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[2 * UNIX_HASH_SIZE];
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;
125 static struct hlist_head *unix_sockets_unbound(void *addr)
127 unsigned long hash = (unsigned long)addr;
131 hash %= UNIX_HASH_SIZE;
132 return &unix_socket_table[UNIX_HASH_SIZE + hash];
135 #define UNIX_ABSTRACT(sk) (unix_sk(sk)->addr->hash < UNIX_HASH_SIZE)
137 #ifdef CONFIG_SECURITY_NETWORK
138 static void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb)
140 memcpy(UNIXSID(skb), &scm->secid, sizeof(u32));
143 static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb)
145 scm->secid = *UNIXSID(skb);
148 static inline void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb)
151 static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb)
153 #endif /* CONFIG_SECURITY_NETWORK */
156 * SMP locking strategy:
157 * hash table is protected with spinlock unix_table_lock
158 * each socket state is protected by separate spin lock.
161 static inline unsigned int unix_hash_fold(__wsum n)
163 unsigned int hash = (__force unsigned int)n;
167 return hash&(UNIX_HASH_SIZE-1);
170 #define unix_peer(sk) (unix_sk(sk)->peer)
172 static inline int unix_our_peer(struct sock *sk, struct sock *osk)
174 return unix_peer(osk) == sk;
177 static inline int unix_may_send(struct sock *sk, struct sock *osk)
179 return unix_peer(osk) == NULL || unix_our_peer(sk, osk);
182 static inline int unix_recvq_full(struct sock const *sk)
184 return skb_queue_len(&sk->sk_receive_queue) > sk->sk_max_ack_backlog;
187 struct sock *unix_peer_get(struct sock *s)
195 unix_state_unlock(s);
198 EXPORT_SYMBOL_GPL(unix_peer_get);
200 static inline void unix_release_addr(struct unix_address *addr)
202 if (atomic_dec_and_test(&addr->refcnt))
207 * Check unix socket name:
208 * - should be not zero length.
209 * - if started by not zero, should be NULL terminated (FS object)
210 * - if started by zero, it is abstract name.
213 static int unix_mkname(struct sockaddr_un *sunaddr, int len, unsigned int *hashp)
215 if (len <= sizeof(short) || len > sizeof(*sunaddr))
217 if (!sunaddr || sunaddr->sun_family != AF_UNIX)
219 if (sunaddr->sun_path[0]) {
221 * This may look like an off by one error but it is a bit more
222 * subtle. 108 is the longest valid AF_UNIX path for a binding.
223 * sun_path[108] doesn't as such exist. However in kernel space
224 * we are guaranteed that it is a valid memory location in our
225 * kernel address buffer.
227 ((char *)sunaddr)[len] = 0;
228 len = strlen(sunaddr->sun_path)+1+sizeof(short);
232 *hashp = unix_hash_fold(csum_partial(sunaddr, len, 0));
236 static void __unix_remove_socket(struct sock *sk)
238 sk_del_node_init(sk);
241 static void __unix_insert_socket(struct hlist_head *list, struct sock *sk)
243 WARN_ON(!sk_unhashed(sk));
244 sk_add_node(sk, list);
247 static inline void unix_remove_socket(struct sock *sk)
249 spin_lock(&unix_table_lock);
250 __unix_remove_socket(sk);
251 spin_unlock(&unix_table_lock);
254 static inline void unix_insert_socket(struct hlist_head *list, struct sock *sk)
256 spin_lock(&unix_table_lock);
257 __unix_insert_socket(list, sk);
258 spin_unlock(&unix_table_lock);
261 static struct sock *__unix_find_socket_byname(struct net *net,
262 struct sockaddr_un *sunname,
263 int len, int type, unsigned int hash)
266 struct hlist_node *node;
268 sk_for_each(s, node, &unix_socket_table[hash ^ type]) {
269 struct unix_sock *u = unix_sk(s);
271 if (!net_eq(sock_net(s), net))
274 if (u->addr->len == len &&
275 !memcmp(u->addr->name, sunname, len))
283 static inline struct sock *unix_find_socket_byname(struct net *net,
284 struct sockaddr_un *sunname,
290 spin_lock(&unix_table_lock);
291 s = __unix_find_socket_byname(net, sunname, len, type, hash);
294 spin_unlock(&unix_table_lock);
298 static struct sock *unix_find_socket_byinode(struct inode *i)
301 struct hlist_node *node;
303 spin_lock(&unix_table_lock);
305 &unix_socket_table[i->i_ino & (UNIX_HASH_SIZE - 1)]) {
306 struct dentry *dentry = unix_sk(s)->path.dentry;
308 if (dentry && dentry->d_inode == i) {
315 spin_unlock(&unix_table_lock);
319 static inline int unix_writable(struct sock *sk)
321 return (atomic_read(&sk->sk_wmem_alloc) << 2) <= sk->sk_sndbuf;
324 static void unix_write_space(struct sock *sk)
326 struct socket_wq *wq;
329 if (unix_writable(sk)) {
330 wq = rcu_dereference(sk->sk_wq);
331 if (wq_has_sleeper(wq))
332 wake_up_interruptible_sync_poll(&wq->wait,
333 POLLOUT | POLLWRNORM | POLLWRBAND);
334 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
339 /* When dgram socket disconnects (or changes its peer), we clear its receive
340 * queue of packets arrived from previous peer. First, it allows to do
341 * flow control based only on wmem_alloc; second, sk connected to peer
342 * may receive messages only from that peer. */
343 static void unix_dgram_disconnected(struct sock *sk, struct sock *other)
345 if (!skb_queue_empty(&sk->sk_receive_queue)) {
346 skb_queue_purge(&sk->sk_receive_queue);
347 wake_up_interruptible_all(&unix_sk(sk)->peer_wait);
349 /* If one link of bidirectional dgram pipe is disconnected,
350 * we signal error. Messages are lost. Do not make this,
351 * when peer was not connected to us.
353 if (!sock_flag(other, SOCK_DEAD) && unix_peer(other) == sk) {
354 other->sk_err = ECONNRESET;
355 other->sk_error_report(other);
360 static void unix_sock_destructor(struct sock *sk)
362 struct unix_sock *u = unix_sk(sk);
364 skb_queue_purge(&sk->sk_receive_queue);
366 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
367 WARN_ON(!sk_unhashed(sk));
368 WARN_ON(sk->sk_socket);
369 if (!sock_flag(sk, SOCK_DEAD)) {
370 printk(KERN_INFO "Attempt to release alive unix socket: %p\n", sk);
375 unix_release_addr(u->addr);
377 atomic_long_dec(&unix_nr_socks);
379 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
381 #ifdef UNIX_REFCNT_DEBUG
382 printk(KERN_DEBUG "UNIX %p is destroyed, %ld are still alive.\n", sk,
383 atomic_long_read(&unix_nr_socks));
387 static int unix_release_sock(struct sock *sk, int embrion)
389 struct unix_sock *u = unix_sk(sk);
395 unix_remove_socket(sk);
400 sk->sk_shutdown = SHUTDOWN_MASK;
402 u->path.dentry = NULL;
404 state = sk->sk_state;
405 sk->sk_state = TCP_CLOSE;
406 unix_state_unlock(sk);
408 wake_up_interruptible_all(&u->peer_wait);
410 skpair = unix_peer(sk);
412 if (skpair != NULL) {
413 if (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) {
414 unix_state_lock(skpair);
416 skpair->sk_shutdown = SHUTDOWN_MASK;
417 if (!skb_queue_empty(&sk->sk_receive_queue) || embrion)
418 skpair->sk_err = ECONNRESET;
419 unix_state_unlock(skpair);
420 skpair->sk_state_change(skpair);
421 sk_wake_async(skpair, SOCK_WAKE_WAITD, POLL_HUP);
423 sock_put(skpair); /* It may now die */
424 unix_peer(sk) = NULL;
427 /* Try to flush out this socket. Throw out buffers at least */
429 while ((skb = skb_dequeue(&sk->sk_receive_queue)) != NULL) {
430 if (state == TCP_LISTEN)
431 unix_release_sock(skb->sk, 1);
432 /* passed fds are erased in the kfree_skb hook */
441 /* ---- Socket is dead now and most probably destroyed ---- */
444 * Fixme: BSD difference: In BSD all sockets connected to use get
445 * ECONNRESET and we die on the spot. In Linux we behave
446 * like files and pipes do and wait for the last
449 * Can't we simply set sock->err?
451 * What the above comment does talk about? --ANK(980817)
454 if (unix_tot_inflight)
455 unix_gc(); /* Garbage collect fds */
460 static void init_peercred(struct sock *sk)
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(task_tgid(current));
466 sk->sk_peer_cred = get_current_cred();
469 static void copy_peercred(struct sock *sk, struct sock *peersk)
471 put_pid(sk->sk_peer_pid);
472 if (sk->sk_peer_cred)
473 put_cred(sk->sk_peer_cred);
474 sk->sk_peer_pid = get_pid(peersk->sk_peer_pid);
475 sk->sk_peer_cred = get_cred(peersk->sk_peer_cred);
478 static int unix_listen(struct socket *sock, int backlog)
481 struct sock *sk = sock->sk;
482 struct unix_sock *u = unix_sk(sk);
483 struct pid *old_pid = NULL;
484 const struct cred *old_cred = NULL;
487 if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
488 goto out; /* Only stream/seqpacket sockets accept */
491 goto out; /* No listens on an unbound socket */
493 if (sk->sk_state != TCP_CLOSE && sk->sk_state != TCP_LISTEN)
495 if (backlog > sk->sk_max_ack_backlog)
496 wake_up_interruptible_all(&u->peer_wait);
497 sk->sk_max_ack_backlog = backlog;
498 sk->sk_state = TCP_LISTEN;
499 /* set credentials so connect can copy them */
504 unix_state_unlock(sk);
512 static int unix_release(struct socket *);
513 static int unix_bind(struct socket *, struct sockaddr *, int);
514 static int unix_stream_connect(struct socket *, struct sockaddr *,
515 int addr_len, int flags);
516 static int unix_socketpair(struct socket *, struct socket *);
517 static int unix_accept(struct socket *, struct socket *, int);
518 static int unix_getname(struct socket *, struct sockaddr *, int *, int);
519 static unsigned int unix_poll(struct file *, struct socket *, poll_table *);
520 static unsigned int unix_dgram_poll(struct file *, struct socket *,
522 static int unix_ioctl(struct socket *, unsigned int, unsigned long);
523 static int unix_shutdown(struct socket *, int);
524 static int unix_stream_sendmsg(struct kiocb *, struct socket *,
525 struct msghdr *, size_t);
526 static int unix_stream_recvmsg(struct kiocb *, struct socket *,
527 struct msghdr *, size_t, int);
528 static int unix_dgram_sendmsg(struct kiocb *, struct socket *,
529 struct msghdr *, size_t);
530 static int unix_dgram_recvmsg(struct kiocb *, struct socket *,
531 struct msghdr *, size_t, int);
532 static int unix_dgram_connect(struct socket *, struct sockaddr *,
534 static int unix_seqpacket_sendmsg(struct kiocb *, struct socket *,
535 struct msghdr *, size_t);
536 static int unix_seqpacket_recvmsg(struct kiocb *, struct socket *,
537 struct msghdr *, size_t, int);
539 static void unix_set_peek_off(struct sock *sk, int val)
541 struct unix_sock *u = unix_sk(sk);
543 mutex_lock(&u->readlock);
544 sk->sk_peek_off = val;
545 mutex_unlock(&u->readlock);
549 static const struct proto_ops unix_stream_ops = {
551 .owner = THIS_MODULE,
552 .release = unix_release,
554 .connect = unix_stream_connect,
555 .socketpair = unix_socketpair,
556 .accept = unix_accept,
557 .getname = unix_getname,
560 .listen = unix_listen,
561 .shutdown = unix_shutdown,
562 .setsockopt = sock_no_setsockopt,
563 .getsockopt = sock_no_getsockopt,
564 .sendmsg = unix_stream_sendmsg,
565 .recvmsg = unix_stream_recvmsg,
566 .mmap = sock_no_mmap,
567 .sendpage = sock_no_sendpage,
568 .set_peek_off = unix_set_peek_off,
571 static const struct proto_ops unix_dgram_ops = {
573 .owner = THIS_MODULE,
574 .release = unix_release,
576 .connect = unix_dgram_connect,
577 .socketpair = unix_socketpair,
578 .accept = sock_no_accept,
579 .getname = unix_getname,
580 .poll = unix_dgram_poll,
582 .listen = sock_no_listen,
583 .shutdown = unix_shutdown,
584 .setsockopt = sock_no_setsockopt,
585 .getsockopt = sock_no_getsockopt,
586 .sendmsg = unix_dgram_sendmsg,
587 .recvmsg = unix_dgram_recvmsg,
588 .mmap = sock_no_mmap,
589 .sendpage = sock_no_sendpage,
590 .set_peek_off = unix_set_peek_off,
593 static const struct proto_ops unix_seqpacket_ops = {
595 .owner = THIS_MODULE,
596 .release = unix_release,
598 .connect = unix_stream_connect,
599 .socketpair = unix_socketpair,
600 .accept = unix_accept,
601 .getname = unix_getname,
602 .poll = unix_dgram_poll,
604 .listen = unix_listen,
605 .shutdown = unix_shutdown,
606 .setsockopt = sock_no_setsockopt,
607 .getsockopt = sock_no_getsockopt,
608 .sendmsg = unix_seqpacket_sendmsg,
609 .recvmsg = unix_seqpacket_recvmsg,
610 .mmap = sock_no_mmap,
611 .sendpage = sock_no_sendpage,
612 .set_peek_off = unix_set_peek_off,
615 static struct proto unix_proto = {
617 .owner = THIS_MODULE,
618 .obj_size = sizeof(struct unix_sock),
622 * AF_UNIX sockets do not interact with hardware, hence they
623 * dont trigger interrupts - so it's safe for them to have
624 * bh-unsafe locking for their sk_receive_queue.lock. Split off
625 * this special lock-class by reinitializing the spinlock key:
627 static struct lock_class_key af_unix_sk_receive_queue_lock_key;
629 static struct sock *unix_create1(struct net *net, struct socket *sock)
631 struct sock *sk = NULL;
634 atomic_long_inc(&unix_nr_socks);
635 if (atomic_long_read(&unix_nr_socks) > 2 * get_max_files())
638 sk = sk_alloc(net, PF_UNIX, GFP_KERNEL, &unix_proto);
642 sock_init_data(sock, sk);
643 lockdep_set_class(&sk->sk_receive_queue.lock,
644 &af_unix_sk_receive_queue_lock_key);
646 sk->sk_write_space = unix_write_space;
647 sk->sk_max_ack_backlog = net->unx.sysctl_max_dgram_qlen;
648 sk->sk_destruct = unix_sock_destructor;
650 u->path.dentry = NULL;
652 spin_lock_init(&u->lock);
653 atomic_long_set(&u->inflight, 0);
654 INIT_LIST_HEAD(&u->link);
655 mutex_init(&u->readlock); /* single task reading lock */
656 init_waitqueue_head(&u->peer_wait);
657 unix_insert_socket(unix_sockets_unbound(sk), sk);
660 atomic_long_dec(&unix_nr_socks);
663 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
669 static int unix_create(struct net *net, struct socket *sock, int protocol,
672 if (protocol && protocol != PF_UNIX)
673 return -EPROTONOSUPPORT;
675 sock->state = SS_UNCONNECTED;
677 switch (sock->type) {
679 sock->ops = &unix_stream_ops;
682 * Believe it or not BSD has AF_UNIX, SOCK_RAW though
686 sock->type = SOCK_DGRAM;
688 sock->ops = &unix_dgram_ops;
691 sock->ops = &unix_seqpacket_ops;
694 return -ESOCKTNOSUPPORT;
697 return unix_create1(net, sock) ? 0 : -ENOMEM;
700 static int unix_release(struct socket *sock)
702 struct sock *sk = sock->sk;
709 return unix_release_sock(sk, 0);
712 static int unix_autobind(struct socket *sock)
714 struct sock *sk = sock->sk;
715 struct net *net = sock_net(sk);
716 struct unix_sock *u = unix_sk(sk);
717 static u32 ordernum = 1;
718 struct unix_address *addr;
720 unsigned int retries = 0;
722 mutex_lock(&u->readlock);
729 addr = kzalloc(sizeof(*addr) + sizeof(short) + 16, GFP_KERNEL);
733 addr->name->sun_family = AF_UNIX;
734 atomic_set(&addr->refcnt, 1);
737 addr->len = sprintf(addr->name->sun_path+1, "%05x", ordernum) + 1 + sizeof(short);
738 addr->hash = unix_hash_fold(csum_partial(addr->name, addr->len, 0));
740 spin_lock(&unix_table_lock);
741 ordernum = (ordernum+1)&0xFFFFF;
743 if (__unix_find_socket_byname(net, addr->name, addr->len, sock->type,
745 spin_unlock(&unix_table_lock);
747 * __unix_find_socket_byname() may take long time if many names
748 * are already in use.
751 /* Give up if all names seems to be in use. */
752 if (retries++ == 0xFFFFF) {
759 addr->hash ^= sk->sk_type;
761 __unix_remove_socket(sk);
763 __unix_insert_socket(&unix_socket_table[addr->hash], sk);
764 spin_unlock(&unix_table_lock);
767 out: mutex_unlock(&u->readlock);
771 static struct sock *unix_find_other(struct net *net,
772 struct sockaddr_un *sunname, int len,
773 int type, unsigned int hash, int *error)
779 if (sunname->sun_path[0]) {
781 err = kern_path(sunname->sun_path, LOOKUP_FOLLOW, &path);
784 inode = path.dentry->d_inode;
785 err = inode_permission(inode, MAY_WRITE);
790 if (!S_ISSOCK(inode->i_mode))
792 u = unix_find_socket_byinode(inode);
796 if (u->sk_type == type)
802 if (u->sk_type != type) {
808 u = unix_find_socket_byname(net, sunname, len, type, hash);
810 struct dentry *dentry;
811 dentry = unix_sk(u)->path.dentry;
813 touch_atime(&unix_sk(u)->path);
827 static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
829 struct sock *sk = sock->sk;
830 struct net *net = sock_net(sk);
831 struct unix_sock *u = unix_sk(sk);
832 struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
833 char *sun_path = sunaddr->sun_path;
834 struct dentry *dentry = NULL;
838 struct unix_address *addr;
839 struct hlist_head *list;
842 if (sunaddr->sun_family != AF_UNIX)
845 if (addr_len == sizeof(short)) {
846 err = unix_autobind(sock);
850 err = unix_mkname(sunaddr, addr_len, &hash);
855 mutex_lock(&u->readlock);
862 addr = kmalloc(sizeof(*addr)+addr_len, GFP_KERNEL);
866 memcpy(addr->name, sunaddr, addr_len);
867 addr->len = addr_len;
868 addr->hash = hash ^ sk->sk_type;
869 atomic_set(&addr->refcnt, 1);
875 * Get the parent directory, calculate the hash for last
878 dentry = kern_path_create(AT_FDCWD, sun_path, &path, 0);
879 err = PTR_ERR(dentry);
881 goto out_mknod_parent;
884 * All right, let's create it.
887 (SOCK_INODE(sock)->i_mode & ~current_umask());
888 err = mnt_want_write(path.mnt);
891 err = security_path_mknod(&path, dentry, mode, 0);
893 goto out_mknod_drop_write;
894 err = vfs_mknod(path.dentry->d_inode, dentry, mode, 0);
895 out_mknod_drop_write:
896 mnt_drop_write(path.mnt);
899 mutex_unlock(&path.dentry->d_inode->i_mutex);
901 path.dentry = dentry;
903 addr->hash = UNIX_HASH_SIZE;
906 spin_lock(&unix_table_lock);
910 if (__unix_find_socket_byname(net, sunaddr, addr_len,
911 sk->sk_type, hash)) {
912 unix_release_addr(addr);
916 list = &unix_socket_table[addr->hash];
918 list = &unix_socket_table[dentry->d_inode->i_ino & (UNIX_HASH_SIZE-1)];
923 __unix_remove_socket(sk);
925 __unix_insert_socket(list, sk);
928 spin_unlock(&unix_table_lock);
930 mutex_unlock(&u->readlock);
936 mutex_unlock(&path.dentry->d_inode->i_mutex);
941 unix_release_addr(addr);
945 static void unix_state_double_lock(struct sock *sk1, struct sock *sk2)
947 if (unlikely(sk1 == sk2) || !sk2) {
948 unix_state_lock(sk1);
952 unix_state_lock(sk1);
953 unix_state_lock_nested(sk2);
955 unix_state_lock(sk2);
956 unix_state_lock_nested(sk1);
960 static void unix_state_double_unlock(struct sock *sk1, struct sock *sk2)
962 if (unlikely(sk1 == sk2) || !sk2) {
963 unix_state_unlock(sk1);
966 unix_state_unlock(sk1);
967 unix_state_unlock(sk2);
970 static int unix_dgram_connect(struct socket *sock, struct sockaddr *addr,
973 struct sock *sk = sock->sk;
974 struct net *net = sock_net(sk);
975 struct sockaddr_un *sunaddr = (struct sockaddr_un *)addr;
980 if (addr->sa_family != AF_UNSPEC) {
981 err = unix_mkname(sunaddr, alen, &hash);
986 if (test_bit(SOCK_PASSCRED, &sock->flags) &&
987 !unix_sk(sk)->addr && (err = unix_autobind(sock)) != 0)
991 other = unix_find_other(net, sunaddr, alen, sock->type, hash, &err);
995 unix_state_double_lock(sk, other);
997 /* Apparently VFS overslept socket death. Retry. */
998 if (sock_flag(other, SOCK_DEAD)) {
999 unix_state_double_unlock(sk, other);
1005 if (!unix_may_send(sk, other))
1008 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1014 * 1003.1g breaking connected state with AF_UNSPEC
1017 unix_state_double_lock(sk, other);
1021 * If it was connected, reconnect.
1023 if (unix_peer(sk)) {
1024 struct sock *old_peer = unix_peer(sk);
1025 unix_peer(sk) = other;
1026 unix_state_double_unlock(sk, other);
1028 if (other != old_peer)
1029 unix_dgram_disconnected(sk, old_peer);
1032 unix_peer(sk) = other;
1033 unix_state_double_unlock(sk, other);
1038 unix_state_double_unlock(sk, other);
1044 static long unix_wait_for_peer(struct sock *other, long timeo)
1046 struct unix_sock *u = unix_sk(other);
1050 prepare_to_wait_exclusive(&u->peer_wait, &wait, TASK_INTERRUPTIBLE);
1052 sched = !sock_flag(other, SOCK_DEAD) &&
1053 !(other->sk_shutdown & RCV_SHUTDOWN) &&
1054 unix_recvq_full(other);
1056 unix_state_unlock(other);
1059 timeo = schedule_timeout(timeo);
1061 finish_wait(&u->peer_wait, &wait);
1065 static int unix_stream_connect(struct socket *sock, struct sockaddr *uaddr,
1066 int addr_len, int flags)
1068 struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
1069 struct sock *sk = sock->sk;
1070 struct net *net = sock_net(sk);
1071 struct unix_sock *u = unix_sk(sk), *newu, *otheru;
1072 struct sock *newsk = NULL;
1073 struct sock *other = NULL;
1074 struct sk_buff *skb = NULL;
1080 err = unix_mkname(sunaddr, addr_len, &hash);
1085 if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr &&
1086 (err = unix_autobind(sock)) != 0)
1089 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
1091 /* First of all allocate resources.
1092 If we will make it after state is locked,
1093 we will have to recheck all again in any case.
1098 /* create new sock for complete connection */
1099 newsk = unix_create1(sock_net(sk), NULL);
1103 /* Allocate skb for sending to listening sock */
1104 skb = sock_wmalloc(newsk, 1, 0, GFP_KERNEL);
1109 /* Find listening sock. */
1110 other = unix_find_other(net, sunaddr, addr_len, sk->sk_type, hash, &err);
1114 /* Latch state of peer */
1115 unix_state_lock(other);
1117 /* Apparently VFS overslept socket death. Retry. */
1118 if (sock_flag(other, SOCK_DEAD)) {
1119 unix_state_unlock(other);
1124 err = -ECONNREFUSED;
1125 if (other->sk_state != TCP_LISTEN)
1127 if (other->sk_shutdown & RCV_SHUTDOWN)
1130 if (unix_recvq_full(other)) {
1135 timeo = unix_wait_for_peer(other, timeo);
1137 err = sock_intr_errno(timeo);
1138 if (signal_pending(current))
1146 It is tricky place. We need to grab our state lock and cannot
1147 drop lock on peer. It is dangerous because deadlock is
1148 possible. Connect to self case and simultaneous
1149 attempt to connect are eliminated by checking socket
1150 state. other is TCP_LISTEN, if sk is TCP_LISTEN we
1151 check this before attempt to grab lock.
1153 Well, and we have to recheck the state after socket locked.
1159 /* This is ok... continue with connect */
1161 case TCP_ESTABLISHED:
1162 /* Socket is already connected */
1170 unix_state_lock_nested(sk);
1172 if (sk->sk_state != st) {
1173 unix_state_unlock(sk);
1174 unix_state_unlock(other);
1179 err = security_unix_stream_connect(sk, other, newsk);
1181 unix_state_unlock(sk);
1185 /* The way is open! Fastly set all the necessary fields... */
1188 unix_peer(newsk) = sk;
1189 newsk->sk_state = TCP_ESTABLISHED;
1190 newsk->sk_type = sk->sk_type;
1191 init_peercred(newsk);
1192 newu = unix_sk(newsk);
1193 RCU_INIT_POINTER(newsk->sk_wq, &newu->peer_wq);
1194 otheru = unix_sk(other);
1196 /* copy address information from listening to new sock*/
1198 atomic_inc(&otheru->addr->refcnt);
1199 newu->addr = otheru->addr;
1201 if (otheru->path.dentry) {
1202 path_get(&otheru->path);
1203 newu->path = otheru->path;
1206 /* Set credentials */
1207 copy_peercred(sk, other);
1209 sock->state = SS_CONNECTED;
1210 sk->sk_state = TCP_ESTABLISHED;
1213 smp_mb__after_atomic_inc(); /* sock_hold() does an atomic_inc() */
1214 unix_peer(sk) = newsk;
1216 unix_state_unlock(sk);
1218 /* take ten and and send info to listening sock */
1219 spin_lock(&other->sk_receive_queue.lock);
1220 __skb_queue_tail(&other->sk_receive_queue, skb);
1221 spin_unlock(&other->sk_receive_queue.lock);
1222 unix_state_unlock(other);
1223 other->sk_data_ready(other, 0);
1229 unix_state_unlock(other);
1234 unix_release_sock(newsk, 0);
1240 static int unix_socketpair(struct socket *socka, struct socket *sockb)
1242 struct sock *ska = socka->sk, *skb = sockb->sk;
1244 /* Join our sockets back to back */
1247 unix_peer(ska) = skb;
1248 unix_peer(skb) = ska;
1252 if (ska->sk_type != SOCK_DGRAM) {
1253 ska->sk_state = TCP_ESTABLISHED;
1254 skb->sk_state = TCP_ESTABLISHED;
1255 socka->state = SS_CONNECTED;
1256 sockb->state = SS_CONNECTED;
1261 static int unix_accept(struct socket *sock, struct socket *newsock, int flags)
1263 struct sock *sk = sock->sk;
1265 struct sk_buff *skb;
1269 if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
1273 if (sk->sk_state != TCP_LISTEN)
1276 /* If socket state is TCP_LISTEN it cannot change (for now...),
1277 * so that no locks are necessary.
1280 skb = skb_recv_datagram(sk, 0, flags&O_NONBLOCK, &err);
1282 /* This means receive shutdown. */
1289 skb_free_datagram(sk, skb);
1290 wake_up_interruptible(&unix_sk(sk)->peer_wait);
1292 /* attach accepted sock to socket */
1293 unix_state_lock(tsk);
1294 newsock->state = SS_CONNECTED;
1295 sock_graft(tsk, newsock);
1296 unix_state_unlock(tsk);
1304 static int unix_getname(struct socket *sock, struct sockaddr *uaddr, int *uaddr_len, int peer)
1306 struct sock *sk = sock->sk;
1307 struct unix_sock *u;
1308 DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr, uaddr);
1312 sk = unix_peer_get(sk);
1323 unix_state_lock(sk);
1325 sunaddr->sun_family = AF_UNIX;
1326 sunaddr->sun_path[0] = 0;
1327 *uaddr_len = sizeof(short);
1329 struct unix_address *addr = u->addr;
1331 *uaddr_len = addr->len;
1332 memcpy(sunaddr, addr->name, *uaddr_len);
1334 unix_state_unlock(sk);
1340 static void unix_detach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1344 scm->fp = UNIXCB(skb).fp;
1345 UNIXCB(skb).fp = NULL;
1347 for (i = scm->fp->count-1; i >= 0; i--)
1348 unix_notinflight(scm->fp->fp[i]);
1351 static void unix_destruct_scm(struct sk_buff *skb)
1353 struct scm_cookie scm;
1354 memset(&scm, 0, sizeof(scm));
1355 scm.pid = UNIXCB(skb).pid;
1356 scm.cred = UNIXCB(skb).cred;
1358 unix_detach_fds(&scm, skb);
1360 /* Alas, it calls VFS */
1361 /* So fscking what? fput() had been SMP-safe since the last Summer */
1366 #define MAX_RECURSION_LEVEL 4
1368 static int unix_attach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1371 unsigned char max_level = 0;
1372 int unix_sock_count = 0;
1374 for (i = scm->fp->count - 1; i >= 0; i--) {
1375 struct sock *sk = unix_get_socket(scm->fp->fp[i]);
1379 max_level = max(max_level,
1380 unix_sk(sk)->recursion_level);
1383 if (unlikely(max_level > MAX_RECURSION_LEVEL))
1384 return -ETOOMANYREFS;
1387 * Need to duplicate file references for the sake of garbage
1388 * collection. Otherwise a socket in the fps might become a
1389 * candidate for GC while the skb is not yet queued.
1391 UNIXCB(skb).fp = scm_fp_dup(scm->fp);
1392 if (!UNIXCB(skb).fp)
1395 if (unix_sock_count) {
1396 for (i = scm->fp->count - 1; i >= 0; i--)
1397 unix_inflight(scm->fp->fp[i]);
1402 static int unix_scm_to_skb(struct scm_cookie *scm, struct sk_buff *skb, bool send_fds)
1406 UNIXCB(skb).pid = get_pid(scm->pid);
1408 UNIXCB(skb).cred = get_cred(scm->cred);
1409 UNIXCB(skb).fp = NULL;
1410 if (scm->fp && send_fds)
1411 err = unix_attach_fds(scm, skb);
1413 skb->destructor = unix_destruct_scm;
1418 * Some apps rely on write() giving SCM_CREDENTIALS
1419 * We include credentials if source or destination socket
1420 * asserted SOCK_PASSCRED.
1422 static void maybe_add_creds(struct sk_buff *skb, const struct socket *sock,
1423 const struct sock *other)
1425 if (UNIXCB(skb).cred)
1427 if (test_bit(SOCK_PASSCRED, &sock->flags) ||
1428 !other->sk_socket ||
1429 test_bit(SOCK_PASSCRED, &other->sk_socket->flags)) {
1430 UNIXCB(skb).pid = get_pid(task_tgid(current));
1431 UNIXCB(skb).cred = get_current_cred();
1436 * Send AF_UNIX data.
1439 static int unix_dgram_sendmsg(struct kiocb *kiocb, struct socket *sock,
1440 struct msghdr *msg, size_t len)
1442 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1443 struct sock *sk = sock->sk;
1444 struct net *net = sock_net(sk);
1445 struct unix_sock *u = unix_sk(sk);
1446 struct sockaddr_un *sunaddr = msg->msg_name;
1447 struct sock *other = NULL;
1448 int namelen = 0; /* fake GCC */
1451 struct sk_buff *skb;
1453 struct scm_cookie tmp_scm;
1457 if (NULL == siocb->scm)
1458 siocb->scm = &tmp_scm;
1460 err = scm_send(sock, msg, siocb->scm);
1465 if (msg->msg_flags&MSG_OOB)
1468 if (msg->msg_namelen) {
1469 err = unix_mkname(sunaddr, msg->msg_namelen, &hash);
1476 other = unix_peer_get(sk);
1481 if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr
1482 && (err = unix_autobind(sock)) != 0)
1486 if (len > sk->sk_sndbuf - 32)
1489 if (len > SKB_MAX_ALLOC)
1490 data_len = min_t(size_t,
1491 len - SKB_MAX_ALLOC,
1492 MAX_SKB_FRAGS * PAGE_SIZE);
1494 skb = sock_alloc_send_pskb(sk, len - data_len, data_len,
1495 msg->msg_flags & MSG_DONTWAIT, &err);
1499 err = unix_scm_to_skb(siocb->scm, skb, true);
1502 max_level = err + 1;
1503 unix_get_secdata(siocb->scm, skb);
1505 skb_put(skb, len - data_len);
1506 skb->data_len = data_len;
1508 err = skb_copy_datagram_from_iovec(skb, 0, msg->msg_iov, 0, len);
1512 timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1517 if (sunaddr == NULL)
1520 other = unix_find_other(net, sunaddr, namelen, sk->sk_type,
1526 if (sk_filter(other, skb) < 0) {
1527 /* Toss the packet but do not return any error to the sender */
1532 unix_state_lock(other);
1534 if (!unix_may_send(sk, other))
1537 if (sock_flag(other, SOCK_DEAD)) {
1539 * Check with 1003.1g - what should
1542 unix_state_unlock(other);
1546 unix_state_lock(sk);
1547 if (unix_peer(sk) == other) {
1548 unix_peer(sk) = NULL;
1549 unix_state_unlock(sk);
1551 unix_dgram_disconnected(sk, other);
1553 err = -ECONNREFUSED;
1555 unix_state_unlock(sk);
1565 if (other->sk_shutdown & RCV_SHUTDOWN)
1568 if (sk->sk_type != SOCK_SEQPACKET) {
1569 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1574 if (unix_peer(other) != sk && unix_recvq_full(other)) {
1580 timeo = unix_wait_for_peer(other, timeo);
1582 err = sock_intr_errno(timeo);
1583 if (signal_pending(current))
1589 if (sock_flag(other, SOCK_RCVTSTAMP))
1590 __net_timestamp(skb);
1591 maybe_add_creds(skb, sock, other);
1592 skb_queue_tail(&other->sk_receive_queue, skb);
1593 if (max_level > unix_sk(other)->recursion_level)
1594 unix_sk(other)->recursion_level = max_level;
1595 unix_state_unlock(other);
1596 other->sk_data_ready(other, len);
1598 scm_destroy(siocb->scm);
1602 unix_state_unlock(other);
1608 scm_destroy(siocb->scm);
1613 static int unix_stream_sendmsg(struct kiocb *kiocb, struct socket *sock,
1614 struct msghdr *msg, size_t len)
1616 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1617 struct sock *sk = sock->sk;
1618 struct sock *other = NULL;
1620 struct sk_buff *skb;
1622 struct scm_cookie tmp_scm;
1623 bool fds_sent = false;
1626 if (NULL == siocb->scm)
1627 siocb->scm = &tmp_scm;
1629 err = scm_send(sock, msg, siocb->scm);
1634 if (msg->msg_flags&MSG_OOB)
1637 if (msg->msg_namelen) {
1638 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
1642 other = unix_peer(sk);
1647 if (sk->sk_shutdown & SEND_SHUTDOWN)
1650 while (sent < len) {
1652 * Optimisation for the fact that under 0.01% of X
1653 * messages typically need breaking up.
1658 /* Keep two messages in the pipe so it schedules better */
1659 if (size > ((sk->sk_sndbuf >> 1) - 64))
1660 size = (sk->sk_sndbuf >> 1) - 64;
1662 if (size > SKB_MAX_ALLOC)
1663 size = SKB_MAX_ALLOC;
1669 skb = sock_alloc_send_skb(sk, size, msg->msg_flags&MSG_DONTWAIT,
1676 * If you pass two values to the sock_alloc_send_skb
1677 * it tries to grab the large buffer with GFP_NOFS
1678 * (which can fail easily), and if it fails grab the
1679 * fallback size buffer which is under a page and will
1682 size = min_t(int, size, skb_tailroom(skb));
1685 /* Only send the fds in the first buffer */
1686 err = unix_scm_to_skb(siocb->scm, skb, !fds_sent);
1691 max_level = err + 1;
1694 err = memcpy_fromiovec(skb_put(skb, size), msg->msg_iov, size);
1700 unix_state_lock(other);
1702 if (sock_flag(other, SOCK_DEAD) ||
1703 (other->sk_shutdown & RCV_SHUTDOWN))
1706 maybe_add_creds(skb, sock, other);
1707 skb_queue_tail(&other->sk_receive_queue, skb);
1708 if (max_level > unix_sk(other)->recursion_level)
1709 unix_sk(other)->recursion_level = max_level;
1710 unix_state_unlock(other);
1711 other->sk_data_ready(other, size);
1715 scm_destroy(siocb->scm);
1721 unix_state_unlock(other);
1724 if (sent == 0 && !(msg->msg_flags&MSG_NOSIGNAL))
1725 send_sig(SIGPIPE, current, 0);
1728 scm_destroy(siocb->scm);
1730 return sent ? : err;
1733 static int unix_seqpacket_sendmsg(struct kiocb *kiocb, struct socket *sock,
1734 struct msghdr *msg, size_t len)
1737 struct sock *sk = sock->sk;
1739 err = sock_error(sk);
1743 if (sk->sk_state != TCP_ESTABLISHED)
1746 if (msg->msg_namelen)
1747 msg->msg_namelen = 0;
1749 return unix_dgram_sendmsg(kiocb, sock, msg, len);
1752 static int unix_seqpacket_recvmsg(struct kiocb *iocb, struct socket *sock,
1753 struct msghdr *msg, size_t size,
1756 struct sock *sk = sock->sk;
1758 if (sk->sk_state != TCP_ESTABLISHED)
1761 return unix_dgram_recvmsg(iocb, sock, msg, size, flags);
1764 static void unix_copy_addr(struct msghdr *msg, struct sock *sk)
1766 struct unix_sock *u = unix_sk(sk);
1768 msg->msg_namelen = 0;
1770 msg->msg_namelen = u->addr->len;
1771 memcpy(msg->msg_name, u->addr->name, u->addr->len);
1775 static int unix_dgram_recvmsg(struct kiocb *iocb, struct socket *sock,
1776 struct msghdr *msg, size_t size,
1779 struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1780 struct scm_cookie tmp_scm;
1781 struct sock *sk = sock->sk;
1782 struct unix_sock *u = unix_sk(sk);
1783 int noblock = flags & MSG_DONTWAIT;
1784 struct sk_buff *skb;
1792 msg->msg_namelen = 0;
1794 err = mutex_lock_interruptible(&u->readlock);
1796 err = sock_intr_errno(sock_rcvtimeo(sk, noblock));
1800 skip = sk_peek_offset(sk, flags);
1802 skb = __skb_recv_datagram(sk, flags, &peeked, &skip, &err);
1804 unix_state_lock(sk);
1805 /* Signal EOF on disconnected non-blocking SEQPACKET socket. */
1806 if (sk->sk_type == SOCK_SEQPACKET && err == -EAGAIN &&
1807 (sk->sk_shutdown & RCV_SHUTDOWN))
1809 unix_state_unlock(sk);
1813 wake_up_interruptible_sync_poll(&u->peer_wait,
1814 POLLOUT | POLLWRNORM | POLLWRBAND);
1817 unix_copy_addr(msg, skb->sk);
1819 if (size > skb->len - skip)
1820 size = skb->len - skip;
1821 else if (size < skb->len - skip)
1822 msg->msg_flags |= MSG_TRUNC;
1824 err = skb_copy_datagram_iovec(skb, skip, msg->msg_iov, size);
1828 if (sock_flag(sk, SOCK_RCVTSTAMP))
1829 __sock_recv_timestamp(msg, sk, skb);
1832 siocb->scm = &tmp_scm;
1833 memset(&tmp_scm, 0, sizeof(tmp_scm));
1835 scm_set_cred(siocb->scm, UNIXCB(skb).pid, UNIXCB(skb).cred);
1836 unix_set_secdata(siocb->scm, skb);
1838 if (!(flags & MSG_PEEK)) {
1840 unix_detach_fds(siocb->scm, skb);
1842 sk_peek_offset_bwd(sk, skb->len);
1844 /* It is questionable: on PEEK we could:
1845 - do not return fds - good, but too simple 8)
1846 - return fds, and do not return them on read (old strategy,
1848 - clone fds (I chose it for now, it is the most universal
1851 POSIX 1003.1g does not actually define this clearly
1852 at all. POSIX 1003.1g doesn't define a lot of things
1857 sk_peek_offset_fwd(sk, size);
1860 siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1862 err = (flags & MSG_TRUNC) ? skb->len - skip : size;
1864 scm_recv(sock, msg, siocb->scm, flags);
1867 skb_free_datagram(sk, skb);
1869 mutex_unlock(&u->readlock);
1875 * Sleep until data has arrive. But check for races..
1878 static long unix_stream_data_wait(struct sock *sk, long timeo)
1882 unix_state_lock(sk);
1885 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1887 if (!skb_queue_empty(&sk->sk_receive_queue) ||
1889 (sk->sk_shutdown & RCV_SHUTDOWN) ||
1890 signal_pending(current) ||
1894 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1895 unix_state_unlock(sk);
1896 timeo = schedule_timeout(timeo);
1897 unix_state_lock(sk);
1898 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1901 finish_wait(sk_sleep(sk), &wait);
1902 unix_state_unlock(sk);
1908 static int unix_stream_recvmsg(struct kiocb *iocb, struct socket *sock,
1909 struct msghdr *msg, size_t size,
1912 struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1913 struct scm_cookie tmp_scm;
1914 struct sock *sk = sock->sk;
1915 struct unix_sock *u = unix_sk(sk);
1916 struct sockaddr_un *sunaddr = msg->msg_name;
1918 int check_creds = 0;
1925 if (sk->sk_state != TCP_ESTABLISHED)
1932 target = sock_rcvlowat(sk, flags&MSG_WAITALL, size);
1933 timeo = sock_rcvtimeo(sk, flags&MSG_DONTWAIT);
1935 msg->msg_namelen = 0;
1937 /* Lock the socket to prevent queue disordering
1938 * while sleeps in memcpy_tomsg
1942 siocb->scm = &tmp_scm;
1943 memset(&tmp_scm, 0, sizeof(tmp_scm));
1946 err = mutex_lock_interruptible(&u->readlock);
1948 err = sock_intr_errno(timeo);
1952 skip = sk_peek_offset(sk, flags);
1956 struct sk_buff *skb;
1958 unix_state_lock(sk);
1959 skb = skb_peek(&sk->sk_receive_queue);
1962 unix_sk(sk)->recursion_level = 0;
1963 if (copied >= target)
1967 * POSIX 1003.1g mandates this order.
1970 err = sock_error(sk);
1973 if (sk->sk_shutdown & RCV_SHUTDOWN)
1976 unix_state_unlock(sk);
1980 mutex_unlock(&u->readlock);
1982 timeo = unix_stream_data_wait(sk, timeo);
1984 if (signal_pending(current)
1985 || mutex_lock_interruptible(&u->readlock)) {
1986 err = sock_intr_errno(timeo);
1992 unix_state_unlock(sk);
1996 if (skip >= skb->len) {
1998 skb = skb_peek_next(skb, &sk->sk_receive_queue);
2002 unix_state_unlock(sk);
2005 /* Never glue messages from different writers */
2006 if ((UNIXCB(skb).pid != siocb->scm->pid) ||
2007 (UNIXCB(skb).cred != siocb->scm->cred))
2010 /* Copy credentials */
2011 scm_set_cred(siocb->scm, UNIXCB(skb).pid, UNIXCB(skb).cred);
2015 /* Copy address just once */
2017 unix_copy_addr(msg, skb->sk);
2021 chunk = min_t(unsigned int, skb->len - skip, size);
2022 if (memcpy_toiovec(msg->msg_iov, skb->data + skip, chunk)) {
2030 /* Mark read part of skb as used */
2031 if (!(flags & MSG_PEEK)) {
2032 skb_pull(skb, chunk);
2034 sk_peek_offset_bwd(sk, chunk);
2037 unix_detach_fds(siocb->scm, skb);
2042 skb_unlink(skb, &sk->sk_receive_queue);
2048 /* It is questionable, see note in unix_dgram_recvmsg.
2051 siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
2053 sk_peek_offset_fwd(sk, chunk);
2059 mutex_unlock(&u->readlock);
2060 scm_recv(sock, msg, siocb->scm, flags);
2062 return copied ? : err;
2065 static int unix_shutdown(struct socket *sock, int mode)
2067 struct sock *sk = sock->sk;
2070 mode = (mode+1)&(RCV_SHUTDOWN|SEND_SHUTDOWN);
2075 unix_state_lock(sk);
2076 sk->sk_shutdown |= mode;
2077 other = unix_peer(sk);
2080 unix_state_unlock(sk);
2081 sk->sk_state_change(sk);
2084 (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET)) {
2088 if (mode&RCV_SHUTDOWN)
2089 peer_mode |= SEND_SHUTDOWN;
2090 if (mode&SEND_SHUTDOWN)
2091 peer_mode |= RCV_SHUTDOWN;
2092 unix_state_lock(other);
2093 other->sk_shutdown |= peer_mode;
2094 unix_state_unlock(other);
2095 other->sk_state_change(other);
2096 if (peer_mode == SHUTDOWN_MASK)
2097 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_HUP);
2098 else if (peer_mode & RCV_SHUTDOWN)
2099 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_IN);
2107 long unix_inq_len(struct sock *sk)
2109 struct sk_buff *skb;
2112 if (sk->sk_state == TCP_LISTEN)
2115 spin_lock(&sk->sk_receive_queue.lock);
2116 if (sk->sk_type == SOCK_STREAM ||
2117 sk->sk_type == SOCK_SEQPACKET) {
2118 skb_queue_walk(&sk->sk_receive_queue, skb)
2121 skb = skb_peek(&sk->sk_receive_queue);
2125 spin_unlock(&sk->sk_receive_queue.lock);
2129 EXPORT_SYMBOL_GPL(unix_inq_len);
2131 long unix_outq_len(struct sock *sk)
2133 return sk_wmem_alloc_get(sk);
2135 EXPORT_SYMBOL_GPL(unix_outq_len);
2137 static int unix_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2139 struct sock *sk = sock->sk;
2145 amount = unix_outq_len(sk);
2146 err = put_user(amount, (int __user *)arg);
2149 amount = unix_inq_len(sk);
2153 err = put_user(amount, (int __user *)arg);
2162 static unsigned int unix_poll(struct file *file, struct socket *sock, poll_table *wait)
2164 struct sock *sk = sock->sk;
2167 sock_poll_wait(file, sk_sleep(sk), wait);
2170 /* exceptional events? */
2173 if (sk->sk_shutdown == SHUTDOWN_MASK)
2175 if (sk->sk_shutdown & RCV_SHUTDOWN)
2176 mask |= POLLRDHUP | POLLIN | POLLRDNORM;
2179 if (!skb_queue_empty(&sk->sk_receive_queue))
2180 mask |= POLLIN | POLLRDNORM;
2182 /* Connection-based need to check for termination and startup */
2183 if ((sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) &&
2184 sk->sk_state == TCP_CLOSE)
2188 * we set writable also when the other side has shut down the
2189 * connection. This prevents stuck sockets.
2191 if (unix_writable(sk))
2192 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2197 static unsigned int unix_dgram_poll(struct file *file, struct socket *sock,
2200 struct sock *sk = sock->sk, *other;
2201 unsigned int mask, writable;
2203 sock_poll_wait(file, sk_sleep(sk), wait);
2206 /* exceptional events? */
2207 if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue))
2209 if (sk->sk_shutdown & RCV_SHUTDOWN)
2210 mask |= POLLRDHUP | POLLIN | POLLRDNORM;
2211 if (sk->sk_shutdown == SHUTDOWN_MASK)
2215 if (!skb_queue_empty(&sk->sk_receive_queue))
2216 mask |= POLLIN | POLLRDNORM;
2218 /* Connection-based need to check for termination and startup */
2219 if (sk->sk_type == SOCK_SEQPACKET) {
2220 if (sk->sk_state == TCP_CLOSE)
2222 /* connection hasn't started yet? */
2223 if (sk->sk_state == TCP_SYN_SENT)
2227 /* No write status requested, avoid expensive OUT tests. */
2228 if (!(poll_requested_events(wait) & (POLLWRBAND|POLLWRNORM|POLLOUT)))
2231 writable = unix_writable(sk);
2232 other = unix_peer_get(sk);
2234 if (unix_peer(other) != sk) {
2235 sock_poll_wait(file, &unix_sk(other)->peer_wait, wait);
2236 if (unix_recvq_full(other))
2243 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2245 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
2250 #ifdef CONFIG_PROC_FS
2252 #define BUCKET_SPACE (BITS_PER_LONG - (UNIX_HASH_BITS + 1) - 1)
2254 #define get_bucket(x) ((x) >> BUCKET_SPACE)
2255 #define get_offset(x) ((x) & ((1L << BUCKET_SPACE) - 1))
2256 #define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
2258 static struct sock *unix_from_bucket(struct seq_file *seq, loff_t *pos)
2260 unsigned long offset = get_offset(*pos);
2261 unsigned long bucket = get_bucket(*pos);
2263 unsigned long count = 0;
2265 for (sk = sk_head(&unix_socket_table[bucket]); sk; sk = sk_next(sk)) {
2266 if (sock_net(sk) != seq_file_net(seq))
2268 if (++count == offset)
2275 static struct sock *unix_next_socket(struct seq_file *seq,
2279 unsigned long bucket;
2281 while (sk > (struct sock *)SEQ_START_TOKEN) {
2285 if (sock_net(sk) == seq_file_net(seq))
2290 sk = unix_from_bucket(seq, pos);
2295 bucket = get_bucket(*pos) + 1;
2296 *pos = set_bucket_offset(bucket, 1);
2297 } while (bucket < ARRAY_SIZE(unix_socket_table));
2302 static void *unix_seq_start(struct seq_file *seq, loff_t *pos)
2303 __acquires(unix_table_lock)
2305 spin_lock(&unix_table_lock);
2308 return SEQ_START_TOKEN;
2310 if (get_bucket(*pos) >= ARRAY_SIZE(unix_socket_table))
2313 return unix_next_socket(seq, NULL, pos);
2316 static void *unix_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2319 return unix_next_socket(seq, v, pos);
2322 static void unix_seq_stop(struct seq_file *seq, void *v)
2323 __releases(unix_table_lock)
2325 spin_unlock(&unix_table_lock);
2328 static int unix_seq_show(struct seq_file *seq, void *v)
2331 if (v == SEQ_START_TOKEN)
2332 seq_puts(seq, "Num RefCount Protocol Flags Type St "
2336 struct unix_sock *u = unix_sk(s);
2339 seq_printf(seq, "%pK: %08X %08X %08X %04X %02X %5lu",
2341 atomic_read(&s->sk_refcnt),
2343 s->sk_state == TCP_LISTEN ? __SO_ACCEPTCON : 0,
2346 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTED : SS_UNCONNECTED) :
2347 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTING : SS_DISCONNECTING),
2355 len = u->addr->len - sizeof(short);
2356 if (!UNIX_ABSTRACT(s))
2362 for ( ; i < len; i++)
2363 seq_putc(seq, u->addr->name->sun_path[i]);
2365 unix_state_unlock(s);
2366 seq_putc(seq, '\n');
2372 static const struct seq_operations unix_seq_ops = {
2373 .start = unix_seq_start,
2374 .next = unix_seq_next,
2375 .stop = unix_seq_stop,
2376 .show = unix_seq_show,
2379 static int unix_seq_open(struct inode *inode, struct file *file)
2381 return seq_open_net(inode, file, &unix_seq_ops,
2382 sizeof(struct seq_net_private));
2385 static const struct file_operations unix_seq_fops = {
2386 .owner = THIS_MODULE,
2387 .open = unix_seq_open,
2389 .llseek = seq_lseek,
2390 .release = seq_release_net,
2395 static const struct net_proto_family unix_family_ops = {
2397 .create = unix_create,
2398 .owner = THIS_MODULE,
2402 static int __net_init unix_net_init(struct net *net)
2404 int error = -ENOMEM;
2406 net->unx.sysctl_max_dgram_qlen = 10;
2407 if (unix_sysctl_register(net))
2410 #ifdef CONFIG_PROC_FS
2411 if (!proc_net_fops_create(net, "unix", 0, &unix_seq_fops)) {
2412 unix_sysctl_unregister(net);
2421 static void __net_exit unix_net_exit(struct net *net)
2423 unix_sysctl_unregister(net);
2424 proc_net_remove(net, "unix");
2427 static struct pernet_operations unix_net_ops = {
2428 .init = unix_net_init,
2429 .exit = unix_net_exit,
2432 static int __init af_unix_init(void)
2435 struct sk_buff *dummy_skb;
2437 BUILD_BUG_ON(sizeof(struct unix_skb_parms) > sizeof(dummy_skb->cb));
2439 rc = proto_register(&unix_proto, 1);
2441 printk(KERN_CRIT "%s: Cannot create unix_sock SLAB cache!\n",
2446 sock_register(&unix_family_ops);
2447 register_pernet_subsys(&unix_net_ops);
2452 static void __exit af_unix_exit(void)
2454 sock_unregister(PF_UNIX);
2455 proto_unregister(&unix_proto);
2456 unregister_pernet_subsys(&unix_net_ops);
2459 /* Earlier than device_initcall() so that other drivers invoking
2460 request_module() don't end up in a loop when modprobe tries
2461 to use a UNIX socket. But later than subsys_initcall() because
2462 we depend on stuff initialised there */
2463 fs_initcall(af_unix_init);
2464 module_exit(af_unix_exit);
2466 MODULE_LICENSE("GPL");
2467 MODULE_ALIAS_NETPROTO(PF_UNIX);