2 * NET4: Implementation of BSD Unix domain sockets.
4 * Authors: Alan Cox, <alan@lxorguk.ukuu.org.uk>
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License
8 * as published by the Free Software Foundation; either version
9 * 2 of the License, or (at your option) any later version.
12 * Linus Torvalds : Assorted bug cures.
13 * Niibe Yutaka : async I/O support.
14 * Carsten Paeth : PF_UNIX check, address fixes.
15 * Alan Cox : Limit size of allocated blocks.
16 * Alan Cox : Fixed the stupid socketpair bug.
17 * Alan Cox : BSD compatibility fine tuning.
18 * Alan Cox : Fixed a bug in connect when interrupted.
19 * Alan Cox : Sorted out a proper draft version of
20 * file descriptor passing hacked up from
22 * Marty Leisner : Fixes to fd passing
23 * Nick Nevin : recvmsg bugfix.
24 * Alan Cox : Started proper garbage collector
25 * Heiko EiBfeldt : Missing verify_area check
26 * Alan Cox : Started POSIXisms
27 * Andreas Schwab : Replace inode by dentry for proper
29 * Kirk Petersen : Made this a module
30 * Christoph Rohland : Elegant non-blocking accept/connect algorithm.
32 * Alexey Kuznetosv : Repaired (I hope) bugs introduces
33 * by above two patches.
34 * Andrea Arcangeli : If possible we block in connect(2)
35 * if the max backlog of the listen socket
36 * is been reached. This won't break
37 * old apps and it will avoid huge amount
38 * of socks hashed (this for unix_gc()
39 * performances reasons).
40 * Security fix that limits the max
41 * number of socks to 2*max_files and
42 * the number of skb queueable in the
44 * Artur Skawina : Hash function optimizations
45 * Alexey Kuznetsov : Full scale SMP. Lot of bugs are introduced 8)
46 * Malcolm Beattie : Set peercred for socketpair
47 * Michal Ostrowski : Module initialization cleanup.
48 * Arnaldo C. Melo : Remove MOD_{INC,DEC}_USE_COUNT,
49 * the core infrastructure is doing that
50 * for all net proto families now (2.5.69+)
53 * Known differences from reference BSD that was tested:
56 * ECONNREFUSED is not returned from one end of a connected() socket to the
57 * other the moment one end closes.
58 * fstat() doesn't return st_dev=0, and give the blksize as high water mark
59 * and a fake inode identifier (nor the BSD first socket fstat twice bug).
61 * accept() returns a path name even if the connecting socket has closed
62 * in the meantime (BSD loses the path and gives up).
63 * accept() returns 0 length path for an unbound connector. BSD returns 16
64 * and a null first byte in the path (but not for gethost/peername - BSD bug ??)
65 * socketpair(...SOCK_RAW..) doesn't panic the kernel.
66 * BSD af_unix apparently has connect forgetting to block properly.
67 * (need to check this with the POSIX spec in detail)
69 * Differences from 2.0.0-11-... (ANK)
70 * Bug fixes and improvements.
71 * - client shutdown killed server socket.
72 * - removed all useless cli/sti pairs.
74 * Semantic changes/extensions.
75 * - generic control message passing.
76 * - SCM_CREDENTIALS control message.
77 * - "Abstract" (not FS based) socket bindings.
78 * Abstract names are sequences of bytes (not zero terminated)
79 * started by 0, so that this name space does not intersect
83 #include <linux/module.h>
84 #include <linux/kernel.h>
85 #include <linux/signal.h>
86 #include <linux/sched.h>
87 #include <linux/errno.h>
88 #include <linux/string.h>
89 #include <linux/stat.h>
90 #include <linux/dcache.h>
91 #include <linux/namei.h>
92 #include <linux/socket.h>
94 #include <linux/fcntl.h>
95 #include <linux/termios.h>
96 #include <linux/sockios.h>
97 #include <linux/net.h>
100 #include <linux/slab.h>
101 #include <asm/uaccess.h>
102 #include <linux/skbuff.h>
103 #include <linux/netdevice.h>
104 #include <net/net_namespace.h>
105 #include <net/sock.h>
106 #include <net/tcp_states.h>
107 #include <net/af_unix.h>
108 #include <linux/proc_fs.h>
109 #include <linux/seq_file.h>
111 #include <linux/init.h>
112 #include <linux/poll.h>
113 #include <linux/rtnetlink.h>
114 #include <linux/mount.h>
115 #include <net/checksum.h>
116 #include <linux/security.h>
118 struct hlist_head unix_socket_table
[UNIX_HASH_SIZE
+ 1];
119 EXPORT_SYMBOL_GPL(unix_socket_table
);
120 DEFINE_SPINLOCK(unix_table_lock
);
121 EXPORT_SYMBOL_GPL(unix_table_lock
);
122 static atomic_long_t unix_nr_socks
;
124 #define unix_sockets_unbound (&unix_socket_table[UNIX_HASH_SIZE])
126 #define UNIX_ABSTRACT(sk) (unix_sk(sk)->addr->hash != UNIX_HASH_SIZE)
128 #ifdef CONFIG_SECURITY_NETWORK
129 static void unix_get_secdata(struct scm_cookie
*scm
, struct sk_buff
*skb
)
131 memcpy(UNIXSID(skb
), &scm
->secid
, sizeof(u32
));
134 static inline void unix_set_secdata(struct scm_cookie
*scm
, struct sk_buff
*skb
)
136 scm
->secid
= *UNIXSID(skb
);
139 static inline void unix_get_secdata(struct scm_cookie
*scm
, struct sk_buff
*skb
)
142 static inline void unix_set_secdata(struct scm_cookie
*scm
, struct sk_buff
*skb
)
144 #endif /* CONFIG_SECURITY_NETWORK */
147 * SMP locking strategy:
148 * hash table is protected with spinlock unix_table_lock
149 * each socket state is protected by separate spin lock.
152 static inline unsigned unix_hash_fold(__wsum n
)
154 unsigned hash
= (__force
unsigned)n
;
157 return hash
&(UNIX_HASH_SIZE
-1);
160 #define unix_peer(sk) (unix_sk(sk)->peer)
162 static inline int unix_our_peer(struct sock
*sk
, struct sock
*osk
)
164 return unix_peer(osk
) == sk
;
167 static inline int unix_may_send(struct sock
*sk
, struct sock
*osk
)
169 return unix_peer(osk
) == NULL
|| unix_our_peer(sk
, osk
);
172 static inline int unix_recvq_full(struct sock
const *sk
)
174 return skb_queue_len(&sk
->sk_receive_queue
) > sk
->sk_max_ack_backlog
;
177 struct sock
*unix_peer_get(struct sock
*s
)
185 unix_state_unlock(s
);
188 EXPORT_SYMBOL_GPL(unix_peer_get
);
190 static inline void unix_release_addr(struct unix_address
*addr
)
192 if (atomic_dec_and_test(&addr
->refcnt
))
197 * Check unix socket name:
198 * - should be not zero length.
199 * - if started by not zero, should be NULL terminated (FS object)
200 * - if started by zero, it is abstract name.
203 static int unix_mkname(struct sockaddr_un
*sunaddr
, int len
, unsigned *hashp
)
205 if (len
<= sizeof(short) || len
> sizeof(*sunaddr
))
207 if (!sunaddr
|| sunaddr
->sun_family
!= AF_UNIX
)
209 if (sunaddr
->sun_path
[0]) {
211 * This may look like an off by one error but it is a bit more
212 * subtle. 108 is the longest valid AF_UNIX path for a binding.
213 * sun_path[108] doesn't as such exist. However in kernel space
214 * we are guaranteed that it is a valid memory location in our
215 * kernel address buffer.
217 ((char *)sunaddr
)[len
] = 0;
218 len
= strlen(sunaddr
->sun_path
)+1+sizeof(short);
222 *hashp
= unix_hash_fold(csum_partial(sunaddr
, len
, 0));
226 static void __unix_remove_socket(struct sock
*sk
)
228 sk_del_node_init(sk
);
231 static void __unix_insert_socket(struct hlist_head
*list
, struct sock
*sk
)
233 WARN_ON(!sk_unhashed(sk
));
234 sk_add_node(sk
, list
);
237 static inline void unix_remove_socket(struct sock
*sk
)
239 spin_lock(&unix_table_lock
);
240 __unix_remove_socket(sk
);
241 spin_unlock(&unix_table_lock
);
244 static inline void unix_insert_socket(struct hlist_head
*list
, struct sock
*sk
)
246 spin_lock(&unix_table_lock
);
247 __unix_insert_socket(list
, sk
);
248 spin_unlock(&unix_table_lock
);
251 static struct sock
*__unix_find_socket_byname(struct net
*net
,
252 struct sockaddr_un
*sunname
,
253 int len
, int type
, unsigned hash
)
256 struct hlist_node
*node
;
258 sk_for_each(s
, node
, &unix_socket_table
[hash
^ type
]) {
259 struct unix_sock
*u
= unix_sk(s
);
261 if (!net_eq(sock_net(s
), net
))
264 if (u
->addr
->len
== len
&&
265 !memcmp(u
->addr
->name
, sunname
, len
))
273 static inline struct sock
*unix_find_socket_byname(struct net
*net
,
274 struct sockaddr_un
*sunname
,
280 spin_lock(&unix_table_lock
);
281 s
= __unix_find_socket_byname(net
, sunname
, len
, type
, hash
);
284 spin_unlock(&unix_table_lock
);
288 static struct sock
*unix_find_socket_byinode(struct inode
*i
)
291 struct hlist_node
*node
;
293 spin_lock(&unix_table_lock
);
295 &unix_socket_table
[i
->i_ino
& (UNIX_HASH_SIZE
- 1)]) {
296 struct dentry
*dentry
= unix_sk(s
)->dentry
;
298 if (dentry
&& dentry
->d_inode
== i
) {
305 spin_unlock(&unix_table_lock
);
309 static inline int unix_writable(struct sock
*sk
)
311 return (atomic_read(&sk
->sk_wmem_alloc
) << 2) <= sk
->sk_sndbuf
;
314 static void unix_write_space(struct sock
*sk
)
316 struct socket_wq
*wq
;
319 if (unix_writable(sk
)) {
320 wq
= rcu_dereference(sk
->sk_wq
);
321 if (wq_has_sleeper(wq
))
322 wake_up_interruptible_sync_poll(&wq
->wait
,
323 POLLOUT
| POLLWRNORM
| POLLWRBAND
);
324 sk_wake_async(sk
, SOCK_WAKE_SPACE
, POLL_OUT
);
329 /* When dgram socket disconnects (or changes its peer), we clear its receive
330 * queue of packets arrived from previous peer. First, it allows to do
331 * flow control based only on wmem_alloc; second, sk connected to peer
332 * may receive messages only from that peer. */
333 static void unix_dgram_disconnected(struct sock
*sk
, struct sock
*other
)
335 if (!skb_queue_empty(&sk
->sk_receive_queue
)) {
336 skb_queue_purge(&sk
->sk_receive_queue
);
337 wake_up_interruptible_all(&unix_sk(sk
)->peer_wait
);
339 /* If one link of bidirectional dgram pipe is disconnected,
340 * we signal error. Messages are lost. Do not make this,
341 * when peer was not connected to us.
343 if (!sock_flag(other
, SOCK_DEAD
) && unix_peer(other
) == sk
) {
344 other
->sk_err
= ECONNRESET
;
345 other
->sk_error_report(other
);
350 static void unix_sock_destructor(struct sock
*sk
)
352 struct unix_sock
*u
= unix_sk(sk
);
354 skb_queue_purge(&sk
->sk_receive_queue
);
356 WARN_ON(atomic_read(&sk
->sk_wmem_alloc
));
357 WARN_ON(!sk_unhashed(sk
));
358 WARN_ON(sk
->sk_socket
);
359 if (!sock_flag(sk
, SOCK_DEAD
)) {
360 printk(KERN_INFO
"Attempt to release alive unix socket: %p\n", sk
);
365 unix_release_addr(u
->addr
);
367 atomic_long_dec(&unix_nr_socks
);
369 sock_prot_inuse_add(sock_net(sk
), sk
->sk_prot
, -1);
371 #ifdef UNIX_REFCNT_DEBUG
372 printk(KERN_DEBUG
"UNIX %p is destroyed, %ld are still alive.\n", sk
,
373 atomic_long_read(&unix_nr_socks
));
377 static int unix_release_sock(struct sock
*sk
, int embrion
)
379 struct unix_sock
*u
= unix_sk(sk
);
380 struct dentry
*dentry
;
381 struct vfsmount
*mnt
;
386 unix_remove_socket(sk
);
391 sk
->sk_shutdown
= SHUTDOWN_MASK
;
396 state
= sk
->sk_state
;
397 sk
->sk_state
= TCP_CLOSE
;
398 unix_state_unlock(sk
);
400 wake_up_interruptible_all(&u
->peer_wait
);
402 skpair
= unix_peer(sk
);
404 if (skpair
!= NULL
) {
405 if (sk
->sk_type
== SOCK_STREAM
|| sk
->sk_type
== SOCK_SEQPACKET
) {
406 unix_state_lock(skpair
);
408 skpair
->sk_shutdown
= SHUTDOWN_MASK
;
409 if (!skb_queue_empty(&sk
->sk_receive_queue
) || embrion
)
410 skpair
->sk_err
= ECONNRESET
;
411 unix_state_unlock(skpair
);
412 skpair
->sk_state_change(skpair
);
413 sk_wake_async(skpair
, SOCK_WAKE_WAITD
, POLL_HUP
);
415 sock_put(skpair
); /* It may now die */
416 unix_peer(sk
) = NULL
;
419 /* Try to flush out this socket. Throw out buffers at least */
421 while ((skb
= skb_dequeue(&sk
->sk_receive_queue
)) != NULL
) {
422 if (state
== TCP_LISTEN
)
423 unix_release_sock(skb
->sk
, 1);
424 /* passed fds are erased in the kfree_skb hook */
435 /* ---- Socket is dead now and most probably destroyed ---- */
438 * Fixme: BSD difference: In BSD all sockets connected to use get
439 * ECONNRESET and we die on the spot. In Linux we behave
440 * like files and pipes do and wait for the last
443 * Can't we simply set sock->err?
445 * What the above comment does talk about? --ANK(980817)
448 if (unix_tot_inflight
)
449 unix_gc(); /* Garbage collect fds */
454 static void init_peercred(struct sock
*sk
)
456 put_pid(sk
->sk_peer_pid
);
457 if (sk
->sk_peer_cred
)
458 put_cred(sk
->sk_peer_cred
);
459 sk
->sk_peer_pid
= get_pid(task_tgid(current
));
460 sk
->sk_peer_cred
= get_current_cred();
463 static void copy_peercred(struct sock
*sk
, struct sock
*peersk
)
465 put_pid(sk
->sk_peer_pid
);
466 if (sk
->sk_peer_cred
)
467 put_cred(sk
->sk_peer_cred
);
468 sk
->sk_peer_pid
= get_pid(peersk
->sk_peer_pid
);
469 sk
->sk_peer_cred
= get_cred(peersk
->sk_peer_cred
);
472 static int unix_listen(struct socket
*sock
, int backlog
)
475 struct sock
*sk
= sock
->sk
;
476 struct unix_sock
*u
= unix_sk(sk
);
477 struct pid
*old_pid
= NULL
;
478 const struct cred
*old_cred
= NULL
;
481 if (sock
->type
!= SOCK_STREAM
&& sock
->type
!= SOCK_SEQPACKET
)
482 goto out
; /* Only stream/seqpacket sockets accept */
485 goto out
; /* No listens on an unbound socket */
487 if (sk
->sk_state
!= TCP_CLOSE
&& sk
->sk_state
!= TCP_LISTEN
)
489 if (backlog
> sk
->sk_max_ack_backlog
)
490 wake_up_interruptible_all(&u
->peer_wait
);
491 sk
->sk_max_ack_backlog
= backlog
;
492 sk
->sk_state
= TCP_LISTEN
;
493 /* set credentials so connect can copy them */
498 unix_state_unlock(sk
);
506 static int unix_release(struct socket
*);
507 static int unix_bind(struct socket
*, struct sockaddr
*, int);
508 static int unix_stream_connect(struct socket
*, struct sockaddr
*,
509 int addr_len
, int flags
);
510 static int unix_socketpair(struct socket
*, struct socket
*);
511 static int unix_accept(struct socket
*, struct socket
*, int);
512 static int unix_getname(struct socket
*, struct sockaddr
*, int *, int);
513 static unsigned int unix_poll(struct file
*, struct socket
*, poll_table
*);
514 static unsigned int unix_dgram_poll(struct file
*, struct socket
*,
516 static int unix_ioctl(struct socket
*, unsigned int, unsigned long);
517 static int unix_shutdown(struct socket
*, int);
518 static int unix_stream_sendmsg(struct kiocb
*, struct socket
*,
519 struct msghdr
*, size_t);
520 static int unix_stream_recvmsg(struct kiocb
*, struct socket
*,
521 struct msghdr
*, size_t, int);
522 static int unix_dgram_sendmsg(struct kiocb
*, struct socket
*,
523 struct msghdr
*, size_t);
524 static int unix_dgram_recvmsg(struct kiocb
*, struct socket
*,
525 struct msghdr
*, size_t, int);
526 static int unix_dgram_connect(struct socket
*, struct sockaddr
*,
528 static int unix_seqpacket_sendmsg(struct kiocb
*, struct socket
*,
529 struct msghdr
*, size_t);
530 static int unix_seqpacket_recvmsg(struct kiocb
*, struct socket
*,
531 struct msghdr
*, size_t, int);
533 static void unix_set_peek_off(struct sock
*sk
, int val
)
535 struct unix_sock
*u
= unix_sk(sk
);
537 mutex_lock(&u
->readlock
);
538 sk
->sk_peek_off
= val
;
539 mutex_unlock(&u
->readlock
);
543 static const struct proto_ops unix_stream_ops
= {
545 .owner
= THIS_MODULE
,
546 .release
= unix_release
,
548 .connect
= unix_stream_connect
,
549 .socketpair
= unix_socketpair
,
550 .accept
= unix_accept
,
551 .getname
= unix_getname
,
554 .listen
= unix_listen
,
555 .shutdown
= unix_shutdown
,
556 .setsockopt
= sock_no_setsockopt
,
557 .getsockopt
= sock_no_getsockopt
,
558 .sendmsg
= unix_stream_sendmsg
,
559 .recvmsg
= unix_stream_recvmsg
,
560 .mmap
= sock_no_mmap
,
561 .sendpage
= sock_no_sendpage
,
564 static const struct proto_ops unix_dgram_ops
= {
566 .owner
= THIS_MODULE
,
567 .release
= unix_release
,
569 .connect
= unix_dgram_connect
,
570 .socketpair
= unix_socketpair
,
571 .accept
= sock_no_accept
,
572 .getname
= unix_getname
,
573 .poll
= unix_dgram_poll
,
575 .listen
= sock_no_listen
,
576 .shutdown
= unix_shutdown
,
577 .setsockopt
= sock_no_setsockopt
,
578 .getsockopt
= sock_no_getsockopt
,
579 .sendmsg
= unix_dgram_sendmsg
,
580 .recvmsg
= unix_dgram_recvmsg
,
581 .mmap
= sock_no_mmap
,
582 .sendpage
= sock_no_sendpage
,
583 .set_peek_off
= unix_set_peek_off
,
586 static const struct proto_ops unix_seqpacket_ops
= {
588 .owner
= THIS_MODULE
,
589 .release
= unix_release
,
591 .connect
= unix_stream_connect
,
592 .socketpair
= unix_socketpair
,
593 .accept
= unix_accept
,
594 .getname
= unix_getname
,
595 .poll
= unix_dgram_poll
,
597 .listen
= unix_listen
,
598 .shutdown
= unix_shutdown
,
599 .setsockopt
= sock_no_setsockopt
,
600 .getsockopt
= sock_no_getsockopt
,
601 .sendmsg
= unix_seqpacket_sendmsg
,
602 .recvmsg
= unix_seqpacket_recvmsg
,
603 .mmap
= sock_no_mmap
,
604 .sendpage
= sock_no_sendpage
,
605 .set_peek_off
= unix_set_peek_off
,
608 static struct proto unix_proto
= {
610 .owner
= THIS_MODULE
,
611 .obj_size
= sizeof(struct unix_sock
),
615 * AF_UNIX sockets do not interact with hardware, hence they
616 * dont trigger interrupts - so it's safe for them to have
617 * bh-unsafe locking for their sk_receive_queue.lock. Split off
618 * this special lock-class by reinitializing the spinlock key:
620 static struct lock_class_key af_unix_sk_receive_queue_lock_key
;
622 static struct sock
*unix_create1(struct net
*net
, struct socket
*sock
)
624 struct sock
*sk
= NULL
;
627 atomic_long_inc(&unix_nr_socks
);
628 if (atomic_long_read(&unix_nr_socks
) > 2 * get_max_files())
631 sk
= sk_alloc(net
, PF_UNIX
, GFP_KERNEL
, &unix_proto
);
635 sock_init_data(sock
, sk
);
636 lockdep_set_class(&sk
->sk_receive_queue
.lock
,
637 &af_unix_sk_receive_queue_lock_key
);
639 sk
->sk_write_space
= unix_write_space
;
640 sk
->sk_max_ack_backlog
= net
->unx
.sysctl_max_dgram_qlen
;
641 sk
->sk_destruct
= unix_sock_destructor
;
645 spin_lock_init(&u
->lock
);
646 atomic_long_set(&u
->inflight
, 0);
647 INIT_LIST_HEAD(&u
->link
);
648 mutex_init(&u
->readlock
); /* single task reading lock */
649 init_waitqueue_head(&u
->peer_wait
);
650 unix_insert_socket(unix_sockets_unbound
, sk
);
653 atomic_long_dec(&unix_nr_socks
);
656 sock_prot_inuse_add(sock_net(sk
), sk
->sk_prot
, 1);
662 static int unix_create(struct net
*net
, struct socket
*sock
, int protocol
,
665 if (protocol
&& protocol
!= PF_UNIX
)
666 return -EPROTONOSUPPORT
;
668 sock
->state
= SS_UNCONNECTED
;
670 switch (sock
->type
) {
672 sock
->ops
= &unix_stream_ops
;
675 * Believe it or not BSD has AF_UNIX, SOCK_RAW though
679 sock
->type
= SOCK_DGRAM
;
681 sock
->ops
= &unix_dgram_ops
;
684 sock
->ops
= &unix_seqpacket_ops
;
687 return -ESOCKTNOSUPPORT
;
690 return unix_create1(net
, sock
) ? 0 : -ENOMEM
;
693 static int unix_release(struct socket
*sock
)
695 struct sock
*sk
= sock
->sk
;
702 return unix_release_sock(sk
, 0);
705 static int unix_autobind(struct socket
*sock
)
707 struct sock
*sk
= sock
->sk
;
708 struct net
*net
= sock_net(sk
);
709 struct unix_sock
*u
= unix_sk(sk
);
710 static u32 ordernum
= 1;
711 struct unix_address
*addr
;
713 unsigned int retries
= 0;
715 mutex_lock(&u
->readlock
);
722 addr
= kzalloc(sizeof(*addr
) + sizeof(short) + 16, GFP_KERNEL
);
726 addr
->name
->sun_family
= AF_UNIX
;
727 atomic_set(&addr
->refcnt
, 1);
730 addr
->len
= sprintf(addr
->name
->sun_path
+1, "%05x", ordernum
) + 1 + sizeof(short);
731 addr
->hash
= unix_hash_fold(csum_partial(addr
->name
, addr
->len
, 0));
733 spin_lock(&unix_table_lock
);
734 ordernum
= (ordernum
+1)&0xFFFFF;
736 if (__unix_find_socket_byname(net
, addr
->name
, addr
->len
, sock
->type
,
738 spin_unlock(&unix_table_lock
);
740 * __unix_find_socket_byname() may take long time if many names
741 * are already in use.
744 /* Give up if all names seems to be in use. */
745 if (retries
++ == 0xFFFFF) {
752 addr
->hash
^= sk
->sk_type
;
754 __unix_remove_socket(sk
);
756 __unix_insert_socket(&unix_socket_table
[addr
->hash
], sk
);
757 spin_unlock(&unix_table_lock
);
760 out
: mutex_unlock(&u
->readlock
);
764 static struct sock
*unix_find_other(struct net
*net
,
765 struct sockaddr_un
*sunname
, int len
,
766 int type
, unsigned hash
, int *error
)
772 if (sunname
->sun_path
[0]) {
774 err
= kern_path(sunname
->sun_path
, LOOKUP_FOLLOW
, &path
);
777 inode
= path
.dentry
->d_inode
;
778 err
= inode_permission(inode
, MAY_WRITE
);
783 if (!S_ISSOCK(inode
->i_mode
))
785 u
= unix_find_socket_byinode(inode
);
789 if (u
->sk_type
== type
)
790 touch_atime(path
.mnt
, path
.dentry
);
795 if (u
->sk_type
!= type
) {
801 u
= unix_find_socket_byname(net
, sunname
, len
, type
, hash
);
803 struct dentry
*dentry
;
804 dentry
= unix_sk(u
)->dentry
;
806 touch_atime(unix_sk(u
)->mnt
, dentry
);
820 static int unix_bind(struct socket
*sock
, struct sockaddr
*uaddr
, int addr_len
)
822 struct sock
*sk
= sock
->sk
;
823 struct net
*net
= sock_net(sk
);
824 struct unix_sock
*u
= unix_sk(sk
);
825 struct sockaddr_un
*sunaddr
= (struct sockaddr_un
*)uaddr
;
826 char *sun_path
= sunaddr
->sun_path
;
827 struct dentry
*dentry
= NULL
;
831 struct unix_address
*addr
;
832 struct hlist_head
*list
;
835 if (sunaddr
->sun_family
!= AF_UNIX
)
838 if (addr_len
== sizeof(short)) {
839 err
= unix_autobind(sock
);
843 err
= unix_mkname(sunaddr
, addr_len
, &hash
);
848 mutex_lock(&u
->readlock
);
855 addr
= kmalloc(sizeof(*addr
)+addr_len
, GFP_KERNEL
);
859 memcpy(addr
->name
, sunaddr
, addr_len
);
860 addr
->len
= addr_len
;
861 addr
->hash
= hash
^ sk
->sk_type
;
862 atomic_set(&addr
->refcnt
, 1);
868 * Get the parent directory, calculate the hash for last
871 dentry
= kern_path_create(AT_FDCWD
, sun_path
, &path
, 0);
872 err
= PTR_ERR(dentry
);
874 goto out_mknod_parent
;
877 * All right, let's create it.
880 (SOCK_INODE(sock
)->i_mode
& ~current_umask());
881 err
= mnt_want_write(path
.mnt
);
884 err
= security_path_mknod(&path
, dentry
, mode
, 0);
886 goto out_mknod_drop_write
;
887 err
= vfs_mknod(path
.dentry
->d_inode
, dentry
, mode
, 0);
888 out_mknod_drop_write
:
889 mnt_drop_write(path
.mnt
);
892 mutex_unlock(&path
.dentry
->d_inode
->i_mutex
);
894 path
.dentry
= dentry
;
896 addr
->hash
= UNIX_HASH_SIZE
;
899 spin_lock(&unix_table_lock
);
903 if (__unix_find_socket_byname(net
, sunaddr
, addr_len
,
904 sk
->sk_type
, hash
)) {
905 unix_release_addr(addr
);
909 list
= &unix_socket_table
[addr
->hash
];
911 list
= &unix_socket_table
[dentry
->d_inode
->i_ino
& (UNIX_HASH_SIZE
-1)];
912 u
->dentry
= path
.dentry
;
917 __unix_remove_socket(sk
);
919 __unix_insert_socket(list
, sk
);
922 spin_unlock(&unix_table_lock
);
924 mutex_unlock(&u
->readlock
);
930 mutex_unlock(&path
.dentry
->d_inode
->i_mutex
);
935 unix_release_addr(addr
);
939 static void unix_state_double_lock(struct sock
*sk1
, struct sock
*sk2
)
941 if (unlikely(sk1
== sk2
) || !sk2
) {
942 unix_state_lock(sk1
);
946 unix_state_lock(sk1
);
947 unix_state_lock_nested(sk2
);
949 unix_state_lock(sk2
);
950 unix_state_lock_nested(sk1
);
954 static void unix_state_double_unlock(struct sock
*sk1
, struct sock
*sk2
)
956 if (unlikely(sk1
== sk2
) || !sk2
) {
957 unix_state_unlock(sk1
);
960 unix_state_unlock(sk1
);
961 unix_state_unlock(sk2
);
964 static int unix_dgram_connect(struct socket
*sock
, struct sockaddr
*addr
,
967 struct sock
*sk
= sock
->sk
;
968 struct net
*net
= sock_net(sk
);
969 struct sockaddr_un
*sunaddr
= (struct sockaddr_un
*)addr
;
974 if (addr
->sa_family
!= AF_UNSPEC
) {
975 err
= unix_mkname(sunaddr
, alen
, &hash
);
980 if (test_bit(SOCK_PASSCRED
, &sock
->flags
) &&
981 !unix_sk(sk
)->addr
&& (err
= unix_autobind(sock
)) != 0)
985 other
= unix_find_other(net
, sunaddr
, alen
, sock
->type
, hash
, &err
);
989 unix_state_double_lock(sk
, other
);
991 /* Apparently VFS overslept socket death. Retry. */
992 if (sock_flag(other
, SOCK_DEAD
)) {
993 unix_state_double_unlock(sk
, other
);
999 if (!unix_may_send(sk
, other
))
1002 err
= security_unix_may_send(sk
->sk_socket
, other
->sk_socket
);
1008 * 1003.1g breaking connected state with AF_UNSPEC
1011 unix_state_double_lock(sk
, other
);
1015 * If it was connected, reconnect.
1017 if (unix_peer(sk
)) {
1018 struct sock
*old_peer
= unix_peer(sk
);
1019 unix_peer(sk
) = other
;
1020 unix_state_double_unlock(sk
, other
);
1022 if (other
!= old_peer
)
1023 unix_dgram_disconnected(sk
, old_peer
);
1026 unix_peer(sk
) = other
;
1027 unix_state_double_unlock(sk
, other
);
1032 unix_state_double_unlock(sk
, other
);
1038 static long unix_wait_for_peer(struct sock
*other
, long timeo
)
1040 struct unix_sock
*u
= unix_sk(other
);
1044 prepare_to_wait_exclusive(&u
->peer_wait
, &wait
, TASK_INTERRUPTIBLE
);
1046 sched
= !sock_flag(other
, SOCK_DEAD
) &&
1047 !(other
->sk_shutdown
& RCV_SHUTDOWN
) &&
1048 unix_recvq_full(other
);
1050 unix_state_unlock(other
);
1053 timeo
= schedule_timeout(timeo
);
1055 finish_wait(&u
->peer_wait
, &wait
);
1059 static int unix_stream_connect(struct socket
*sock
, struct sockaddr
*uaddr
,
1060 int addr_len
, int flags
)
1062 struct sockaddr_un
*sunaddr
= (struct sockaddr_un
*)uaddr
;
1063 struct sock
*sk
= sock
->sk
;
1064 struct net
*net
= sock_net(sk
);
1065 struct unix_sock
*u
= unix_sk(sk
), *newu
, *otheru
;
1066 struct sock
*newsk
= NULL
;
1067 struct sock
*other
= NULL
;
1068 struct sk_buff
*skb
= NULL
;
1074 err
= unix_mkname(sunaddr
, addr_len
, &hash
);
1079 if (test_bit(SOCK_PASSCRED
, &sock
->flags
) && !u
->addr
&&
1080 (err
= unix_autobind(sock
)) != 0)
1083 timeo
= sock_sndtimeo(sk
, flags
& O_NONBLOCK
);
1085 /* First of all allocate resources.
1086 If we will make it after state is locked,
1087 we will have to recheck all again in any case.
1092 /* create new sock for complete connection */
1093 newsk
= unix_create1(sock_net(sk
), NULL
);
1097 /* Allocate skb for sending to listening sock */
1098 skb
= sock_wmalloc(newsk
, 1, 0, GFP_KERNEL
);
1103 /* Find listening sock. */
1104 other
= unix_find_other(net
, sunaddr
, addr_len
, sk
->sk_type
, hash
, &err
);
1108 /* Latch state of peer */
1109 unix_state_lock(other
);
1111 /* Apparently VFS overslept socket death. Retry. */
1112 if (sock_flag(other
, SOCK_DEAD
)) {
1113 unix_state_unlock(other
);
1118 err
= -ECONNREFUSED
;
1119 if (other
->sk_state
!= TCP_LISTEN
)
1121 if (other
->sk_shutdown
& RCV_SHUTDOWN
)
1124 if (unix_recvq_full(other
)) {
1129 timeo
= unix_wait_for_peer(other
, timeo
);
1131 err
= sock_intr_errno(timeo
);
1132 if (signal_pending(current
))
1140 It is tricky place. We need to grab our state lock and cannot
1141 drop lock on peer. It is dangerous because deadlock is
1142 possible. Connect to self case and simultaneous
1143 attempt to connect are eliminated by checking socket
1144 state. other is TCP_LISTEN, if sk is TCP_LISTEN we
1145 check this before attempt to grab lock.
1147 Well, and we have to recheck the state after socket locked.
1153 /* This is ok... continue with connect */
1155 case TCP_ESTABLISHED
:
1156 /* Socket is already connected */
1164 unix_state_lock_nested(sk
);
1166 if (sk
->sk_state
!= st
) {
1167 unix_state_unlock(sk
);
1168 unix_state_unlock(other
);
1173 err
= security_unix_stream_connect(sk
, other
, newsk
);
1175 unix_state_unlock(sk
);
1179 /* The way is open! Fastly set all the necessary fields... */
1182 unix_peer(newsk
) = sk
;
1183 newsk
->sk_state
= TCP_ESTABLISHED
;
1184 newsk
->sk_type
= sk
->sk_type
;
1185 init_peercred(newsk
);
1186 newu
= unix_sk(newsk
);
1187 RCU_INIT_POINTER(newsk
->sk_wq
, &newu
->peer_wq
);
1188 otheru
= unix_sk(other
);
1190 /* copy address information from listening to new sock*/
1192 atomic_inc(&otheru
->addr
->refcnt
);
1193 newu
->addr
= otheru
->addr
;
1195 if (otheru
->dentry
) {
1196 newu
->dentry
= dget(otheru
->dentry
);
1197 newu
->mnt
= mntget(otheru
->mnt
);
1200 /* Set credentials */
1201 copy_peercred(sk
, other
);
1203 sock
->state
= SS_CONNECTED
;
1204 sk
->sk_state
= TCP_ESTABLISHED
;
1207 smp_mb__after_atomic_inc(); /* sock_hold() does an atomic_inc() */
1208 unix_peer(sk
) = newsk
;
1210 unix_state_unlock(sk
);
1212 /* take ten and and send info to listening sock */
1213 spin_lock(&other
->sk_receive_queue
.lock
);
1214 __skb_queue_tail(&other
->sk_receive_queue
, skb
);
1215 spin_unlock(&other
->sk_receive_queue
.lock
);
1216 unix_state_unlock(other
);
1217 other
->sk_data_ready(other
, 0);
1223 unix_state_unlock(other
);
1228 unix_release_sock(newsk
, 0);
1234 static int unix_socketpair(struct socket
*socka
, struct socket
*sockb
)
1236 struct sock
*ska
= socka
->sk
, *skb
= sockb
->sk
;
1238 /* Join our sockets back to back */
1241 unix_peer(ska
) = skb
;
1242 unix_peer(skb
) = ska
;
1246 if (ska
->sk_type
!= SOCK_DGRAM
) {
1247 ska
->sk_state
= TCP_ESTABLISHED
;
1248 skb
->sk_state
= TCP_ESTABLISHED
;
1249 socka
->state
= SS_CONNECTED
;
1250 sockb
->state
= SS_CONNECTED
;
1255 static int unix_accept(struct socket
*sock
, struct socket
*newsock
, int flags
)
1257 struct sock
*sk
= sock
->sk
;
1259 struct sk_buff
*skb
;
1263 if (sock
->type
!= SOCK_STREAM
&& sock
->type
!= SOCK_SEQPACKET
)
1267 if (sk
->sk_state
!= TCP_LISTEN
)
1270 /* If socket state is TCP_LISTEN it cannot change (for now...),
1271 * so that no locks are necessary.
1274 skb
= skb_recv_datagram(sk
, 0, flags
&O_NONBLOCK
, &err
);
1276 /* This means receive shutdown. */
1283 skb_free_datagram(sk
, skb
);
1284 wake_up_interruptible(&unix_sk(sk
)->peer_wait
);
1286 /* attach accepted sock to socket */
1287 unix_state_lock(tsk
);
1288 newsock
->state
= SS_CONNECTED
;
1289 sock_graft(tsk
, newsock
);
1290 unix_state_unlock(tsk
);
1298 static int unix_getname(struct socket
*sock
, struct sockaddr
*uaddr
, int *uaddr_len
, int peer
)
1300 struct sock
*sk
= sock
->sk
;
1301 struct unix_sock
*u
;
1302 DECLARE_SOCKADDR(struct sockaddr_un
*, sunaddr
, uaddr
);
1306 sk
= unix_peer_get(sk
);
1317 unix_state_lock(sk
);
1319 sunaddr
->sun_family
= AF_UNIX
;
1320 sunaddr
->sun_path
[0] = 0;
1321 *uaddr_len
= sizeof(short);
1323 struct unix_address
*addr
= u
->addr
;
1325 *uaddr_len
= addr
->len
;
1326 memcpy(sunaddr
, addr
->name
, *uaddr_len
);
1328 unix_state_unlock(sk
);
1334 static void unix_detach_fds(struct scm_cookie
*scm
, struct sk_buff
*skb
)
1338 scm
->fp
= UNIXCB(skb
).fp
;
1339 UNIXCB(skb
).fp
= NULL
;
1341 for (i
= scm
->fp
->count
-1; i
>= 0; i
--)
1342 unix_notinflight(scm
->fp
->fp
[i
]);
1345 static void unix_destruct_scm(struct sk_buff
*skb
)
1347 struct scm_cookie scm
;
1348 memset(&scm
, 0, sizeof(scm
));
1349 scm
.pid
= UNIXCB(skb
).pid
;
1350 scm
.cred
= UNIXCB(skb
).cred
;
1352 unix_detach_fds(&scm
, skb
);
1354 /* Alas, it calls VFS */
1355 /* So fscking what? fput() had been SMP-safe since the last Summer */
1360 #define MAX_RECURSION_LEVEL 4
1362 static int unix_attach_fds(struct scm_cookie
*scm
, struct sk_buff
*skb
)
1365 unsigned char max_level
= 0;
1366 int unix_sock_count
= 0;
1368 for (i
= scm
->fp
->count
- 1; i
>= 0; i
--) {
1369 struct sock
*sk
= unix_get_socket(scm
->fp
->fp
[i
]);
1373 max_level
= max(max_level
,
1374 unix_sk(sk
)->recursion_level
);
1377 if (unlikely(max_level
> MAX_RECURSION_LEVEL
))
1378 return -ETOOMANYREFS
;
1381 * Need to duplicate file references for the sake of garbage
1382 * collection. Otherwise a socket in the fps might become a
1383 * candidate for GC while the skb is not yet queued.
1385 UNIXCB(skb
).fp
= scm_fp_dup(scm
->fp
);
1386 if (!UNIXCB(skb
).fp
)
1389 if (unix_sock_count
) {
1390 for (i
= scm
->fp
->count
- 1; i
>= 0; i
--)
1391 unix_inflight(scm
->fp
->fp
[i
]);
1396 static int unix_scm_to_skb(struct scm_cookie
*scm
, struct sk_buff
*skb
, bool send_fds
)
1400 UNIXCB(skb
).pid
= get_pid(scm
->pid
);
1402 UNIXCB(skb
).cred
= get_cred(scm
->cred
);
1403 UNIXCB(skb
).fp
= NULL
;
1404 if (scm
->fp
&& send_fds
)
1405 err
= unix_attach_fds(scm
, skb
);
1407 skb
->destructor
= unix_destruct_scm
;
1412 * Some apps rely on write() giving SCM_CREDENTIALS
1413 * We include credentials if source or destination socket
1414 * asserted SOCK_PASSCRED.
1416 static void maybe_add_creds(struct sk_buff
*skb
, const struct socket
*sock
,
1417 const struct sock
*other
)
1419 if (UNIXCB(skb
).cred
)
1421 if (test_bit(SOCK_PASSCRED
, &sock
->flags
) ||
1422 !other
->sk_socket
||
1423 test_bit(SOCK_PASSCRED
, &other
->sk_socket
->flags
)) {
1424 UNIXCB(skb
).pid
= get_pid(task_tgid(current
));
1425 UNIXCB(skb
).cred
= get_current_cred();
1430 * Send AF_UNIX data.
1433 static int unix_dgram_sendmsg(struct kiocb
*kiocb
, struct socket
*sock
,
1434 struct msghdr
*msg
, size_t len
)
1436 struct sock_iocb
*siocb
= kiocb_to_siocb(kiocb
);
1437 struct sock
*sk
= sock
->sk
;
1438 struct net
*net
= sock_net(sk
);
1439 struct unix_sock
*u
= unix_sk(sk
);
1440 struct sockaddr_un
*sunaddr
= msg
->msg_name
;
1441 struct sock
*other
= NULL
;
1442 int namelen
= 0; /* fake GCC */
1445 struct sk_buff
*skb
;
1447 struct scm_cookie tmp_scm
;
1450 if (NULL
== siocb
->scm
)
1451 siocb
->scm
= &tmp_scm
;
1453 err
= scm_send(sock
, msg
, siocb
->scm
);
1458 if (msg
->msg_flags
&MSG_OOB
)
1461 if (msg
->msg_namelen
) {
1462 err
= unix_mkname(sunaddr
, msg
->msg_namelen
, &hash
);
1469 other
= unix_peer_get(sk
);
1474 if (test_bit(SOCK_PASSCRED
, &sock
->flags
) && !u
->addr
1475 && (err
= unix_autobind(sock
)) != 0)
1479 if (len
> sk
->sk_sndbuf
- 32)
1482 skb
= sock_alloc_send_skb(sk
, len
, msg
->msg_flags
&MSG_DONTWAIT
, &err
);
1486 err
= unix_scm_to_skb(siocb
->scm
, skb
, true);
1489 max_level
= err
+ 1;
1490 unix_get_secdata(siocb
->scm
, skb
);
1492 skb_reset_transport_header(skb
);
1493 err
= memcpy_fromiovec(skb_put(skb
, len
), msg
->msg_iov
, len
);
1497 timeo
= sock_sndtimeo(sk
, msg
->msg_flags
& MSG_DONTWAIT
);
1502 if (sunaddr
== NULL
)
1505 other
= unix_find_other(net
, sunaddr
, namelen
, sk
->sk_type
,
1511 if (sk_filter(other
, skb
) < 0) {
1512 /* Toss the packet but do not return any error to the sender */
1517 unix_state_lock(other
);
1519 if (!unix_may_send(sk
, other
))
1522 if (sock_flag(other
, SOCK_DEAD
)) {
1524 * Check with 1003.1g - what should
1527 unix_state_unlock(other
);
1531 unix_state_lock(sk
);
1532 if (unix_peer(sk
) == other
) {
1533 unix_peer(sk
) = NULL
;
1534 unix_state_unlock(sk
);
1536 unix_dgram_disconnected(sk
, other
);
1538 err
= -ECONNREFUSED
;
1540 unix_state_unlock(sk
);
1550 if (other
->sk_shutdown
& RCV_SHUTDOWN
)
1553 if (sk
->sk_type
!= SOCK_SEQPACKET
) {
1554 err
= security_unix_may_send(sk
->sk_socket
, other
->sk_socket
);
1559 if (unix_peer(other
) != sk
&& unix_recvq_full(other
)) {
1565 timeo
= unix_wait_for_peer(other
, timeo
);
1567 err
= sock_intr_errno(timeo
);
1568 if (signal_pending(current
))
1574 if (sock_flag(other
, SOCK_RCVTSTAMP
))
1575 __net_timestamp(skb
);
1576 maybe_add_creds(skb
, sock
, other
);
1577 skb_queue_tail(&other
->sk_receive_queue
, skb
);
1578 if (max_level
> unix_sk(other
)->recursion_level
)
1579 unix_sk(other
)->recursion_level
= max_level
;
1580 unix_state_unlock(other
);
1581 other
->sk_data_ready(other
, len
);
1583 scm_destroy(siocb
->scm
);
1587 unix_state_unlock(other
);
1593 scm_destroy(siocb
->scm
);
1598 static int unix_stream_sendmsg(struct kiocb
*kiocb
, struct socket
*sock
,
1599 struct msghdr
*msg
, size_t len
)
1601 struct sock_iocb
*siocb
= kiocb_to_siocb(kiocb
);
1602 struct sock
*sk
= sock
->sk
;
1603 struct sock
*other
= NULL
;
1605 struct sk_buff
*skb
;
1607 struct scm_cookie tmp_scm
;
1608 bool fds_sent
= false;
1611 if (NULL
== siocb
->scm
)
1612 siocb
->scm
= &tmp_scm
;
1614 err
= scm_send(sock
, msg
, siocb
->scm
);
1619 if (msg
->msg_flags
&MSG_OOB
)
1622 if (msg
->msg_namelen
) {
1623 err
= sk
->sk_state
== TCP_ESTABLISHED
? -EISCONN
: -EOPNOTSUPP
;
1627 other
= unix_peer(sk
);
1632 if (sk
->sk_shutdown
& SEND_SHUTDOWN
)
1635 while (sent
< len
) {
1637 * Optimisation for the fact that under 0.01% of X
1638 * messages typically need breaking up.
1643 /* Keep two messages in the pipe so it schedules better */
1644 if (size
> ((sk
->sk_sndbuf
>> 1) - 64))
1645 size
= (sk
->sk_sndbuf
>> 1) - 64;
1647 if (size
> SKB_MAX_ALLOC
)
1648 size
= SKB_MAX_ALLOC
;
1654 skb
= sock_alloc_send_skb(sk
, size
, msg
->msg_flags
&MSG_DONTWAIT
,
1661 * If you pass two values to the sock_alloc_send_skb
1662 * it tries to grab the large buffer with GFP_NOFS
1663 * (which can fail easily), and if it fails grab the
1664 * fallback size buffer which is under a page and will
1667 size
= min_t(int, size
, skb_tailroom(skb
));
1670 /* Only send the fds in the first buffer */
1671 err
= unix_scm_to_skb(siocb
->scm
, skb
, !fds_sent
);
1676 max_level
= err
+ 1;
1679 err
= memcpy_fromiovec(skb_put(skb
, size
), msg
->msg_iov
, size
);
1685 unix_state_lock(other
);
1687 if (sock_flag(other
, SOCK_DEAD
) ||
1688 (other
->sk_shutdown
& RCV_SHUTDOWN
))
1691 maybe_add_creds(skb
, sock
, other
);
1692 skb_queue_tail(&other
->sk_receive_queue
, skb
);
1693 if (max_level
> unix_sk(other
)->recursion_level
)
1694 unix_sk(other
)->recursion_level
= max_level
;
1695 unix_state_unlock(other
);
1696 other
->sk_data_ready(other
, size
);
1700 scm_destroy(siocb
->scm
);
1706 unix_state_unlock(other
);
1709 if (sent
== 0 && !(msg
->msg_flags
&MSG_NOSIGNAL
))
1710 send_sig(SIGPIPE
, current
, 0);
1713 scm_destroy(siocb
->scm
);
1715 return sent
? : err
;
1718 static int unix_seqpacket_sendmsg(struct kiocb
*kiocb
, struct socket
*sock
,
1719 struct msghdr
*msg
, size_t len
)
1722 struct sock
*sk
= sock
->sk
;
1724 err
= sock_error(sk
);
1728 if (sk
->sk_state
!= TCP_ESTABLISHED
)
1731 if (msg
->msg_namelen
)
1732 msg
->msg_namelen
= 0;
1734 return unix_dgram_sendmsg(kiocb
, sock
, msg
, len
);
1737 static int unix_seqpacket_recvmsg(struct kiocb
*iocb
, struct socket
*sock
,
1738 struct msghdr
*msg
, size_t size
,
1741 struct sock
*sk
= sock
->sk
;
1743 if (sk
->sk_state
!= TCP_ESTABLISHED
)
1746 return unix_dgram_recvmsg(iocb
, sock
, msg
, size
, flags
);
1749 static void unix_copy_addr(struct msghdr
*msg
, struct sock
*sk
)
1751 struct unix_sock
*u
= unix_sk(sk
);
1753 msg
->msg_namelen
= 0;
1755 msg
->msg_namelen
= u
->addr
->len
;
1756 memcpy(msg
->msg_name
, u
->addr
->name
, u
->addr
->len
);
1760 static int unix_dgram_recvmsg(struct kiocb
*iocb
, struct socket
*sock
,
1761 struct msghdr
*msg
, size_t size
,
1764 struct sock_iocb
*siocb
= kiocb_to_siocb(iocb
);
1765 struct scm_cookie tmp_scm
;
1766 struct sock
*sk
= sock
->sk
;
1767 struct unix_sock
*u
= unix_sk(sk
);
1768 int noblock
= flags
& MSG_DONTWAIT
;
1769 struct sk_buff
*skb
;
1777 msg
->msg_namelen
= 0;
1779 err
= mutex_lock_interruptible(&u
->readlock
);
1781 err
= sock_intr_errno(sock_rcvtimeo(sk
, noblock
));
1785 skip
= sk_peek_offset(sk
, flags
);
1787 skb
= __skb_recv_datagram(sk
, flags
, &peeked
, &skip
, &err
);
1789 unix_state_lock(sk
);
1790 /* Signal EOF on disconnected non-blocking SEQPACKET socket. */
1791 if (sk
->sk_type
== SOCK_SEQPACKET
&& err
== -EAGAIN
&&
1792 (sk
->sk_shutdown
& RCV_SHUTDOWN
))
1794 unix_state_unlock(sk
);
1798 wake_up_interruptible_sync_poll(&u
->peer_wait
,
1799 POLLOUT
| POLLWRNORM
| POLLWRBAND
);
1802 unix_copy_addr(msg
, skb
->sk
);
1804 if (size
> skb
->len
- skip
)
1805 size
= skb
->len
- skip
;
1806 else if (size
< skb
->len
- skip
)
1807 msg
->msg_flags
|= MSG_TRUNC
;
1809 err
= skb_copy_datagram_iovec(skb
, skip
, msg
->msg_iov
, size
);
1813 if (sock_flag(sk
, SOCK_RCVTSTAMP
))
1814 __sock_recv_timestamp(msg
, sk
, skb
);
1817 siocb
->scm
= &tmp_scm
;
1818 memset(&tmp_scm
, 0, sizeof(tmp_scm
));
1820 scm_set_cred(siocb
->scm
, UNIXCB(skb
).pid
, UNIXCB(skb
).cred
);
1821 unix_set_secdata(siocb
->scm
, skb
);
1823 if (!(flags
& MSG_PEEK
)) {
1825 unix_detach_fds(siocb
->scm
, skb
);
1827 sk_peek_offset_bwd(sk
, skb
->len
);
1829 /* It is questionable: on PEEK we could:
1830 - do not return fds - good, but too simple 8)
1831 - return fds, and do not return them on read (old strategy,
1833 - clone fds (I chose it for now, it is the most universal
1836 POSIX 1003.1g does not actually define this clearly
1837 at all. POSIX 1003.1g doesn't define a lot of things
1842 sk_peek_offset_fwd(sk
, size
);
1845 siocb
->scm
->fp
= scm_fp_dup(UNIXCB(skb
).fp
);
1849 scm_recv(sock
, msg
, siocb
->scm
, flags
);
1852 skb_free_datagram(sk
, skb
);
1854 mutex_unlock(&u
->readlock
);
1860 * Sleep until data has arrive. But check for races..
1863 static long unix_stream_data_wait(struct sock
*sk
, long timeo
)
1867 unix_state_lock(sk
);
1870 prepare_to_wait(sk_sleep(sk
), &wait
, TASK_INTERRUPTIBLE
);
1872 if (!skb_queue_empty(&sk
->sk_receive_queue
) ||
1874 (sk
->sk_shutdown
& RCV_SHUTDOWN
) ||
1875 signal_pending(current
) ||
1879 set_bit(SOCK_ASYNC_WAITDATA
, &sk
->sk_socket
->flags
);
1880 unix_state_unlock(sk
);
1881 timeo
= schedule_timeout(timeo
);
1882 unix_state_lock(sk
);
1883 clear_bit(SOCK_ASYNC_WAITDATA
, &sk
->sk_socket
->flags
);
1886 finish_wait(sk_sleep(sk
), &wait
);
1887 unix_state_unlock(sk
);
1893 static int unix_stream_recvmsg(struct kiocb
*iocb
, struct socket
*sock
,
1894 struct msghdr
*msg
, size_t size
,
1897 struct sock_iocb
*siocb
= kiocb_to_siocb(iocb
);
1898 struct scm_cookie tmp_scm
;
1899 struct sock
*sk
= sock
->sk
;
1900 struct unix_sock
*u
= unix_sk(sk
);
1901 struct sockaddr_un
*sunaddr
= msg
->msg_name
;
1903 int check_creds
= 0;
1909 if (sk
->sk_state
!= TCP_ESTABLISHED
)
1916 target
= sock_rcvlowat(sk
, flags
&MSG_WAITALL
, size
);
1917 timeo
= sock_rcvtimeo(sk
, flags
&MSG_DONTWAIT
);
1919 msg
->msg_namelen
= 0;
1921 /* Lock the socket to prevent queue disordering
1922 * while sleeps in memcpy_tomsg
1926 siocb
->scm
= &tmp_scm
;
1927 memset(&tmp_scm
, 0, sizeof(tmp_scm
));
1930 err
= mutex_lock_interruptible(&u
->readlock
);
1932 err
= sock_intr_errno(timeo
);
1938 struct sk_buff
*skb
;
1940 unix_state_lock(sk
);
1941 skb
= skb_peek(&sk
->sk_receive_queue
);
1943 unix_sk(sk
)->recursion_level
= 0;
1944 if (copied
>= target
)
1948 * POSIX 1003.1g mandates this order.
1951 err
= sock_error(sk
);
1954 if (sk
->sk_shutdown
& RCV_SHUTDOWN
)
1957 unix_state_unlock(sk
);
1961 mutex_unlock(&u
->readlock
);
1963 timeo
= unix_stream_data_wait(sk
, timeo
);
1965 if (signal_pending(current
)
1966 || mutex_lock_interruptible(&u
->readlock
)) {
1967 err
= sock_intr_errno(timeo
);
1973 unix_state_unlock(sk
);
1976 unix_state_unlock(sk
);
1979 /* Never glue messages from different writers */
1980 if ((UNIXCB(skb
).pid
!= siocb
->scm
->pid
) ||
1981 (UNIXCB(skb
).cred
!= siocb
->scm
->cred
))
1984 /* Copy credentials */
1985 scm_set_cred(siocb
->scm
, UNIXCB(skb
).pid
, UNIXCB(skb
).cred
);
1989 /* Copy address just once */
1991 unix_copy_addr(msg
, skb
->sk
);
1995 chunk
= min_t(unsigned int, skb
->len
, size
);
1996 if (memcpy_toiovec(msg
->msg_iov
, skb
->data
, chunk
)) {
2004 /* Mark read part of skb as used */
2005 if (!(flags
& MSG_PEEK
)) {
2006 skb_pull(skb
, chunk
);
2009 unix_detach_fds(siocb
->scm
, skb
);
2014 skb_unlink(skb
, &sk
->sk_receive_queue
);
2020 /* It is questionable, see note in unix_dgram_recvmsg.
2023 siocb
->scm
->fp
= scm_fp_dup(UNIXCB(skb
).fp
);
2029 mutex_unlock(&u
->readlock
);
2030 scm_recv(sock
, msg
, siocb
->scm
, flags
);
2032 return copied
? : err
;
2035 static int unix_shutdown(struct socket
*sock
, int mode
)
2037 struct sock
*sk
= sock
->sk
;
2040 mode
= (mode
+1)&(RCV_SHUTDOWN
|SEND_SHUTDOWN
);
2045 unix_state_lock(sk
);
2046 sk
->sk_shutdown
|= mode
;
2047 other
= unix_peer(sk
);
2050 unix_state_unlock(sk
);
2051 sk
->sk_state_change(sk
);
2054 (sk
->sk_type
== SOCK_STREAM
|| sk
->sk_type
== SOCK_SEQPACKET
)) {
2058 if (mode
&RCV_SHUTDOWN
)
2059 peer_mode
|= SEND_SHUTDOWN
;
2060 if (mode
&SEND_SHUTDOWN
)
2061 peer_mode
|= RCV_SHUTDOWN
;
2062 unix_state_lock(other
);
2063 other
->sk_shutdown
|= peer_mode
;
2064 unix_state_unlock(other
);
2065 other
->sk_state_change(other
);
2066 if (peer_mode
== SHUTDOWN_MASK
)
2067 sk_wake_async(other
, SOCK_WAKE_WAITD
, POLL_HUP
);
2068 else if (peer_mode
& RCV_SHUTDOWN
)
2069 sk_wake_async(other
, SOCK_WAKE_WAITD
, POLL_IN
);
2077 long unix_inq_len(struct sock
*sk
)
2079 struct sk_buff
*skb
;
2082 if (sk
->sk_state
== TCP_LISTEN
)
2085 spin_lock(&sk
->sk_receive_queue
.lock
);
2086 if (sk
->sk_type
== SOCK_STREAM
||
2087 sk
->sk_type
== SOCK_SEQPACKET
) {
2088 skb_queue_walk(&sk
->sk_receive_queue
, skb
)
2091 skb
= skb_peek(&sk
->sk_receive_queue
);
2095 spin_unlock(&sk
->sk_receive_queue
.lock
);
2099 EXPORT_SYMBOL_GPL(unix_inq_len
);
2101 long unix_outq_len(struct sock
*sk
)
2103 return sk_wmem_alloc_get(sk
);
2105 EXPORT_SYMBOL_GPL(unix_outq_len
);
2107 static int unix_ioctl(struct socket
*sock
, unsigned int cmd
, unsigned long arg
)
2109 struct sock
*sk
= sock
->sk
;
2115 amount
= unix_outq_len(sk
);
2116 err
= put_user(amount
, (int __user
*)arg
);
2119 amount
= unix_inq_len(sk
);
2123 err
= put_user(amount
, (int __user
*)arg
);
2132 static unsigned int unix_poll(struct file
*file
, struct socket
*sock
, poll_table
*wait
)
2134 struct sock
*sk
= sock
->sk
;
2137 sock_poll_wait(file
, sk_sleep(sk
), wait
);
2140 /* exceptional events? */
2143 if (sk
->sk_shutdown
== SHUTDOWN_MASK
)
2145 if (sk
->sk_shutdown
& RCV_SHUTDOWN
)
2146 mask
|= POLLRDHUP
| POLLIN
| POLLRDNORM
;
2149 if (!skb_queue_empty(&sk
->sk_receive_queue
))
2150 mask
|= POLLIN
| POLLRDNORM
;
2152 /* Connection-based need to check for termination and startup */
2153 if ((sk
->sk_type
== SOCK_STREAM
|| sk
->sk_type
== SOCK_SEQPACKET
) &&
2154 sk
->sk_state
== TCP_CLOSE
)
2158 * we set writable also when the other side has shut down the
2159 * connection. This prevents stuck sockets.
2161 if (unix_writable(sk
))
2162 mask
|= POLLOUT
| POLLWRNORM
| POLLWRBAND
;
2167 static unsigned int unix_dgram_poll(struct file
*file
, struct socket
*sock
,
2170 struct sock
*sk
= sock
->sk
, *other
;
2171 unsigned int mask
, writable
;
2173 sock_poll_wait(file
, sk_sleep(sk
), wait
);
2176 /* exceptional events? */
2177 if (sk
->sk_err
|| !skb_queue_empty(&sk
->sk_error_queue
))
2179 if (sk
->sk_shutdown
& RCV_SHUTDOWN
)
2180 mask
|= POLLRDHUP
| POLLIN
| POLLRDNORM
;
2181 if (sk
->sk_shutdown
== SHUTDOWN_MASK
)
2185 if (!skb_queue_empty(&sk
->sk_receive_queue
))
2186 mask
|= POLLIN
| POLLRDNORM
;
2188 /* Connection-based need to check for termination and startup */
2189 if (sk
->sk_type
== SOCK_SEQPACKET
) {
2190 if (sk
->sk_state
== TCP_CLOSE
)
2192 /* connection hasn't started yet? */
2193 if (sk
->sk_state
== TCP_SYN_SENT
)
2197 /* No write status requested, avoid expensive OUT tests. */
2198 if (wait
&& !(wait
->key
& (POLLWRBAND
| POLLWRNORM
| POLLOUT
)))
2201 writable
= unix_writable(sk
);
2202 other
= unix_peer_get(sk
);
2204 if (unix_peer(other
) != sk
) {
2205 sock_poll_wait(file
, &unix_sk(other
)->peer_wait
, wait
);
2206 if (unix_recvq_full(other
))
2213 mask
|= POLLOUT
| POLLWRNORM
| POLLWRBAND
;
2215 set_bit(SOCK_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
);
2220 #ifdef CONFIG_PROC_FS
2221 static struct sock
*first_unix_socket(int *i
)
2223 for (*i
= 0; *i
<= UNIX_HASH_SIZE
; (*i
)++) {
2224 if (!hlist_empty(&unix_socket_table
[*i
]))
2225 return __sk_head(&unix_socket_table
[*i
]);
2230 static struct sock
*next_unix_socket(int *i
, struct sock
*s
)
2232 struct sock
*next
= sk_next(s
);
2233 /* More in this chain? */
2236 /* Look for next non-empty chain. */
2237 for ((*i
)++; *i
<= UNIX_HASH_SIZE
; (*i
)++) {
2238 if (!hlist_empty(&unix_socket_table
[*i
]))
2239 return __sk_head(&unix_socket_table
[*i
]);
2244 struct unix_iter_state
{
2245 struct seq_net_private p
;
2249 static struct sock
*unix_seq_idx(struct seq_file
*seq
, loff_t pos
)
2251 struct unix_iter_state
*iter
= seq
->private;
2255 for (s
= first_unix_socket(&iter
->i
); s
; s
= next_unix_socket(&iter
->i
, s
)) {
2256 if (sock_net(s
) != seq_file_net(seq
))
2265 static void *unix_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2266 __acquires(unix_table_lock
)
2268 spin_lock(&unix_table_lock
);
2269 return *pos
? unix_seq_idx(seq
, *pos
- 1) : SEQ_START_TOKEN
;
2272 static void *unix_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2274 struct unix_iter_state
*iter
= seq
->private;
2275 struct sock
*sk
= v
;
2278 if (v
== SEQ_START_TOKEN
)
2279 sk
= first_unix_socket(&iter
->i
);
2281 sk
= next_unix_socket(&iter
->i
, sk
);
2282 while (sk
&& (sock_net(sk
) != seq_file_net(seq
)))
2283 sk
= next_unix_socket(&iter
->i
, sk
);
2287 static void unix_seq_stop(struct seq_file
*seq
, void *v
)
2288 __releases(unix_table_lock
)
2290 spin_unlock(&unix_table_lock
);
2293 static int unix_seq_show(struct seq_file
*seq
, void *v
)
2296 if (v
== SEQ_START_TOKEN
)
2297 seq_puts(seq
, "Num RefCount Protocol Flags Type St "
2301 struct unix_sock
*u
= unix_sk(s
);
2304 seq_printf(seq
, "%pK: %08X %08X %08X %04X %02X %5lu",
2306 atomic_read(&s
->sk_refcnt
),
2308 s
->sk_state
== TCP_LISTEN
? __SO_ACCEPTCON
: 0,
2311 (s
->sk_state
== TCP_ESTABLISHED
? SS_CONNECTED
: SS_UNCONNECTED
) :
2312 (s
->sk_state
== TCP_ESTABLISHED
? SS_CONNECTING
: SS_DISCONNECTING
),
2320 len
= u
->addr
->len
- sizeof(short);
2321 if (!UNIX_ABSTRACT(s
))
2327 for ( ; i
< len
; i
++)
2328 seq_putc(seq
, u
->addr
->name
->sun_path
[i
]);
2330 unix_state_unlock(s
);
2331 seq_putc(seq
, '\n');
2337 static const struct seq_operations unix_seq_ops
= {
2338 .start
= unix_seq_start
,
2339 .next
= unix_seq_next
,
2340 .stop
= unix_seq_stop
,
2341 .show
= unix_seq_show
,
2344 static int unix_seq_open(struct inode
*inode
, struct file
*file
)
2346 return seq_open_net(inode
, file
, &unix_seq_ops
,
2347 sizeof(struct unix_iter_state
));
2350 static const struct file_operations unix_seq_fops
= {
2351 .owner
= THIS_MODULE
,
2352 .open
= unix_seq_open
,
2354 .llseek
= seq_lseek
,
2355 .release
= seq_release_net
,
2360 static const struct net_proto_family unix_family_ops
= {
2362 .create
= unix_create
,
2363 .owner
= THIS_MODULE
,
2367 static int __net_init
unix_net_init(struct net
*net
)
2369 int error
= -ENOMEM
;
2371 net
->unx
.sysctl_max_dgram_qlen
= 10;
2372 if (unix_sysctl_register(net
))
2375 #ifdef CONFIG_PROC_FS
2376 if (!proc_net_fops_create(net
, "unix", 0, &unix_seq_fops
)) {
2377 unix_sysctl_unregister(net
);
2386 static void __net_exit
unix_net_exit(struct net
*net
)
2388 unix_sysctl_unregister(net
);
2389 proc_net_remove(net
, "unix");
2392 static struct pernet_operations unix_net_ops
= {
2393 .init
= unix_net_init
,
2394 .exit
= unix_net_exit
,
2397 static int __init
af_unix_init(void)
2400 struct sk_buff
*dummy_skb
;
2402 BUILD_BUG_ON(sizeof(struct unix_skb_parms
) > sizeof(dummy_skb
->cb
));
2404 rc
= proto_register(&unix_proto
, 1);
2406 printk(KERN_CRIT
"%s: Cannot create unix_sock SLAB cache!\n",
2411 sock_register(&unix_family_ops
);
2412 register_pernet_subsys(&unix_net_ops
);
2417 static void __exit
af_unix_exit(void)
2419 sock_unregister(PF_UNIX
);
2420 proto_unregister(&unix_proto
);
2421 unregister_pernet_subsys(&unix_net_ops
);
2424 /* Earlier than device_initcall() so that other drivers invoking
2425 request_module() don't end up in a loop when modprobe tries
2426 to use a UNIX socket. But later than subsys_initcall() because
2427 we depend on stuff initialised there */
2428 fs_initcall(af_unix_init
);
2429 module_exit(af_unix_exit
);
2431 MODULE_LICENSE("GPL");
2432 MODULE_ALIAS_NETPROTO(PF_UNIX
);