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/signal.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 <linux/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>
120 #include <linux/file.h>
124 struct hlist_head unix_socket_table[2 * UNIX_HASH_SIZE];
125 EXPORT_SYMBOL_GPL(unix_socket_table);
126 DEFINE_SPINLOCK(unix_table_lock);
127 EXPORT_SYMBOL_GPL(unix_table_lock);
128 static atomic_long_t unix_nr_socks;
131 static struct hlist_head *unix_sockets_unbound(void *addr)
133 unsigned long hash = (unsigned long)addr;
137 hash %= UNIX_HASH_SIZE;
138 return &unix_socket_table[UNIX_HASH_SIZE + hash];
141 #define UNIX_ABSTRACT(sk) (unix_sk(sk)->addr->hash < UNIX_HASH_SIZE)
143 #ifdef CONFIG_SECURITY_NETWORK
144 static void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb)
146 UNIXCB(skb).secid = scm->secid;
149 static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb)
151 scm->secid = UNIXCB(skb).secid;
154 static inline bool unix_secdata_eq(struct scm_cookie *scm, struct sk_buff *skb)
156 return (scm->secid == UNIXCB(skb).secid);
159 static inline void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb)
162 static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb)
165 static inline bool unix_secdata_eq(struct scm_cookie *scm, struct sk_buff *skb)
169 #endif /* CONFIG_SECURITY_NETWORK */
172 * SMP locking strategy:
173 * hash table is protected with spinlock unix_table_lock
174 * each socket state is protected by separate spin lock.
177 static inline unsigned int unix_hash_fold(__wsum n)
179 unsigned int hash = (__force unsigned int)csum_fold(n);
182 return hash&(UNIX_HASH_SIZE-1);
185 #define unix_peer(sk) (unix_sk(sk)->peer)
187 static inline int unix_our_peer(struct sock *sk, struct sock *osk)
189 return unix_peer(osk) == sk;
192 static inline int unix_may_send(struct sock *sk, struct sock *osk)
194 return unix_peer(osk) == NULL || unix_our_peer(sk, osk);
197 static inline int unix_recvq_full(const struct sock *sk)
199 return skb_queue_len(&sk->sk_receive_queue) > sk->sk_max_ack_backlog;
202 static inline int unix_recvq_full_lockless(const struct sock *sk)
204 return skb_queue_len_lockless(&sk->sk_receive_queue) >
205 READ_ONCE(sk->sk_max_ack_backlog);
208 struct sock *unix_peer_get(struct sock *s)
216 unix_state_unlock(s);
219 EXPORT_SYMBOL_GPL(unix_peer_get);
221 static inline void unix_release_addr(struct unix_address *addr)
223 if (refcount_dec_and_test(&addr->refcnt))
228 * Check unix socket name:
229 * - should be not zero length.
230 * - if started by not zero, should be NULL terminated (FS object)
231 * - if started by zero, it is abstract name.
234 static int unix_mkname(struct sockaddr_un *sunaddr, int len, unsigned int *hashp)
238 if (len <= sizeof(short) || len > sizeof(*sunaddr))
240 if (!sunaddr || sunaddr->sun_family != AF_UNIX)
242 if (sunaddr->sun_path[0]) {
244 * This may look like an off by one error but it is a bit more
245 * subtle. 108 is the longest valid AF_UNIX path for a binding.
246 * sun_path[108] doesn't as such exist. However in kernel space
247 * we are guaranteed that it is a valid memory location in our
248 * kernel address buffer.
250 ((char *)sunaddr)[len] = 0;
251 len = strlen(sunaddr->sun_path)+1+sizeof(short);
255 *hashp = unix_hash_fold(csum_partial(sunaddr, len, 0));
259 static void __unix_remove_socket(struct sock *sk)
261 sk_del_node_init(sk);
264 static void __unix_insert_socket(struct hlist_head *list, struct sock *sk)
266 WARN_ON(!sk_unhashed(sk));
267 sk_add_node(sk, list);
270 static inline void unix_remove_socket(struct sock *sk)
272 spin_lock(&unix_table_lock);
273 __unix_remove_socket(sk);
274 spin_unlock(&unix_table_lock);
277 static inline void unix_insert_socket(struct hlist_head *list, struct sock *sk)
279 spin_lock(&unix_table_lock);
280 __unix_insert_socket(list, sk);
281 spin_unlock(&unix_table_lock);
284 static struct sock *__unix_find_socket_byname(struct net *net,
285 struct sockaddr_un *sunname,
286 int len, int type, unsigned int hash)
290 sk_for_each(s, &unix_socket_table[hash ^ type]) {
291 struct unix_sock *u = unix_sk(s);
293 if (!net_eq(sock_net(s), net))
296 if (u->addr->len == len &&
297 !memcmp(u->addr->name, sunname, len))
305 static inline struct sock *unix_find_socket_byname(struct net *net,
306 struct sockaddr_un *sunname,
312 spin_lock(&unix_table_lock);
313 s = __unix_find_socket_byname(net, sunname, len, type, hash);
316 spin_unlock(&unix_table_lock);
320 static struct sock *unix_find_socket_byinode(struct inode *i)
324 spin_lock(&unix_table_lock);
326 &unix_socket_table[i->i_ino & (UNIX_HASH_SIZE - 1)]) {
327 struct dentry *dentry = unix_sk(s)->path.dentry;
329 if (dentry && d_backing_inode(dentry) == i) {
336 spin_unlock(&unix_table_lock);
340 /* Support code for asymmetrically connected dgram sockets
342 * If a datagram socket is connected to a socket not itself connected
343 * to the first socket (eg, /dev/log), clients may only enqueue more
344 * messages if the present receive queue of the server socket is not
345 * "too large". This means there's a second writeability condition
346 * poll and sendmsg need to test. The dgram recv code will do a wake
347 * up on the peer_wait wait queue of a socket upon reception of a
348 * datagram which needs to be propagated to sleeping would-be writers
349 * since these might not have sent anything so far. This can't be
350 * accomplished via poll_wait because the lifetime of the server
351 * socket might be less than that of its clients if these break their
352 * association with it or if the server socket is closed while clients
353 * are still connected to it and there's no way to inform "a polling
354 * implementation" that it should let go of a certain wait queue
356 * In order to propagate a wake up, a wait_queue_entry_t of the client
357 * socket is enqueued on the peer_wait queue of the server socket
358 * whose wake function does a wake_up on the ordinary client socket
359 * wait queue. This connection is established whenever a write (or
360 * poll for write) hit the flow control condition and broken when the
361 * association to the server socket is dissolved or after a wake up
365 static int unix_dgram_peer_wake_relay(wait_queue_entry_t *q, unsigned mode, int flags,
369 wait_queue_head_t *u_sleep;
371 u = container_of(q, struct unix_sock, peer_wake);
373 __remove_wait_queue(&unix_sk(u->peer_wake.private)->peer_wait,
375 u->peer_wake.private = NULL;
377 /* relaying can only happen while the wq still exists */
378 u_sleep = sk_sleep(&u->sk);
380 wake_up_interruptible_poll(u_sleep, key);
385 static int unix_dgram_peer_wake_connect(struct sock *sk, struct sock *other)
387 struct unix_sock *u, *u_other;
391 u_other = unix_sk(other);
393 spin_lock(&u_other->peer_wait.lock);
395 if (!u->peer_wake.private) {
396 u->peer_wake.private = other;
397 __add_wait_queue(&u_other->peer_wait, &u->peer_wake);
402 spin_unlock(&u_other->peer_wait.lock);
406 static void unix_dgram_peer_wake_disconnect(struct sock *sk,
409 struct unix_sock *u, *u_other;
412 u_other = unix_sk(other);
413 spin_lock(&u_other->peer_wait.lock);
415 if (u->peer_wake.private == other) {
416 __remove_wait_queue(&u_other->peer_wait, &u->peer_wake);
417 u->peer_wake.private = NULL;
420 spin_unlock(&u_other->peer_wait.lock);
423 static void unix_dgram_peer_wake_disconnect_wakeup(struct sock *sk,
426 unix_dgram_peer_wake_disconnect(sk, other);
427 wake_up_interruptible_poll(sk_sleep(sk),
434 * - unix_peer(sk) == other
435 * - association is stable
437 static int unix_dgram_peer_wake_me(struct sock *sk, struct sock *other)
441 connected = unix_dgram_peer_wake_connect(sk, other);
443 if (unix_recvq_full(other))
447 unix_dgram_peer_wake_disconnect(sk, other);
452 static int unix_writable(const struct sock *sk)
454 return sk->sk_state != TCP_LISTEN &&
455 (refcount_read(&sk->sk_wmem_alloc) << 2) <= sk->sk_sndbuf;
458 static void unix_write_space(struct sock *sk)
460 struct socket_wq *wq;
463 if (unix_writable(sk)) {
464 wq = rcu_dereference(sk->sk_wq);
465 if (skwq_has_sleeper(wq))
466 wake_up_interruptible_sync_poll(&wq->wait,
467 POLLOUT | POLLWRNORM | POLLWRBAND);
468 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
473 /* When dgram socket disconnects (or changes its peer), we clear its receive
474 * queue of packets arrived from previous peer. First, it allows to do
475 * flow control based only on wmem_alloc; second, sk connected to peer
476 * may receive messages only from that peer. */
477 static void unix_dgram_disconnected(struct sock *sk, struct sock *other)
479 if (!skb_queue_empty(&sk->sk_receive_queue)) {
480 skb_queue_purge(&sk->sk_receive_queue);
481 wake_up_interruptible_all(&unix_sk(sk)->peer_wait);
483 /* If one link of bidirectional dgram pipe is disconnected,
484 * we signal error. Messages are lost. Do not make this,
485 * when peer was not connected to us.
487 if (!sock_flag(other, SOCK_DEAD) && unix_peer(other) == sk) {
488 other->sk_err = ECONNRESET;
489 other->sk_error_report(other);
494 static void unix_sock_destructor(struct sock *sk)
496 struct unix_sock *u = unix_sk(sk);
498 skb_queue_purge(&sk->sk_receive_queue);
500 WARN_ON(refcount_read(&sk->sk_wmem_alloc));
501 WARN_ON(!sk_unhashed(sk));
502 WARN_ON(sk->sk_socket);
503 if (!sock_flag(sk, SOCK_DEAD)) {
504 pr_info("Attempt to release alive unix socket: %p\n", sk);
509 unix_release_addr(u->addr);
511 atomic_long_dec(&unix_nr_socks);
513 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
515 #ifdef UNIX_REFCNT_DEBUG
516 pr_debug("UNIX %p is destroyed, %ld are still alive.\n", sk,
517 atomic_long_read(&unix_nr_socks));
521 static void unix_release_sock(struct sock *sk, int embrion)
523 struct unix_sock *u = unix_sk(sk);
529 unix_remove_socket(sk);
534 sk->sk_shutdown = SHUTDOWN_MASK;
536 u->path.dentry = NULL;
538 state = sk->sk_state;
539 sk->sk_state = TCP_CLOSE;
541 skpair = unix_peer(sk);
542 unix_peer(sk) = NULL;
544 unix_state_unlock(sk);
546 wake_up_interruptible_all(&u->peer_wait);
548 if (skpair != NULL) {
549 if (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) {
550 unix_state_lock(skpair);
552 skpair->sk_shutdown = SHUTDOWN_MASK;
553 if (!skb_queue_empty(&sk->sk_receive_queue) || embrion)
554 skpair->sk_err = ECONNRESET;
555 unix_state_unlock(skpair);
556 skpair->sk_state_change(skpair);
557 sk_wake_async(skpair, SOCK_WAKE_WAITD, POLL_HUP);
560 unix_dgram_peer_wake_disconnect(sk, skpair);
561 sock_put(skpair); /* It may now die */
564 /* Try to flush out this socket. Throw out buffers at least */
566 while ((skb = skb_dequeue(&sk->sk_receive_queue)) != NULL) {
567 if (state == TCP_LISTEN)
568 unix_release_sock(skb->sk, 1);
569 /* passed fds are erased in the kfree_skb hook */
570 UNIXCB(skb).consumed = skb->len;
579 /* ---- Socket is dead now and most probably destroyed ---- */
582 * Fixme: BSD difference: In BSD all sockets connected to us get
583 * ECONNRESET and we die on the spot. In Linux we behave
584 * like files and pipes do and wait for the last
587 * Can't we simply set sock->err?
589 * What the above comment does talk about? --ANK(980817)
592 if (unix_tot_inflight)
593 unix_gc(); /* Garbage collect fds */
596 static void init_peercred(struct sock *sk)
598 const struct cred *old_cred;
601 spin_lock(&sk->sk_peer_lock);
602 old_pid = sk->sk_peer_pid;
603 old_cred = sk->sk_peer_cred;
604 sk->sk_peer_pid = get_pid(task_tgid(current));
605 sk->sk_peer_cred = get_current_cred();
606 spin_unlock(&sk->sk_peer_lock);
612 static void copy_peercred(struct sock *sk, struct sock *peersk)
614 const struct cred *old_cred;
618 spin_lock(&sk->sk_peer_lock);
619 spin_lock_nested(&peersk->sk_peer_lock, SINGLE_DEPTH_NESTING);
621 spin_lock(&peersk->sk_peer_lock);
622 spin_lock_nested(&sk->sk_peer_lock, SINGLE_DEPTH_NESTING);
624 old_pid = sk->sk_peer_pid;
625 old_cred = sk->sk_peer_cred;
626 sk->sk_peer_pid = get_pid(peersk->sk_peer_pid);
627 sk->sk_peer_cred = get_cred(peersk->sk_peer_cred);
629 spin_unlock(&sk->sk_peer_lock);
630 spin_unlock(&peersk->sk_peer_lock);
636 static int unix_listen(struct socket *sock, int backlog)
639 struct sock *sk = sock->sk;
640 struct unix_sock *u = unix_sk(sk);
641 struct pid *old_pid = NULL;
644 if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
645 goto out; /* Only stream/seqpacket sockets accept */
648 goto out; /* No listens on an unbound socket */
650 if (sk->sk_state != TCP_CLOSE && sk->sk_state != TCP_LISTEN)
652 if (backlog > sk->sk_max_ack_backlog)
653 wake_up_interruptible_all(&u->peer_wait);
654 sk->sk_max_ack_backlog = backlog;
655 sk->sk_state = TCP_LISTEN;
656 /* set credentials so connect can copy them */
661 unix_state_unlock(sk);
667 static int unix_release(struct socket *);
668 static int unix_bind(struct socket *, struct sockaddr *, int);
669 static int unix_stream_connect(struct socket *, struct sockaddr *,
670 int addr_len, int flags);
671 static int unix_socketpair(struct socket *, struct socket *);
672 static int unix_accept(struct socket *, struct socket *, int, bool);
673 static int unix_getname(struct socket *, struct sockaddr *, int *, int);
674 static unsigned int unix_poll(struct file *, struct socket *, poll_table *);
675 static unsigned int unix_dgram_poll(struct file *, struct socket *,
677 static int unix_ioctl(struct socket *, unsigned int, unsigned long);
679 static int unix_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg);
681 static int unix_shutdown(struct socket *, int);
682 static int unix_stream_sendmsg(struct socket *, struct msghdr *, size_t);
683 static int unix_stream_recvmsg(struct socket *, struct msghdr *, size_t, int);
684 static ssize_t unix_stream_sendpage(struct socket *, struct page *, int offset,
685 size_t size, int flags);
686 static ssize_t unix_stream_splice_read(struct socket *, loff_t *ppos,
687 struct pipe_inode_info *, size_t size,
689 static int unix_dgram_sendmsg(struct socket *, struct msghdr *, size_t);
690 static int unix_dgram_recvmsg(struct socket *, struct msghdr *, size_t, int);
691 static int unix_dgram_connect(struct socket *, struct sockaddr *,
693 static int unix_seqpacket_sendmsg(struct socket *, struct msghdr *, size_t);
694 static int unix_seqpacket_recvmsg(struct socket *, struct msghdr *, size_t,
697 static int unix_set_peek_off(struct sock *sk, int val)
699 struct unix_sock *u = unix_sk(sk);
701 if (mutex_lock_interruptible(&u->iolock))
704 sk->sk_peek_off = val;
705 mutex_unlock(&u->iolock);
711 static const struct proto_ops unix_stream_ops = {
713 .owner = THIS_MODULE,
714 .release = unix_release,
716 .connect = unix_stream_connect,
717 .socketpair = unix_socketpair,
718 .accept = unix_accept,
719 .getname = unix_getname,
723 .compat_ioctl = unix_compat_ioctl,
725 .listen = unix_listen,
726 .shutdown = unix_shutdown,
727 .setsockopt = sock_no_setsockopt,
728 .getsockopt = sock_no_getsockopt,
729 .sendmsg = unix_stream_sendmsg,
730 .recvmsg = unix_stream_recvmsg,
731 .mmap = sock_no_mmap,
732 .sendpage = unix_stream_sendpage,
733 .splice_read = unix_stream_splice_read,
734 .set_peek_off = unix_set_peek_off,
737 static const struct proto_ops unix_dgram_ops = {
739 .owner = THIS_MODULE,
740 .release = unix_release,
742 .connect = unix_dgram_connect,
743 .socketpair = unix_socketpair,
744 .accept = sock_no_accept,
745 .getname = unix_getname,
746 .poll = unix_dgram_poll,
749 .compat_ioctl = unix_compat_ioctl,
751 .listen = sock_no_listen,
752 .shutdown = unix_shutdown,
753 .setsockopt = sock_no_setsockopt,
754 .getsockopt = sock_no_getsockopt,
755 .sendmsg = unix_dgram_sendmsg,
756 .recvmsg = unix_dgram_recvmsg,
757 .mmap = sock_no_mmap,
758 .sendpage = sock_no_sendpage,
759 .set_peek_off = unix_set_peek_off,
762 static const struct proto_ops unix_seqpacket_ops = {
764 .owner = THIS_MODULE,
765 .release = unix_release,
767 .connect = unix_stream_connect,
768 .socketpair = unix_socketpair,
769 .accept = unix_accept,
770 .getname = unix_getname,
771 .poll = unix_dgram_poll,
774 .compat_ioctl = unix_compat_ioctl,
776 .listen = unix_listen,
777 .shutdown = unix_shutdown,
778 .setsockopt = sock_no_setsockopt,
779 .getsockopt = sock_no_getsockopt,
780 .sendmsg = unix_seqpacket_sendmsg,
781 .recvmsg = unix_seqpacket_recvmsg,
782 .mmap = sock_no_mmap,
783 .sendpage = sock_no_sendpage,
784 .set_peek_off = unix_set_peek_off,
787 static struct proto unix_proto = {
789 .owner = THIS_MODULE,
790 .obj_size = sizeof(struct unix_sock),
794 * AF_UNIX sockets do not interact with hardware, hence they
795 * dont trigger interrupts - so it's safe for them to have
796 * bh-unsafe locking for their sk_receive_queue.lock. Split off
797 * this special lock-class by reinitializing the spinlock key:
799 static struct lock_class_key af_unix_sk_receive_queue_lock_key;
801 static struct sock *unix_create1(struct net *net, struct socket *sock, int kern)
803 struct sock *sk = NULL;
806 atomic_long_inc(&unix_nr_socks);
807 if (atomic_long_read(&unix_nr_socks) > 2 * get_max_files())
810 sk = sk_alloc(net, PF_UNIX, GFP_KERNEL, &unix_proto, kern);
814 sock_init_data(sock, sk);
815 lockdep_set_class(&sk->sk_receive_queue.lock,
816 &af_unix_sk_receive_queue_lock_key);
818 sk->sk_allocation = GFP_KERNEL_ACCOUNT;
819 sk->sk_write_space = unix_write_space;
820 sk->sk_max_ack_backlog = net->unx.sysctl_max_dgram_qlen;
821 sk->sk_destruct = unix_sock_destructor;
823 u->path.dentry = NULL;
825 spin_lock_init(&u->lock);
826 atomic_long_set(&u->inflight, 0);
827 INIT_LIST_HEAD(&u->link);
828 mutex_init(&u->iolock); /* single task reading lock */
829 mutex_init(&u->bindlock); /* single task binding lock */
830 init_waitqueue_head(&u->peer_wait);
831 init_waitqueue_func_entry(&u->peer_wake, unix_dgram_peer_wake_relay);
832 unix_insert_socket(unix_sockets_unbound(sk), sk);
835 atomic_long_dec(&unix_nr_socks);
838 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
844 static int unix_create(struct net *net, struct socket *sock, int protocol,
847 if (protocol && protocol != PF_UNIX)
848 return -EPROTONOSUPPORT;
850 sock->state = SS_UNCONNECTED;
852 switch (sock->type) {
854 sock->ops = &unix_stream_ops;
857 * Believe it or not BSD has AF_UNIX, SOCK_RAW though
861 sock->type = SOCK_DGRAM;
863 sock->ops = &unix_dgram_ops;
866 sock->ops = &unix_seqpacket_ops;
869 return -ESOCKTNOSUPPORT;
872 return unix_create1(net, sock, kern) ? 0 : -ENOMEM;
875 static int unix_release(struct socket *sock)
877 struct sock *sk = sock->sk;
882 unix_release_sock(sk, 0);
888 static int unix_autobind(struct socket *sock)
890 struct sock *sk = sock->sk;
891 struct net *net = sock_net(sk);
892 struct unix_sock *u = unix_sk(sk);
893 static u32 ordernum = 1;
894 struct unix_address *addr;
896 unsigned int retries = 0;
898 err = mutex_lock_interruptible(&u->bindlock);
907 addr = kzalloc(sizeof(*addr) + sizeof(short) + 16, GFP_KERNEL);
911 addr->name->sun_family = AF_UNIX;
912 refcount_set(&addr->refcnt, 1);
915 addr->len = sprintf(addr->name->sun_path+1, "%05x", ordernum) + 1 + sizeof(short);
916 addr->hash = unix_hash_fold(csum_partial(addr->name, addr->len, 0));
918 spin_lock(&unix_table_lock);
919 ordernum = (ordernum+1)&0xFFFFF;
921 if (__unix_find_socket_byname(net, addr->name, addr->len, sock->type,
923 spin_unlock(&unix_table_lock);
925 * __unix_find_socket_byname() may take long time if many names
926 * are already in use.
929 /* Give up if all names seems to be in use. */
930 if (retries++ == 0xFFFFF) {
937 addr->hash ^= sk->sk_type;
939 __unix_remove_socket(sk);
940 smp_store_release(&u->addr, addr);
941 __unix_insert_socket(&unix_socket_table[addr->hash], sk);
942 spin_unlock(&unix_table_lock);
945 out: mutex_unlock(&u->bindlock);
949 static struct sock *unix_find_other(struct net *net,
950 struct sockaddr_un *sunname, int len,
951 int type, unsigned int hash, int *error)
957 if (sunname->sun_path[0]) {
959 err = kern_path(sunname->sun_path, LOOKUP_FOLLOW, &path);
962 inode = d_backing_inode(path.dentry);
963 err = inode_permission(inode, MAY_WRITE);
968 if (!S_ISSOCK(inode->i_mode))
970 u = unix_find_socket_byinode(inode);
974 if (u->sk_type == type)
980 if (u->sk_type != type) {
986 u = unix_find_socket_byname(net, sunname, len, type, hash);
988 struct dentry *dentry;
989 dentry = unix_sk(u)->path.dentry;
991 touch_atime(&unix_sk(u)->path);
1004 static int unix_mknod(const char *sun_path, umode_t mode, struct path *res)
1006 struct dentry *dentry;
1010 * Get the parent directory, calculate the hash for last
1013 dentry = kern_path_create(AT_FDCWD, sun_path, &path, 0);
1014 err = PTR_ERR(dentry);
1019 * All right, let's create it.
1021 err = security_path_mknod(&path, dentry, mode, 0);
1023 err = vfs_mknod(d_inode(path.dentry), dentry, mode, 0);
1025 res->mnt = mntget(path.mnt);
1026 res->dentry = dget(dentry);
1029 done_path_create(&path, dentry);
1033 static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
1035 struct sock *sk = sock->sk;
1036 struct net *net = sock_net(sk);
1037 struct unix_sock *u = unix_sk(sk);
1038 struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
1039 char *sun_path = sunaddr->sun_path;
1042 struct unix_address *addr;
1043 struct hlist_head *list;
1044 struct path path = { };
1047 if (addr_len < offsetofend(struct sockaddr_un, sun_family) ||
1048 sunaddr->sun_family != AF_UNIX)
1051 if (addr_len == sizeof(short)) {
1052 err = unix_autobind(sock);
1056 err = unix_mkname(sunaddr, addr_len, &hash);
1062 umode_t mode = S_IFSOCK |
1063 (SOCK_INODE(sock)->i_mode & ~current_umask());
1064 err = unix_mknod(sun_path, mode, &path);
1072 err = mutex_lock_interruptible(&u->bindlock);
1081 addr = kmalloc(sizeof(*addr)+addr_len, GFP_KERNEL);
1085 memcpy(addr->name, sunaddr, addr_len);
1086 addr->len = addr_len;
1087 addr->hash = hash ^ sk->sk_type;
1088 refcount_set(&addr->refcnt, 1);
1091 addr->hash = UNIX_HASH_SIZE;
1092 hash = d_backing_inode(path.dentry)->i_ino & (UNIX_HASH_SIZE - 1);
1093 spin_lock(&unix_table_lock);
1095 list = &unix_socket_table[hash];
1097 spin_lock(&unix_table_lock);
1099 if (__unix_find_socket_byname(net, sunaddr, addr_len,
1100 sk->sk_type, hash)) {
1101 unix_release_addr(addr);
1105 list = &unix_socket_table[addr->hash];
1109 __unix_remove_socket(sk);
1110 smp_store_release(&u->addr, addr);
1111 __unix_insert_socket(list, sk);
1114 spin_unlock(&unix_table_lock);
1116 mutex_unlock(&u->bindlock);
1124 static void unix_state_double_lock(struct sock *sk1, struct sock *sk2)
1126 if (unlikely(sk1 == sk2) || !sk2) {
1127 unix_state_lock(sk1);
1131 unix_state_lock(sk1);
1132 unix_state_lock_nested(sk2);
1134 unix_state_lock(sk2);
1135 unix_state_lock_nested(sk1);
1139 static void unix_state_double_unlock(struct sock *sk1, struct sock *sk2)
1141 if (unlikely(sk1 == sk2) || !sk2) {
1142 unix_state_unlock(sk1);
1145 unix_state_unlock(sk1);
1146 unix_state_unlock(sk2);
1149 static int unix_dgram_connect(struct socket *sock, struct sockaddr *addr,
1150 int alen, int flags)
1152 struct sock *sk = sock->sk;
1153 struct net *net = sock_net(sk);
1154 struct sockaddr_un *sunaddr = (struct sockaddr_un *)addr;
1160 if (alen < offsetofend(struct sockaddr, sa_family))
1163 if (addr->sa_family != AF_UNSPEC) {
1164 err = unix_mkname(sunaddr, alen, &hash);
1169 if (test_bit(SOCK_PASSCRED, &sock->flags) &&
1170 !unix_sk(sk)->addr && (err = unix_autobind(sock)) != 0)
1174 other = unix_find_other(net, sunaddr, alen, sock->type, hash, &err);
1178 unix_state_double_lock(sk, other);
1180 /* Apparently VFS overslept socket death. Retry. */
1181 if (sock_flag(other, SOCK_DEAD)) {
1182 unix_state_double_unlock(sk, other);
1188 if (!unix_may_send(sk, other))
1191 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1197 * 1003.1g breaking connected state with AF_UNSPEC
1200 unix_state_double_lock(sk, other);
1204 * If it was connected, reconnect.
1206 if (unix_peer(sk)) {
1207 struct sock *old_peer = unix_peer(sk);
1208 unix_peer(sk) = other;
1209 unix_dgram_peer_wake_disconnect_wakeup(sk, old_peer);
1211 unix_state_double_unlock(sk, other);
1213 if (other != old_peer)
1214 unix_dgram_disconnected(sk, old_peer);
1217 unix_peer(sk) = other;
1218 unix_state_double_unlock(sk, other);
1223 unix_state_double_unlock(sk, other);
1229 static long unix_wait_for_peer(struct sock *other, long timeo)
1231 struct unix_sock *u = unix_sk(other);
1235 prepare_to_wait_exclusive(&u->peer_wait, &wait, TASK_INTERRUPTIBLE);
1237 sched = !sock_flag(other, SOCK_DEAD) &&
1238 !(other->sk_shutdown & RCV_SHUTDOWN) &&
1239 unix_recvq_full(other);
1241 unix_state_unlock(other);
1244 timeo = schedule_timeout(timeo);
1246 finish_wait(&u->peer_wait, &wait);
1250 static int unix_stream_connect(struct socket *sock, struct sockaddr *uaddr,
1251 int addr_len, int flags)
1253 struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
1254 struct sock *sk = sock->sk;
1255 struct net *net = sock_net(sk);
1256 struct unix_sock *u = unix_sk(sk), *newu, *otheru;
1257 struct sock *newsk = NULL;
1258 struct sock *other = NULL;
1259 struct sk_buff *skb = NULL;
1265 err = unix_mkname(sunaddr, addr_len, &hash);
1270 if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr &&
1271 (err = unix_autobind(sock)) != 0)
1274 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
1276 /* First of all allocate resources.
1277 If we will make it after state is locked,
1278 we will have to recheck all again in any case.
1283 /* create new sock for complete connection */
1284 newsk = unix_create1(sock_net(sk), NULL, 0);
1288 /* Allocate skb for sending to listening sock */
1289 skb = sock_wmalloc(newsk, 1, 0, GFP_KERNEL);
1294 /* Find listening sock. */
1295 other = unix_find_other(net, sunaddr, addr_len, sk->sk_type, hash, &err);
1299 /* Latch state of peer */
1300 unix_state_lock(other);
1302 /* Apparently VFS overslept socket death. Retry. */
1303 if (sock_flag(other, SOCK_DEAD)) {
1304 unix_state_unlock(other);
1309 err = -ECONNREFUSED;
1310 if (other->sk_state != TCP_LISTEN)
1312 if (other->sk_shutdown & RCV_SHUTDOWN)
1315 if (unix_recvq_full(other)) {
1320 timeo = unix_wait_for_peer(other, timeo);
1322 err = sock_intr_errno(timeo);
1323 if (signal_pending(current))
1331 It is tricky place. We need to grab our state lock and cannot
1332 drop lock on peer. It is dangerous because deadlock is
1333 possible. Connect to self case and simultaneous
1334 attempt to connect are eliminated by checking socket
1335 state. other is TCP_LISTEN, if sk is TCP_LISTEN we
1336 check this before attempt to grab lock.
1338 Well, and we have to recheck the state after socket locked.
1344 /* This is ok... continue with connect */
1346 case TCP_ESTABLISHED:
1347 /* Socket is already connected */
1355 unix_state_lock_nested(sk);
1357 if (sk->sk_state != st) {
1358 unix_state_unlock(sk);
1359 unix_state_unlock(other);
1364 err = security_unix_stream_connect(sk, other, newsk);
1366 unix_state_unlock(sk);
1370 /* The way is open! Fastly set all the necessary fields... */
1373 unix_peer(newsk) = sk;
1374 newsk->sk_state = TCP_ESTABLISHED;
1375 newsk->sk_type = sk->sk_type;
1376 init_peercred(newsk);
1377 newu = unix_sk(newsk);
1378 RCU_INIT_POINTER(newsk->sk_wq, &newu->peer_wq);
1379 otheru = unix_sk(other);
1381 /* copy address information from listening to new sock
1383 * The contents of *(otheru->addr) and otheru->path
1384 * are seen fully set up here, since we have found
1385 * otheru in hash under unix_table_lock. Insertion
1386 * into the hash chain we'd found it in had been done
1387 * in an earlier critical area protected by unix_table_lock,
1388 * the same one where we'd set *(otheru->addr) contents,
1389 * as well as otheru->path and otheru->addr itself.
1391 * Using smp_store_release() here to set newu->addr
1392 * is enough to make those stores, as well as stores
1393 * to newu->path visible to anyone who gets newu->addr
1394 * by smp_load_acquire(). IOW, the same warranties
1395 * as for unix_sock instances bound in unix_bind() or
1396 * in unix_autobind().
1398 if (otheru->path.dentry) {
1399 path_get(&otheru->path);
1400 newu->path = otheru->path;
1402 refcount_inc(&otheru->addr->refcnt);
1403 smp_store_release(&newu->addr, otheru->addr);
1405 /* Set credentials */
1406 copy_peercred(sk, other);
1408 sock->state = SS_CONNECTED;
1409 sk->sk_state = TCP_ESTABLISHED;
1412 smp_mb__after_atomic(); /* sock_hold() does an atomic_inc() */
1413 unix_peer(sk) = newsk;
1415 unix_state_unlock(sk);
1417 /* take ten and and send info to listening sock */
1418 spin_lock(&other->sk_receive_queue.lock);
1419 __skb_queue_tail(&other->sk_receive_queue, skb);
1420 spin_unlock(&other->sk_receive_queue.lock);
1421 unix_state_unlock(other);
1422 other->sk_data_ready(other);
1428 unix_state_unlock(other);
1433 unix_release_sock(newsk, 0);
1439 static int unix_socketpair(struct socket *socka, struct socket *sockb)
1441 struct sock *ska = socka->sk, *skb = sockb->sk;
1443 /* Join our sockets back to back */
1446 unix_peer(ska) = skb;
1447 unix_peer(skb) = ska;
1451 if (ska->sk_type != SOCK_DGRAM) {
1452 ska->sk_state = TCP_ESTABLISHED;
1453 skb->sk_state = TCP_ESTABLISHED;
1454 socka->state = SS_CONNECTED;
1455 sockb->state = SS_CONNECTED;
1460 static void unix_sock_inherit_flags(const struct socket *old,
1463 if (test_bit(SOCK_PASSCRED, &old->flags))
1464 set_bit(SOCK_PASSCRED, &new->flags);
1465 if (test_bit(SOCK_PASSSEC, &old->flags))
1466 set_bit(SOCK_PASSSEC, &new->flags);
1469 static int unix_accept(struct socket *sock, struct socket *newsock, int flags,
1472 struct sock *sk = sock->sk;
1474 struct sk_buff *skb;
1478 if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
1482 if (sk->sk_state != TCP_LISTEN)
1485 /* If socket state is TCP_LISTEN it cannot change (for now...),
1486 * so that no locks are necessary.
1489 skb = skb_recv_datagram(sk, 0, flags&O_NONBLOCK, &err);
1491 /* This means receive shutdown. */
1498 skb_free_datagram(sk, skb);
1499 wake_up_interruptible(&unix_sk(sk)->peer_wait);
1501 /* attach accepted sock to socket */
1502 unix_state_lock(tsk);
1503 newsock->state = SS_CONNECTED;
1504 unix_sock_inherit_flags(sock, newsock);
1505 sock_graft(tsk, newsock);
1506 unix_state_unlock(tsk);
1514 static int unix_getname(struct socket *sock, struct sockaddr *uaddr, int *uaddr_len, int peer)
1516 struct sock *sk = sock->sk;
1517 struct unix_address *addr;
1518 DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr, uaddr);
1522 sk = unix_peer_get(sk);
1532 addr = smp_load_acquire(&unix_sk(sk)->addr);
1534 sunaddr->sun_family = AF_UNIX;
1535 sunaddr->sun_path[0] = 0;
1536 *uaddr_len = sizeof(short);
1538 *uaddr_len = addr->len;
1539 memcpy(sunaddr, addr->name, *uaddr_len);
1546 static void unix_peek_fds(struct scm_cookie *scm, struct sk_buff *skb)
1548 scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1551 * Garbage collection of unix sockets starts by selecting a set of
1552 * candidate sockets which have reference only from being in flight
1553 * (total_refs == inflight_refs). This condition is checked once during
1554 * the candidate collection phase, and candidates are marked as such, so
1555 * that non-candidates can later be ignored. While inflight_refs is
1556 * protected by unix_gc_lock, total_refs (file count) is not, hence this
1557 * is an instantaneous decision.
1559 * Once a candidate, however, the socket must not be reinstalled into a
1560 * file descriptor while the garbage collection is in progress.
1562 * If the above conditions are met, then the directed graph of
1563 * candidates (*) does not change while unix_gc_lock is held.
1565 * Any operations that changes the file count through file descriptors
1566 * (dup, close, sendmsg) does not change the graph since candidates are
1567 * not installed in fds.
1569 * Dequeing a candidate via recvmsg would install it into an fd, but
1570 * that takes unix_gc_lock to decrement the inflight count, so it's
1571 * serialized with garbage collection.
1573 * MSG_PEEK is special in that it does not change the inflight count,
1574 * yet does install the socket into an fd. The following lock/unlock
1575 * pair is to ensure serialization with garbage collection. It must be
1576 * done between incrementing the file count and installing the file into
1579 * If garbage collection starts after the barrier provided by the
1580 * lock/unlock, then it will see the elevated refcount and not mark this
1581 * as a candidate. If a garbage collection is already in progress
1582 * before the file count was incremented, then the lock/unlock pair will
1583 * ensure that garbage collection is finished before progressing to
1584 * installing the fd.
1586 * (*) A -> B where B is on the queue of A or B is on the queue of C
1587 * which is on the queue of listening socket A.
1589 spin_lock(&unix_gc_lock);
1590 spin_unlock(&unix_gc_lock);
1593 static int unix_scm_to_skb(struct scm_cookie *scm, struct sk_buff *skb, bool send_fds)
1597 UNIXCB(skb).pid = get_pid(scm->pid);
1598 UNIXCB(skb).uid = scm->creds.uid;
1599 UNIXCB(skb).gid = scm->creds.gid;
1600 UNIXCB(skb).fp = NULL;
1601 unix_get_secdata(scm, skb);
1602 if (scm->fp && send_fds)
1603 err = unix_attach_fds(scm, skb);
1605 skb->destructor = unix_destruct_scm;
1609 static bool unix_passcred_enabled(const struct socket *sock,
1610 const struct sock *other)
1612 return test_bit(SOCK_PASSCRED, &sock->flags) ||
1613 !other->sk_socket ||
1614 test_bit(SOCK_PASSCRED, &other->sk_socket->flags);
1618 * Some apps rely on write() giving SCM_CREDENTIALS
1619 * We include credentials if source or destination socket
1620 * asserted SOCK_PASSCRED.
1622 static void maybe_add_creds(struct sk_buff *skb, const struct socket *sock,
1623 const struct sock *other)
1625 if (UNIXCB(skb).pid)
1627 if (unix_passcred_enabled(sock, other)) {
1628 UNIXCB(skb).pid = get_pid(task_tgid(current));
1629 current_uid_gid(&UNIXCB(skb).uid, &UNIXCB(skb).gid);
1633 static int maybe_init_creds(struct scm_cookie *scm,
1634 struct socket *socket,
1635 const struct sock *other)
1638 struct msghdr msg = { .msg_controllen = 0 };
1640 err = scm_send(socket, &msg, scm, false);
1644 if (unix_passcred_enabled(socket, other)) {
1645 scm->pid = get_pid(task_tgid(current));
1646 current_uid_gid(&scm->creds.uid, &scm->creds.gid);
1651 static bool unix_skb_scm_eq(struct sk_buff *skb,
1652 struct scm_cookie *scm)
1654 const struct unix_skb_parms *u = &UNIXCB(skb);
1656 return u->pid == scm->pid &&
1657 uid_eq(u->uid, scm->creds.uid) &&
1658 gid_eq(u->gid, scm->creds.gid) &&
1659 unix_secdata_eq(scm, skb);
1663 * Send AF_UNIX data.
1666 static int unix_dgram_sendmsg(struct socket *sock, struct msghdr *msg,
1669 struct sock *sk = sock->sk;
1670 struct net *net = sock_net(sk);
1671 struct unix_sock *u = unix_sk(sk);
1672 DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr, msg->msg_name);
1673 struct sock *other = NULL;
1674 int namelen = 0; /* fake GCC */
1677 struct sk_buff *skb;
1679 struct scm_cookie scm;
1684 err = scm_send(sock, msg, &scm, false);
1689 if (msg->msg_flags&MSG_OOB)
1692 if (msg->msg_namelen) {
1693 err = unix_mkname(sunaddr, msg->msg_namelen, &hash);
1700 other = unix_peer_get(sk);
1705 if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr
1706 && (err = unix_autobind(sock)) != 0)
1710 if (len > sk->sk_sndbuf - 32)
1713 if (len > SKB_MAX_ALLOC) {
1714 data_len = min_t(size_t,
1715 len - SKB_MAX_ALLOC,
1716 MAX_SKB_FRAGS * PAGE_SIZE);
1717 data_len = PAGE_ALIGN(data_len);
1719 BUILD_BUG_ON(SKB_MAX_ALLOC < PAGE_SIZE);
1722 skb = sock_alloc_send_pskb(sk, len - data_len, data_len,
1723 msg->msg_flags & MSG_DONTWAIT, &err,
1724 PAGE_ALLOC_COSTLY_ORDER);
1728 err = unix_scm_to_skb(&scm, skb, true);
1732 skb_put(skb, len - data_len);
1733 skb->data_len = data_len;
1735 err = skb_copy_datagram_from_iter(skb, 0, &msg->msg_iter, len);
1739 timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1744 if (sunaddr == NULL)
1747 other = unix_find_other(net, sunaddr, namelen, sk->sk_type,
1753 if (sk_filter(other, skb) < 0) {
1754 /* Toss the packet but do not return any error to the sender */
1760 unix_state_lock(other);
1763 if (!unix_may_send(sk, other))
1766 if (unlikely(sock_flag(other, SOCK_DEAD))) {
1768 * Check with 1003.1g - what should
1771 unix_state_unlock(other);
1775 unix_state_lock(sk);
1778 if (unix_peer(sk) == other) {
1779 unix_peer(sk) = NULL;
1780 unix_dgram_peer_wake_disconnect_wakeup(sk, other);
1782 unix_state_unlock(sk);
1784 unix_dgram_disconnected(sk, other);
1786 err = -ECONNREFUSED;
1788 unix_state_unlock(sk);
1798 if (other->sk_shutdown & RCV_SHUTDOWN)
1801 if (sk->sk_type != SOCK_SEQPACKET) {
1802 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1807 /* other == sk && unix_peer(other) != sk if
1808 * - unix_peer(sk) == NULL, destination address bound to sk
1809 * - unix_peer(sk) == sk by time of get but disconnected before lock
1812 unlikely(unix_peer(other) != sk &&
1813 unix_recvq_full_lockless(other))) {
1815 timeo = unix_wait_for_peer(other, timeo);
1817 err = sock_intr_errno(timeo);
1818 if (signal_pending(current))
1825 unix_state_unlock(other);
1826 unix_state_double_lock(sk, other);
1829 if (unix_peer(sk) != other ||
1830 unix_dgram_peer_wake_me(sk, other)) {
1838 goto restart_locked;
1842 if (unlikely(sk_locked))
1843 unix_state_unlock(sk);
1845 if (sock_flag(other, SOCK_RCVTSTAMP))
1846 __net_timestamp(skb);
1847 maybe_add_creds(skb, sock, other);
1848 skb_queue_tail(&other->sk_receive_queue, skb);
1849 unix_state_unlock(other);
1850 other->sk_data_ready(other);
1857 unix_state_unlock(sk);
1858 unix_state_unlock(other);
1868 /* We use paged skbs for stream sockets, and limit occupancy to 32768
1869 * bytes, and a minimun of a full page.
1871 #define UNIX_SKB_FRAGS_SZ (PAGE_SIZE << get_order(32768))
1873 static int unix_stream_sendmsg(struct socket *sock, struct msghdr *msg,
1876 struct sock *sk = sock->sk;
1877 struct sock *other = NULL;
1879 struct sk_buff *skb;
1881 struct scm_cookie scm;
1882 bool fds_sent = false;
1886 err = scm_send(sock, msg, &scm, false);
1891 if (msg->msg_flags&MSG_OOB)
1894 if (msg->msg_namelen) {
1895 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
1899 other = unix_peer(sk);
1904 if (sk->sk_shutdown & SEND_SHUTDOWN)
1907 while (sent < len) {
1910 /* Keep two messages in the pipe so it schedules better */
1911 size = min_t(int, size, (sk->sk_sndbuf >> 1) - 64);
1913 /* allow fallback to order-0 allocations */
1914 size = min_t(int, size, SKB_MAX_HEAD(0) + UNIX_SKB_FRAGS_SZ);
1916 data_len = max_t(int, 0, size - SKB_MAX_HEAD(0));
1918 data_len = min_t(size_t, size, PAGE_ALIGN(data_len));
1920 skb = sock_alloc_send_pskb(sk, size - data_len, data_len,
1921 msg->msg_flags & MSG_DONTWAIT, &err,
1922 get_order(UNIX_SKB_FRAGS_SZ));
1926 /* Only send the fds in the first buffer */
1927 err = unix_scm_to_skb(&scm, skb, !fds_sent);
1934 skb_put(skb, size - data_len);
1935 skb->data_len = data_len;
1937 err = skb_copy_datagram_from_iter(skb, 0, &msg->msg_iter, size);
1943 unix_state_lock(other);
1945 if (sock_flag(other, SOCK_DEAD) ||
1946 (other->sk_shutdown & RCV_SHUTDOWN))
1949 maybe_add_creds(skb, sock, other);
1950 skb_queue_tail(&other->sk_receive_queue, skb);
1951 unix_state_unlock(other);
1952 other->sk_data_ready(other);
1961 unix_state_unlock(other);
1964 if (sent == 0 && !(msg->msg_flags&MSG_NOSIGNAL))
1965 send_sig(SIGPIPE, current, 0);
1969 return sent ? : err;
1972 static ssize_t unix_stream_sendpage(struct socket *socket, struct page *page,
1973 int offset, size_t size, int flags)
1976 bool send_sigpipe = false;
1977 bool init_scm = true;
1978 struct scm_cookie scm;
1979 struct sock *other, *sk = socket->sk;
1980 struct sk_buff *skb, *newskb = NULL, *tail = NULL;
1982 if (flags & MSG_OOB)
1985 other = unix_peer(sk);
1986 if (!other || sk->sk_state != TCP_ESTABLISHED)
1991 unix_state_unlock(other);
1992 mutex_unlock(&unix_sk(other)->iolock);
1993 newskb = sock_alloc_send_pskb(sk, 0, 0, flags & MSG_DONTWAIT,
1999 /* we must acquire iolock as we modify already present
2000 * skbs in the sk_receive_queue and mess with skb->len
2002 err = mutex_lock_interruptible(&unix_sk(other)->iolock);
2004 err = flags & MSG_DONTWAIT ? -EAGAIN : -ERESTARTSYS;
2008 if (sk->sk_shutdown & SEND_SHUTDOWN) {
2010 send_sigpipe = true;
2014 unix_state_lock(other);
2016 if (sock_flag(other, SOCK_DEAD) ||
2017 other->sk_shutdown & RCV_SHUTDOWN) {
2019 send_sigpipe = true;
2020 goto err_state_unlock;
2024 err = maybe_init_creds(&scm, socket, other);
2026 goto err_state_unlock;
2030 skb = skb_peek_tail(&other->sk_receive_queue);
2031 if (tail && tail == skb) {
2033 } else if (!skb || !unix_skb_scm_eq(skb, &scm)) {
2040 } else if (newskb) {
2041 /* this is fast path, we don't necessarily need to
2042 * call to kfree_skb even though with newskb == NULL
2043 * this - does no harm
2045 consume_skb(newskb);
2049 if (skb_append_pagefrags(skb, page, offset, size)) {
2055 skb->data_len += size;
2056 skb->truesize += size;
2057 refcount_add(size, &sk->sk_wmem_alloc);
2060 err = unix_scm_to_skb(&scm, skb, false);
2062 goto err_state_unlock;
2063 spin_lock(&other->sk_receive_queue.lock);
2064 __skb_queue_tail(&other->sk_receive_queue, newskb);
2065 spin_unlock(&other->sk_receive_queue.lock);
2068 unix_state_unlock(other);
2069 mutex_unlock(&unix_sk(other)->iolock);
2071 other->sk_data_ready(other);
2076 unix_state_unlock(other);
2078 mutex_unlock(&unix_sk(other)->iolock);
2081 if (send_sigpipe && !(flags & MSG_NOSIGNAL))
2082 send_sig(SIGPIPE, current, 0);
2088 static int unix_seqpacket_sendmsg(struct socket *sock, struct msghdr *msg,
2092 struct sock *sk = sock->sk;
2094 err = sock_error(sk);
2098 if (sk->sk_state != TCP_ESTABLISHED)
2101 if (msg->msg_namelen)
2102 msg->msg_namelen = 0;
2104 return unix_dgram_sendmsg(sock, msg, len);
2107 static int unix_seqpacket_recvmsg(struct socket *sock, struct msghdr *msg,
2108 size_t size, int flags)
2110 struct sock *sk = sock->sk;
2112 if (sk->sk_state != TCP_ESTABLISHED)
2115 return unix_dgram_recvmsg(sock, msg, size, flags);
2118 static void unix_copy_addr(struct msghdr *msg, struct sock *sk)
2120 struct unix_address *addr = smp_load_acquire(&unix_sk(sk)->addr);
2123 msg->msg_namelen = addr->len;
2124 memcpy(msg->msg_name, addr->name, addr->len);
2128 static int unix_dgram_recvmsg(struct socket *sock, struct msghdr *msg,
2129 size_t size, int flags)
2131 struct scm_cookie scm;
2132 struct sock *sk = sock->sk;
2133 struct unix_sock *u = unix_sk(sk);
2134 struct sk_buff *skb, *last;
2143 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
2146 mutex_lock(&u->iolock);
2148 skip = sk_peek_offset(sk, flags);
2149 skb = __skb_try_recv_datagram(sk, flags, NULL, &peeked, &skip,
2154 mutex_unlock(&u->iolock);
2159 !__skb_wait_for_more_packets(sk, &err, &timeo, last));
2161 if (!skb) { /* implies iolock unlocked */
2162 unix_state_lock(sk);
2163 /* Signal EOF on disconnected non-blocking SEQPACKET socket. */
2164 if (sk->sk_type == SOCK_SEQPACKET && err == -EAGAIN &&
2165 (sk->sk_shutdown & RCV_SHUTDOWN))
2167 unix_state_unlock(sk);
2171 if (wq_has_sleeper(&u->peer_wait))
2172 wake_up_interruptible_sync_poll(&u->peer_wait,
2173 POLLOUT | POLLWRNORM |
2177 unix_copy_addr(msg, skb->sk);
2179 if (size > skb->len - skip)
2180 size = skb->len - skip;
2181 else if (size < skb->len - skip)
2182 msg->msg_flags |= MSG_TRUNC;
2184 err = skb_copy_datagram_msg(skb, skip, msg, size);
2188 if (sock_flag(sk, SOCK_RCVTSTAMP))
2189 __sock_recv_timestamp(msg, sk, skb);
2191 memset(&scm, 0, sizeof(scm));
2193 scm_set_cred(&scm, UNIXCB(skb).pid, UNIXCB(skb).uid, UNIXCB(skb).gid);
2194 unix_set_secdata(&scm, skb);
2196 if (!(flags & MSG_PEEK)) {
2198 unix_detach_fds(&scm, skb);
2200 sk_peek_offset_bwd(sk, skb->len);
2202 /* It is questionable: on PEEK we could:
2203 - do not return fds - good, but too simple 8)
2204 - return fds, and do not return them on read (old strategy,
2206 - clone fds (I chose it for now, it is the most universal
2209 POSIX 1003.1g does not actually define this clearly
2210 at all. POSIX 1003.1g doesn't define a lot of things
2215 sk_peek_offset_fwd(sk, size);
2218 unix_peek_fds(&scm, skb);
2220 err = (flags & MSG_TRUNC) ? skb->len - skip : size;
2222 scm_recv(sock, msg, &scm, flags);
2225 skb_free_datagram(sk, skb);
2226 mutex_unlock(&u->iolock);
2232 * Sleep until more data has arrived. But check for races..
2234 static long unix_stream_data_wait(struct sock *sk, long timeo,
2235 struct sk_buff *last, unsigned int last_len,
2238 struct sk_buff *tail;
2241 unix_state_lock(sk);
2244 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2246 tail = skb_peek_tail(&sk->sk_receive_queue);
2248 (tail && tail->len != last_len) ||
2250 (sk->sk_shutdown & RCV_SHUTDOWN) ||
2251 signal_pending(current) ||
2255 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2256 unix_state_unlock(sk);
2258 timeo = freezable_schedule_timeout(timeo);
2260 timeo = schedule_timeout(timeo);
2261 unix_state_lock(sk);
2263 if (sock_flag(sk, SOCK_DEAD))
2266 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2269 finish_wait(sk_sleep(sk), &wait);
2270 unix_state_unlock(sk);
2274 static unsigned int unix_skb_len(const struct sk_buff *skb)
2276 return skb->len - UNIXCB(skb).consumed;
2279 struct unix_stream_read_state {
2280 int (*recv_actor)(struct sk_buff *, int, int,
2281 struct unix_stream_read_state *);
2282 struct socket *socket;
2284 struct pipe_inode_info *pipe;
2287 unsigned int splice_flags;
2290 static int unix_stream_read_generic(struct unix_stream_read_state *state,
2293 struct scm_cookie scm;
2294 struct socket *sock = state->socket;
2295 struct sock *sk = sock->sk;
2296 struct unix_sock *u = unix_sk(sk);
2298 int flags = state->flags;
2299 int noblock = flags & MSG_DONTWAIT;
2300 bool check_creds = false;
2305 size_t size = state->size;
2306 unsigned int last_len;
2308 if (unlikely(sk->sk_state != TCP_ESTABLISHED)) {
2313 if (unlikely(flags & MSG_OOB)) {
2318 target = sock_rcvlowat(sk, flags & MSG_WAITALL, size);
2319 timeo = sock_rcvtimeo(sk, noblock);
2321 memset(&scm, 0, sizeof(scm));
2323 /* Lock the socket to prevent queue disordering
2324 * while sleeps in memcpy_tomsg
2326 mutex_lock(&u->iolock);
2328 skip = max(sk_peek_offset(sk, flags), 0);
2333 struct sk_buff *skb, *last;
2336 unix_state_lock(sk);
2337 if (sock_flag(sk, SOCK_DEAD)) {
2341 last = skb = skb_peek(&sk->sk_receive_queue);
2342 last_len = last ? last->len : 0;
2345 if (copied >= target)
2349 * POSIX 1003.1g mandates this order.
2352 err = sock_error(sk);
2355 if (sk->sk_shutdown & RCV_SHUTDOWN)
2358 unix_state_unlock(sk);
2364 mutex_unlock(&u->iolock);
2366 timeo = unix_stream_data_wait(sk, timeo, last,
2367 last_len, freezable);
2369 if (signal_pending(current)) {
2370 err = sock_intr_errno(timeo);
2375 mutex_lock(&u->iolock);
2378 unix_state_unlock(sk);
2382 while (skip >= unix_skb_len(skb)) {
2383 skip -= unix_skb_len(skb);
2385 last_len = skb->len;
2386 skb = skb_peek_next(skb, &sk->sk_receive_queue);
2391 unix_state_unlock(sk);
2394 /* Never glue messages from different writers */
2395 if (!unix_skb_scm_eq(skb, &scm))
2397 } else if (test_bit(SOCK_PASSCRED, &sock->flags)) {
2398 /* Copy credentials */
2399 scm_set_cred(&scm, UNIXCB(skb).pid, UNIXCB(skb).uid, UNIXCB(skb).gid);
2400 unix_set_secdata(&scm, skb);
2404 /* Copy address just once */
2405 if (state->msg && state->msg->msg_name) {
2406 DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr,
2407 state->msg->msg_name);
2408 unix_copy_addr(state->msg, skb->sk);
2412 chunk = min_t(unsigned int, unix_skb_len(skb) - skip, size);
2414 chunk = state->recv_actor(skb, skip, chunk, state);
2415 drop_skb = !unix_skb_len(skb);
2416 /* skb is only safe to use if !drop_skb */
2427 /* the skb was touched by a concurrent reader;
2428 * we should not expect anything from this skb
2429 * anymore and assume it invalid - we can be
2430 * sure it was dropped from the socket queue
2432 * let's report a short read
2438 /* Mark read part of skb as used */
2439 if (!(flags & MSG_PEEK)) {
2440 UNIXCB(skb).consumed += chunk;
2442 sk_peek_offset_bwd(sk, chunk);
2445 unix_detach_fds(&scm, skb);
2447 if (unix_skb_len(skb))
2450 skb_unlink(skb, &sk->sk_receive_queue);
2456 /* It is questionable, see note in unix_dgram_recvmsg.
2459 unix_peek_fds(&scm, skb);
2461 sk_peek_offset_fwd(sk, chunk);
2468 last_len = skb->len;
2469 unix_state_lock(sk);
2470 skb = skb_peek_next(skb, &sk->sk_receive_queue);
2473 unix_state_unlock(sk);
2478 mutex_unlock(&u->iolock);
2480 scm_recv(sock, state->msg, &scm, flags);
2484 return copied ? : err;
2487 static int unix_stream_read_actor(struct sk_buff *skb,
2488 int skip, int chunk,
2489 struct unix_stream_read_state *state)
2493 ret = skb_copy_datagram_msg(skb, UNIXCB(skb).consumed + skip,
2495 return ret ?: chunk;
2498 static int unix_stream_recvmsg(struct socket *sock, struct msghdr *msg,
2499 size_t size, int flags)
2501 struct unix_stream_read_state state = {
2502 .recv_actor = unix_stream_read_actor,
2509 return unix_stream_read_generic(&state, true);
2512 static int unix_stream_splice_actor(struct sk_buff *skb,
2513 int skip, int chunk,
2514 struct unix_stream_read_state *state)
2516 return skb_splice_bits(skb, state->socket->sk,
2517 UNIXCB(skb).consumed + skip,
2518 state->pipe, chunk, state->splice_flags);
2521 static ssize_t unix_stream_splice_read(struct socket *sock, loff_t *ppos,
2522 struct pipe_inode_info *pipe,
2523 size_t size, unsigned int flags)
2525 struct unix_stream_read_state state = {
2526 .recv_actor = unix_stream_splice_actor,
2530 .splice_flags = flags,
2533 if (unlikely(*ppos))
2536 if (sock->file->f_flags & O_NONBLOCK ||
2537 flags & SPLICE_F_NONBLOCK)
2538 state.flags = MSG_DONTWAIT;
2540 return unix_stream_read_generic(&state, false);
2543 static int unix_shutdown(struct socket *sock, int mode)
2545 struct sock *sk = sock->sk;
2548 if (mode < SHUT_RD || mode > SHUT_RDWR)
2551 * SHUT_RD (0) -> RCV_SHUTDOWN (1)
2552 * SHUT_WR (1) -> SEND_SHUTDOWN (2)
2553 * SHUT_RDWR (2) -> SHUTDOWN_MASK (3)
2557 unix_state_lock(sk);
2558 sk->sk_shutdown |= mode;
2559 other = unix_peer(sk);
2562 unix_state_unlock(sk);
2563 sk->sk_state_change(sk);
2566 (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET)) {
2570 if (mode&RCV_SHUTDOWN)
2571 peer_mode |= SEND_SHUTDOWN;
2572 if (mode&SEND_SHUTDOWN)
2573 peer_mode |= RCV_SHUTDOWN;
2574 unix_state_lock(other);
2575 other->sk_shutdown |= peer_mode;
2576 unix_state_unlock(other);
2577 other->sk_state_change(other);
2578 if (peer_mode == SHUTDOWN_MASK)
2579 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_HUP);
2580 else if (peer_mode & RCV_SHUTDOWN)
2581 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_IN);
2589 long unix_inq_len(struct sock *sk)
2591 struct sk_buff *skb;
2594 if (sk->sk_state == TCP_LISTEN)
2597 spin_lock(&sk->sk_receive_queue.lock);
2598 if (sk->sk_type == SOCK_STREAM ||
2599 sk->sk_type == SOCK_SEQPACKET) {
2600 skb_queue_walk(&sk->sk_receive_queue, skb)
2601 amount += unix_skb_len(skb);
2603 skb = skb_peek(&sk->sk_receive_queue);
2607 spin_unlock(&sk->sk_receive_queue.lock);
2611 EXPORT_SYMBOL_GPL(unix_inq_len);
2613 long unix_outq_len(struct sock *sk)
2615 return sk_wmem_alloc_get(sk);
2617 EXPORT_SYMBOL_GPL(unix_outq_len);
2619 static int unix_open_file(struct sock *sk)
2625 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
2628 if (!smp_load_acquire(&unix_sk(sk)->addr))
2631 path = unix_sk(sk)->path;
2637 fd = get_unused_fd_flags(O_CLOEXEC);
2641 f = dentry_open(&path, O_PATH, current_cred());
2655 static int unix_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2657 struct sock *sk = sock->sk;
2663 amount = unix_outq_len(sk);
2664 err = put_user(amount, (int __user *)arg);
2667 amount = unix_inq_len(sk);
2671 err = put_user(amount, (int __user *)arg);
2674 err = unix_open_file(sk);
2683 #ifdef CONFIG_COMPAT
2684 static int unix_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2686 return unix_ioctl(sock, cmd, (unsigned long)compat_ptr(arg));
2690 static unsigned int unix_poll(struct file *file, struct socket *sock, poll_table *wait)
2692 struct sock *sk = sock->sk;
2695 sock_poll_wait(file, sk_sleep(sk), wait);
2698 /* exceptional events? */
2701 if (sk->sk_shutdown == SHUTDOWN_MASK)
2703 if (sk->sk_shutdown & RCV_SHUTDOWN)
2704 mask |= POLLRDHUP | POLLIN | POLLRDNORM;
2707 if (!skb_queue_empty_lockless(&sk->sk_receive_queue))
2708 mask |= POLLIN | POLLRDNORM;
2710 /* Connection-based need to check for termination and startup */
2711 if ((sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) &&
2712 sk->sk_state == TCP_CLOSE)
2716 * we set writable also when the other side has shut down the
2717 * connection. This prevents stuck sockets.
2719 if (unix_writable(sk))
2720 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2725 static unsigned int unix_dgram_poll(struct file *file, struct socket *sock,
2728 struct sock *sk = sock->sk, *other;
2729 unsigned int mask, writable;
2731 sock_poll_wait(file, sk_sleep(sk), wait);
2734 /* exceptional events? */
2735 if (sk->sk_err || !skb_queue_empty_lockless(&sk->sk_error_queue))
2737 (sock_flag(sk, SOCK_SELECT_ERR_QUEUE) ? POLLPRI : 0);
2739 if (sk->sk_shutdown & RCV_SHUTDOWN)
2740 mask |= POLLRDHUP | POLLIN | POLLRDNORM;
2741 if (sk->sk_shutdown == SHUTDOWN_MASK)
2745 if (!skb_queue_empty_lockless(&sk->sk_receive_queue))
2746 mask |= POLLIN | POLLRDNORM;
2748 /* Connection-based need to check for termination and startup */
2749 if (sk->sk_type == SOCK_SEQPACKET) {
2750 if (sk->sk_state == TCP_CLOSE)
2752 /* connection hasn't started yet? */
2753 if (sk->sk_state == TCP_SYN_SENT)
2757 /* No write status requested, avoid expensive OUT tests. */
2758 if (!(poll_requested_events(wait) & (POLLWRBAND|POLLWRNORM|POLLOUT)))
2761 writable = unix_writable(sk);
2763 unix_state_lock(sk);
2765 other = unix_peer(sk);
2766 if (other && unix_peer(other) != sk &&
2767 unix_recvq_full_lockless(other) &&
2768 unix_dgram_peer_wake_me(sk, other))
2771 unix_state_unlock(sk);
2775 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2777 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2782 #ifdef CONFIG_PROC_FS
2784 #define BUCKET_SPACE (BITS_PER_LONG - (UNIX_HASH_BITS + 1) - 1)
2786 #define get_bucket(x) ((x) >> BUCKET_SPACE)
2787 #define get_offset(x) ((x) & ((1L << BUCKET_SPACE) - 1))
2788 #define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
2790 static struct sock *unix_from_bucket(struct seq_file *seq, loff_t *pos)
2792 unsigned long offset = get_offset(*pos);
2793 unsigned long bucket = get_bucket(*pos);
2795 unsigned long count = 0;
2797 for (sk = sk_head(&unix_socket_table[bucket]); sk; sk = sk_next(sk)) {
2798 if (sock_net(sk) != seq_file_net(seq))
2800 if (++count == offset)
2807 static struct sock *unix_next_socket(struct seq_file *seq,
2811 unsigned long bucket;
2813 while (sk > (struct sock *)SEQ_START_TOKEN) {
2817 if (sock_net(sk) == seq_file_net(seq))
2822 sk = unix_from_bucket(seq, pos);
2827 bucket = get_bucket(*pos) + 1;
2828 *pos = set_bucket_offset(bucket, 1);
2829 } while (bucket < ARRAY_SIZE(unix_socket_table));
2834 static void *unix_seq_start(struct seq_file *seq, loff_t *pos)
2835 __acquires(unix_table_lock)
2837 spin_lock(&unix_table_lock);
2840 return SEQ_START_TOKEN;
2842 if (get_bucket(*pos) >= ARRAY_SIZE(unix_socket_table))
2845 return unix_next_socket(seq, NULL, pos);
2848 static void *unix_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2851 return unix_next_socket(seq, v, pos);
2854 static void unix_seq_stop(struct seq_file *seq, void *v)
2855 __releases(unix_table_lock)
2857 spin_unlock(&unix_table_lock);
2860 static int unix_seq_show(struct seq_file *seq, void *v)
2863 if (v == SEQ_START_TOKEN)
2864 seq_puts(seq, "Num RefCount Protocol Flags Type St "
2868 struct unix_sock *u = unix_sk(s);
2871 seq_printf(seq, "%pK: %08X %08X %08X %04X %02X %5lu",
2873 refcount_read(&s->sk_refcnt),
2875 s->sk_state == TCP_LISTEN ? __SO_ACCEPTCON : 0,
2878 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTED : SS_UNCONNECTED) :
2879 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTING : SS_DISCONNECTING),
2882 if (u->addr) { // under unix_table_lock here
2887 len = u->addr->len - sizeof(short);
2888 if (!UNIX_ABSTRACT(s))
2894 for ( ; i < len; i++)
2895 seq_putc(seq, u->addr->name->sun_path[i] ?:
2898 unix_state_unlock(s);
2899 seq_putc(seq, '\n');
2905 static const struct seq_operations unix_seq_ops = {
2906 .start = unix_seq_start,
2907 .next = unix_seq_next,
2908 .stop = unix_seq_stop,
2909 .show = unix_seq_show,
2912 static int unix_seq_open(struct inode *inode, struct file *file)
2914 return seq_open_net(inode, file, &unix_seq_ops,
2915 sizeof(struct seq_net_private));
2918 static const struct file_operations unix_seq_fops = {
2919 .owner = THIS_MODULE,
2920 .open = unix_seq_open,
2922 .llseek = seq_lseek,
2923 .release = seq_release_net,
2928 static const struct net_proto_family unix_family_ops = {
2930 .create = unix_create,
2931 .owner = THIS_MODULE,
2935 static int __net_init unix_net_init(struct net *net)
2937 int error = -ENOMEM;
2939 net->unx.sysctl_max_dgram_qlen = 10;
2940 if (unix_sysctl_register(net))
2943 #ifdef CONFIG_PROC_FS
2944 if (!proc_create("unix", 0, net->proc_net, &unix_seq_fops)) {
2945 unix_sysctl_unregister(net);
2954 static void __net_exit unix_net_exit(struct net *net)
2956 unix_sysctl_unregister(net);
2957 remove_proc_entry("unix", net->proc_net);
2960 static struct pernet_operations unix_net_ops = {
2961 .init = unix_net_init,
2962 .exit = unix_net_exit,
2965 static int __init af_unix_init(void)
2969 BUILD_BUG_ON(sizeof(struct unix_skb_parms) > FIELD_SIZEOF(struct sk_buff, cb));
2971 rc = proto_register(&unix_proto, 1);
2973 pr_crit("%s: Cannot create unix_sock SLAB cache!\n", __func__);
2977 sock_register(&unix_family_ops);
2978 register_pernet_subsys(&unix_net_ops);
2983 static void __exit af_unix_exit(void)
2985 sock_unregister(PF_UNIX);
2986 proto_unregister(&unix_proto);
2987 unregister_pernet_subsys(&unix_net_ops);
2990 /* Earlier than device_initcall() so that other drivers invoking
2991 request_module() don't end up in a loop when modprobe tries
2992 to use a UNIX socket. But later than subsys_initcall() because
2993 we depend on stuff initialised there */
2994 fs_initcall(af_unix_init);
2995 module_exit(af_unix_exit);
2997 MODULE_LICENSE("GPL");
2998 MODULE_ALIAS_NETPROTO(PF_UNIX);