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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
85 #include <linux/module.h>
86 #include <linux/kernel.h>
87 #include <linux/signal.h>
88 #include <linux/sched.h>
89 #include <linux/errno.h>
90 #include <linux/string.h>
91 #include <linux/stat.h>
92 #include <linux/dcache.h>
93 #include <linux/namei.h>
94 #include <linux/socket.h>
96 #include <linux/fcntl.h>
97 #include <linux/termios.h>
98 #include <linux/sockios.h>
99 #include <linux/net.h>
100 #include <linux/in.h>
101 #include <linux/fs.h>
102 #include <linux/slab.h>
103 #include <asm/uaccess.h>
104 #include <linux/skbuff.h>
105 #include <linux/netdevice.h>
106 #include <net/net_namespace.h>
107 #include <net/sock.h>
108 #include <net/tcp_states.h>
109 #include <net/af_unix.h>
110 #include <linux/proc_fs.h>
111 #include <linux/seq_file.h>
113 #include <linux/init.h>
114 #include <linux/poll.h>
115 #include <linux/rtnetlink.h>
116 #include <linux/mount.h>
117 #include <net/checksum.h>
118 #include <linux/security.h>
119 #include <linux/freezer.h>
121 struct hlist_head unix_socket_table[2 * UNIX_HASH_SIZE];
122 EXPORT_SYMBOL_GPL(unix_socket_table);
123 DEFINE_SPINLOCK(unix_table_lock);
124 EXPORT_SYMBOL_GPL(unix_table_lock);
125 static atomic_long_t unix_nr_socks;
128 static struct hlist_head *unix_sockets_unbound(void *addr)
130 unsigned long hash = (unsigned long)addr;
134 hash %= UNIX_HASH_SIZE;
135 return &unix_socket_table[UNIX_HASH_SIZE + hash];
138 #define UNIX_ABSTRACT(sk) (unix_sk(sk)->addr->hash < UNIX_HASH_SIZE)
140 #ifdef CONFIG_SECURITY_NETWORK
141 static void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb)
143 UNIXCB(skb).secid = scm->secid;
146 static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb)
148 scm->secid = UNIXCB(skb).secid;
151 static inline bool unix_secdata_eq(struct scm_cookie *scm, struct sk_buff *skb)
153 return (scm->secid == UNIXCB(skb).secid);
156 static inline void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb)
159 static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb)
162 static inline bool unix_secdata_eq(struct scm_cookie *scm, struct sk_buff *skb)
166 #endif /* CONFIG_SECURITY_NETWORK */
169 * SMP locking strategy:
170 * hash table is protected with spinlock unix_table_lock
171 * each socket state is protected by separate spin lock.
174 static inline unsigned int unix_hash_fold(__wsum n)
176 unsigned int hash = (__force unsigned int)csum_fold(n);
179 return hash&(UNIX_HASH_SIZE-1);
182 #define unix_peer(sk) (unix_sk(sk)->peer)
184 static inline int unix_our_peer(struct sock *sk, struct sock *osk)
186 return unix_peer(osk) == sk;
189 static inline int unix_may_send(struct sock *sk, struct sock *osk)
191 return unix_peer(osk) == NULL || unix_our_peer(sk, osk);
194 static inline int unix_recvq_full(struct sock const *sk)
196 return skb_queue_len(&sk->sk_receive_queue) > sk->sk_max_ack_backlog;
199 struct sock *unix_peer_get(struct sock *s)
207 unix_state_unlock(s);
210 EXPORT_SYMBOL_GPL(unix_peer_get);
212 static inline void unix_release_addr(struct unix_address *addr)
214 if (atomic_dec_and_test(&addr->refcnt))
219 * Check unix socket name:
220 * - should be not zero length.
221 * - if started by not zero, should be NULL terminated (FS object)
222 * - if started by zero, it is abstract name.
225 static int unix_mkname(struct sockaddr_un *sunaddr, int len, unsigned int *hashp)
227 if (len <= sizeof(short) || len > sizeof(*sunaddr))
229 if (!sunaddr || sunaddr->sun_family != AF_UNIX)
231 if (sunaddr->sun_path[0]) {
233 * This may look like an off by one error but it is a bit more
234 * subtle. 108 is the longest valid AF_UNIX path for a binding.
235 * sun_path[108] doesn't as such exist. However in kernel space
236 * we are guaranteed that it is a valid memory location in our
237 * kernel address buffer.
239 ((char *)sunaddr)[len] = 0;
240 len = strlen(sunaddr->sun_path)+1+sizeof(short);
244 *hashp = unix_hash_fold(csum_partial(sunaddr, len, 0));
248 static void __unix_remove_socket(struct sock *sk)
250 sk_del_node_init(sk);
253 static void __unix_insert_socket(struct hlist_head *list, struct sock *sk)
255 WARN_ON(!sk_unhashed(sk));
256 sk_add_node(sk, list);
259 static inline void unix_remove_socket(struct sock *sk)
261 spin_lock(&unix_table_lock);
262 __unix_remove_socket(sk);
263 spin_unlock(&unix_table_lock);
266 static inline void unix_insert_socket(struct hlist_head *list, struct sock *sk)
268 spin_lock(&unix_table_lock);
269 __unix_insert_socket(list, sk);
270 spin_unlock(&unix_table_lock);
273 static struct sock *__unix_find_socket_byname(struct net *net,
274 struct sockaddr_un *sunname,
275 int len, int type, unsigned int hash)
279 sk_for_each(s, &unix_socket_table[hash ^ type]) {
280 struct unix_sock *u = unix_sk(s);
282 if (!net_eq(sock_net(s), net))
285 if (u->addr->len == len &&
286 !memcmp(u->addr->name, sunname, len))
294 static inline struct sock *unix_find_socket_byname(struct net *net,
295 struct sockaddr_un *sunname,
301 spin_lock(&unix_table_lock);
302 s = __unix_find_socket_byname(net, sunname, len, type, hash);
305 spin_unlock(&unix_table_lock);
309 static struct sock *unix_find_socket_byinode(struct inode *i)
313 spin_lock(&unix_table_lock);
315 &unix_socket_table[i->i_ino & (UNIX_HASH_SIZE - 1)]) {
316 struct dentry *dentry = unix_sk(s)->path.dentry;
318 if (dentry && d_real_inode(dentry) == i) {
325 spin_unlock(&unix_table_lock);
329 /* Support code for asymmetrically connected dgram sockets
331 * If a datagram socket is connected to a socket not itself connected
332 * to the first socket (eg, /dev/log), clients may only enqueue more
333 * messages if the present receive queue of the server socket is not
334 * "too large". This means there's a second writeability condition
335 * poll and sendmsg need to test. The dgram recv code will do a wake
336 * up on the peer_wait wait queue of a socket upon reception of a
337 * datagram which needs to be propagated to sleeping would-be writers
338 * since these might not have sent anything so far. This can't be
339 * accomplished via poll_wait because the lifetime of the server
340 * socket might be less than that of its clients if these break their
341 * association with it or if the server socket is closed while clients
342 * are still connected to it and there's no way to inform "a polling
343 * implementation" that it should let go of a certain wait queue
345 * In order to propagate a wake up, a wait_queue_t of the client
346 * socket is enqueued on the peer_wait queue of the server socket
347 * whose wake function does a wake_up on the ordinary client socket
348 * wait queue. This connection is established whenever a write (or
349 * poll for write) hit the flow control condition and broken when the
350 * association to the server socket is dissolved or after a wake up
354 static int unix_dgram_peer_wake_relay(wait_queue_t *q, unsigned mode, int flags,
358 wait_queue_head_t *u_sleep;
360 u = container_of(q, struct unix_sock, peer_wake);
362 __remove_wait_queue(&unix_sk(u->peer_wake.private)->peer_wait,
364 u->peer_wake.private = NULL;
366 /* relaying can only happen while the wq still exists */
367 u_sleep = sk_sleep(&u->sk);
369 wake_up_interruptible_poll(u_sleep, key);
374 static int unix_dgram_peer_wake_connect(struct sock *sk, struct sock *other)
376 struct unix_sock *u, *u_other;
380 u_other = unix_sk(other);
382 spin_lock(&u_other->peer_wait.lock);
384 if (!u->peer_wake.private) {
385 u->peer_wake.private = other;
386 __add_wait_queue(&u_other->peer_wait, &u->peer_wake);
391 spin_unlock(&u_other->peer_wait.lock);
395 static void unix_dgram_peer_wake_disconnect(struct sock *sk,
398 struct unix_sock *u, *u_other;
401 u_other = unix_sk(other);
402 spin_lock(&u_other->peer_wait.lock);
404 if (u->peer_wake.private == other) {
405 __remove_wait_queue(&u_other->peer_wait, &u->peer_wake);
406 u->peer_wake.private = NULL;
409 spin_unlock(&u_other->peer_wait.lock);
412 static void unix_dgram_peer_wake_disconnect_wakeup(struct sock *sk,
415 unix_dgram_peer_wake_disconnect(sk, other);
416 wake_up_interruptible_poll(sk_sleep(sk),
423 * - unix_peer(sk) == other
424 * - association is stable
426 static int unix_dgram_peer_wake_me(struct sock *sk, struct sock *other)
430 connected = unix_dgram_peer_wake_connect(sk, other);
432 if (unix_recvq_full(other))
436 unix_dgram_peer_wake_disconnect(sk, other);
441 static int unix_writable(const struct sock *sk)
443 return sk->sk_state != TCP_LISTEN &&
444 (atomic_read(&sk->sk_wmem_alloc) << 2) <= sk->sk_sndbuf;
447 static void unix_write_space(struct sock *sk)
449 struct socket_wq *wq;
452 if (unix_writable(sk)) {
453 wq = rcu_dereference(sk->sk_wq);
454 if (wq_has_sleeper(wq))
455 wake_up_interruptible_sync_poll(&wq->wait,
456 POLLOUT | POLLWRNORM | POLLWRBAND);
457 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
462 /* When dgram socket disconnects (or changes its peer), we clear its receive
463 * queue of packets arrived from previous peer. First, it allows to do
464 * flow control based only on wmem_alloc; second, sk connected to peer
465 * may receive messages only from that peer. */
466 static void unix_dgram_disconnected(struct sock *sk, struct sock *other)
468 if (!skb_queue_empty(&sk->sk_receive_queue)) {
469 skb_queue_purge(&sk->sk_receive_queue);
470 wake_up_interruptible_all(&unix_sk(sk)->peer_wait);
472 /* If one link of bidirectional dgram pipe is disconnected,
473 * we signal error. Messages are lost. Do not make this,
474 * when peer was not connected to us.
476 if (!sock_flag(other, SOCK_DEAD) && unix_peer(other) == sk) {
477 other->sk_err = ECONNRESET;
478 other->sk_error_report(other);
483 static void unix_sock_destructor(struct sock *sk)
485 struct unix_sock *u = unix_sk(sk);
487 skb_queue_purge(&sk->sk_receive_queue);
489 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
490 WARN_ON(!sk_unhashed(sk));
491 WARN_ON(sk->sk_socket);
492 if (!sock_flag(sk, SOCK_DEAD)) {
493 pr_info("Attempt to release alive unix socket: %p\n", sk);
498 unix_release_addr(u->addr);
500 atomic_long_dec(&unix_nr_socks);
502 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
504 #ifdef UNIX_REFCNT_DEBUG
505 pr_debug("UNIX %p is destroyed, %ld are still alive.\n", sk,
506 atomic_long_read(&unix_nr_socks));
510 static void unix_release_sock(struct sock *sk, int embrion)
512 struct unix_sock *u = unix_sk(sk);
518 unix_remove_socket(sk);
523 sk->sk_shutdown = SHUTDOWN_MASK;
525 u->path.dentry = NULL;
527 state = sk->sk_state;
528 sk->sk_state = TCP_CLOSE;
529 unix_state_unlock(sk);
531 wake_up_interruptible_all(&u->peer_wait);
533 skpair = unix_peer(sk);
535 if (skpair != NULL) {
536 if (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) {
537 unix_state_lock(skpair);
539 skpair->sk_shutdown = SHUTDOWN_MASK;
540 if (!skb_queue_empty(&sk->sk_receive_queue) || embrion)
541 skpair->sk_err = ECONNRESET;
542 unix_state_unlock(skpair);
543 skpair->sk_state_change(skpair);
544 sk_wake_async(skpair, SOCK_WAKE_WAITD, POLL_HUP);
547 unix_dgram_peer_wake_disconnect(sk, skpair);
548 sock_put(skpair); /* It may now die */
549 unix_peer(sk) = NULL;
552 /* Try to flush out this socket. Throw out buffers at least */
554 while ((skb = skb_dequeue(&sk->sk_receive_queue)) != NULL) {
555 if (state == TCP_LISTEN)
556 unix_release_sock(skb->sk, 1);
557 /* passed fds are erased in the kfree_skb hook */
558 UNIXCB(skb).consumed = skb->len;
567 /* ---- Socket is dead now and most probably destroyed ---- */
570 * Fixme: BSD difference: In BSD all sockets connected to us get
571 * ECONNRESET and we die on the spot. In Linux we behave
572 * like files and pipes do and wait for the last
575 * Can't we simply set sock->err?
577 * What the above comment does talk about? --ANK(980817)
580 if (unix_tot_inflight)
581 unix_gc(); /* Garbage collect fds */
584 static void init_peercred(struct sock *sk)
586 put_pid(sk->sk_peer_pid);
587 if (sk->sk_peer_cred)
588 put_cred(sk->sk_peer_cred);
589 sk->sk_peer_pid = get_pid(task_tgid(current));
590 sk->sk_peer_cred = get_current_cred();
593 static void copy_peercred(struct sock *sk, struct sock *peersk)
595 put_pid(sk->sk_peer_pid);
596 if (sk->sk_peer_cred)
597 put_cred(sk->sk_peer_cred);
598 sk->sk_peer_pid = get_pid(peersk->sk_peer_pid);
599 sk->sk_peer_cred = get_cred(peersk->sk_peer_cred);
602 static int unix_listen(struct socket *sock, int backlog)
605 struct sock *sk = sock->sk;
606 struct unix_sock *u = unix_sk(sk);
607 struct pid *old_pid = NULL;
610 if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
611 goto out; /* Only stream/seqpacket sockets accept */
614 goto out; /* No listens on an unbound socket */
616 if (sk->sk_state != TCP_CLOSE && sk->sk_state != TCP_LISTEN)
618 if (backlog > sk->sk_max_ack_backlog)
619 wake_up_interruptible_all(&u->peer_wait);
620 sk->sk_max_ack_backlog = backlog;
621 sk->sk_state = TCP_LISTEN;
622 /* set credentials so connect can copy them */
627 unix_state_unlock(sk);
633 static int unix_release(struct socket *);
634 static int unix_bind(struct socket *, struct sockaddr *, int);
635 static int unix_stream_connect(struct socket *, struct sockaddr *,
636 int addr_len, int flags);
637 static int unix_socketpair(struct socket *, struct socket *);
638 static int unix_accept(struct socket *, struct socket *, int);
639 static int unix_getname(struct socket *, struct sockaddr *, int *, int);
640 static unsigned int unix_poll(struct file *, struct socket *, poll_table *);
641 static unsigned int unix_dgram_poll(struct file *, struct socket *,
643 static int unix_ioctl(struct socket *, unsigned int, unsigned long);
644 static int unix_shutdown(struct socket *, int);
645 static int unix_stream_sendmsg(struct socket *, struct msghdr *, size_t);
646 static int unix_stream_recvmsg(struct socket *, struct msghdr *, size_t, int);
647 static ssize_t unix_stream_sendpage(struct socket *, struct page *, int offset,
648 size_t size, int flags);
649 static ssize_t unix_stream_splice_read(struct socket *, loff_t *ppos,
650 struct pipe_inode_info *, size_t size,
652 static int unix_dgram_sendmsg(struct socket *, struct msghdr *, size_t);
653 static int unix_dgram_recvmsg(struct socket *, struct msghdr *, size_t, int);
654 static int unix_dgram_connect(struct socket *, struct sockaddr *,
656 static int unix_seqpacket_sendmsg(struct socket *, struct msghdr *, size_t);
657 static int unix_seqpacket_recvmsg(struct socket *, struct msghdr *, size_t,
660 static int unix_set_peek_off(struct sock *sk, int val)
662 struct unix_sock *u = unix_sk(sk);
664 if (mutex_lock_interruptible(&u->iolock))
667 sk->sk_peek_off = val;
668 mutex_unlock(&u->iolock);
674 static const struct proto_ops unix_stream_ops = {
676 .owner = THIS_MODULE,
677 .release = unix_release,
679 .connect = unix_stream_connect,
680 .socketpair = unix_socketpair,
681 .accept = unix_accept,
682 .getname = unix_getname,
685 .listen = unix_listen,
686 .shutdown = unix_shutdown,
687 .setsockopt = sock_no_setsockopt,
688 .getsockopt = sock_no_getsockopt,
689 .sendmsg = unix_stream_sendmsg,
690 .recvmsg = unix_stream_recvmsg,
691 .mmap = sock_no_mmap,
692 .sendpage = unix_stream_sendpage,
693 .splice_read = unix_stream_splice_read,
694 .set_peek_off = unix_set_peek_off,
697 static const struct proto_ops unix_dgram_ops = {
699 .owner = THIS_MODULE,
700 .release = unix_release,
702 .connect = unix_dgram_connect,
703 .socketpair = unix_socketpair,
704 .accept = sock_no_accept,
705 .getname = unix_getname,
706 .poll = unix_dgram_poll,
708 .listen = sock_no_listen,
709 .shutdown = unix_shutdown,
710 .setsockopt = sock_no_setsockopt,
711 .getsockopt = sock_no_getsockopt,
712 .sendmsg = unix_dgram_sendmsg,
713 .recvmsg = unix_dgram_recvmsg,
714 .mmap = sock_no_mmap,
715 .sendpage = sock_no_sendpage,
716 .set_peek_off = unix_set_peek_off,
719 static const struct proto_ops unix_seqpacket_ops = {
721 .owner = THIS_MODULE,
722 .release = unix_release,
724 .connect = unix_stream_connect,
725 .socketpair = unix_socketpair,
726 .accept = unix_accept,
727 .getname = unix_getname,
728 .poll = unix_dgram_poll,
730 .listen = unix_listen,
731 .shutdown = unix_shutdown,
732 .setsockopt = sock_no_setsockopt,
733 .getsockopt = sock_no_getsockopt,
734 .sendmsg = unix_seqpacket_sendmsg,
735 .recvmsg = unix_seqpacket_recvmsg,
736 .mmap = sock_no_mmap,
737 .sendpage = sock_no_sendpage,
738 .set_peek_off = unix_set_peek_off,
741 static struct proto unix_proto = {
743 .owner = THIS_MODULE,
744 .obj_size = sizeof(struct unix_sock),
748 * AF_UNIX sockets do not interact with hardware, hence they
749 * dont trigger interrupts - so it's safe for them to have
750 * bh-unsafe locking for their sk_receive_queue.lock. Split off
751 * this special lock-class by reinitializing the spinlock key:
753 static struct lock_class_key af_unix_sk_receive_queue_lock_key;
755 static struct sock *unix_create1(struct net *net, struct socket *sock, int kern)
757 struct sock *sk = NULL;
760 atomic_long_inc(&unix_nr_socks);
761 if (atomic_long_read(&unix_nr_socks) > 2 * get_max_files())
764 sk = sk_alloc(net, PF_UNIX, GFP_KERNEL, &unix_proto, kern);
768 sock_init_data(sock, sk);
769 lockdep_set_class(&sk->sk_receive_queue.lock,
770 &af_unix_sk_receive_queue_lock_key);
772 sk->sk_write_space = unix_write_space;
773 sk->sk_max_ack_backlog = net->unx.sysctl_max_dgram_qlen;
774 sk->sk_destruct = unix_sock_destructor;
776 u->path.dentry = NULL;
778 spin_lock_init(&u->lock);
779 atomic_long_set(&u->inflight, 0);
780 INIT_LIST_HEAD(&u->link);
781 mutex_init(&u->iolock); /* single task reading lock */
782 mutex_init(&u->bindlock); /* single task binding lock */
783 init_waitqueue_head(&u->peer_wait);
784 init_waitqueue_func_entry(&u->peer_wake, unix_dgram_peer_wake_relay);
785 unix_insert_socket(unix_sockets_unbound(sk), sk);
788 atomic_long_dec(&unix_nr_socks);
791 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
797 static int unix_create(struct net *net, struct socket *sock, int protocol,
800 if (protocol && protocol != PF_UNIX)
801 return -EPROTONOSUPPORT;
803 sock->state = SS_UNCONNECTED;
805 switch (sock->type) {
807 sock->ops = &unix_stream_ops;
810 * Believe it or not BSD has AF_UNIX, SOCK_RAW though
814 sock->type = SOCK_DGRAM;
816 sock->ops = &unix_dgram_ops;
819 sock->ops = &unix_seqpacket_ops;
822 return -ESOCKTNOSUPPORT;
825 return unix_create1(net, sock, kern) ? 0 : -ENOMEM;
828 static int unix_release(struct socket *sock)
830 struct sock *sk = sock->sk;
835 unix_release_sock(sk, 0);
841 static int unix_autobind(struct socket *sock)
843 struct sock *sk = sock->sk;
844 struct net *net = sock_net(sk);
845 struct unix_sock *u = unix_sk(sk);
846 static u32 ordernum = 1;
847 struct unix_address *addr;
849 unsigned int retries = 0;
851 err = mutex_lock_interruptible(&u->bindlock);
860 addr = kzalloc(sizeof(*addr) + sizeof(short) + 16, GFP_KERNEL);
864 addr->name->sun_family = AF_UNIX;
865 atomic_set(&addr->refcnt, 1);
868 addr->len = sprintf(addr->name->sun_path+1, "%05x", ordernum) + 1 + sizeof(short);
869 addr->hash = unix_hash_fold(csum_partial(addr->name, addr->len, 0));
871 spin_lock(&unix_table_lock);
872 ordernum = (ordernum+1)&0xFFFFF;
874 if (__unix_find_socket_byname(net, addr->name, addr->len, sock->type,
876 spin_unlock(&unix_table_lock);
878 * __unix_find_socket_byname() may take long time if many names
879 * are already in use.
882 /* Give up if all names seems to be in use. */
883 if (retries++ == 0xFFFFF) {
890 addr->hash ^= sk->sk_type;
892 __unix_remove_socket(sk);
894 __unix_insert_socket(&unix_socket_table[addr->hash], sk);
895 spin_unlock(&unix_table_lock);
898 out: mutex_unlock(&u->bindlock);
902 static struct sock *unix_find_other(struct net *net,
903 struct sockaddr_un *sunname, int len,
904 int type, unsigned int hash, int *error)
910 if (sunname->sun_path[0]) {
912 err = kern_path(sunname->sun_path, LOOKUP_FOLLOW, &path);
915 inode = d_real_inode(path.dentry);
916 err = inode_permission(inode, MAY_WRITE);
921 if (!S_ISSOCK(inode->i_mode))
923 u = unix_find_socket_byinode(inode);
927 if (u->sk_type == type)
933 if (u->sk_type != type) {
939 u = unix_find_socket_byname(net, sunname, len, type, hash);
941 struct dentry *dentry;
942 dentry = unix_sk(u)->path.dentry;
944 touch_atime(&unix_sk(u)->path);
957 static int unix_mknod(const char *sun_path, umode_t mode, struct path *res)
959 struct dentry *dentry;
963 * Get the parent directory, calculate the hash for last
966 dentry = kern_path_create(AT_FDCWD, sun_path, &path, 0);
967 err = PTR_ERR(dentry);
972 * All right, let's create it.
974 err = security_path_mknod(&path, dentry, mode, 0);
976 err = vfs_mknod(d_inode(path.dentry), dentry, mode, 0);
978 res->mnt = mntget(path.mnt);
979 res->dentry = dget(dentry);
982 done_path_create(&path, dentry);
986 static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
988 struct sock *sk = sock->sk;
989 struct net *net = sock_net(sk);
990 struct unix_sock *u = unix_sk(sk);
991 struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
992 char *sun_path = sunaddr->sun_path;
995 struct unix_address *addr;
996 struct hlist_head *list;
997 struct path path = { NULL, NULL };
1000 if (sunaddr->sun_family != AF_UNIX)
1003 if (addr_len == sizeof(short)) {
1004 err = unix_autobind(sock);
1008 err = unix_mkname(sunaddr, addr_len, &hash);
1014 umode_t mode = S_IFSOCK |
1015 (SOCK_INODE(sock)->i_mode & ~current_umask());
1016 err = unix_mknod(sun_path, mode, &path);
1024 err = mutex_lock_interruptible(&u->bindlock);
1033 addr = kmalloc(sizeof(*addr)+addr_len, GFP_KERNEL);
1037 memcpy(addr->name, sunaddr, addr_len);
1038 addr->len = addr_len;
1039 addr->hash = hash ^ sk->sk_type;
1040 atomic_set(&addr->refcnt, 1);
1043 addr->hash = UNIX_HASH_SIZE;
1044 hash = d_real_inode(path.dentry)->i_ino & (UNIX_HASH_SIZE - 1);
1045 spin_lock(&unix_table_lock);
1047 list = &unix_socket_table[hash];
1049 spin_lock(&unix_table_lock);
1051 if (__unix_find_socket_byname(net, sunaddr, addr_len,
1052 sk->sk_type, hash)) {
1053 unix_release_addr(addr);
1057 list = &unix_socket_table[addr->hash];
1061 __unix_remove_socket(sk);
1063 __unix_insert_socket(list, sk);
1066 spin_unlock(&unix_table_lock);
1068 mutex_unlock(&u->bindlock);
1076 static void unix_state_double_lock(struct sock *sk1, struct sock *sk2)
1078 if (unlikely(sk1 == sk2) || !sk2) {
1079 unix_state_lock(sk1);
1083 unix_state_lock(sk1);
1084 unix_state_lock_nested(sk2);
1086 unix_state_lock(sk2);
1087 unix_state_lock_nested(sk1);
1091 static void unix_state_double_unlock(struct sock *sk1, struct sock *sk2)
1093 if (unlikely(sk1 == sk2) || !sk2) {
1094 unix_state_unlock(sk1);
1097 unix_state_unlock(sk1);
1098 unix_state_unlock(sk2);
1101 static int unix_dgram_connect(struct socket *sock, struct sockaddr *addr,
1102 int alen, int flags)
1104 struct sock *sk = sock->sk;
1105 struct net *net = sock_net(sk);
1106 struct sockaddr_un *sunaddr = (struct sockaddr_un *)addr;
1111 if (addr->sa_family != AF_UNSPEC) {
1112 err = unix_mkname(sunaddr, alen, &hash);
1117 if (test_bit(SOCK_PASSCRED, &sock->flags) &&
1118 !unix_sk(sk)->addr && (err = unix_autobind(sock)) != 0)
1122 other = unix_find_other(net, sunaddr, alen, sock->type, hash, &err);
1126 unix_state_double_lock(sk, other);
1128 /* Apparently VFS overslept socket death. Retry. */
1129 if (sock_flag(other, SOCK_DEAD)) {
1130 unix_state_double_unlock(sk, other);
1136 if (!unix_may_send(sk, other))
1139 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1145 * 1003.1g breaking connected state with AF_UNSPEC
1148 unix_state_double_lock(sk, other);
1152 * If it was connected, reconnect.
1154 if (unix_peer(sk)) {
1155 struct sock *old_peer = unix_peer(sk);
1156 unix_peer(sk) = other;
1157 unix_dgram_peer_wake_disconnect_wakeup(sk, old_peer);
1159 unix_state_double_unlock(sk, other);
1161 if (other != old_peer)
1162 unix_dgram_disconnected(sk, old_peer);
1165 unix_peer(sk) = other;
1166 unix_state_double_unlock(sk, other);
1171 unix_state_double_unlock(sk, other);
1177 static long unix_wait_for_peer(struct sock *other, long timeo)
1179 struct unix_sock *u = unix_sk(other);
1183 prepare_to_wait_exclusive(&u->peer_wait, &wait, TASK_INTERRUPTIBLE);
1185 sched = !sock_flag(other, SOCK_DEAD) &&
1186 !(other->sk_shutdown & RCV_SHUTDOWN) &&
1187 unix_recvq_full(other);
1189 unix_state_unlock(other);
1192 timeo = schedule_timeout(timeo);
1194 finish_wait(&u->peer_wait, &wait);
1198 static int unix_stream_connect(struct socket *sock, struct sockaddr *uaddr,
1199 int addr_len, int flags)
1201 struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
1202 struct sock *sk = sock->sk;
1203 struct net *net = sock_net(sk);
1204 struct unix_sock *u = unix_sk(sk), *newu, *otheru;
1205 struct sock *newsk = NULL;
1206 struct sock *other = NULL;
1207 struct sk_buff *skb = NULL;
1213 err = unix_mkname(sunaddr, addr_len, &hash);
1218 if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr &&
1219 (err = unix_autobind(sock)) != 0)
1222 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
1224 /* First of all allocate resources.
1225 If we will make it after state is locked,
1226 we will have to recheck all again in any case.
1231 /* create new sock for complete connection */
1232 newsk = unix_create1(sock_net(sk), NULL, 0);
1236 /* Allocate skb for sending to listening sock */
1237 skb = sock_wmalloc(newsk, 1, 0, GFP_KERNEL);
1242 /* Find listening sock. */
1243 other = unix_find_other(net, sunaddr, addr_len, sk->sk_type, hash, &err);
1247 /* Latch state of peer */
1248 unix_state_lock(other);
1250 /* Apparently VFS overslept socket death. Retry. */
1251 if (sock_flag(other, SOCK_DEAD)) {
1252 unix_state_unlock(other);
1257 err = -ECONNREFUSED;
1258 if (other->sk_state != TCP_LISTEN)
1260 if (other->sk_shutdown & RCV_SHUTDOWN)
1263 if (unix_recvq_full(other)) {
1268 timeo = unix_wait_for_peer(other, timeo);
1270 err = sock_intr_errno(timeo);
1271 if (signal_pending(current))
1279 It is tricky place. We need to grab our state lock and cannot
1280 drop lock on peer. It is dangerous because deadlock is
1281 possible. Connect to self case and simultaneous
1282 attempt to connect are eliminated by checking socket
1283 state. other is TCP_LISTEN, if sk is TCP_LISTEN we
1284 check this before attempt to grab lock.
1286 Well, and we have to recheck the state after socket locked.
1292 /* This is ok... continue with connect */
1294 case TCP_ESTABLISHED:
1295 /* Socket is already connected */
1303 unix_state_lock_nested(sk);
1305 if (sk->sk_state != st) {
1306 unix_state_unlock(sk);
1307 unix_state_unlock(other);
1312 err = security_unix_stream_connect(sk, other, newsk);
1314 unix_state_unlock(sk);
1318 /* The way is open! Fastly set all the necessary fields... */
1321 unix_peer(newsk) = sk;
1322 newsk->sk_state = TCP_ESTABLISHED;
1323 newsk->sk_type = sk->sk_type;
1324 init_peercred(newsk);
1325 newu = unix_sk(newsk);
1326 RCU_INIT_POINTER(newsk->sk_wq, &newu->peer_wq);
1327 otheru = unix_sk(other);
1329 /* copy address information from listening to new sock*/
1331 atomic_inc(&otheru->addr->refcnt);
1332 newu->addr = otheru->addr;
1334 if (otheru->path.dentry) {
1335 path_get(&otheru->path);
1336 newu->path = otheru->path;
1339 /* Set credentials */
1340 copy_peercred(sk, other);
1342 sock->state = SS_CONNECTED;
1343 sk->sk_state = TCP_ESTABLISHED;
1346 smp_mb__after_atomic(); /* sock_hold() does an atomic_inc() */
1347 unix_peer(sk) = newsk;
1349 unix_state_unlock(sk);
1351 /* take ten and and send info to listening sock */
1352 spin_lock(&other->sk_receive_queue.lock);
1353 __skb_queue_tail(&other->sk_receive_queue, skb);
1354 spin_unlock(&other->sk_receive_queue.lock);
1355 unix_state_unlock(other);
1356 other->sk_data_ready(other);
1362 unix_state_unlock(other);
1367 unix_release_sock(newsk, 0);
1373 static int unix_socketpair(struct socket *socka, struct socket *sockb)
1375 struct sock *ska = socka->sk, *skb = sockb->sk;
1377 /* Join our sockets back to back */
1380 unix_peer(ska) = skb;
1381 unix_peer(skb) = ska;
1385 if (ska->sk_type != SOCK_DGRAM) {
1386 ska->sk_state = TCP_ESTABLISHED;
1387 skb->sk_state = TCP_ESTABLISHED;
1388 socka->state = SS_CONNECTED;
1389 sockb->state = SS_CONNECTED;
1394 static void unix_sock_inherit_flags(const struct socket *old,
1397 if (test_bit(SOCK_PASSCRED, &old->flags))
1398 set_bit(SOCK_PASSCRED, &new->flags);
1399 if (test_bit(SOCK_PASSSEC, &old->flags))
1400 set_bit(SOCK_PASSSEC, &new->flags);
1403 static int unix_accept(struct socket *sock, struct socket *newsock, int flags)
1405 struct sock *sk = sock->sk;
1407 struct sk_buff *skb;
1411 if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
1415 if (sk->sk_state != TCP_LISTEN)
1418 /* If socket state is TCP_LISTEN it cannot change (for now...),
1419 * so that no locks are necessary.
1422 skb = skb_recv_datagram(sk, 0, flags&O_NONBLOCK, &err);
1424 /* This means receive shutdown. */
1431 skb_free_datagram(sk, skb);
1432 wake_up_interruptible(&unix_sk(sk)->peer_wait);
1434 /* attach accepted sock to socket */
1435 unix_state_lock(tsk);
1436 newsock->state = SS_CONNECTED;
1437 unix_sock_inherit_flags(sock, newsock);
1438 sock_graft(tsk, newsock);
1439 unix_state_unlock(tsk);
1447 static int unix_getname(struct socket *sock, struct sockaddr *uaddr, int *uaddr_len, int peer)
1449 struct sock *sk = sock->sk;
1450 struct unix_sock *u;
1451 DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr, uaddr);
1455 sk = unix_peer_get(sk);
1466 unix_state_lock(sk);
1468 sunaddr->sun_family = AF_UNIX;
1469 sunaddr->sun_path[0] = 0;
1470 *uaddr_len = sizeof(short);
1472 struct unix_address *addr = u->addr;
1474 *uaddr_len = addr->len;
1475 memcpy(sunaddr, addr->name, *uaddr_len);
1477 unix_state_unlock(sk);
1483 static void unix_detach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1487 scm->fp = UNIXCB(skb).fp;
1488 UNIXCB(skb).fp = NULL;
1490 for (i = scm->fp->count-1; i >= 0; i--)
1491 unix_notinflight(scm->fp->user, scm->fp->fp[i]);
1494 static void unix_destruct_scm(struct sk_buff *skb)
1496 struct scm_cookie scm;
1497 memset(&scm, 0, sizeof(scm));
1498 scm.pid = UNIXCB(skb).pid;
1500 unix_detach_fds(&scm, skb);
1502 /* Alas, it calls VFS */
1503 /* So fscking what? fput() had been SMP-safe since the last Summer */
1509 * The "user->unix_inflight" variable is protected by the garbage
1510 * collection lock, and we just read it locklessly here. If you go
1511 * over the limit, there might be a tiny race in actually noticing
1512 * it across threads. Tough.
1514 static inline bool too_many_unix_fds(struct task_struct *p)
1516 struct user_struct *user = current_user();
1518 if (unlikely(user->unix_inflight > task_rlimit(p, RLIMIT_NOFILE)))
1519 return !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN);
1523 #define MAX_RECURSION_LEVEL 4
1525 static int unix_attach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1528 unsigned char max_level = 0;
1529 int unix_sock_count = 0;
1531 if (too_many_unix_fds(current))
1532 return -ETOOMANYREFS;
1534 for (i = scm->fp->count - 1; i >= 0; i--) {
1535 struct sock *sk = unix_get_socket(scm->fp->fp[i]);
1539 max_level = max(max_level,
1540 unix_sk(sk)->recursion_level);
1543 if (unlikely(max_level > MAX_RECURSION_LEVEL))
1544 return -ETOOMANYREFS;
1547 * Need to duplicate file references for the sake of garbage
1548 * collection. Otherwise a socket in the fps might become a
1549 * candidate for GC while the skb is not yet queued.
1551 UNIXCB(skb).fp = scm_fp_dup(scm->fp);
1552 if (!UNIXCB(skb).fp)
1555 for (i = scm->fp->count - 1; i >= 0; i--)
1556 unix_inflight(scm->fp->user, scm->fp->fp[i]);
1560 static int unix_scm_to_skb(struct scm_cookie *scm, struct sk_buff *skb, bool send_fds)
1564 UNIXCB(skb).pid = get_pid(scm->pid);
1565 UNIXCB(skb).uid = scm->creds.uid;
1566 UNIXCB(skb).gid = scm->creds.gid;
1567 UNIXCB(skb).fp = NULL;
1568 unix_get_secdata(scm, skb);
1569 if (scm->fp && send_fds)
1570 err = unix_attach_fds(scm, skb);
1572 skb->destructor = unix_destruct_scm;
1576 static bool unix_passcred_enabled(const struct socket *sock,
1577 const struct sock *other)
1579 return test_bit(SOCK_PASSCRED, &sock->flags) ||
1580 !other->sk_socket ||
1581 test_bit(SOCK_PASSCRED, &other->sk_socket->flags);
1585 * Some apps rely on write() giving SCM_CREDENTIALS
1586 * We include credentials if source or destination socket
1587 * asserted SOCK_PASSCRED.
1589 static void maybe_add_creds(struct sk_buff *skb, const struct socket *sock,
1590 const struct sock *other)
1592 if (UNIXCB(skb).pid)
1594 if (unix_passcred_enabled(sock, other)) {
1595 UNIXCB(skb).pid = get_pid(task_tgid(current));
1596 current_uid_gid(&UNIXCB(skb).uid, &UNIXCB(skb).gid);
1600 static int maybe_init_creds(struct scm_cookie *scm,
1601 struct socket *socket,
1602 const struct sock *other)
1605 struct msghdr msg = { .msg_controllen = 0 };
1607 err = scm_send(socket, &msg, scm, false);
1611 if (unix_passcred_enabled(socket, other)) {
1612 scm->pid = get_pid(task_tgid(current));
1613 current_uid_gid(&scm->creds.uid, &scm->creds.gid);
1618 static bool unix_skb_scm_eq(struct sk_buff *skb,
1619 struct scm_cookie *scm)
1621 const struct unix_skb_parms *u = &UNIXCB(skb);
1623 return u->pid == scm->pid &&
1624 uid_eq(u->uid, scm->creds.uid) &&
1625 gid_eq(u->gid, scm->creds.gid) &&
1626 unix_secdata_eq(scm, skb);
1630 * Send AF_UNIX data.
1633 static int unix_dgram_sendmsg(struct socket *sock, struct msghdr *msg,
1636 struct sock *sk = sock->sk;
1637 struct net *net = sock_net(sk);
1638 struct unix_sock *u = unix_sk(sk);
1639 DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr, msg->msg_name);
1640 struct sock *other = NULL;
1641 int namelen = 0; /* fake GCC */
1644 struct sk_buff *skb;
1646 struct scm_cookie scm;
1652 err = scm_send(sock, msg, &scm, false);
1657 if (msg->msg_flags&MSG_OOB)
1660 if (msg->msg_namelen) {
1661 err = unix_mkname(sunaddr, msg->msg_namelen, &hash);
1668 other = unix_peer_get(sk);
1673 if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr
1674 && (err = unix_autobind(sock)) != 0)
1678 if (len > sk->sk_sndbuf - 32)
1681 if (len > SKB_MAX_ALLOC) {
1682 data_len = min_t(size_t,
1683 len - SKB_MAX_ALLOC,
1684 MAX_SKB_FRAGS * PAGE_SIZE);
1685 data_len = PAGE_ALIGN(data_len);
1687 BUILD_BUG_ON(SKB_MAX_ALLOC < PAGE_SIZE);
1690 skb = sock_alloc_send_pskb(sk, len - data_len, data_len,
1691 msg->msg_flags & MSG_DONTWAIT, &err,
1692 PAGE_ALLOC_COSTLY_ORDER);
1696 err = unix_scm_to_skb(&scm, skb, true);
1699 max_level = err + 1;
1701 skb_put(skb, len - data_len);
1702 skb->data_len = data_len;
1704 err = skb_copy_datagram_from_iter(skb, 0, &msg->msg_iter, len);
1708 timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1713 if (sunaddr == NULL)
1716 other = unix_find_other(net, sunaddr, namelen, sk->sk_type,
1722 if (sk_filter(other, skb) < 0) {
1723 /* Toss the packet but do not return any error to the sender */
1729 unix_state_lock(other);
1732 if (!unix_may_send(sk, other))
1735 if (unlikely(sock_flag(other, SOCK_DEAD))) {
1737 * Check with 1003.1g - what should
1740 unix_state_unlock(other);
1744 unix_state_lock(sk);
1747 if (unix_peer(sk) == other) {
1748 unix_peer(sk) = NULL;
1749 unix_dgram_peer_wake_disconnect_wakeup(sk, other);
1751 unix_state_unlock(sk);
1753 unix_dgram_disconnected(sk, other);
1755 err = -ECONNREFUSED;
1757 unix_state_unlock(sk);
1767 if (other->sk_shutdown & RCV_SHUTDOWN)
1770 if (sk->sk_type != SOCK_SEQPACKET) {
1771 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1776 /* other == sk && unix_peer(other) != sk if
1777 * - unix_peer(sk) == NULL, destination address bound to sk
1778 * - unix_peer(sk) == sk by time of get but disconnected before lock
1781 unlikely(unix_peer(other) != sk && unix_recvq_full(other))) {
1783 timeo = unix_wait_for_peer(other, timeo);
1785 err = sock_intr_errno(timeo);
1786 if (signal_pending(current))
1793 unix_state_unlock(other);
1794 unix_state_double_lock(sk, other);
1797 if (unix_peer(sk) != other ||
1798 unix_dgram_peer_wake_me(sk, other)) {
1806 goto restart_locked;
1810 if (unlikely(sk_locked))
1811 unix_state_unlock(sk);
1813 if (sock_flag(other, SOCK_RCVTSTAMP))
1814 __net_timestamp(skb);
1815 maybe_add_creds(skb, sock, other);
1816 skb_queue_tail(&other->sk_receive_queue, skb);
1817 if (max_level > unix_sk(other)->recursion_level)
1818 unix_sk(other)->recursion_level = max_level;
1819 unix_state_unlock(other);
1820 other->sk_data_ready(other);
1827 unix_state_unlock(sk);
1828 unix_state_unlock(other);
1838 /* We use paged skbs for stream sockets, and limit occupancy to 32768
1839 * bytes, and a minimun of a full page.
1841 #define UNIX_SKB_FRAGS_SZ (PAGE_SIZE << get_order(32768))
1843 static int unix_stream_sendmsg(struct socket *sock, struct msghdr *msg,
1846 struct sock *sk = sock->sk;
1847 struct sock *other = NULL;
1849 struct sk_buff *skb;
1851 struct scm_cookie scm;
1852 bool fds_sent = false;
1857 err = scm_send(sock, msg, &scm, false);
1862 if (msg->msg_flags&MSG_OOB)
1865 if (msg->msg_namelen) {
1866 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
1870 other = unix_peer(sk);
1875 if (sk->sk_shutdown & SEND_SHUTDOWN)
1878 while (sent < len) {
1881 /* Keep two messages in the pipe so it schedules better */
1882 size = min_t(int, size, (sk->sk_sndbuf >> 1) - 64);
1884 /* allow fallback to order-0 allocations */
1885 size = min_t(int, size, SKB_MAX_HEAD(0) + UNIX_SKB_FRAGS_SZ);
1887 data_len = max_t(int, 0, size - SKB_MAX_HEAD(0));
1889 data_len = min_t(size_t, size, PAGE_ALIGN(data_len));
1891 skb = sock_alloc_send_pskb(sk, size - data_len, data_len,
1892 msg->msg_flags & MSG_DONTWAIT, &err,
1893 get_order(UNIX_SKB_FRAGS_SZ));
1897 /* Only send the fds in the first buffer */
1898 err = unix_scm_to_skb(&scm, skb, !fds_sent);
1903 max_level = err + 1;
1906 skb_put(skb, size - data_len);
1907 skb->data_len = data_len;
1909 err = skb_copy_datagram_from_iter(skb, 0, &msg->msg_iter, size);
1915 unix_state_lock(other);
1917 if (sock_flag(other, SOCK_DEAD) ||
1918 (other->sk_shutdown & RCV_SHUTDOWN))
1921 maybe_add_creds(skb, sock, other);
1922 skb_queue_tail(&other->sk_receive_queue, skb);
1923 if (max_level > unix_sk(other)->recursion_level)
1924 unix_sk(other)->recursion_level = max_level;
1925 unix_state_unlock(other);
1926 other->sk_data_ready(other);
1935 unix_state_unlock(other);
1938 if (sent == 0 && !(msg->msg_flags&MSG_NOSIGNAL))
1939 send_sig(SIGPIPE, current, 0);
1943 return sent ? : err;
1946 static ssize_t unix_stream_sendpage(struct socket *socket, struct page *page,
1947 int offset, size_t size, int flags)
1950 bool send_sigpipe = false;
1951 bool init_scm = true;
1952 struct scm_cookie scm;
1953 struct sock *other, *sk = socket->sk;
1954 struct sk_buff *skb, *newskb = NULL, *tail = NULL;
1956 if (flags & MSG_OOB)
1959 other = unix_peer(sk);
1960 if (!other || sk->sk_state != TCP_ESTABLISHED)
1965 unix_state_unlock(other);
1966 mutex_unlock(&unix_sk(other)->iolock);
1967 newskb = sock_alloc_send_pskb(sk, 0, 0, flags & MSG_DONTWAIT,
1973 /* we must acquire iolock as we modify already present
1974 * skbs in the sk_receive_queue and mess with skb->len
1976 err = mutex_lock_interruptible(&unix_sk(other)->iolock);
1978 err = flags & MSG_DONTWAIT ? -EAGAIN : -ERESTARTSYS;
1982 if (sk->sk_shutdown & SEND_SHUTDOWN) {
1984 send_sigpipe = true;
1988 unix_state_lock(other);
1990 if (sock_flag(other, SOCK_DEAD) ||
1991 other->sk_shutdown & RCV_SHUTDOWN) {
1993 send_sigpipe = true;
1994 goto err_state_unlock;
1998 err = maybe_init_creds(&scm, socket, other);
2000 goto err_state_unlock;
2004 skb = skb_peek_tail(&other->sk_receive_queue);
2005 if (tail && tail == skb) {
2007 } else if (!skb || !unix_skb_scm_eq(skb, &scm)) {
2014 } else if (newskb) {
2015 /* this is fast path, we don't necessarily need to
2016 * call to kfree_skb even though with newskb == NULL
2017 * this - does no harm
2019 consume_skb(newskb);
2023 if (skb_append_pagefrags(skb, page, offset, size)) {
2029 skb->data_len += size;
2030 skb->truesize += size;
2031 atomic_add(size, &sk->sk_wmem_alloc);
2034 err = unix_scm_to_skb(&scm, skb, false);
2036 goto err_state_unlock;
2037 spin_lock(&other->sk_receive_queue.lock);
2038 __skb_queue_tail(&other->sk_receive_queue, newskb);
2039 spin_unlock(&other->sk_receive_queue.lock);
2042 unix_state_unlock(other);
2043 mutex_unlock(&unix_sk(other)->iolock);
2045 other->sk_data_ready(other);
2050 unix_state_unlock(other);
2052 mutex_unlock(&unix_sk(other)->iolock);
2055 if (send_sigpipe && !(flags & MSG_NOSIGNAL))
2056 send_sig(SIGPIPE, current, 0);
2062 static int unix_seqpacket_sendmsg(struct socket *sock, struct msghdr *msg,
2066 struct sock *sk = sock->sk;
2068 err = sock_error(sk);
2072 if (sk->sk_state != TCP_ESTABLISHED)
2075 if (msg->msg_namelen)
2076 msg->msg_namelen = 0;
2078 return unix_dgram_sendmsg(sock, msg, len);
2081 static int unix_seqpacket_recvmsg(struct socket *sock, struct msghdr *msg,
2082 size_t size, int flags)
2084 struct sock *sk = sock->sk;
2086 if (sk->sk_state != TCP_ESTABLISHED)
2089 return unix_dgram_recvmsg(sock, msg, size, flags);
2092 static void unix_copy_addr(struct msghdr *msg, struct sock *sk)
2094 struct unix_sock *u = unix_sk(sk);
2097 msg->msg_namelen = u->addr->len;
2098 memcpy(msg->msg_name, u->addr->name, u->addr->len);
2102 static int unix_dgram_recvmsg(struct socket *sock, struct msghdr *msg,
2103 size_t size, int flags)
2105 struct scm_cookie scm;
2106 struct sock *sk = sock->sk;
2107 struct unix_sock *u = unix_sk(sk);
2108 int noblock = flags & MSG_DONTWAIT;
2109 struct sk_buff *skb;
2117 err = mutex_lock_interruptible(&u->iolock);
2118 if (unlikely(err)) {
2119 /* recvmsg() in non blocking mode is supposed to return -EAGAIN
2120 * sk_rcvtimeo is not honored by mutex_lock_interruptible()
2122 err = noblock ? -EAGAIN : -ERESTARTSYS;
2126 skip = sk_peek_offset(sk, flags);
2128 skb = __skb_recv_datagram(sk, flags, &peeked, &skip, &err);
2130 unix_state_lock(sk);
2131 /* Signal EOF on disconnected non-blocking SEQPACKET socket. */
2132 if (sk->sk_type == SOCK_SEQPACKET && err == -EAGAIN &&
2133 (sk->sk_shutdown & RCV_SHUTDOWN))
2135 unix_state_unlock(sk);
2139 wake_up_interruptible_sync_poll(&u->peer_wait,
2140 POLLOUT | POLLWRNORM | POLLWRBAND);
2143 unix_copy_addr(msg, skb->sk);
2145 if (size > skb->len - skip)
2146 size = skb->len - skip;
2147 else if (size < skb->len - skip)
2148 msg->msg_flags |= MSG_TRUNC;
2150 err = skb_copy_datagram_msg(skb, skip, msg, size);
2154 if (sock_flag(sk, SOCK_RCVTSTAMP))
2155 __sock_recv_timestamp(msg, sk, skb);
2157 memset(&scm, 0, sizeof(scm));
2159 scm_set_cred(&scm, UNIXCB(skb).pid, UNIXCB(skb).uid, UNIXCB(skb).gid);
2160 unix_set_secdata(&scm, skb);
2162 if (!(flags & MSG_PEEK)) {
2164 unix_detach_fds(&scm, skb);
2166 sk_peek_offset_bwd(sk, skb->len);
2168 /* It is questionable: on PEEK we could:
2169 - do not return fds - good, but too simple 8)
2170 - return fds, and do not return them on read (old strategy,
2172 - clone fds (I chose it for now, it is the most universal
2175 POSIX 1003.1g does not actually define this clearly
2176 at all. POSIX 1003.1g doesn't define a lot of things
2181 sk_peek_offset_fwd(sk, size);
2184 scm.fp = scm_fp_dup(UNIXCB(skb).fp);
2186 err = (flags & MSG_TRUNC) ? skb->len - skip : size;
2188 scm_recv(sock, msg, &scm, flags);
2191 skb_free_datagram(sk, skb);
2193 mutex_unlock(&u->iolock);
2199 * Sleep until more data has arrived. But check for races..
2201 static long unix_stream_data_wait(struct sock *sk, long timeo,
2202 struct sk_buff *last, unsigned int last_len,
2205 struct sk_buff *tail;
2208 unix_state_lock(sk);
2211 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2213 tail = skb_peek_tail(&sk->sk_receive_queue);
2215 (tail && tail->len != last_len) ||
2217 (sk->sk_shutdown & RCV_SHUTDOWN) ||
2218 signal_pending(current) ||
2222 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2223 unix_state_unlock(sk);
2225 timeo = freezable_schedule_timeout(timeo);
2227 timeo = schedule_timeout(timeo);
2228 unix_state_lock(sk);
2230 if (sock_flag(sk, SOCK_DEAD))
2233 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2236 finish_wait(sk_sleep(sk), &wait);
2237 unix_state_unlock(sk);
2241 static unsigned int unix_skb_len(const struct sk_buff *skb)
2243 return skb->len - UNIXCB(skb).consumed;
2246 struct unix_stream_read_state {
2247 int (*recv_actor)(struct sk_buff *, int, int,
2248 struct unix_stream_read_state *);
2249 struct socket *socket;
2251 struct pipe_inode_info *pipe;
2254 unsigned int splice_flags;
2257 static int unix_stream_read_generic(struct unix_stream_read_state *state,
2260 struct scm_cookie scm;
2261 struct socket *sock = state->socket;
2262 struct sock *sk = sock->sk;
2263 struct unix_sock *u = unix_sk(sk);
2265 int flags = state->flags;
2266 int noblock = flags & MSG_DONTWAIT;
2267 bool check_creds = false;
2272 size_t size = state->size;
2273 unsigned int last_len;
2275 if (unlikely(sk->sk_state != TCP_ESTABLISHED)) {
2280 if (unlikely(flags & MSG_OOB)) {
2285 target = sock_rcvlowat(sk, flags & MSG_WAITALL, size);
2286 timeo = sock_rcvtimeo(sk, noblock);
2288 memset(&scm, 0, sizeof(scm));
2290 /* Lock the socket to prevent queue disordering
2291 * while sleeps in memcpy_tomsg
2293 mutex_lock(&u->iolock);
2295 if (flags & MSG_PEEK)
2296 skip = sk_peek_offset(sk, flags);
2303 struct sk_buff *skb, *last;
2305 unix_state_lock(sk);
2306 if (sock_flag(sk, SOCK_DEAD)) {
2310 last = skb = skb_peek(&sk->sk_receive_queue);
2311 last_len = last ? last->len : 0;
2314 unix_sk(sk)->recursion_level = 0;
2315 if (copied >= target)
2319 * POSIX 1003.1g mandates this order.
2322 err = sock_error(sk);
2325 if (sk->sk_shutdown & RCV_SHUTDOWN)
2328 unix_state_unlock(sk);
2334 mutex_unlock(&u->iolock);
2336 timeo = unix_stream_data_wait(sk, timeo, last,
2337 last_len, freezable);
2339 if (signal_pending(current)) {
2340 err = sock_intr_errno(timeo);
2345 mutex_lock(&u->iolock);
2348 unix_state_unlock(sk);
2352 while (skip >= unix_skb_len(skb)) {
2353 skip -= unix_skb_len(skb);
2355 last_len = skb->len;
2356 skb = skb_peek_next(skb, &sk->sk_receive_queue);
2361 unix_state_unlock(sk);
2364 /* Never glue messages from different writers */
2365 if (!unix_skb_scm_eq(skb, &scm))
2367 } else if (test_bit(SOCK_PASSCRED, &sock->flags)) {
2368 /* Copy credentials */
2369 scm_set_cred(&scm, UNIXCB(skb).pid, UNIXCB(skb).uid, UNIXCB(skb).gid);
2370 unix_set_secdata(&scm, skb);
2374 /* Copy address just once */
2375 if (state->msg && state->msg->msg_name) {
2376 DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr,
2377 state->msg->msg_name);
2378 unix_copy_addr(state->msg, skb->sk);
2382 chunk = min_t(unsigned int, unix_skb_len(skb) - skip, size);
2384 chunk = state->recv_actor(skb, skip, chunk, state);
2385 drop_skb = !unix_skb_len(skb);
2386 /* skb is only safe to use if !drop_skb */
2397 /* the skb was touched by a concurrent reader;
2398 * we should not expect anything from this skb
2399 * anymore and assume it invalid - we can be
2400 * sure it was dropped from the socket queue
2402 * let's report a short read
2408 /* Mark read part of skb as used */
2409 if (!(flags & MSG_PEEK)) {
2410 UNIXCB(skb).consumed += chunk;
2412 sk_peek_offset_bwd(sk, chunk);
2415 unix_detach_fds(&scm, skb);
2417 if (unix_skb_len(skb))
2420 skb_unlink(skb, &sk->sk_receive_queue);
2426 /* It is questionable, see note in unix_dgram_recvmsg.
2429 scm.fp = scm_fp_dup(UNIXCB(skb).fp);
2431 sk_peek_offset_fwd(sk, chunk);
2438 last_len = skb->len;
2439 unix_state_lock(sk);
2440 skb = skb_peek_next(skb, &sk->sk_receive_queue);
2443 unix_state_unlock(sk);
2448 mutex_unlock(&u->iolock);
2450 scm_recv(sock, state->msg, &scm, flags);
2454 return copied ? : err;
2457 static int unix_stream_read_actor(struct sk_buff *skb,
2458 int skip, int chunk,
2459 struct unix_stream_read_state *state)
2463 ret = skb_copy_datagram_msg(skb, UNIXCB(skb).consumed + skip,
2465 return ret ?: chunk;
2468 static int unix_stream_recvmsg(struct socket *sock, struct msghdr *msg,
2469 size_t size, int flags)
2471 struct unix_stream_read_state state = {
2472 .recv_actor = unix_stream_read_actor,
2479 return unix_stream_read_generic(&state, true);
2482 static ssize_t skb_unix_socket_splice(struct sock *sk,
2483 struct pipe_inode_info *pipe,
2484 struct splice_pipe_desc *spd)
2487 struct unix_sock *u = unix_sk(sk);
2489 mutex_unlock(&u->iolock);
2490 ret = splice_to_pipe(pipe, spd);
2491 mutex_lock(&u->iolock);
2496 static int unix_stream_splice_actor(struct sk_buff *skb,
2497 int skip, int chunk,
2498 struct unix_stream_read_state *state)
2500 return skb_splice_bits(skb, state->socket->sk,
2501 UNIXCB(skb).consumed + skip,
2502 state->pipe, chunk, state->splice_flags,
2503 skb_unix_socket_splice);
2506 static ssize_t unix_stream_splice_read(struct socket *sock, loff_t *ppos,
2507 struct pipe_inode_info *pipe,
2508 size_t size, unsigned int flags)
2510 struct unix_stream_read_state state = {
2511 .recv_actor = unix_stream_splice_actor,
2515 .splice_flags = flags,
2518 if (unlikely(*ppos))
2521 if (sock->file->f_flags & O_NONBLOCK ||
2522 flags & SPLICE_F_NONBLOCK)
2523 state.flags = MSG_DONTWAIT;
2525 return unix_stream_read_generic(&state, false);
2528 static int unix_shutdown(struct socket *sock, int mode)
2530 struct sock *sk = sock->sk;
2533 if (mode < SHUT_RD || mode > SHUT_RDWR)
2536 * SHUT_RD (0) -> RCV_SHUTDOWN (1)
2537 * SHUT_WR (1) -> SEND_SHUTDOWN (2)
2538 * SHUT_RDWR (2) -> SHUTDOWN_MASK (3)
2542 unix_state_lock(sk);
2543 sk->sk_shutdown |= mode;
2544 other = unix_peer(sk);
2547 unix_state_unlock(sk);
2548 sk->sk_state_change(sk);
2551 (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET)) {
2555 if (mode&RCV_SHUTDOWN)
2556 peer_mode |= SEND_SHUTDOWN;
2557 if (mode&SEND_SHUTDOWN)
2558 peer_mode |= RCV_SHUTDOWN;
2559 unix_state_lock(other);
2560 other->sk_shutdown |= peer_mode;
2561 unix_state_unlock(other);
2562 other->sk_state_change(other);
2563 if (peer_mode == SHUTDOWN_MASK)
2564 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_HUP);
2565 else if (peer_mode & RCV_SHUTDOWN)
2566 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_IN);
2574 long unix_inq_len(struct sock *sk)
2576 struct sk_buff *skb;
2579 if (sk->sk_state == TCP_LISTEN)
2582 spin_lock(&sk->sk_receive_queue.lock);
2583 if (sk->sk_type == SOCK_STREAM ||
2584 sk->sk_type == SOCK_SEQPACKET) {
2585 skb_queue_walk(&sk->sk_receive_queue, skb)
2586 amount += unix_skb_len(skb);
2588 skb = skb_peek(&sk->sk_receive_queue);
2592 spin_unlock(&sk->sk_receive_queue.lock);
2596 EXPORT_SYMBOL_GPL(unix_inq_len);
2598 long unix_outq_len(struct sock *sk)
2600 return sk_wmem_alloc_get(sk);
2602 EXPORT_SYMBOL_GPL(unix_outq_len);
2604 static int unix_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2606 struct sock *sk = sock->sk;
2612 amount = unix_outq_len(sk);
2613 err = put_user(amount, (int __user *)arg);
2616 amount = unix_inq_len(sk);
2620 err = put_user(amount, (int __user *)arg);
2629 static unsigned int unix_poll(struct file *file, struct socket *sock, poll_table *wait)
2631 struct sock *sk = sock->sk;
2634 sock_poll_wait(file, sk_sleep(sk), wait);
2637 /* exceptional events? */
2640 if (sk->sk_shutdown == SHUTDOWN_MASK)
2642 if (sk->sk_shutdown & RCV_SHUTDOWN)
2643 mask |= POLLRDHUP | POLLIN | POLLRDNORM;
2646 if (!skb_queue_empty(&sk->sk_receive_queue))
2647 mask |= POLLIN | POLLRDNORM;
2649 /* Connection-based need to check for termination and startup */
2650 if ((sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) &&
2651 sk->sk_state == TCP_CLOSE)
2655 * we set writable also when the other side has shut down the
2656 * connection. This prevents stuck sockets.
2658 if (unix_writable(sk))
2659 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2664 static unsigned int unix_dgram_poll(struct file *file, struct socket *sock,
2667 struct sock *sk = sock->sk, *other;
2668 unsigned int mask, writable;
2670 sock_poll_wait(file, sk_sleep(sk), wait);
2673 /* exceptional events? */
2674 if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue))
2676 (sock_flag(sk, SOCK_SELECT_ERR_QUEUE) ? POLLPRI : 0);
2678 if (sk->sk_shutdown & RCV_SHUTDOWN)
2679 mask |= POLLRDHUP | POLLIN | POLLRDNORM;
2680 if (sk->sk_shutdown == SHUTDOWN_MASK)
2684 if (!skb_queue_empty(&sk->sk_receive_queue))
2685 mask |= POLLIN | POLLRDNORM;
2687 /* Connection-based need to check for termination and startup */
2688 if (sk->sk_type == SOCK_SEQPACKET) {
2689 if (sk->sk_state == TCP_CLOSE)
2691 /* connection hasn't started yet? */
2692 if (sk->sk_state == TCP_SYN_SENT)
2696 /* No write status requested, avoid expensive OUT tests. */
2697 if (!(poll_requested_events(wait) & (POLLWRBAND|POLLWRNORM|POLLOUT)))
2700 writable = unix_writable(sk);
2702 unix_state_lock(sk);
2704 other = unix_peer(sk);
2705 if (other && unix_peer(other) != sk &&
2706 unix_recvq_full(other) &&
2707 unix_dgram_peer_wake_me(sk, other))
2710 unix_state_unlock(sk);
2714 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2716 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2721 #ifdef CONFIG_PROC_FS
2723 #define BUCKET_SPACE (BITS_PER_LONG - (UNIX_HASH_BITS + 1) - 1)
2725 #define get_bucket(x) ((x) >> BUCKET_SPACE)
2726 #define get_offset(x) ((x) & ((1L << BUCKET_SPACE) - 1))
2727 #define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
2729 static struct sock *unix_from_bucket(struct seq_file *seq, loff_t *pos)
2731 unsigned long offset = get_offset(*pos);
2732 unsigned long bucket = get_bucket(*pos);
2734 unsigned long count = 0;
2736 for (sk = sk_head(&unix_socket_table[bucket]); sk; sk = sk_next(sk)) {
2737 if (sock_net(sk) != seq_file_net(seq))
2739 if (++count == offset)
2746 static struct sock *unix_next_socket(struct seq_file *seq,
2750 unsigned long bucket;
2752 while (sk > (struct sock *)SEQ_START_TOKEN) {
2756 if (sock_net(sk) == seq_file_net(seq))
2761 sk = unix_from_bucket(seq, pos);
2766 bucket = get_bucket(*pos) + 1;
2767 *pos = set_bucket_offset(bucket, 1);
2768 } while (bucket < ARRAY_SIZE(unix_socket_table));
2773 static void *unix_seq_start(struct seq_file *seq, loff_t *pos)
2774 __acquires(unix_table_lock)
2776 spin_lock(&unix_table_lock);
2779 return SEQ_START_TOKEN;
2781 if (get_bucket(*pos) >= ARRAY_SIZE(unix_socket_table))
2784 return unix_next_socket(seq, NULL, pos);
2787 static void *unix_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2790 return unix_next_socket(seq, v, pos);
2793 static void unix_seq_stop(struct seq_file *seq, void *v)
2794 __releases(unix_table_lock)
2796 spin_unlock(&unix_table_lock);
2799 static int unix_seq_show(struct seq_file *seq, void *v)
2802 if (v == SEQ_START_TOKEN)
2803 seq_puts(seq, "Num RefCount Protocol Flags Type St "
2807 struct unix_sock *u = unix_sk(s);
2810 seq_printf(seq, "%pK: %08X %08X %08X %04X %02X %5lu",
2812 atomic_read(&s->sk_refcnt),
2814 s->sk_state == TCP_LISTEN ? __SO_ACCEPTCON : 0,
2817 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTED : SS_UNCONNECTED) :
2818 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTING : SS_DISCONNECTING),
2826 len = u->addr->len - sizeof(short);
2827 if (!UNIX_ABSTRACT(s))
2833 for ( ; i < len; i++)
2834 seq_putc(seq, u->addr->name->sun_path[i]);
2836 unix_state_unlock(s);
2837 seq_putc(seq, '\n');
2843 static const struct seq_operations unix_seq_ops = {
2844 .start = unix_seq_start,
2845 .next = unix_seq_next,
2846 .stop = unix_seq_stop,
2847 .show = unix_seq_show,
2850 static int unix_seq_open(struct inode *inode, struct file *file)
2852 return seq_open_net(inode, file, &unix_seq_ops,
2853 sizeof(struct seq_net_private));
2856 static const struct file_operations unix_seq_fops = {
2857 .owner = THIS_MODULE,
2858 .open = unix_seq_open,
2860 .llseek = seq_lseek,
2861 .release = seq_release_net,
2866 static const struct net_proto_family unix_family_ops = {
2868 .create = unix_create,
2869 .owner = THIS_MODULE,
2873 static int __net_init unix_net_init(struct net *net)
2875 int error = -ENOMEM;
2877 net->unx.sysctl_max_dgram_qlen = 10;
2878 if (unix_sysctl_register(net))
2881 #ifdef CONFIG_PROC_FS
2882 if (!proc_create("unix", 0, net->proc_net, &unix_seq_fops)) {
2883 unix_sysctl_unregister(net);
2892 static void __net_exit unix_net_exit(struct net *net)
2894 unix_sysctl_unregister(net);
2895 remove_proc_entry("unix", net->proc_net);
2898 static struct pernet_operations unix_net_ops = {
2899 .init = unix_net_init,
2900 .exit = unix_net_exit,
2903 static int __init af_unix_init(void)
2907 BUILD_BUG_ON(sizeof(struct unix_skb_parms) > FIELD_SIZEOF(struct sk_buff, cb));
2909 rc = proto_register(&unix_proto, 1);
2911 pr_crit("%s: Cannot create unix_sock SLAB cache!\n", __func__);
2915 sock_register(&unix_family_ops);
2916 register_pernet_subsys(&unix_net_ops);
2921 static void __exit af_unix_exit(void)
2923 sock_unregister(PF_UNIX);
2924 proto_unregister(&unix_proto);
2925 unregister_pernet_subsys(&unix_net_ops);
2928 /* Earlier than device_initcall() so that other drivers invoking
2929 request_module() don't end up in a loop when modprobe tries
2930 to use a UNIX socket. But later than subsys_initcall() because
2931 we depend on stuff initialised there */
2932 fs_initcall(af_unix_init);
2933 module_exit(af_unix_exit);
2935 MODULE_LICENSE("GPL");
2936 MODULE_ALIAS_NETPROTO(PF_UNIX);