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 struct hlist_head unix_socket_table[UNIX_HASH_SIZE + 1];
119 EXPORT_SYMBOL_GPL(unix_socket_table);
120 DEFINE_SPINLOCK(unix_table_lock);
121 EXPORT_SYMBOL_GPL(unix_table_lock);
122 static atomic_long_t unix_nr_socks;
124 #define unix_sockets_unbound (&unix_socket_table[UNIX_HASH_SIZE])
126 #define UNIX_ABSTRACT(sk) (unix_sk(sk)->addr->hash != UNIX_HASH_SIZE)
128 #ifdef CONFIG_SECURITY_NETWORK
129 static void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb)
131 memcpy(UNIXSID(skb), &scm->secid, sizeof(u32));
134 static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb)
136 scm->secid = *UNIXSID(skb);
139 static inline void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb)
142 static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb)
144 #endif /* CONFIG_SECURITY_NETWORK */
147 * SMP locking strategy:
148 * hash table is protected with spinlock unix_table_lock
149 * each socket state is protected by separate spin lock.
152 static inline unsigned int unix_hash_fold(__wsum n)
154 unsigned int hash = (__force unsigned int)n;
158 return hash&(UNIX_HASH_SIZE-1);
161 #define unix_peer(sk) (unix_sk(sk)->peer)
163 static inline int unix_our_peer(struct sock *sk, struct sock *osk)
165 return unix_peer(osk) == sk;
168 static inline int unix_may_send(struct sock *sk, struct sock *osk)
170 return unix_peer(osk) == NULL || unix_our_peer(sk, osk);
173 static inline int unix_recvq_full(struct sock const *sk)
175 return skb_queue_len(&sk->sk_receive_queue) > sk->sk_max_ack_backlog;
178 struct sock *unix_peer_get(struct sock *s)
186 unix_state_unlock(s);
189 EXPORT_SYMBOL_GPL(unix_peer_get);
191 static inline void unix_release_addr(struct unix_address *addr)
193 if (atomic_dec_and_test(&addr->refcnt))
198 * Check unix socket name:
199 * - should be not zero length.
200 * - if started by not zero, should be NULL terminated (FS object)
201 * - if started by zero, it is abstract name.
204 static int unix_mkname(struct sockaddr_un *sunaddr, int len, unsigned int *hashp)
206 if (len <= sizeof(short) || len > sizeof(*sunaddr))
208 if (!sunaddr || sunaddr->sun_family != AF_UNIX)
210 if (sunaddr->sun_path[0]) {
212 * This may look like an off by one error but it is a bit more
213 * subtle. 108 is the longest valid AF_UNIX path for a binding.
214 * sun_path[108] doesn't as such exist. However in kernel space
215 * we are guaranteed that it is a valid memory location in our
216 * kernel address buffer.
218 ((char *)sunaddr)[len] = 0;
219 len = strlen(sunaddr->sun_path)+1+sizeof(short);
223 *hashp = unix_hash_fold(csum_partial(sunaddr, len, 0));
227 static void __unix_remove_socket(struct sock *sk)
229 sk_del_node_init(sk);
232 static void __unix_insert_socket(struct hlist_head *list, struct sock *sk)
234 WARN_ON(!sk_unhashed(sk));
235 sk_add_node(sk, list);
238 static inline void unix_remove_socket(struct sock *sk)
240 spin_lock(&unix_table_lock);
241 __unix_remove_socket(sk);
242 spin_unlock(&unix_table_lock);
245 static inline void unix_insert_socket(struct hlist_head *list, struct sock *sk)
247 spin_lock(&unix_table_lock);
248 __unix_insert_socket(list, sk);
249 spin_unlock(&unix_table_lock);
252 static struct sock *__unix_find_socket_byname(struct net *net,
253 struct sockaddr_un *sunname,
254 int len, int type, unsigned int hash)
257 struct hlist_node *node;
259 sk_for_each(s, node, &unix_socket_table[hash ^ type]) {
260 struct unix_sock *u = unix_sk(s);
262 if (!net_eq(sock_net(s), net))
265 if (u->addr->len == len &&
266 !memcmp(u->addr->name, sunname, len))
274 static inline struct sock *unix_find_socket_byname(struct net *net,
275 struct sockaddr_un *sunname,
281 spin_lock(&unix_table_lock);
282 s = __unix_find_socket_byname(net, sunname, len, type, hash);
285 spin_unlock(&unix_table_lock);
289 static struct sock *unix_find_socket_byinode(struct inode *i)
292 struct hlist_node *node;
294 spin_lock(&unix_table_lock);
296 &unix_socket_table[i->i_ino & (UNIX_HASH_SIZE - 1)]) {
297 struct dentry *dentry = unix_sk(s)->path.dentry;
299 if (dentry && dentry->d_inode == i) {
306 spin_unlock(&unix_table_lock);
310 static inline int unix_writable(struct sock *sk)
312 return (atomic_read(&sk->sk_wmem_alloc) << 2) <= sk->sk_sndbuf;
315 static void unix_write_space(struct sock *sk)
317 struct socket_wq *wq;
320 if (unix_writable(sk)) {
321 wq = rcu_dereference(sk->sk_wq);
322 if (wq_has_sleeper(wq))
323 wake_up_interruptible_sync_poll(&wq->wait,
324 POLLOUT | POLLWRNORM | POLLWRBAND);
325 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
330 /* When dgram socket disconnects (or changes its peer), we clear its receive
331 * queue of packets arrived from previous peer. First, it allows to do
332 * flow control based only on wmem_alloc; second, sk connected to peer
333 * may receive messages only from that peer. */
334 static void unix_dgram_disconnected(struct sock *sk, struct sock *other)
336 if (!skb_queue_empty(&sk->sk_receive_queue)) {
337 skb_queue_purge(&sk->sk_receive_queue);
338 wake_up_interruptible_all(&unix_sk(sk)->peer_wait);
340 /* If one link of bidirectional dgram pipe is disconnected,
341 * we signal error. Messages are lost. Do not make this,
342 * when peer was not connected to us.
344 if (!sock_flag(other, SOCK_DEAD) && unix_peer(other) == sk) {
345 other->sk_err = ECONNRESET;
346 other->sk_error_report(other);
351 static void unix_sock_destructor(struct sock *sk)
353 struct unix_sock *u = unix_sk(sk);
355 skb_queue_purge(&sk->sk_receive_queue);
357 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
358 WARN_ON(!sk_unhashed(sk));
359 WARN_ON(sk->sk_socket);
360 if (!sock_flag(sk, SOCK_DEAD)) {
361 printk(KERN_INFO "Attempt to release alive unix socket: %p\n", sk);
366 unix_release_addr(u->addr);
368 atomic_long_dec(&unix_nr_socks);
370 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
372 #ifdef UNIX_REFCNT_DEBUG
373 printk(KERN_DEBUG "UNIX %p is destroyed, %ld are still alive.\n", sk,
374 atomic_long_read(&unix_nr_socks));
378 static int unix_release_sock(struct sock *sk, int embrion)
380 struct unix_sock *u = unix_sk(sk);
386 unix_remove_socket(sk);
391 sk->sk_shutdown = SHUTDOWN_MASK;
393 u->path.dentry = NULL;
395 state = sk->sk_state;
396 sk->sk_state = TCP_CLOSE;
397 unix_state_unlock(sk);
399 wake_up_interruptible_all(&u->peer_wait);
401 skpair = unix_peer(sk);
403 if (skpair != NULL) {
404 if (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) {
405 unix_state_lock(skpair);
407 skpair->sk_shutdown = SHUTDOWN_MASK;
408 if (!skb_queue_empty(&sk->sk_receive_queue) || embrion)
409 skpair->sk_err = ECONNRESET;
410 unix_state_unlock(skpair);
411 skpair->sk_state_change(skpair);
412 sk_wake_async(skpair, SOCK_WAKE_WAITD, POLL_HUP);
414 sock_put(skpair); /* It may now die */
415 unix_peer(sk) = NULL;
418 /* Try to flush out this socket. Throw out buffers at least */
420 while ((skb = skb_dequeue(&sk->sk_receive_queue)) != NULL) {
421 if (state == TCP_LISTEN)
422 unix_release_sock(skb->sk, 1);
423 /* passed fds are erased in the kfree_skb hook */
432 /* ---- Socket is dead now and most probably destroyed ---- */
435 * Fixme: BSD difference: In BSD all sockets connected to use get
436 * ECONNRESET and we die on the spot. In Linux we behave
437 * like files and pipes do and wait for the last
440 * Can't we simply set sock->err?
442 * What the above comment does talk about? --ANK(980817)
445 if (unix_tot_inflight)
446 unix_gc(); /* Garbage collect fds */
451 static void init_peercred(struct sock *sk)
453 put_pid(sk->sk_peer_pid);
454 if (sk->sk_peer_cred)
455 put_cred(sk->sk_peer_cred);
456 sk->sk_peer_pid = get_pid(task_tgid(current));
457 sk->sk_peer_cred = get_current_cred();
460 static void copy_peercred(struct sock *sk, struct sock *peersk)
462 put_pid(sk->sk_peer_pid);
463 if (sk->sk_peer_cred)
464 put_cred(sk->sk_peer_cred);
465 sk->sk_peer_pid = get_pid(peersk->sk_peer_pid);
466 sk->sk_peer_cred = get_cred(peersk->sk_peer_cred);
469 static int unix_listen(struct socket *sock, int backlog)
472 struct sock *sk = sock->sk;
473 struct unix_sock *u = unix_sk(sk);
474 struct pid *old_pid = NULL;
475 const struct cred *old_cred = NULL;
478 if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
479 goto out; /* Only stream/seqpacket sockets accept */
482 goto out; /* No listens on an unbound socket */
484 if (sk->sk_state != TCP_CLOSE && sk->sk_state != TCP_LISTEN)
486 if (backlog > sk->sk_max_ack_backlog)
487 wake_up_interruptible_all(&u->peer_wait);
488 sk->sk_max_ack_backlog = backlog;
489 sk->sk_state = TCP_LISTEN;
490 /* set credentials so connect can copy them */
495 unix_state_unlock(sk);
503 static int unix_release(struct socket *);
504 static int unix_bind(struct socket *, struct sockaddr *, int);
505 static int unix_stream_connect(struct socket *, struct sockaddr *,
506 int addr_len, int flags);
507 static int unix_socketpair(struct socket *, struct socket *);
508 static int unix_accept(struct socket *, struct socket *, int);
509 static int unix_getname(struct socket *, struct sockaddr *, int *, int);
510 static unsigned int unix_poll(struct file *, struct socket *, poll_table *);
511 static unsigned int unix_dgram_poll(struct file *, struct socket *,
513 static int unix_ioctl(struct socket *, unsigned int, unsigned long);
514 static int unix_shutdown(struct socket *, int);
515 static int unix_stream_sendmsg(struct kiocb *, struct socket *,
516 struct msghdr *, size_t);
517 static int unix_stream_recvmsg(struct kiocb *, struct socket *,
518 struct msghdr *, size_t, int);
519 static int unix_dgram_sendmsg(struct kiocb *, struct socket *,
520 struct msghdr *, size_t);
521 static int unix_dgram_recvmsg(struct kiocb *, struct socket *,
522 struct msghdr *, size_t, int);
523 static int unix_dgram_connect(struct socket *, struct sockaddr *,
525 static int unix_seqpacket_sendmsg(struct kiocb *, struct socket *,
526 struct msghdr *, size_t);
527 static int unix_seqpacket_recvmsg(struct kiocb *, struct socket *,
528 struct msghdr *, size_t, int);
530 static void unix_set_peek_off(struct sock *sk, int val)
532 struct unix_sock *u = unix_sk(sk);
534 mutex_lock(&u->readlock);
535 sk->sk_peek_off = val;
536 mutex_unlock(&u->readlock);
540 static const struct proto_ops unix_stream_ops = {
542 .owner = THIS_MODULE,
543 .release = unix_release,
545 .connect = unix_stream_connect,
546 .socketpair = unix_socketpair,
547 .accept = unix_accept,
548 .getname = unix_getname,
551 .listen = unix_listen,
552 .shutdown = unix_shutdown,
553 .setsockopt = sock_no_setsockopt,
554 .getsockopt = sock_no_getsockopt,
555 .sendmsg = unix_stream_sendmsg,
556 .recvmsg = unix_stream_recvmsg,
557 .mmap = sock_no_mmap,
558 .sendpage = sock_no_sendpage,
559 .set_peek_off = unix_set_peek_off,
562 static const struct proto_ops unix_dgram_ops = {
564 .owner = THIS_MODULE,
565 .release = unix_release,
567 .connect = unix_dgram_connect,
568 .socketpair = unix_socketpair,
569 .accept = sock_no_accept,
570 .getname = unix_getname,
571 .poll = unix_dgram_poll,
573 .listen = sock_no_listen,
574 .shutdown = unix_shutdown,
575 .setsockopt = sock_no_setsockopt,
576 .getsockopt = sock_no_getsockopt,
577 .sendmsg = unix_dgram_sendmsg,
578 .recvmsg = unix_dgram_recvmsg,
579 .mmap = sock_no_mmap,
580 .sendpage = sock_no_sendpage,
581 .set_peek_off = unix_set_peek_off,
584 static const struct proto_ops unix_seqpacket_ops = {
586 .owner = THIS_MODULE,
587 .release = unix_release,
589 .connect = unix_stream_connect,
590 .socketpair = unix_socketpair,
591 .accept = unix_accept,
592 .getname = unix_getname,
593 .poll = unix_dgram_poll,
595 .listen = unix_listen,
596 .shutdown = unix_shutdown,
597 .setsockopt = sock_no_setsockopt,
598 .getsockopt = sock_no_getsockopt,
599 .sendmsg = unix_seqpacket_sendmsg,
600 .recvmsg = unix_seqpacket_recvmsg,
601 .mmap = sock_no_mmap,
602 .sendpage = sock_no_sendpage,
603 .set_peek_off = unix_set_peek_off,
606 static struct proto unix_proto = {
608 .owner = THIS_MODULE,
609 .obj_size = sizeof(struct unix_sock),
613 * AF_UNIX sockets do not interact with hardware, hence they
614 * dont trigger interrupts - so it's safe for them to have
615 * bh-unsafe locking for their sk_receive_queue.lock. Split off
616 * this special lock-class by reinitializing the spinlock key:
618 static struct lock_class_key af_unix_sk_receive_queue_lock_key;
620 static struct sock *unix_create1(struct net *net, struct socket *sock)
622 struct sock *sk = NULL;
625 atomic_long_inc(&unix_nr_socks);
626 if (atomic_long_read(&unix_nr_socks) > 2 * get_max_files())
629 sk = sk_alloc(net, PF_UNIX, GFP_KERNEL, &unix_proto);
633 sock_init_data(sock, sk);
634 lockdep_set_class(&sk->sk_receive_queue.lock,
635 &af_unix_sk_receive_queue_lock_key);
637 sk->sk_write_space = unix_write_space;
638 sk->sk_max_ack_backlog = net->unx.sysctl_max_dgram_qlen;
639 sk->sk_destruct = unix_sock_destructor;
641 u->path.dentry = NULL;
643 spin_lock_init(&u->lock);
644 atomic_long_set(&u->inflight, 0);
645 INIT_LIST_HEAD(&u->link);
646 mutex_init(&u->readlock); /* single task reading lock */
647 init_waitqueue_head(&u->peer_wait);
648 unix_insert_socket(unix_sockets_unbound, sk);
651 atomic_long_dec(&unix_nr_socks);
654 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
660 static int unix_create(struct net *net, struct socket *sock, int protocol,
663 if (protocol && protocol != PF_UNIX)
664 return -EPROTONOSUPPORT;
666 sock->state = SS_UNCONNECTED;
668 switch (sock->type) {
670 sock->ops = &unix_stream_ops;
673 * Believe it or not BSD has AF_UNIX, SOCK_RAW though
677 sock->type = SOCK_DGRAM;
679 sock->ops = &unix_dgram_ops;
682 sock->ops = &unix_seqpacket_ops;
685 return -ESOCKTNOSUPPORT;
688 return unix_create1(net, sock) ? 0 : -ENOMEM;
691 static int unix_release(struct socket *sock)
693 struct sock *sk = sock->sk;
700 return unix_release_sock(sk, 0);
703 static int unix_autobind(struct socket *sock)
705 struct sock *sk = sock->sk;
706 struct net *net = sock_net(sk);
707 struct unix_sock *u = unix_sk(sk);
708 static u32 ordernum = 1;
709 struct unix_address *addr;
711 unsigned int retries = 0;
713 mutex_lock(&u->readlock);
720 addr = kzalloc(sizeof(*addr) + sizeof(short) + 16, GFP_KERNEL);
724 addr->name->sun_family = AF_UNIX;
725 atomic_set(&addr->refcnt, 1);
728 addr->len = sprintf(addr->name->sun_path+1, "%05x", ordernum) + 1 + sizeof(short);
729 addr->hash = unix_hash_fold(csum_partial(addr->name, addr->len, 0));
731 spin_lock(&unix_table_lock);
732 ordernum = (ordernum+1)&0xFFFFF;
734 if (__unix_find_socket_byname(net, addr->name, addr->len, sock->type,
736 spin_unlock(&unix_table_lock);
738 * __unix_find_socket_byname() may take long time if many names
739 * are already in use.
742 /* Give up if all names seems to be in use. */
743 if (retries++ == 0xFFFFF) {
750 addr->hash ^= sk->sk_type;
752 __unix_remove_socket(sk);
754 __unix_insert_socket(&unix_socket_table[addr->hash], sk);
755 spin_unlock(&unix_table_lock);
758 out: mutex_unlock(&u->readlock);
762 static struct sock *unix_find_other(struct net *net,
763 struct sockaddr_un *sunname, int len,
764 int type, unsigned int hash, int *error)
770 if (sunname->sun_path[0]) {
772 err = kern_path(sunname->sun_path, LOOKUP_FOLLOW, &path);
775 inode = path.dentry->d_inode;
776 err = inode_permission(inode, MAY_WRITE);
781 if (!S_ISSOCK(inode->i_mode))
783 u = unix_find_socket_byinode(inode);
787 if (u->sk_type == type)
793 if (u->sk_type != type) {
799 u = unix_find_socket_byname(net, sunname, len, type, hash);
801 struct dentry *dentry;
802 dentry = unix_sk(u)->path.dentry;
804 touch_atime(&unix_sk(u)->path);
818 static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
820 struct sock *sk = sock->sk;
821 struct net *net = sock_net(sk);
822 struct unix_sock *u = unix_sk(sk);
823 struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
824 char *sun_path = sunaddr->sun_path;
825 struct dentry *dentry = NULL;
829 struct unix_address *addr;
830 struct hlist_head *list;
833 if (sunaddr->sun_family != AF_UNIX)
836 if (addr_len == sizeof(short)) {
837 err = unix_autobind(sock);
841 err = unix_mkname(sunaddr, addr_len, &hash);
846 mutex_lock(&u->readlock);
853 addr = kmalloc(sizeof(*addr)+addr_len, GFP_KERNEL);
857 memcpy(addr->name, sunaddr, addr_len);
858 addr->len = addr_len;
859 addr->hash = hash ^ sk->sk_type;
860 atomic_set(&addr->refcnt, 1);
866 * Get the parent directory, calculate the hash for last
869 dentry = kern_path_create(AT_FDCWD, sun_path, &path, 0);
870 err = PTR_ERR(dentry);
872 goto out_mknod_parent;
875 * All right, let's create it.
878 (SOCK_INODE(sock)->i_mode & ~current_umask());
879 err = security_path_mknod(&path, dentry, mode, 0);
881 goto out_mknod_drop_write;
882 err = vfs_mknod(path.dentry->d_inode, dentry, mode, 0);
883 out_mknod_drop_write:
888 done_path_create(&path, dentry);
889 path.dentry = dentry;
891 addr->hash = UNIX_HASH_SIZE;
894 spin_lock(&unix_table_lock);
898 if (__unix_find_socket_byname(net, sunaddr, addr_len,
899 sk->sk_type, hash)) {
900 unix_release_addr(addr);
904 list = &unix_socket_table[addr->hash];
906 list = &unix_socket_table[dentry->d_inode->i_ino & (UNIX_HASH_SIZE-1)];
911 __unix_remove_socket(sk);
913 __unix_insert_socket(list, sk);
916 spin_unlock(&unix_table_lock);
918 mutex_unlock(&u->readlock);
923 done_path_create(&path, dentry);
927 unix_release_addr(addr);
931 static void unix_state_double_lock(struct sock *sk1, struct sock *sk2)
933 if (unlikely(sk1 == sk2) || !sk2) {
934 unix_state_lock(sk1);
938 unix_state_lock(sk1);
939 unix_state_lock_nested(sk2);
941 unix_state_lock(sk2);
942 unix_state_lock_nested(sk1);
946 static void unix_state_double_unlock(struct sock *sk1, struct sock *sk2)
948 if (unlikely(sk1 == sk2) || !sk2) {
949 unix_state_unlock(sk1);
952 unix_state_unlock(sk1);
953 unix_state_unlock(sk2);
956 static int unix_dgram_connect(struct socket *sock, struct sockaddr *addr,
959 struct sock *sk = sock->sk;
960 struct net *net = sock_net(sk);
961 struct sockaddr_un *sunaddr = (struct sockaddr_un *)addr;
966 if (addr->sa_family != AF_UNSPEC) {
967 err = unix_mkname(sunaddr, alen, &hash);
972 if (test_bit(SOCK_PASSCRED, &sock->flags) &&
973 !unix_sk(sk)->addr && (err = unix_autobind(sock)) != 0)
977 other = unix_find_other(net, sunaddr, alen, sock->type, hash, &err);
981 unix_state_double_lock(sk, other);
983 /* Apparently VFS overslept socket death. Retry. */
984 if (sock_flag(other, SOCK_DEAD)) {
985 unix_state_double_unlock(sk, other);
991 if (!unix_may_send(sk, other))
994 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1000 * 1003.1g breaking connected state with AF_UNSPEC
1003 unix_state_double_lock(sk, other);
1007 * If it was connected, reconnect.
1009 if (unix_peer(sk)) {
1010 struct sock *old_peer = unix_peer(sk);
1011 unix_peer(sk) = other;
1012 unix_state_double_unlock(sk, other);
1014 if (other != old_peer)
1015 unix_dgram_disconnected(sk, old_peer);
1018 unix_peer(sk) = other;
1019 unix_state_double_unlock(sk, other);
1024 unix_state_double_unlock(sk, other);
1030 static long unix_wait_for_peer(struct sock *other, long timeo)
1032 struct unix_sock *u = unix_sk(other);
1036 prepare_to_wait_exclusive(&u->peer_wait, &wait, TASK_INTERRUPTIBLE);
1038 sched = !sock_flag(other, SOCK_DEAD) &&
1039 !(other->sk_shutdown & RCV_SHUTDOWN) &&
1040 unix_recvq_full(other);
1042 unix_state_unlock(other);
1045 timeo = schedule_timeout(timeo);
1047 finish_wait(&u->peer_wait, &wait);
1051 static int unix_stream_connect(struct socket *sock, struct sockaddr *uaddr,
1052 int addr_len, int flags)
1054 struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
1055 struct sock *sk = sock->sk;
1056 struct net *net = sock_net(sk);
1057 struct unix_sock *u = unix_sk(sk), *newu, *otheru;
1058 struct sock *newsk = NULL;
1059 struct sock *other = NULL;
1060 struct sk_buff *skb = NULL;
1066 err = unix_mkname(sunaddr, addr_len, &hash);
1071 if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr &&
1072 (err = unix_autobind(sock)) != 0)
1075 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
1077 /* First of all allocate resources.
1078 If we will make it after state is locked,
1079 we will have to recheck all again in any case.
1084 /* create new sock for complete connection */
1085 newsk = unix_create1(sock_net(sk), NULL);
1089 /* Allocate skb for sending to listening sock */
1090 skb = sock_wmalloc(newsk, 1, 0, GFP_KERNEL);
1095 /* Find listening sock. */
1096 other = unix_find_other(net, sunaddr, addr_len, sk->sk_type, hash, &err);
1100 /* Latch state of peer */
1101 unix_state_lock(other);
1103 /* Apparently VFS overslept socket death. Retry. */
1104 if (sock_flag(other, SOCK_DEAD)) {
1105 unix_state_unlock(other);
1110 err = -ECONNREFUSED;
1111 if (other->sk_state != TCP_LISTEN)
1113 if (other->sk_shutdown & RCV_SHUTDOWN)
1116 if (unix_recvq_full(other)) {
1121 timeo = unix_wait_for_peer(other, timeo);
1123 err = sock_intr_errno(timeo);
1124 if (signal_pending(current))
1132 It is tricky place. We need to grab our state lock and cannot
1133 drop lock on peer. It is dangerous because deadlock is
1134 possible. Connect to self case and simultaneous
1135 attempt to connect are eliminated by checking socket
1136 state. other is TCP_LISTEN, if sk is TCP_LISTEN we
1137 check this before attempt to grab lock.
1139 Well, and we have to recheck the state after socket locked.
1145 /* This is ok... continue with connect */
1147 case TCP_ESTABLISHED:
1148 /* Socket is already connected */
1156 unix_state_lock_nested(sk);
1158 if (sk->sk_state != st) {
1159 unix_state_unlock(sk);
1160 unix_state_unlock(other);
1165 err = security_unix_stream_connect(sk, other, newsk);
1167 unix_state_unlock(sk);
1171 /* The way is open! Fastly set all the necessary fields... */
1174 unix_peer(newsk) = sk;
1175 newsk->sk_state = TCP_ESTABLISHED;
1176 newsk->sk_type = sk->sk_type;
1177 init_peercred(newsk);
1178 newu = unix_sk(newsk);
1179 RCU_INIT_POINTER(newsk->sk_wq, &newu->peer_wq);
1180 otheru = unix_sk(other);
1182 /* copy address information from listening to new sock*/
1184 atomic_inc(&otheru->addr->refcnt);
1185 newu->addr = otheru->addr;
1187 if (otheru->path.dentry) {
1188 path_get(&otheru->path);
1189 newu->path = otheru->path;
1192 /* Set credentials */
1193 copy_peercred(sk, other);
1195 sock->state = SS_CONNECTED;
1196 sk->sk_state = TCP_ESTABLISHED;
1199 smp_mb__after_atomic_inc(); /* sock_hold() does an atomic_inc() */
1200 unix_peer(sk) = newsk;
1202 unix_state_unlock(sk);
1204 /* take ten and and send info to listening sock */
1205 spin_lock(&other->sk_receive_queue.lock);
1206 __skb_queue_tail(&other->sk_receive_queue, skb);
1207 spin_unlock(&other->sk_receive_queue.lock);
1208 unix_state_unlock(other);
1209 other->sk_data_ready(other, 0);
1215 unix_state_unlock(other);
1220 unix_release_sock(newsk, 0);
1226 static int unix_socketpair(struct socket *socka, struct socket *sockb)
1228 struct sock *ska = socka->sk, *skb = sockb->sk;
1230 /* Join our sockets back to back */
1233 unix_peer(ska) = skb;
1234 unix_peer(skb) = ska;
1238 if (ska->sk_type != SOCK_DGRAM) {
1239 ska->sk_state = TCP_ESTABLISHED;
1240 skb->sk_state = TCP_ESTABLISHED;
1241 socka->state = SS_CONNECTED;
1242 sockb->state = SS_CONNECTED;
1247 static int unix_accept(struct socket *sock, struct socket *newsock, int flags)
1249 struct sock *sk = sock->sk;
1251 struct sk_buff *skb;
1255 if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
1259 if (sk->sk_state != TCP_LISTEN)
1262 /* If socket state is TCP_LISTEN it cannot change (for now...),
1263 * so that no locks are necessary.
1266 skb = skb_recv_datagram(sk, 0, flags&O_NONBLOCK, &err);
1268 /* This means receive shutdown. */
1275 skb_free_datagram(sk, skb);
1276 wake_up_interruptible(&unix_sk(sk)->peer_wait);
1278 /* attach accepted sock to socket */
1279 unix_state_lock(tsk);
1280 newsock->state = SS_CONNECTED;
1281 sock_graft(tsk, newsock);
1282 unix_state_unlock(tsk);
1290 static int unix_getname(struct socket *sock, struct sockaddr *uaddr, int *uaddr_len, int peer)
1292 struct sock *sk = sock->sk;
1293 struct unix_sock *u;
1294 DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr, uaddr);
1298 sk = unix_peer_get(sk);
1309 unix_state_lock(sk);
1311 sunaddr->sun_family = AF_UNIX;
1312 sunaddr->sun_path[0] = 0;
1313 *uaddr_len = sizeof(short);
1315 struct unix_address *addr = u->addr;
1317 *uaddr_len = addr->len;
1318 memcpy(sunaddr, addr->name, *uaddr_len);
1320 unix_state_unlock(sk);
1326 static void unix_detach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1330 scm->fp = UNIXCB(skb).fp;
1331 UNIXCB(skb).fp = NULL;
1333 for (i = scm->fp->count-1; i >= 0; i--)
1334 unix_notinflight(scm->fp->fp[i]);
1337 static void unix_destruct_scm(struct sk_buff *skb)
1339 struct scm_cookie scm;
1340 memset(&scm, 0, sizeof(scm));
1341 scm.pid = UNIXCB(skb).pid;
1342 scm.cred = UNIXCB(skb).cred;
1344 unix_detach_fds(&scm, skb);
1346 /* Alas, it calls VFS */
1347 /* So fscking what? fput() had been SMP-safe since the last Summer */
1352 #define MAX_RECURSION_LEVEL 4
1354 static int unix_attach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1357 unsigned char max_level = 0;
1358 int unix_sock_count = 0;
1360 for (i = scm->fp->count - 1; i >= 0; i--) {
1361 struct sock *sk = unix_get_socket(scm->fp->fp[i]);
1365 max_level = max(max_level,
1366 unix_sk(sk)->recursion_level);
1369 if (unlikely(max_level > MAX_RECURSION_LEVEL))
1370 return -ETOOMANYREFS;
1373 * Need to duplicate file references for the sake of garbage
1374 * collection. Otherwise a socket in the fps might become a
1375 * candidate for GC while the skb is not yet queued.
1377 UNIXCB(skb).fp = scm_fp_dup(scm->fp);
1378 if (!UNIXCB(skb).fp)
1381 if (unix_sock_count) {
1382 for (i = scm->fp->count - 1; i >= 0; i--)
1383 unix_inflight(scm->fp->fp[i]);
1388 static int unix_scm_to_skb(struct scm_cookie *scm, struct sk_buff *skb, bool send_fds)
1392 UNIXCB(skb).pid = get_pid(scm->pid);
1394 UNIXCB(skb).cred = get_cred(scm->cred);
1395 UNIXCB(skb).fp = NULL;
1396 if (scm->fp && send_fds)
1397 err = unix_attach_fds(scm, skb);
1399 skb->destructor = unix_destruct_scm;
1404 * Some apps rely on write() giving SCM_CREDENTIALS
1405 * We include credentials if source or destination socket
1406 * asserted SOCK_PASSCRED.
1408 static void maybe_add_creds(struct sk_buff *skb, const struct socket *sock,
1409 const struct sock *other)
1411 if (UNIXCB(skb).cred)
1413 if (test_bit(SOCK_PASSCRED, &sock->flags) ||
1414 !other->sk_socket ||
1415 test_bit(SOCK_PASSCRED, &other->sk_socket->flags)) {
1416 UNIXCB(skb).pid = get_pid(task_tgid(current));
1417 UNIXCB(skb).cred = get_current_cred();
1422 * Send AF_UNIX data.
1425 static int unix_dgram_sendmsg(struct kiocb *kiocb, struct socket *sock,
1426 struct msghdr *msg, size_t len)
1428 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1429 struct sock *sk = sock->sk;
1430 struct net *net = sock_net(sk);
1431 struct unix_sock *u = unix_sk(sk);
1432 struct sockaddr_un *sunaddr = msg->msg_name;
1433 struct sock *other = NULL;
1434 int namelen = 0; /* fake GCC */
1437 struct sk_buff *skb;
1439 struct scm_cookie tmp_scm;
1443 if (NULL == siocb->scm)
1444 siocb->scm = &tmp_scm;
1446 err = scm_send(sock, msg, siocb->scm);
1451 if (msg->msg_flags&MSG_OOB)
1454 if (msg->msg_namelen) {
1455 err = unix_mkname(sunaddr, msg->msg_namelen, &hash);
1462 other = unix_peer_get(sk);
1467 if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr
1468 && (err = unix_autobind(sock)) != 0)
1472 if (len > sk->sk_sndbuf - 32)
1475 if (len > SKB_MAX_ALLOC)
1476 data_len = min_t(size_t,
1477 len - SKB_MAX_ALLOC,
1478 MAX_SKB_FRAGS * PAGE_SIZE);
1480 skb = sock_alloc_send_pskb(sk, len - data_len, data_len,
1481 msg->msg_flags & MSG_DONTWAIT, &err);
1485 err = unix_scm_to_skb(siocb->scm, skb, true);
1488 max_level = err + 1;
1489 unix_get_secdata(siocb->scm, skb);
1491 skb_put(skb, len - data_len);
1492 skb->data_len = data_len;
1494 err = skb_copy_datagram_from_iovec(skb, 0, msg->msg_iov, 0, len);
1498 timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1503 if (sunaddr == NULL)
1506 other = unix_find_other(net, sunaddr, namelen, sk->sk_type,
1512 if (sk_filter(other, skb) < 0) {
1513 /* Toss the packet but do not return any error to the sender */
1518 unix_state_lock(other);
1520 if (!unix_may_send(sk, other))
1523 if (sock_flag(other, SOCK_DEAD)) {
1525 * Check with 1003.1g - what should
1528 unix_state_unlock(other);
1532 unix_state_lock(sk);
1533 if (unix_peer(sk) == other) {
1534 unix_peer(sk) = NULL;
1535 unix_state_unlock(sk);
1537 unix_dgram_disconnected(sk, other);
1539 err = -ECONNREFUSED;
1541 unix_state_unlock(sk);
1551 if (other->sk_shutdown & RCV_SHUTDOWN)
1554 if (sk->sk_type != SOCK_SEQPACKET) {
1555 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1560 if (unix_peer(other) != sk && unix_recvq_full(other)) {
1566 timeo = unix_wait_for_peer(other, timeo);
1568 err = sock_intr_errno(timeo);
1569 if (signal_pending(current))
1575 if (sock_flag(other, SOCK_RCVTSTAMP))
1576 __net_timestamp(skb);
1577 maybe_add_creds(skb, sock, other);
1578 skb_queue_tail(&other->sk_receive_queue, skb);
1579 if (max_level > unix_sk(other)->recursion_level)
1580 unix_sk(other)->recursion_level = max_level;
1581 unix_state_unlock(other);
1582 other->sk_data_ready(other, len);
1584 scm_destroy(siocb->scm);
1588 unix_state_unlock(other);
1594 scm_destroy(siocb->scm);
1599 static int unix_stream_sendmsg(struct kiocb *kiocb, struct socket *sock,
1600 struct msghdr *msg, size_t len)
1602 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1603 struct sock *sk = sock->sk;
1604 struct sock *other = NULL;
1606 struct sk_buff *skb;
1608 struct scm_cookie tmp_scm;
1609 bool fds_sent = false;
1612 if (NULL == siocb->scm)
1613 siocb->scm = &tmp_scm;
1615 err = scm_send(sock, msg, siocb->scm);
1620 if (msg->msg_flags&MSG_OOB)
1623 if (msg->msg_namelen) {
1624 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
1628 other = unix_peer(sk);
1633 if (sk->sk_shutdown & SEND_SHUTDOWN)
1636 while (sent < len) {
1638 * Optimisation for the fact that under 0.01% of X
1639 * messages typically need breaking up.
1644 /* Keep two messages in the pipe so it schedules better */
1645 if (size > ((sk->sk_sndbuf >> 1) - 64))
1646 size = (sk->sk_sndbuf >> 1) - 64;
1648 if (size > SKB_MAX_ALLOC)
1649 size = SKB_MAX_ALLOC;
1655 skb = sock_alloc_send_skb(sk, size, msg->msg_flags&MSG_DONTWAIT,
1662 * If you pass two values to the sock_alloc_send_skb
1663 * it tries to grab the large buffer with GFP_NOFS
1664 * (which can fail easily), and if it fails grab the
1665 * fallback size buffer which is under a page and will
1668 size = min_t(int, size, skb_tailroom(skb));
1671 /* Only send the fds in the first buffer */
1672 err = unix_scm_to_skb(siocb->scm, skb, !fds_sent);
1677 max_level = err + 1;
1680 err = memcpy_fromiovec(skb_put(skb, size), msg->msg_iov, size);
1686 unix_state_lock(other);
1688 if (sock_flag(other, SOCK_DEAD) ||
1689 (other->sk_shutdown & RCV_SHUTDOWN))
1692 maybe_add_creds(skb, sock, other);
1693 skb_queue_tail(&other->sk_receive_queue, skb);
1694 if (max_level > unix_sk(other)->recursion_level)
1695 unix_sk(other)->recursion_level = max_level;
1696 unix_state_unlock(other);
1697 other->sk_data_ready(other, size);
1701 scm_destroy(siocb->scm);
1707 unix_state_unlock(other);
1710 if (sent == 0 && !(msg->msg_flags&MSG_NOSIGNAL))
1711 send_sig(SIGPIPE, current, 0);
1714 scm_destroy(siocb->scm);
1716 return sent ? : err;
1719 static int unix_seqpacket_sendmsg(struct kiocb *kiocb, struct socket *sock,
1720 struct msghdr *msg, size_t len)
1723 struct sock *sk = sock->sk;
1725 err = sock_error(sk);
1729 if (sk->sk_state != TCP_ESTABLISHED)
1732 if (msg->msg_namelen)
1733 msg->msg_namelen = 0;
1735 return unix_dgram_sendmsg(kiocb, sock, msg, len);
1738 static int unix_seqpacket_recvmsg(struct kiocb *iocb, struct socket *sock,
1739 struct msghdr *msg, size_t size,
1742 struct sock *sk = sock->sk;
1744 if (sk->sk_state != TCP_ESTABLISHED)
1747 return unix_dgram_recvmsg(iocb, sock, msg, size, flags);
1750 static void unix_copy_addr(struct msghdr *msg, struct sock *sk)
1752 struct unix_sock *u = unix_sk(sk);
1754 msg->msg_namelen = 0;
1756 msg->msg_namelen = u->addr->len;
1757 memcpy(msg->msg_name, u->addr->name, u->addr->len);
1761 static int unix_dgram_recvmsg(struct kiocb *iocb, struct socket *sock,
1762 struct msghdr *msg, size_t size,
1765 struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1766 struct scm_cookie tmp_scm;
1767 struct sock *sk = sock->sk;
1768 struct unix_sock *u = unix_sk(sk);
1769 int noblock = flags & MSG_DONTWAIT;
1770 struct sk_buff *skb;
1778 msg->msg_namelen = 0;
1780 err = mutex_lock_interruptible(&u->readlock);
1782 err = sock_intr_errno(sock_rcvtimeo(sk, noblock));
1786 skip = sk_peek_offset(sk, flags);
1788 skb = __skb_recv_datagram(sk, flags, &peeked, &skip, &err);
1790 unix_state_lock(sk);
1791 /* Signal EOF on disconnected non-blocking SEQPACKET socket. */
1792 if (sk->sk_type == SOCK_SEQPACKET && err == -EAGAIN &&
1793 (sk->sk_shutdown & RCV_SHUTDOWN))
1795 unix_state_unlock(sk);
1799 wake_up_interruptible_sync_poll(&u->peer_wait,
1800 POLLOUT | POLLWRNORM | POLLWRBAND);
1803 unix_copy_addr(msg, skb->sk);
1805 if (size > skb->len - skip)
1806 size = skb->len - skip;
1807 else if (size < skb->len - skip)
1808 msg->msg_flags |= MSG_TRUNC;
1810 err = skb_copy_datagram_iovec(skb, skip, msg->msg_iov, size);
1814 if (sock_flag(sk, SOCK_RCVTSTAMP))
1815 __sock_recv_timestamp(msg, sk, skb);
1818 siocb->scm = &tmp_scm;
1819 memset(&tmp_scm, 0, sizeof(tmp_scm));
1821 scm_set_cred(siocb->scm, UNIXCB(skb).pid, UNIXCB(skb).cred);
1822 unix_set_secdata(siocb->scm, skb);
1824 if (!(flags & MSG_PEEK)) {
1826 unix_detach_fds(siocb->scm, skb);
1828 sk_peek_offset_bwd(sk, skb->len);
1830 /* It is questionable: on PEEK we could:
1831 - do not return fds - good, but too simple 8)
1832 - return fds, and do not return them on read (old strategy,
1834 - clone fds (I chose it for now, it is the most universal
1837 POSIX 1003.1g does not actually define this clearly
1838 at all. POSIX 1003.1g doesn't define a lot of things
1843 sk_peek_offset_fwd(sk, size);
1846 siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1848 err = (flags & MSG_TRUNC) ? skb->len - skip : size;
1850 scm_recv(sock, msg, siocb->scm, flags);
1853 skb_free_datagram(sk, skb);
1855 mutex_unlock(&u->readlock);
1861 * Sleep until data has arrive. But check for races..
1864 static long unix_stream_data_wait(struct sock *sk, long timeo)
1868 unix_state_lock(sk);
1871 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1873 if (!skb_queue_empty(&sk->sk_receive_queue) ||
1875 (sk->sk_shutdown & RCV_SHUTDOWN) ||
1876 signal_pending(current) ||
1880 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1881 unix_state_unlock(sk);
1882 timeo = schedule_timeout(timeo);
1883 unix_state_lock(sk);
1884 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1887 finish_wait(sk_sleep(sk), &wait);
1888 unix_state_unlock(sk);
1894 static int unix_stream_recvmsg(struct kiocb *iocb, struct socket *sock,
1895 struct msghdr *msg, size_t size,
1898 struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1899 struct scm_cookie tmp_scm;
1900 struct sock *sk = sock->sk;
1901 struct unix_sock *u = unix_sk(sk);
1902 struct sockaddr_un *sunaddr = msg->msg_name;
1904 int check_creds = 0;
1911 if (sk->sk_state != TCP_ESTABLISHED)
1918 target = sock_rcvlowat(sk, flags&MSG_WAITALL, size);
1919 timeo = sock_rcvtimeo(sk, flags&MSG_DONTWAIT);
1921 msg->msg_namelen = 0;
1923 /* Lock the socket to prevent queue disordering
1924 * while sleeps in memcpy_tomsg
1928 siocb->scm = &tmp_scm;
1929 memset(&tmp_scm, 0, sizeof(tmp_scm));
1932 err = mutex_lock_interruptible(&u->readlock);
1934 err = sock_intr_errno(timeo);
1938 skip = sk_peek_offset(sk, flags);
1942 struct sk_buff *skb;
1944 unix_state_lock(sk);
1945 skb = skb_peek(&sk->sk_receive_queue);
1948 unix_sk(sk)->recursion_level = 0;
1949 if (copied >= target)
1953 * POSIX 1003.1g mandates this order.
1956 err = sock_error(sk);
1959 if (sk->sk_shutdown & RCV_SHUTDOWN)
1962 unix_state_unlock(sk);
1966 mutex_unlock(&u->readlock);
1968 timeo = unix_stream_data_wait(sk, timeo);
1970 if (signal_pending(current)
1971 || mutex_lock_interruptible(&u->readlock)) {
1972 err = sock_intr_errno(timeo);
1978 unix_state_unlock(sk);
1982 if (skip >= skb->len) {
1984 skb = skb_peek_next(skb, &sk->sk_receive_queue);
1988 unix_state_unlock(sk);
1991 /* Never glue messages from different writers */
1992 if ((UNIXCB(skb).pid != siocb->scm->pid) ||
1993 (UNIXCB(skb).cred != siocb->scm->cred))
1996 /* Copy credentials */
1997 scm_set_cred(siocb->scm, UNIXCB(skb).pid, UNIXCB(skb).cred);
2001 /* Copy address just once */
2003 unix_copy_addr(msg, skb->sk);
2007 chunk = min_t(unsigned int, skb->len - skip, size);
2008 if (memcpy_toiovec(msg->msg_iov, skb->data + skip, chunk)) {
2016 /* Mark read part of skb as used */
2017 if (!(flags & MSG_PEEK)) {
2018 skb_pull(skb, chunk);
2020 sk_peek_offset_bwd(sk, chunk);
2023 unix_detach_fds(siocb->scm, skb);
2028 skb_unlink(skb, &sk->sk_receive_queue);
2034 /* It is questionable, see note in unix_dgram_recvmsg.
2037 siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
2039 sk_peek_offset_fwd(sk, chunk);
2045 mutex_unlock(&u->readlock);
2046 scm_recv(sock, msg, siocb->scm, flags);
2048 return copied ? : err;
2051 static int unix_shutdown(struct socket *sock, int mode)
2053 struct sock *sk = sock->sk;
2056 mode = (mode+1)&(RCV_SHUTDOWN|SEND_SHUTDOWN);
2061 unix_state_lock(sk);
2062 sk->sk_shutdown |= mode;
2063 other = unix_peer(sk);
2066 unix_state_unlock(sk);
2067 sk->sk_state_change(sk);
2070 (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET)) {
2074 if (mode&RCV_SHUTDOWN)
2075 peer_mode |= SEND_SHUTDOWN;
2076 if (mode&SEND_SHUTDOWN)
2077 peer_mode |= RCV_SHUTDOWN;
2078 unix_state_lock(other);
2079 other->sk_shutdown |= peer_mode;
2080 unix_state_unlock(other);
2081 other->sk_state_change(other);
2082 if (peer_mode == SHUTDOWN_MASK)
2083 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_HUP);
2084 else if (peer_mode & RCV_SHUTDOWN)
2085 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_IN);
2093 long unix_inq_len(struct sock *sk)
2095 struct sk_buff *skb;
2098 if (sk->sk_state == TCP_LISTEN)
2101 spin_lock(&sk->sk_receive_queue.lock);
2102 if (sk->sk_type == SOCK_STREAM ||
2103 sk->sk_type == SOCK_SEQPACKET) {
2104 skb_queue_walk(&sk->sk_receive_queue, skb)
2107 skb = skb_peek(&sk->sk_receive_queue);
2111 spin_unlock(&sk->sk_receive_queue.lock);
2115 EXPORT_SYMBOL_GPL(unix_inq_len);
2117 long unix_outq_len(struct sock *sk)
2119 return sk_wmem_alloc_get(sk);
2121 EXPORT_SYMBOL_GPL(unix_outq_len);
2123 static int unix_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2125 struct sock *sk = sock->sk;
2131 amount = unix_outq_len(sk);
2132 err = put_user(amount, (int __user *)arg);
2135 amount = unix_inq_len(sk);
2139 err = put_user(amount, (int __user *)arg);
2148 static unsigned int unix_poll(struct file *file, struct socket *sock, poll_table *wait)
2150 struct sock *sk = sock->sk;
2153 sock_poll_wait(file, sk_sleep(sk), wait);
2156 /* exceptional events? */
2159 if (sk->sk_shutdown == SHUTDOWN_MASK)
2161 if (sk->sk_shutdown & RCV_SHUTDOWN)
2162 mask |= POLLRDHUP | POLLIN | POLLRDNORM;
2165 if (!skb_queue_empty(&sk->sk_receive_queue))
2166 mask |= POLLIN | POLLRDNORM;
2168 /* Connection-based need to check for termination and startup */
2169 if ((sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) &&
2170 sk->sk_state == TCP_CLOSE)
2174 * we set writable also when the other side has shut down the
2175 * connection. This prevents stuck sockets.
2177 if (unix_writable(sk))
2178 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2183 static unsigned int unix_dgram_poll(struct file *file, struct socket *sock,
2186 struct sock *sk = sock->sk, *other;
2187 unsigned int mask, writable;
2189 sock_poll_wait(file, sk_sleep(sk), wait);
2192 /* exceptional events? */
2193 if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue))
2195 if (sk->sk_shutdown & RCV_SHUTDOWN)
2196 mask |= POLLRDHUP | POLLIN | POLLRDNORM;
2197 if (sk->sk_shutdown == SHUTDOWN_MASK)
2201 if (!skb_queue_empty(&sk->sk_receive_queue))
2202 mask |= POLLIN | POLLRDNORM;
2204 /* Connection-based need to check for termination and startup */
2205 if (sk->sk_type == SOCK_SEQPACKET) {
2206 if (sk->sk_state == TCP_CLOSE)
2208 /* connection hasn't started yet? */
2209 if (sk->sk_state == TCP_SYN_SENT)
2213 /* No write status requested, avoid expensive OUT tests. */
2214 if (!(poll_requested_events(wait) & (POLLWRBAND|POLLWRNORM|POLLOUT)))
2217 writable = unix_writable(sk);
2218 other = unix_peer_get(sk);
2220 if (unix_peer(other) != sk) {
2221 sock_poll_wait(file, &unix_sk(other)->peer_wait, wait);
2222 if (unix_recvq_full(other))
2229 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2231 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
2236 #ifdef CONFIG_PROC_FS
2237 static struct sock *first_unix_socket(int *i)
2239 for (*i = 0; *i <= UNIX_HASH_SIZE; (*i)++) {
2240 if (!hlist_empty(&unix_socket_table[*i]))
2241 return __sk_head(&unix_socket_table[*i]);
2246 static struct sock *next_unix_socket(int *i, struct sock *s)
2248 struct sock *next = sk_next(s);
2249 /* More in this chain? */
2252 /* Look for next non-empty chain. */
2253 for ((*i)++; *i <= UNIX_HASH_SIZE; (*i)++) {
2254 if (!hlist_empty(&unix_socket_table[*i]))
2255 return __sk_head(&unix_socket_table[*i]);
2260 struct unix_iter_state {
2261 struct seq_net_private p;
2265 static struct sock *unix_seq_idx(struct seq_file *seq, loff_t pos)
2267 struct unix_iter_state *iter = seq->private;
2271 for (s = first_unix_socket(&iter->i); s; s = next_unix_socket(&iter->i, s)) {
2272 if (sock_net(s) != seq_file_net(seq))
2281 static void *unix_seq_start(struct seq_file *seq, loff_t *pos)
2282 __acquires(unix_table_lock)
2284 spin_lock(&unix_table_lock);
2285 return *pos ? unix_seq_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2288 static void *unix_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2290 struct unix_iter_state *iter = seq->private;
2291 struct sock *sk = v;
2294 if (v == SEQ_START_TOKEN)
2295 sk = first_unix_socket(&iter->i);
2297 sk = next_unix_socket(&iter->i, sk);
2298 while (sk && (sock_net(sk) != seq_file_net(seq)))
2299 sk = next_unix_socket(&iter->i, sk);
2303 static void unix_seq_stop(struct seq_file *seq, void *v)
2304 __releases(unix_table_lock)
2306 spin_unlock(&unix_table_lock);
2309 static int unix_seq_show(struct seq_file *seq, void *v)
2312 if (v == SEQ_START_TOKEN)
2313 seq_puts(seq, "Num RefCount Protocol Flags Type St "
2317 struct unix_sock *u = unix_sk(s);
2320 seq_printf(seq, "%pK: %08X %08X %08X %04X %02X %5lu",
2322 atomic_read(&s->sk_refcnt),
2324 s->sk_state == TCP_LISTEN ? __SO_ACCEPTCON : 0,
2327 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTED : SS_UNCONNECTED) :
2328 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTING : SS_DISCONNECTING),
2336 len = u->addr->len - sizeof(short);
2337 if (!UNIX_ABSTRACT(s))
2343 for ( ; i < len; i++)
2344 seq_putc(seq, u->addr->name->sun_path[i]);
2346 unix_state_unlock(s);
2347 seq_putc(seq, '\n');
2353 static const struct seq_operations unix_seq_ops = {
2354 .start = unix_seq_start,
2355 .next = unix_seq_next,
2356 .stop = unix_seq_stop,
2357 .show = unix_seq_show,
2360 static int unix_seq_open(struct inode *inode, struct file *file)
2362 return seq_open_net(inode, file, &unix_seq_ops,
2363 sizeof(struct unix_iter_state));
2366 static const struct file_operations unix_seq_fops = {
2367 .owner = THIS_MODULE,
2368 .open = unix_seq_open,
2370 .llseek = seq_lseek,
2371 .release = seq_release_net,
2376 static const struct net_proto_family unix_family_ops = {
2378 .create = unix_create,
2379 .owner = THIS_MODULE,
2383 static int __net_init unix_net_init(struct net *net)
2385 int error = -ENOMEM;
2387 net->unx.sysctl_max_dgram_qlen = 10;
2388 if (unix_sysctl_register(net))
2391 #ifdef CONFIG_PROC_FS
2392 if (!proc_net_fops_create(net, "unix", 0, &unix_seq_fops)) {
2393 unix_sysctl_unregister(net);
2402 static void __net_exit unix_net_exit(struct net *net)
2404 unix_sysctl_unregister(net);
2405 proc_net_remove(net, "unix");
2408 static struct pernet_operations unix_net_ops = {
2409 .init = unix_net_init,
2410 .exit = unix_net_exit,
2413 static int __init af_unix_init(void)
2416 struct sk_buff *dummy_skb;
2418 BUILD_BUG_ON(sizeof(struct unix_skb_parms) > sizeof(dummy_skb->cb));
2420 rc = proto_register(&unix_proto, 1);
2422 printk(KERN_CRIT "%s: Cannot create unix_sock SLAB cache!\n",
2427 sock_register(&unix_family_ops);
2428 register_pernet_subsys(&unix_net_ops);
2433 static void __exit af_unix_exit(void)
2435 sock_unregister(PF_UNIX);
2436 proto_unregister(&unix_proto);
2437 unregister_pernet_subsys(&unix_net_ops);
2440 /* Earlier than device_initcall() so that other drivers invoking
2441 request_module() don't end up in a loop when modprobe tries
2442 to use a UNIX socket. But later than subsys_initcall() because
2443 we depend on stuff initialised there */
2444 fs_initcall(af_unix_init);
2445 module_exit(af_unix_exit);
2447 MODULE_LICENSE("GPL");
2448 MODULE_ALIAS_NETPROTO(PF_UNIX);