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)
306 spin_unlock(&unix_table_lock);
310 static inline int unix_writable(struct sock *sk)
312 return (atomic_read(&sk->sk_wmem_alloc) << 2) <= sk->sk_sndbuf;
315 static void unix_write_space(struct sock *sk)
317 read_lock(&sk->sk_callback_lock);
318 if (unix_writable(sk)) {
319 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
320 wake_up_interruptible_sync(sk->sk_sleep);
321 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
323 read_unlock(&sk->sk_callback_lock);
326 /* When dgram socket disconnects (or changes its peer), we clear its receive
327 * queue of packets arrived from previous peer. First, it allows to do
328 * flow control based only on wmem_alloc; second, sk connected to peer
329 * may receive messages only from that peer. */
330 static void unix_dgram_disconnected(struct sock *sk, struct sock *other)
332 if (!skb_queue_empty(&sk->sk_receive_queue)) {
333 skb_queue_purge(&sk->sk_receive_queue);
334 wake_up_interruptible_all(&unix_sk(sk)->peer_wait);
336 /* If one link of bidirectional dgram pipe is disconnected,
337 * we signal error. Messages are lost. Do not make this,
338 * when peer was not connected to us.
340 if (!sock_flag(other, SOCK_DEAD) && unix_peer(other) == sk) {
341 other->sk_err = ECONNRESET;
342 other->sk_error_report(other);
347 static void unix_sock_destructor(struct sock *sk)
349 struct unix_sock *u = unix_sk(sk);
351 skb_queue_purge(&sk->sk_receive_queue);
353 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
354 WARN_ON(!sk_unhashed(sk));
355 WARN_ON(sk->sk_socket);
356 if (!sock_flag(sk, SOCK_DEAD)) {
357 printk("Attempt to release alive unix socket: %p\n", sk);
362 unix_release_addr(u->addr);
364 atomic_dec(&unix_nr_socks);
365 #ifdef UNIX_REFCNT_DEBUG
366 printk(KERN_DEBUG "UNIX %p is destroyed, %d are still alive.\n", sk, 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);
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 current_euid_egid(&sk->sk_peercred.uid, &sk->sk_peercred.gid);
474 unix_state_unlock(sk);
479 static int unix_release(struct socket *);
480 static int unix_bind(struct socket *, struct sockaddr *, int);
481 static int unix_stream_connect(struct socket *, struct sockaddr *,
482 int addr_len, int flags);
483 static int unix_socketpair(struct socket *, struct socket *);
484 static int unix_accept(struct socket *, struct socket *, int);
485 static int unix_getname(struct socket *, struct sockaddr *, int *, int);
486 static unsigned int unix_poll(struct file *, struct socket *, poll_table *);
487 static unsigned int unix_dgram_poll(struct file *, struct socket *,
489 static int unix_ioctl(struct socket *, unsigned int, unsigned long);
490 static int unix_shutdown(struct socket *, int);
491 static int unix_stream_sendmsg(struct kiocb *, struct socket *,
492 struct msghdr *, size_t);
493 static int unix_stream_recvmsg(struct kiocb *, struct socket *,
494 struct msghdr *, size_t, int);
495 static int unix_dgram_sendmsg(struct kiocb *, struct socket *,
496 struct msghdr *, size_t);
497 static int unix_dgram_recvmsg(struct kiocb *, struct socket *,
498 struct msghdr *, size_t, int);
499 static int unix_dgram_connect(struct socket *, struct sockaddr *,
501 static int unix_seqpacket_sendmsg(struct kiocb *, struct socket *,
502 struct msghdr *, size_t);
504 static const struct proto_ops unix_stream_ops = {
506 .owner = THIS_MODULE,
507 .release = unix_release,
509 .connect = unix_stream_connect,
510 .socketpair = unix_socketpair,
511 .accept = unix_accept,
512 .getname = unix_getname,
515 .listen = unix_listen,
516 .shutdown = unix_shutdown,
517 .setsockopt = sock_no_setsockopt,
518 .getsockopt = sock_no_getsockopt,
519 .sendmsg = unix_stream_sendmsg,
520 .recvmsg = unix_stream_recvmsg,
521 .mmap = sock_no_mmap,
522 .sendpage = sock_no_sendpage,
525 static const struct proto_ops unix_dgram_ops = {
527 .owner = THIS_MODULE,
528 .release = unix_release,
530 .connect = unix_dgram_connect,
531 .socketpair = unix_socketpair,
532 .accept = sock_no_accept,
533 .getname = unix_getname,
534 .poll = unix_dgram_poll,
536 .listen = sock_no_listen,
537 .shutdown = unix_shutdown,
538 .setsockopt = sock_no_setsockopt,
539 .getsockopt = sock_no_getsockopt,
540 .sendmsg = unix_dgram_sendmsg,
541 .recvmsg = unix_dgram_recvmsg,
542 .mmap = sock_no_mmap,
543 .sendpage = sock_no_sendpage,
546 static const struct proto_ops unix_seqpacket_ops = {
548 .owner = THIS_MODULE,
549 .release = unix_release,
551 .connect = unix_stream_connect,
552 .socketpair = unix_socketpair,
553 .accept = unix_accept,
554 .getname = unix_getname,
555 .poll = unix_dgram_poll,
557 .listen = unix_listen,
558 .shutdown = unix_shutdown,
559 .setsockopt = sock_no_setsockopt,
560 .getsockopt = sock_no_getsockopt,
561 .sendmsg = unix_seqpacket_sendmsg,
562 .recvmsg = unix_dgram_recvmsg,
563 .mmap = sock_no_mmap,
564 .sendpage = sock_no_sendpage,
567 static struct proto unix_proto = {
569 .owner = THIS_MODULE,
570 .obj_size = sizeof(struct unix_sock),
574 * AF_UNIX sockets do not interact with hardware, hence they
575 * dont trigger interrupts - so it's safe for them to have
576 * bh-unsafe locking for their sk_receive_queue.lock. Split off
577 * this special lock-class by reinitializing the spinlock key:
579 static struct lock_class_key af_unix_sk_receive_queue_lock_key;
581 static struct sock * unix_create1(struct net *net, struct socket *sock)
583 struct sock *sk = NULL;
586 atomic_inc(&unix_nr_socks);
587 if (atomic_read(&unix_nr_socks) > 2 * get_max_files())
590 sk = sk_alloc(net, PF_UNIX, GFP_KERNEL, &unix_proto);
594 sock_init_data(sock,sk);
595 lockdep_set_class(&sk->sk_receive_queue.lock,
596 &af_unix_sk_receive_queue_lock_key);
598 sk->sk_write_space = unix_write_space;
599 sk->sk_max_ack_backlog = net->unx.sysctl_max_dgram_qlen;
600 sk->sk_destruct = unix_sock_destructor;
604 spin_lock_init(&u->lock);
605 atomic_long_set(&u->inflight, 0);
606 INIT_LIST_HEAD(&u->link);
607 mutex_init(&u->readlock); /* single task reading lock */
608 init_waitqueue_head(&u->peer_wait);
609 unix_insert_socket(unix_sockets_unbound, sk);
612 atomic_dec(&unix_nr_socks);
616 static int unix_create(struct net *net, struct socket *sock, int protocol)
618 if (protocol && protocol != PF_UNIX)
619 return -EPROTONOSUPPORT;
621 sock->state = SS_UNCONNECTED;
623 switch (sock->type) {
625 sock->ops = &unix_stream_ops;
628 * Believe it or not BSD has AF_UNIX, SOCK_RAW though
632 sock->type=SOCK_DGRAM;
634 sock->ops = &unix_dgram_ops;
637 sock->ops = &unix_seqpacket_ops;
640 return -ESOCKTNOSUPPORT;
643 return unix_create1(net, sock) ? 0 : -ENOMEM;
646 static int unix_release(struct socket *sock)
648 struct sock *sk = sock->sk;
655 return unix_release_sock (sk, 0);
658 static int unix_autobind(struct socket *sock)
660 struct sock *sk = sock->sk;
661 struct net *net = sock_net(sk);
662 struct unix_sock *u = unix_sk(sk);
663 static u32 ordernum = 1;
664 struct unix_address * addr;
667 mutex_lock(&u->readlock);
674 addr = kzalloc(sizeof(*addr) + sizeof(short) + 16, GFP_KERNEL);
678 addr->name->sun_family = AF_UNIX;
679 atomic_set(&addr->refcnt, 1);
682 addr->len = sprintf(addr->name->sun_path+1, "%05x", ordernum) + 1 + sizeof(short);
683 addr->hash = unix_hash_fold(csum_partial((void*)addr->name, addr->len, 0));
685 spin_lock(&unix_table_lock);
686 ordernum = (ordernum+1)&0xFFFFF;
688 if (__unix_find_socket_byname(net, addr->name, addr->len, sock->type,
690 spin_unlock(&unix_table_lock);
691 /* Sanity yield. It is unusual case, but yet... */
692 if (!(ordernum&0xFF))
696 addr->hash ^= sk->sk_type;
698 __unix_remove_socket(sk);
700 __unix_insert_socket(&unix_socket_table[addr->hash], sk);
701 spin_unlock(&unix_table_lock);
704 out: mutex_unlock(&u->readlock);
708 static struct sock *unix_find_other(struct net *net,
709 struct sockaddr_un *sunname, int len,
710 int type, unsigned hash, int *error)
716 if (sunname->sun_path[0]) {
718 err = kern_path(sunname->sun_path, LOOKUP_FOLLOW, &path);
721 inode = path.dentry->d_inode;
722 err = inode_permission(inode, MAY_WRITE);
727 if (!S_ISSOCK(inode->i_mode))
729 u = unix_find_socket_byinode(net, inode);
733 if (u->sk_type == type)
734 touch_atime(path.mnt, path.dentry);
739 if (u->sk_type != type) {
745 u=unix_find_socket_byname(net, sunname, len, type, hash);
747 struct dentry *dentry;
748 dentry = unix_sk(u)->dentry;
750 touch_atime(unix_sk(u)->mnt, dentry);
764 static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
766 struct sock *sk = sock->sk;
767 struct net *net = sock_net(sk);
768 struct unix_sock *u = unix_sk(sk);
769 struct sockaddr_un *sunaddr=(struct sockaddr_un *)uaddr;
770 struct dentry * dentry = NULL;
774 struct unix_address *addr;
775 struct hlist_head *list;
778 if (sunaddr->sun_family != AF_UNIX)
781 if (addr_len==sizeof(short)) {
782 err = unix_autobind(sock);
786 err = unix_mkname(sunaddr, addr_len, &hash);
791 mutex_lock(&u->readlock);
798 addr = kmalloc(sizeof(*addr)+addr_len, GFP_KERNEL);
802 memcpy(addr->name, sunaddr, addr_len);
803 addr->len = addr_len;
804 addr->hash = hash ^ sk->sk_type;
805 atomic_set(&addr->refcnt, 1);
807 if (sunaddr->sun_path[0]) {
811 * Get the parent directory, calculate the hash for last
814 err = path_lookup(sunaddr->sun_path, LOOKUP_PARENT, &nd);
816 goto out_mknod_parent;
818 dentry = lookup_create(&nd, 0);
819 err = PTR_ERR(dentry);
821 goto out_mknod_unlock;
824 * All right, let's create it.
827 (SOCK_INODE(sock)->i_mode & ~current->fs->umask);
828 err = mnt_want_write(nd.path.mnt);
831 err = vfs_mknod(nd.path.dentry->d_inode, dentry, mode, 0);
832 mnt_drop_write(nd.path.mnt);
835 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
836 dput(nd.path.dentry);
837 nd.path.dentry = dentry;
839 addr->hash = UNIX_HASH_SIZE;
842 spin_lock(&unix_table_lock);
844 if (!sunaddr->sun_path[0]) {
846 if (__unix_find_socket_byname(net, sunaddr, addr_len,
847 sk->sk_type, hash)) {
848 unix_release_addr(addr);
852 list = &unix_socket_table[addr->hash];
854 list = &unix_socket_table[dentry->d_inode->i_ino & (UNIX_HASH_SIZE-1)];
855 u->dentry = nd.path.dentry;
856 u->mnt = nd.path.mnt;
860 __unix_remove_socket(sk);
862 __unix_insert_socket(list, sk);
865 spin_unlock(&unix_table_lock);
867 mutex_unlock(&u->readlock);
874 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
879 unix_release_addr(addr);
883 static void unix_state_double_lock(struct sock *sk1, struct sock *sk2)
885 if (unlikely(sk1 == sk2) || !sk2) {
886 unix_state_lock(sk1);
890 unix_state_lock(sk1);
891 unix_state_lock_nested(sk2);
893 unix_state_lock(sk2);
894 unix_state_lock_nested(sk1);
898 static void unix_state_double_unlock(struct sock *sk1, struct sock *sk2)
900 if (unlikely(sk1 == sk2) || !sk2) {
901 unix_state_unlock(sk1);
904 unix_state_unlock(sk1);
905 unix_state_unlock(sk2);
908 static int unix_dgram_connect(struct socket *sock, struct sockaddr *addr,
911 struct sock *sk = sock->sk;
912 struct net *net = sock_net(sk);
913 struct sockaddr_un *sunaddr=(struct sockaddr_un*)addr;
918 if (addr->sa_family != AF_UNSPEC) {
919 err = unix_mkname(sunaddr, alen, &hash);
924 if (test_bit(SOCK_PASSCRED, &sock->flags) &&
925 !unix_sk(sk)->addr && (err = unix_autobind(sock)) != 0)
929 other=unix_find_other(net, sunaddr, alen, sock->type, hash, &err);
933 unix_state_double_lock(sk, other);
935 /* Apparently VFS overslept socket death. Retry. */
936 if (sock_flag(other, SOCK_DEAD)) {
937 unix_state_double_unlock(sk, other);
943 if (!unix_may_send(sk, other))
946 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
952 * 1003.1g breaking connected state with AF_UNSPEC
955 unix_state_double_lock(sk, other);
959 * If it was connected, reconnect.
962 struct sock *old_peer = unix_peer(sk);
964 unix_state_double_unlock(sk, other);
966 if (other != old_peer)
967 unix_dgram_disconnected(sk, old_peer);
971 unix_state_double_unlock(sk, other);
976 unix_state_double_unlock(sk, other);
982 static long unix_wait_for_peer(struct sock *other, long timeo)
984 struct unix_sock *u = unix_sk(other);
988 prepare_to_wait_exclusive(&u->peer_wait, &wait, TASK_INTERRUPTIBLE);
990 sched = !sock_flag(other, SOCK_DEAD) &&
991 !(other->sk_shutdown & RCV_SHUTDOWN) &&
992 unix_recvq_full(other);
994 unix_state_unlock(other);
997 timeo = schedule_timeout(timeo);
999 finish_wait(&u->peer_wait, &wait);
1003 static int unix_stream_connect(struct socket *sock, struct sockaddr *uaddr,
1004 int addr_len, int flags)
1006 struct sockaddr_un *sunaddr=(struct sockaddr_un *)uaddr;
1007 struct sock *sk = sock->sk;
1008 struct net *net = sock_net(sk);
1009 struct unix_sock *u = unix_sk(sk), *newu, *otheru;
1010 struct sock *newsk = NULL;
1011 struct sock *other = NULL;
1012 struct sk_buff *skb = NULL;
1018 err = unix_mkname(sunaddr, addr_len, &hash);
1023 if (test_bit(SOCK_PASSCRED, &sock->flags)
1024 && !u->addr && (err = unix_autobind(sock)) != 0)
1027 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
1029 /* First of all allocate resources.
1030 If we will make it after state is locked,
1031 we will have to recheck all again in any case.
1036 /* create new sock for complete connection */
1037 newsk = unix_create1(sock_net(sk), NULL);
1041 /* Allocate skb for sending to listening sock */
1042 skb = sock_wmalloc(newsk, 1, 0, GFP_KERNEL);
1047 /* Find listening sock. */
1048 other = unix_find_other(net, sunaddr, addr_len, sk->sk_type, hash, &err);
1052 /* Latch state of peer */
1053 unix_state_lock(other);
1055 /* Apparently VFS overslept socket death. Retry. */
1056 if (sock_flag(other, SOCK_DEAD)) {
1057 unix_state_unlock(other);
1062 err = -ECONNREFUSED;
1063 if (other->sk_state != TCP_LISTEN)
1066 if (unix_recvq_full(other)) {
1071 timeo = unix_wait_for_peer(other, timeo);
1073 err = sock_intr_errno(timeo);
1074 if (signal_pending(current))
1082 It is tricky place. We need to grab write lock and cannot
1083 drop lock on peer. It is dangerous because deadlock is
1084 possible. Connect to self case and simultaneous
1085 attempt to connect are eliminated by checking socket
1086 state. other is TCP_LISTEN, if sk is TCP_LISTEN we
1087 check this before attempt to grab lock.
1089 Well, and we have to recheck the state after socket locked.
1095 /* This is ok... continue with connect */
1097 case TCP_ESTABLISHED:
1098 /* Socket is already connected */
1106 unix_state_lock_nested(sk);
1108 if (sk->sk_state != st) {
1109 unix_state_unlock(sk);
1110 unix_state_unlock(other);
1115 err = security_unix_stream_connect(sock, other->sk_socket, newsk);
1117 unix_state_unlock(sk);
1121 /* The way is open! Fastly set all the necessary fields... */
1124 unix_peer(newsk) = sk;
1125 newsk->sk_state = TCP_ESTABLISHED;
1126 newsk->sk_type = sk->sk_type;
1127 newsk->sk_peercred.pid = task_tgid_vnr(current);
1128 current_euid_egid(&newsk->sk_peercred.uid, &newsk->sk_peercred.gid);
1129 newu = unix_sk(newsk);
1130 newsk->sk_sleep = &newu->peer_wait;
1131 otheru = unix_sk(other);
1133 /* copy address information from listening to new sock*/
1135 atomic_inc(&otheru->addr->refcnt);
1136 newu->addr = otheru->addr;
1138 if (otheru->dentry) {
1139 newu->dentry = dget(otheru->dentry);
1140 newu->mnt = mntget(otheru->mnt);
1143 /* Set credentials */
1144 sk->sk_peercred = other->sk_peercred;
1146 sock->state = SS_CONNECTED;
1147 sk->sk_state = TCP_ESTABLISHED;
1150 smp_mb__after_atomic_inc(); /* sock_hold() does an atomic_inc() */
1151 unix_peer(sk) = newsk;
1153 unix_state_unlock(sk);
1155 /* take ten and and send info to listening sock */
1156 spin_lock(&other->sk_receive_queue.lock);
1157 __skb_queue_tail(&other->sk_receive_queue, skb);
1158 spin_unlock(&other->sk_receive_queue.lock);
1159 unix_state_unlock(other);
1160 other->sk_data_ready(other, 0);
1166 unix_state_unlock(other);
1172 unix_release_sock(newsk, 0);
1178 static int unix_socketpair(struct socket *socka, struct socket *sockb)
1180 struct sock *ska=socka->sk, *skb = sockb->sk;
1182 /* Join our sockets back to back */
1187 ska->sk_peercred.pid = skb->sk_peercred.pid = task_tgid_vnr(current);
1188 current_euid_egid(&skb->sk_peercred.uid, &skb->sk_peercred.gid);
1189 ska->sk_peercred.uid = skb->sk_peercred.uid;
1190 ska->sk_peercred.gid = skb->sk_peercred.gid;
1192 if (ska->sk_type != SOCK_DGRAM) {
1193 ska->sk_state = TCP_ESTABLISHED;
1194 skb->sk_state = TCP_ESTABLISHED;
1195 socka->state = SS_CONNECTED;
1196 sockb->state = SS_CONNECTED;
1201 static int unix_accept(struct socket *sock, struct socket *newsock, int flags)
1203 struct sock *sk = sock->sk;
1205 struct sk_buff *skb;
1209 if (sock->type!=SOCK_STREAM && sock->type!=SOCK_SEQPACKET)
1213 if (sk->sk_state != TCP_LISTEN)
1216 /* If socket state is TCP_LISTEN it cannot change (for now...),
1217 * so that no locks are necessary.
1220 skb = skb_recv_datagram(sk, 0, flags&O_NONBLOCK, &err);
1222 /* This means receive shutdown. */
1229 skb_free_datagram(sk, skb);
1230 wake_up_interruptible(&unix_sk(sk)->peer_wait);
1232 /* attach accepted sock to socket */
1233 unix_state_lock(tsk);
1234 newsock->state = SS_CONNECTED;
1235 sock_graft(tsk, newsock);
1236 unix_state_unlock(tsk);
1244 static int unix_getname(struct socket *sock, struct sockaddr *uaddr, int *uaddr_len, int peer)
1246 struct sock *sk = sock->sk;
1247 struct unix_sock *u;
1248 struct sockaddr_un *sunaddr=(struct sockaddr_un *)uaddr;
1252 sk = unix_peer_get(sk);
1263 unix_state_lock(sk);
1265 sunaddr->sun_family = AF_UNIX;
1266 sunaddr->sun_path[0] = 0;
1267 *uaddr_len = sizeof(short);
1269 struct unix_address *addr = u->addr;
1271 *uaddr_len = addr->len;
1272 memcpy(sunaddr, addr->name, *uaddr_len);
1274 unix_state_unlock(sk);
1280 static void unix_detach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1284 scm->fp = UNIXCB(skb).fp;
1285 skb->destructor = sock_wfree;
1286 UNIXCB(skb).fp = NULL;
1288 for (i=scm->fp->count-1; i>=0; i--)
1289 unix_notinflight(scm->fp->fp[i]);
1292 static void unix_destruct_fds(struct sk_buff *skb)
1294 struct scm_cookie scm;
1295 memset(&scm, 0, sizeof(scm));
1296 unix_detach_fds(&scm, skb);
1298 /* Alas, it calls VFS */
1299 /* So fscking what? fput() had been SMP-safe since the last Summer */
1304 static void unix_attach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1307 for (i=scm->fp->count-1; i>=0; i--)
1308 unix_inflight(scm->fp->fp[i]);
1309 UNIXCB(skb).fp = scm->fp;
1310 skb->destructor = unix_destruct_fds;
1315 * Send AF_UNIX data.
1318 static int unix_dgram_sendmsg(struct kiocb *kiocb, struct socket *sock,
1319 struct msghdr *msg, size_t len)
1321 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1322 struct sock *sk = sock->sk;
1323 struct net *net = sock_net(sk);
1324 struct unix_sock *u = unix_sk(sk);
1325 struct sockaddr_un *sunaddr=msg->msg_name;
1326 struct sock *other = NULL;
1327 int namelen = 0; /* fake GCC */
1330 struct sk_buff *skb;
1332 struct scm_cookie tmp_scm;
1334 if (NULL == siocb->scm)
1335 siocb->scm = &tmp_scm;
1336 err = scm_send(sock, msg, siocb->scm);
1341 if (msg->msg_flags&MSG_OOB)
1344 if (msg->msg_namelen) {
1345 err = unix_mkname(sunaddr, msg->msg_namelen, &hash);
1352 other = unix_peer_get(sk);
1357 if (test_bit(SOCK_PASSCRED, &sock->flags)
1358 && !u->addr && (err = unix_autobind(sock)) != 0)
1362 if (len > sk->sk_sndbuf - 32)
1365 skb = sock_alloc_send_skb(sk, len, msg->msg_flags&MSG_DONTWAIT, &err);
1369 memcpy(UNIXCREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1371 unix_attach_fds(siocb->scm, skb);
1372 unix_get_secdata(siocb->scm, skb);
1374 skb_reset_transport_header(skb);
1375 err = memcpy_fromiovec(skb_put(skb,len), msg->msg_iov, len);
1379 timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1384 if (sunaddr == NULL)
1387 other = unix_find_other(net, sunaddr, namelen, sk->sk_type,
1393 unix_state_lock(other);
1395 if (!unix_may_send(sk, other))
1398 if (sock_flag(other, SOCK_DEAD)) {
1400 * Check with 1003.1g - what should
1403 unix_state_unlock(other);
1407 unix_state_lock(sk);
1408 if (unix_peer(sk) == other) {
1410 unix_state_unlock(sk);
1412 unix_dgram_disconnected(sk, other);
1414 err = -ECONNREFUSED;
1416 unix_state_unlock(sk);
1426 if (other->sk_shutdown & RCV_SHUTDOWN)
1429 if (sk->sk_type != SOCK_SEQPACKET) {
1430 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1435 if (unix_peer(other) != sk && unix_recvq_full(other)) {
1441 timeo = unix_wait_for_peer(other, timeo);
1443 err = sock_intr_errno(timeo);
1444 if (signal_pending(current))
1450 skb_queue_tail(&other->sk_receive_queue, skb);
1451 unix_state_unlock(other);
1452 other->sk_data_ready(other, len);
1454 scm_destroy(siocb->scm);
1458 unix_state_unlock(other);
1464 scm_destroy(siocb->scm);
1469 static int unix_stream_sendmsg(struct kiocb *kiocb, struct socket *sock,
1470 struct msghdr *msg, size_t len)
1472 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1473 struct sock *sk = sock->sk;
1474 struct sock *other = NULL;
1475 struct sockaddr_un *sunaddr=msg->msg_name;
1477 struct sk_buff *skb;
1479 struct scm_cookie tmp_scm;
1481 if (NULL == siocb->scm)
1482 siocb->scm = &tmp_scm;
1483 err = scm_send(sock, msg, siocb->scm);
1488 if (msg->msg_flags&MSG_OOB)
1491 if (msg->msg_namelen) {
1492 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
1497 other = unix_peer(sk);
1502 if (sk->sk_shutdown & SEND_SHUTDOWN)
1508 * Optimisation for the fact that under 0.01% of X
1509 * messages typically need breaking up.
1514 /* Keep two messages in the pipe so it schedules better */
1515 if (size > ((sk->sk_sndbuf >> 1) - 64))
1516 size = (sk->sk_sndbuf >> 1) - 64;
1518 if (size > SKB_MAX_ALLOC)
1519 size = SKB_MAX_ALLOC;
1525 skb=sock_alloc_send_skb(sk,size,msg->msg_flags&MSG_DONTWAIT, &err);
1531 * If you pass two values to the sock_alloc_send_skb
1532 * it tries to grab the large buffer with GFP_NOFS
1533 * (which can fail easily), and if it fails grab the
1534 * fallback size buffer which is under a page and will
1537 size = min_t(int, size, skb_tailroom(skb));
1539 memcpy(UNIXCREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1541 unix_attach_fds(siocb->scm, skb);
1543 if ((err = memcpy_fromiovec(skb_put(skb,size), msg->msg_iov, size)) != 0) {
1548 unix_state_lock(other);
1550 if (sock_flag(other, SOCK_DEAD) ||
1551 (other->sk_shutdown & RCV_SHUTDOWN))
1554 skb_queue_tail(&other->sk_receive_queue, skb);
1555 unix_state_unlock(other);
1556 other->sk_data_ready(other, size);
1560 scm_destroy(siocb->scm);
1566 unix_state_unlock(other);
1569 if (sent==0 && !(msg->msg_flags&MSG_NOSIGNAL))
1570 send_sig(SIGPIPE,current,0);
1573 scm_destroy(siocb->scm);
1575 return sent ? : err;
1578 static int unix_seqpacket_sendmsg(struct kiocb *kiocb, struct socket *sock,
1579 struct msghdr *msg, size_t len)
1582 struct sock *sk = sock->sk;
1584 err = sock_error(sk);
1588 if (sk->sk_state != TCP_ESTABLISHED)
1591 if (msg->msg_namelen)
1592 msg->msg_namelen = 0;
1594 return unix_dgram_sendmsg(kiocb, sock, msg, len);
1597 static void unix_copy_addr(struct msghdr *msg, struct sock *sk)
1599 struct unix_sock *u = unix_sk(sk);
1601 msg->msg_namelen = 0;
1603 msg->msg_namelen = u->addr->len;
1604 memcpy(msg->msg_name, u->addr->name, u->addr->len);
1608 static int unix_dgram_recvmsg(struct kiocb *iocb, struct socket *sock,
1609 struct msghdr *msg, size_t size,
1612 struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1613 struct scm_cookie tmp_scm;
1614 struct sock *sk = sock->sk;
1615 struct unix_sock *u = unix_sk(sk);
1616 int noblock = flags & MSG_DONTWAIT;
1617 struct sk_buff *skb;
1624 msg->msg_namelen = 0;
1626 mutex_lock(&u->readlock);
1628 skb = skb_recv_datagram(sk, flags, noblock, &err);
1630 unix_state_lock(sk);
1631 /* Signal EOF on disconnected non-blocking SEQPACKET socket. */
1632 if (sk->sk_type == SOCK_SEQPACKET && err == -EAGAIN &&
1633 (sk->sk_shutdown & RCV_SHUTDOWN))
1635 unix_state_unlock(sk);
1639 wake_up_interruptible_sync(&u->peer_wait);
1642 unix_copy_addr(msg, skb->sk);
1644 if (size > skb->len)
1646 else if (size < skb->len)
1647 msg->msg_flags |= MSG_TRUNC;
1649 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, size);
1654 siocb->scm = &tmp_scm;
1655 memset(&tmp_scm, 0, sizeof(tmp_scm));
1657 siocb->scm->creds = *UNIXCREDS(skb);
1658 unix_set_secdata(siocb->scm, skb);
1660 if (!(flags & MSG_PEEK))
1663 unix_detach_fds(siocb->scm, skb);
1667 /* It is questionable: on PEEK we could:
1668 - do not return fds - good, but too simple 8)
1669 - return fds, and do not return them on read (old strategy,
1671 - clone fds (I chose it for now, it is the most universal
1674 POSIX 1003.1g does not actually define this clearly
1675 at all. POSIX 1003.1g doesn't define a lot of things
1680 siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1684 scm_recv(sock, msg, siocb->scm, flags);
1687 skb_free_datagram(sk,skb);
1689 mutex_unlock(&u->readlock);
1695 * Sleep until data has arrive. But check for races..
1698 static long unix_stream_data_wait(struct sock * sk, long timeo)
1702 unix_state_lock(sk);
1705 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1707 if (!skb_queue_empty(&sk->sk_receive_queue) ||
1709 (sk->sk_shutdown & RCV_SHUTDOWN) ||
1710 signal_pending(current) ||
1714 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1715 unix_state_unlock(sk);
1716 timeo = schedule_timeout(timeo);
1717 unix_state_lock(sk);
1718 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1721 finish_wait(sk->sk_sleep, &wait);
1722 unix_state_unlock(sk);
1728 static int unix_stream_recvmsg(struct kiocb *iocb, struct socket *sock,
1729 struct msghdr *msg, size_t size,
1732 struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1733 struct scm_cookie tmp_scm;
1734 struct sock *sk = sock->sk;
1735 struct unix_sock *u = unix_sk(sk);
1736 struct sockaddr_un *sunaddr=msg->msg_name;
1738 int check_creds = 0;
1744 if (sk->sk_state != TCP_ESTABLISHED)
1751 target = sock_rcvlowat(sk, flags&MSG_WAITALL, size);
1752 timeo = sock_rcvtimeo(sk, flags&MSG_DONTWAIT);
1754 msg->msg_namelen = 0;
1756 /* Lock the socket to prevent queue disordering
1757 * while sleeps in memcpy_tomsg
1761 siocb->scm = &tmp_scm;
1762 memset(&tmp_scm, 0, sizeof(tmp_scm));
1765 mutex_lock(&u->readlock);
1770 struct sk_buff *skb;
1772 unix_state_lock(sk);
1773 skb = skb_dequeue(&sk->sk_receive_queue);
1776 if (copied >= target)
1780 * POSIX 1003.1g mandates this order.
1783 if ((err = sock_error(sk)) != 0)
1785 if (sk->sk_shutdown & RCV_SHUTDOWN)
1788 unix_state_unlock(sk);
1792 mutex_unlock(&u->readlock);
1794 timeo = unix_stream_data_wait(sk, timeo);
1796 if (signal_pending(current)) {
1797 err = sock_intr_errno(timeo);
1800 mutex_lock(&u->readlock);
1803 unix_state_unlock(sk);
1806 unix_state_unlock(sk);
1809 /* Never glue messages from different writers */
1810 if (memcmp(UNIXCREDS(skb), &siocb->scm->creds, sizeof(siocb->scm->creds)) != 0) {
1811 skb_queue_head(&sk->sk_receive_queue, skb);
1815 /* Copy credentials */
1816 siocb->scm->creds = *UNIXCREDS(skb);
1820 /* Copy address just once */
1823 unix_copy_addr(msg, skb->sk);
1827 chunk = min_t(unsigned int, skb->len, size);
1828 if (memcpy_toiovec(msg->msg_iov, skb->data, chunk)) {
1829 skb_queue_head(&sk->sk_receive_queue, skb);
1837 /* Mark read part of skb as used */
1838 if (!(flags & MSG_PEEK))
1840 skb_pull(skb, chunk);
1843 unix_detach_fds(siocb->scm, skb);
1845 /* put the skb back if we didn't use it up.. */
1848 skb_queue_head(&sk->sk_receive_queue, skb);
1859 /* It is questionable, see note in unix_dgram_recvmsg.
1862 siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1864 /* put message back and return */
1865 skb_queue_head(&sk->sk_receive_queue, skb);
1870 mutex_unlock(&u->readlock);
1871 scm_recv(sock, msg, siocb->scm, flags);
1873 return copied ? : err;
1876 static int unix_shutdown(struct socket *sock, int mode)
1878 struct sock *sk = sock->sk;
1881 mode = (mode+1)&(RCV_SHUTDOWN|SEND_SHUTDOWN);
1884 unix_state_lock(sk);
1885 sk->sk_shutdown |= mode;
1886 other=unix_peer(sk);
1889 unix_state_unlock(sk);
1890 sk->sk_state_change(sk);
1893 (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET)) {
1897 if (mode&RCV_SHUTDOWN)
1898 peer_mode |= SEND_SHUTDOWN;
1899 if (mode&SEND_SHUTDOWN)
1900 peer_mode |= RCV_SHUTDOWN;
1901 unix_state_lock(other);
1902 other->sk_shutdown |= peer_mode;
1903 unix_state_unlock(other);
1904 other->sk_state_change(other);
1905 read_lock(&other->sk_callback_lock);
1906 if (peer_mode == SHUTDOWN_MASK)
1907 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_HUP);
1908 else if (peer_mode & RCV_SHUTDOWN)
1909 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_IN);
1910 read_unlock(&other->sk_callback_lock);
1918 static int unix_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1920 struct sock *sk = sock->sk;
1927 amount = atomic_read(&sk->sk_wmem_alloc);
1928 err = put_user(amount, (int __user *)arg);
1932 struct sk_buff *skb;
1934 if (sk->sk_state == TCP_LISTEN) {
1939 spin_lock(&sk->sk_receive_queue.lock);
1940 if (sk->sk_type == SOCK_STREAM ||
1941 sk->sk_type == SOCK_SEQPACKET) {
1942 skb_queue_walk(&sk->sk_receive_queue, skb)
1945 skb = skb_peek(&sk->sk_receive_queue);
1949 spin_unlock(&sk->sk_receive_queue.lock);
1950 err = put_user(amount, (int __user *)arg);
1961 static unsigned int unix_poll(struct file * file, struct socket *sock, poll_table *wait)
1963 struct sock *sk = sock->sk;
1966 poll_wait(file, sk->sk_sleep, wait);
1969 /* exceptional events? */
1972 if (sk->sk_shutdown == SHUTDOWN_MASK)
1974 if (sk->sk_shutdown & RCV_SHUTDOWN)
1978 if (!skb_queue_empty(&sk->sk_receive_queue) ||
1979 (sk->sk_shutdown & RCV_SHUTDOWN))
1980 mask |= POLLIN | POLLRDNORM;
1982 /* Connection-based need to check for termination and startup */
1983 if ((sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) && sk->sk_state == TCP_CLOSE)
1987 * we set writable also when the other side has shut down the
1988 * connection. This prevents stuck sockets.
1990 if (unix_writable(sk))
1991 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
1996 static unsigned int unix_dgram_poll(struct file *file, struct socket *sock,
1999 struct sock *sk = sock->sk, *other;
2000 unsigned int mask, writable;
2002 poll_wait(file, sk->sk_sleep, wait);
2005 /* exceptional events? */
2006 if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue))
2008 if (sk->sk_shutdown & RCV_SHUTDOWN)
2010 if (sk->sk_shutdown == SHUTDOWN_MASK)
2014 if (!skb_queue_empty(&sk->sk_receive_queue) ||
2015 (sk->sk_shutdown & RCV_SHUTDOWN))
2016 mask |= POLLIN | POLLRDNORM;
2018 /* Connection-based need to check for termination and startup */
2019 if (sk->sk_type == SOCK_SEQPACKET) {
2020 if (sk->sk_state == TCP_CLOSE)
2022 /* connection hasn't started yet? */
2023 if (sk->sk_state == TCP_SYN_SENT)
2028 writable = unix_writable(sk);
2030 other = unix_peer_get(sk);
2032 if (unix_peer(other) != sk) {
2033 poll_wait(file, &unix_sk(other)->peer_wait,
2035 if (unix_recvq_full(other))
2044 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2046 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
2051 #ifdef CONFIG_PROC_FS
2052 static struct sock *first_unix_socket(int *i)
2054 for (*i = 0; *i <= UNIX_HASH_SIZE; (*i)++) {
2055 if (!hlist_empty(&unix_socket_table[*i]))
2056 return __sk_head(&unix_socket_table[*i]);
2061 static struct sock *next_unix_socket(int *i, struct sock *s)
2063 struct sock *next = sk_next(s);
2064 /* More in this chain? */
2067 /* Look for next non-empty chain. */
2068 for ((*i)++; *i <= UNIX_HASH_SIZE; (*i)++) {
2069 if (!hlist_empty(&unix_socket_table[*i]))
2070 return __sk_head(&unix_socket_table[*i]);
2075 struct unix_iter_state {
2076 struct seq_net_private p;
2079 static struct sock *unix_seq_idx(struct seq_file *seq, loff_t pos)
2081 struct unix_iter_state *iter = seq->private;
2085 for (s = first_unix_socket(&iter->i); s; s = next_unix_socket(&iter->i, s)) {
2086 if (sock_net(s) != seq_file_net(seq))
2096 static void *unix_seq_start(struct seq_file *seq, loff_t *pos)
2097 __acquires(unix_table_lock)
2099 spin_lock(&unix_table_lock);
2100 return *pos ? unix_seq_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2103 static void *unix_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2105 struct unix_iter_state *iter = seq->private;
2106 struct sock *sk = v;
2109 if (v == SEQ_START_TOKEN)
2110 sk = first_unix_socket(&iter->i);
2112 sk = next_unix_socket(&iter->i, sk);
2113 while (sk && (sock_net(sk) != seq_file_net(seq)))
2114 sk = next_unix_socket(&iter->i, sk);
2118 static void unix_seq_stop(struct seq_file *seq, void *v)
2119 __releases(unix_table_lock)
2121 spin_unlock(&unix_table_lock);
2124 static int unix_seq_show(struct seq_file *seq, void *v)
2127 if (v == SEQ_START_TOKEN)
2128 seq_puts(seq, "Num RefCount Protocol Flags Type St "
2132 struct unix_sock *u = unix_sk(s);
2135 seq_printf(seq, "%p: %08X %08X %08X %04X %02X %5lu",
2137 atomic_read(&s->sk_refcnt),
2139 s->sk_state == TCP_LISTEN ? __SO_ACCEPTCON : 0,
2142 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTED : SS_UNCONNECTED) :
2143 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTING : SS_DISCONNECTING),
2151 len = u->addr->len - sizeof(short);
2152 if (!UNIX_ABSTRACT(s))
2158 for ( ; i < len; i++)
2159 seq_putc(seq, u->addr->name->sun_path[i]);
2161 unix_state_unlock(s);
2162 seq_putc(seq, '\n');
2168 static const struct seq_operations unix_seq_ops = {
2169 .start = unix_seq_start,
2170 .next = unix_seq_next,
2171 .stop = unix_seq_stop,
2172 .show = unix_seq_show,
2176 static int unix_seq_open(struct inode *inode, struct file *file)
2178 return seq_open_net(inode, file, &unix_seq_ops,
2179 sizeof(struct unix_iter_state));
2182 static const struct file_operations unix_seq_fops = {
2183 .owner = THIS_MODULE,
2184 .open = unix_seq_open,
2186 .llseek = seq_lseek,
2187 .release = seq_release_net,
2192 static struct net_proto_family unix_family_ops = {
2194 .create = unix_create,
2195 .owner = THIS_MODULE,
2199 static int unix_net_init(struct net *net)
2201 int error = -ENOMEM;
2203 net->unx.sysctl_max_dgram_qlen = 10;
2204 if (unix_sysctl_register(net))
2207 #ifdef CONFIG_PROC_FS
2208 if (!proc_net_fops_create(net, "unix", 0, &unix_seq_fops)) {
2209 unix_sysctl_unregister(net);
2218 static void unix_net_exit(struct net *net)
2220 unix_sysctl_unregister(net);
2221 proc_net_remove(net, "unix");
2224 static struct pernet_operations unix_net_ops = {
2225 .init = unix_net_init,
2226 .exit = unix_net_exit,
2229 static int __init af_unix_init(void)
2232 struct sk_buff *dummy_skb;
2234 BUILD_BUG_ON(sizeof(struct unix_skb_parms) > sizeof(dummy_skb->cb));
2236 rc = proto_register(&unix_proto, 1);
2238 printk(KERN_CRIT "%s: Cannot create unix_sock SLAB cache!\n",
2243 sock_register(&unix_family_ops);
2244 register_pernet_subsys(&unix_net_ops);
2249 static void __exit af_unix_exit(void)
2251 sock_unregister(PF_UNIX);
2252 proto_unregister(&unix_proto);
2253 unregister_pernet_subsys(&unix_net_ops);
2256 /* Earlier than device_initcall() so that other drivers invoking
2257 request_module() don't end up in a loop when modprobe tries
2258 to use a UNIX socket. But later than subsys_initcall() because
2259 we depend on stuff initialised there */
2260 fs_initcall(af_unix_init);
2261 module_exit(af_unix_exit);
2263 MODULE_LICENSE("GPL");
2264 MODULE_ALIAS_NETPROTO(PF_UNIX);