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.
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 struct sock *unix_peer_get(struct sock *s)
178 unix_state_unlock(s);
182 static inline void unix_release_addr(struct unix_address *addr)
184 if (atomic_dec_and_test(&addr->refcnt))
189 * Check unix socket name:
190 * - should be not zero length.
191 * - if started by not zero, should be NULL terminated (FS object)
192 * - if started by zero, it is abstract name.
195 static int unix_mkname(struct sockaddr_un * sunaddr, int len, unsigned *hashp)
197 if (len <= sizeof(short) || len > sizeof(*sunaddr))
199 if (!sunaddr || sunaddr->sun_family != AF_UNIX)
201 if (sunaddr->sun_path[0]) {
203 * This may look like an off by one error but it is a bit more
204 * subtle. 108 is the longest valid AF_UNIX path for a binding.
205 * sun_path[108] doesnt as such exist. However in kernel space
206 * we are guaranteed that it is a valid memory location in our
207 * kernel address buffer.
209 ((char *)sunaddr)[len]=0;
210 len = strlen(sunaddr->sun_path)+1+sizeof(short);
214 *hashp = unix_hash_fold(csum_partial((char*)sunaddr, len, 0));
218 static void __unix_remove_socket(struct sock *sk)
220 sk_del_node_init(sk);
223 static void __unix_insert_socket(struct hlist_head *list, struct sock *sk)
225 BUG_TRAP(sk_unhashed(sk));
226 sk_add_node(sk, list);
229 static inline void unix_remove_socket(struct sock *sk)
231 spin_lock(&unix_table_lock);
232 __unix_remove_socket(sk);
233 spin_unlock(&unix_table_lock);
236 static inline void unix_insert_socket(struct hlist_head *list, struct sock *sk)
238 spin_lock(&unix_table_lock);
239 __unix_insert_socket(list, sk);
240 spin_unlock(&unix_table_lock);
243 static struct sock *__unix_find_socket_byname(struct net *net,
244 struct sockaddr_un *sunname,
245 int len, int type, unsigned hash)
248 struct hlist_node *node;
250 sk_for_each(s, node, &unix_socket_table[hash ^ type]) {
251 struct unix_sock *u = unix_sk(s);
253 if (!net_eq(sock_net(s), net))
256 if (u->addr->len == len &&
257 !memcmp(u->addr->name, sunname, len))
265 static inline struct sock *unix_find_socket_byname(struct net *net,
266 struct sockaddr_un *sunname,
272 spin_lock(&unix_table_lock);
273 s = __unix_find_socket_byname(net, sunname, len, type, hash);
276 spin_unlock(&unix_table_lock);
280 static struct sock *unix_find_socket_byinode(struct net *net, struct inode *i)
283 struct hlist_node *node;
285 spin_lock(&unix_table_lock);
287 &unix_socket_table[i->i_ino & (UNIX_HASH_SIZE - 1)]) {
288 struct dentry *dentry = unix_sk(s)->dentry;
290 if (!net_eq(sock_net(s), net))
293 if(dentry && dentry->d_inode == i)
301 spin_unlock(&unix_table_lock);
305 static inline int unix_writable(struct sock *sk)
307 return (atomic_read(&sk->sk_wmem_alloc) << 2) <= sk->sk_sndbuf;
310 static void unix_write_space(struct sock *sk)
312 read_lock(&sk->sk_callback_lock);
313 if (unix_writable(sk)) {
314 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
315 wake_up_interruptible_sync(sk->sk_sleep);
316 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
318 read_unlock(&sk->sk_callback_lock);
321 /* When dgram socket disconnects (or changes its peer), we clear its receive
322 * queue of packets arrived from previous peer. First, it allows to do
323 * flow control based only on wmem_alloc; second, sk connected to peer
324 * may receive messages only from that peer. */
325 static void unix_dgram_disconnected(struct sock *sk, struct sock *other)
327 if (!skb_queue_empty(&sk->sk_receive_queue)) {
328 skb_queue_purge(&sk->sk_receive_queue);
329 wake_up_interruptible_all(&unix_sk(sk)->peer_wait);
331 /* If one link of bidirectional dgram pipe is disconnected,
332 * we signal error. Messages are lost. Do not make this,
333 * when peer was not connected to us.
335 if (!sock_flag(other, SOCK_DEAD) && unix_peer(other) == sk) {
336 other->sk_err = ECONNRESET;
337 other->sk_error_report(other);
342 static void unix_sock_destructor(struct sock *sk)
344 struct unix_sock *u = unix_sk(sk);
346 skb_queue_purge(&sk->sk_receive_queue);
348 BUG_TRAP(!atomic_read(&sk->sk_wmem_alloc));
349 BUG_TRAP(sk_unhashed(sk));
350 BUG_TRAP(!sk->sk_socket);
351 if (!sock_flag(sk, SOCK_DEAD)) {
352 printk("Attempt to release alive unix socket: %p\n", sk);
357 unix_release_addr(u->addr);
359 atomic_dec(&unix_nr_socks);
360 #ifdef UNIX_REFCNT_DEBUG
361 printk(KERN_DEBUG "UNIX %p is destroyed, %d are still alive.\n", sk, atomic_read(&unix_nr_socks));
365 static int unix_release_sock (struct sock *sk, int embrion)
367 struct unix_sock *u = unix_sk(sk);
368 struct dentry *dentry;
369 struct vfsmount *mnt;
374 unix_remove_socket(sk);
379 sk->sk_shutdown = SHUTDOWN_MASK;
384 state = sk->sk_state;
385 sk->sk_state = TCP_CLOSE;
386 unix_state_unlock(sk);
388 wake_up_interruptible_all(&u->peer_wait);
390 skpair=unix_peer(sk);
393 if (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) {
394 unix_state_lock(skpair);
396 skpair->sk_shutdown = SHUTDOWN_MASK;
397 if (!skb_queue_empty(&sk->sk_receive_queue) || embrion)
398 skpair->sk_err = ECONNRESET;
399 unix_state_unlock(skpair);
400 skpair->sk_state_change(skpair);
401 read_lock(&skpair->sk_callback_lock);
402 sk_wake_async(skpair, SOCK_WAKE_WAITD, POLL_HUP);
403 read_unlock(&skpair->sk_callback_lock);
405 sock_put(skpair); /* It may now die */
406 unix_peer(sk) = NULL;
409 /* Try to flush out this socket. Throw out buffers at least */
411 while ((skb = skb_dequeue(&sk->sk_receive_queue)) != NULL) {
412 if (state==TCP_LISTEN)
413 unix_release_sock(skb->sk, 1);
414 /* passed fds are erased in the kfree_skb hook */
425 /* ---- Socket is dead now and most probably destroyed ---- */
428 * Fixme: BSD difference: In BSD all sockets connected to use get
429 * ECONNRESET and we die on the spot. In Linux we behave
430 * like files and pipes do and wait for the last
433 * Can't we simply set sock->err?
435 * What the above comment does talk about? --ANK(980817)
438 if (unix_tot_inflight)
439 unix_gc(); /* Garbage collect fds */
444 static int unix_listen(struct socket *sock, int backlog)
447 struct sock *sk = sock->sk;
448 struct unix_sock *u = unix_sk(sk);
451 if (sock->type!=SOCK_STREAM && sock->type!=SOCK_SEQPACKET)
452 goto out; /* Only stream/seqpacket sockets accept */
455 goto out; /* No listens on an unbound socket */
457 if (sk->sk_state != TCP_CLOSE && sk->sk_state != TCP_LISTEN)
459 if (backlog > sk->sk_max_ack_backlog)
460 wake_up_interruptible_all(&u->peer_wait);
461 sk->sk_max_ack_backlog = backlog;
462 sk->sk_state = TCP_LISTEN;
463 /* set credentials so connect can copy them */
464 sk->sk_peercred.pid = task_tgid_vnr(current);
465 sk->sk_peercred.uid = current->euid;
466 sk->sk_peercred.gid = current->egid;
470 unix_state_unlock(sk);
475 static int unix_release(struct socket *);
476 static int unix_bind(struct socket *, struct sockaddr *, int);
477 static int unix_stream_connect(struct socket *, struct sockaddr *,
478 int addr_len, int flags);
479 static int unix_socketpair(struct socket *, struct socket *);
480 static int unix_accept(struct socket *, struct socket *, int);
481 static int unix_getname(struct socket *, struct sockaddr *, int *, int);
482 static unsigned int unix_poll(struct file *, struct socket *, poll_table *);
483 static int unix_ioctl(struct socket *, unsigned int, unsigned long);
484 static int unix_shutdown(struct socket *, int);
485 static int unix_stream_sendmsg(struct kiocb *, struct socket *,
486 struct msghdr *, size_t);
487 static int unix_stream_recvmsg(struct kiocb *, struct socket *,
488 struct msghdr *, size_t, int);
489 static int unix_dgram_sendmsg(struct kiocb *, struct socket *,
490 struct msghdr *, size_t);
491 static int unix_dgram_recvmsg(struct kiocb *, struct socket *,
492 struct msghdr *, size_t, int);
493 static int unix_dgram_connect(struct socket *, struct sockaddr *,
495 static int unix_seqpacket_sendmsg(struct kiocb *, struct socket *,
496 struct msghdr *, size_t);
498 static const struct proto_ops unix_stream_ops = {
500 .owner = THIS_MODULE,
501 .release = unix_release,
503 .connect = unix_stream_connect,
504 .socketpair = unix_socketpair,
505 .accept = unix_accept,
506 .getname = unix_getname,
509 .listen = unix_listen,
510 .shutdown = unix_shutdown,
511 .setsockopt = sock_no_setsockopt,
512 .getsockopt = sock_no_getsockopt,
513 .sendmsg = unix_stream_sendmsg,
514 .recvmsg = unix_stream_recvmsg,
515 .mmap = sock_no_mmap,
516 .sendpage = sock_no_sendpage,
519 static const struct proto_ops unix_dgram_ops = {
521 .owner = THIS_MODULE,
522 .release = unix_release,
524 .connect = unix_dgram_connect,
525 .socketpair = unix_socketpair,
526 .accept = sock_no_accept,
527 .getname = unix_getname,
528 .poll = datagram_poll,
530 .listen = sock_no_listen,
531 .shutdown = unix_shutdown,
532 .setsockopt = sock_no_setsockopt,
533 .getsockopt = sock_no_getsockopt,
534 .sendmsg = unix_dgram_sendmsg,
535 .recvmsg = unix_dgram_recvmsg,
536 .mmap = sock_no_mmap,
537 .sendpage = sock_no_sendpage,
540 static const struct proto_ops unix_seqpacket_ops = {
542 .owner = THIS_MODULE,
543 .release = unix_release,
545 .connect = unix_stream_connect,
546 .socketpair = unix_socketpair,
547 .accept = unix_accept,
548 .getname = unix_getname,
549 .poll = datagram_poll,
551 .listen = unix_listen,
552 .shutdown = unix_shutdown,
553 .setsockopt = sock_no_setsockopt,
554 .getsockopt = sock_no_getsockopt,
555 .sendmsg = unix_seqpacket_sendmsg,
556 .recvmsg = unix_dgram_recvmsg,
557 .mmap = sock_no_mmap,
558 .sendpage = sock_no_sendpage,
561 static struct proto unix_proto = {
563 .owner = THIS_MODULE,
564 .obj_size = sizeof(struct unix_sock),
568 * AF_UNIX sockets do not interact with hardware, hence they
569 * dont trigger interrupts - so it's safe for them to have
570 * bh-unsafe locking for their sk_receive_queue.lock. Split off
571 * this special lock-class by reinitializing the spinlock key:
573 static struct lock_class_key af_unix_sk_receive_queue_lock_key;
575 static struct sock * unix_create1(struct net *net, struct socket *sock)
577 struct sock *sk = NULL;
580 atomic_inc(&unix_nr_socks);
581 if (atomic_read(&unix_nr_socks) > 2 * get_max_files())
584 sk = sk_alloc(net, PF_UNIX, GFP_KERNEL, &unix_proto);
588 sock_init_data(sock,sk);
589 lockdep_set_class(&sk->sk_receive_queue.lock,
590 &af_unix_sk_receive_queue_lock_key);
592 sk->sk_write_space = unix_write_space;
593 sk->sk_max_ack_backlog = net->unx.sysctl_max_dgram_qlen;
594 sk->sk_destruct = unix_sock_destructor;
598 spin_lock_init(&u->lock);
599 atomic_set(&u->inflight, 0);
600 INIT_LIST_HEAD(&u->link);
601 mutex_init(&u->readlock); /* single task reading lock */
602 init_waitqueue_head(&u->peer_wait);
603 unix_insert_socket(unix_sockets_unbound, sk);
606 atomic_dec(&unix_nr_socks);
610 static int unix_create(struct net *net, struct socket *sock, int protocol)
612 if (protocol && protocol != PF_UNIX)
613 return -EPROTONOSUPPORT;
615 sock->state = SS_UNCONNECTED;
617 switch (sock->type) {
619 sock->ops = &unix_stream_ops;
622 * Believe it or not BSD has AF_UNIX, SOCK_RAW though
626 sock->type=SOCK_DGRAM;
628 sock->ops = &unix_dgram_ops;
631 sock->ops = &unix_seqpacket_ops;
634 return -ESOCKTNOSUPPORT;
637 return unix_create1(net, sock) ? 0 : -ENOMEM;
640 static int unix_release(struct socket *sock)
642 struct sock *sk = sock->sk;
649 return unix_release_sock (sk, 0);
652 static int unix_autobind(struct socket *sock)
654 struct sock *sk = sock->sk;
655 struct net *net = sock_net(sk);
656 struct unix_sock *u = unix_sk(sk);
657 static u32 ordernum = 1;
658 struct unix_address * addr;
661 mutex_lock(&u->readlock);
668 addr = kzalloc(sizeof(*addr) + sizeof(short) + 16, GFP_KERNEL);
672 addr->name->sun_family = AF_UNIX;
673 atomic_set(&addr->refcnt, 1);
676 addr->len = sprintf(addr->name->sun_path+1, "%05x", ordernum) + 1 + sizeof(short);
677 addr->hash = unix_hash_fold(csum_partial((void*)addr->name, addr->len, 0));
679 spin_lock(&unix_table_lock);
680 ordernum = (ordernum+1)&0xFFFFF;
682 if (__unix_find_socket_byname(net, addr->name, addr->len, sock->type,
684 spin_unlock(&unix_table_lock);
685 /* Sanity yield. It is unusual case, but yet... */
686 if (!(ordernum&0xFF))
690 addr->hash ^= sk->sk_type;
692 __unix_remove_socket(sk);
694 __unix_insert_socket(&unix_socket_table[addr->hash], sk);
695 spin_unlock(&unix_table_lock);
698 out: mutex_unlock(&u->readlock);
702 static struct sock *unix_find_other(struct net *net,
703 struct sockaddr_un *sunname, int len,
704 int type, unsigned hash, int *error)
710 if (sunname->sun_path[0]) {
711 err = path_lookup(sunname->sun_path, LOOKUP_FOLLOW, &nd);
714 err = vfs_permission(&nd, MAY_WRITE);
719 if (!S_ISSOCK(nd.path.dentry->d_inode->i_mode))
721 u = unix_find_socket_byinode(net, nd.path.dentry->d_inode);
725 if (u->sk_type == type)
726 touch_atime(nd.path.mnt, nd.path.dentry);
731 if (u->sk_type != type) {
737 u=unix_find_socket_byname(net, sunname, len, type, hash);
739 struct dentry *dentry;
740 dentry = unix_sk(u)->dentry;
742 touch_atime(unix_sk(u)->mnt, dentry);
756 static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
758 struct sock *sk = sock->sk;
759 struct net *net = sock_net(sk);
760 struct unix_sock *u = unix_sk(sk);
761 struct sockaddr_un *sunaddr=(struct sockaddr_un *)uaddr;
762 struct dentry * dentry = NULL;
766 struct unix_address *addr;
767 struct hlist_head *list;
770 if (sunaddr->sun_family != AF_UNIX)
773 if (addr_len==sizeof(short)) {
774 err = unix_autobind(sock);
778 err = unix_mkname(sunaddr, addr_len, &hash);
783 mutex_lock(&u->readlock);
790 addr = kmalloc(sizeof(*addr)+addr_len, GFP_KERNEL);
794 memcpy(addr->name, sunaddr, addr_len);
795 addr->len = addr_len;
796 addr->hash = hash ^ sk->sk_type;
797 atomic_set(&addr->refcnt, 1);
799 if (sunaddr->sun_path[0]) {
803 * Get the parent directory, calculate the hash for last
806 err = path_lookup(sunaddr->sun_path, LOOKUP_PARENT, &nd);
808 goto out_mknod_parent;
810 dentry = lookup_create(&nd, 0);
811 err = PTR_ERR(dentry);
813 goto out_mknod_unlock;
816 * All right, let's create it.
819 (SOCK_INODE(sock)->i_mode & ~current->fs->umask);
820 err = mnt_want_write(nd.path.mnt);
823 err = vfs_mknod(nd.path.dentry->d_inode, dentry, mode, 0);
824 mnt_drop_write(nd.path.mnt);
827 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
828 dput(nd.path.dentry);
829 nd.path.dentry = dentry;
831 addr->hash = UNIX_HASH_SIZE;
834 spin_lock(&unix_table_lock);
836 if (!sunaddr->sun_path[0]) {
838 if (__unix_find_socket_byname(net, sunaddr, addr_len,
839 sk->sk_type, hash)) {
840 unix_release_addr(addr);
844 list = &unix_socket_table[addr->hash];
846 list = &unix_socket_table[dentry->d_inode->i_ino & (UNIX_HASH_SIZE-1)];
847 u->dentry = nd.path.dentry;
848 u->mnt = nd.path.mnt;
852 __unix_remove_socket(sk);
854 __unix_insert_socket(list, sk);
857 spin_unlock(&unix_table_lock);
859 mutex_unlock(&u->readlock);
866 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
871 unix_release_addr(addr);
875 static void unix_state_double_lock(struct sock *sk1, struct sock *sk2)
877 if (unlikely(sk1 == sk2) || !sk2) {
878 unix_state_lock(sk1);
882 unix_state_lock(sk1);
883 unix_state_lock_nested(sk2);
885 unix_state_lock(sk2);
886 unix_state_lock_nested(sk1);
890 static void unix_state_double_unlock(struct sock *sk1, struct sock *sk2)
892 if (unlikely(sk1 == sk2) || !sk2) {
893 unix_state_unlock(sk1);
896 unix_state_unlock(sk1);
897 unix_state_unlock(sk2);
900 static int unix_dgram_connect(struct socket *sock, struct sockaddr *addr,
903 struct sock *sk = sock->sk;
904 struct net *net = sock_net(sk);
905 struct sockaddr_un *sunaddr=(struct sockaddr_un*)addr;
910 if (addr->sa_family != AF_UNSPEC) {
911 err = unix_mkname(sunaddr, alen, &hash);
916 if (test_bit(SOCK_PASSCRED, &sock->flags) &&
917 !unix_sk(sk)->addr && (err = unix_autobind(sock)) != 0)
921 other=unix_find_other(net, sunaddr, alen, sock->type, hash, &err);
925 unix_state_double_lock(sk, other);
927 /* Apparently VFS overslept socket death. Retry. */
928 if (sock_flag(other, SOCK_DEAD)) {
929 unix_state_double_unlock(sk, other);
935 if (!unix_may_send(sk, other))
938 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
944 * 1003.1g breaking connected state with AF_UNSPEC
947 unix_state_double_lock(sk, other);
951 * If it was connected, reconnect.
954 struct sock *old_peer = unix_peer(sk);
956 unix_state_double_unlock(sk, other);
958 if (other != old_peer)
959 unix_dgram_disconnected(sk, old_peer);
963 unix_state_double_unlock(sk, other);
968 unix_state_double_unlock(sk, other);
974 static long unix_wait_for_peer(struct sock *other, long timeo)
976 struct unix_sock *u = unix_sk(other);
980 prepare_to_wait_exclusive(&u->peer_wait, &wait, TASK_INTERRUPTIBLE);
982 sched = !sock_flag(other, SOCK_DEAD) &&
983 !(other->sk_shutdown & RCV_SHUTDOWN) &&
984 (skb_queue_len(&other->sk_receive_queue) >
985 other->sk_max_ack_backlog);
987 unix_state_unlock(other);
990 timeo = schedule_timeout(timeo);
992 finish_wait(&u->peer_wait, &wait);
996 static int unix_stream_connect(struct socket *sock, struct sockaddr *uaddr,
997 int addr_len, int flags)
999 struct sockaddr_un *sunaddr=(struct sockaddr_un *)uaddr;
1000 struct sock *sk = sock->sk;
1001 struct net *net = sock_net(sk);
1002 struct unix_sock *u = unix_sk(sk), *newu, *otheru;
1003 struct sock *newsk = NULL;
1004 struct sock *other = NULL;
1005 struct sk_buff *skb = NULL;
1011 err = unix_mkname(sunaddr, addr_len, &hash);
1016 if (test_bit(SOCK_PASSCRED, &sock->flags)
1017 && !u->addr && (err = unix_autobind(sock)) != 0)
1020 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
1022 /* First of all allocate resources.
1023 If we will make it after state is locked,
1024 we will have to recheck all again in any case.
1029 /* create new sock for complete connection */
1030 newsk = unix_create1(sock_net(sk), NULL);
1034 /* Allocate skb for sending to listening sock */
1035 skb = sock_wmalloc(newsk, 1, 0, GFP_KERNEL);
1040 /* Find listening sock. */
1041 other = unix_find_other(net, sunaddr, addr_len, sk->sk_type, hash, &err);
1045 /* Latch state of peer */
1046 unix_state_lock(other);
1048 /* Apparently VFS overslept socket death. Retry. */
1049 if (sock_flag(other, SOCK_DEAD)) {
1050 unix_state_unlock(other);
1055 err = -ECONNREFUSED;
1056 if (other->sk_state != TCP_LISTEN)
1059 if (skb_queue_len(&other->sk_receive_queue) >
1060 other->sk_max_ack_backlog) {
1065 timeo = unix_wait_for_peer(other, timeo);
1067 err = sock_intr_errno(timeo);
1068 if (signal_pending(current))
1076 It is tricky place. We need to grab write lock and cannot
1077 drop lock on peer. It is dangerous because deadlock is
1078 possible. Connect to self case and simultaneous
1079 attempt to connect are eliminated by checking socket
1080 state. other is TCP_LISTEN, if sk is TCP_LISTEN we
1081 check this before attempt to grab lock.
1083 Well, and we have to recheck the state after socket locked.
1089 /* This is ok... continue with connect */
1091 case TCP_ESTABLISHED:
1092 /* Socket is already connected */
1100 unix_state_lock_nested(sk);
1102 if (sk->sk_state != st) {
1103 unix_state_unlock(sk);
1104 unix_state_unlock(other);
1109 err = security_unix_stream_connect(sock, other->sk_socket, newsk);
1111 unix_state_unlock(sk);
1115 /* The way is open! Fastly set all the necessary fields... */
1118 unix_peer(newsk) = sk;
1119 newsk->sk_state = TCP_ESTABLISHED;
1120 newsk->sk_type = sk->sk_type;
1121 newsk->sk_peercred.pid = task_tgid_vnr(current);
1122 newsk->sk_peercred.uid = current->euid;
1123 newsk->sk_peercred.gid = current->egid;
1124 newu = unix_sk(newsk);
1125 newsk->sk_sleep = &newu->peer_wait;
1126 otheru = unix_sk(other);
1128 /* copy address information from listening to new sock*/
1130 atomic_inc(&otheru->addr->refcnt);
1131 newu->addr = otheru->addr;
1133 if (otheru->dentry) {
1134 newu->dentry = dget(otheru->dentry);
1135 newu->mnt = mntget(otheru->mnt);
1138 /* Set credentials */
1139 sk->sk_peercred = other->sk_peercred;
1141 sock->state = SS_CONNECTED;
1142 sk->sk_state = TCP_ESTABLISHED;
1145 smp_mb__after_atomic_inc(); /* sock_hold() does an atomic_inc() */
1146 unix_peer(sk) = newsk;
1148 unix_state_unlock(sk);
1150 /* take ten and and send info to listening sock */
1151 spin_lock(&other->sk_receive_queue.lock);
1152 __skb_queue_tail(&other->sk_receive_queue, skb);
1153 spin_unlock(&other->sk_receive_queue.lock);
1154 unix_state_unlock(other);
1155 other->sk_data_ready(other, 0);
1161 unix_state_unlock(other);
1167 unix_release_sock(newsk, 0);
1173 static int unix_socketpair(struct socket *socka, struct socket *sockb)
1175 struct sock *ska=socka->sk, *skb = sockb->sk;
1177 /* Join our sockets back to back */
1182 ska->sk_peercred.pid = skb->sk_peercred.pid = task_tgid_vnr(current);
1183 ska->sk_peercred.uid = skb->sk_peercred.uid = current->euid;
1184 ska->sk_peercred.gid = skb->sk_peercred.gid = current->egid;
1186 if (ska->sk_type != SOCK_DGRAM) {
1187 ska->sk_state = TCP_ESTABLISHED;
1188 skb->sk_state = TCP_ESTABLISHED;
1189 socka->state = SS_CONNECTED;
1190 sockb->state = SS_CONNECTED;
1195 static int unix_accept(struct socket *sock, struct socket *newsock, int flags)
1197 struct sock *sk = sock->sk;
1199 struct sk_buff *skb;
1203 if (sock->type!=SOCK_STREAM && sock->type!=SOCK_SEQPACKET)
1207 if (sk->sk_state != TCP_LISTEN)
1210 /* If socket state is TCP_LISTEN it cannot change (for now...),
1211 * so that no locks are necessary.
1214 skb = skb_recv_datagram(sk, 0, flags&O_NONBLOCK, &err);
1216 /* This means receive shutdown. */
1223 skb_free_datagram(sk, skb);
1224 wake_up_interruptible(&unix_sk(sk)->peer_wait);
1226 /* attach accepted sock to socket */
1227 unix_state_lock(tsk);
1228 newsock->state = SS_CONNECTED;
1229 sock_graft(tsk, newsock);
1230 unix_state_unlock(tsk);
1238 static int unix_getname(struct socket *sock, struct sockaddr *uaddr, int *uaddr_len, int peer)
1240 struct sock *sk = sock->sk;
1241 struct unix_sock *u;
1242 struct sockaddr_un *sunaddr=(struct sockaddr_un *)uaddr;
1246 sk = unix_peer_get(sk);
1257 unix_state_lock(sk);
1259 sunaddr->sun_family = AF_UNIX;
1260 sunaddr->sun_path[0] = 0;
1261 *uaddr_len = sizeof(short);
1263 struct unix_address *addr = u->addr;
1265 *uaddr_len = addr->len;
1266 memcpy(sunaddr, addr->name, *uaddr_len);
1268 unix_state_unlock(sk);
1274 static void unix_detach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1278 scm->fp = UNIXCB(skb).fp;
1279 skb->destructor = sock_wfree;
1280 UNIXCB(skb).fp = NULL;
1282 for (i=scm->fp->count-1; i>=0; i--)
1283 unix_notinflight(scm->fp->fp[i]);
1286 static void unix_destruct_fds(struct sk_buff *skb)
1288 struct scm_cookie scm;
1289 memset(&scm, 0, sizeof(scm));
1290 unix_detach_fds(&scm, skb);
1292 /* Alas, it calls VFS */
1293 /* So fscking what? fput() had been SMP-safe since the last Summer */
1298 static void unix_attach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1301 for (i=scm->fp->count-1; i>=0; i--)
1302 unix_inflight(scm->fp->fp[i]);
1303 UNIXCB(skb).fp = scm->fp;
1304 skb->destructor = unix_destruct_fds;
1309 * Send AF_UNIX data.
1312 static int unix_dgram_sendmsg(struct kiocb *kiocb, struct socket *sock,
1313 struct msghdr *msg, size_t len)
1315 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1316 struct sock *sk = sock->sk;
1317 struct net *net = sock_net(sk);
1318 struct unix_sock *u = unix_sk(sk);
1319 struct sockaddr_un *sunaddr=msg->msg_name;
1320 struct sock *other = NULL;
1321 int namelen = 0; /* fake GCC */
1324 struct sk_buff *skb;
1326 struct scm_cookie tmp_scm;
1328 if (NULL == siocb->scm)
1329 siocb->scm = &tmp_scm;
1330 err = scm_send(sock, msg, siocb->scm);
1335 if (msg->msg_flags&MSG_OOB)
1338 if (msg->msg_namelen) {
1339 err = unix_mkname(sunaddr, msg->msg_namelen, &hash);
1346 other = unix_peer_get(sk);
1351 if (test_bit(SOCK_PASSCRED, &sock->flags)
1352 && !u->addr && (err = unix_autobind(sock)) != 0)
1356 if (len > sk->sk_sndbuf - 32)
1359 skb = sock_alloc_send_skb(sk, len, msg->msg_flags&MSG_DONTWAIT, &err);
1363 memcpy(UNIXCREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1365 unix_attach_fds(siocb->scm, skb);
1366 unix_get_secdata(siocb->scm, skb);
1368 skb_reset_transport_header(skb);
1369 err = memcpy_fromiovec(skb_put(skb,len), msg->msg_iov, len);
1373 timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1378 if (sunaddr == NULL)
1381 other = unix_find_other(net, sunaddr, namelen, sk->sk_type,
1387 unix_state_lock(other);
1389 if (!unix_may_send(sk, other))
1392 if (sock_flag(other, SOCK_DEAD)) {
1394 * Check with 1003.1g - what should
1397 unix_state_unlock(other);
1401 unix_state_lock(sk);
1402 if (unix_peer(sk) == other) {
1404 unix_state_unlock(sk);
1406 unix_dgram_disconnected(sk, other);
1408 err = -ECONNREFUSED;
1410 unix_state_unlock(sk);
1420 if (other->sk_shutdown & RCV_SHUTDOWN)
1423 if (sk->sk_type != SOCK_SEQPACKET) {
1424 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1429 if (unix_peer(other) != sk &&
1430 (skb_queue_len(&other->sk_receive_queue) >
1431 other->sk_max_ack_backlog)) {
1437 timeo = unix_wait_for_peer(other, timeo);
1439 err = sock_intr_errno(timeo);
1440 if (signal_pending(current))
1446 skb_queue_tail(&other->sk_receive_queue, skb);
1447 unix_state_unlock(other);
1448 other->sk_data_ready(other, len);
1450 scm_destroy(siocb->scm);
1454 unix_state_unlock(other);
1460 scm_destroy(siocb->scm);
1465 static int unix_stream_sendmsg(struct kiocb *kiocb, struct socket *sock,
1466 struct msghdr *msg, size_t len)
1468 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1469 struct sock *sk = sock->sk;
1470 struct sock *other = NULL;
1471 struct sockaddr_un *sunaddr=msg->msg_name;
1473 struct sk_buff *skb;
1475 struct scm_cookie tmp_scm;
1477 if (NULL == siocb->scm)
1478 siocb->scm = &tmp_scm;
1479 err = scm_send(sock, msg, siocb->scm);
1484 if (msg->msg_flags&MSG_OOB)
1487 if (msg->msg_namelen) {
1488 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
1493 other = unix_peer(sk);
1498 if (sk->sk_shutdown & SEND_SHUTDOWN)
1504 * Optimisation for the fact that under 0.01% of X
1505 * messages typically need breaking up.
1510 /* Keep two messages in the pipe so it schedules better */
1511 if (size > ((sk->sk_sndbuf >> 1) - 64))
1512 size = (sk->sk_sndbuf >> 1) - 64;
1514 if (size > SKB_MAX_ALLOC)
1515 size = SKB_MAX_ALLOC;
1521 skb=sock_alloc_send_skb(sk,size,msg->msg_flags&MSG_DONTWAIT, &err);
1527 * If you pass two values to the sock_alloc_send_skb
1528 * it tries to grab the large buffer with GFP_NOFS
1529 * (which can fail easily), and if it fails grab the
1530 * fallback size buffer which is under a page and will
1533 size = min_t(int, size, skb_tailroom(skb));
1535 memcpy(UNIXCREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1537 unix_attach_fds(siocb->scm, skb);
1539 if ((err = memcpy_fromiovec(skb_put(skb,size), msg->msg_iov, size)) != 0) {
1544 unix_state_lock(other);
1546 if (sock_flag(other, SOCK_DEAD) ||
1547 (other->sk_shutdown & RCV_SHUTDOWN))
1550 skb_queue_tail(&other->sk_receive_queue, skb);
1551 unix_state_unlock(other);
1552 other->sk_data_ready(other, size);
1556 scm_destroy(siocb->scm);
1562 unix_state_unlock(other);
1565 if (sent==0 && !(msg->msg_flags&MSG_NOSIGNAL))
1566 send_sig(SIGPIPE,current,0);
1569 scm_destroy(siocb->scm);
1571 return sent ? : err;
1574 static int unix_seqpacket_sendmsg(struct kiocb *kiocb, struct socket *sock,
1575 struct msghdr *msg, size_t len)
1578 struct sock *sk = sock->sk;
1580 err = sock_error(sk);
1584 if (sk->sk_state != TCP_ESTABLISHED)
1587 if (msg->msg_namelen)
1588 msg->msg_namelen = 0;
1590 return unix_dgram_sendmsg(kiocb, sock, msg, len);
1593 static void unix_copy_addr(struct msghdr *msg, struct sock *sk)
1595 struct unix_sock *u = unix_sk(sk);
1597 msg->msg_namelen = 0;
1599 msg->msg_namelen = u->addr->len;
1600 memcpy(msg->msg_name, u->addr->name, u->addr->len);
1604 static int unix_dgram_recvmsg(struct kiocb *iocb, struct socket *sock,
1605 struct msghdr *msg, size_t size,
1608 struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1609 struct scm_cookie tmp_scm;
1610 struct sock *sk = sock->sk;
1611 struct unix_sock *u = unix_sk(sk);
1612 int noblock = flags & MSG_DONTWAIT;
1613 struct sk_buff *skb;
1620 msg->msg_namelen = 0;
1622 mutex_lock(&u->readlock);
1624 skb = skb_recv_datagram(sk, flags, noblock, &err);
1626 unix_state_lock(sk);
1627 /* Signal EOF on disconnected non-blocking SEQPACKET socket. */
1628 if (sk->sk_type == SOCK_SEQPACKET && err == -EAGAIN &&
1629 (sk->sk_shutdown & RCV_SHUTDOWN))
1631 unix_state_unlock(sk);
1635 wake_up_interruptible_sync(&u->peer_wait);
1638 unix_copy_addr(msg, skb->sk);
1640 if (size > skb->len)
1642 else if (size < skb->len)
1643 msg->msg_flags |= MSG_TRUNC;
1645 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, size);
1650 siocb->scm = &tmp_scm;
1651 memset(&tmp_scm, 0, sizeof(tmp_scm));
1653 siocb->scm->creds = *UNIXCREDS(skb);
1654 unix_set_secdata(siocb->scm, skb);
1656 if (!(flags & MSG_PEEK))
1659 unix_detach_fds(siocb->scm, skb);
1663 /* It is questionable: on PEEK we could:
1664 - do not return fds - good, but too simple 8)
1665 - return fds, and do not return them on read (old strategy,
1667 - clone fds (I chose it for now, it is the most universal
1670 POSIX 1003.1g does not actually define this clearly
1671 at all. POSIX 1003.1g doesn't define a lot of things
1676 siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1680 scm_recv(sock, msg, siocb->scm, flags);
1683 skb_free_datagram(sk,skb);
1685 mutex_unlock(&u->readlock);
1691 * Sleep until data has arrive. But check for races..
1694 static long unix_stream_data_wait(struct sock * sk, long timeo)
1698 unix_state_lock(sk);
1701 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1703 if (!skb_queue_empty(&sk->sk_receive_queue) ||
1705 (sk->sk_shutdown & RCV_SHUTDOWN) ||
1706 signal_pending(current) ||
1710 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1711 unix_state_unlock(sk);
1712 timeo = schedule_timeout(timeo);
1713 unix_state_lock(sk);
1714 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1717 finish_wait(sk->sk_sleep, &wait);
1718 unix_state_unlock(sk);
1724 static int unix_stream_recvmsg(struct kiocb *iocb, struct socket *sock,
1725 struct msghdr *msg, size_t size,
1728 struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1729 struct scm_cookie tmp_scm;
1730 struct sock *sk = sock->sk;
1731 struct unix_sock *u = unix_sk(sk);
1732 struct sockaddr_un *sunaddr=msg->msg_name;
1734 int check_creds = 0;
1740 if (sk->sk_state != TCP_ESTABLISHED)
1747 target = sock_rcvlowat(sk, flags&MSG_WAITALL, size);
1748 timeo = sock_rcvtimeo(sk, flags&MSG_DONTWAIT);
1750 msg->msg_namelen = 0;
1752 /* Lock the socket to prevent queue disordering
1753 * while sleeps in memcpy_tomsg
1757 siocb->scm = &tmp_scm;
1758 memset(&tmp_scm, 0, sizeof(tmp_scm));
1761 mutex_lock(&u->readlock);
1766 struct sk_buff *skb;
1768 unix_state_lock(sk);
1769 skb = skb_dequeue(&sk->sk_receive_queue);
1772 if (copied >= target)
1776 * POSIX 1003.1g mandates this order.
1779 if ((err = sock_error(sk)) != 0)
1781 if (sk->sk_shutdown & RCV_SHUTDOWN)
1784 unix_state_unlock(sk);
1788 mutex_unlock(&u->readlock);
1790 timeo = unix_stream_data_wait(sk, timeo);
1792 if (signal_pending(current)) {
1793 err = sock_intr_errno(timeo);
1796 mutex_lock(&u->readlock);
1799 unix_state_unlock(sk);
1802 unix_state_unlock(sk);
1805 /* Never glue messages from different writers */
1806 if (memcmp(UNIXCREDS(skb), &siocb->scm->creds, sizeof(siocb->scm->creds)) != 0) {
1807 skb_queue_head(&sk->sk_receive_queue, skb);
1811 /* Copy credentials */
1812 siocb->scm->creds = *UNIXCREDS(skb);
1816 /* Copy address just once */
1819 unix_copy_addr(msg, skb->sk);
1823 chunk = min_t(unsigned int, skb->len, size);
1824 if (memcpy_toiovec(msg->msg_iov, skb->data, chunk)) {
1825 skb_queue_head(&sk->sk_receive_queue, skb);
1833 /* Mark read part of skb as used */
1834 if (!(flags & MSG_PEEK))
1836 skb_pull(skb, chunk);
1839 unix_detach_fds(siocb->scm, skb);
1841 /* put the skb back if we didn't use it up.. */
1844 skb_queue_head(&sk->sk_receive_queue, skb);
1855 /* It is questionable, see note in unix_dgram_recvmsg.
1858 siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1860 /* put message back and return */
1861 skb_queue_head(&sk->sk_receive_queue, skb);
1866 mutex_unlock(&u->readlock);
1867 scm_recv(sock, msg, siocb->scm, flags);
1869 return copied ? : err;
1872 static int unix_shutdown(struct socket *sock, int mode)
1874 struct sock *sk = sock->sk;
1877 mode = (mode+1)&(RCV_SHUTDOWN|SEND_SHUTDOWN);
1880 unix_state_lock(sk);
1881 sk->sk_shutdown |= mode;
1882 other=unix_peer(sk);
1885 unix_state_unlock(sk);
1886 sk->sk_state_change(sk);
1889 (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET)) {
1893 if (mode&RCV_SHUTDOWN)
1894 peer_mode |= SEND_SHUTDOWN;
1895 if (mode&SEND_SHUTDOWN)
1896 peer_mode |= RCV_SHUTDOWN;
1897 unix_state_lock(other);
1898 other->sk_shutdown |= peer_mode;
1899 unix_state_unlock(other);
1900 other->sk_state_change(other);
1901 read_lock(&other->sk_callback_lock);
1902 if (peer_mode == SHUTDOWN_MASK)
1903 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_HUP);
1904 else if (peer_mode & RCV_SHUTDOWN)
1905 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_IN);
1906 read_unlock(&other->sk_callback_lock);
1914 static int unix_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1916 struct sock *sk = sock->sk;
1923 amount = atomic_read(&sk->sk_wmem_alloc);
1924 err = put_user(amount, (int __user *)arg);
1928 struct sk_buff *skb;
1930 if (sk->sk_state == TCP_LISTEN) {
1935 spin_lock(&sk->sk_receive_queue.lock);
1936 if (sk->sk_type == SOCK_STREAM ||
1937 sk->sk_type == SOCK_SEQPACKET) {
1938 skb_queue_walk(&sk->sk_receive_queue, skb)
1941 skb = skb_peek(&sk->sk_receive_queue);
1945 spin_unlock(&sk->sk_receive_queue.lock);
1946 err = put_user(amount, (int __user *)arg);
1957 static unsigned int unix_poll(struct file * file, struct socket *sock, poll_table *wait)
1959 struct sock *sk = sock->sk;
1962 poll_wait(file, sk->sk_sleep, wait);
1965 /* exceptional events? */
1968 if (sk->sk_shutdown == SHUTDOWN_MASK)
1970 if (sk->sk_shutdown & RCV_SHUTDOWN)
1974 if (!skb_queue_empty(&sk->sk_receive_queue) ||
1975 (sk->sk_shutdown & RCV_SHUTDOWN))
1976 mask |= POLLIN | POLLRDNORM;
1978 /* Connection-based need to check for termination and startup */
1979 if ((sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) && sk->sk_state == TCP_CLOSE)
1983 * we set writable also when the other side has shut down the
1984 * connection. This prevents stuck sockets.
1986 if (unix_writable(sk))
1987 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
1993 #ifdef CONFIG_PROC_FS
1994 static struct sock *first_unix_socket(int *i)
1996 for (*i = 0; *i <= UNIX_HASH_SIZE; (*i)++) {
1997 if (!hlist_empty(&unix_socket_table[*i]))
1998 return __sk_head(&unix_socket_table[*i]);
2003 static struct sock *next_unix_socket(int *i, struct sock *s)
2005 struct sock *next = sk_next(s);
2006 /* More in this chain? */
2009 /* Look for next non-empty chain. */
2010 for ((*i)++; *i <= UNIX_HASH_SIZE; (*i)++) {
2011 if (!hlist_empty(&unix_socket_table[*i]))
2012 return __sk_head(&unix_socket_table[*i]);
2017 struct unix_iter_state {
2018 struct seq_net_private p;
2021 static struct sock *unix_seq_idx(struct seq_file *seq, loff_t pos)
2023 struct unix_iter_state *iter = seq->private;
2027 for (s = first_unix_socket(&iter->i); s; s = next_unix_socket(&iter->i, s)) {
2028 if (sock_net(s) != seq_file_net(seq))
2038 static void *unix_seq_start(struct seq_file *seq, loff_t *pos)
2039 __acquires(unix_table_lock)
2041 spin_lock(&unix_table_lock);
2042 return *pos ? unix_seq_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2045 static void *unix_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2047 struct unix_iter_state *iter = seq->private;
2048 struct sock *sk = v;
2051 if (v == SEQ_START_TOKEN)
2052 sk = first_unix_socket(&iter->i);
2054 sk = next_unix_socket(&iter->i, sk);
2055 while (sk && (sock_net(sk) != seq_file_net(seq)))
2056 sk = next_unix_socket(&iter->i, sk);
2060 static void unix_seq_stop(struct seq_file *seq, void *v)
2061 __releases(unix_table_lock)
2063 spin_unlock(&unix_table_lock);
2066 static int unix_seq_show(struct seq_file *seq, void *v)
2069 if (v == SEQ_START_TOKEN)
2070 seq_puts(seq, "Num RefCount Protocol Flags Type St "
2074 struct unix_sock *u = unix_sk(s);
2077 seq_printf(seq, "%p: %08X %08X %08X %04X %02X %5lu",
2079 atomic_read(&s->sk_refcnt),
2081 s->sk_state == TCP_LISTEN ? __SO_ACCEPTCON : 0,
2084 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTED : SS_UNCONNECTED) :
2085 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTING : SS_DISCONNECTING),
2093 len = u->addr->len - sizeof(short);
2094 if (!UNIX_ABSTRACT(s))
2100 for ( ; i < len; i++)
2101 seq_putc(seq, u->addr->name->sun_path[i]);
2103 unix_state_unlock(s);
2104 seq_putc(seq, '\n');
2110 static const struct seq_operations unix_seq_ops = {
2111 .start = unix_seq_start,
2112 .next = unix_seq_next,
2113 .stop = unix_seq_stop,
2114 .show = unix_seq_show,
2118 static int unix_seq_open(struct inode *inode, struct file *file)
2120 return seq_open_net(inode, file, &unix_seq_ops,
2121 sizeof(struct unix_iter_state));
2124 static const struct file_operations unix_seq_fops = {
2125 .owner = THIS_MODULE,
2126 .open = unix_seq_open,
2128 .llseek = seq_lseek,
2129 .release = seq_release_net,
2134 static struct net_proto_family unix_family_ops = {
2136 .create = unix_create,
2137 .owner = THIS_MODULE,
2141 static int unix_net_init(struct net *net)
2143 int error = -ENOMEM;
2145 net->unx.sysctl_max_dgram_qlen = 10;
2146 if (unix_sysctl_register(net))
2149 #ifdef CONFIG_PROC_FS
2150 if (!proc_net_fops_create(net, "unix", 0, &unix_seq_fops)) {
2151 unix_sysctl_unregister(net);
2160 static void unix_net_exit(struct net *net)
2162 unix_sysctl_unregister(net);
2163 proc_net_remove(net, "unix");
2166 static struct pernet_operations unix_net_ops = {
2167 .init = unix_net_init,
2168 .exit = unix_net_exit,
2171 static int __init af_unix_init(void)
2174 struct sk_buff *dummy_skb;
2176 BUILD_BUG_ON(sizeof(struct unix_skb_parms) > sizeof(dummy_skb->cb));
2178 rc = proto_register(&unix_proto, 1);
2180 printk(KERN_CRIT "%s: Cannot create unix_sock SLAB cache!\n",
2185 sock_register(&unix_family_ops);
2186 register_pernet_subsys(&unix_net_ops);
2191 static void __exit af_unix_exit(void)
2193 sock_unregister(PF_UNIX);
2194 proto_unregister(&unix_proto);
2195 unregister_pernet_subsys(&unix_net_ops);
2198 /* Earlier than device_initcall() so that other drivers invoking
2199 request_module() don't end up in a loop when modprobe tries
2200 to use a UNIX socket. But later than subsys_initcall() because
2201 we depend on stuff initialised there */
2202 fs_initcall(af_unix_init);
2203 module_exit(af_unix_exit);
2205 MODULE_LICENSE("GPL");
2206 MODULE_ALIAS_NETPROTO(PF_UNIX);