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(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_INFO "Attempt to release alive unix socket: %p\n", sk);
361 unix_release_addr(u->addr);
363 atomic_dec(&unix_nr_socks);
365 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
367 #ifdef UNIX_REFCNT_DEBUG
368 printk(KERN_DEBUG "UNIX %p is destroyed, %d are still alive.\n", sk,
369 atomic_read(&unix_nr_socks));
373 static int unix_release_sock(struct sock *sk, int embrion)
375 struct unix_sock *u = unix_sk(sk);
376 struct dentry *dentry;
377 struct vfsmount *mnt;
382 unix_remove_socket(sk);
387 sk->sk_shutdown = SHUTDOWN_MASK;
392 state = sk->sk_state;
393 sk->sk_state = TCP_CLOSE;
394 unix_state_unlock(sk);
396 wake_up_interruptible_all(&u->peer_wait);
398 skpair = unix_peer(sk);
400 if (skpair != NULL) {
401 if (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) {
402 unix_state_lock(skpair);
404 skpair->sk_shutdown = SHUTDOWN_MASK;
405 if (!skb_queue_empty(&sk->sk_receive_queue) || embrion)
406 skpair->sk_err = ECONNRESET;
407 unix_state_unlock(skpair);
408 skpair->sk_state_change(skpair);
409 read_lock(&skpair->sk_callback_lock);
410 sk_wake_async(skpair, SOCK_WAKE_WAITD, POLL_HUP);
411 read_unlock(&skpair->sk_callback_lock);
413 sock_put(skpair); /* It may now die */
414 unix_peer(sk) = NULL;
417 /* Try to flush out this socket. Throw out buffers at least */
419 while ((skb = skb_dequeue(&sk->sk_receive_queue)) != NULL) {
420 if (state == TCP_LISTEN)
421 unix_release_sock(skb->sk, 1);
422 /* passed fds are erased in the kfree_skb hook */
433 /* ---- Socket is dead now and most probably destroyed ---- */
436 * Fixme: BSD difference: In BSD all sockets connected to use get
437 * ECONNRESET and we die on the spot. In Linux we behave
438 * like files and pipes do and wait for the last
441 * Can't we simply set sock->err?
443 * What the above comment does talk about? --ANK(980817)
446 if (unix_tot_inflight)
447 unix_gc(); /* Garbage collect fds */
452 static int unix_listen(struct socket *sock, int backlog)
455 struct sock *sk = sock->sk;
456 struct unix_sock *u = unix_sk(sk);
459 if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
460 goto out; /* Only stream/seqpacket sockets accept */
463 goto out; /* No listens on an unbound socket */
465 if (sk->sk_state != TCP_CLOSE && sk->sk_state != TCP_LISTEN)
467 if (backlog > sk->sk_max_ack_backlog)
468 wake_up_interruptible_all(&u->peer_wait);
469 sk->sk_max_ack_backlog = backlog;
470 sk->sk_state = TCP_LISTEN;
471 /* set credentials so connect can copy them */
472 sk->sk_peercred.pid = task_tgid_vnr(current);
473 sk->sk_peercred.uid = current->euid;
474 sk->sk_peercred.gid = current->egid;
478 unix_state_unlock(sk);
483 static int unix_release(struct socket *);
484 static int unix_bind(struct socket *, struct sockaddr *, int);
485 static int unix_stream_connect(struct socket *, struct sockaddr *,
486 int addr_len, int flags);
487 static int unix_socketpair(struct socket *, struct socket *);
488 static int unix_accept(struct socket *, struct socket *, int);
489 static int unix_getname(struct socket *, struct sockaddr *, int *, int);
490 static unsigned int unix_poll(struct file *, struct socket *, poll_table *);
491 static unsigned int unix_dgram_poll(struct file *, struct socket *,
493 static int unix_ioctl(struct socket *, unsigned int, unsigned long);
494 static int unix_shutdown(struct socket *, int);
495 static int unix_stream_sendmsg(struct kiocb *, struct socket *,
496 struct msghdr *, size_t);
497 static int unix_stream_recvmsg(struct kiocb *, struct socket *,
498 struct msghdr *, size_t, int);
499 static int unix_dgram_sendmsg(struct kiocb *, struct socket *,
500 struct msghdr *, size_t);
501 static int unix_dgram_recvmsg(struct kiocb *, struct socket *,
502 struct msghdr *, size_t, int);
503 static int unix_dgram_connect(struct socket *, struct sockaddr *,
505 static int unix_seqpacket_sendmsg(struct kiocb *, struct socket *,
506 struct msghdr *, size_t);
508 static const struct proto_ops unix_stream_ops = {
510 .owner = THIS_MODULE,
511 .release = unix_release,
513 .connect = unix_stream_connect,
514 .socketpair = unix_socketpair,
515 .accept = unix_accept,
516 .getname = unix_getname,
519 .listen = unix_listen,
520 .shutdown = unix_shutdown,
521 .setsockopt = sock_no_setsockopt,
522 .getsockopt = sock_no_getsockopt,
523 .sendmsg = unix_stream_sendmsg,
524 .recvmsg = unix_stream_recvmsg,
525 .mmap = sock_no_mmap,
526 .sendpage = sock_no_sendpage,
529 static const struct proto_ops unix_dgram_ops = {
531 .owner = THIS_MODULE,
532 .release = unix_release,
534 .connect = unix_dgram_connect,
535 .socketpair = unix_socketpair,
536 .accept = sock_no_accept,
537 .getname = unix_getname,
538 .poll = unix_dgram_poll,
540 .listen = sock_no_listen,
541 .shutdown = unix_shutdown,
542 .setsockopt = sock_no_setsockopt,
543 .getsockopt = sock_no_getsockopt,
544 .sendmsg = unix_dgram_sendmsg,
545 .recvmsg = unix_dgram_recvmsg,
546 .mmap = sock_no_mmap,
547 .sendpage = sock_no_sendpage,
550 static const struct proto_ops unix_seqpacket_ops = {
552 .owner = THIS_MODULE,
553 .release = unix_release,
555 .connect = unix_stream_connect,
556 .socketpair = unix_socketpair,
557 .accept = unix_accept,
558 .getname = unix_getname,
559 .poll = unix_dgram_poll,
561 .listen = unix_listen,
562 .shutdown = unix_shutdown,
563 .setsockopt = sock_no_setsockopt,
564 .getsockopt = sock_no_getsockopt,
565 .sendmsg = unix_seqpacket_sendmsg,
566 .recvmsg = unix_dgram_recvmsg,
567 .mmap = sock_no_mmap,
568 .sendpage = sock_no_sendpage,
571 static struct proto unix_proto = {
573 .owner = THIS_MODULE,
574 .obj_size = sizeof(struct unix_sock),
578 * AF_UNIX sockets do not interact with hardware, hence they
579 * dont trigger interrupts - so it's safe for them to have
580 * bh-unsafe locking for their sk_receive_queue.lock. Split off
581 * this special lock-class by reinitializing the spinlock key:
583 static struct lock_class_key af_unix_sk_receive_queue_lock_key;
585 static struct sock *unix_create1(struct net *net, struct socket *sock)
587 struct sock *sk = NULL;
590 atomic_inc(&unix_nr_socks);
591 if (atomic_read(&unix_nr_socks) > 2 * get_max_files())
594 sk = sk_alloc(net, PF_UNIX, GFP_KERNEL, &unix_proto);
598 sock_init_data(sock, sk);
599 lockdep_set_class(&sk->sk_receive_queue.lock,
600 &af_unix_sk_receive_queue_lock_key);
602 sk->sk_write_space = unix_write_space;
603 sk->sk_max_ack_backlog = net->unx.sysctl_max_dgram_qlen;
604 sk->sk_destruct = unix_sock_destructor;
608 spin_lock_init(&u->lock);
609 atomic_long_set(&u->inflight, 0);
610 INIT_LIST_HEAD(&u->link);
611 mutex_init(&u->readlock); /* single task reading lock */
612 init_waitqueue_head(&u->peer_wait);
613 unix_insert_socket(unix_sockets_unbound, sk);
616 atomic_dec(&unix_nr_socks);
619 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
625 static int unix_create(struct net *net, struct socket *sock, int protocol)
627 if (protocol && protocol != PF_UNIX)
628 return -EPROTONOSUPPORT;
630 sock->state = SS_UNCONNECTED;
632 switch (sock->type) {
634 sock->ops = &unix_stream_ops;
637 * Believe it or not BSD has AF_UNIX, SOCK_RAW though
641 sock->type = SOCK_DGRAM;
643 sock->ops = &unix_dgram_ops;
646 sock->ops = &unix_seqpacket_ops;
649 return -ESOCKTNOSUPPORT;
652 return unix_create1(net, sock) ? 0 : -ENOMEM;
655 static int unix_release(struct socket *sock)
657 struct sock *sk = sock->sk;
664 return unix_release_sock(sk, 0);
667 static int unix_autobind(struct socket *sock)
669 struct sock *sk = sock->sk;
670 struct net *net = sock_net(sk);
671 struct unix_sock *u = unix_sk(sk);
672 static u32 ordernum = 1;
673 struct unix_address *addr;
676 mutex_lock(&u->readlock);
683 addr = kzalloc(sizeof(*addr) + sizeof(short) + 16, GFP_KERNEL);
687 addr->name->sun_family = AF_UNIX;
688 atomic_set(&addr->refcnt, 1);
691 addr->len = sprintf(addr->name->sun_path+1, "%05x", ordernum) + 1 + sizeof(short);
692 addr->hash = unix_hash_fold(csum_partial(addr->name, addr->len, 0));
694 spin_lock(&unix_table_lock);
695 ordernum = (ordernum+1)&0xFFFFF;
697 if (__unix_find_socket_byname(net, addr->name, addr->len, sock->type,
699 spin_unlock(&unix_table_lock);
700 /* Sanity yield. It is unusual case, but yet... */
701 if (!(ordernum&0xFF))
705 addr->hash ^= sk->sk_type;
707 __unix_remove_socket(sk);
709 __unix_insert_socket(&unix_socket_table[addr->hash], sk);
710 spin_unlock(&unix_table_lock);
713 out: mutex_unlock(&u->readlock);
717 static struct sock *unix_find_other(struct net *net,
718 struct sockaddr_un *sunname, int len,
719 int type, unsigned hash, int *error)
725 if (sunname->sun_path[0]) {
727 err = kern_path(sunname->sun_path, LOOKUP_FOLLOW, &path);
730 inode = path.dentry->d_inode;
731 err = inode_permission(inode, MAY_WRITE);
736 if (!S_ISSOCK(inode->i_mode))
738 u = unix_find_socket_byinode(net, inode);
742 if (u->sk_type == type)
743 touch_atime(path.mnt, path.dentry);
748 if (u->sk_type != type) {
754 u = unix_find_socket_byname(net, sunname, len, type, hash);
756 struct dentry *dentry;
757 dentry = unix_sk(u)->dentry;
759 touch_atime(unix_sk(u)->mnt, dentry);
773 static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
775 struct sock *sk = sock->sk;
776 struct net *net = sock_net(sk);
777 struct unix_sock *u = unix_sk(sk);
778 struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
779 struct dentry *dentry = NULL;
783 struct unix_address *addr;
784 struct hlist_head *list;
787 if (sunaddr->sun_family != AF_UNIX)
790 if (addr_len == sizeof(short)) {
791 err = unix_autobind(sock);
795 err = unix_mkname(sunaddr, addr_len, &hash);
800 mutex_lock(&u->readlock);
807 addr = kmalloc(sizeof(*addr)+addr_len, GFP_KERNEL);
811 memcpy(addr->name, sunaddr, addr_len);
812 addr->len = addr_len;
813 addr->hash = hash ^ sk->sk_type;
814 atomic_set(&addr->refcnt, 1);
816 if (sunaddr->sun_path[0]) {
820 * Get the parent directory, calculate the hash for last
823 err = path_lookup(sunaddr->sun_path, LOOKUP_PARENT, &nd);
825 goto out_mknod_parent;
827 dentry = lookup_create(&nd, 0);
828 err = PTR_ERR(dentry);
830 goto out_mknod_unlock;
833 * All right, let's create it.
836 (SOCK_INODE(sock)->i_mode & ~current->fs->umask);
837 err = mnt_want_write(nd.path.mnt);
840 err = vfs_mknod(nd.path.dentry->d_inode, dentry, mode, 0);
841 mnt_drop_write(nd.path.mnt);
844 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
845 dput(nd.path.dentry);
846 nd.path.dentry = dentry;
848 addr->hash = UNIX_HASH_SIZE;
851 spin_lock(&unix_table_lock);
853 if (!sunaddr->sun_path[0]) {
855 if (__unix_find_socket_byname(net, sunaddr, addr_len,
856 sk->sk_type, hash)) {
857 unix_release_addr(addr);
861 list = &unix_socket_table[addr->hash];
863 list = &unix_socket_table[dentry->d_inode->i_ino & (UNIX_HASH_SIZE-1)];
864 u->dentry = nd.path.dentry;
865 u->mnt = nd.path.mnt;
869 __unix_remove_socket(sk);
871 __unix_insert_socket(list, sk);
874 spin_unlock(&unix_table_lock);
876 mutex_unlock(&u->readlock);
883 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
888 unix_release_addr(addr);
892 static void unix_state_double_lock(struct sock *sk1, struct sock *sk2)
894 if (unlikely(sk1 == sk2) || !sk2) {
895 unix_state_lock(sk1);
899 unix_state_lock(sk1);
900 unix_state_lock_nested(sk2);
902 unix_state_lock(sk2);
903 unix_state_lock_nested(sk1);
907 static void unix_state_double_unlock(struct sock *sk1, struct sock *sk2)
909 if (unlikely(sk1 == sk2) || !sk2) {
910 unix_state_unlock(sk1);
913 unix_state_unlock(sk1);
914 unix_state_unlock(sk2);
917 static int unix_dgram_connect(struct socket *sock, struct sockaddr *addr,
920 struct sock *sk = sock->sk;
921 struct net *net = sock_net(sk);
922 struct sockaddr_un *sunaddr = (struct sockaddr_un *)addr;
927 if (addr->sa_family != AF_UNSPEC) {
928 err = unix_mkname(sunaddr, alen, &hash);
933 if (test_bit(SOCK_PASSCRED, &sock->flags) &&
934 !unix_sk(sk)->addr && (err = unix_autobind(sock)) != 0)
938 other = unix_find_other(net, sunaddr, alen, sock->type, hash, &err);
942 unix_state_double_lock(sk, other);
944 /* Apparently VFS overslept socket death. Retry. */
945 if (sock_flag(other, SOCK_DEAD)) {
946 unix_state_double_unlock(sk, other);
952 if (!unix_may_send(sk, other))
955 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
961 * 1003.1g breaking connected state with AF_UNSPEC
964 unix_state_double_lock(sk, other);
968 * If it was connected, reconnect.
971 struct sock *old_peer = unix_peer(sk);
972 unix_peer(sk) = other;
973 unix_state_double_unlock(sk, other);
975 if (other != old_peer)
976 unix_dgram_disconnected(sk, old_peer);
979 unix_peer(sk) = other;
980 unix_state_double_unlock(sk, other);
985 unix_state_double_unlock(sk, other);
991 static long unix_wait_for_peer(struct sock *other, long timeo)
993 struct unix_sock *u = unix_sk(other);
997 prepare_to_wait_exclusive(&u->peer_wait, &wait, TASK_INTERRUPTIBLE);
999 sched = !sock_flag(other, SOCK_DEAD) &&
1000 !(other->sk_shutdown & RCV_SHUTDOWN) &&
1001 unix_recvq_full(other);
1003 unix_state_unlock(other);
1006 timeo = schedule_timeout(timeo);
1008 finish_wait(&u->peer_wait, &wait);
1012 static int unix_stream_connect(struct socket *sock, struct sockaddr *uaddr,
1013 int addr_len, int flags)
1015 struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
1016 struct sock *sk = sock->sk;
1017 struct net *net = sock_net(sk);
1018 struct unix_sock *u = unix_sk(sk), *newu, *otheru;
1019 struct sock *newsk = NULL;
1020 struct sock *other = NULL;
1021 struct sk_buff *skb = NULL;
1027 err = unix_mkname(sunaddr, addr_len, &hash);
1032 if (test_bit(SOCK_PASSCRED, &sock->flags)
1033 && !u->addr && (err = unix_autobind(sock)) != 0)
1036 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
1038 /* First of all allocate resources.
1039 If we will make it after state is locked,
1040 we will have to recheck all again in any case.
1045 /* create new sock for complete connection */
1046 newsk = unix_create1(sock_net(sk), NULL);
1050 /* Allocate skb for sending to listening sock */
1051 skb = sock_wmalloc(newsk, 1, 0, GFP_KERNEL);
1056 /* Find listening sock. */
1057 other = unix_find_other(net, sunaddr, addr_len, sk->sk_type, hash, &err);
1061 /* Latch state of peer */
1062 unix_state_lock(other);
1064 /* Apparently VFS overslept socket death. Retry. */
1065 if (sock_flag(other, SOCK_DEAD)) {
1066 unix_state_unlock(other);
1071 err = -ECONNREFUSED;
1072 if (other->sk_state != TCP_LISTEN)
1075 if (unix_recvq_full(other)) {
1080 timeo = unix_wait_for_peer(other, timeo);
1082 err = sock_intr_errno(timeo);
1083 if (signal_pending(current))
1091 It is tricky place. We need to grab write lock and cannot
1092 drop lock on peer. It is dangerous because deadlock is
1093 possible. Connect to self case and simultaneous
1094 attempt to connect are eliminated by checking socket
1095 state. other is TCP_LISTEN, if sk is TCP_LISTEN we
1096 check this before attempt to grab lock.
1098 Well, and we have to recheck the state after socket locked.
1104 /* This is ok... continue with connect */
1106 case TCP_ESTABLISHED:
1107 /* Socket is already connected */
1115 unix_state_lock_nested(sk);
1117 if (sk->sk_state != st) {
1118 unix_state_unlock(sk);
1119 unix_state_unlock(other);
1124 err = security_unix_stream_connect(sock, other->sk_socket, newsk);
1126 unix_state_unlock(sk);
1130 /* The way is open! Fastly set all the necessary fields... */
1133 unix_peer(newsk) = sk;
1134 newsk->sk_state = TCP_ESTABLISHED;
1135 newsk->sk_type = sk->sk_type;
1136 newsk->sk_peercred.pid = task_tgid_vnr(current);
1137 newsk->sk_peercred.uid = current->euid;
1138 newsk->sk_peercred.gid = current->egid;
1139 newu = unix_sk(newsk);
1140 newsk->sk_sleep = &newu->peer_wait;
1141 otheru = unix_sk(other);
1143 /* copy address information from listening to new sock*/
1145 atomic_inc(&otheru->addr->refcnt);
1146 newu->addr = otheru->addr;
1148 if (otheru->dentry) {
1149 newu->dentry = dget(otheru->dentry);
1150 newu->mnt = mntget(otheru->mnt);
1153 /* Set credentials */
1154 sk->sk_peercred = other->sk_peercred;
1156 sock->state = SS_CONNECTED;
1157 sk->sk_state = TCP_ESTABLISHED;
1160 smp_mb__after_atomic_inc(); /* sock_hold() does an atomic_inc() */
1161 unix_peer(sk) = newsk;
1163 unix_state_unlock(sk);
1165 /* take ten and and send info to listening sock */
1166 spin_lock(&other->sk_receive_queue.lock);
1167 __skb_queue_tail(&other->sk_receive_queue, skb);
1168 spin_unlock(&other->sk_receive_queue.lock);
1169 unix_state_unlock(other);
1170 other->sk_data_ready(other, 0);
1176 unix_state_unlock(other);
1182 unix_release_sock(newsk, 0);
1188 static int unix_socketpair(struct socket *socka, struct socket *sockb)
1190 struct sock *ska = socka->sk, *skb = sockb->sk;
1192 /* Join our sockets back to back */
1195 unix_peer(ska) = skb;
1196 unix_peer(skb) = ska;
1197 ska->sk_peercred.pid = skb->sk_peercred.pid = task_tgid_vnr(current);
1198 ska->sk_peercred.uid = skb->sk_peercred.uid = current->euid;
1199 ska->sk_peercred.gid = skb->sk_peercred.gid = current->egid;
1201 if (ska->sk_type != SOCK_DGRAM) {
1202 ska->sk_state = TCP_ESTABLISHED;
1203 skb->sk_state = TCP_ESTABLISHED;
1204 socka->state = SS_CONNECTED;
1205 sockb->state = SS_CONNECTED;
1210 static int unix_accept(struct socket *sock, struct socket *newsock, int flags)
1212 struct sock *sk = sock->sk;
1214 struct sk_buff *skb;
1218 if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
1222 if (sk->sk_state != TCP_LISTEN)
1225 /* If socket state is TCP_LISTEN it cannot change (for now...),
1226 * so that no locks are necessary.
1229 skb = skb_recv_datagram(sk, 0, flags&O_NONBLOCK, &err);
1231 /* This means receive shutdown. */
1238 skb_free_datagram(sk, skb);
1239 wake_up_interruptible(&unix_sk(sk)->peer_wait);
1241 /* attach accepted sock to socket */
1242 unix_state_lock(tsk);
1243 newsock->state = SS_CONNECTED;
1244 sock_graft(tsk, newsock);
1245 unix_state_unlock(tsk);
1253 static int unix_getname(struct socket *sock, struct sockaddr *uaddr, int *uaddr_len, int peer)
1255 struct sock *sk = sock->sk;
1256 struct unix_sock *u;
1257 struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
1261 sk = unix_peer_get(sk);
1272 unix_state_lock(sk);
1274 sunaddr->sun_family = AF_UNIX;
1275 sunaddr->sun_path[0] = 0;
1276 *uaddr_len = sizeof(short);
1278 struct unix_address *addr = u->addr;
1280 *uaddr_len = addr->len;
1281 memcpy(sunaddr, addr->name, *uaddr_len);
1283 unix_state_unlock(sk);
1289 static void unix_detach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1293 scm->fp = UNIXCB(skb).fp;
1294 skb->destructor = sock_wfree;
1295 UNIXCB(skb).fp = NULL;
1297 for (i = scm->fp->count-1; i >= 0; i--)
1298 unix_notinflight(scm->fp->fp[i]);
1301 static void unix_destruct_fds(struct sk_buff *skb)
1303 struct scm_cookie scm;
1304 memset(&scm, 0, sizeof(scm));
1305 unix_detach_fds(&scm, skb);
1307 /* Alas, it calls VFS */
1308 /* So fscking what? fput() had been SMP-safe since the last Summer */
1313 static int unix_attach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1318 * Need to duplicate file references for the sake of garbage
1319 * collection. Otherwise a socket in the fps might become a
1320 * candidate for GC while the skb is not yet queued.
1322 UNIXCB(skb).fp = scm_fp_dup(scm->fp);
1323 if (!UNIXCB(skb).fp)
1326 for (i = scm->fp->count-1; i >= 0; i--)
1327 unix_inflight(scm->fp->fp[i]);
1328 skb->destructor = unix_destruct_fds;
1333 * Send AF_UNIX data.
1336 static int unix_dgram_sendmsg(struct kiocb *kiocb, struct socket *sock,
1337 struct msghdr *msg, size_t len)
1339 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1340 struct sock *sk = sock->sk;
1341 struct net *net = sock_net(sk);
1342 struct unix_sock *u = unix_sk(sk);
1343 struct sockaddr_un *sunaddr = msg->msg_name;
1344 struct sock *other = NULL;
1345 int namelen = 0; /* fake GCC */
1348 struct sk_buff *skb;
1350 struct scm_cookie tmp_scm;
1352 if (NULL == siocb->scm)
1353 siocb->scm = &tmp_scm;
1354 err = scm_send(sock, msg, siocb->scm);
1359 if (msg->msg_flags&MSG_OOB)
1362 if (msg->msg_namelen) {
1363 err = unix_mkname(sunaddr, msg->msg_namelen, &hash);
1370 other = unix_peer_get(sk);
1375 if (test_bit(SOCK_PASSCRED, &sock->flags)
1376 && !u->addr && (err = unix_autobind(sock)) != 0)
1380 if (len > sk->sk_sndbuf - 32)
1383 skb = sock_alloc_send_skb(sk, len, msg->msg_flags&MSG_DONTWAIT, &err);
1387 memcpy(UNIXCREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1388 if (siocb->scm->fp) {
1389 err = unix_attach_fds(siocb->scm, skb);
1393 unix_get_secdata(siocb->scm, skb);
1395 skb_reset_transport_header(skb);
1396 err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
1400 timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1405 if (sunaddr == NULL)
1408 other = unix_find_other(net, sunaddr, namelen, sk->sk_type,
1414 unix_state_lock(other);
1416 if (!unix_may_send(sk, other))
1419 if (sock_flag(other, SOCK_DEAD)) {
1421 * Check with 1003.1g - what should
1424 unix_state_unlock(other);
1428 unix_state_lock(sk);
1429 if (unix_peer(sk) == other) {
1430 unix_peer(sk) = NULL;
1431 unix_state_unlock(sk);
1433 unix_dgram_disconnected(sk, other);
1435 err = -ECONNREFUSED;
1437 unix_state_unlock(sk);
1447 if (other->sk_shutdown & RCV_SHUTDOWN)
1450 if (sk->sk_type != SOCK_SEQPACKET) {
1451 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1456 if (unix_peer(other) != sk && unix_recvq_full(other)) {
1462 timeo = unix_wait_for_peer(other, timeo);
1464 err = sock_intr_errno(timeo);
1465 if (signal_pending(current))
1471 skb_queue_tail(&other->sk_receive_queue, skb);
1472 unix_state_unlock(other);
1473 other->sk_data_ready(other, len);
1475 scm_destroy(siocb->scm);
1479 unix_state_unlock(other);
1485 scm_destroy(siocb->scm);
1490 static int unix_stream_sendmsg(struct kiocb *kiocb, struct socket *sock,
1491 struct msghdr *msg, size_t len)
1493 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1494 struct sock *sk = sock->sk;
1495 struct sock *other = NULL;
1496 struct sockaddr_un *sunaddr = msg->msg_name;
1498 struct sk_buff *skb;
1500 struct scm_cookie tmp_scm;
1502 if (NULL == siocb->scm)
1503 siocb->scm = &tmp_scm;
1504 err = scm_send(sock, msg, siocb->scm);
1509 if (msg->msg_flags&MSG_OOB)
1512 if (msg->msg_namelen) {
1513 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
1518 other = unix_peer(sk);
1523 if (sk->sk_shutdown & SEND_SHUTDOWN)
1526 while (sent < len) {
1528 * Optimisation for the fact that under 0.01% of X
1529 * messages typically need breaking up.
1534 /* Keep two messages in the pipe so it schedules better */
1535 if (size > ((sk->sk_sndbuf >> 1) - 64))
1536 size = (sk->sk_sndbuf >> 1) - 64;
1538 if (size > SKB_MAX_ALLOC)
1539 size = SKB_MAX_ALLOC;
1545 skb = sock_alloc_send_skb(sk, size, msg->msg_flags&MSG_DONTWAIT,
1552 * If you pass two values to the sock_alloc_send_skb
1553 * it tries to grab the large buffer with GFP_NOFS
1554 * (which can fail easily), and if it fails grab the
1555 * fallback size buffer which is under a page and will
1558 size = min_t(int, size, skb_tailroom(skb));
1560 memcpy(UNIXCREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1561 if (siocb->scm->fp) {
1562 err = unix_attach_fds(siocb->scm, skb);
1569 err = memcpy_fromiovec(skb_put(skb, size), msg->msg_iov, size);
1575 unix_state_lock(other);
1577 if (sock_flag(other, SOCK_DEAD) ||
1578 (other->sk_shutdown & RCV_SHUTDOWN))
1581 skb_queue_tail(&other->sk_receive_queue, skb);
1582 unix_state_unlock(other);
1583 other->sk_data_ready(other, size);
1587 scm_destroy(siocb->scm);
1593 unix_state_unlock(other);
1596 if (sent == 0 && !(msg->msg_flags&MSG_NOSIGNAL))
1597 send_sig(SIGPIPE, current, 0);
1600 scm_destroy(siocb->scm);
1602 return sent ? : err;
1605 static int unix_seqpacket_sendmsg(struct kiocb *kiocb, struct socket *sock,
1606 struct msghdr *msg, size_t len)
1609 struct sock *sk = sock->sk;
1611 err = sock_error(sk);
1615 if (sk->sk_state != TCP_ESTABLISHED)
1618 if (msg->msg_namelen)
1619 msg->msg_namelen = 0;
1621 return unix_dgram_sendmsg(kiocb, sock, msg, len);
1624 static void unix_copy_addr(struct msghdr *msg, struct sock *sk)
1626 struct unix_sock *u = unix_sk(sk);
1628 msg->msg_namelen = 0;
1630 msg->msg_namelen = u->addr->len;
1631 memcpy(msg->msg_name, u->addr->name, u->addr->len);
1635 static int unix_dgram_recvmsg(struct kiocb *iocb, struct socket *sock,
1636 struct msghdr *msg, size_t size,
1639 struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1640 struct scm_cookie tmp_scm;
1641 struct sock *sk = sock->sk;
1642 struct unix_sock *u = unix_sk(sk);
1643 int noblock = flags & MSG_DONTWAIT;
1644 struct sk_buff *skb;
1651 msg->msg_namelen = 0;
1653 mutex_lock(&u->readlock);
1655 skb = skb_recv_datagram(sk, flags, noblock, &err);
1657 unix_state_lock(sk);
1658 /* Signal EOF on disconnected non-blocking SEQPACKET socket. */
1659 if (sk->sk_type == SOCK_SEQPACKET && err == -EAGAIN &&
1660 (sk->sk_shutdown & RCV_SHUTDOWN))
1662 unix_state_unlock(sk);
1666 wake_up_interruptible_sync(&u->peer_wait);
1669 unix_copy_addr(msg, skb->sk);
1671 if (size > skb->len)
1673 else if (size < skb->len)
1674 msg->msg_flags |= MSG_TRUNC;
1676 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, size);
1681 siocb->scm = &tmp_scm;
1682 memset(&tmp_scm, 0, sizeof(tmp_scm));
1684 siocb->scm->creds = *UNIXCREDS(skb);
1685 unix_set_secdata(siocb->scm, skb);
1687 if (!(flags & MSG_PEEK)) {
1689 unix_detach_fds(siocb->scm, skb);
1691 /* It is questionable: on PEEK we could:
1692 - do not return fds - good, but too simple 8)
1693 - return fds, and do not return them on read (old strategy,
1695 - clone fds (I chose it for now, it is the most universal
1698 POSIX 1003.1g does not actually define this clearly
1699 at all. POSIX 1003.1g doesn't define a lot of things
1704 siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1708 scm_recv(sock, msg, siocb->scm, flags);
1711 skb_free_datagram(sk, skb);
1713 mutex_unlock(&u->readlock);
1719 * Sleep until data has arrive. But check for races..
1722 static long unix_stream_data_wait(struct sock *sk, long timeo)
1726 unix_state_lock(sk);
1729 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1731 if (!skb_queue_empty(&sk->sk_receive_queue) ||
1733 (sk->sk_shutdown & RCV_SHUTDOWN) ||
1734 signal_pending(current) ||
1738 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1739 unix_state_unlock(sk);
1740 timeo = schedule_timeout(timeo);
1741 unix_state_lock(sk);
1742 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1745 finish_wait(sk->sk_sleep, &wait);
1746 unix_state_unlock(sk);
1752 static int unix_stream_recvmsg(struct kiocb *iocb, struct socket *sock,
1753 struct msghdr *msg, size_t size,
1756 struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1757 struct scm_cookie tmp_scm;
1758 struct sock *sk = sock->sk;
1759 struct unix_sock *u = unix_sk(sk);
1760 struct sockaddr_un *sunaddr = msg->msg_name;
1762 int check_creds = 0;
1768 if (sk->sk_state != TCP_ESTABLISHED)
1775 target = sock_rcvlowat(sk, flags&MSG_WAITALL, size);
1776 timeo = sock_rcvtimeo(sk, flags&MSG_DONTWAIT);
1778 msg->msg_namelen = 0;
1780 /* Lock the socket to prevent queue disordering
1781 * while sleeps in memcpy_tomsg
1785 siocb->scm = &tmp_scm;
1786 memset(&tmp_scm, 0, sizeof(tmp_scm));
1789 mutex_lock(&u->readlock);
1793 struct sk_buff *skb;
1795 unix_state_lock(sk);
1796 skb = skb_dequeue(&sk->sk_receive_queue);
1798 if (copied >= target)
1802 * POSIX 1003.1g mandates this order.
1805 err = sock_error(sk);
1808 if (sk->sk_shutdown & RCV_SHUTDOWN)
1811 unix_state_unlock(sk);
1815 mutex_unlock(&u->readlock);
1817 timeo = unix_stream_data_wait(sk, timeo);
1819 if (signal_pending(current)) {
1820 err = sock_intr_errno(timeo);
1823 mutex_lock(&u->readlock);
1826 unix_state_unlock(sk);
1829 unix_state_unlock(sk);
1832 /* Never glue messages from different writers */
1833 if (memcmp(UNIXCREDS(skb), &siocb->scm->creds,
1834 sizeof(siocb->scm->creds)) != 0) {
1835 skb_queue_head(&sk->sk_receive_queue, skb);
1839 /* Copy credentials */
1840 siocb->scm->creds = *UNIXCREDS(skb);
1844 /* Copy address just once */
1846 unix_copy_addr(msg, skb->sk);
1850 chunk = min_t(unsigned int, skb->len, size);
1851 if (memcpy_toiovec(msg->msg_iov, skb->data, chunk)) {
1852 skb_queue_head(&sk->sk_receive_queue, skb);
1860 /* Mark read part of skb as used */
1861 if (!(flags & MSG_PEEK)) {
1862 skb_pull(skb, chunk);
1865 unix_detach_fds(siocb->scm, skb);
1867 /* put the skb back if we didn't use it up.. */
1869 skb_queue_head(&sk->sk_receive_queue, skb);
1878 /* It is questionable, see note in unix_dgram_recvmsg.
1881 siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1883 /* put message back and return */
1884 skb_queue_head(&sk->sk_receive_queue, skb);
1889 mutex_unlock(&u->readlock);
1890 scm_recv(sock, msg, siocb->scm, flags);
1892 return copied ? : err;
1895 static int unix_shutdown(struct socket *sock, int mode)
1897 struct sock *sk = sock->sk;
1900 mode = (mode+1)&(RCV_SHUTDOWN|SEND_SHUTDOWN);
1903 unix_state_lock(sk);
1904 sk->sk_shutdown |= mode;
1905 other = unix_peer(sk);
1908 unix_state_unlock(sk);
1909 sk->sk_state_change(sk);
1912 (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET)) {
1916 if (mode&RCV_SHUTDOWN)
1917 peer_mode |= SEND_SHUTDOWN;
1918 if (mode&SEND_SHUTDOWN)
1919 peer_mode |= RCV_SHUTDOWN;
1920 unix_state_lock(other);
1921 other->sk_shutdown |= peer_mode;
1922 unix_state_unlock(other);
1923 other->sk_state_change(other);
1924 read_lock(&other->sk_callback_lock);
1925 if (peer_mode == SHUTDOWN_MASK)
1926 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_HUP);
1927 else if (peer_mode & RCV_SHUTDOWN)
1928 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_IN);
1929 read_unlock(&other->sk_callback_lock);
1937 static int unix_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1939 struct sock *sk = sock->sk;
1945 amount = atomic_read(&sk->sk_wmem_alloc);
1946 err = put_user(amount, (int __user *)arg);
1950 struct sk_buff *skb;
1952 if (sk->sk_state == TCP_LISTEN) {
1957 spin_lock(&sk->sk_receive_queue.lock);
1958 if (sk->sk_type == SOCK_STREAM ||
1959 sk->sk_type == SOCK_SEQPACKET) {
1960 skb_queue_walk(&sk->sk_receive_queue, skb)
1963 skb = skb_peek(&sk->sk_receive_queue);
1967 spin_unlock(&sk->sk_receive_queue.lock);
1968 err = put_user(amount, (int __user *)arg);
1979 static unsigned int unix_poll(struct file *file, struct socket *sock, poll_table *wait)
1981 struct sock *sk = sock->sk;
1984 poll_wait(file, sk->sk_sleep, wait);
1987 /* exceptional events? */
1990 if (sk->sk_shutdown == SHUTDOWN_MASK)
1992 if (sk->sk_shutdown & RCV_SHUTDOWN)
1996 if (!skb_queue_empty(&sk->sk_receive_queue) ||
1997 (sk->sk_shutdown & RCV_SHUTDOWN))
1998 mask |= POLLIN | POLLRDNORM;
2000 /* Connection-based need to check for termination and startup */
2001 if ((sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) &&
2002 sk->sk_state == TCP_CLOSE)
2006 * we set writable also when the other side has shut down the
2007 * connection. This prevents stuck sockets.
2009 if (unix_writable(sk))
2010 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2015 static unsigned int unix_dgram_poll(struct file *file, struct socket *sock,
2018 struct sock *sk = sock->sk, *other;
2019 unsigned int mask, writable;
2021 poll_wait(file, sk->sk_sleep, wait);
2024 /* exceptional events? */
2025 if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue))
2027 if (sk->sk_shutdown & RCV_SHUTDOWN)
2029 if (sk->sk_shutdown == SHUTDOWN_MASK)
2033 if (!skb_queue_empty(&sk->sk_receive_queue) ||
2034 (sk->sk_shutdown & RCV_SHUTDOWN))
2035 mask |= POLLIN | POLLRDNORM;
2037 /* Connection-based need to check for termination and startup */
2038 if (sk->sk_type == SOCK_SEQPACKET) {
2039 if (sk->sk_state == TCP_CLOSE)
2041 /* connection hasn't started yet? */
2042 if (sk->sk_state == TCP_SYN_SENT)
2047 writable = unix_writable(sk);
2049 other = unix_peer_get(sk);
2051 if (unix_peer(other) != sk) {
2052 poll_wait(file, &unix_sk(other)->peer_wait,
2054 if (unix_recvq_full(other))
2063 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2065 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
2070 #ifdef CONFIG_PROC_FS
2071 static struct sock *first_unix_socket(int *i)
2073 for (*i = 0; *i <= UNIX_HASH_SIZE; (*i)++) {
2074 if (!hlist_empty(&unix_socket_table[*i]))
2075 return __sk_head(&unix_socket_table[*i]);
2080 static struct sock *next_unix_socket(int *i, struct sock *s)
2082 struct sock *next = sk_next(s);
2083 /* More in this chain? */
2086 /* Look for next non-empty chain. */
2087 for ((*i)++; *i <= UNIX_HASH_SIZE; (*i)++) {
2088 if (!hlist_empty(&unix_socket_table[*i]))
2089 return __sk_head(&unix_socket_table[*i]);
2094 struct unix_iter_state {
2095 struct seq_net_private p;
2099 static struct sock *unix_seq_idx(struct seq_file *seq, loff_t pos)
2101 struct unix_iter_state *iter = seq->private;
2105 for (s = first_unix_socket(&iter->i); s; s = next_unix_socket(&iter->i, s)) {
2106 if (sock_net(s) != seq_file_net(seq))
2115 static void *unix_seq_start(struct seq_file *seq, loff_t *pos)
2116 __acquires(unix_table_lock)
2118 spin_lock(&unix_table_lock);
2119 return *pos ? unix_seq_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2122 static void *unix_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2124 struct unix_iter_state *iter = seq->private;
2125 struct sock *sk = v;
2128 if (v == SEQ_START_TOKEN)
2129 sk = first_unix_socket(&iter->i);
2131 sk = next_unix_socket(&iter->i, sk);
2132 while (sk && (sock_net(sk) != seq_file_net(seq)))
2133 sk = next_unix_socket(&iter->i, sk);
2137 static void unix_seq_stop(struct seq_file *seq, void *v)
2138 __releases(unix_table_lock)
2140 spin_unlock(&unix_table_lock);
2143 static int unix_seq_show(struct seq_file *seq, void *v)
2146 if (v == SEQ_START_TOKEN)
2147 seq_puts(seq, "Num RefCount Protocol Flags Type St "
2151 struct unix_sock *u = unix_sk(s);
2154 seq_printf(seq, "%p: %08X %08X %08X %04X %02X %5lu",
2156 atomic_read(&s->sk_refcnt),
2158 s->sk_state == TCP_LISTEN ? __SO_ACCEPTCON : 0,
2161 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTED : SS_UNCONNECTED) :
2162 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTING : SS_DISCONNECTING),
2170 len = u->addr->len - sizeof(short);
2171 if (!UNIX_ABSTRACT(s))
2177 for ( ; i < len; i++)
2178 seq_putc(seq, u->addr->name->sun_path[i]);
2180 unix_state_unlock(s);
2181 seq_putc(seq, '\n');
2187 static const struct seq_operations unix_seq_ops = {
2188 .start = unix_seq_start,
2189 .next = unix_seq_next,
2190 .stop = unix_seq_stop,
2191 .show = unix_seq_show,
2194 static int unix_seq_open(struct inode *inode, struct file *file)
2196 return seq_open_net(inode, file, &unix_seq_ops,
2197 sizeof(struct unix_iter_state));
2200 static const struct file_operations unix_seq_fops = {
2201 .owner = THIS_MODULE,
2202 .open = unix_seq_open,
2204 .llseek = seq_lseek,
2205 .release = seq_release_net,
2210 static struct net_proto_family unix_family_ops = {
2212 .create = unix_create,
2213 .owner = THIS_MODULE,
2217 static int unix_net_init(struct net *net)
2219 int error = -ENOMEM;
2221 net->unx.sysctl_max_dgram_qlen = 10;
2222 if (unix_sysctl_register(net))
2225 #ifdef CONFIG_PROC_FS
2226 if (!proc_net_fops_create(net, "unix", 0, &unix_seq_fops)) {
2227 unix_sysctl_unregister(net);
2236 static void unix_net_exit(struct net *net)
2238 unix_sysctl_unregister(net);
2239 proc_net_remove(net, "unix");
2242 static struct pernet_operations unix_net_ops = {
2243 .init = unix_net_init,
2244 .exit = unix_net_exit,
2247 static int __init af_unix_init(void)
2250 struct sk_buff *dummy_skb;
2252 BUILD_BUG_ON(sizeof(struct unix_skb_parms) > sizeof(dummy_skb->cb));
2254 rc = proto_register(&unix_proto, 1);
2256 printk(KERN_CRIT "%s: Cannot create unix_sock SLAB cache!\n",
2261 sock_register(&unix_family_ops);
2262 register_pernet_subsys(&unix_net_ops);
2267 static void __exit af_unix_exit(void)
2269 sock_unregister(PF_UNIX);
2270 proto_unregister(&unix_proto);
2271 unregister_pernet_subsys(&unix_net_ops);
2274 /* Earlier than device_initcall() so that other drivers invoking
2275 request_module() don't end up in a loop when modprobe tries
2276 to use a UNIX socket. But later than subsys_initcall() because
2277 we depend on stuff initialised there */
2278 fs_initcall(af_unix_init);
2279 module_exit(af_unix_exit);
2281 MODULE_LICENSE("GPL");
2282 MODULE_ALIAS_NETPROTO(PF_UNIX);