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 (atomic_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_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_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 (atomic_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(atomic_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);
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 atomic_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
= { NULL
, NULL
};
1002 if (sunaddr
->sun_family
!= AF_UNIX
)
1005 if (addr_len
== sizeof(short)) {
1006 err
= unix_autobind(sock
);
1010 err
= unix_mkname(sunaddr
, addr_len
, &hash
);
1016 umode_t mode
= S_IFSOCK
|
1017 (SOCK_INODE(sock
)->i_mode
& ~current_umask());
1018 err
= unix_mknod(sun_path
, mode
, &path
);
1026 err
= mutex_lock_interruptible(&u
->bindlock
);
1035 addr
= kmalloc(sizeof(*addr
)+addr_len
, GFP_KERNEL
);
1039 memcpy(addr
->name
, sunaddr
, addr_len
);
1040 addr
->len
= addr_len
;
1041 addr
->hash
= hash
^ sk
->sk_type
;
1042 atomic_set(&addr
->refcnt
, 1);
1045 addr
->hash
= UNIX_HASH_SIZE
;
1046 hash
= d_backing_inode(path
.dentry
)->i_ino
& (UNIX_HASH_SIZE
- 1);
1047 spin_lock(&unix_table_lock
);
1049 list
= &unix_socket_table
[hash
];
1051 spin_lock(&unix_table_lock
);
1053 if (__unix_find_socket_byname(net
, sunaddr
, addr_len
,
1054 sk
->sk_type
, hash
)) {
1055 unix_release_addr(addr
);
1059 list
= &unix_socket_table
[addr
->hash
];
1063 __unix_remove_socket(sk
);
1065 __unix_insert_socket(list
, sk
);
1068 spin_unlock(&unix_table_lock
);
1070 mutex_unlock(&u
->bindlock
);
1078 static void unix_state_double_lock(struct sock
*sk1
, struct sock
*sk2
)
1080 if (unlikely(sk1
== sk2
) || !sk2
) {
1081 unix_state_lock(sk1
);
1085 unix_state_lock(sk1
);
1086 unix_state_lock_nested(sk2
);
1088 unix_state_lock(sk2
);
1089 unix_state_lock_nested(sk1
);
1093 static void unix_state_double_unlock(struct sock
*sk1
, struct sock
*sk2
)
1095 if (unlikely(sk1
== sk2
) || !sk2
) {
1096 unix_state_unlock(sk1
);
1099 unix_state_unlock(sk1
);
1100 unix_state_unlock(sk2
);
1103 static int unix_dgram_connect(struct socket
*sock
, struct sockaddr
*addr
,
1104 int alen
, int flags
)
1106 struct sock
*sk
= sock
->sk
;
1107 struct net
*net
= sock_net(sk
);
1108 struct sockaddr_un
*sunaddr
= (struct sockaddr_un
*)addr
;
1113 if (addr
->sa_family
!= AF_UNSPEC
) {
1114 err
= unix_mkname(sunaddr
, alen
, &hash
);
1119 if (test_bit(SOCK_PASSCRED
, &sock
->flags
) &&
1120 !unix_sk(sk
)->addr
&& (err
= unix_autobind(sock
)) != 0)
1124 other
= unix_find_other(net
, sunaddr
, alen
, sock
->type
, hash
, &err
);
1128 unix_state_double_lock(sk
, other
);
1130 /* Apparently VFS overslept socket death. Retry. */
1131 if (sock_flag(other
, SOCK_DEAD
)) {
1132 unix_state_double_unlock(sk
, other
);
1138 if (!unix_may_send(sk
, other
))
1141 err
= security_unix_may_send(sk
->sk_socket
, other
->sk_socket
);
1147 * 1003.1g breaking connected state with AF_UNSPEC
1150 unix_state_double_lock(sk
, other
);
1154 * If it was connected, reconnect.
1156 if (unix_peer(sk
)) {
1157 struct sock
*old_peer
= unix_peer(sk
);
1158 unix_peer(sk
) = other
;
1159 unix_dgram_peer_wake_disconnect_wakeup(sk
, old_peer
);
1161 unix_state_double_unlock(sk
, other
);
1163 if (other
!= old_peer
)
1164 unix_dgram_disconnected(sk
, old_peer
);
1167 unix_peer(sk
) = other
;
1168 unix_state_double_unlock(sk
, other
);
1173 unix_state_double_unlock(sk
, other
);
1179 static long unix_wait_for_peer(struct sock
*other
, long timeo
)
1181 struct unix_sock
*u
= unix_sk(other
);
1185 prepare_to_wait_exclusive(&u
->peer_wait
, &wait
, TASK_INTERRUPTIBLE
);
1187 sched
= !sock_flag(other
, SOCK_DEAD
) &&
1188 !(other
->sk_shutdown
& RCV_SHUTDOWN
) &&
1189 unix_recvq_full(other
);
1191 unix_state_unlock(other
);
1194 timeo
= schedule_timeout(timeo
);
1196 finish_wait(&u
->peer_wait
, &wait
);
1200 static int unix_stream_connect(struct socket
*sock
, struct sockaddr
*uaddr
,
1201 int addr_len
, int flags
)
1203 struct sockaddr_un
*sunaddr
= (struct sockaddr_un
*)uaddr
;
1204 struct sock
*sk
= sock
->sk
;
1205 struct net
*net
= sock_net(sk
);
1206 struct unix_sock
*u
= unix_sk(sk
), *newu
, *otheru
;
1207 struct sock
*newsk
= NULL
;
1208 struct sock
*other
= NULL
;
1209 struct sk_buff
*skb
= NULL
;
1215 err
= unix_mkname(sunaddr
, addr_len
, &hash
);
1220 if (test_bit(SOCK_PASSCRED
, &sock
->flags
) && !u
->addr
&&
1221 (err
= unix_autobind(sock
)) != 0)
1224 timeo
= sock_sndtimeo(sk
, flags
& O_NONBLOCK
);
1226 /* First of all allocate resources.
1227 If we will make it after state is locked,
1228 we will have to recheck all again in any case.
1233 /* create new sock for complete connection */
1234 newsk
= unix_create1(sock_net(sk
), NULL
, 0);
1238 /* Allocate skb for sending to listening sock */
1239 skb
= sock_wmalloc(newsk
, 1, 0, GFP_KERNEL
);
1244 /* Find listening sock. */
1245 other
= unix_find_other(net
, sunaddr
, addr_len
, sk
->sk_type
, hash
, &err
);
1249 /* Latch state of peer */
1250 unix_state_lock(other
);
1252 /* Apparently VFS overslept socket death. Retry. */
1253 if (sock_flag(other
, SOCK_DEAD
)) {
1254 unix_state_unlock(other
);
1259 err
= -ECONNREFUSED
;
1260 if (other
->sk_state
!= TCP_LISTEN
)
1262 if (other
->sk_shutdown
& RCV_SHUTDOWN
)
1265 if (unix_recvq_full(other
)) {
1270 timeo
= unix_wait_for_peer(other
, timeo
);
1272 err
= sock_intr_errno(timeo
);
1273 if (signal_pending(current
))
1281 It is tricky place. We need to grab our state lock and cannot
1282 drop lock on peer. It is dangerous because deadlock is
1283 possible. Connect to self case and simultaneous
1284 attempt to connect are eliminated by checking socket
1285 state. other is TCP_LISTEN, if sk is TCP_LISTEN we
1286 check this before attempt to grab lock.
1288 Well, and we have to recheck the state after socket locked.
1294 /* This is ok... continue with connect */
1296 case TCP_ESTABLISHED
:
1297 /* Socket is already connected */
1305 unix_state_lock_nested(sk
);
1307 if (sk
->sk_state
!= st
) {
1308 unix_state_unlock(sk
);
1309 unix_state_unlock(other
);
1314 err
= security_unix_stream_connect(sk
, other
, newsk
);
1316 unix_state_unlock(sk
);
1320 /* The way is open! Fastly set all the necessary fields... */
1323 unix_peer(newsk
) = sk
;
1324 newsk
->sk_state
= TCP_ESTABLISHED
;
1325 newsk
->sk_type
= sk
->sk_type
;
1326 init_peercred(newsk
);
1327 newu
= unix_sk(newsk
);
1328 RCU_INIT_POINTER(newsk
->sk_wq
, &newu
->peer_wq
);
1329 otheru
= unix_sk(other
);
1331 /* copy address information from listening to new sock*/
1333 atomic_inc(&otheru
->addr
->refcnt
);
1334 newu
->addr
= otheru
->addr
;
1336 if (otheru
->path
.dentry
) {
1337 path_get(&otheru
->path
);
1338 newu
->path
= otheru
->path
;
1341 /* Set credentials */
1342 copy_peercred(sk
, other
);
1344 sock
->state
= SS_CONNECTED
;
1345 sk
->sk_state
= TCP_ESTABLISHED
;
1348 smp_mb__after_atomic(); /* sock_hold() does an atomic_inc() */
1349 unix_peer(sk
) = newsk
;
1351 unix_state_unlock(sk
);
1353 /* take ten and and send info to listening sock */
1354 spin_lock(&other
->sk_receive_queue
.lock
);
1355 __skb_queue_tail(&other
->sk_receive_queue
, skb
);
1356 spin_unlock(&other
->sk_receive_queue
.lock
);
1357 unix_state_unlock(other
);
1358 other
->sk_data_ready(other
);
1364 unix_state_unlock(other
);
1369 unix_release_sock(newsk
, 0);
1375 static int unix_socketpair(struct socket
*socka
, struct socket
*sockb
)
1377 struct sock
*ska
= socka
->sk
, *skb
= sockb
->sk
;
1379 /* Join our sockets back to back */
1382 unix_peer(ska
) = skb
;
1383 unix_peer(skb
) = ska
;
1387 if (ska
->sk_type
!= SOCK_DGRAM
) {
1388 ska
->sk_state
= TCP_ESTABLISHED
;
1389 skb
->sk_state
= TCP_ESTABLISHED
;
1390 socka
->state
= SS_CONNECTED
;
1391 sockb
->state
= SS_CONNECTED
;
1396 static void unix_sock_inherit_flags(const struct socket
*old
,
1399 if (test_bit(SOCK_PASSCRED
, &old
->flags
))
1400 set_bit(SOCK_PASSCRED
, &new->flags
);
1401 if (test_bit(SOCK_PASSSEC
, &old
->flags
))
1402 set_bit(SOCK_PASSSEC
, &new->flags
);
1405 static int unix_accept(struct socket
*sock
, struct socket
*newsock
, int flags
)
1407 struct sock
*sk
= sock
->sk
;
1409 struct sk_buff
*skb
;
1413 if (sock
->type
!= SOCK_STREAM
&& sock
->type
!= SOCK_SEQPACKET
)
1417 if (sk
->sk_state
!= TCP_LISTEN
)
1420 /* If socket state is TCP_LISTEN it cannot change (for now...),
1421 * so that no locks are necessary.
1424 skb
= skb_recv_datagram(sk
, 0, flags
&O_NONBLOCK
, &err
);
1426 /* This means receive shutdown. */
1433 skb_free_datagram(sk
, skb
);
1434 wake_up_interruptible(&unix_sk(sk
)->peer_wait
);
1436 /* attach accepted sock to socket */
1437 unix_state_lock(tsk
);
1438 newsock
->state
= SS_CONNECTED
;
1439 unix_sock_inherit_flags(sock
, newsock
);
1440 sock_graft(tsk
, newsock
);
1441 unix_state_unlock(tsk
);
1449 static int unix_getname(struct socket
*sock
, struct sockaddr
*uaddr
, int *uaddr_len
, int peer
)
1451 struct sock
*sk
= sock
->sk
;
1452 struct unix_sock
*u
;
1453 DECLARE_SOCKADDR(struct sockaddr_un
*, sunaddr
, uaddr
);
1457 sk
= unix_peer_get(sk
);
1468 unix_state_lock(sk
);
1470 sunaddr
->sun_family
= AF_UNIX
;
1471 sunaddr
->sun_path
[0] = 0;
1472 *uaddr_len
= sizeof(short);
1474 struct unix_address
*addr
= u
->addr
;
1476 *uaddr_len
= addr
->len
;
1477 memcpy(sunaddr
, addr
->name
, *uaddr_len
);
1479 unix_state_unlock(sk
);
1485 static void unix_detach_fds(struct scm_cookie
*scm
, struct sk_buff
*skb
)
1489 scm
->fp
= UNIXCB(skb
).fp
;
1490 UNIXCB(skb
).fp
= NULL
;
1492 for (i
= scm
->fp
->count
-1; i
>= 0; i
--)
1493 unix_notinflight(scm
->fp
->user
, scm
->fp
->fp
[i
]);
1496 static void unix_destruct_scm(struct sk_buff
*skb
)
1498 struct scm_cookie scm
;
1499 memset(&scm
, 0, sizeof(scm
));
1500 scm
.pid
= UNIXCB(skb
).pid
;
1502 unix_detach_fds(&scm
, skb
);
1504 /* Alas, it calls VFS */
1505 /* So fscking what? fput() had been SMP-safe since the last Summer */
1511 * The "user->unix_inflight" variable is protected by the garbage
1512 * collection lock, and we just read it locklessly here. If you go
1513 * over the limit, there might be a tiny race in actually noticing
1514 * it across threads. Tough.
1516 static inline bool too_many_unix_fds(struct task_struct
*p
)
1518 struct user_struct
*user
= current_user();
1520 if (unlikely(user
->unix_inflight
> task_rlimit(p
, RLIMIT_NOFILE
)))
1521 return !capable(CAP_SYS_RESOURCE
) && !capable(CAP_SYS_ADMIN
);
1525 #define MAX_RECURSION_LEVEL 4
1527 static int unix_attach_fds(struct scm_cookie
*scm
, struct sk_buff
*skb
)
1530 unsigned char max_level
= 0;
1532 if (too_many_unix_fds(current
))
1533 return -ETOOMANYREFS
;
1535 for (i
= scm
->fp
->count
- 1; i
>= 0; i
--) {
1536 struct sock
*sk
= unix_get_socket(scm
->fp
->fp
[i
]);
1539 max_level
= max(max_level
,
1540 unix_sk(sk
)->recursion_level
);
1542 if (unlikely(max_level
> MAX_RECURSION_LEVEL
))
1543 return -ETOOMANYREFS
;
1546 * Need to duplicate file references for the sake of garbage
1547 * collection. Otherwise a socket in the fps might become a
1548 * candidate for GC while the skb is not yet queued.
1550 UNIXCB(skb
).fp
= scm_fp_dup(scm
->fp
);
1551 if (!UNIXCB(skb
).fp
)
1554 for (i
= scm
->fp
->count
- 1; i
>= 0; i
--)
1555 unix_inflight(scm
->fp
->user
, scm
->fp
->fp
[i
]);
1559 static int unix_scm_to_skb(struct scm_cookie
*scm
, struct sk_buff
*skb
, bool send_fds
)
1563 UNIXCB(skb
).pid
= get_pid(scm
->pid
);
1564 UNIXCB(skb
).uid
= scm
->creds
.uid
;
1565 UNIXCB(skb
).gid
= scm
->creds
.gid
;
1566 UNIXCB(skb
).fp
= NULL
;
1567 unix_get_secdata(scm
, skb
);
1568 if (scm
->fp
&& send_fds
)
1569 err
= unix_attach_fds(scm
, skb
);
1571 skb
->destructor
= unix_destruct_scm
;
1575 static bool unix_passcred_enabled(const struct socket
*sock
,
1576 const struct sock
*other
)
1578 return test_bit(SOCK_PASSCRED
, &sock
->flags
) ||
1579 !other
->sk_socket
||
1580 test_bit(SOCK_PASSCRED
, &other
->sk_socket
->flags
);
1584 * Some apps rely on write() giving SCM_CREDENTIALS
1585 * We include credentials if source or destination socket
1586 * asserted SOCK_PASSCRED.
1588 static void maybe_add_creds(struct sk_buff
*skb
, const struct socket
*sock
,
1589 const struct sock
*other
)
1591 if (UNIXCB(skb
).pid
)
1593 if (unix_passcred_enabled(sock
, other
)) {
1594 UNIXCB(skb
).pid
= get_pid(task_tgid(current
));
1595 current_uid_gid(&UNIXCB(skb
).uid
, &UNIXCB(skb
).gid
);
1599 static int maybe_init_creds(struct scm_cookie
*scm
,
1600 struct socket
*socket
,
1601 const struct sock
*other
)
1604 struct msghdr msg
= { .msg_controllen
= 0 };
1606 err
= scm_send(socket
, &msg
, scm
, false);
1610 if (unix_passcred_enabled(socket
, other
)) {
1611 scm
->pid
= get_pid(task_tgid(current
));
1612 current_uid_gid(&scm
->creds
.uid
, &scm
->creds
.gid
);
1617 static bool unix_skb_scm_eq(struct sk_buff
*skb
,
1618 struct scm_cookie
*scm
)
1620 const struct unix_skb_parms
*u
= &UNIXCB(skb
);
1622 return u
->pid
== scm
->pid
&&
1623 uid_eq(u
->uid
, scm
->creds
.uid
) &&
1624 gid_eq(u
->gid
, scm
->creds
.gid
) &&
1625 unix_secdata_eq(scm
, skb
);
1629 * Send AF_UNIX data.
1632 static int unix_dgram_sendmsg(struct socket
*sock
, struct msghdr
*msg
,
1635 struct sock
*sk
= sock
->sk
;
1636 struct net
*net
= sock_net(sk
);
1637 struct unix_sock
*u
= unix_sk(sk
);
1638 DECLARE_SOCKADDR(struct sockaddr_un
*, sunaddr
, msg
->msg_name
);
1639 struct sock
*other
= NULL
;
1640 int namelen
= 0; /* fake GCC */
1643 struct sk_buff
*skb
;
1645 struct scm_cookie scm
;
1651 err
= scm_send(sock
, msg
, &scm
, false);
1656 if (msg
->msg_flags
&MSG_OOB
)
1659 if (msg
->msg_namelen
) {
1660 err
= unix_mkname(sunaddr
, msg
->msg_namelen
, &hash
);
1667 other
= unix_peer_get(sk
);
1672 if (test_bit(SOCK_PASSCRED
, &sock
->flags
) && !u
->addr
1673 && (err
= unix_autobind(sock
)) != 0)
1677 if (len
> sk
->sk_sndbuf
- 32)
1680 if (len
> SKB_MAX_ALLOC
) {
1681 data_len
= min_t(size_t,
1682 len
- SKB_MAX_ALLOC
,
1683 MAX_SKB_FRAGS
* PAGE_SIZE
);
1684 data_len
= PAGE_ALIGN(data_len
);
1686 BUILD_BUG_ON(SKB_MAX_ALLOC
< PAGE_SIZE
);
1689 skb
= sock_alloc_send_pskb(sk
, len
- data_len
, data_len
,
1690 msg
->msg_flags
& MSG_DONTWAIT
, &err
,
1691 PAGE_ALLOC_COSTLY_ORDER
);
1695 err
= unix_scm_to_skb(&scm
, skb
, true);
1698 max_level
= err
+ 1;
1700 skb_put(skb
, len
- data_len
);
1701 skb
->data_len
= data_len
;
1703 err
= skb_copy_datagram_from_iter(skb
, 0, &msg
->msg_iter
, len
);
1707 timeo
= sock_sndtimeo(sk
, msg
->msg_flags
& MSG_DONTWAIT
);
1712 if (sunaddr
== NULL
)
1715 other
= unix_find_other(net
, sunaddr
, namelen
, sk
->sk_type
,
1721 if (sk_filter(other
, skb
) < 0) {
1722 /* Toss the packet but do not return any error to the sender */
1728 unix_state_lock(other
);
1731 if (!unix_may_send(sk
, other
))
1734 if (unlikely(sock_flag(other
, SOCK_DEAD
))) {
1736 * Check with 1003.1g - what should
1739 unix_state_unlock(other
);
1743 unix_state_lock(sk
);
1746 if (unix_peer(sk
) == other
) {
1747 unix_peer(sk
) = NULL
;
1748 unix_dgram_peer_wake_disconnect_wakeup(sk
, other
);
1750 unix_state_unlock(sk
);
1752 unix_dgram_disconnected(sk
, other
);
1754 err
= -ECONNREFUSED
;
1756 unix_state_unlock(sk
);
1766 if (other
->sk_shutdown
& RCV_SHUTDOWN
)
1769 if (sk
->sk_type
!= SOCK_SEQPACKET
) {
1770 err
= security_unix_may_send(sk
->sk_socket
, other
->sk_socket
);
1775 /* other == sk && unix_peer(other) != sk if
1776 * - unix_peer(sk) == NULL, destination address bound to sk
1777 * - unix_peer(sk) == sk by time of get but disconnected before lock
1780 unlikely(unix_peer(other
) != sk
&& unix_recvq_full(other
))) {
1782 timeo
= unix_wait_for_peer(other
, timeo
);
1784 err
= sock_intr_errno(timeo
);
1785 if (signal_pending(current
))
1792 unix_state_unlock(other
);
1793 unix_state_double_lock(sk
, other
);
1796 if (unix_peer(sk
) != other
||
1797 unix_dgram_peer_wake_me(sk
, other
)) {
1805 goto restart_locked
;
1809 if (unlikely(sk_locked
))
1810 unix_state_unlock(sk
);
1812 if (sock_flag(other
, SOCK_RCVTSTAMP
))
1813 __net_timestamp(skb
);
1814 maybe_add_creds(skb
, sock
, other
);
1815 skb_queue_tail(&other
->sk_receive_queue
, skb
);
1816 if (max_level
> unix_sk(other
)->recursion_level
)
1817 unix_sk(other
)->recursion_level
= max_level
;
1818 unix_state_unlock(other
);
1819 other
->sk_data_ready(other
);
1826 unix_state_unlock(sk
);
1827 unix_state_unlock(other
);
1837 /* We use paged skbs for stream sockets, and limit occupancy to 32768
1838 * bytes, and a minimun of a full page.
1840 #define UNIX_SKB_FRAGS_SZ (PAGE_SIZE << get_order(32768))
1842 static int unix_stream_sendmsg(struct socket
*sock
, struct msghdr
*msg
,
1845 struct sock
*sk
= sock
->sk
;
1846 struct sock
*other
= NULL
;
1848 struct sk_buff
*skb
;
1850 struct scm_cookie scm
;
1851 bool fds_sent
= false;
1856 err
= scm_send(sock
, msg
, &scm
, false);
1861 if (msg
->msg_flags
&MSG_OOB
)
1864 if (msg
->msg_namelen
) {
1865 err
= sk
->sk_state
== TCP_ESTABLISHED
? -EISCONN
: -EOPNOTSUPP
;
1869 other
= unix_peer(sk
);
1874 if (sk
->sk_shutdown
& SEND_SHUTDOWN
)
1877 while (sent
< len
) {
1880 /* Keep two messages in the pipe so it schedules better */
1881 size
= min_t(int, size
, (sk
->sk_sndbuf
>> 1) - 64);
1883 /* allow fallback to order-0 allocations */
1884 size
= min_t(int, size
, SKB_MAX_HEAD(0) + UNIX_SKB_FRAGS_SZ
);
1886 data_len
= max_t(int, 0, size
- SKB_MAX_HEAD(0));
1888 data_len
= min_t(size_t, size
, PAGE_ALIGN(data_len
));
1890 skb
= sock_alloc_send_pskb(sk
, size
- data_len
, data_len
,
1891 msg
->msg_flags
& MSG_DONTWAIT
, &err
,
1892 get_order(UNIX_SKB_FRAGS_SZ
));
1896 /* Only send the fds in the first buffer */
1897 err
= unix_scm_to_skb(&scm
, skb
, !fds_sent
);
1902 max_level
= err
+ 1;
1905 skb_put(skb
, size
- data_len
);
1906 skb
->data_len
= data_len
;
1908 err
= skb_copy_datagram_from_iter(skb
, 0, &msg
->msg_iter
, size
);
1914 unix_state_lock(other
);
1916 if (sock_flag(other
, SOCK_DEAD
) ||
1917 (other
->sk_shutdown
& RCV_SHUTDOWN
))
1920 maybe_add_creds(skb
, sock
, other
);
1921 skb_queue_tail(&other
->sk_receive_queue
, skb
);
1922 if (max_level
> unix_sk(other
)->recursion_level
)
1923 unix_sk(other
)->recursion_level
= max_level
;
1924 unix_state_unlock(other
);
1925 other
->sk_data_ready(other
);
1934 unix_state_unlock(other
);
1937 if (sent
== 0 && !(msg
->msg_flags
&MSG_NOSIGNAL
))
1938 send_sig(SIGPIPE
, current
, 0);
1942 return sent
? : err
;
1945 static ssize_t
unix_stream_sendpage(struct socket
*socket
, struct page
*page
,
1946 int offset
, size_t size
, int flags
)
1949 bool send_sigpipe
= false;
1950 bool init_scm
= true;
1951 struct scm_cookie scm
;
1952 struct sock
*other
, *sk
= socket
->sk
;
1953 struct sk_buff
*skb
, *newskb
= NULL
, *tail
= NULL
;
1955 if (flags
& MSG_OOB
)
1958 other
= unix_peer(sk
);
1959 if (!other
|| sk
->sk_state
!= TCP_ESTABLISHED
)
1964 unix_state_unlock(other
);
1965 mutex_unlock(&unix_sk(other
)->iolock
);
1966 newskb
= sock_alloc_send_pskb(sk
, 0, 0, flags
& MSG_DONTWAIT
,
1972 /* we must acquire iolock as we modify already present
1973 * skbs in the sk_receive_queue and mess with skb->len
1975 err
= mutex_lock_interruptible(&unix_sk(other
)->iolock
);
1977 err
= flags
& MSG_DONTWAIT
? -EAGAIN
: -ERESTARTSYS
;
1981 if (sk
->sk_shutdown
& SEND_SHUTDOWN
) {
1983 send_sigpipe
= true;
1987 unix_state_lock(other
);
1989 if (sock_flag(other
, SOCK_DEAD
) ||
1990 other
->sk_shutdown
& RCV_SHUTDOWN
) {
1992 send_sigpipe
= true;
1993 goto err_state_unlock
;
1997 err
= maybe_init_creds(&scm
, socket
, other
);
1999 goto err_state_unlock
;
2003 skb
= skb_peek_tail(&other
->sk_receive_queue
);
2004 if (tail
&& tail
== skb
) {
2006 } else if (!skb
|| !unix_skb_scm_eq(skb
, &scm
)) {
2013 } else if (newskb
) {
2014 /* this is fast path, we don't necessarily need to
2015 * call to kfree_skb even though with newskb == NULL
2016 * this - does no harm
2018 consume_skb(newskb
);
2022 if (skb_append_pagefrags(skb
, page
, offset
, size
)) {
2028 skb
->data_len
+= size
;
2029 skb
->truesize
+= size
;
2030 atomic_add(size
, &sk
->sk_wmem_alloc
);
2033 err
= unix_scm_to_skb(&scm
, skb
, false);
2035 goto err_state_unlock
;
2036 spin_lock(&other
->sk_receive_queue
.lock
);
2037 __skb_queue_tail(&other
->sk_receive_queue
, newskb
);
2038 spin_unlock(&other
->sk_receive_queue
.lock
);
2041 unix_state_unlock(other
);
2042 mutex_unlock(&unix_sk(other
)->iolock
);
2044 other
->sk_data_ready(other
);
2049 unix_state_unlock(other
);
2051 mutex_unlock(&unix_sk(other
)->iolock
);
2054 if (send_sigpipe
&& !(flags
& MSG_NOSIGNAL
))
2055 send_sig(SIGPIPE
, current
, 0);
2061 static int unix_seqpacket_sendmsg(struct socket
*sock
, struct msghdr
*msg
,
2065 struct sock
*sk
= sock
->sk
;
2067 err
= sock_error(sk
);
2071 if (sk
->sk_state
!= TCP_ESTABLISHED
)
2074 if (msg
->msg_namelen
)
2075 msg
->msg_namelen
= 0;
2077 return unix_dgram_sendmsg(sock
, msg
, len
);
2080 static int unix_seqpacket_recvmsg(struct socket
*sock
, struct msghdr
*msg
,
2081 size_t size
, int flags
)
2083 struct sock
*sk
= sock
->sk
;
2085 if (sk
->sk_state
!= TCP_ESTABLISHED
)
2088 return unix_dgram_recvmsg(sock
, msg
, size
, flags
);
2091 static void unix_copy_addr(struct msghdr
*msg
, struct sock
*sk
)
2093 struct unix_sock
*u
= unix_sk(sk
);
2096 msg
->msg_namelen
= u
->addr
->len
;
2097 memcpy(msg
->msg_name
, u
->addr
->name
, u
->addr
->len
);
2101 static int unix_dgram_recvmsg(struct socket
*sock
, struct msghdr
*msg
,
2102 size_t size
, int flags
)
2104 struct scm_cookie scm
;
2105 struct sock
*sk
= sock
->sk
;
2106 struct unix_sock
*u
= unix_sk(sk
);
2107 struct sk_buff
*skb
, *last
;
2116 timeo
= sock_rcvtimeo(sk
, flags
& MSG_DONTWAIT
);
2119 mutex_lock(&u
->iolock
);
2121 skip
= sk_peek_offset(sk
, flags
);
2122 skb
= __skb_try_recv_datagram(sk
, flags
, NULL
, &peeked
, &skip
,
2127 mutex_unlock(&u
->iolock
);
2132 !__skb_wait_for_more_packets(sk
, &err
, &timeo
, last
));
2134 if (!skb
) { /* implies iolock unlocked */
2135 unix_state_lock(sk
);
2136 /* Signal EOF on disconnected non-blocking SEQPACKET socket. */
2137 if (sk
->sk_type
== SOCK_SEQPACKET
&& err
== -EAGAIN
&&
2138 (sk
->sk_shutdown
& RCV_SHUTDOWN
))
2140 unix_state_unlock(sk
);
2144 if (wq_has_sleeper(&u
->peer_wait
))
2145 wake_up_interruptible_sync_poll(&u
->peer_wait
,
2146 POLLOUT
| POLLWRNORM
|
2150 unix_copy_addr(msg
, skb
->sk
);
2152 if (size
> skb
->len
- skip
)
2153 size
= skb
->len
- skip
;
2154 else if (size
< skb
->len
- skip
)
2155 msg
->msg_flags
|= MSG_TRUNC
;
2157 err
= skb_copy_datagram_msg(skb
, skip
, msg
, size
);
2161 if (sock_flag(sk
, SOCK_RCVTSTAMP
))
2162 __sock_recv_timestamp(msg
, sk
, skb
);
2164 memset(&scm
, 0, sizeof(scm
));
2166 scm_set_cred(&scm
, UNIXCB(skb
).pid
, UNIXCB(skb
).uid
, UNIXCB(skb
).gid
);
2167 unix_set_secdata(&scm
, skb
);
2169 if (!(flags
& MSG_PEEK
)) {
2171 unix_detach_fds(&scm
, skb
);
2173 sk_peek_offset_bwd(sk
, skb
->len
);
2175 /* It is questionable: on PEEK we could:
2176 - do not return fds - good, but too simple 8)
2177 - return fds, and do not return them on read (old strategy,
2179 - clone fds (I chose it for now, it is the most universal
2182 POSIX 1003.1g does not actually define this clearly
2183 at all. POSIX 1003.1g doesn't define a lot of things
2188 sk_peek_offset_fwd(sk
, size
);
2191 scm
.fp
= scm_fp_dup(UNIXCB(skb
).fp
);
2193 err
= (flags
& MSG_TRUNC
) ? skb
->len
- skip
: size
;
2195 scm_recv(sock
, msg
, &scm
, flags
);
2198 skb_free_datagram(sk
, skb
);
2199 mutex_unlock(&u
->iolock
);
2205 * Sleep until more data has arrived. But check for races..
2207 static long unix_stream_data_wait(struct sock
*sk
, long timeo
,
2208 struct sk_buff
*last
, unsigned int last_len
,
2211 struct sk_buff
*tail
;
2214 unix_state_lock(sk
);
2217 prepare_to_wait(sk_sleep(sk
), &wait
, TASK_INTERRUPTIBLE
);
2219 tail
= skb_peek_tail(&sk
->sk_receive_queue
);
2221 (tail
&& tail
->len
!= last_len
) ||
2223 (sk
->sk_shutdown
& RCV_SHUTDOWN
) ||
2224 signal_pending(current
) ||
2228 sk_set_bit(SOCKWQ_ASYNC_WAITDATA
, sk
);
2229 unix_state_unlock(sk
);
2231 timeo
= freezable_schedule_timeout(timeo
);
2233 timeo
= schedule_timeout(timeo
);
2234 unix_state_lock(sk
);
2236 if (sock_flag(sk
, SOCK_DEAD
))
2239 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA
, sk
);
2242 finish_wait(sk_sleep(sk
), &wait
);
2243 unix_state_unlock(sk
);
2247 static unsigned int unix_skb_len(const struct sk_buff
*skb
)
2249 return skb
->len
- UNIXCB(skb
).consumed
;
2252 struct unix_stream_read_state
{
2253 int (*recv_actor
)(struct sk_buff
*, int, int,
2254 struct unix_stream_read_state
*);
2255 struct socket
*socket
;
2257 struct pipe_inode_info
*pipe
;
2260 unsigned int splice_flags
;
2263 static int unix_stream_read_generic(struct unix_stream_read_state
*state
,
2266 struct scm_cookie scm
;
2267 struct socket
*sock
= state
->socket
;
2268 struct sock
*sk
= sock
->sk
;
2269 struct unix_sock
*u
= unix_sk(sk
);
2271 int flags
= state
->flags
;
2272 int noblock
= flags
& MSG_DONTWAIT
;
2273 bool check_creds
= false;
2278 size_t size
= state
->size
;
2279 unsigned int last_len
;
2281 if (unlikely(sk
->sk_state
!= TCP_ESTABLISHED
)) {
2286 if (unlikely(flags
& MSG_OOB
)) {
2291 target
= sock_rcvlowat(sk
, flags
& MSG_WAITALL
, size
);
2292 timeo
= sock_rcvtimeo(sk
, noblock
);
2294 memset(&scm
, 0, sizeof(scm
));
2296 /* Lock the socket to prevent queue disordering
2297 * while sleeps in memcpy_tomsg
2299 mutex_lock(&u
->iolock
);
2301 if (flags
& MSG_PEEK
)
2302 skip
= sk_peek_offset(sk
, flags
);
2309 struct sk_buff
*skb
, *last
;
2312 unix_state_lock(sk
);
2313 if (sock_flag(sk
, SOCK_DEAD
)) {
2317 last
= skb
= skb_peek(&sk
->sk_receive_queue
);
2318 last_len
= last
? last
->len
: 0;
2321 unix_sk(sk
)->recursion_level
= 0;
2322 if (copied
>= target
)
2326 * POSIX 1003.1g mandates this order.
2329 err
= sock_error(sk
);
2332 if (sk
->sk_shutdown
& RCV_SHUTDOWN
)
2335 unix_state_unlock(sk
);
2341 mutex_unlock(&u
->iolock
);
2343 timeo
= unix_stream_data_wait(sk
, timeo
, last
,
2344 last_len
, freezable
);
2346 if (signal_pending(current
)) {
2347 err
= sock_intr_errno(timeo
);
2352 mutex_lock(&u
->iolock
);
2355 unix_state_unlock(sk
);
2359 while (skip
>= unix_skb_len(skb
)) {
2360 skip
-= unix_skb_len(skb
);
2362 last_len
= skb
->len
;
2363 skb
= skb_peek_next(skb
, &sk
->sk_receive_queue
);
2368 unix_state_unlock(sk
);
2371 /* Never glue messages from different writers */
2372 if (!unix_skb_scm_eq(skb
, &scm
))
2374 } else if (test_bit(SOCK_PASSCRED
, &sock
->flags
)) {
2375 /* Copy credentials */
2376 scm_set_cred(&scm
, UNIXCB(skb
).pid
, UNIXCB(skb
).uid
, UNIXCB(skb
).gid
);
2377 unix_set_secdata(&scm
, skb
);
2381 /* Copy address just once */
2382 if (state
->msg
&& state
->msg
->msg_name
) {
2383 DECLARE_SOCKADDR(struct sockaddr_un
*, sunaddr
,
2384 state
->msg
->msg_name
);
2385 unix_copy_addr(state
->msg
, skb
->sk
);
2389 chunk
= min_t(unsigned int, unix_skb_len(skb
) - skip
, size
);
2391 chunk
= state
->recv_actor(skb
, skip
, chunk
, state
);
2392 drop_skb
= !unix_skb_len(skb
);
2393 /* skb is only safe to use if !drop_skb */
2404 /* the skb was touched by a concurrent reader;
2405 * we should not expect anything from this skb
2406 * anymore and assume it invalid - we can be
2407 * sure it was dropped from the socket queue
2409 * let's report a short read
2415 /* Mark read part of skb as used */
2416 if (!(flags
& MSG_PEEK
)) {
2417 UNIXCB(skb
).consumed
+= chunk
;
2419 sk_peek_offset_bwd(sk
, chunk
);
2422 unix_detach_fds(&scm
, skb
);
2424 if (unix_skb_len(skb
))
2427 skb_unlink(skb
, &sk
->sk_receive_queue
);
2433 /* It is questionable, see note in unix_dgram_recvmsg.
2436 scm
.fp
= scm_fp_dup(UNIXCB(skb
).fp
);
2438 sk_peek_offset_fwd(sk
, chunk
);
2445 last_len
= skb
->len
;
2446 unix_state_lock(sk
);
2447 skb
= skb_peek_next(skb
, &sk
->sk_receive_queue
);
2450 unix_state_unlock(sk
);
2455 mutex_unlock(&u
->iolock
);
2457 scm_recv(sock
, state
->msg
, &scm
, flags
);
2461 return copied
? : err
;
2464 static int unix_stream_read_actor(struct sk_buff
*skb
,
2465 int skip
, int chunk
,
2466 struct unix_stream_read_state
*state
)
2470 ret
= skb_copy_datagram_msg(skb
, UNIXCB(skb
).consumed
+ skip
,
2472 return ret
?: chunk
;
2475 static int unix_stream_recvmsg(struct socket
*sock
, struct msghdr
*msg
,
2476 size_t size
, int flags
)
2478 struct unix_stream_read_state state
= {
2479 .recv_actor
= unix_stream_read_actor
,
2486 return unix_stream_read_generic(&state
, true);
2489 static int unix_stream_splice_actor(struct sk_buff
*skb
,
2490 int skip
, int chunk
,
2491 struct unix_stream_read_state
*state
)
2493 return skb_splice_bits(skb
, state
->socket
->sk
,
2494 UNIXCB(skb
).consumed
+ skip
,
2495 state
->pipe
, chunk
, state
->splice_flags
);
2498 static ssize_t
unix_stream_splice_read(struct socket
*sock
, loff_t
*ppos
,
2499 struct pipe_inode_info
*pipe
,
2500 size_t size
, unsigned int flags
)
2502 struct unix_stream_read_state state
= {
2503 .recv_actor
= unix_stream_splice_actor
,
2507 .splice_flags
= flags
,
2510 if (unlikely(*ppos
))
2513 if (sock
->file
->f_flags
& O_NONBLOCK
||
2514 flags
& SPLICE_F_NONBLOCK
)
2515 state
.flags
= MSG_DONTWAIT
;
2517 return unix_stream_read_generic(&state
, false);
2520 static int unix_shutdown(struct socket
*sock
, int mode
)
2522 struct sock
*sk
= sock
->sk
;
2525 if (mode
< SHUT_RD
|| mode
> SHUT_RDWR
)
2528 * SHUT_RD (0) -> RCV_SHUTDOWN (1)
2529 * SHUT_WR (1) -> SEND_SHUTDOWN (2)
2530 * SHUT_RDWR (2) -> SHUTDOWN_MASK (3)
2534 unix_state_lock(sk
);
2535 sk
->sk_shutdown
|= mode
;
2536 other
= unix_peer(sk
);
2539 unix_state_unlock(sk
);
2540 sk
->sk_state_change(sk
);
2543 (sk
->sk_type
== SOCK_STREAM
|| sk
->sk_type
== SOCK_SEQPACKET
)) {
2547 if (mode
&RCV_SHUTDOWN
)
2548 peer_mode
|= SEND_SHUTDOWN
;
2549 if (mode
&SEND_SHUTDOWN
)
2550 peer_mode
|= RCV_SHUTDOWN
;
2551 unix_state_lock(other
);
2552 other
->sk_shutdown
|= peer_mode
;
2553 unix_state_unlock(other
);
2554 other
->sk_state_change(other
);
2555 if (peer_mode
== SHUTDOWN_MASK
)
2556 sk_wake_async(other
, SOCK_WAKE_WAITD
, POLL_HUP
);
2557 else if (peer_mode
& RCV_SHUTDOWN
)
2558 sk_wake_async(other
, SOCK_WAKE_WAITD
, POLL_IN
);
2566 long unix_inq_len(struct sock
*sk
)
2568 struct sk_buff
*skb
;
2571 if (sk
->sk_state
== TCP_LISTEN
)
2574 spin_lock(&sk
->sk_receive_queue
.lock
);
2575 if (sk
->sk_type
== SOCK_STREAM
||
2576 sk
->sk_type
== SOCK_SEQPACKET
) {
2577 skb_queue_walk(&sk
->sk_receive_queue
, skb
)
2578 amount
+= unix_skb_len(skb
);
2580 skb
= skb_peek(&sk
->sk_receive_queue
);
2584 spin_unlock(&sk
->sk_receive_queue
.lock
);
2588 EXPORT_SYMBOL_GPL(unix_inq_len
);
2590 long unix_outq_len(struct sock
*sk
)
2592 return sk_wmem_alloc_get(sk
);
2594 EXPORT_SYMBOL_GPL(unix_outq_len
);
2596 static int unix_open_file(struct sock
*sk
)
2602 if (!ns_capable(sock_net(sk
)->user_ns
, CAP_NET_ADMIN
))
2605 unix_state_lock(sk
);
2606 path
= unix_sk(sk
)->path
;
2608 unix_state_unlock(sk
);
2613 unix_state_unlock(sk
);
2615 fd
= get_unused_fd_flags(O_CLOEXEC
);
2619 f
= dentry_open(&path
, O_PATH
, current_cred());
2633 static int unix_ioctl(struct socket
*sock
, unsigned int cmd
, unsigned long arg
)
2635 struct sock
*sk
= sock
->sk
;
2641 amount
= unix_outq_len(sk
);
2642 err
= put_user(amount
, (int __user
*)arg
);
2645 amount
= unix_inq_len(sk
);
2649 err
= put_user(amount
, (int __user
*)arg
);
2652 err
= unix_open_file(sk
);
2661 static unsigned int unix_poll(struct file
*file
, struct socket
*sock
, poll_table
*wait
)
2663 struct sock
*sk
= sock
->sk
;
2666 sock_poll_wait(file
, sk_sleep(sk
), wait
);
2669 /* exceptional events? */
2672 if (sk
->sk_shutdown
== SHUTDOWN_MASK
)
2674 if (sk
->sk_shutdown
& RCV_SHUTDOWN
)
2675 mask
|= POLLRDHUP
| POLLIN
| POLLRDNORM
;
2678 if (!skb_queue_empty(&sk
->sk_receive_queue
))
2679 mask
|= POLLIN
| POLLRDNORM
;
2681 /* Connection-based need to check for termination and startup */
2682 if ((sk
->sk_type
== SOCK_STREAM
|| sk
->sk_type
== SOCK_SEQPACKET
) &&
2683 sk
->sk_state
== TCP_CLOSE
)
2687 * we set writable also when the other side has shut down the
2688 * connection. This prevents stuck sockets.
2690 if (unix_writable(sk
))
2691 mask
|= POLLOUT
| POLLWRNORM
| POLLWRBAND
;
2696 static unsigned int unix_dgram_poll(struct file
*file
, struct socket
*sock
,
2699 struct sock
*sk
= sock
->sk
, *other
;
2700 unsigned int mask
, writable
;
2702 sock_poll_wait(file
, sk_sleep(sk
), wait
);
2705 /* exceptional events? */
2706 if (sk
->sk_err
|| !skb_queue_empty(&sk
->sk_error_queue
))
2708 (sock_flag(sk
, SOCK_SELECT_ERR_QUEUE
) ? POLLPRI
: 0);
2710 if (sk
->sk_shutdown
& RCV_SHUTDOWN
)
2711 mask
|= POLLRDHUP
| POLLIN
| POLLRDNORM
;
2712 if (sk
->sk_shutdown
== SHUTDOWN_MASK
)
2716 if (!skb_queue_empty(&sk
->sk_receive_queue
))
2717 mask
|= POLLIN
| POLLRDNORM
;
2719 /* Connection-based need to check for termination and startup */
2720 if (sk
->sk_type
== SOCK_SEQPACKET
) {
2721 if (sk
->sk_state
== TCP_CLOSE
)
2723 /* connection hasn't started yet? */
2724 if (sk
->sk_state
== TCP_SYN_SENT
)
2728 /* No write status requested, avoid expensive OUT tests. */
2729 if (!(poll_requested_events(wait
) & (POLLWRBAND
|POLLWRNORM
|POLLOUT
)))
2732 writable
= unix_writable(sk
);
2734 unix_state_lock(sk
);
2736 other
= unix_peer(sk
);
2737 if (other
&& unix_peer(other
) != sk
&&
2738 unix_recvq_full(other
) &&
2739 unix_dgram_peer_wake_me(sk
, other
))
2742 unix_state_unlock(sk
);
2746 mask
|= POLLOUT
| POLLWRNORM
| POLLWRBAND
;
2748 sk_set_bit(SOCKWQ_ASYNC_NOSPACE
, sk
);
2753 #ifdef CONFIG_PROC_FS
2755 #define BUCKET_SPACE (BITS_PER_LONG - (UNIX_HASH_BITS + 1) - 1)
2757 #define get_bucket(x) ((x) >> BUCKET_SPACE)
2758 #define get_offset(x) ((x) & ((1L << BUCKET_SPACE) - 1))
2759 #define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
2761 static struct sock
*unix_from_bucket(struct seq_file
*seq
, loff_t
*pos
)
2763 unsigned long offset
= get_offset(*pos
);
2764 unsigned long bucket
= get_bucket(*pos
);
2766 unsigned long count
= 0;
2768 for (sk
= sk_head(&unix_socket_table
[bucket
]); sk
; sk
= sk_next(sk
)) {
2769 if (sock_net(sk
) != seq_file_net(seq
))
2771 if (++count
== offset
)
2778 static struct sock
*unix_next_socket(struct seq_file
*seq
,
2782 unsigned long bucket
;
2784 while (sk
> (struct sock
*)SEQ_START_TOKEN
) {
2788 if (sock_net(sk
) == seq_file_net(seq
))
2793 sk
= unix_from_bucket(seq
, pos
);
2798 bucket
= get_bucket(*pos
) + 1;
2799 *pos
= set_bucket_offset(bucket
, 1);
2800 } while (bucket
< ARRAY_SIZE(unix_socket_table
));
2805 static void *unix_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2806 __acquires(unix_table_lock
)
2808 spin_lock(&unix_table_lock
);
2811 return SEQ_START_TOKEN
;
2813 if (get_bucket(*pos
) >= ARRAY_SIZE(unix_socket_table
))
2816 return unix_next_socket(seq
, NULL
, pos
);
2819 static void *unix_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2822 return unix_next_socket(seq
, v
, pos
);
2825 static void unix_seq_stop(struct seq_file
*seq
, void *v
)
2826 __releases(unix_table_lock
)
2828 spin_unlock(&unix_table_lock
);
2831 static int unix_seq_show(struct seq_file
*seq
, void *v
)
2834 if (v
== SEQ_START_TOKEN
)
2835 seq_puts(seq
, "Num RefCount Protocol Flags Type St "
2839 struct unix_sock
*u
= unix_sk(s
);
2842 seq_printf(seq
, "%pK: %08X %08X %08X %04X %02X %5lu",
2844 atomic_read(&s
->sk_refcnt
),
2846 s
->sk_state
== TCP_LISTEN
? __SO_ACCEPTCON
: 0,
2849 (s
->sk_state
== TCP_ESTABLISHED
? SS_CONNECTED
: SS_UNCONNECTED
) :
2850 (s
->sk_state
== TCP_ESTABLISHED
? SS_CONNECTING
: SS_DISCONNECTING
),
2858 len
= u
->addr
->len
- sizeof(short);
2859 if (!UNIX_ABSTRACT(s
))
2865 for ( ; i
< len
; i
++)
2866 seq_putc(seq
, u
->addr
->name
->sun_path
[i
] ?:
2869 unix_state_unlock(s
);
2870 seq_putc(seq
, '\n');
2876 static const struct seq_operations unix_seq_ops
= {
2877 .start
= unix_seq_start
,
2878 .next
= unix_seq_next
,
2879 .stop
= unix_seq_stop
,
2880 .show
= unix_seq_show
,
2883 static int unix_seq_open(struct inode
*inode
, struct file
*file
)
2885 return seq_open_net(inode
, file
, &unix_seq_ops
,
2886 sizeof(struct seq_net_private
));
2889 static const struct file_operations unix_seq_fops
= {
2890 .owner
= THIS_MODULE
,
2891 .open
= unix_seq_open
,
2893 .llseek
= seq_lseek
,
2894 .release
= seq_release_net
,
2899 static const struct net_proto_family unix_family_ops
= {
2901 .create
= unix_create
,
2902 .owner
= THIS_MODULE
,
2906 static int __net_init
unix_net_init(struct net
*net
)
2908 int error
= -ENOMEM
;
2910 net
->unx
.sysctl_max_dgram_qlen
= 10;
2911 if (unix_sysctl_register(net
))
2914 #ifdef CONFIG_PROC_FS
2915 if (!proc_create("unix", 0, net
->proc_net
, &unix_seq_fops
)) {
2916 unix_sysctl_unregister(net
);
2925 static void __net_exit
unix_net_exit(struct net
*net
)
2927 unix_sysctl_unregister(net
);
2928 remove_proc_entry("unix", net
->proc_net
);
2931 static struct pernet_operations unix_net_ops
= {
2932 .init
= unix_net_init
,
2933 .exit
= unix_net_exit
,
2936 static int __init
af_unix_init(void)
2940 BUILD_BUG_ON(sizeof(struct unix_skb_parms
) > FIELD_SIZEOF(struct sk_buff
, cb
));
2942 rc
= proto_register(&unix_proto
, 1);
2944 pr_crit("%s: Cannot create unix_sock SLAB cache!\n", __func__
);
2948 sock_register(&unix_family_ops
);
2949 register_pernet_subsys(&unix_net_ops
);
2954 static void __exit
af_unix_exit(void)
2956 sock_unregister(PF_UNIX
);
2957 proto_unregister(&unix_proto
);
2958 unregister_pernet_subsys(&unix_net_ops
);
2961 /* Earlier than device_initcall() so that other drivers invoking
2962 request_module() don't end up in a loop when modprobe tries
2963 to use a UNIX socket. But later than subsys_initcall() because
2964 we depend on stuff initialised there */
2965 fs_initcall(af_unix_init
);
2966 module_exit(af_unix_exit
);
2968 MODULE_LICENSE("GPL");
2969 MODULE_ALIAS_NETPROTO(PF_UNIX
);