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_long_t unix_nr_socks;
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 spin lock.
150 static inline unsigned unix_hash_fold(__wsum n)
152 unsigned int hash = (__force unsigned int)csum_fold(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] doesn't 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 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 (dentry && dentry->d_inode == i) {
302 spin_unlock(&unix_table_lock);
306 /* Support code for asymmetrically connected dgram sockets
308 * If a datagram socket is connected to a socket not itself connected
309 * to the first socket (eg, /dev/log), clients may only enqueue more
310 * messages if the present receive queue of the server socket is not
311 * "too large". This means there's a second writeability condition
312 * poll and sendmsg need to test. The dgram recv code will do a wake
313 * up on the peer_wait wait queue of a socket upon reception of a
314 * datagram which needs to be propagated to sleeping would-be writers
315 * since these might not have sent anything so far. This can't be
316 * accomplished via poll_wait because the lifetime of the server
317 * socket might be less than that of its clients if these break their
318 * association with it or if the server socket is closed while clients
319 * are still connected to it and there's no way to inform "a polling
320 * implementation" that it should let go of a certain wait queue
322 * In order to propagate a wake up, a wait_queue_t of the client
323 * socket is enqueued on the peer_wait queue of the server socket
324 * whose wake function does a wake_up on the ordinary client socket
325 * wait queue. This connection is established whenever a write (or
326 * poll for write) hit the flow control condition and broken when the
327 * association to the server socket is dissolved or after a wake up
331 static int unix_dgram_peer_wake_relay(wait_queue_t *q, unsigned mode, int flags,
335 wait_queue_head_t *u_sleep;
337 u = container_of(q, struct unix_sock, peer_wake);
339 __remove_wait_queue(&unix_sk(u->peer_wake.private)->peer_wait,
341 u->peer_wake.private = NULL;
343 /* relaying can only happen while the wq still exists */
344 u_sleep = sk_sleep(&u->sk);
346 wake_up_interruptible_poll(u_sleep, key);
351 static int unix_dgram_peer_wake_connect(struct sock *sk, struct sock *other)
353 struct unix_sock *u, *u_other;
357 u_other = unix_sk(other);
359 spin_lock(&u_other->peer_wait.lock);
361 if (!u->peer_wake.private) {
362 u->peer_wake.private = other;
363 __add_wait_queue(&u_other->peer_wait, &u->peer_wake);
368 spin_unlock(&u_other->peer_wait.lock);
372 static void unix_dgram_peer_wake_disconnect(struct sock *sk,
375 struct unix_sock *u, *u_other;
378 u_other = unix_sk(other);
379 spin_lock(&u_other->peer_wait.lock);
381 if (u->peer_wake.private == other) {
382 __remove_wait_queue(&u_other->peer_wait, &u->peer_wake);
383 u->peer_wake.private = NULL;
386 spin_unlock(&u_other->peer_wait.lock);
389 static void unix_dgram_peer_wake_disconnect_wakeup(struct sock *sk,
392 unix_dgram_peer_wake_disconnect(sk, other);
393 wake_up_interruptible_poll(sk_sleep(sk),
400 * - unix_peer(sk) == other
401 * - association is stable
403 static int unix_dgram_peer_wake_me(struct sock *sk, struct sock *other)
407 connected = unix_dgram_peer_wake_connect(sk, other);
409 if (unix_recvq_full(other))
413 unix_dgram_peer_wake_disconnect(sk, other);
418 static inline int unix_writable(struct sock *sk)
420 return (atomic_read(&sk->sk_wmem_alloc) << 2) <= sk->sk_sndbuf;
423 static void unix_write_space(struct sock *sk)
425 struct socket_wq *wq;
428 if (unix_writable(sk)) {
429 wq = rcu_dereference(sk->sk_wq);
430 if (wq_has_sleeper(wq))
431 wake_up_interruptible_sync_poll(&wq->wait,
432 POLLOUT | POLLWRNORM | POLLWRBAND);
433 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
438 /* When dgram socket disconnects (or changes its peer), we clear its receive
439 * queue of packets arrived from previous peer. First, it allows to do
440 * flow control based only on wmem_alloc; second, sk connected to peer
441 * may receive messages only from that peer. */
442 static void unix_dgram_disconnected(struct sock *sk, struct sock *other)
444 if (!skb_queue_empty(&sk->sk_receive_queue)) {
445 skb_queue_purge(&sk->sk_receive_queue);
446 wake_up_interruptible_all(&unix_sk(sk)->peer_wait);
448 /* If one link of bidirectional dgram pipe is disconnected,
449 * we signal error. Messages are lost. Do not make this,
450 * when peer was not connected to us.
452 if (!sock_flag(other, SOCK_DEAD) && unix_peer(other) == sk) {
453 other->sk_err = ECONNRESET;
454 other->sk_error_report(other);
459 static void unix_sock_destructor(struct sock *sk)
461 struct unix_sock *u = unix_sk(sk);
463 skb_queue_purge(&sk->sk_receive_queue);
465 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
466 WARN_ON(!sk_unhashed(sk));
467 WARN_ON(sk->sk_socket);
468 if (!sock_flag(sk, SOCK_DEAD)) {
469 printk(KERN_INFO "Attempt to release alive unix socket: %p\n", sk);
474 unix_release_addr(u->addr);
476 atomic_long_dec(&unix_nr_socks);
478 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
480 #ifdef UNIX_REFCNT_DEBUG
481 printk(KERN_DEBUG "UNIX %p is destroyed, %ld are still alive.\n", sk,
482 atomic_long_read(&unix_nr_socks));
486 static void unix_release_sock(struct sock *sk, int embrion)
488 struct unix_sock *u = unix_sk(sk);
489 struct dentry *dentry;
490 struct vfsmount *mnt;
495 unix_remove_socket(sk);
500 sk->sk_shutdown = SHUTDOWN_MASK;
505 state = sk->sk_state;
506 sk->sk_state = TCP_CLOSE;
507 unix_state_unlock(sk);
509 wake_up_interruptible_all(&u->peer_wait);
511 skpair = unix_peer(sk);
513 if (skpair != NULL) {
514 if (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) {
515 unix_state_lock(skpair);
517 skpair->sk_shutdown = SHUTDOWN_MASK;
518 if (!skb_queue_empty(&sk->sk_receive_queue) || embrion)
519 skpair->sk_err = ECONNRESET;
520 unix_state_unlock(skpair);
521 skpair->sk_state_change(skpair);
522 sk_wake_async(skpair, SOCK_WAKE_WAITD, POLL_HUP);
525 unix_dgram_peer_wake_disconnect(sk, skpair);
526 sock_put(skpair); /* It may now die */
527 unix_peer(sk) = NULL;
530 /* Try to flush out this socket. Throw out buffers at least */
532 while ((skb = skb_dequeue(&sk->sk_receive_queue)) != NULL) {
533 if (state == TCP_LISTEN)
534 unix_release_sock(skb->sk, 1);
535 /* passed fds are erased in the kfree_skb hook */
546 /* ---- Socket is dead now and most probably destroyed ---- */
549 * Fixme: BSD difference: In BSD all sockets connected to use get
550 * ECONNRESET and we die on the spot. In Linux we behave
551 * like files and pipes do and wait for the last
554 * Can't we simply set sock->err?
556 * What the above comment does talk about? --ANK(980817)
559 if (unix_tot_inflight)
560 unix_gc(); /* Garbage collect fds */
563 static void init_peercred(struct sock *sk)
565 put_pid(sk->sk_peer_pid);
566 if (sk->sk_peer_cred)
567 put_cred(sk->sk_peer_cred);
568 sk->sk_peer_pid = get_pid(task_tgid(current));
569 sk->sk_peer_cred = get_current_cred();
572 static void copy_peercred(struct sock *sk, struct sock *peersk)
574 put_pid(sk->sk_peer_pid);
575 if (sk->sk_peer_cred)
576 put_cred(sk->sk_peer_cred);
577 sk->sk_peer_pid = get_pid(peersk->sk_peer_pid);
578 sk->sk_peer_cred = get_cred(peersk->sk_peer_cred);
581 static int unix_listen(struct socket *sock, int backlog)
584 struct sock *sk = sock->sk;
585 struct unix_sock *u = unix_sk(sk);
586 struct pid *old_pid = NULL;
587 const struct cred *old_cred = NULL;
590 if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
591 goto out; /* Only stream/seqpacket sockets accept */
594 goto out; /* No listens on an unbound socket */
596 if (sk->sk_state != TCP_CLOSE && sk->sk_state != TCP_LISTEN)
598 if (backlog > sk->sk_max_ack_backlog)
599 wake_up_interruptible_all(&u->peer_wait);
600 sk->sk_max_ack_backlog = backlog;
601 sk->sk_state = TCP_LISTEN;
602 /* set credentials so connect can copy them */
607 unix_state_unlock(sk);
615 static int unix_release(struct socket *);
616 static int unix_bind(struct socket *, struct sockaddr *, int);
617 static int unix_stream_connect(struct socket *, struct sockaddr *,
618 int addr_len, int flags);
619 static int unix_socketpair(struct socket *, struct socket *);
620 static int unix_accept(struct socket *, struct socket *, int);
621 static int unix_getname(struct socket *, struct sockaddr *, int *, int);
622 static unsigned int unix_poll(struct file *, struct socket *, poll_table *);
623 static unsigned int unix_dgram_poll(struct file *, struct socket *,
625 static int unix_ioctl(struct socket *, unsigned int, unsigned long);
626 static int unix_shutdown(struct socket *, int);
627 static int unix_stream_sendmsg(struct kiocb *, struct socket *,
628 struct msghdr *, size_t);
629 static int unix_stream_recvmsg(struct kiocb *, struct socket *,
630 struct msghdr *, size_t, int);
631 static int unix_dgram_sendmsg(struct kiocb *, struct socket *,
632 struct msghdr *, size_t);
633 static int unix_dgram_recvmsg(struct kiocb *, struct socket *,
634 struct msghdr *, size_t, int);
635 static int unix_dgram_connect(struct socket *, struct sockaddr *,
637 static int unix_seqpacket_sendmsg(struct kiocb *, struct socket *,
638 struct msghdr *, size_t);
639 static int unix_seqpacket_recvmsg(struct kiocb *, struct socket *,
640 struct msghdr *, size_t, int);
642 static const struct proto_ops unix_stream_ops = {
644 .owner = THIS_MODULE,
645 .release = unix_release,
647 .connect = unix_stream_connect,
648 .socketpair = unix_socketpair,
649 .accept = unix_accept,
650 .getname = unix_getname,
653 .listen = unix_listen,
654 .shutdown = unix_shutdown,
655 .setsockopt = sock_no_setsockopt,
656 .getsockopt = sock_no_getsockopt,
657 .sendmsg = unix_stream_sendmsg,
658 .recvmsg = unix_stream_recvmsg,
659 .mmap = sock_no_mmap,
660 .sendpage = sock_no_sendpage,
663 static const struct proto_ops unix_dgram_ops = {
665 .owner = THIS_MODULE,
666 .release = unix_release,
668 .connect = unix_dgram_connect,
669 .socketpair = unix_socketpair,
670 .accept = sock_no_accept,
671 .getname = unix_getname,
672 .poll = unix_dgram_poll,
674 .listen = sock_no_listen,
675 .shutdown = unix_shutdown,
676 .setsockopt = sock_no_setsockopt,
677 .getsockopt = sock_no_getsockopt,
678 .sendmsg = unix_dgram_sendmsg,
679 .recvmsg = unix_dgram_recvmsg,
680 .mmap = sock_no_mmap,
681 .sendpage = sock_no_sendpage,
684 static const struct proto_ops unix_seqpacket_ops = {
686 .owner = THIS_MODULE,
687 .release = unix_release,
689 .connect = unix_stream_connect,
690 .socketpair = unix_socketpair,
691 .accept = unix_accept,
692 .getname = unix_getname,
693 .poll = unix_dgram_poll,
695 .listen = unix_listen,
696 .shutdown = unix_shutdown,
697 .setsockopt = sock_no_setsockopt,
698 .getsockopt = sock_no_getsockopt,
699 .sendmsg = unix_seqpacket_sendmsg,
700 .recvmsg = unix_seqpacket_recvmsg,
701 .mmap = sock_no_mmap,
702 .sendpage = sock_no_sendpage,
705 static struct proto unix_proto = {
707 .owner = THIS_MODULE,
708 .obj_size = sizeof(struct unix_sock),
712 * AF_UNIX sockets do not interact with hardware, hence they
713 * dont trigger interrupts - so it's safe for them to have
714 * bh-unsafe locking for their sk_receive_queue.lock. Split off
715 * this special lock-class by reinitializing the spinlock key:
717 static struct lock_class_key af_unix_sk_receive_queue_lock_key;
719 static struct sock *unix_create1(struct net *net, struct socket *sock)
721 struct sock *sk = NULL;
724 atomic_long_inc(&unix_nr_socks);
725 if (atomic_long_read(&unix_nr_socks) > 2 * get_max_files())
728 sk = sk_alloc(net, PF_UNIX, GFP_KERNEL, &unix_proto);
732 sock_init_data(sock, sk);
733 lockdep_set_class(&sk->sk_receive_queue.lock,
734 &af_unix_sk_receive_queue_lock_key);
736 sk->sk_write_space = unix_write_space;
737 sk->sk_max_ack_backlog = net->unx.sysctl_max_dgram_qlen;
738 sk->sk_destruct = unix_sock_destructor;
742 spin_lock_init(&u->lock);
743 atomic_long_set(&u->inflight, 0);
744 INIT_LIST_HEAD(&u->link);
745 mutex_init(&u->readlock); /* single task reading lock */
746 init_waitqueue_head(&u->peer_wait);
747 init_waitqueue_func_entry(&u->peer_wake, unix_dgram_peer_wake_relay);
748 unix_insert_socket(unix_sockets_unbound, sk);
751 atomic_long_dec(&unix_nr_socks);
754 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
760 static int unix_create(struct net *net, struct socket *sock, int protocol,
763 if (protocol && protocol != PF_UNIX)
764 return -EPROTONOSUPPORT;
766 sock->state = SS_UNCONNECTED;
768 switch (sock->type) {
770 sock->ops = &unix_stream_ops;
773 * Believe it or not BSD has AF_UNIX, SOCK_RAW though
777 sock->type = SOCK_DGRAM;
779 sock->ops = &unix_dgram_ops;
782 sock->ops = &unix_seqpacket_ops;
785 return -ESOCKTNOSUPPORT;
788 return unix_create1(net, sock) ? 0 : -ENOMEM;
791 static int unix_release(struct socket *sock)
793 struct sock *sk = sock->sk;
798 unix_release_sock(sk, 0);
804 static int unix_autobind(struct socket *sock)
806 struct sock *sk = sock->sk;
807 struct net *net = sock_net(sk);
808 struct unix_sock *u = unix_sk(sk);
809 static u32 ordernum = 1;
810 struct unix_address *addr;
812 unsigned int retries = 0;
814 err = mutex_lock_interruptible(&u->readlock);
823 addr = kzalloc(sizeof(*addr) + sizeof(short) + 16, GFP_KERNEL);
827 addr->name->sun_family = AF_UNIX;
828 atomic_set(&addr->refcnt, 1);
831 addr->len = sprintf(addr->name->sun_path+1, "%05x", ordernum) + 1 + sizeof(short);
832 addr->hash = unix_hash_fold(csum_partial(addr->name, addr->len, 0));
834 spin_lock(&unix_table_lock);
835 ordernum = (ordernum+1)&0xFFFFF;
837 if (__unix_find_socket_byname(net, addr->name, addr->len, sock->type,
839 spin_unlock(&unix_table_lock);
841 * __unix_find_socket_byname() may take long time if many names
842 * are already in use.
845 /* Give up if all names seems to be in use. */
846 if (retries++ == 0xFFFFF) {
853 addr->hash ^= sk->sk_type;
855 __unix_remove_socket(sk);
857 __unix_insert_socket(&unix_socket_table[addr->hash], sk);
858 spin_unlock(&unix_table_lock);
861 out: mutex_unlock(&u->readlock);
865 static struct sock *unix_find_other(struct net *net,
866 struct sockaddr_un *sunname, int len,
867 int type, unsigned hash, int *error)
873 if (sunname->sun_path[0]) {
875 err = kern_path(sunname->sun_path, LOOKUP_FOLLOW, &path);
878 inode = path.dentry->d_inode;
879 err = inode_permission(inode, MAY_WRITE);
884 if (!S_ISSOCK(inode->i_mode))
886 u = unix_find_socket_byinode(inode);
890 if (u->sk_type == type)
891 touch_atime(path.mnt, path.dentry);
896 if (u->sk_type != type) {
902 u = unix_find_socket_byname(net, sunname, len, type, hash);
904 struct dentry *dentry;
905 dentry = unix_sk(u)->dentry;
907 touch_atime(unix_sk(u)->mnt, dentry);
921 static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
923 struct sock *sk = sock->sk;
924 struct net *net = sock_net(sk);
925 struct unix_sock *u = unix_sk(sk);
926 struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
927 char *sun_path = sunaddr->sun_path;
928 struct dentry *dentry = NULL;
932 struct unix_address *addr;
933 struct hlist_head *list;
936 if (sunaddr->sun_family != AF_UNIX)
939 if (addr_len == sizeof(short)) {
940 err = unix_autobind(sock);
944 err = unix_mkname(sunaddr, addr_len, &hash);
949 err = mutex_lock_interruptible(&u->readlock);
958 addr = kmalloc(sizeof(*addr)+addr_len, GFP_KERNEL);
962 memcpy(addr->name, sunaddr, addr_len);
963 addr->len = addr_len;
964 addr->hash = hash ^ sk->sk_type;
965 atomic_set(&addr->refcnt, 1);
971 * Get the parent directory, calculate the hash for last
974 dentry = kern_path_create(AT_FDCWD, sun_path, &path, 0);
975 err = PTR_ERR(dentry);
977 goto out_mknod_parent;
980 * All right, let's create it.
983 (SOCK_INODE(sock)->i_mode & ~current_umask());
984 err = mnt_want_write(path.mnt);
987 err = security_path_mknod(&path, dentry, mode, 0);
989 goto out_mknod_drop_write;
990 err = vfs_mknod(path.dentry->d_inode, dentry, mode, 0);
991 out_mknod_drop_write:
992 mnt_drop_write(path.mnt);
995 mutex_unlock(&path.dentry->d_inode->i_mutex);
997 path.dentry = dentry;
999 addr->hash = UNIX_HASH_SIZE;
1002 spin_lock(&unix_table_lock);
1006 if (__unix_find_socket_byname(net, sunaddr, addr_len,
1007 sk->sk_type, hash)) {
1008 unix_release_addr(addr);
1012 list = &unix_socket_table[addr->hash];
1014 list = &unix_socket_table[dentry->d_inode->i_ino & (UNIX_HASH_SIZE-1)];
1015 u->dentry = path.dentry;
1020 __unix_remove_socket(sk);
1022 __unix_insert_socket(list, sk);
1025 spin_unlock(&unix_table_lock);
1027 mutex_unlock(&u->readlock);
1033 mutex_unlock(&path.dentry->d_inode->i_mutex);
1038 unix_release_addr(addr);
1042 static void unix_state_double_lock(struct sock *sk1, struct sock *sk2)
1044 if (unlikely(sk1 == sk2) || !sk2) {
1045 unix_state_lock(sk1);
1049 unix_state_lock(sk1);
1050 unix_state_lock_nested(sk2);
1052 unix_state_lock(sk2);
1053 unix_state_lock_nested(sk1);
1057 static void unix_state_double_unlock(struct sock *sk1, struct sock *sk2)
1059 if (unlikely(sk1 == sk2) || !sk2) {
1060 unix_state_unlock(sk1);
1063 unix_state_unlock(sk1);
1064 unix_state_unlock(sk2);
1067 static int unix_dgram_connect(struct socket *sock, struct sockaddr *addr,
1068 int alen, int flags)
1070 struct sock *sk = sock->sk;
1071 struct net *net = sock_net(sk);
1072 struct sockaddr_un *sunaddr = (struct sockaddr_un *)addr;
1077 if (addr->sa_family != AF_UNSPEC) {
1078 err = unix_mkname(sunaddr, alen, &hash);
1083 if (test_bit(SOCK_PASSCRED, &sock->flags) &&
1084 !unix_sk(sk)->addr && (err = unix_autobind(sock)) != 0)
1088 other = unix_find_other(net, sunaddr, alen, sock->type, hash, &err);
1092 unix_state_double_lock(sk, other);
1094 /* Apparently VFS overslept socket death. Retry. */
1095 if (sock_flag(other, SOCK_DEAD)) {
1096 unix_state_double_unlock(sk, other);
1102 if (!unix_may_send(sk, other))
1105 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1111 * 1003.1g breaking connected state with AF_UNSPEC
1114 unix_state_double_lock(sk, other);
1118 * If it was connected, reconnect.
1120 if (unix_peer(sk)) {
1121 struct sock *old_peer = unix_peer(sk);
1122 unix_peer(sk) = other;
1123 unix_dgram_peer_wake_disconnect_wakeup(sk, old_peer);
1125 unix_state_double_unlock(sk, other);
1127 if (other != old_peer)
1128 unix_dgram_disconnected(sk, old_peer);
1131 unix_peer(sk) = other;
1132 unix_state_double_unlock(sk, other);
1137 unix_state_double_unlock(sk, other);
1143 static long unix_wait_for_peer(struct sock *other, long timeo)
1145 struct unix_sock *u = unix_sk(other);
1149 prepare_to_wait_exclusive(&u->peer_wait, &wait, TASK_INTERRUPTIBLE);
1151 sched = !sock_flag(other, SOCK_DEAD) &&
1152 !(other->sk_shutdown & RCV_SHUTDOWN) &&
1153 unix_recvq_full(other);
1155 unix_state_unlock(other);
1158 timeo = schedule_timeout(timeo);
1160 finish_wait(&u->peer_wait, &wait);
1164 static int unix_stream_connect(struct socket *sock, struct sockaddr *uaddr,
1165 int addr_len, int flags)
1167 struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
1168 struct sock *sk = sock->sk;
1169 struct net *net = sock_net(sk);
1170 struct unix_sock *u = unix_sk(sk), *newu, *otheru;
1171 struct sock *newsk = NULL;
1172 struct sock *other = NULL;
1173 struct sk_buff *skb = NULL;
1179 err = unix_mkname(sunaddr, addr_len, &hash);
1184 if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr &&
1185 (err = unix_autobind(sock)) != 0)
1188 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
1190 /* First of all allocate resources.
1191 If we will make it after state is locked,
1192 we will have to recheck all again in any case.
1197 /* create new sock for complete connection */
1198 newsk = unix_create1(sock_net(sk), NULL);
1202 /* Allocate skb for sending to listening sock */
1203 skb = sock_wmalloc(newsk, 1, 0, GFP_KERNEL);
1208 /* Find listening sock. */
1209 other = unix_find_other(net, sunaddr, addr_len, sk->sk_type, hash, &err);
1213 /* Latch state of peer */
1214 unix_state_lock(other);
1216 /* Apparently VFS overslept socket death. Retry. */
1217 if (sock_flag(other, SOCK_DEAD)) {
1218 unix_state_unlock(other);
1223 err = -ECONNREFUSED;
1224 if (other->sk_state != TCP_LISTEN)
1226 if (other->sk_shutdown & RCV_SHUTDOWN)
1229 if (unix_recvq_full(other)) {
1234 timeo = unix_wait_for_peer(other, timeo);
1236 err = sock_intr_errno(timeo);
1237 if (signal_pending(current))
1245 It is tricky place. We need to grab our state lock and cannot
1246 drop lock on peer. It is dangerous because deadlock is
1247 possible. Connect to self case and simultaneous
1248 attempt to connect are eliminated by checking socket
1249 state. other is TCP_LISTEN, if sk is TCP_LISTEN we
1250 check this before attempt to grab lock.
1252 Well, and we have to recheck the state after socket locked.
1258 /* This is ok... continue with connect */
1260 case TCP_ESTABLISHED:
1261 /* Socket is already connected */
1269 unix_state_lock_nested(sk);
1271 if (sk->sk_state != st) {
1272 unix_state_unlock(sk);
1273 unix_state_unlock(other);
1278 err = security_unix_stream_connect(sk, other, newsk);
1280 unix_state_unlock(sk);
1284 /* The way is open! Fastly set all the necessary fields... */
1287 unix_peer(newsk) = sk;
1288 newsk->sk_state = TCP_ESTABLISHED;
1289 newsk->sk_type = sk->sk_type;
1290 init_peercred(newsk);
1291 newu = unix_sk(newsk);
1292 RCU_INIT_POINTER(newsk->sk_wq, &newu->peer_wq);
1293 otheru = unix_sk(other);
1295 /* copy address information from listening to new sock*/
1297 atomic_inc(&otheru->addr->refcnt);
1298 newu->addr = otheru->addr;
1300 if (otheru->dentry) {
1301 newu->dentry = dget(otheru->dentry);
1302 newu->mnt = mntget(otheru->mnt);
1305 /* Set credentials */
1306 copy_peercred(sk, other);
1308 sock->state = SS_CONNECTED;
1309 sk->sk_state = TCP_ESTABLISHED;
1312 smp_mb__after_atomic_inc(); /* sock_hold() does an atomic_inc() */
1313 unix_peer(sk) = newsk;
1315 unix_state_unlock(sk);
1317 /* take ten and and send info to listening sock */
1318 spin_lock(&other->sk_receive_queue.lock);
1319 __skb_queue_tail(&other->sk_receive_queue, skb);
1320 spin_unlock(&other->sk_receive_queue.lock);
1321 unix_state_unlock(other);
1322 other->sk_data_ready(other, 0);
1328 unix_state_unlock(other);
1333 unix_release_sock(newsk, 0);
1339 static int unix_socketpair(struct socket *socka, struct socket *sockb)
1341 struct sock *ska = socka->sk, *skb = sockb->sk;
1343 /* Join our sockets back to back */
1346 unix_peer(ska) = skb;
1347 unix_peer(skb) = ska;
1351 if (ska->sk_type != SOCK_DGRAM) {
1352 ska->sk_state = TCP_ESTABLISHED;
1353 skb->sk_state = TCP_ESTABLISHED;
1354 socka->state = SS_CONNECTED;
1355 sockb->state = SS_CONNECTED;
1360 static void unix_sock_inherit_flags(const struct socket *old,
1363 if (test_bit(SOCK_PASSCRED, &old->flags))
1364 set_bit(SOCK_PASSCRED, &new->flags);
1365 if (test_bit(SOCK_PASSSEC, &old->flags))
1366 set_bit(SOCK_PASSSEC, &new->flags);
1369 static int unix_accept(struct socket *sock, struct socket *newsock, int flags)
1371 struct sock *sk = sock->sk;
1373 struct sk_buff *skb;
1377 if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
1381 if (sk->sk_state != TCP_LISTEN)
1384 /* If socket state is TCP_LISTEN it cannot change (for now...),
1385 * so that no locks are necessary.
1388 skb = skb_recv_datagram(sk, 0, flags&O_NONBLOCK, &err);
1390 /* This means receive shutdown. */
1397 skb_free_datagram(sk, skb);
1398 wake_up_interruptible(&unix_sk(sk)->peer_wait);
1400 /* attach accepted sock to socket */
1401 unix_state_lock(tsk);
1402 newsock->state = SS_CONNECTED;
1403 unix_sock_inherit_flags(sock, newsock);
1404 sock_graft(tsk, newsock);
1405 unix_state_unlock(tsk);
1413 static int unix_getname(struct socket *sock, struct sockaddr *uaddr, int *uaddr_len, int peer)
1415 struct sock *sk = sock->sk;
1416 struct unix_sock *u;
1417 DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr, uaddr);
1421 sk = unix_peer_get(sk);
1432 unix_state_lock(sk);
1434 sunaddr->sun_family = AF_UNIX;
1435 sunaddr->sun_path[0] = 0;
1436 *uaddr_len = sizeof(short);
1438 struct unix_address *addr = u->addr;
1440 *uaddr_len = addr->len;
1441 memcpy(sunaddr, addr->name, *uaddr_len);
1443 unix_state_unlock(sk);
1449 static void unix_detach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1453 scm->fp = UNIXCB(skb).fp;
1454 UNIXCB(skb).fp = NULL;
1456 for (i = scm->fp->count-1; i >= 0; i--)
1457 unix_notinflight(scm->fp->user, scm->fp->fp[i]);
1460 static void unix_destruct_scm(struct sk_buff *skb)
1462 struct scm_cookie scm;
1463 memset(&scm, 0, sizeof(scm));
1464 scm.pid = UNIXCB(skb).pid;
1465 scm.cred = UNIXCB(skb).cred;
1467 unix_detach_fds(&scm, skb);
1469 /* Alas, it calls VFS */
1470 /* So fscking what? fput() had been SMP-safe since the last Summer */
1476 * The "user->unix_inflight" variable is protected by the garbage
1477 * collection lock, and we just read it locklessly here. If you go
1478 * over the limit, there might be a tiny race in actually noticing
1479 * it across threads. Tough.
1481 static inline bool too_many_unix_fds(struct task_struct *p)
1483 struct user_struct *user = current_user();
1485 if (unlikely(user->unix_inflight > task_rlimit(p, RLIMIT_NOFILE)))
1486 return !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN);
1490 #define MAX_RECURSION_LEVEL 4
1492 static int unix_attach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1495 unsigned char max_level = 0;
1496 int unix_sock_count = 0;
1498 if (too_many_unix_fds(current))
1499 return -ETOOMANYREFS;
1501 for (i = scm->fp->count - 1; i >= 0; i--) {
1502 struct sock *sk = unix_get_socket(scm->fp->fp[i]);
1506 max_level = max(max_level,
1507 unix_sk(sk)->recursion_level);
1510 if (unlikely(max_level > MAX_RECURSION_LEVEL))
1511 return -ETOOMANYREFS;
1514 * Need to duplicate file references for the sake of garbage
1515 * collection. Otherwise a socket in the fps might become a
1516 * candidate for GC while the skb is not yet queued.
1518 UNIXCB(skb).fp = scm_fp_dup(scm->fp);
1519 if (!UNIXCB(skb).fp)
1522 for (i = scm->fp->count - 1; i >= 0; i--)
1523 unix_inflight(scm->fp->user, scm->fp->fp[i]);
1527 static int unix_scm_to_skb(struct scm_cookie *scm, struct sk_buff *skb, bool send_fds)
1531 UNIXCB(skb).pid = get_pid(scm->pid);
1533 UNIXCB(skb).cred = get_cred(scm->cred);
1534 UNIXCB(skb).fp = NULL;
1535 if (scm->fp && send_fds)
1536 err = unix_attach_fds(scm, skb);
1538 skb->destructor = unix_destruct_scm;
1543 * Some apps rely on write() giving SCM_CREDENTIALS
1544 * We include credentials if source or destination socket
1545 * asserted SOCK_PASSCRED.
1547 static void maybe_add_creds(struct sk_buff *skb, const struct socket *sock,
1548 const struct sock *other)
1550 if (UNIXCB(skb).cred)
1552 if (test_bit(SOCK_PASSCRED, &sock->flags) ||
1553 !other->sk_socket ||
1554 test_bit(SOCK_PASSCRED, &other->sk_socket->flags)) {
1555 UNIXCB(skb).pid = get_pid(task_tgid(current));
1556 UNIXCB(skb).cred = get_current_cred();
1561 * Send AF_UNIX data.
1564 static int unix_dgram_sendmsg(struct kiocb *kiocb, struct socket *sock,
1565 struct msghdr *msg, size_t len)
1567 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1568 struct sock *sk = sock->sk;
1569 struct net *net = sock_net(sk);
1570 struct unix_sock *u = unix_sk(sk);
1571 struct sockaddr_un *sunaddr = msg->msg_name;
1572 struct sock *other = NULL;
1573 int namelen = 0; /* fake GCC */
1576 struct sk_buff *skb;
1578 struct scm_cookie tmp_scm;
1582 if (NULL == siocb->scm)
1583 siocb->scm = &tmp_scm;
1585 err = scm_send(sock, msg, siocb->scm, false);
1590 if (msg->msg_flags&MSG_OOB)
1593 if (msg->msg_namelen) {
1594 err = unix_mkname(sunaddr, msg->msg_namelen, &hash);
1601 other = unix_peer_get(sk);
1606 if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr
1607 && (err = unix_autobind(sock)) != 0)
1611 if (len > sk->sk_sndbuf - 32)
1614 skb = sock_alloc_send_skb(sk, len, msg->msg_flags&MSG_DONTWAIT, &err);
1618 err = unix_scm_to_skb(siocb->scm, skb, true);
1621 max_level = err + 1;
1622 unix_get_secdata(siocb->scm, skb);
1624 skb_reset_transport_header(skb);
1625 err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
1629 timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1634 if (sunaddr == NULL)
1637 other = unix_find_other(net, sunaddr, namelen, sk->sk_type,
1643 if (sk_filter(other, skb) < 0) {
1644 /* Toss the packet but do not return any error to the sender */
1650 unix_state_lock(other);
1653 if (!unix_may_send(sk, other))
1656 if (unlikely(sock_flag(other, SOCK_DEAD))) {
1658 * Check with 1003.1g - what should
1661 unix_state_unlock(other);
1665 unix_state_lock(sk);
1668 if (unix_peer(sk) == other) {
1669 unix_peer(sk) = NULL;
1670 unix_dgram_peer_wake_disconnect_wakeup(sk, other);
1672 unix_state_unlock(sk);
1674 unix_dgram_disconnected(sk, other);
1676 err = -ECONNREFUSED;
1678 unix_state_unlock(sk);
1688 if (other->sk_shutdown & RCV_SHUTDOWN)
1691 if (sk->sk_type != SOCK_SEQPACKET) {
1692 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1697 /* other == sk && unix_peer(other) != sk if
1698 * - unix_peer(sk) == NULL, destination address bound to sk
1699 * - unix_peer(sk) == sk by time of get but disconnected before lock
1702 unlikely(unix_peer(other) != sk && unix_recvq_full(other))) {
1704 timeo = unix_wait_for_peer(other, timeo);
1706 err = sock_intr_errno(timeo);
1707 if (signal_pending(current))
1714 unix_state_unlock(other);
1715 unix_state_double_lock(sk, other);
1718 if (unix_peer(sk) != other ||
1719 unix_dgram_peer_wake_me(sk, other)) {
1727 goto restart_locked;
1731 if (unlikely(sk_locked))
1732 unix_state_unlock(sk);
1734 if (sock_flag(other, SOCK_RCVTSTAMP))
1735 __net_timestamp(skb);
1736 maybe_add_creds(skb, sock, other);
1737 skb_queue_tail(&other->sk_receive_queue, skb);
1738 if (max_level > unix_sk(other)->recursion_level)
1739 unix_sk(other)->recursion_level = max_level;
1740 unix_state_unlock(other);
1741 other->sk_data_ready(other, len);
1743 scm_destroy(siocb->scm);
1748 unix_state_unlock(sk);
1749 unix_state_unlock(other);
1755 scm_destroy(siocb->scm);
1760 static int unix_stream_sendmsg(struct kiocb *kiocb, struct socket *sock,
1761 struct msghdr *msg, size_t len)
1763 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1764 struct sock *sk = sock->sk;
1765 struct sock *other = NULL;
1767 struct sk_buff *skb;
1769 struct scm_cookie tmp_scm;
1770 bool fds_sent = false;
1773 if (NULL == siocb->scm)
1774 siocb->scm = &tmp_scm;
1776 err = scm_send(sock, msg, siocb->scm, false);
1781 if (msg->msg_flags&MSG_OOB)
1784 if (msg->msg_namelen) {
1785 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
1789 other = unix_peer(sk);
1794 if (sk->sk_shutdown & SEND_SHUTDOWN)
1797 while (sent < len) {
1799 * Optimisation for the fact that under 0.01% of X
1800 * messages typically need breaking up.
1805 /* Keep two messages in the pipe so it schedules better */
1806 if (size > ((sk->sk_sndbuf >> 1) - 64))
1807 size = (sk->sk_sndbuf >> 1) - 64;
1809 if (size > SKB_MAX_ALLOC)
1810 size = SKB_MAX_ALLOC;
1816 skb = sock_alloc_send_skb(sk, size, msg->msg_flags&MSG_DONTWAIT,
1823 * If you pass two values to the sock_alloc_send_skb
1824 * it tries to grab the large buffer with GFP_NOFS
1825 * (which can fail easily), and if it fails grab the
1826 * fallback size buffer which is under a page and will
1829 size = min_t(int, size, skb_tailroom(skb));
1832 /* Only send the fds in the first buffer */
1833 err = unix_scm_to_skb(siocb->scm, skb, !fds_sent);
1838 max_level = err + 1;
1841 err = memcpy_fromiovec(skb_put(skb, size), msg->msg_iov, size);
1847 unix_state_lock(other);
1849 if (sock_flag(other, SOCK_DEAD) ||
1850 (other->sk_shutdown & RCV_SHUTDOWN))
1853 maybe_add_creds(skb, sock, other);
1854 skb_queue_tail(&other->sk_receive_queue, skb);
1855 if (max_level > unix_sk(other)->recursion_level)
1856 unix_sk(other)->recursion_level = max_level;
1857 unix_state_unlock(other);
1858 other->sk_data_ready(other, size);
1862 scm_destroy(siocb->scm);
1868 unix_state_unlock(other);
1871 if (sent == 0 && !(msg->msg_flags&MSG_NOSIGNAL))
1872 send_sig(SIGPIPE, current, 0);
1875 scm_destroy(siocb->scm);
1877 return sent ? : err;
1880 static int unix_seqpacket_sendmsg(struct kiocb *kiocb, struct socket *sock,
1881 struct msghdr *msg, size_t len)
1884 struct sock *sk = sock->sk;
1886 err = sock_error(sk);
1890 if (sk->sk_state != TCP_ESTABLISHED)
1893 if (msg->msg_namelen)
1894 msg->msg_namelen = 0;
1896 return unix_dgram_sendmsg(kiocb, sock, msg, len);
1899 static int unix_seqpacket_recvmsg(struct kiocb *iocb, struct socket *sock,
1900 struct msghdr *msg, size_t size,
1903 struct sock *sk = sock->sk;
1905 if (sk->sk_state != TCP_ESTABLISHED)
1908 return unix_dgram_recvmsg(iocb, sock, msg, size, flags);
1911 static void unix_copy_addr(struct msghdr *msg, struct sock *sk)
1913 struct unix_sock *u = unix_sk(sk);
1916 msg->msg_namelen = u->addr->len;
1917 memcpy(msg->msg_name, u->addr->name, u->addr->len);
1921 static int unix_dgram_recvmsg(struct kiocb *iocb, struct socket *sock,
1922 struct msghdr *msg, size_t size,
1925 struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1926 struct scm_cookie tmp_scm;
1927 struct sock *sk = sock->sk;
1928 struct unix_sock *u = unix_sk(sk);
1929 int noblock = flags & MSG_DONTWAIT;
1930 struct sk_buff *skb;
1937 err = mutex_lock_interruptible(&u->readlock);
1938 if (unlikely(err)) {
1939 /* recvmsg() in non blocking mode is supposed to return -EAGAIN
1940 * sk_rcvtimeo is not honored by mutex_lock_interruptible()
1942 err = noblock ? -EAGAIN : -ERESTARTSYS;
1946 skb = skb_recv_datagram(sk, flags, noblock, &err);
1948 unix_state_lock(sk);
1949 /* Signal EOF on disconnected non-blocking SEQPACKET socket. */
1950 if (sk->sk_type == SOCK_SEQPACKET && err == -EAGAIN &&
1951 (sk->sk_shutdown & RCV_SHUTDOWN))
1953 unix_state_unlock(sk);
1957 wake_up_interruptible_sync_poll(&u->peer_wait,
1958 POLLOUT | POLLWRNORM | POLLWRBAND);
1961 unix_copy_addr(msg, skb->sk);
1963 if (size > skb->len)
1965 else if (size < skb->len)
1966 msg->msg_flags |= MSG_TRUNC;
1968 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, size);
1972 if (sock_flag(sk, SOCK_RCVTSTAMP))
1973 __sock_recv_timestamp(msg, sk, skb);
1976 siocb->scm = &tmp_scm;
1977 memset(&tmp_scm, 0, sizeof(tmp_scm));
1979 scm_set_cred(siocb->scm, UNIXCB(skb).pid, UNIXCB(skb).cred);
1980 unix_set_secdata(siocb->scm, skb);
1982 if (!(flags & MSG_PEEK)) {
1984 unix_detach_fds(siocb->scm, skb);
1986 /* It is questionable: on PEEK we could:
1987 - do not return fds - good, but too simple 8)
1988 - return fds, and do not return them on read (old strategy,
1990 - clone fds (I chose it for now, it is the most universal
1993 POSIX 1003.1g does not actually define this clearly
1994 at all. POSIX 1003.1g doesn't define a lot of things
1999 siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
2003 scm_recv(sock, msg, siocb->scm, flags);
2006 skb_free_datagram(sk, skb);
2008 mutex_unlock(&u->readlock);
2014 * Sleep until data has arrive. But check for races..
2017 static long unix_stream_data_wait(struct sock *sk, long timeo)
2021 unix_state_lock(sk);
2024 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2026 if (!skb_queue_empty(&sk->sk_receive_queue) ||
2028 (sk->sk_shutdown & RCV_SHUTDOWN) ||
2029 signal_pending(current) ||
2033 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
2034 unix_state_unlock(sk);
2035 timeo = schedule_timeout(timeo);
2036 unix_state_lock(sk);
2038 if (sock_flag(sk, SOCK_DEAD))
2041 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
2044 finish_wait(sk_sleep(sk), &wait);
2045 unix_state_unlock(sk);
2051 static int unix_stream_recvmsg(struct kiocb *iocb, struct socket *sock,
2052 struct msghdr *msg, size_t size,
2055 struct sock_iocb *siocb = kiocb_to_siocb(iocb);
2056 struct scm_cookie tmp_scm;
2057 struct sock *sk = sock->sk;
2058 struct unix_sock *u = unix_sk(sk);
2059 struct sockaddr_un *sunaddr = msg->msg_name;
2061 int noblock = flags & MSG_DONTWAIT;
2062 int check_creds = 0;
2067 if (unlikely(sk->sk_state != TCP_ESTABLISHED)) {
2072 if (unlikely(flags & MSG_OOB)) {
2077 target = sock_rcvlowat(sk, flags&MSG_WAITALL, size);
2078 timeo = sock_rcvtimeo(sk, noblock);
2080 /* Lock the socket to prevent queue disordering
2081 * while sleeps in memcpy_tomsg
2085 siocb->scm = &tmp_scm;
2086 memset(&tmp_scm, 0, sizeof(tmp_scm));
2089 mutex_lock(&u->readlock);
2093 struct sk_buff *skb;
2095 unix_state_lock(sk);
2096 if (sock_flag(sk, SOCK_DEAD)) {
2100 skb = skb_peek(&sk->sk_receive_queue);
2102 unix_sk(sk)->recursion_level = 0;
2103 if (copied >= target)
2107 * POSIX 1003.1g mandates this order.
2110 err = sock_error(sk);
2113 if (sk->sk_shutdown & RCV_SHUTDOWN)
2116 unix_state_unlock(sk);
2122 mutex_unlock(&u->readlock);
2124 timeo = unix_stream_data_wait(sk, timeo);
2126 if (signal_pending(current)) {
2127 err = sock_intr_errno(timeo);
2128 scm_destroy(siocb->scm);
2132 mutex_lock(&u->readlock);
2135 unix_state_unlock(sk);
2138 unix_state_unlock(sk);
2141 /* Never glue messages from different writers */
2142 if ((UNIXCB(skb).pid != siocb->scm->pid) ||
2143 (UNIXCB(skb).cred != siocb->scm->cred))
2145 } else if (test_bit(SOCK_PASSCRED, &sock->flags)) {
2146 /* Copy credentials */
2147 scm_set_cred(siocb->scm, UNIXCB(skb).pid, UNIXCB(skb).cred);
2151 /* Copy address just once */
2153 unix_copy_addr(msg, skb->sk);
2157 chunk = min_t(unsigned int, skb->len, size);
2158 if (memcpy_toiovec(msg->msg_iov, skb->data, chunk)) {
2166 /* Mark read part of skb as used */
2167 if (!(flags & MSG_PEEK)) {
2168 skb_pull(skb, chunk);
2171 unix_detach_fds(siocb->scm, skb);
2176 skb_unlink(skb, &sk->sk_receive_queue);
2182 /* It is questionable, see note in unix_dgram_recvmsg.
2185 siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
2191 mutex_unlock(&u->readlock);
2192 scm_recv(sock, msg, siocb->scm, flags);
2194 return copied ? : err;
2197 static int unix_shutdown(struct socket *sock, int mode)
2199 struct sock *sk = sock->sk;
2202 mode = (mode+1)&(RCV_SHUTDOWN|SEND_SHUTDOWN);
2207 unix_state_lock(sk);
2208 sk->sk_shutdown |= mode;
2209 other = unix_peer(sk);
2212 unix_state_unlock(sk);
2213 sk->sk_state_change(sk);
2216 (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET)) {
2220 if (mode&RCV_SHUTDOWN)
2221 peer_mode |= SEND_SHUTDOWN;
2222 if (mode&SEND_SHUTDOWN)
2223 peer_mode |= RCV_SHUTDOWN;
2224 unix_state_lock(other);
2225 other->sk_shutdown |= peer_mode;
2226 unix_state_unlock(other);
2227 other->sk_state_change(other);
2228 if (peer_mode == SHUTDOWN_MASK)
2229 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_HUP);
2230 else if (peer_mode & RCV_SHUTDOWN)
2231 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_IN);
2239 static int unix_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2241 struct sock *sk = sock->sk;
2247 amount = sk_wmem_alloc_get(sk);
2248 err = put_user(amount, (int __user *)arg);
2252 struct sk_buff *skb;
2254 if (sk->sk_state == TCP_LISTEN) {
2259 spin_lock(&sk->sk_receive_queue.lock);
2260 if (sk->sk_type == SOCK_STREAM ||
2261 sk->sk_type == SOCK_SEQPACKET) {
2262 skb_queue_walk(&sk->sk_receive_queue, skb)
2265 skb = skb_peek(&sk->sk_receive_queue);
2269 spin_unlock(&sk->sk_receive_queue.lock);
2270 err = put_user(amount, (int __user *)arg);
2281 static unsigned int unix_poll(struct file *file, struct socket *sock, poll_table *wait)
2283 struct sock *sk = sock->sk;
2286 sock_poll_wait(file, sk_sleep(sk), wait);
2289 /* exceptional events? */
2292 if (sk->sk_shutdown == SHUTDOWN_MASK)
2294 if (sk->sk_shutdown & RCV_SHUTDOWN)
2295 mask |= POLLRDHUP | POLLIN | POLLRDNORM;
2298 if (!skb_queue_empty(&sk->sk_receive_queue))
2299 mask |= POLLIN | POLLRDNORM;
2301 /* Connection-based need to check for termination and startup */
2302 if ((sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) &&
2303 sk->sk_state == TCP_CLOSE)
2307 * we set writable also when the other side has shut down the
2308 * connection. This prevents stuck sockets.
2310 if (unix_writable(sk))
2311 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2316 static unsigned int unix_dgram_poll(struct file *file, struct socket *sock,
2319 struct sock *sk = sock->sk, *other;
2320 unsigned int mask, writable;
2322 sock_poll_wait(file, sk_sleep(sk), wait);
2325 /* exceptional events? */
2326 if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue))
2328 if (sk->sk_shutdown & RCV_SHUTDOWN)
2329 mask |= POLLRDHUP | POLLIN | POLLRDNORM;
2330 if (sk->sk_shutdown == SHUTDOWN_MASK)
2334 if (!skb_queue_empty(&sk->sk_receive_queue))
2335 mask |= POLLIN | POLLRDNORM;
2337 /* Connection-based need to check for termination and startup */
2338 if (sk->sk_type == SOCK_SEQPACKET) {
2339 if (sk->sk_state == TCP_CLOSE)
2341 /* connection hasn't started yet? */
2342 if (sk->sk_state == TCP_SYN_SENT)
2346 /* No write status requested, avoid expensive OUT tests. */
2347 if (wait && !(wait->key & (POLLWRBAND | POLLWRNORM | POLLOUT)))
2350 writable = unix_writable(sk);
2352 unix_state_lock(sk);
2354 other = unix_peer(sk);
2355 if (other && unix_peer(other) != sk &&
2356 unix_recvq_full(other) &&
2357 unix_dgram_peer_wake_me(sk, other))
2360 unix_state_unlock(sk);
2364 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2366 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
2371 #ifdef CONFIG_PROC_FS
2372 static struct sock *first_unix_socket(int *i)
2374 for (*i = 0; *i <= UNIX_HASH_SIZE; (*i)++) {
2375 if (!hlist_empty(&unix_socket_table[*i]))
2376 return __sk_head(&unix_socket_table[*i]);
2381 static struct sock *next_unix_socket(int *i, struct sock *s)
2383 struct sock *next = sk_next(s);
2384 /* More in this chain? */
2387 /* Look for next non-empty chain. */
2388 for ((*i)++; *i <= UNIX_HASH_SIZE; (*i)++) {
2389 if (!hlist_empty(&unix_socket_table[*i]))
2390 return __sk_head(&unix_socket_table[*i]);
2395 struct unix_iter_state {
2396 struct seq_net_private p;
2400 static struct sock *unix_seq_idx(struct seq_file *seq, loff_t pos)
2402 struct unix_iter_state *iter = seq->private;
2406 for (s = first_unix_socket(&iter->i); s; s = next_unix_socket(&iter->i, s)) {
2407 if (sock_net(s) != seq_file_net(seq))
2416 static void *unix_seq_start(struct seq_file *seq, loff_t *pos)
2417 __acquires(unix_table_lock)
2419 spin_lock(&unix_table_lock);
2420 return *pos ? unix_seq_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2423 static void *unix_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2425 struct unix_iter_state *iter = seq->private;
2426 struct sock *sk = v;
2429 if (v == SEQ_START_TOKEN)
2430 sk = first_unix_socket(&iter->i);
2432 sk = next_unix_socket(&iter->i, sk);
2433 while (sk && (sock_net(sk) != seq_file_net(seq)))
2434 sk = next_unix_socket(&iter->i, sk);
2438 static void unix_seq_stop(struct seq_file *seq, void *v)
2439 __releases(unix_table_lock)
2441 spin_unlock(&unix_table_lock);
2444 static int unix_seq_show(struct seq_file *seq, void *v)
2447 if (v == SEQ_START_TOKEN)
2448 seq_puts(seq, "Num RefCount Protocol Flags Type St "
2452 struct unix_sock *u = unix_sk(s);
2455 seq_printf(seq, "%pK: %08X %08X %08X %04X %02X %5lu",
2457 atomic_read(&s->sk_refcnt),
2459 s->sk_state == TCP_LISTEN ? __SO_ACCEPTCON : 0,
2462 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTED : SS_UNCONNECTED) :
2463 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTING : SS_DISCONNECTING),
2471 len = u->addr->len - sizeof(short);
2472 if (!UNIX_ABSTRACT(s))
2478 for ( ; i < len; i++)
2479 seq_putc(seq, u->addr->name->sun_path[i]);
2481 unix_state_unlock(s);
2482 seq_putc(seq, '\n');
2488 static const struct seq_operations unix_seq_ops = {
2489 .start = unix_seq_start,
2490 .next = unix_seq_next,
2491 .stop = unix_seq_stop,
2492 .show = unix_seq_show,
2495 static int unix_seq_open(struct inode *inode, struct file *file)
2497 return seq_open_net(inode, file, &unix_seq_ops,
2498 sizeof(struct unix_iter_state));
2501 static const struct file_operations unix_seq_fops = {
2502 .owner = THIS_MODULE,
2503 .open = unix_seq_open,
2505 .llseek = seq_lseek,
2506 .release = seq_release_net,
2511 static const struct net_proto_family unix_family_ops = {
2513 .create = unix_create,
2514 .owner = THIS_MODULE,
2518 static int __net_init unix_net_init(struct net *net)
2520 int error = -ENOMEM;
2522 net->unx.sysctl_max_dgram_qlen = 10;
2523 if (unix_sysctl_register(net))
2526 #ifdef CONFIG_PROC_FS
2527 if (!proc_net_fops_create(net, "unix", 0, &unix_seq_fops)) {
2528 unix_sysctl_unregister(net);
2537 static void __net_exit unix_net_exit(struct net *net)
2539 unix_sysctl_unregister(net);
2540 proc_net_remove(net, "unix");
2543 static struct pernet_operations unix_net_ops = {
2544 .init = unix_net_init,
2545 .exit = unix_net_exit,
2548 static int __init af_unix_init(void)
2551 struct sk_buff *dummy_skb;
2553 BUILD_BUG_ON(sizeof(struct unix_skb_parms) > sizeof(dummy_skb->cb));
2555 rc = proto_register(&unix_proto, 1);
2557 printk(KERN_CRIT "%s: Cannot create unix_sock SLAB cache!\n",
2562 sock_register(&unix_family_ops);
2563 register_pernet_subsys(&unix_net_ops);
2568 static void __exit af_unix_exit(void)
2570 sock_unregister(PF_UNIX);
2571 proto_unregister(&unix_proto);
2572 unregister_pernet_subsys(&unix_net_ops);
2575 /* Earlier than device_initcall() so that other drivers invoking
2576 request_module() don't end up in a loop when modprobe tries
2577 to use a UNIX socket. But later than subsys_initcall() because
2578 we depend on stuff initialised there */
2579 fs_initcall(af_unix_init);
2580 module_exit(af_unix_exit);
2582 MODULE_LICENSE("GPL");
2583 MODULE_ALIAS_NETPROTO(PF_UNIX);