072 static struct sock *next_unix_socket(int *i, struct sock *s)
2074 struct sock *next = sk_next(s);
2075 /* More in this chain? */
2078 /* Look for next non-empty chain. */
2079 for ((*i)++; *i <= UNIX_HASH_SIZE; (*i)++) {
2080 if (!hlist_empty(&unix_socket_table[*i]))
2081 return __sk_head(&unix_socket_table[*i]);
2086 struct unix_iter_state {
2087 struct seq_net_private p;
2091 static struct sock *unix_seq_idx(struct seq_file *seq, loff_t pos)
2093 struct unix_iter_state *iter = seq->private;
2097 for (s = first_unix_socket(&iter->i); s; s = next_unix_socket(&iter->i, s)) {
2098 if (sock_net(s) != seq_file_net(seq))
2107 static void *unix_seq_start(struct seq_file *seq, loff_t *pos)
2108 __acquires(unix_table_lock)
2110 spin_lock(&unix_table_lock);
2111 return *pos ? unix_seq_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2114 static void *unix_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2116 struct unix_iter_state *iter = seq->private;
2117 struct sock *sk = v;
2120 if (v == SEQ_START_TOKEN)
2121 sk = first_unix_socket(&iter->i);
2123 sk = next_unix_socket(&iter->i, sk);
2124 while (sk && (sock_net(sk) != seq_file_net(seq)))
2125 sk = next_unix_socket(&iter->i, sk);
2129 static void unix_seq_stop(struct seq_file *seq, void *v)
2130 __releases(unix_table_lock)
2132 spin_unlock(&unix_table_lock);
2135 static int unix_seq_show(struct seq_file *seq, void *v)
2138 if (v == SEQ_START_TOKEN)
2139 seq_puts(seq, "Num RefCount Protocol Flags Type St "
2143 struct unix_sock *u = unix_sk(s);
2146 seq_printf(seq, "%p: %08X %08X %08X %04X %02X %5lu",
2148 atomic_read(&s->sk_refcnt),
2150 s->sk_state == TCP_LISTEN ? __SO_ACCEPTCON : 0,
2153 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTED : SS_UNCONNECTED) :
2154 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTING : SS_DISCONNECTING),
2162 len = u->addr->len - sizeof(short);
2163 if (!UNIX_ABSTRACT(s))
2169 for ( ; i < len; i++)
2170 seq_putc(seq, u->addr->name->sun_path[i]);
2172 unix_state_unlock(s);
2173 seq_putc(seq, '\n');
2179 static const struct seq_operations unix_seq_ops = {
2180 .start = unix_seq_start,
2181 .next = unix_seq_next,
2182 .stop = unix_seq_stop,
2183 .show = unix_seq_show,
2186 static int unix_seq_open(struct inode *inode, struct file *file)
2188 return seq_open_net(inode, file, &unix_seq_ops,
2189 sizeof(struct unix_iter_state));
2192 static const struct file_operations unix_seq_fops = {
2193 .owner = THIS_MODULE,
2194 .open = unix_seq_open,
2196 .llseek = seq_lseek,
2197 .release = seq_release_net,
2202 static struct net_proto_family unix_family_ops = {
2204 .create = unix_create,
2205 .owner = THIS_MODULE,
2209 static int unix_net_init(struct net *net)
2211 int error = -ENOMEM;
2213 net->unx.sysctl_max_dgram_qlen = 10;
2214 if (unix_sysctl_register(net))
2217 #ifdef CONFIG_PROC_FS
2218 if (!proc_net_fops_create(net, "unix", 0, &unix_seq_fops)) {
2219 unix_sysctl_unregister(net);
2228 static void unix_net_exit(struct net *net)
2230 unix_sysctl_unregister(net);
2231 proc_net_remove(net, "unix");
2234 static struct pernet_operations unix_net_ops = {
2235 .init = unix_net_init,
2236 .exit = unix_net_exit,
2239 static int __init af_unix_init(void)
2242 struct sk_buff *dummy_skb;
2244 BUILD_BUG_ON(sizeof(struct unix_skb_parms) > sizeof(dummy_skb->cb));
2246 rc = proto_register(&unix_proto, 1);
2248 printk(KERN_CRIT "%s: Cannot create unix_sock SLAB cache!\n",
2253 sock_register(&unix_family_ops);
2254 register_pernet_subsys(&unix_net_ops);
2259 static void __exit af_unix_exit(void)
2261 sock_unregister(PF_UNIX);
2262 proto_unregister(&unix_proto);
2263 unregister_pernet_subsys(&unix_net_ops);
2266 /* Earlier than device_initcall() so that other drivers invoking
2267 request_module() don't end up in a loop when modprobe tries
2268 to use a UNIX socket. But later than subsys_initcall() because
2269 we depend on stuff initialised there */
2270 fs_initcall(af_unix_init);
2271 module_exit(af_unix_exit);
2273 MODULE_LICENSE("GPL");
2274 MODULE_ALIAS_NETPROTO(PF_UNIX);