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 unix_hash_fold(__wsum n)
154 unsigned hash = (__force unsigned)n;
157 return hash&(UNIX_HASH_SIZE-1);
160 #define unix_peer(sk) (unix_sk(sk)->peer)
162 static inline int unix_our_peer(struct sock *sk, struct sock *osk)
164 return unix_peer(osk) == sk;
167 static inline int unix_may_send(struct sock *sk, struct sock *osk)
169 return unix_peer(osk) == NULL || unix_our_peer(sk, osk);
172 static inline int unix_recvq_full(struct sock const *sk)
174 return skb_queue_len(&sk->sk_receive_queue) > sk->sk_max_ack_backlog;
177 struct sock *unix_peer_get(struct sock *s)
185 unix_state_unlock(s);
188 EXPORT_SYMBOL_GPL(unix_peer_get);
190 static inline void unix_release_addr(struct unix_address *addr)
192 if (atomic_dec_and_test(&addr->refcnt))
197 * Check unix socket name:
198 * - should be not zero length.
199 * - if started by not zero, should be NULL terminated (FS object)
200 * - if started by zero, it is abstract name.
203 static int unix_mkname(struct sockaddr_un *sunaddr, int len, unsigned *hashp)
205 if (len <= sizeof(short) || len > sizeof(*sunaddr))
207 if (!sunaddr || sunaddr->sun_family != AF_UNIX)
209 if (sunaddr->sun_path[0]) {
211 * This may look like an off by one error but it is a bit more
212 * subtle. 108 is the longest valid AF_UNIX path for a binding.
213 * sun_path[108] doesn't as such exist. However in kernel space
214 * we are guaranteed that it is a valid memory location in our
215 * kernel address buffer.
217 ((char *)sunaddr)[len] = 0;
218 len = strlen(sunaddr->sun_path)+1+sizeof(short);
222 *hashp = unix_hash_fold(csum_partial(sunaddr, len, 0));
226 static void __unix_remove_socket(struct sock *sk)
228 sk_del_node_init(sk);
231 static void __unix_insert_socket(struct hlist_head *list, struct sock *sk)
233 WARN_ON(!sk_unhashed(sk));
234 sk_add_node(sk, list);
237 static inline void unix_remove_socket(struct sock *sk)
239 spin_lock(&unix_table_lock);
240 __unix_remove_socket(sk);
241 spin_unlock(&unix_table_lock);
244 static inline void unix_insert_socket(struct hlist_head *list, struct sock *sk)
246 spin_lock(&unix_table_lock);
247 __unix_insert_socket(list, sk);
248 spin_unlock(&unix_table_lock);
251 static struct sock *__unix_find_socket_byname(struct net *net,
252 struct sockaddr_un *sunname,
253 int len, int type, unsigned hash)
256 struct hlist_node *node;
258 sk_for_each(s, node, &unix_socket_table[hash ^ type]) {
259 struct unix_sock *u = unix_sk(s);
261 if (!net_eq(sock_net(s), net))
264 if (u->addr->len == len &&
265 !memcmp(u->addr->name, sunname, len))
273 static inline struct sock *unix_find_socket_byname(struct net *net,
274 struct sockaddr_un *sunname,
280 spin_lock(&unix_table_lock);
281 s = __unix_find_socket_byname(net, sunname, len, type, hash);
284 spin_unlock(&unix_table_lock);
288 static struct sock *unix_find_socket_byinode(struct inode *i)
291 struct hlist_node *node;
293 spin_lock(&unix_table_lock);
295 &unix_socket_table[i->i_ino & (UNIX_HASH_SIZE - 1)]) {
296 struct dentry *dentry = unix_sk(s)->path.dentry;
298 if (dentry && dentry->d_inode == i) {
305 spin_unlock(&unix_table_lock);
309 static inline int unix_writable(struct sock *sk)
311 return (atomic_read(&sk->sk_wmem_alloc) << 2) <= sk->sk_sndbuf;
314 static void unix_write_space(struct sock *sk)
316 struct socket_wq *wq;
319 if (unix_writable(sk)) {
320 wq = rcu_dereference(sk->sk_wq);
321 if (wq_has_sleeper(wq))
322 wake_up_interruptible_sync_poll(&wq->wait,
323 POLLOUT | POLLWRNORM | POLLWRBAND);
324 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
329 /* When dgram socket disconnects (or changes its peer), we clear its receive
330 * queue of packets arrived from previous peer. First, it allows to do
331 * flow control based only on wmem_alloc; second, sk connected to peer
332 * may receive messages only from that peer. */
333 static void unix_dgram_disconnected(struct sock *sk, struct sock *other)
335 if (!skb_queue_empty(&sk->sk_receive_queue)) {
336 skb_queue_purge(&sk->sk_receive_queue);
337 wake_up_interruptible_all(&unix_sk(sk)->peer_wait);
339 /* If one link of bidirectional dgram pipe is disconnected,
340 * we signal error. Messages are lost. Do not make this,
341 * when peer was not connected to us.
343 if (!sock_flag(other, SOCK_DEAD) && unix_peer(other) == sk) {
344 other->sk_err = ECONNRESET;
345 other->sk_error_report(other);
350 static void unix_sock_destructor(struct sock *sk)
352 struct unix_sock *u = unix_sk(sk);
354 skb_queue_purge(&sk->sk_receive_queue);
356 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
357 WARN_ON(!sk_unhashed(sk));
358 WARN_ON(sk->sk_socket);
359 if (!sock_flag(sk, SOCK_DEAD)) {
360 printk(KERN_INFO "Attempt to release alive unix socket: %p\n", sk);
365 unix_release_addr(u->addr);
367 atomic_long_dec(&unix_nr_socks);
369 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
371 #ifdef UNIX_REFCNT_DEBUG
372 printk(KERN_DEBUG "UNIX %p is destroyed, %ld are still alive.\n", sk,
373 atomic_long_read(&unix_nr_socks));
377 static int unix_release_sock(struct sock *sk, int embrion)
379 struct unix_sock *u = unix_sk(sk);
385 unix_remove_socket(sk);
390 sk->sk_shutdown = SHUTDOWN_MASK;
392 u->path.dentry = NULL;
394 state = sk->sk_state;
395 sk->sk_state = TCP_CLOSE;
396 unix_state_unlock(sk);
398 wake_up_interruptible_all(&u->peer_wait);
400 skpair = unix_peer(sk);
402 if (skpair != NULL) {
403 if (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) {
404 unix_state_lock(skpair);
406 skpair->sk_shutdown = SHUTDOWN_MASK;
407 if (!skb_queue_empty(&sk->sk_receive_queue) || embrion)
408 skpair->sk_err = ECONNRESET;
409 unix_state_unlock(skpair);
410 skpair->sk_state_change(skpair);
411 sk_wake_async(skpair, SOCK_WAKE_WAITD, POLL_HUP);
413 sock_put(skpair); /* It may now die */
414 unix_peer(sk) = NULL;
417 /* Try to flush out this socket. Throw out buffers at least */
419 while ((skb = skb_dequeue(&sk->sk_receive_queue)) != NULL) {
420 if (state == TCP_LISTEN)
421 unix_release_sock(skb->sk, 1);
422 /* passed fds are erased in the kfree_skb hook */
431 /* ---- Socket is dead now and most probably destroyed ---- */
434 * Fixme: BSD difference: In BSD all sockets connected to use get
435 * ECONNRESET and we die on the spot. In Linux we behave
436 * like files and pipes do and wait for the last
439 * Can't we simply set sock->err?
441 * What the above comment does talk about? --ANK(980817)
444 if (unix_tot_inflight)
445 unix_gc(); /* Garbage collect fds */
450 static void init_peercred(struct sock *sk)
452 put_pid(sk->sk_peer_pid);
453 if (sk->sk_peer_cred)
454 put_cred(sk->sk_peer_cred);
455 sk->sk_peer_pid = get_pid(task_tgid(current));
456 sk->sk_peer_cred = get_current_cred();
459 static void copy_peercred(struct sock *sk, struct sock *peersk)
461 put_pid(sk->sk_peer_pid);
462 if (sk->sk_peer_cred)
463 put_cred(sk->sk_peer_cred);
464 sk->sk_peer_pid = get_pid(peersk->sk_peer_pid);
465 sk->sk_peer_cred = get_cred(peersk->sk_peer_cred);
468 static int unix_listen(struct socket *sock, int backlog)
471 struct sock *sk = sock->sk;
472 struct unix_sock *u = unix_sk(sk);
473 struct pid *old_pid = NULL;
474 const struct cred *old_cred = NULL;
477 if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
478 goto out; /* Only stream/seqpacket sockets accept */
481 goto out; /* No listens on an unbound socket */
483 if (sk->sk_state != TCP_CLOSE && sk->sk_state != TCP_LISTEN)
485 if (backlog > sk->sk_max_ack_backlog)
486 wake_up_interruptible_all(&u->peer_wait);
487 sk->sk_max_ack_backlog = backlog;
488 sk->sk_state = TCP_LISTEN;
489 /* set credentials so connect can copy them */
494 unix_state_unlock(sk);
502 static int unix_release(struct socket *);
503 static int unix_bind(struct socket *, struct sockaddr *, int);
504 static int unix_stream_connect(struct socket *, struct sockaddr *,
505 int addr_len, int flags);
506 static int unix_socketpair(struct socket *, struct socket *);
507 static int unix_accept(struct socket *, struct socket *, int);
508 static int unix_getname(struct socket *, struct sockaddr *, int *, int);
509 static unsigned int unix_poll(struct file *, struct socket *, poll_table *);
510 static unsigned int unix_dgram_poll(struct file *, struct socket *,
512 static int unix_ioctl(struct socket *, unsigned int, unsigned long);
513 static int unix_shutdown(struct socket *, int);
514 static int unix_stream_sendmsg(struct kiocb *, struct socket *,
515 struct msghdr *, size_t);
516 static int unix_stream_recvmsg(struct kiocb *, struct socket *,
517 struct msghdr *, size_t, int);
518 static int unix_dgram_sendmsg(struct kiocb *, struct socket *,
519 struct msghdr *, size_t);
520 static int unix_dgram_recvmsg(struct kiocb *, struct socket *,
521 struct msghdr *, size_t, int);
522 static int unix_dgram_connect(struct socket *, struct sockaddr *,
524 static int unix_seqpacket_sendmsg(struct kiocb *, struct socket *,
525 struct msghdr *, size_t);
526 static int unix_seqpacket_recvmsg(struct kiocb *, struct socket *,
527 struct msghdr *, size_t, int);
529 static void unix_set_peek_off(struct sock *sk, int val)
531 struct unix_sock *u = unix_sk(sk);
533 mutex_lock(&u->readlock);
534 sk->sk_peek_off = val;
535 mutex_unlock(&u->readlock);
539 static const struct proto_ops unix_stream_ops = {
541 .owner = THIS_MODULE,
542 .release = unix_release,
544 .connect = unix_stream_connect,
545 .socketpair = unix_socketpair,
546 .accept = unix_accept,
547 .getname = unix_getname,
550 .listen = unix_listen,
551 .shutdown = unix_shutdown,
552 .setsockopt = sock_no_setsockopt,
553 .getsockopt = sock_no_getsockopt,
554 .sendmsg = unix_stream_sendmsg,
555 .recvmsg = unix_stream_recvmsg,
556 .mmap = sock_no_mmap,
557 .sendpage = sock_no_sendpage,
558 .set_peek_off = unix_set_peek_off,
561 static const struct proto_ops unix_dgram_ops = {
563 .owner = THIS_MODULE,
564 .release = unix_release,
566 .connect = unix_dgram_connect,
567 .socketpair = unix_socketpair,
568 .accept = sock_no_accept,
569 .getname = unix_getname,
570 .poll = unix_dgram_poll,
572 .listen = sock_no_listen,
573 .shutdown = unix_shutdown,
574 .setsockopt = sock_no_setsockopt,
575 .getsockopt = sock_no_getsockopt,
576 .sendmsg = unix_dgram_sendmsg,
577 .recvmsg = unix_dgram_recvmsg,
578 .mmap = sock_no_mmap,
579 .sendpage = sock_no_sendpage,
580 .set_peek_off = unix_set_peek_off,
583 static const struct proto_ops unix_seqpacket_ops = {
585 .owner = THIS_MODULE,
586 .release = unix_release,
588 .connect = unix_stream_connect,
589 .socketpair = unix_socketpair,
590 .accept = unix_accept,
591 .getname = unix_getname,
592 .poll = unix_dgram_poll,
594 .listen = unix_listen,
595 .shutdown = unix_shutdown,
596 .setsockopt = sock_no_setsockopt,
597 .getsockopt = sock_no_getsockopt,
598 .sendmsg = unix_seqpacket_sendmsg,
599 .recvmsg = unix_seqpacket_recvmsg,
600 .mmap = sock_no_mmap,
601 .sendpage = sock_no_sendpage,
602 .set_peek_off = unix_set_peek_off,
605 static struct proto unix_proto = {
607 .owner = THIS_MODULE,
608 .obj_size = sizeof(struct unix_sock),
612 * AF_UNIX sockets do not interact with hardware, hence they
613 * dont trigger interrupts - so it's safe for them to have
614 * bh-unsafe locking for their sk_receive_queue.lock. Split off
615 * this special lock-class by reinitializing the spinlock key:
617 static struct lock_class_key af_unix_sk_receive_queue_lock_key;
619 static struct sock *unix_create1(struct net *net, struct socket *sock)
621 struct sock *sk = NULL;
624 atomic_long_inc(&unix_nr_socks);
625 if (atomic_long_read(&unix_nr_socks) > 2 * get_max_files())
628 sk = sk_alloc(net, PF_UNIX, GFP_KERNEL, &unix_proto);
632 sock_init_data(sock, sk);
633 lockdep_set_class(&sk->sk_receive_queue.lock,
634 &af_unix_sk_receive_queue_lock_key);
636 sk->sk_write_space = unix_write_space;
637 sk->sk_max_ack_backlog = net->unx.sysctl_max_dgram_qlen;
638 sk->sk_destruct = unix_sock_destructor;
640 u->path.dentry = NULL;
642 spin_lock_init(&u->lock);
643 atomic_long_set(&u->inflight, 0);
644 INIT_LIST_HEAD(&u->link);
645 mutex_init(&u->readlock); /* single task reading lock */
646 init_waitqueue_head(&u->peer_wait);
647 unix_insert_socket(unix_sockets_unbound, sk);
650 atomic_long_dec(&unix_nr_socks);
653 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
659 static int unix_create(struct net *net, struct socket *sock, int protocol,
662 if (protocol && protocol != PF_UNIX)
663 return -EPROTONOSUPPORT;
665 sock->state = SS_UNCONNECTED;
667 switch (sock->type) {
669 sock->ops = &unix_stream_ops;
672 * Believe it or not BSD has AF_UNIX, SOCK_RAW though
676 sock->type = SOCK_DGRAM;
678 sock->ops = &unix_dgram_ops;
681 sock->ops = &unix_seqpacket_ops;
684 return -ESOCKTNOSUPPORT;
687 return unix_create1(net, sock) ? 0 : -ENOMEM;
690 static int unix_release(struct socket *sock)
692 struct sock *sk = sock->sk;
699 return unix_release_sock(sk, 0);
702 static int unix_autobind(struct socket *sock)
704 struct sock *sk = sock->sk;
705 struct net *net = sock_net(sk);
706 struct unix_sock *u = unix_sk(sk);
707 static u32 ordernum = 1;
708 struct unix_address *addr;
710 unsigned int retries = 0;
712 mutex_lock(&u->readlock);
719 addr = kzalloc(sizeof(*addr) + sizeof(short) + 16, GFP_KERNEL);
723 addr->name->sun_family = AF_UNIX;
724 atomic_set(&addr->refcnt, 1);
727 addr->len = sprintf(addr->name->sun_path+1, "%05x", ordernum) + 1 + sizeof(short);
728 addr->hash = unix_hash_fold(csum_partial(addr->name, addr->len, 0));
730 spin_lock(&unix_table_lock);
731 ordernum = (ordernum+1)&0xFFFFF;
733 if (__unix_find_socket_byname(net, addr->name, addr->len, sock->type,
735 spin_unlock(&unix_table_lock);
737 * __unix_find_socket_byname() may take long time if many names
738 * are already in use.
741 /* Give up if all names seems to be in use. */
742 if (retries++ == 0xFFFFF) {
749 addr->hash ^= sk->sk_type;
751 __unix_remove_socket(sk);
753 __unix_insert_socket(&unix_socket_table[addr->hash], sk);
754 spin_unlock(&unix_table_lock);
757 out: mutex_unlock(&u->readlock);
761 static struct sock *unix_find_other(struct net *net,
762 struct sockaddr_un *sunname, int len,
763 int type, unsigned hash, int *error)
769 if (sunname->sun_path[0]) {
771 err = kern_path(sunname->sun_path, LOOKUP_FOLLOW, &path);
774 inode = path.dentry->d_inode;
775 err = inode_permission(inode, MAY_WRITE);
780 if (!S_ISSOCK(inode->i_mode))
782 u = unix_find_socket_byinode(inode);
786 if (u->sk_type == type)
792 if (u->sk_type != type) {
798 u = unix_find_socket_byname(net, sunname, len, type, hash);
800 struct dentry *dentry;
801 dentry = unix_sk(u)->path.dentry;
803 touch_atime(&unix_sk(u)->path);
817 static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
819 struct sock *sk = sock->sk;
820 struct net *net = sock_net(sk);
821 struct unix_sock *u = unix_sk(sk);
822 struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
823 char *sun_path = sunaddr->sun_path;
824 struct dentry *dentry = NULL;
828 struct unix_address *addr;
829 struct hlist_head *list;
832 if (sunaddr->sun_family != AF_UNIX)
835 if (addr_len == sizeof(short)) {
836 err = unix_autobind(sock);
840 err = unix_mkname(sunaddr, addr_len, &hash);
845 mutex_lock(&u->readlock);
852 addr = kmalloc(sizeof(*addr)+addr_len, GFP_KERNEL);
856 memcpy(addr->name, sunaddr, addr_len);
857 addr->len = addr_len;
858 addr->hash = hash ^ sk->sk_type;
859 atomic_set(&addr->refcnt, 1);
865 * Get the parent directory, calculate the hash for last
868 dentry = kern_path_create(AT_FDCWD, sun_path, &path, 0);
869 err = PTR_ERR(dentry);
871 goto out_mknod_parent;
874 * All right, let's create it.
877 (SOCK_INODE(sock)->i_mode & ~current_umask());
878 err = mnt_want_write(path.mnt);
881 err = security_path_mknod(&path, dentry, mode, 0);
883 goto out_mknod_drop_write;
884 err = vfs_mknod(path.dentry->d_inode, dentry, mode, 0);
885 out_mknod_drop_write:
886 mnt_drop_write(path.mnt);
889 mutex_unlock(&path.dentry->d_inode->i_mutex);
891 path.dentry = dentry;
893 addr->hash = UNIX_HASH_SIZE;
896 spin_lock(&unix_table_lock);
900 if (__unix_find_socket_byname(net, sunaddr, addr_len,
901 sk->sk_type, hash)) {
902 unix_release_addr(addr);
906 list = &unix_socket_table[addr->hash];
908 list = &unix_socket_table[dentry->d_inode->i_ino & (UNIX_HASH_SIZE-1)];
913 __unix_remove_socket(sk);
915 __unix_insert_socket(list, sk);
918 spin_unlock(&unix_table_lock);
920 mutex_unlock(&u->readlock);
926 mutex_unlock(&path.dentry->d_inode->i_mutex);
931 unix_release_addr(addr);
935 static void unix_state_double_lock(struct sock *sk1, struct sock *sk2)
937 if (unlikely(sk1 == sk2) || !sk2) {
938 unix_state_lock(sk1);
942 unix_state_lock(sk1);
943 unix_state_lock_nested(sk2);
945 unix_state_lock(sk2);
946 unix_state_lock_nested(sk1);
950 static void unix_state_double_unlock(struct sock *sk1, struct sock *sk2)
952 if (unlikely(sk1 == sk2) || !sk2) {
953 unix_state_unlock(sk1);
956 unix_state_unlock(sk1);
957 unix_state_unlock(sk2);
960 static int unix_dgram_connect(struct socket *sock, struct sockaddr *addr,
963 struct sock *sk = sock->sk;
964 struct net *net = sock_net(sk);
965 struct sockaddr_un *sunaddr = (struct sockaddr_un *)addr;
970 if (addr->sa_family != AF_UNSPEC) {
971 err = unix_mkname(sunaddr, alen, &hash);
976 if (test_bit(SOCK_PASSCRED, &sock->flags) &&
977 !unix_sk(sk)->addr && (err = unix_autobind(sock)) != 0)
981 other = unix_find_other(net, sunaddr, alen, sock->type, hash, &err);
985 unix_state_double_lock(sk, other);
987 /* Apparently VFS overslept socket death. Retry. */
988 if (sock_flag(other, SOCK_DEAD)) {
989 unix_state_double_unlock(sk, other);
995 if (!unix_may_send(sk, other))
998 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1004 * 1003.1g breaking connected state with AF_UNSPEC
1007 unix_state_double_lock(sk, other);
1011 * If it was connected, reconnect.
1013 if (unix_peer(sk)) {
1014 struct sock *old_peer = unix_peer(sk);
1015 unix_peer(sk) = other;
1016 unix_state_double_unlock(sk, other);
1018 if (other != old_peer)
1019 unix_dgram_disconnected(sk, old_peer);
1022 unix_peer(sk) = other;
1023 unix_state_double_unlock(sk, other);
1028 unix_state_double_unlock(sk, other);
1034 static long unix_wait_for_peer(struct sock *other, long timeo)
1036 struct unix_sock *u = unix_sk(other);
1040 prepare_to_wait_exclusive(&u->peer_wait, &wait, TASK_INTERRUPTIBLE);
1042 sched = !sock_flag(other, SOCK_DEAD) &&
1043 !(other->sk_shutdown & RCV_SHUTDOWN) &&
1044 unix_recvq_full(other);
1046 unix_state_unlock(other);
1049 timeo = schedule_timeout(timeo);
1051 finish_wait(&u->peer_wait, &wait);
1055 static int unix_stream_connect(struct socket *sock, struct sockaddr *uaddr,
1056 int addr_len, int flags)
1058 struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
1059 struct sock *sk = sock->sk;
1060 struct net *net = sock_net(sk);
1061 struct unix_sock *u = unix_sk(sk), *newu, *otheru;
1062 struct sock *newsk = NULL;
1063 struct sock *other = NULL;
1064 struct sk_buff *skb = NULL;
1070 err = unix_mkname(sunaddr, addr_len, &hash);
1075 if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr &&
1076 (err = unix_autobind(sock)) != 0)
1079 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
1081 /* First of all allocate resources.
1082 If we will make it after state is locked,
1083 we will have to recheck all again in any case.
1088 /* create new sock for complete connection */
1089 newsk = unix_create1(sock_net(sk), NULL);
1093 /* Allocate skb for sending to listening sock */
1094 skb = sock_wmalloc(newsk, 1, 0, GFP_KERNEL);
1099 /* Find listening sock. */
1100 other = unix_find_other(net, sunaddr, addr_len, sk->sk_type, hash, &err);
1104 /* Latch state of peer */
1105 unix_state_lock(other);
1107 /* Apparently VFS overslept socket death. Retry. */
1108 if (sock_flag(other, SOCK_DEAD)) {
1109 unix_state_unlock(other);
1114 err = -ECONNREFUSED;
1115 if (other->sk_state != TCP_LISTEN)
1117 if (other->sk_shutdown & RCV_SHUTDOWN)
1120 if (unix_recvq_full(other)) {
1125 timeo = unix_wait_for_peer(other, timeo);
1127 err = sock_intr_errno(timeo);
1128 if (signal_pending(current))
1136 It is tricky place. We need to grab our state lock and cannot
1137 drop lock on peer. It is dangerous because deadlock is
1138 possible. Connect to self case and simultaneous
1139 attempt to connect are eliminated by checking socket
1140 state. other is TCP_LISTEN, if sk is TCP_LISTEN we
1141 check this before attempt to grab lock.
1143 Well, and we have to recheck the state after socket locked.
1149 /* This is ok... continue with connect */
1151 case TCP_ESTABLISHED:
1152 /* Socket is already connected */
1160 unix_state_lock_nested(sk);
1162 if (sk->sk_state != st) {
1163 unix_state_unlock(sk);
1164 unix_state_unlock(other);
1169 err = security_unix_stream_connect(sk, other, newsk);
1171 unix_state_unlock(sk);
1175 /* The way is open! Fastly set all the necessary fields... */
1178 unix_peer(newsk) = sk;
1179 newsk->sk_state = TCP_ESTABLISHED;
1180 newsk->sk_type = sk->sk_type;
1181 init_peercred(newsk);
1182 newu = unix_sk(newsk);
1183 RCU_INIT_POINTER(newsk->sk_wq, &newu->peer_wq);
1184 otheru = unix_sk(other);
1186 /* copy address information from listening to new sock*/
1188 atomic_inc(&otheru->addr->refcnt);
1189 newu->addr = otheru->addr;
1191 if (otheru->path.dentry) {
1192 path_get(&otheru->path);
1193 newu->path = otheru->path;
1196 /* Set credentials */
1197 copy_peercred(sk, other);
1199 sock->state = SS_CONNECTED;
1200 sk->sk_state = TCP_ESTABLISHED;
1203 smp_mb__after_atomic_inc(); /* sock_hold() does an atomic_inc() */
1204 unix_peer(sk) = newsk;
1206 unix_state_unlock(sk);
1208 /* take ten and and send info to listening sock */
1209 spin_lock(&other->sk_receive_queue.lock);
1210 __skb_queue_tail(&other->sk_receive_queue, skb);
1211 spin_unlock(&other->sk_receive_queue.lock);
1212 unix_state_unlock(other);
1213 other->sk_data_ready(other, 0);
1219 unix_state_unlock(other);
1224 unix_release_sock(newsk, 0);
1230 static int unix_socketpair(struct socket *socka, struct socket *sockb)
1232 struct sock *ska = socka->sk, *skb = sockb->sk;
1234 /* Join our sockets back to back */
1237 unix_peer(ska) = skb;
1238 unix_peer(skb) = ska;
1242 if (ska->sk_type != SOCK_DGRAM) {
1243 ska->sk_state = TCP_ESTABLISHED;
1244 skb->sk_state = TCP_ESTABLISHED;
1245 socka->state = SS_CONNECTED;
1246 sockb->state = SS_CONNECTED;
1251 static int unix_accept(struct socket *sock, struct socket *newsock, int flags)
1253 struct sock *sk = sock->sk;
1255 struct sk_buff *skb;
1259 if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
1263 if (sk->sk_state != TCP_LISTEN)
1266 /* If socket state is TCP_LISTEN it cannot change (for now...),
1267 * so that no locks are necessary.
1270 skb = skb_recv_datagram(sk, 0, flags&O_NONBLOCK, &err);
1272 /* This means receive shutdown. */
1279 skb_free_datagram(sk, skb);
1280 wake_up_interruptible(&unix_sk(sk)->peer_wait);
1282 /* attach accepted sock to socket */
1283 unix_state_lock(tsk);
1284 newsock->state = SS_CONNECTED;
1285 sock_graft(tsk, newsock);
1286 unix_state_unlock(tsk);
1294 static int unix_getname(struct socket *sock, struct sockaddr *uaddr, int *uaddr_len, int peer)
1296 struct sock *sk = sock->sk;
1297 struct unix_sock *u;
1298 DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr, uaddr);
1302 sk = unix_peer_get(sk);
1313 unix_state_lock(sk);
1315 sunaddr->sun_family = AF_UNIX;
1316 sunaddr->sun_path[0] = 0;
1317 *uaddr_len = sizeof(short);
1319 struct unix_address *addr = u->addr;
1321 *uaddr_len = addr->len;
1322 memcpy(sunaddr, addr->name, *uaddr_len);
1324 unix_state_unlock(sk);
1330 static void unix_detach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1334 scm->fp = UNIXCB(skb).fp;
1335 UNIXCB(skb).fp = NULL;
1337 for (i = scm->fp->count-1; i >= 0; i--)
1338 unix_notinflight(scm->fp->fp[i]);
1341 static void unix_destruct_scm(struct sk_buff *skb)
1343 struct scm_cookie scm;
1344 memset(&scm, 0, sizeof(scm));
1345 scm.pid = UNIXCB(skb).pid;
1346 scm.cred = UNIXCB(skb).cred;
1348 unix_detach_fds(&scm, skb);
1350 /* Alas, it calls VFS */
1351 /* So fscking what? fput() had been SMP-safe since the last Summer */
1356 #define MAX_RECURSION_LEVEL 4
1358 static int unix_attach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1361 unsigned char max_level = 0;
1362 int unix_sock_count = 0;
1364 for (i = scm->fp->count - 1; i >= 0; i--) {
1365 struct sock *sk = unix_get_socket(scm->fp->fp[i]);
1369 max_level = max(max_level,
1370 unix_sk(sk)->recursion_level);
1373 if (unlikely(max_level > MAX_RECURSION_LEVEL))
1374 return -ETOOMANYREFS;
1377 * Need to duplicate file references for the sake of garbage
1378 * collection. Otherwise a socket in the fps might become a
1379 * candidate for GC while the skb is not yet queued.
1381 UNIXCB(skb).fp = scm_fp_dup(scm->fp);
1382 if (!UNIXCB(skb).fp)
1385 if (unix_sock_count) {
1386 for (i = scm->fp->count - 1; i >= 0; i--)
1387 unix_inflight(scm->fp->fp[i]);
1392 static int unix_scm_to_skb(struct scm_cookie *scm, struct sk_buff *skb, bool send_fds)
1396 UNIXCB(skb).pid = get_pid(scm->pid);
1398 UNIXCB(skb).cred = get_cred(scm->cred);
1399 UNIXCB(skb).fp = NULL;
1400 if (scm->fp && send_fds)
1401 err = unix_attach_fds(scm, skb);
1403 skb->destructor = unix_destruct_scm;
1408 * Some apps rely on write() giving SCM_CREDENTIALS
1409 * We include credentials if source or destination socket
1410 * asserted SOCK_PASSCRED.
1412 static void maybe_add_creds(struct sk_buff *skb, const struct socket *sock,
1413 const struct sock *other)
1415 if (UNIXCB(skb).cred)
1417 if (test_bit(SOCK_PASSCRED, &sock->flags) ||
1418 !other->sk_socket ||
1419 test_bit(SOCK_PASSCRED, &other->sk_socket->flags)) {
1420 UNIXCB(skb).pid = get_pid(task_tgid(current));
1421 UNIXCB(skb).cred = get_current_cred();
1426 * Send AF_UNIX data.
1429 static int unix_dgram_sendmsg(struct kiocb *kiocb, struct socket *sock,
1430 struct msghdr *msg, size_t len)
1432 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1433 struct sock *sk = sock->sk;
1434 struct net *net = sock_net(sk);
1435 struct unix_sock *u = unix_sk(sk);
1436 struct sockaddr_un *sunaddr = msg->msg_name;
1437 struct sock *other = NULL;
1438 int namelen = 0; /* fake GCC */
1441 struct sk_buff *skb;
1443 struct scm_cookie tmp_scm;
1447 if (NULL == siocb->scm)
1448 siocb->scm = &tmp_scm;
1450 err = scm_send(sock, msg, siocb->scm);
1455 if (msg->msg_flags&MSG_OOB)
1458 if (msg->msg_namelen) {
1459 err = unix_mkname(sunaddr, msg->msg_namelen, &hash);
1466 other = unix_peer_get(sk);
1471 if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr
1472 && (err = unix_autobind(sock)) != 0)
1476 if (len > sk->sk_sndbuf - 32)
1479 if (len > SKB_MAX_ALLOC)
1480 data_len = min_t(size_t,
1481 len - SKB_MAX_ALLOC,
1482 MAX_SKB_FRAGS * PAGE_SIZE);
1484 skb = sock_alloc_send_pskb(sk, len - data_len, data_len,
1485 msg->msg_flags & MSG_DONTWAIT, &err);
1489 err = unix_scm_to_skb(siocb->scm, skb, true);
1492 max_level = err + 1;
1493 unix_get_secdata(siocb->scm, skb);
1495 skb_put(skb, len - data_len);
1496 skb->data_len = data_len;
1498 err = skb_copy_datagram_from_iovec(skb, 0, msg->msg_iov, 0, len);
1502 timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1507 if (sunaddr == NULL)
1510 other = unix_find_other(net, sunaddr, namelen, sk->sk_type,
1516 if (sk_filter(other, skb) < 0) {
1517 /* Toss the packet but do not return any error to the sender */
1522 unix_state_lock(other);
1524 if (!unix_may_send(sk, other))
1527 if (sock_flag(other, SOCK_DEAD)) {
1529 * Check with 1003.1g - what should
1532 unix_state_unlock(other);
1536 unix_state_lock(sk);
1537 if (unix_peer(sk) == other) {
1538 unix_peer(sk) = NULL;
1539 unix_state_unlock(sk);
1541 unix_dgram_disconnected(sk, other);
1543 err = -ECONNREFUSED;
1545 unix_state_unlock(sk);
1555 if (other->sk_shutdown & RCV_SHUTDOWN)
1558 if (sk->sk_type != SOCK_SEQPACKET) {
1559 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1564 if (unix_peer(other) != sk && unix_recvq_full(other)) {
1570 timeo = unix_wait_for_peer(other, timeo);
1572 err = sock_intr_errno(timeo);
1573 if (signal_pending(current))
1579 if (sock_flag(other, SOCK_RCVTSTAMP))
1580 __net_timestamp(skb);
1581 maybe_add_creds(skb, sock, other);
1582 skb_queue_tail(&other->sk_receive_queue, skb);
1583 if (max_level > unix_sk(other)->recursion_level)
1584 unix_sk(other)->recursion_level = max_level;
1585 unix_state_unlock(other);
1586 other->sk_data_ready(other, len);
1588 scm_destroy(siocb->scm);
1592 unix_state_unlock(other);
1598 scm_destroy(siocb->scm);
1603 static int unix_stream_sendmsg(struct kiocb *kiocb, struct socket *sock,
1604 struct msghdr *msg, size_t len)
1606 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1607 struct sock *sk = sock->sk;
1608 struct sock *other = NULL;
1610 struct sk_buff *skb;
1612 struct scm_cookie tmp_scm;
1613 bool fds_sent = false;
1616 if (NULL == siocb->scm)
1617 siocb->scm = &tmp_scm;
1619 err = scm_send(sock, msg, siocb->scm);
1624 if (msg->msg_flags&MSG_OOB)
1627 if (msg->msg_namelen) {
1628 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
1632 other = unix_peer(sk);
1637 if (sk->sk_shutdown & SEND_SHUTDOWN)
1640 while (sent < len) {
1642 * Optimisation for the fact that under 0.01% of X
1643 * messages typically need breaking up.
1648 /* Keep two messages in the pipe so it schedules better */
1649 if (size > ((sk->sk_sndbuf >> 1) - 64))
1650 size = (sk->sk_sndbuf >> 1) - 64;
1652 if (size > SKB_MAX_ALLOC)
1653 size = SKB_MAX_ALLOC;
1659 skb = sock_alloc_send_skb(sk, size, msg->msg_flags&MSG_DONTWAIT,
1666 * If you pass two values to the sock_alloc_send_skb
1667 * it tries to grab the large buffer with GFP_NOFS
1668 * (which can fail easily), and if it fails grab the
1669 * fallback size buffer which is under a page and will
1672 size = min_t(int, size, skb_tailroom(skb));
1675 /* Only send the fds in the first buffer */
1676 err = unix_scm_to_skb(siocb->scm, skb, !fds_sent);
1681 max_level = err + 1;
1684 err = memcpy_fromiovec(skb_put(skb, size), msg->msg_iov, size);
1690 unix_state_lock(other);
1692 if (sock_flag(other, SOCK_DEAD) ||
1693 (other->sk_shutdown & RCV_SHUTDOWN))
1696 maybe_add_creds(skb, sock, other);
1697 skb_queue_tail(&other->sk_receive_queue, skb);
1698 if (max_level > unix_sk(other)->recursion_level)
1699 unix_sk(other)->recursion_level = max_level;
1700 unix_state_unlock(other);
1701 other->sk_data_ready(other, size);
1705 scm_destroy(siocb->scm);
1711 unix_state_unlock(other);
1714 if (sent == 0 && !(msg->msg_flags&MSG_NOSIGNAL))
1715 send_sig(SIGPIPE, current, 0);
1718 scm_destroy(siocb->scm);
1720 return sent ? : err;
1723 static int unix_seqpacket_sendmsg(struct kiocb *kiocb, struct socket *sock,
1724 struct msghdr *msg, size_t len)
1727 struct sock *sk = sock->sk;
1729 err = sock_error(sk);
1733 if (sk->sk_state != TCP_ESTABLISHED)
1736 if (msg->msg_namelen)
1737 msg->msg_namelen = 0;
1739 return unix_dgram_sendmsg(kiocb, sock, msg, len);
1742 static int unix_seqpacket_recvmsg(struct kiocb *iocb, struct socket *sock,
1743 struct msghdr *msg, size_t size,
1746 struct sock *sk = sock->sk;
1748 if (sk->sk_state != TCP_ESTABLISHED)
1751 return unix_dgram_recvmsg(iocb, sock, msg, size, flags);
1754 static void unix_copy_addr(struct msghdr *msg, struct sock *sk)
1756 struct unix_sock *u = unix_sk(sk);
1758 msg->msg_namelen = 0;
1760 msg->msg_namelen = u->addr->len;
1761 memcpy(msg->msg_name, u->addr->name, u->addr->len);
1765 static int unix_dgram_recvmsg(struct kiocb *iocb, struct socket *sock,
1766 struct msghdr *msg, size_t size,
1769 struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1770 struct scm_cookie tmp_scm;
1771 struct sock *sk = sock->sk;
1772 struct unix_sock *u = unix_sk(sk);
1773 int noblock = flags & MSG_DONTWAIT;
1774 struct sk_buff *skb;
1782 msg->msg_namelen = 0;
1784 err = mutex_lock_interruptible(&u->readlock);
1786 err = sock_intr_errno(sock_rcvtimeo(sk, noblock));
1790 skip = sk_peek_offset(sk, flags);
1792 skb = __skb_recv_datagram(sk, flags, &peeked, &skip, &err);
1794 unix_state_lock(sk);
1795 /* Signal EOF on disconnected non-blocking SEQPACKET socket. */
1796 if (sk->sk_type == SOCK_SEQPACKET && err == -EAGAIN &&
1797 (sk->sk_shutdown & RCV_SHUTDOWN))
1799 unix_state_unlock(sk);
1803 wake_up_interruptible_sync_poll(&u->peer_wait,
1804 POLLOUT | POLLWRNORM | POLLWRBAND);
1807 unix_copy_addr(msg, skb->sk);
1809 if (size > skb->len - skip)
1810 size = skb->len - skip;
1811 else if (size < skb->len - skip)
1812 msg->msg_flags |= MSG_TRUNC;
1814 err = skb_copy_datagram_iovec(skb, skip, msg->msg_iov, size);
1818 if (sock_flag(sk, SOCK_RCVTSTAMP))
1819 __sock_recv_timestamp(msg, sk, skb);
1822 siocb->scm = &tmp_scm;
1823 memset(&tmp_scm, 0, sizeof(tmp_scm));
1825 scm_set_cred(siocb->scm, UNIXCB(skb).pid, UNIXCB(skb).cred);
1826 unix_set_secdata(siocb->scm, skb);
1828 if (!(flags & MSG_PEEK)) {
1830 unix_detach_fds(siocb->scm, skb);
1832 sk_peek_offset_bwd(sk, skb->len);
1834 /* It is questionable: on PEEK we could:
1835 - do not return fds - good, but too simple 8)
1836 - return fds, and do not return them on read (old strategy,
1838 - clone fds (I chose it for now, it is the most universal
1841 POSIX 1003.1g does not actually define this clearly
1842 at all. POSIX 1003.1g doesn't define a lot of things
1847 sk_peek_offset_fwd(sk, size);
1850 siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1852 err = (flags & MSG_TRUNC) ? skb->len - skip : size;
1854 scm_recv(sock, msg, siocb->scm, flags);
1857 skb_free_datagram(sk, skb);
1859 mutex_unlock(&u->readlock);
1865 * Sleep until data has arrive. But check for races..
1868 static long unix_stream_data_wait(struct sock *sk, long timeo)
1872 unix_state_lock(sk);
1875 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1877 if (!skb_queue_empty(&sk->sk_receive_queue) ||
1879 (sk->sk_shutdown & RCV_SHUTDOWN) ||
1880 signal_pending(current) ||
1884 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1885 unix_state_unlock(sk);
1886 timeo = schedule_timeout(timeo);
1887 unix_state_lock(sk);
1888 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1891 finish_wait(sk_sleep(sk), &wait);
1892 unix_state_unlock(sk);
1898 static int unix_stream_recvmsg(struct kiocb *iocb, struct socket *sock,
1899 struct msghdr *msg, size_t size,
1902 struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1903 struct scm_cookie tmp_scm;
1904 struct sock *sk = sock->sk;
1905 struct unix_sock *u = unix_sk(sk);
1906 struct sockaddr_un *sunaddr = msg->msg_name;
1908 int check_creds = 0;
1915 if (sk->sk_state != TCP_ESTABLISHED)
1922 target = sock_rcvlowat(sk, flags&MSG_WAITALL, size);
1923 timeo = sock_rcvtimeo(sk, flags&MSG_DONTWAIT);
1925 msg->msg_namelen = 0;
1927 /* Lock the socket to prevent queue disordering
1928 * while sleeps in memcpy_tomsg
1932 siocb->scm = &tmp_scm;
1933 memset(&tmp_scm, 0, sizeof(tmp_scm));
1936 err = mutex_lock_interruptible(&u->readlock);
1938 err = sock_intr_errno(timeo);
1942 skip = sk_peek_offset(sk, flags);
1946 struct sk_buff *skb;
1948 unix_state_lock(sk);
1949 skb = skb_peek(&sk->sk_receive_queue);
1952 unix_sk(sk)->recursion_level = 0;
1953 if (copied >= target)
1957 * POSIX 1003.1g mandates this order.
1960 err = sock_error(sk);
1963 if (sk->sk_shutdown & RCV_SHUTDOWN)
1966 unix_state_unlock(sk);
1970 mutex_unlock(&u->readlock);
1972 timeo = unix_stream_data_wait(sk, timeo);
1974 if (signal_pending(current)
1975 || mutex_lock_interruptible(&u->readlock)) {
1976 err = sock_intr_errno(timeo);
1982 unix_state_unlock(sk);
1986 if (skip >= skb->len) {
1988 skb = skb_peek_next(skb, &sk->sk_receive_queue);
1992 unix_state_unlock(sk);
1995 /* Never glue messages from different writers */
1996 if ((UNIXCB(skb).pid != siocb->scm->pid) ||
1997 (UNIXCB(skb).cred != siocb->scm->cred))
2000 /* Copy credentials */
2001 scm_set_cred(siocb->scm, UNIXCB(skb).pid, UNIXCB(skb).cred);
2005 /* Copy address just once */
2007 unix_copy_addr(msg, skb->sk);
2011 chunk = min_t(unsigned int, skb->len - skip, size);
2012 if (memcpy_toiovec(msg->msg_iov, skb->data + skip, chunk)) {
2020 /* Mark read part of skb as used */
2021 if (!(flags & MSG_PEEK)) {
2022 skb_pull(skb, chunk);
2024 sk_peek_offset_bwd(sk, chunk);
2027 unix_detach_fds(siocb->scm, skb);
2032 skb_unlink(skb, &sk->sk_receive_queue);
2038 /* It is questionable, see note in unix_dgram_recvmsg.
2041 siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
2043 sk_peek_offset_fwd(sk, chunk);
2049 mutex_unlock(&u->readlock);
2050 scm_recv(sock, msg, siocb->scm, flags);
2052 return copied ? : err;
2055 static int unix_shutdown(struct socket *sock, int mode)
2057 struct sock *sk = sock->sk;
2060 mode = (mode+1)&(RCV_SHUTDOWN|SEND_SHUTDOWN);
2065 unix_state_lock(sk);
2066 sk->sk_shutdown |= mode;
2067 other = unix_peer(sk);
2070 unix_state_unlock(sk);
2071 sk->sk_state_change(sk);
2074 (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET)) {
2078 if (mode&RCV_SHUTDOWN)
2079 peer_mode |= SEND_SHUTDOWN;
2080 if (mode&SEND_SHUTDOWN)
2081 peer_mode |= RCV_SHUTDOWN;
2082 unix_state_lock(other);
2083 other->sk_shutdown |= peer_mode;
2084 unix_state_unlock(other);
2085 other->sk_state_change(other);
2086 if (peer_mode == SHUTDOWN_MASK)
2087 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_HUP);
2088 else if (peer_mode & RCV_SHUTDOWN)
2089 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_IN);
2097 long unix_inq_len(struct sock *sk)
2099 struct sk_buff *skb;
2102 if (sk->sk_state == TCP_LISTEN)
2105 spin_lock(&sk->sk_receive_queue.lock);
2106 if (sk->sk_type == SOCK_STREAM ||
2107 sk->sk_type == SOCK_SEQPACKET) {
2108 skb_queue_walk(&sk->sk_receive_queue, skb)
2111 skb = skb_peek(&sk->sk_receive_queue);
2115 spin_unlock(&sk->sk_receive_queue.lock);
2119 EXPORT_SYMBOL_GPL(unix_inq_len);
2121 long unix_outq_len(struct sock *sk)
2123 return sk_wmem_alloc_get(sk);
2125 EXPORT_SYMBOL_GPL(unix_outq_len);
2127 static int unix_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2129 struct sock *sk = sock->sk;
2135 amount = unix_outq_len(sk);
2136 err = put_user(amount, (int __user *)arg);
2139 amount = unix_inq_len(sk);
2143 err = put_user(amount, (int __user *)arg);
2152 static unsigned int unix_poll(struct file *file, struct socket *sock, poll_table *wait)
2154 struct sock *sk = sock->sk;
2157 sock_poll_wait(file, sk_sleep(sk), wait);
2160 /* exceptional events? */
2163 if (sk->sk_shutdown == SHUTDOWN_MASK)
2165 if (sk->sk_shutdown & RCV_SHUTDOWN)
2166 mask |= POLLRDHUP | POLLIN | POLLRDNORM;
2169 if (!skb_queue_empty(&sk->sk_receive_queue))
2170 mask |= POLLIN | POLLRDNORM;
2172 /* Connection-based need to check for termination and startup */
2173 if ((sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) &&
2174 sk->sk_state == TCP_CLOSE)
2178 * we set writable also when the other side has shut down the
2179 * connection. This prevents stuck sockets.
2181 if (unix_writable(sk))
2182 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2187 static unsigned int unix_dgram_poll(struct file *file, struct socket *sock,
2190 struct sock *sk = sock->sk, *other;
2191 unsigned int mask, writable;
2193 sock_poll_wait(file, sk_sleep(sk), wait);
2196 /* exceptional events? */
2197 if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue))
2199 if (sk->sk_shutdown & RCV_SHUTDOWN)
2200 mask |= POLLRDHUP | POLLIN | POLLRDNORM;
2201 if (sk->sk_shutdown == SHUTDOWN_MASK)
2205 if (!skb_queue_empty(&sk->sk_receive_queue))
2206 mask |= POLLIN | POLLRDNORM;
2208 /* Connection-based need to check for termination and startup */
2209 if (sk->sk_type == SOCK_SEQPACKET) {
2210 if (sk->sk_state == TCP_CLOSE)
2212 /* connection hasn't started yet? */
2213 if (sk->sk_state == TCP_SYN_SENT)
2217 /* No write status requested, avoid expensive OUT tests. */
2218 if (!(poll_requested_events(wait) & (POLLWRBAND|POLLWRNORM|POLLOUT)))
2221 writable = unix_writable(sk);
2222 other = unix_peer_get(sk);
2224 if (unix_peer(other) != sk) {
2225 sock_poll_wait(file, &unix_sk(other)->peer_wait, wait);
2226 if (unix_recvq_full(other))
2233 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2235 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
2240 #ifdef CONFIG_PROC_FS
2241 static struct sock *first_unix_socket(int *i)
2243 for (*i = 0; *i <= UNIX_HASH_SIZE; (*i)++) {
2244 if (!hlist_empty(&unix_socket_table[*i]))
2245 return __sk_head(&unix_socket_table[*i]);
2250 static struct sock *next_unix_socket(int *i, struct sock *s)
2252 struct sock *next = sk_next(s);
2253 /* More in this chain? */
2256 /* Look for next non-empty chain. */
2257 for ((*i)++; *i <= UNIX_HASH_SIZE; (*i)++) {
2258 if (!hlist_empty(&unix_socket_table[*i]))
2259 return __sk_head(&unix_socket_table[*i]);
2264 struct unix_iter_state {
2265 struct seq_net_private p;
2269 static struct sock *unix_seq_idx(struct seq_file *seq, loff_t pos)
2271 struct unix_iter_state *iter = seq->private;
2275 for (s = first_unix_socket(&iter->i); s; s = next_unix_socket(&iter->i, s)) {
2276 if (sock_net(s) != seq_file_net(seq))
2285 static void *unix_seq_start(struct seq_file *seq, loff_t *pos)
2286 __acquires(unix_table_lock)
2288 spin_lock(&unix_table_lock);
2289 return *pos ? unix_seq_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2292 static void *unix_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2294 struct unix_iter_state *iter = seq->private;
2295 struct sock *sk = v;
2298 if (v == SEQ_START_TOKEN)
2299 sk = first_unix_socket(&iter->i);
2301 sk = next_unix_socket(&iter->i, sk);
2302 while (sk && (sock_net(sk) != seq_file_net(seq)))
2303 sk = next_unix_socket(&iter->i, sk);
2307 static void unix_seq_stop(struct seq_file *seq, void *v)
2308 __releases(unix_table_lock)
2310 spin_unlock(&unix_table_lock);
2313 static int unix_seq_show(struct seq_file *seq, void *v)
2316 if (v == SEQ_START_TOKEN)
2317 seq_puts(seq, "Num RefCount Protocol Flags Type St "
2321 struct unix_sock *u = unix_sk(s);
2324 seq_printf(seq, "%pK: %08X %08X %08X %04X %02X %5lu",
2326 atomic_read(&s->sk_refcnt),
2328 s->sk_state == TCP_LISTEN ? __SO_ACCEPTCON : 0,
2331 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTED : SS_UNCONNECTED) :
2332 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTING : SS_DISCONNECTING),
2340 len = u->addr->len - sizeof(short);
2341 if (!UNIX_ABSTRACT(s))
2347 for ( ; i < len; i++)
2348 seq_putc(seq, u->addr->name->sun_path[i]);
2350 unix_state_unlock(s);
2351 seq_putc(seq, '\n');
2357 static const struct seq_operations unix_seq_ops = {
2358 .start = unix_seq_start,
2359 .next = unix_seq_next,
2360 .stop = unix_seq_stop,
2361 .show = unix_seq_show,
2364 static int unix_seq_open(struct inode *inode, struct file *file)
2366 return seq_open_net(inode, file, &unix_seq_ops,
2367 sizeof(struct unix_iter_state));
2370 static const struct file_operations unix_seq_fops = {
2371 .owner = THIS_MODULE,
2372 .open = unix_seq_open,
2374 .llseek = seq_lseek,
2375 .release = seq_release_net,
2380 static const struct net_proto_family unix_family_ops = {
2382 .create = unix_create,
2383 .owner = THIS_MODULE,
2387 static int __net_init unix_net_init(struct net *net)
2389 int error = -ENOMEM;
2391 net->unx.sysctl_max_dgram_qlen = 10;
2392 if (unix_sysctl_register(net))
2395 #ifdef CONFIG_PROC_FS
2396 if (!proc_net_fops_create(net, "unix", 0, &unix_seq_fops)) {
2397 unix_sysctl_unregister(net);
2406 static void __net_exit unix_net_exit(struct net *net)
2408 unix_sysctl_unregister(net);
2409 proc_net_remove(net, "unix");
2412 static struct pernet_operations unix_net_ops = {
2413 .init = unix_net_init,
2414 .exit = unix_net_exit,
2417 static int __init af_unix_init(void)
2420 struct sk_buff *dummy_skb;
2422 BUILD_BUG_ON(sizeof(struct unix_skb_parms) > sizeof(dummy_skb->cb));
2424 rc = proto_register(&unix_proto, 1);
2426 printk(KERN_CRIT "%s: Cannot create unix_sock SLAB cache!\n",
2431 sock_register(&unix_family_ops);
2432 register_pernet_subsys(&unix_net_ops);
2437 static void __exit af_unix_exit(void)
2439 sock_unregister(PF_UNIX);
2440 proto_unregister(&unix_proto);
2441 unregister_pernet_subsys(&unix_net_ops);
2444 /* Earlier than device_initcall() so that other drivers invoking
2445 request_module() don't end up in a loop when modprobe tries
2446 to use a UNIX socket. But later than subsys_initcall() because
2447 we depend on stuff initialised there */
2448 fs_initcall(af_unix_init);
2449 module_exit(af_unix_exit);
2451 MODULE_LICENSE("GPL");
2452 MODULE_ALIAS_NETPROTO(PF_UNIX);