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>
122 struct hlist_head unix_socket_table
[2 * UNIX_HASH_SIZE
];
123 EXPORT_SYMBOL_GPL(unix_socket_table
);
124 DEFINE_SPINLOCK(unix_table_lock
);
125 EXPORT_SYMBOL_GPL(unix_table_lock
);
126 static atomic_long_t unix_nr_socks
;
129 static struct hlist_head
*unix_sockets_unbound(void *addr
)
131 unsigned long hash
= (unsigned long)addr
;
135 hash
%= UNIX_HASH_SIZE
;
136 return &unix_socket_table
[UNIX_HASH_SIZE
+ hash
];
139 #define UNIX_ABSTRACT(sk) (unix_sk(sk)->addr->hash < UNIX_HASH_SIZE)
141 #ifdef CONFIG_SECURITY_NETWORK
142 static void unix_get_secdata(struct scm_cookie
*scm
, struct sk_buff
*skb
)
144 UNIXCB(skb
).secid
= scm
->secid
;
147 static inline void unix_set_secdata(struct scm_cookie
*scm
, struct sk_buff
*skb
)
149 scm
->secid
= UNIXCB(skb
).secid
;
152 static inline bool unix_secdata_eq(struct scm_cookie
*scm
, struct sk_buff
*skb
)
154 return (scm
->secid
== UNIXCB(skb
).secid
);
157 static inline void unix_get_secdata(struct scm_cookie
*scm
, struct sk_buff
*skb
)
160 static inline void unix_set_secdata(struct scm_cookie
*scm
, struct sk_buff
*skb
)
163 static inline bool unix_secdata_eq(struct scm_cookie
*scm
, struct sk_buff
*skb
)
167 #endif /* CONFIG_SECURITY_NETWORK */
170 * SMP locking strategy:
171 * hash table is protected with spinlock unix_table_lock
172 * each socket state is protected by separate spin lock.
175 static inline unsigned int unix_hash_fold(__wsum n
)
177 unsigned int hash
= (__force
unsigned int)csum_fold(n
);
180 return hash
&(UNIX_HASH_SIZE
-1);
183 #define unix_peer(sk) (unix_sk(sk)->peer)
185 static inline int unix_our_peer(struct sock
*sk
, struct sock
*osk
)
187 return unix_peer(osk
) == sk
;
190 static inline int unix_may_send(struct sock
*sk
, struct sock
*osk
)
192 return unix_peer(osk
) == NULL
|| unix_our_peer(sk
, osk
);
195 static inline int unix_recvq_full(struct sock
const *sk
)
197 return skb_queue_len(&sk
->sk_receive_queue
) > sk
->sk_max_ack_backlog
;
200 struct sock
*unix_peer_get(struct sock
*s
)
208 unix_state_unlock(s
);
211 EXPORT_SYMBOL_GPL(unix_peer_get
);
213 static inline void unix_release_addr(struct unix_address
*addr
)
215 if (refcount_dec_and_test(&addr
->refcnt
))
220 * Check unix socket name:
221 * - should be not zero length.
222 * - if started by not zero, should be NULL terminated (FS object)
223 * - if started by zero, it is abstract name.
226 static int unix_mkname(struct sockaddr_un
*sunaddr
, int len
, unsigned int *hashp
)
228 if (len
<= sizeof(short) || len
> sizeof(*sunaddr
))
230 if (!sunaddr
|| sunaddr
->sun_family
!= AF_UNIX
)
232 if (sunaddr
->sun_path
[0]) {
234 * This may look like an off by one error but it is a bit more
235 * subtle. 108 is the longest valid AF_UNIX path for a binding.
236 * sun_path[108] doesn't as such exist. However in kernel space
237 * we are guaranteed that it is a valid memory location in our
238 * kernel address buffer.
240 ((char *)sunaddr
)[len
] = 0;
241 len
= strlen(sunaddr
->sun_path
)+1+sizeof(short);
245 *hashp
= unix_hash_fold(csum_partial(sunaddr
, len
, 0));
249 static void __unix_remove_socket(struct sock
*sk
)
251 sk_del_node_init(sk
);
254 static void __unix_insert_socket(struct hlist_head
*list
, struct sock
*sk
)
256 WARN_ON(!sk_unhashed(sk
));
257 sk_add_node(sk
, list
);
260 static inline void unix_remove_socket(struct sock
*sk
)
262 spin_lock(&unix_table_lock
);
263 __unix_remove_socket(sk
);
264 spin_unlock(&unix_table_lock
);
267 static inline void unix_insert_socket(struct hlist_head
*list
, struct sock
*sk
)
269 spin_lock(&unix_table_lock
);
270 __unix_insert_socket(list
, sk
);
271 spin_unlock(&unix_table_lock
);
274 static struct sock
*__unix_find_socket_byname(struct net
*net
,
275 struct sockaddr_un
*sunname
,
276 int len
, int type
, unsigned int hash
)
280 sk_for_each(s
, &unix_socket_table
[hash
^ type
]) {
281 struct unix_sock
*u
= unix_sk(s
);
283 if (!net_eq(sock_net(s
), net
))
286 if (u
->addr
->len
== len
&&
287 !memcmp(u
->addr
->name
, sunname
, len
))
295 static inline struct sock
*unix_find_socket_byname(struct net
*net
,
296 struct sockaddr_un
*sunname
,
302 spin_lock(&unix_table_lock
);
303 s
= __unix_find_socket_byname(net
, sunname
, len
, type
, hash
);
306 spin_unlock(&unix_table_lock
);
310 static struct sock
*unix_find_socket_byinode(struct inode
*i
)
314 spin_lock(&unix_table_lock
);
316 &unix_socket_table
[i
->i_ino
& (UNIX_HASH_SIZE
- 1)]) {
317 struct dentry
*dentry
= unix_sk(s
)->path
.dentry
;
319 if (dentry
&& d_backing_inode(dentry
) == i
) {
326 spin_unlock(&unix_table_lock
);
330 /* Support code for asymmetrically connected dgram sockets
332 * If a datagram socket is connected to a socket not itself connected
333 * to the first socket (eg, /dev/log), clients may only enqueue more
334 * messages if the present receive queue of the server socket is not
335 * "too large". This means there's a second writeability condition
336 * poll and sendmsg need to test. The dgram recv code will do a wake
337 * up on the peer_wait wait queue of a socket upon reception of a
338 * datagram which needs to be propagated to sleeping would-be writers
339 * since these might not have sent anything so far. This can't be
340 * accomplished via poll_wait because the lifetime of the server
341 * socket might be less than that of its clients if these break their
342 * association with it or if the server socket is closed while clients
343 * are still connected to it and there's no way to inform "a polling
344 * implementation" that it should let go of a certain wait queue
346 * In order to propagate a wake up, a wait_queue_entry_t of the client
347 * socket is enqueued on the peer_wait queue of the server socket
348 * whose wake function does a wake_up on the ordinary client socket
349 * wait queue. This connection is established whenever a write (or
350 * poll for write) hit the flow control condition and broken when the
351 * association to the server socket is dissolved or after a wake up
355 static int unix_dgram_peer_wake_relay(wait_queue_entry_t
*q
, unsigned mode
, int flags
,
359 wait_queue_head_t
*u_sleep
;
361 u
= container_of(q
, struct unix_sock
, peer_wake
);
363 __remove_wait_queue(&unix_sk(u
->peer_wake
.private)->peer_wait
,
365 u
->peer_wake
.private = NULL
;
367 /* relaying can only happen while the wq still exists */
368 u_sleep
= sk_sleep(&u
->sk
);
370 wake_up_interruptible_poll(u_sleep
, key
);
375 static int unix_dgram_peer_wake_connect(struct sock
*sk
, struct sock
*other
)
377 struct unix_sock
*u
, *u_other
;
381 u_other
= unix_sk(other
);
383 spin_lock(&u_other
->peer_wait
.lock
);
385 if (!u
->peer_wake
.private) {
386 u
->peer_wake
.private = other
;
387 __add_wait_queue(&u_other
->peer_wait
, &u
->peer_wake
);
392 spin_unlock(&u_other
->peer_wait
.lock
);
396 static void unix_dgram_peer_wake_disconnect(struct sock
*sk
,
399 struct unix_sock
*u
, *u_other
;
402 u_other
= unix_sk(other
);
403 spin_lock(&u_other
->peer_wait
.lock
);
405 if (u
->peer_wake
.private == other
) {
406 __remove_wait_queue(&u_other
->peer_wait
, &u
->peer_wake
);
407 u
->peer_wake
.private = NULL
;
410 spin_unlock(&u_other
->peer_wait
.lock
);
413 static void unix_dgram_peer_wake_disconnect_wakeup(struct sock
*sk
,
416 unix_dgram_peer_wake_disconnect(sk
, other
);
417 wake_up_interruptible_poll(sk_sleep(sk
),
424 * - unix_peer(sk) == other
425 * - association is stable
427 static int unix_dgram_peer_wake_me(struct sock
*sk
, struct sock
*other
)
431 connected
= unix_dgram_peer_wake_connect(sk
, other
);
433 if (unix_recvq_full(other
))
437 unix_dgram_peer_wake_disconnect(sk
, other
);
442 static int unix_writable(const struct sock
*sk
)
444 return sk
->sk_state
!= TCP_LISTEN
&&
445 (refcount_read(&sk
->sk_wmem_alloc
) << 2) <= sk
->sk_sndbuf
;
448 static void unix_write_space(struct sock
*sk
)
450 struct socket_wq
*wq
;
453 if (unix_writable(sk
)) {
454 wq
= rcu_dereference(sk
->sk_wq
);
455 if (skwq_has_sleeper(wq
))
456 wake_up_interruptible_sync_poll(&wq
->wait
,
457 POLLOUT
| POLLWRNORM
| POLLWRBAND
);
458 sk_wake_async(sk
, SOCK_WAKE_SPACE
, POLL_OUT
);
463 /* When dgram socket disconnects (or changes its peer), we clear its receive
464 * queue of packets arrived from previous peer. First, it allows to do
465 * flow control based only on wmem_alloc; second, sk connected to peer
466 * may receive messages only from that peer. */
467 static void unix_dgram_disconnected(struct sock
*sk
, struct sock
*other
)
469 if (!skb_queue_empty(&sk
->sk_receive_queue
)) {
470 skb_queue_purge(&sk
->sk_receive_queue
);
471 wake_up_interruptible_all(&unix_sk(sk
)->peer_wait
);
473 /* If one link of bidirectional dgram pipe is disconnected,
474 * we signal error. Messages are lost. Do not make this,
475 * when peer was not connected to us.
477 if (!sock_flag(other
, SOCK_DEAD
) && unix_peer(other
) == sk
) {
478 other
->sk_err
= ECONNRESET
;
479 other
->sk_error_report(other
);
484 static void unix_sock_destructor(struct sock
*sk
)
486 struct unix_sock
*u
= unix_sk(sk
);
488 skb_queue_purge(&sk
->sk_receive_queue
);
490 WARN_ON(refcount_read(&sk
->sk_wmem_alloc
));
491 WARN_ON(!sk_unhashed(sk
));
492 WARN_ON(sk
->sk_socket
);
493 if (!sock_flag(sk
, SOCK_DEAD
)) {
494 pr_info("Attempt to release alive unix socket: %p\n", sk
);
499 unix_release_addr(u
->addr
);
501 atomic_long_dec(&unix_nr_socks
);
503 sock_prot_inuse_add(sock_net(sk
), sk
->sk_prot
, -1);
505 #ifdef UNIX_REFCNT_DEBUG
506 pr_debug("UNIX %p is destroyed, %ld are still alive.\n", sk
,
507 atomic_long_read(&unix_nr_socks
));
511 static void unix_release_sock(struct sock
*sk
, int embrion
)
513 struct unix_sock
*u
= unix_sk(sk
);
519 unix_remove_socket(sk
);
524 sk
->sk_shutdown
= SHUTDOWN_MASK
;
526 u
->path
.dentry
= NULL
;
528 state
= sk
->sk_state
;
529 sk
->sk_state
= TCP_CLOSE
;
530 unix_state_unlock(sk
);
532 wake_up_interruptible_all(&u
->peer_wait
);
534 skpair
= unix_peer(sk
);
536 if (skpair
!= NULL
) {
537 if (sk
->sk_type
== SOCK_STREAM
|| sk
->sk_type
== SOCK_SEQPACKET
) {
538 unix_state_lock(skpair
);
540 skpair
->sk_shutdown
= SHUTDOWN_MASK
;
541 if (!skb_queue_empty(&sk
->sk_receive_queue
) || embrion
)
542 skpair
->sk_err
= ECONNRESET
;
543 unix_state_unlock(skpair
);
544 skpair
->sk_state_change(skpair
);
545 sk_wake_async(skpair
, SOCK_WAKE_WAITD
, POLL_HUP
);
548 unix_dgram_peer_wake_disconnect(sk
, skpair
);
549 sock_put(skpair
); /* It may now die */
550 unix_peer(sk
) = NULL
;
553 /* Try to flush out this socket. Throw out buffers at least */
555 while ((skb
= skb_dequeue(&sk
->sk_receive_queue
)) != NULL
) {
556 if (state
== TCP_LISTEN
)
557 unix_release_sock(skb
->sk
, 1);
558 /* passed fds are erased in the kfree_skb hook */
559 UNIXCB(skb
).consumed
= skb
->len
;
568 /* ---- Socket is dead now and most probably destroyed ---- */
571 * Fixme: BSD difference: In BSD all sockets connected to us get
572 * ECONNRESET and we die on the spot. In Linux we behave
573 * like files and pipes do and wait for the last
576 * Can't we simply set sock->err?
578 * What the above comment does talk about? --ANK(980817)
581 if (unix_tot_inflight
)
582 unix_gc(); /* Garbage collect fds */
585 static void init_peercred(struct sock
*sk
)
587 put_pid(sk
->sk_peer_pid
);
588 if (sk
->sk_peer_cred
)
589 put_cred(sk
->sk_peer_cred
);
590 sk
->sk_peer_pid
= get_pid(task_tgid(current
));
591 sk
->sk_peer_cred
= get_current_cred();
594 static void copy_peercred(struct sock
*sk
, struct sock
*peersk
)
596 put_pid(sk
->sk_peer_pid
);
597 if (sk
->sk_peer_cred
)
598 put_cred(sk
->sk_peer_cred
);
599 sk
->sk_peer_pid
= get_pid(peersk
->sk_peer_pid
);
600 sk
->sk_peer_cred
= get_cred(peersk
->sk_peer_cred
);
603 static int unix_listen(struct socket
*sock
, int backlog
)
606 struct sock
*sk
= sock
->sk
;
607 struct unix_sock
*u
= unix_sk(sk
);
608 struct pid
*old_pid
= NULL
;
611 if (sock
->type
!= SOCK_STREAM
&& sock
->type
!= SOCK_SEQPACKET
)
612 goto out
; /* Only stream/seqpacket sockets accept */
615 goto out
; /* No listens on an unbound socket */
617 if (sk
->sk_state
!= TCP_CLOSE
&& sk
->sk_state
!= TCP_LISTEN
)
619 if (backlog
> sk
->sk_max_ack_backlog
)
620 wake_up_interruptible_all(&u
->peer_wait
);
621 sk
->sk_max_ack_backlog
= backlog
;
622 sk
->sk_state
= TCP_LISTEN
;
623 /* set credentials so connect can copy them */
628 unix_state_unlock(sk
);
634 static int unix_release(struct socket
*);
635 static int unix_bind(struct socket
*, struct sockaddr
*, int);
636 static int unix_stream_connect(struct socket
*, struct sockaddr
*,
637 int addr_len
, int flags
);
638 static int unix_socketpair(struct socket
*, struct socket
*);
639 static int unix_accept(struct socket
*, struct socket
*, int, bool);
640 static int unix_getname(struct socket
*, struct sockaddr
*, int *, int);
641 static unsigned int unix_poll(struct file
*, struct socket
*, poll_table
*);
642 static unsigned int unix_dgram_poll(struct file
*, struct socket
*,
644 static int unix_ioctl(struct socket
*, unsigned int, unsigned long);
645 static int unix_shutdown(struct socket
*, int);
646 static int unix_stream_sendmsg(struct socket
*, struct msghdr
*, size_t);
647 static int unix_stream_recvmsg(struct socket
*, struct msghdr
*, size_t, int);
648 static ssize_t
unix_stream_sendpage(struct socket
*, struct page
*, int offset
,
649 size_t size
, int flags
);
650 static ssize_t
unix_stream_splice_read(struct socket
*, loff_t
*ppos
,
651 struct pipe_inode_info
*, size_t size
,
653 static int unix_dgram_sendmsg(struct socket
*, struct msghdr
*, size_t);
654 static int unix_dgram_recvmsg(struct socket
*, struct msghdr
*, size_t, int);
655 static int unix_dgram_connect(struct socket
*, struct sockaddr
*,
657 static int unix_seqpacket_sendmsg(struct socket
*, struct msghdr
*, size_t);
658 static int unix_seqpacket_recvmsg(struct socket
*, struct msghdr
*, size_t,
661 static int unix_set_peek_off(struct sock
*sk
, int val
)
663 struct unix_sock
*u
= unix_sk(sk
);
665 if (mutex_lock_interruptible(&u
->iolock
))
668 sk
->sk_peek_off
= val
;
669 mutex_unlock(&u
->iolock
);
675 static const struct proto_ops unix_stream_ops
= {
677 .owner
= THIS_MODULE
,
678 .release
= unix_release
,
680 .connect
= unix_stream_connect
,
681 .socketpair
= unix_socketpair
,
682 .accept
= unix_accept
,
683 .getname
= unix_getname
,
686 .listen
= unix_listen
,
687 .shutdown
= unix_shutdown
,
688 .setsockopt
= sock_no_setsockopt
,
689 .getsockopt
= sock_no_getsockopt
,
690 .sendmsg
= unix_stream_sendmsg
,
691 .recvmsg
= unix_stream_recvmsg
,
692 .mmap
= sock_no_mmap
,
693 .sendpage
= unix_stream_sendpage
,
694 .splice_read
= unix_stream_splice_read
,
695 .set_peek_off
= unix_set_peek_off
,
698 static const struct proto_ops unix_dgram_ops
= {
700 .owner
= THIS_MODULE
,
701 .release
= unix_release
,
703 .connect
= unix_dgram_connect
,
704 .socketpair
= unix_socketpair
,
705 .accept
= sock_no_accept
,
706 .getname
= unix_getname
,
707 .poll
= unix_dgram_poll
,
709 .listen
= sock_no_listen
,
710 .shutdown
= unix_shutdown
,
711 .setsockopt
= sock_no_setsockopt
,
712 .getsockopt
= sock_no_getsockopt
,
713 .sendmsg
= unix_dgram_sendmsg
,
714 .recvmsg
= unix_dgram_recvmsg
,
715 .mmap
= sock_no_mmap
,
716 .sendpage
= sock_no_sendpage
,
717 .set_peek_off
= unix_set_peek_off
,
720 static const struct proto_ops unix_seqpacket_ops
= {
722 .owner
= THIS_MODULE
,
723 .release
= unix_release
,
725 .connect
= unix_stream_connect
,
726 .socketpair
= unix_socketpair
,
727 .accept
= unix_accept
,
728 .getname
= unix_getname
,
729 .poll
= unix_dgram_poll
,
731 .listen
= unix_listen
,
732 .shutdown
= unix_shutdown
,
733 .setsockopt
= sock_no_setsockopt
,
734 .getsockopt
= sock_no_getsockopt
,
735 .sendmsg
= unix_seqpacket_sendmsg
,
736 .recvmsg
= unix_seqpacket_recvmsg
,
737 .mmap
= sock_no_mmap
,
738 .sendpage
= sock_no_sendpage
,
739 .set_peek_off
= unix_set_peek_off
,
742 static struct proto unix_proto
= {
744 .owner
= THIS_MODULE
,
745 .obj_size
= sizeof(struct unix_sock
),
749 * AF_UNIX sockets do not interact with hardware, hence they
750 * dont trigger interrupts - so it's safe for them to have
751 * bh-unsafe locking for their sk_receive_queue.lock. Split off
752 * this special lock-class by reinitializing the spinlock key:
754 static struct lock_class_key af_unix_sk_receive_queue_lock_key
;
756 static struct sock
*unix_create1(struct net
*net
, struct socket
*sock
, int kern
)
758 struct sock
*sk
= NULL
;
761 atomic_long_inc(&unix_nr_socks
);
762 if (atomic_long_read(&unix_nr_socks
) > 2 * get_max_files())
765 sk
= sk_alloc(net
, PF_UNIX
, GFP_KERNEL
, &unix_proto
, kern
);
769 sock_init_data(sock
, sk
);
770 lockdep_set_class(&sk
->sk_receive_queue
.lock
,
771 &af_unix_sk_receive_queue_lock_key
);
773 sk
->sk_allocation
= GFP_KERNEL_ACCOUNT
;
774 sk
->sk_write_space
= unix_write_space
;
775 sk
->sk_max_ack_backlog
= net
->unx
.sysctl_max_dgram_qlen
;
776 sk
->sk_destruct
= unix_sock_destructor
;
778 u
->path
.dentry
= NULL
;
780 spin_lock_init(&u
->lock
);
781 atomic_long_set(&u
->inflight
, 0);
782 INIT_LIST_HEAD(&u
->link
);
783 mutex_init(&u
->iolock
); /* single task reading lock */
784 mutex_init(&u
->bindlock
); /* single task binding lock */
785 init_waitqueue_head(&u
->peer_wait
);
786 init_waitqueue_func_entry(&u
->peer_wake
, unix_dgram_peer_wake_relay
);
787 unix_insert_socket(unix_sockets_unbound(sk
), sk
);
790 atomic_long_dec(&unix_nr_socks
);
793 sock_prot_inuse_add(sock_net(sk
), sk
->sk_prot
, 1);
799 static int unix_create(struct net
*net
, struct socket
*sock
, int protocol
,
802 if (protocol
&& protocol
!= PF_UNIX
)
803 return -EPROTONOSUPPORT
;
805 sock
->state
= SS_UNCONNECTED
;
807 switch (sock
->type
) {
809 sock
->ops
= &unix_stream_ops
;
812 * Believe it or not BSD has AF_UNIX, SOCK_RAW though
816 sock
->type
= SOCK_DGRAM
;
818 sock
->ops
= &unix_dgram_ops
;
821 sock
->ops
= &unix_seqpacket_ops
;
824 return -ESOCKTNOSUPPORT
;
827 return unix_create1(net
, sock
, kern
) ? 0 : -ENOMEM
;
830 static int unix_release(struct socket
*sock
)
832 struct sock
*sk
= sock
->sk
;
837 unix_release_sock(sk
, 0);
843 static int unix_autobind(struct socket
*sock
)
845 struct sock
*sk
= sock
->sk
;
846 struct net
*net
= sock_net(sk
);
847 struct unix_sock
*u
= unix_sk(sk
);
848 static u32 ordernum
= 1;
849 struct unix_address
*addr
;
851 unsigned int retries
= 0;
853 err
= mutex_lock_interruptible(&u
->bindlock
);
862 addr
= kzalloc(sizeof(*addr
) + sizeof(short) + 16, GFP_KERNEL
);
866 addr
->name
->sun_family
= AF_UNIX
;
867 refcount_set(&addr
->refcnt
, 1);
870 addr
->len
= sprintf(addr
->name
->sun_path
+1, "%05x", ordernum
) + 1 + sizeof(short);
871 addr
->hash
= unix_hash_fold(csum_partial(addr
->name
, addr
->len
, 0));
873 spin_lock(&unix_table_lock
);
874 ordernum
= (ordernum
+1)&0xFFFFF;
876 if (__unix_find_socket_byname(net
, addr
->name
, addr
->len
, sock
->type
,
878 spin_unlock(&unix_table_lock
);
880 * __unix_find_socket_byname() may take long time if many names
881 * are already in use.
884 /* Give up if all names seems to be in use. */
885 if (retries
++ == 0xFFFFF) {
892 addr
->hash
^= sk
->sk_type
;
894 __unix_remove_socket(sk
);
896 __unix_insert_socket(&unix_socket_table
[addr
->hash
], sk
);
897 spin_unlock(&unix_table_lock
);
900 out
: mutex_unlock(&u
->bindlock
);
904 static struct sock
*unix_find_other(struct net
*net
,
905 struct sockaddr_un
*sunname
, int len
,
906 int type
, unsigned int hash
, int *error
)
912 if (sunname
->sun_path
[0]) {
914 err
= kern_path(sunname
->sun_path
, LOOKUP_FOLLOW
, &path
);
917 inode
= d_backing_inode(path
.dentry
);
918 err
= inode_permission(inode
, MAY_WRITE
);
923 if (!S_ISSOCK(inode
->i_mode
))
925 u
= unix_find_socket_byinode(inode
);
929 if (u
->sk_type
== type
)
935 if (u
->sk_type
!= type
) {
941 u
= unix_find_socket_byname(net
, sunname
, len
, type
, hash
);
943 struct dentry
*dentry
;
944 dentry
= unix_sk(u
)->path
.dentry
;
946 touch_atime(&unix_sk(u
)->path
);
959 static int unix_mknod(const char *sun_path
, umode_t mode
, struct path
*res
)
961 struct dentry
*dentry
;
965 * Get the parent directory, calculate the hash for last
968 dentry
= kern_path_create(AT_FDCWD
, sun_path
, &path
, 0);
969 err
= PTR_ERR(dentry
);
974 * All right, let's create it.
976 err
= security_path_mknod(&path
, dentry
, mode
, 0);
978 err
= vfs_mknod(d_inode(path
.dentry
), dentry
, mode
, 0);
980 res
->mnt
= mntget(path
.mnt
);
981 res
->dentry
= dget(dentry
);
984 done_path_create(&path
, dentry
);
988 static int unix_bind(struct socket
*sock
, struct sockaddr
*uaddr
, int addr_len
)
990 struct sock
*sk
= sock
->sk
;
991 struct net
*net
= sock_net(sk
);
992 struct unix_sock
*u
= unix_sk(sk
);
993 struct sockaddr_un
*sunaddr
= (struct sockaddr_un
*)uaddr
;
994 char *sun_path
= sunaddr
->sun_path
;
997 struct unix_address
*addr
;
998 struct hlist_head
*list
;
999 struct path path
= { };
1002 if (addr_len
< offsetofend(struct sockaddr_un
, sun_family
) ||
1003 sunaddr
->sun_family
!= AF_UNIX
)
1006 if (addr_len
== sizeof(short)) {
1007 err
= unix_autobind(sock
);
1011 err
= unix_mkname(sunaddr
, addr_len
, &hash
);
1017 umode_t mode
= S_IFSOCK
|
1018 (SOCK_INODE(sock
)->i_mode
& ~current_umask());
1019 err
= unix_mknod(sun_path
, mode
, &path
);
1027 err
= mutex_lock_interruptible(&u
->bindlock
);
1036 addr
= kmalloc(sizeof(*addr
)+addr_len
, GFP_KERNEL
);
1040 memcpy(addr
->name
, sunaddr
, addr_len
);
1041 addr
->len
= addr_len
;
1042 addr
->hash
= hash
^ sk
->sk_type
;
1043 refcount_set(&addr
->refcnt
, 1);
1046 addr
->hash
= UNIX_HASH_SIZE
;
1047 hash
= d_backing_inode(path
.dentry
)->i_ino
& (UNIX_HASH_SIZE
- 1);
1048 spin_lock(&unix_table_lock
);
1050 list
= &unix_socket_table
[hash
];
1052 spin_lock(&unix_table_lock
);
1054 if (__unix_find_socket_byname(net
, sunaddr
, addr_len
,
1055 sk
->sk_type
, hash
)) {
1056 unix_release_addr(addr
);
1060 list
= &unix_socket_table
[addr
->hash
];
1064 __unix_remove_socket(sk
);
1066 __unix_insert_socket(list
, sk
);
1069 spin_unlock(&unix_table_lock
);
1071 mutex_unlock(&u
->bindlock
);
1079 static void unix_state_double_lock(struct sock
*sk1
, struct sock
*sk2
)
1081 if (unlikely(sk1
== sk2
) || !sk2
) {
1082 unix_state_lock(sk1
);
1086 unix_state_lock(sk1
);
1087 unix_state_lock_nested(sk2
);
1089 unix_state_lock(sk2
);
1090 unix_state_lock_nested(sk1
);
1094 static void unix_state_double_unlock(struct sock
*sk1
, struct sock
*sk2
)
1096 if (unlikely(sk1
== sk2
) || !sk2
) {
1097 unix_state_unlock(sk1
);
1100 unix_state_unlock(sk1
);
1101 unix_state_unlock(sk2
);
1104 static int unix_dgram_connect(struct socket
*sock
, struct sockaddr
*addr
,
1105 int alen
, int flags
)
1107 struct sock
*sk
= sock
->sk
;
1108 struct net
*net
= sock_net(sk
);
1109 struct sockaddr_un
*sunaddr
= (struct sockaddr_un
*)addr
;
1115 if (alen
< offsetofend(struct sockaddr
, sa_family
))
1118 if (addr
->sa_family
!= AF_UNSPEC
) {
1119 err
= unix_mkname(sunaddr
, alen
, &hash
);
1124 if (test_bit(SOCK_PASSCRED
, &sock
->flags
) &&
1125 !unix_sk(sk
)->addr
&& (err
= unix_autobind(sock
)) != 0)
1129 other
= unix_find_other(net
, sunaddr
, alen
, sock
->type
, hash
, &err
);
1133 unix_state_double_lock(sk
, other
);
1135 /* Apparently VFS overslept socket death. Retry. */
1136 if (sock_flag(other
, SOCK_DEAD
)) {
1137 unix_state_double_unlock(sk
, other
);
1143 if (!unix_may_send(sk
, other
))
1146 err
= security_unix_may_send(sk
->sk_socket
, other
->sk_socket
);
1152 * 1003.1g breaking connected state with AF_UNSPEC
1155 unix_state_double_lock(sk
, other
);
1159 * If it was connected, reconnect.
1161 if (unix_peer(sk
)) {
1162 struct sock
*old_peer
= unix_peer(sk
);
1163 unix_peer(sk
) = other
;
1164 unix_dgram_peer_wake_disconnect_wakeup(sk
, old_peer
);
1166 unix_state_double_unlock(sk
, other
);
1168 if (other
!= old_peer
)
1169 unix_dgram_disconnected(sk
, old_peer
);
1172 unix_peer(sk
) = other
;
1173 unix_state_double_unlock(sk
, other
);
1178 unix_state_double_unlock(sk
, other
);
1184 static long unix_wait_for_peer(struct sock
*other
, long timeo
)
1186 struct unix_sock
*u
= unix_sk(other
);
1190 prepare_to_wait_exclusive(&u
->peer_wait
, &wait
, TASK_INTERRUPTIBLE
);
1192 sched
= !sock_flag(other
, SOCK_DEAD
) &&
1193 !(other
->sk_shutdown
& RCV_SHUTDOWN
) &&
1194 unix_recvq_full(other
);
1196 unix_state_unlock(other
);
1199 timeo
= schedule_timeout(timeo
);
1201 finish_wait(&u
->peer_wait
, &wait
);
1205 static int unix_stream_connect(struct socket
*sock
, struct sockaddr
*uaddr
,
1206 int addr_len
, int flags
)
1208 struct sockaddr_un
*sunaddr
= (struct sockaddr_un
*)uaddr
;
1209 struct sock
*sk
= sock
->sk
;
1210 struct net
*net
= sock_net(sk
);
1211 struct unix_sock
*u
= unix_sk(sk
), *newu
, *otheru
;
1212 struct sock
*newsk
= NULL
;
1213 struct sock
*other
= NULL
;
1214 struct sk_buff
*skb
= NULL
;
1220 err
= unix_mkname(sunaddr
, addr_len
, &hash
);
1225 if (test_bit(SOCK_PASSCRED
, &sock
->flags
) && !u
->addr
&&
1226 (err
= unix_autobind(sock
)) != 0)
1229 timeo
= sock_sndtimeo(sk
, flags
& O_NONBLOCK
);
1231 /* First of all allocate resources.
1232 If we will make it after state is locked,
1233 we will have to recheck all again in any case.
1238 /* create new sock for complete connection */
1239 newsk
= unix_create1(sock_net(sk
), NULL
, 0);
1243 /* Allocate skb for sending to listening sock */
1244 skb
= sock_wmalloc(newsk
, 1, 0, GFP_KERNEL
);
1249 /* Find listening sock. */
1250 other
= unix_find_other(net
, sunaddr
, addr_len
, sk
->sk_type
, hash
, &err
);
1254 /* Latch state of peer */
1255 unix_state_lock(other
);
1257 /* Apparently VFS overslept socket death. Retry. */
1258 if (sock_flag(other
, SOCK_DEAD
)) {
1259 unix_state_unlock(other
);
1264 err
= -ECONNREFUSED
;
1265 if (other
->sk_state
!= TCP_LISTEN
)
1267 if (other
->sk_shutdown
& RCV_SHUTDOWN
)
1270 if (unix_recvq_full(other
)) {
1275 timeo
= unix_wait_for_peer(other
, timeo
);
1277 err
= sock_intr_errno(timeo
);
1278 if (signal_pending(current
))
1286 It is tricky place. We need to grab our state lock and cannot
1287 drop lock on peer. It is dangerous because deadlock is
1288 possible. Connect to self case and simultaneous
1289 attempt to connect are eliminated by checking socket
1290 state. other is TCP_LISTEN, if sk is TCP_LISTEN we
1291 check this before attempt to grab lock.
1293 Well, and we have to recheck the state after socket locked.
1299 /* This is ok... continue with connect */
1301 case TCP_ESTABLISHED
:
1302 /* Socket is already connected */
1310 unix_state_lock_nested(sk
);
1312 if (sk
->sk_state
!= st
) {
1313 unix_state_unlock(sk
);
1314 unix_state_unlock(other
);
1319 err
= security_unix_stream_connect(sk
, other
, newsk
);
1321 unix_state_unlock(sk
);
1325 /* The way is open! Fastly set all the necessary fields... */
1328 unix_peer(newsk
) = sk
;
1329 newsk
->sk_state
= TCP_ESTABLISHED
;
1330 newsk
->sk_type
= sk
->sk_type
;
1331 init_peercred(newsk
);
1332 newu
= unix_sk(newsk
);
1333 RCU_INIT_POINTER(newsk
->sk_wq
, &newu
->peer_wq
);
1334 otheru
= unix_sk(other
);
1336 /* copy address information from listening to new sock*/
1338 refcount_inc(&otheru
->addr
->refcnt
);
1339 newu
->addr
= otheru
->addr
;
1341 if (otheru
->path
.dentry
) {
1342 path_get(&otheru
->path
);
1343 newu
->path
= otheru
->path
;
1346 /* Set credentials */
1347 copy_peercred(sk
, other
);
1349 sock
->state
= SS_CONNECTED
;
1350 sk
->sk_state
= TCP_ESTABLISHED
;
1353 smp_mb__after_atomic(); /* sock_hold() does an atomic_inc() */
1354 unix_peer(sk
) = newsk
;
1356 unix_state_unlock(sk
);
1358 /* take ten and and send info to listening sock */
1359 spin_lock(&other
->sk_receive_queue
.lock
);
1360 __skb_queue_tail(&other
->sk_receive_queue
, skb
);
1361 spin_unlock(&other
->sk_receive_queue
.lock
);
1362 unix_state_unlock(other
);
1363 other
->sk_data_ready(other
);
1369 unix_state_unlock(other
);
1374 unix_release_sock(newsk
, 0);
1380 static int unix_socketpair(struct socket
*socka
, struct socket
*sockb
)
1382 struct sock
*ska
= socka
->sk
, *skb
= sockb
->sk
;
1384 /* Join our sockets back to back */
1387 unix_peer(ska
) = skb
;
1388 unix_peer(skb
) = ska
;
1392 if (ska
->sk_type
!= SOCK_DGRAM
) {
1393 ska
->sk_state
= TCP_ESTABLISHED
;
1394 skb
->sk_state
= TCP_ESTABLISHED
;
1395 socka
->state
= SS_CONNECTED
;
1396 sockb
->state
= SS_CONNECTED
;
1401 static void unix_sock_inherit_flags(const struct socket
*old
,
1404 if (test_bit(SOCK_PASSCRED
, &old
->flags
))
1405 set_bit(SOCK_PASSCRED
, &new->flags
);
1406 if (test_bit(SOCK_PASSSEC
, &old
->flags
))
1407 set_bit(SOCK_PASSSEC
, &new->flags
);
1410 static int unix_accept(struct socket
*sock
, struct socket
*newsock
, int flags
,
1413 struct sock
*sk
= sock
->sk
;
1415 struct sk_buff
*skb
;
1419 if (sock
->type
!= SOCK_STREAM
&& sock
->type
!= SOCK_SEQPACKET
)
1423 if (sk
->sk_state
!= TCP_LISTEN
)
1426 /* If socket state is TCP_LISTEN it cannot change (for now...),
1427 * so that no locks are necessary.
1430 skb
= skb_recv_datagram(sk
, 0, flags
&O_NONBLOCK
, &err
);
1432 /* This means receive shutdown. */
1439 skb_free_datagram(sk
, skb
);
1440 wake_up_interruptible(&unix_sk(sk
)->peer_wait
);
1442 /* attach accepted sock to socket */
1443 unix_state_lock(tsk
);
1444 newsock
->state
= SS_CONNECTED
;
1445 unix_sock_inherit_flags(sock
, newsock
);
1446 sock_graft(tsk
, newsock
);
1447 unix_state_unlock(tsk
);
1455 static int unix_getname(struct socket
*sock
, struct sockaddr
*uaddr
, int *uaddr_len
, int peer
)
1457 struct sock
*sk
= sock
->sk
;
1458 struct unix_sock
*u
;
1459 DECLARE_SOCKADDR(struct sockaddr_un
*, sunaddr
, uaddr
);
1463 sk
= unix_peer_get(sk
);
1474 unix_state_lock(sk
);
1476 sunaddr
->sun_family
= AF_UNIX
;
1477 sunaddr
->sun_path
[0] = 0;
1478 *uaddr_len
= sizeof(short);
1480 struct unix_address
*addr
= u
->addr
;
1482 *uaddr_len
= addr
->len
;
1483 memcpy(sunaddr
, addr
->name
, *uaddr_len
);
1485 unix_state_unlock(sk
);
1491 static void unix_detach_fds(struct scm_cookie
*scm
, struct sk_buff
*skb
)
1495 scm
->fp
= UNIXCB(skb
).fp
;
1496 UNIXCB(skb
).fp
= NULL
;
1498 for (i
= scm
->fp
->count
-1; i
>= 0; i
--)
1499 unix_notinflight(scm
->fp
->user
, scm
->fp
->fp
[i
]);
1502 static void unix_destruct_scm(struct sk_buff
*skb
)
1504 struct scm_cookie scm
;
1505 memset(&scm
, 0, sizeof(scm
));
1506 scm
.pid
= UNIXCB(skb
).pid
;
1508 unix_detach_fds(&scm
, skb
);
1510 /* Alas, it calls VFS */
1511 /* So fscking what? fput() had been SMP-safe since the last Summer */
1517 * The "user->unix_inflight" variable is protected by the garbage
1518 * collection lock, and we just read it locklessly here. If you go
1519 * over the limit, there might be a tiny race in actually noticing
1520 * it across threads. Tough.
1522 static inline bool too_many_unix_fds(struct task_struct
*p
)
1524 struct user_struct
*user
= current_user();
1526 if (unlikely(user
->unix_inflight
> task_rlimit(p
, RLIMIT_NOFILE
)))
1527 return !capable(CAP_SYS_RESOURCE
) && !capable(CAP_SYS_ADMIN
);
1531 #define MAX_RECURSION_LEVEL 4
1533 static int unix_attach_fds(struct scm_cookie
*scm
, struct sk_buff
*skb
)
1536 unsigned char max_level
= 0;
1538 if (too_many_unix_fds(current
))
1539 return -ETOOMANYREFS
;
1541 for (i
= scm
->fp
->count
- 1; i
>= 0; i
--) {
1542 struct sock
*sk
= unix_get_socket(scm
->fp
->fp
[i
]);
1545 max_level
= max(max_level
,
1546 unix_sk(sk
)->recursion_level
);
1548 if (unlikely(max_level
> MAX_RECURSION_LEVEL
))
1549 return -ETOOMANYREFS
;
1552 * Need to duplicate file references for the sake of garbage
1553 * collection. Otherwise a socket in the fps might become a
1554 * candidate for GC while the skb is not yet queued.
1556 UNIXCB(skb
).fp
= scm_fp_dup(scm
->fp
);
1557 if (!UNIXCB(skb
).fp
)
1560 for (i
= scm
->fp
->count
- 1; i
>= 0; i
--)
1561 unix_inflight(scm
->fp
->user
, scm
->fp
->fp
[i
]);
1565 static int unix_scm_to_skb(struct scm_cookie
*scm
, struct sk_buff
*skb
, bool send_fds
)
1569 UNIXCB(skb
).pid
= get_pid(scm
->pid
);
1570 UNIXCB(skb
).uid
= scm
->creds
.uid
;
1571 UNIXCB(skb
).gid
= scm
->creds
.gid
;
1572 UNIXCB(skb
).fp
= NULL
;
1573 unix_get_secdata(scm
, skb
);
1574 if (scm
->fp
&& send_fds
)
1575 err
= unix_attach_fds(scm
, skb
);
1577 skb
->destructor
= unix_destruct_scm
;
1581 static bool unix_passcred_enabled(const struct socket
*sock
,
1582 const struct sock
*other
)
1584 return test_bit(SOCK_PASSCRED
, &sock
->flags
) ||
1585 !other
->sk_socket
||
1586 test_bit(SOCK_PASSCRED
, &other
->sk_socket
->flags
);
1590 * Some apps rely on write() giving SCM_CREDENTIALS
1591 * We include credentials if source or destination socket
1592 * asserted SOCK_PASSCRED.
1594 static void maybe_add_creds(struct sk_buff
*skb
, const struct socket
*sock
,
1595 const struct sock
*other
)
1597 if (UNIXCB(skb
).pid
)
1599 if (unix_passcred_enabled(sock
, other
)) {
1600 UNIXCB(skb
).pid
= get_pid(task_tgid(current
));
1601 current_uid_gid(&UNIXCB(skb
).uid
, &UNIXCB(skb
).gid
);
1605 static int maybe_init_creds(struct scm_cookie
*scm
,
1606 struct socket
*socket
,
1607 const struct sock
*other
)
1610 struct msghdr msg
= { .msg_controllen
= 0 };
1612 err
= scm_send(socket
, &msg
, scm
, false);
1616 if (unix_passcred_enabled(socket
, other
)) {
1617 scm
->pid
= get_pid(task_tgid(current
));
1618 current_uid_gid(&scm
->creds
.uid
, &scm
->creds
.gid
);
1623 static bool unix_skb_scm_eq(struct sk_buff
*skb
,
1624 struct scm_cookie
*scm
)
1626 const struct unix_skb_parms
*u
= &UNIXCB(skb
);
1628 return u
->pid
== scm
->pid
&&
1629 uid_eq(u
->uid
, scm
->creds
.uid
) &&
1630 gid_eq(u
->gid
, scm
->creds
.gid
) &&
1631 unix_secdata_eq(scm
, skb
);
1635 * Send AF_UNIX data.
1638 static int unix_dgram_sendmsg(struct socket
*sock
, struct msghdr
*msg
,
1641 struct sock
*sk
= sock
->sk
;
1642 struct net
*net
= sock_net(sk
);
1643 struct unix_sock
*u
= unix_sk(sk
);
1644 DECLARE_SOCKADDR(struct sockaddr_un
*, sunaddr
, msg
->msg_name
);
1645 struct sock
*other
= NULL
;
1646 int namelen
= 0; /* fake GCC */
1649 struct sk_buff
*skb
;
1651 struct scm_cookie scm
;
1657 err
= scm_send(sock
, msg
, &scm
, false);
1662 if (msg
->msg_flags
&MSG_OOB
)
1665 if (msg
->msg_namelen
) {
1666 err
= unix_mkname(sunaddr
, msg
->msg_namelen
, &hash
);
1673 other
= unix_peer_get(sk
);
1678 if (test_bit(SOCK_PASSCRED
, &sock
->flags
) && !u
->addr
1679 && (err
= unix_autobind(sock
)) != 0)
1683 if (len
> sk
->sk_sndbuf
- 32)
1686 if (len
> SKB_MAX_ALLOC
) {
1687 data_len
= min_t(size_t,
1688 len
- SKB_MAX_ALLOC
,
1689 MAX_SKB_FRAGS
* PAGE_SIZE
);
1690 data_len
= PAGE_ALIGN(data_len
);
1692 BUILD_BUG_ON(SKB_MAX_ALLOC
< PAGE_SIZE
);
1695 skb
= sock_alloc_send_pskb(sk
, len
- data_len
, data_len
,
1696 msg
->msg_flags
& MSG_DONTWAIT
, &err
,
1697 PAGE_ALLOC_COSTLY_ORDER
);
1701 err
= unix_scm_to_skb(&scm
, skb
, true);
1704 max_level
= err
+ 1;
1706 skb_put(skb
, len
- data_len
);
1707 skb
->data_len
= data_len
;
1709 err
= skb_copy_datagram_from_iter(skb
, 0, &msg
->msg_iter
, len
);
1713 timeo
= sock_sndtimeo(sk
, msg
->msg_flags
& MSG_DONTWAIT
);
1718 if (sunaddr
== NULL
)
1721 other
= unix_find_other(net
, sunaddr
, namelen
, sk
->sk_type
,
1727 if (sk_filter(other
, skb
) < 0) {
1728 /* Toss the packet but do not return any error to the sender */
1734 unix_state_lock(other
);
1737 if (!unix_may_send(sk
, other
))
1740 if (unlikely(sock_flag(other
, SOCK_DEAD
))) {
1742 * Check with 1003.1g - what should
1745 unix_state_unlock(other
);
1749 unix_state_lock(sk
);
1752 if (unix_peer(sk
) == other
) {
1753 unix_peer(sk
) = NULL
;
1754 unix_dgram_peer_wake_disconnect_wakeup(sk
, other
);
1756 unix_state_unlock(sk
);
1758 unix_dgram_disconnected(sk
, other
);
1760 err
= -ECONNREFUSED
;
1762 unix_state_unlock(sk
);
1772 if (other
->sk_shutdown
& RCV_SHUTDOWN
)
1775 if (sk
->sk_type
!= SOCK_SEQPACKET
) {
1776 err
= security_unix_may_send(sk
->sk_socket
, other
->sk_socket
);
1781 /* other == sk && unix_peer(other) != sk if
1782 * - unix_peer(sk) == NULL, destination address bound to sk
1783 * - unix_peer(sk) == sk by time of get but disconnected before lock
1786 unlikely(unix_peer(other
) != sk
&& unix_recvq_full(other
))) {
1788 timeo
= unix_wait_for_peer(other
, timeo
);
1790 err
= sock_intr_errno(timeo
);
1791 if (signal_pending(current
))
1798 unix_state_unlock(other
);
1799 unix_state_double_lock(sk
, other
);
1802 if (unix_peer(sk
) != other
||
1803 unix_dgram_peer_wake_me(sk
, other
)) {
1811 goto restart_locked
;
1815 if (unlikely(sk_locked
))
1816 unix_state_unlock(sk
);
1818 if (sock_flag(other
, SOCK_RCVTSTAMP
))
1819 __net_timestamp(skb
);
1820 maybe_add_creds(skb
, sock
, other
);
1821 skb_queue_tail(&other
->sk_receive_queue
, skb
);
1822 if (max_level
> unix_sk(other
)->recursion_level
)
1823 unix_sk(other
)->recursion_level
= max_level
;
1824 unix_state_unlock(other
);
1825 other
->sk_data_ready(other
);
1832 unix_state_unlock(sk
);
1833 unix_state_unlock(other
);
1843 /* We use paged skbs for stream sockets, and limit occupancy to 32768
1844 * bytes, and a minimun of a full page.
1846 #define UNIX_SKB_FRAGS_SZ (PAGE_SIZE << get_order(32768))
1848 static int unix_stream_sendmsg(struct socket
*sock
, struct msghdr
*msg
,
1851 struct sock
*sk
= sock
->sk
;
1852 struct sock
*other
= NULL
;
1854 struct sk_buff
*skb
;
1856 struct scm_cookie scm
;
1857 bool fds_sent
= false;
1862 err
= scm_send(sock
, msg
, &scm
, false);
1867 if (msg
->msg_flags
&MSG_OOB
)
1870 if (msg
->msg_namelen
) {
1871 err
= sk
->sk_state
== TCP_ESTABLISHED
? -EISCONN
: -EOPNOTSUPP
;
1875 other
= unix_peer(sk
);
1880 if (sk
->sk_shutdown
& SEND_SHUTDOWN
)
1883 while (sent
< len
) {
1886 /* Keep two messages in the pipe so it schedules better */
1887 size
= min_t(int, size
, (sk
->sk_sndbuf
>> 1) - 64);
1889 /* allow fallback to order-0 allocations */
1890 size
= min_t(int, size
, SKB_MAX_HEAD(0) + UNIX_SKB_FRAGS_SZ
);
1892 data_len
= max_t(int, 0, size
- SKB_MAX_HEAD(0));
1894 data_len
= min_t(size_t, size
, PAGE_ALIGN(data_len
));
1896 skb
= sock_alloc_send_pskb(sk
, size
- data_len
, data_len
,
1897 msg
->msg_flags
& MSG_DONTWAIT
, &err
,
1898 get_order(UNIX_SKB_FRAGS_SZ
));
1902 /* Only send the fds in the first buffer */
1903 err
= unix_scm_to_skb(&scm
, skb
, !fds_sent
);
1908 max_level
= err
+ 1;
1911 skb_put(skb
, size
- data_len
);
1912 skb
->data_len
= data_len
;
1914 err
= skb_copy_datagram_from_iter(skb
, 0, &msg
->msg_iter
, size
);
1920 unix_state_lock(other
);
1922 if (sock_flag(other
, SOCK_DEAD
) ||
1923 (other
->sk_shutdown
& RCV_SHUTDOWN
))
1926 maybe_add_creds(skb
, sock
, other
);
1927 skb_queue_tail(&other
->sk_receive_queue
, skb
);
1928 if (max_level
> unix_sk(other
)->recursion_level
)
1929 unix_sk(other
)->recursion_level
= max_level
;
1930 unix_state_unlock(other
);
1931 other
->sk_data_ready(other
);
1940 unix_state_unlock(other
);
1943 if (sent
== 0 && !(msg
->msg_flags
&MSG_NOSIGNAL
))
1944 send_sig(SIGPIPE
, current
, 0);
1948 return sent
? : err
;
1951 static ssize_t
unix_stream_sendpage(struct socket
*socket
, struct page
*page
,
1952 int offset
, size_t size
, int flags
)
1955 bool send_sigpipe
= false;
1956 bool init_scm
= true;
1957 struct scm_cookie scm
;
1958 struct sock
*other
, *sk
= socket
->sk
;
1959 struct sk_buff
*skb
, *newskb
= NULL
, *tail
= NULL
;
1961 if (flags
& MSG_OOB
)
1964 other
= unix_peer(sk
);
1965 if (!other
|| sk
->sk_state
!= TCP_ESTABLISHED
)
1970 unix_state_unlock(other
);
1971 mutex_unlock(&unix_sk(other
)->iolock
);
1972 newskb
= sock_alloc_send_pskb(sk
, 0, 0, flags
& MSG_DONTWAIT
,
1978 /* we must acquire iolock as we modify already present
1979 * skbs in the sk_receive_queue and mess with skb->len
1981 err
= mutex_lock_interruptible(&unix_sk(other
)->iolock
);
1983 err
= flags
& MSG_DONTWAIT
? -EAGAIN
: -ERESTARTSYS
;
1987 if (sk
->sk_shutdown
& SEND_SHUTDOWN
) {
1989 send_sigpipe
= true;
1993 unix_state_lock(other
);
1995 if (sock_flag(other
, SOCK_DEAD
) ||
1996 other
->sk_shutdown
& RCV_SHUTDOWN
) {
1998 send_sigpipe
= true;
1999 goto err_state_unlock
;
2003 err
= maybe_init_creds(&scm
, socket
, other
);
2005 goto err_state_unlock
;
2009 skb
= skb_peek_tail(&other
->sk_receive_queue
);
2010 if (tail
&& tail
== skb
) {
2012 } else if (!skb
|| !unix_skb_scm_eq(skb
, &scm
)) {
2019 } else if (newskb
) {
2020 /* this is fast path, we don't necessarily need to
2021 * call to kfree_skb even though with newskb == NULL
2022 * this - does no harm
2024 consume_skb(newskb
);
2028 if (skb_append_pagefrags(skb
, page
, offset
, size
)) {
2034 skb
->data_len
+= size
;
2035 skb
->truesize
+= size
;
2036 refcount_add(size
, &sk
->sk_wmem_alloc
);
2039 err
= unix_scm_to_skb(&scm
, skb
, false);
2041 goto err_state_unlock
;
2042 spin_lock(&other
->sk_receive_queue
.lock
);
2043 __skb_queue_tail(&other
->sk_receive_queue
, newskb
);
2044 spin_unlock(&other
->sk_receive_queue
.lock
);
2047 unix_state_unlock(other
);
2048 mutex_unlock(&unix_sk(other
)->iolock
);
2050 other
->sk_data_ready(other
);
2055 unix_state_unlock(other
);
2057 mutex_unlock(&unix_sk(other
)->iolock
);
2060 if (send_sigpipe
&& !(flags
& MSG_NOSIGNAL
))
2061 send_sig(SIGPIPE
, current
, 0);
2067 static int unix_seqpacket_sendmsg(struct socket
*sock
, struct msghdr
*msg
,
2071 struct sock
*sk
= sock
->sk
;
2073 err
= sock_error(sk
);
2077 if (sk
->sk_state
!= TCP_ESTABLISHED
)
2080 if (msg
->msg_namelen
)
2081 msg
->msg_namelen
= 0;
2083 return unix_dgram_sendmsg(sock
, msg
, len
);
2086 static int unix_seqpacket_recvmsg(struct socket
*sock
, struct msghdr
*msg
,
2087 size_t size
, int flags
)
2089 struct sock
*sk
= sock
->sk
;
2091 if (sk
->sk_state
!= TCP_ESTABLISHED
)
2094 return unix_dgram_recvmsg(sock
, msg
, size
, flags
);
2097 static void unix_copy_addr(struct msghdr
*msg
, struct sock
*sk
)
2099 struct unix_sock
*u
= unix_sk(sk
);
2102 msg
->msg_namelen
= u
->addr
->len
;
2103 memcpy(msg
->msg_name
, u
->addr
->name
, u
->addr
->len
);
2107 static int unix_dgram_recvmsg(struct socket
*sock
, struct msghdr
*msg
,
2108 size_t size
, int flags
)
2110 struct scm_cookie scm
;
2111 struct sock
*sk
= sock
->sk
;
2112 struct unix_sock
*u
= unix_sk(sk
);
2113 struct sk_buff
*skb
, *last
;
2122 timeo
= sock_rcvtimeo(sk
, flags
& MSG_DONTWAIT
);
2125 mutex_lock(&u
->iolock
);
2127 skip
= sk_peek_offset(sk
, flags
);
2128 skb
= __skb_try_recv_datagram(sk
, flags
, NULL
, &peeked
, &skip
,
2133 mutex_unlock(&u
->iolock
);
2138 !__skb_wait_for_more_packets(sk
, &err
, &timeo
, last
));
2140 if (!skb
) { /* implies iolock unlocked */
2141 unix_state_lock(sk
);
2142 /* Signal EOF on disconnected non-blocking SEQPACKET socket. */
2143 if (sk
->sk_type
== SOCK_SEQPACKET
&& err
== -EAGAIN
&&
2144 (sk
->sk_shutdown
& RCV_SHUTDOWN
))
2146 unix_state_unlock(sk
);
2150 if (wq_has_sleeper(&u
->peer_wait
))
2151 wake_up_interruptible_sync_poll(&u
->peer_wait
,
2152 POLLOUT
| POLLWRNORM
|
2156 unix_copy_addr(msg
, skb
->sk
);
2158 if (size
> skb
->len
- skip
)
2159 size
= skb
->len
- skip
;
2160 else if (size
< skb
->len
- skip
)
2161 msg
->msg_flags
|= MSG_TRUNC
;
2163 err
= skb_copy_datagram_msg(skb
, skip
, msg
, size
);
2167 if (sock_flag(sk
, SOCK_RCVTSTAMP
))
2168 __sock_recv_timestamp(msg
, sk
, skb
);
2170 memset(&scm
, 0, sizeof(scm
));
2172 scm_set_cred(&scm
, UNIXCB(skb
).pid
, UNIXCB(skb
).uid
, UNIXCB(skb
).gid
);
2173 unix_set_secdata(&scm
, skb
);
2175 if (!(flags
& MSG_PEEK
)) {
2177 unix_detach_fds(&scm
, skb
);
2179 sk_peek_offset_bwd(sk
, skb
->len
);
2181 /* It is questionable: on PEEK we could:
2182 - do not return fds - good, but too simple 8)
2183 - return fds, and do not return them on read (old strategy,
2185 - clone fds (I chose it for now, it is the most universal
2188 POSIX 1003.1g does not actually define this clearly
2189 at all. POSIX 1003.1g doesn't define a lot of things
2194 sk_peek_offset_fwd(sk
, size
);
2197 scm
.fp
= scm_fp_dup(UNIXCB(skb
).fp
);
2199 err
= (flags
& MSG_TRUNC
) ? skb
->len
- skip
: size
;
2201 scm_recv(sock
, msg
, &scm
, flags
);
2204 skb_free_datagram(sk
, skb
);
2205 mutex_unlock(&u
->iolock
);
2211 * Sleep until more data has arrived. But check for races..
2213 static long unix_stream_data_wait(struct sock
*sk
, long timeo
,
2214 struct sk_buff
*last
, unsigned int last_len
,
2217 struct sk_buff
*tail
;
2220 unix_state_lock(sk
);
2223 prepare_to_wait(sk_sleep(sk
), &wait
, TASK_INTERRUPTIBLE
);
2225 tail
= skb_peek_tail(&sk
->sk_receive_queue
);
2227 (tail
&& tail
->len
!= last_len
) ||
2229 (sk
->sk_shutdown
& RCV_SHUTDOWN
) ||
2230 signal_pending(current
) ||
2234 sk_set_bit(SOCKWQ_ASYNC_WAITDATA
, sk
);
2235 unix_state_unlock(sk
);
2237 timeo
= freezable_schedule_timeout(timeo
);
2239 timeo
= schedule_timeout(timeo
);
2240 unix_state_lock(sk
);
2242 if (sock_flag(sk
, SOCK_DEAD
))
2245 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA
, sk
);
2248 finish_wait(sk_sleep(sk
), &wait
);
2249 unix_state_unlock(sk
);
2253 static unsigned int unix_skb_len(const struct sk_buff
*skb
)
2255 return skb
->len
- UNIXCB(skb
).consumed
;
2258 struct unix_stream_read_state
{
2259 int (*recv_actor
)(struct sk_buff
*, int, int,
2260 struct unix_stream_read_state
*);
2261 struct socket
*socket
;
2263 struct pipe_inode_info
*pipe
;
2266 unsigned int splice_flags
;
2269 static int unix_stream_read_generic(struct unix_stream_read_state
*state
,
2272 struct scm_cookie scm
;
2273 struct socket
*sock
= state
->socket
;
2274 struct sock
*sk
= sock
->sk
;
2275 struct unix_sock
*u
= unix_sk(sk
);
2277 int flags
= state
->flags
;
2278 int noblock
= flags
& MSG_DONTWAIT
;
2279 bool check_creds
= false;
2284 size_t size
= state
->size
;
2285 unsigned int last_len
;
2287 if (unlikely(sk
->sk_state
!= TCP_ESTABLISHED
)) {
2292 if (unlikely(flags
& MSG_OOB
)) {
2297 target
= sock_rcvlowat(sk
, flags
& MSG_WAITALL
, size
);
2298 timeo
= sock_rcvtimeo(sk
, noblock
);
2300 memset(&scm
, 0, sizeof(scm
));
2302 /* Lock the socket to prevent queue disordering
2303 * while sleeps in memcpy_tomsg
2305 mutex_lock(&u
->iolock
);
2307 skip
= max(sk_peek_offset(sk
, flags
), 0);
2312 struct sk_buff
*skb
, *last
;
2315 unix_state_lock(sk
);
2316 if (sock_flag(sk
, SOCK_DEAD
)) {
2320 last
= skb
= skb_peek(&sk
->sk_receive_queue
);
2321 last_len
= last
? last
->len
: 0;
2324 unix_sk(sk
)->recursion_level
= 0;
2325 if (copied
>= target
)
2329 * POSIX 1003.1g mandates this order.
2332 err
= sock_error(sk
);
2335 if (sk
->sk_shutdown
& RCV_SHUTDOWN
)
2338 unix_state_unlock(sk
);
2344 mutex_unlock(&u
->iolock
);
2346 timeo
= unix_stream_data_wait(sk
, timeo
, last
,
2347 last_len
, freezable
);
2349 if (signal_pending(current
)) {
2350 err
= sock_intr_errno(timeo
);
2355 mutex_lock(&u
->iolock
);
2358 unix_state_unlock(sk
);
2362 while (skip
>= unix_skb_len(skb
)) {
2363 skip
-= unix_skb_len(skb
);
2365 last_len
= skb
->len
;
2366 skb
= skb_peek_next(skb
, &sk
->sk_receive_queue
);
2371 unix_state_unlock(sk
);
2374 /* Never glue messages from different writers */
2375 if (!unix_skb_scm_eq(skb
, &scm
))
2377 } else if (test_bit(SOCK_PASSCRED
, &sock
->flags
)) {
2378 /* Copy credentials */
2379 scm_set_cred(&scm
, UNIXCB(skb
).pid
, UNIXCB(skb
).uid
, UNIXCB(skb
).gid
);
2380 unix_set_secdata(&scm
, skb
);
2384 /* Copy address just once */
2385 if (state
->msg
&& state
->msg
->msg_name
) {
2386 DECLARE_SOCKADDR(struct sockaddr_un
*, sunaddr
,
2387 state
->msg
->msg_name
);
2388 unix_copy_addr(state
->msg
, skb
->sk
);
2392 chunk
= min_t(unsigned int, unix_skb_len(skb
) - skip
, size
);
2394 chunk
= state
->recv_actor(skb
, skip
, chunk
, state
);
2395 drop_skb
= !unix_skb_len(skb
);
2396 /* skb is only safe to use if !drop_skb */
2407 /* the skb was touched by a concurrent reader;
2408 * we should not expect anything from this skb
2409 * anymore and assume it invalid - we can be
2410 * sure it was dropped from the socket queue
2412 * let's report a short read
2418 /* Mark read part of skb as used */
2419 if (!(flags
& MSG_PEEK
)) {
2420 UNIXCB(skb
).consumed
+= chunk
;
2422 sk_peek_offset_bwd(sk
, chunk
);
2425 unix_detach_fds(&scm
, skb
);
2427 if (unix_skb_len(skb
))
2430 skb_unlink(skb
, &sk
->sk_receive_queue
);
2436 /* It is questionable, see note in unix_dgram_recvmsg.
2439 scm
.fp
= scm_fp_dup(UNIXCB(skb
).fp
);
2441 sk_peek_offset_fwd(sk
, chunk
);
2448 last_len
= skb
->len
;
2449 unix_state_lock(sk
);
2450 skb
= skb_peek_next(skb
, &sk
->sk_receive_queue
);
2453 unix_state_unlock(sk
);
2458 mutex_unlock(&u
->iolock
);
2460 scm_recv(sock
, state
->msg
, &scm
, flags
);
2464 return copied
? : err
;
2467 static int unix_stream_read_actor(struct sk_buff
*skb
,
2468 int skip
, int chunk
,
2469 struct unix_stream_read_state
*state
)
2473 ret
= skb_copy_datagram_msg(skb
, UNIXCB(skb
).consumed
+ skip
,
2475 return ret
?: chunk
;
2478 static int unix_stream_recvmsg(struct socket
*sock
, struct msghdr
*msg
,
2479 size_t size
, int flags
)
2481 struct unix_stream_read_state state
= {
2482 .recv_actor
= unix_stream_read_actor
,
2489 return unix_stream_read_generic(&state
, true);
2492 static int unix_stream_splice_actor(struct sk_buff
*skb
,
2493 int skip
, int chunk
,
2494 struct unix_stream_read_state
*state
)
2496 return skb_splice_bits(skb
, state
->socket
->sk
,
2497 UNIXCB(skb
).consumed
+ skip
,
2498 state
->pipe
, chunk
, state
->splice_flags
);
2501 static ssize_t
unix_stream_splice_read(struct socket
*sock
, loff_t
*ppos
,
2502 struct pipe_inode_info
*pipe
,
2503 size_t size
, unsigned int flags
)
2505 struct unix_stream_read_state state
= {
2506 .recv_actor
= unix_stream_splice_actor
,
2510 .splice_flags
= flags
,
2513 if (unlikely(*ppos
))
2516 if (sock
->file
->f_flags
& O_NONBLOCK
||
2517 flags
& SPLICE_F_NONBLOCK
)
2518 state
.flags
= MSG_DONTWAIT
;
2520 return unix_stream_read_generic(&state
, false);
2523 static int unix_shutdown(struct socket
*sock
, int mode
)
2525 struct sock
*sk
= sock
->sk
;
2528 if (mode
< SHUT_RD
|| mode
> SHUT_RDWR
)
2531 * SHUT_RD (0) -> RCV_SHUTDOWN (1)
2532 * SHUT_WR (1) -> SEND_SHUTDOWN (2)
2533 * SHUT_RDWR (2) -> SHUTDOWN_MASK (3)
2537 unix_state_lock(sk
);
2538 sk
->sk_shutdown
|= mode
;
2539 other
= unix_peer(sk
);
2542 unix_state_unlock(sk
);
2543 sk
->sk_state_change(sk
);
2546 (sk
->sk_type
== SOCK_STREAM
|| sk
->sk_type
== SOCK_SEQPACKET
)) {
2550 if (mode
&RCV_SHUTDOWN
)
2551 peer_mode
|= SEND_SHUTDOWN
;
2552 if (mode
&SEND_SHUTDOWN
)
2553 peer_mode
|= RCV_SHUTDOWN
;
2554 unix_state_lock(other
);
2555 other
->sk_shutdown
|= peer_mode
;
2556 unix_state_unlock(other
);
2557 other
->sk_state_change(other
);
2558 if (peer_mode
== SHUTDOWN_MASK
)
2559 sk_wake_async(other
, SOCK_WAKE_WAITD
, POLL_HUP
);
2560 else if (peer_mode
& RCV_SHUTDOWN
)
2561 sk_wake_async(other
, SOCK_WAKE_WAITD
, POLL_IN
);
2569 long unix_inq_len(struct sock
*sk
)
2571 struct sk_buff
*skb
;
2574 if (sk
->sk_state
== TCP_LISTEN
)
2577 spin_lock(&sk
->sk_receive_queue
.lock
);
2578 if (sk
->sk_type
== SOCK_STREAM
||
2579 sk
->sk_type
== SOCK_SEQPACKET
) {
2580 skb_queue_walk(&sk
->sk_receive_queue
, skb
)
2581 amount
+= unix_skb_len(skb
);
2583 skb
= skb_peek(&sk
->sk_receive_queue
);
2587 spin_unlock(&sk
->sk_receive_queue
.lock
);
2591 EXPORT_SYMBOL_GPL(unix_inq_len
);
2593 long unix_outq_len(struct sock
*sk
)
2595 return sk_wmem_alloc_get(sk
);
2597 EXPORT_SYMBOL_GPL(unix_outq_len
);
2599 static int unix_open_file(struct sock
*sk
)
2605 if (!ns_capable(sock_net(sk
)->user_ns
, CAP_NET_ADMIN
))
2608 unix_state_lock(sk
);
2609 path
= unix_sk(sk
)->path
;
2611 unix_state_unlock(sk
);
2616 unix_state_unlock(sk
);
2618 fd
= get_unused_fd_flags(O_CLOEXEC
);
2622 f
= dentry_open(&path
, O_PATH
, current_cred());
2636 static int unix_ioctl(struct socket
*sock
, unsigned int cmd
, unsigned long arg
)
2638 struct sock
*sk
= sock
->sk
;
2644 amount
= unix_outq_len(sk
);
2645 err
= put_user(amount
, (int __user
*)arg
);
2648 amount
= unix_inq_len(sk
);
2652 err
= put_user(amount
, (int __user
*)arg
);
2655 err
= unix_open_file(sk
);
2664 static unsigned int unix_poll(struct file
*file
, struct socket
*sock
, poll_table
*wait
)
2666 struct sock
*sk
= sock
->sk
;
2669 sock_poll_wait(file
, sk_sleep(sk
), wait
);
2672 /* exceptional events? */
2675 if (sk
->sk_shutdown
== SHUTDOWN_MASK
)
2677 if (sk
->sk_shutdown
& RCV_SHUTDOWN
)
2678 mask
|= POLLRDHUP
| POLLIN
| POLLRDNORM
;
2681 if (!skb_queue_empty(&sk
->sk_receive_queue
))
2682 mask
|= POLLIN
| POLLRDNORM
;
2684 /* Connection-based need to check for termination and startup */
2685 if ((sk
->sk_type
== SOCK_STREAM
|| sk
->sk_type
== SOCK_SEQPACKET
) &&
2686 sk
->sk_state
== TCP_CLOSE
)
2690 * we set writable also when the other side has shut down the
2691 * connection. This prevents stuck sockets.
2693 if (unix_writable(sk
))
2694 mask
|= POLLOUT
| POLLWRNORM
| POLLWRBAND
;
2699 static unsigned int unix_dgram_poll(struct file
*file
, struct socket
*sock
,
2702 struct sock
*sk
= sock
->sk
, *other
;
2703 unsigned int mask
, writable
;
2705 sock_poll_wait(file
, sk_sleep(sk
), wait
);
2708 /* exceptional events? */
2709 if (sk
->sk_err
|| !skb_queue_empty(&sk
->sk_error_queue
))
2711 (sock_flag(sk
, SOCK_SELECT_ERR_QUEUE
) ? POLLPRI
: 0);
2713 if (sk
->sk_shutdown
& RCV_SHUTDOWN
)
2714 mask
|= POLLRDHUP
| POLLIN
| POLLRDNORM
;
2715 if (sk
->sk_shutdown
== SHUTDOWN_MASK
)
2719 if (!skb_queue_empty(&sk
->sk_receive_queue
))
2720 mask
|= POLLIN
| POLLRDNORM
;
2722 /* Connection-based need to check for termination and startup */
2723 if (sk
->sk_type
== SOCK_SEQPACKET
) {
2724 if (sk
->sk_state
== TCP_CLOSE
)
2726 /* connection hasn't started yet? */
2727 if (sk
->sk_state
== TCP_SYN_SENT
)
2731 /* No write status requested, avoid expensive OUT tests. */
2732 if (!(poll_requested_events(wait
) & (POLLWRBAND
|POLLWRNORM
|POLLOUT
)))
2735 writable
= unix_writable(sk
);
2737 unix_state_lock(sk
);
2739 other
= unix_peer(sk
);
2740 if (other
&& unix_peer(other
) != sk
&&
2741 unix_recvq_full(other
) &&
2742 unix_dgram_peer_wake_me(sk
, other
))
2745 unix_state_unlock(sk
);
2749 mask
|= POLLOUT
| POLLWRNORM
| POLLWRBAND
;
2751 sk_set_bit(SOCKWQ_ASYNC_NOSPACE
, sk
);
2756 #ifdef CONFIG_PROC_FS
2758 #define BUCKET_SPACE (BITS_PER_LONG - (UNIX_HASH_BITS + 1) - 1)
2760 #define get_bucket(x) ((x) >> BUCKET_SPACE)
2761 #define get_offset(x) ((x) & ((1L << BUCKET_SPACE) - 1))
2762 #define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
2764 static struct sock
*unix_from_bucket(struct seq_file
*seq
, loff_t
*pos
)
2766 unsigned long offset
= get_offset(*pos
);
2767 unsigned long bucket
= get_bucket(*pos
);
2769 unsigned long count
= 0;
2771 for (sk
= sk_head(&unix_socket_table
[bucket
]); sk
; sk
= sk_next(sk
)) {
2772 if (sock_net(sk
) != seq_file_net(seq
))
2774 if (++count
== offset
)
2781 static struct sock
*unix_next_socket(struct seq_file
*seq
,
2785 unsigned long bucket
;
2787 while (sk
> (struct sock
*)SEQ_START_TOKEN
) {
2791 if (sock_net(sk
) == seq_file_net(seq
))
2796 sk
= unix_from_bucket(seq
, pos
);
2801 bucket
= get_bucket(*pos
) + 1;
2802 *pos
= set_bucket_offset(bucket
, 1);
2803 } while (bucket
< ARRAY_SIZE(unix_socket_table
));
2808 static void *unix_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2809 __acquires(unix_table_lock
)
2811 spin_lock(&unix_table_lock
);
2814 return SEQ_START_TOKEN
;
2816 if (get_bucket(*pos
) >= ARRAY_SIZE(unix_socket_table
))
2819 return unix_next_socket(seq
, NULL
, pos
);
2822 static void *unix_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2825 return unix_next_socket(seq
, v
, pos
);
2828 static void unix_seq_stop(struct seq_file
*seq
, void *v
)
2829 __releases(unix_table_lock
)
2831 spin_unlock(&unix_table_lock
);
2834 static int unix_seq_show(struct seq_file
*seq
, void *v
)
2837 if (v
== SEQ_START_TOKEN
)
2838 seq_puts(seq
, "Num RefCount Protocol Flags Type St "
2842 struct unix_sock
*u
= unix_sk(s
);
2845 seq_printf(seq
, "%pK: %08X %08X %08X %04X %02X %5lu",
2847 refcount_read(&s
->sk_refcnt
),
2849 s
->sk_state
== TCP_LISTEN
? __SO_ACCEPTCON
: 0,
2852 (s
->sk_state
== TCP_ESTABLISHED
? SS_CONNECTED
: SS_UNCONNECTED
) :
2853 (s
->sk_state
== TCP_ESTABLISHED
? SS_CONNECTING
: SS_DISCONNECTING
),
2861 len
= u
->addr
->len
- sizeof(short);
2862 if (!UNIX_ABSTRACT(s
))
2868 for ( ; i
< len
; i
++)
2869 seq_putc(seq
, u
->addr
->name
->sun_path
[i
] ?:
2872 unix_state_unlock(s
);
2873 seq_putc(seq
, '\n');
2879 static const struct seq_operations unix_seq_ops
= {
2880 .start
= unix_seq_start
,
2881 .next
= unix_seq_next
,
2882 .stop
= unix_seq_stop
,
2883 .show
= unix_seq_show
,
2886 static int unix_seq_open(struct inode
*inode
, struct file
*file
)
2888 return seq_open_net(inode
, file
, &unix_seq_ops
,
2889 sizeof(struct seq_net_private
));
2892 static const struct file_operations unix_seq_fops
= {
2893 .owner
= THIS_MODULE
,
2894 .open
= unix_seq_open
,
2896 .llseek
= seq_lseek
,
2897 .release
= seq_release_net
,
2902 static const struct net_proto_family unix_family_ops
= {
2904 .create
= unix_create
,
2905 .owner
= THIS_MODULE
,
2909 static int __net_init
unix_net_init(struct net
*net
)
2911 int error
= -ENOMEM
;
2913 net
->unx
.sysctl_max_dgram_qlen
= 10;
2914 if (unix_sysctl_register(net
))
2917 #ifdef CONFIG_PROC_FS
2918 if (!proc_create("unix", 0, net
->proc_net
, &unix_seq_fops
)) {
2919 unix_sysctl_unregister(net
);
2928 static void __net_exit
unix_net_exit(struct net
*net
)
2930 unix_sysctl_unregister(net
);
2931 remove_proc_entry("unix", net
->proc_net
);
2934 static struct pernet_operations unix_net_ops
= {
2935 .init
= unix_net_init
,
2936 .exit
= unix_net_exit
,
2939 static int __init
af_unix_init(void)
2943 BUILD_BUG_ON(sizeof(struct unix_skb_parms
) > FIELD_SIZEOF(struct sk_buff
, cb
));
2945 rc
= proto_register(&unix_proto
, 1);
2947 pr_crit("%s: Cannot create unix_sock SLAB cache!\n", __func__
);
2951 sock_register(&unix_family_ops
);
2952 register_pernet_subsys(&unix_net_ops
);
2957 static void __exit
af_unix_exit(void)
2959 sock_unregister(PF_UNIX
);
2960 proto_unregister(&unix_proto
);
2961 unregister_pernet_subsys(&unix_net_ops
);
2964 /* Earlier than device_initcall() so that other drivers invoking
2965 request_module() don't end up in a loop when modprobe tries
2966 to use a UNIX socket. But later than subsys_initcall() because
2967 we depend on stuff initialised there */
2968 fs_initcall(af_unix_init
);
2969 module_exit(af_unix_exit
);
2971 MODULE_LICENSE("GPL");
2972 MODULE_ALIAS_NETPROTO(PF_UNIX
);