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 static struct hlist_head unix_socket_table
[UNIX_HASH_SIZE
+ 1];
119 static DEFINE_SPINLOCK(unix_table_lock
);
120 static atomic_t unix_nr_socks
= ATOMIC_INIT(0);
122 #define unix_sockets_unbound (&unix_socket_table[UNIX_HASH_SIZE])
124 #define UNIX_ABSTRACT(sk) (unix_sk(sk)->addr->hash != UNIX_HASH_SIZE)
126 #ifdef CONFIG_SECURITY_NETWORK
127 static void unix_get_secdata(struct scm_cookie
*scm
, struct sk_buff
*skb
)
129 memcpy(UNIXSID(skb
), &scm
->secid
, sizeof(u32
));
132 static inline void unix_set_secdata(struct scm_cookie
*scm
, struct sk_buff
*skb
)
134 scm
->secid
= *UNIXSID(skb
);
137 static inline void unix_get_secdata(struct scm_cookie
*scm
, struct sk_buff
*skb
)
140 static inline void unix_set_secdata(struct scm_cookie
*scm
, struct sk_buff
*skb
)
142 #endif /* CONFIG_SECURITY_NETWORK */
145 * SMP locking strategy:
146 * hash table is protected with spinlock unix_table_lock
147 * each socket state is protected by separate spin lock.
150 static inline unsigned unix_hash_fold(__wsum n
)
152 unsigned hash
= (__force
unsigned)n
;
155 return hash
&(UNIX_HASH_SIZE
-1);
158 #define unix_peer(sk) (unix_sk(sk)->peer)
160 static inline int unix_our_peer(struct sock
*sk
, struct sock
*osk
)
162 return unix_peer(osk
) == sk
;
165 static inline int unix_may_send(struct sock
*sk
, struct sock
*osk
)
167 return unix_peer(osk
) == NULL
|| unix_our_peer(sk
, osk
);
170 static inline int unix_recvq_full(struct sock
const *sk
)
172 return skb_queue_len(&sk
->sk_receive_queue
) > sk
->sk_max_ack_backlog
;
175 static struct sock
*unix_peer_get(struct sock
*s
)
183 unix_state_unlock(s
);
187 static inline void unix_release_addr(struct unix_address
*addr
)
189 if (atomic_dec_and_test(&addr
->refcnt
))
194 * Check unix socket name:
195 * - should be not zero length.
196 * - if started by not zero, should be NULL terminated (FS object)
197 * - if started by zero, it is abstract name.
200 static int unix_mkname(struct sockaddr_un
*sunaddr
, int len
, unsigned *hashp
)
202 if (len
<= sizeof(short) || len
> sizeof(*sunaddr
))
204 if (!sunaddr
|| sunaddr
->sun_family
!= AF_UNIX
)
206 if (sunaddr
->sun_path
[0]) {
208 * This may look like an off by one error but it is a bit more
209 * subtle. 108 is the longest valid AF_UNIX path for a binding.
210 * sun_path[108] doesnt as such exist. However in kernel space
211 * we are guaranteed that it is a valid memory location in our
212 * kernel address buffer.
214 ((char *)sunaddr
)[len
] = 0;
215 len
= strlen(sunaddr
->sun_path
)+1+sizeof(short);
219 *hashp
= unix_hash_fold(csum_partial(sunaddr
, len
, 0));
223 static void __unix_remove_socket(struct sock
*sk
)
225 sk_del_node_init(sk
);
228 static void __unix_insert_socket(struct hlist_head
*list
, struct sock
*sk
)
230 WARN_ON(!sk_unhashed(sk
));
231 sk_add_node(sk
, list
);
234 static inline void unix_remove_socket(struct sock
*sk
)
236 spin_lock(&unix_table_lock
);
237 __unix_remove_socket(sk
);
238 spin_unlock(&unix_table_lock
);
241 static inline void unix_insert_socket(struct hlist_head
*list
, struct sock
*sk
)
243 spin_lock(&unix_table_lock
);
244 __unix_insert_socket(list
, sk
);
245 spin_unlock(&unix_table_lock
);
248 static struct sock
*__unix_find_socket_byname(struct net
*net
,
249 struct sockaddr_un
*sunname
,
250 int len
, int type
, unsigned hash
)
253 struct hlist_node
*node
;
255 sk_for_each(s
, node
, &unix_socket_table
[hash
^ type
]) {
256 struct unix_sock
*u
= unix_sk(s
);
258 if (!net_eq(sock_net(s
), net
))
261 if (u
->addr
->len
== len
&&
262 !memcmp(u
->addr
->name
, sunname
, len
))
270 static inline struct sock
*unix_find_socket_byname(struct net
*net
,
271 struct sockaddr_un
*sunname
,
277 spin_lock(&unix_table_lock
);
278 s
= __unix_find_socket_byname(net
, sunname
, len
, type
, hash
);
281 spin_unlock(&unix_table_lock
);
285 static struct sock
*unix_find_socket_byinode(struct net
*net
, struct inode
*i
)
288 struct hlist_node
*node
;
290 spin_lock(&unix_table_lock
);
292 &unix_socket_table
[i
->i_ino
& (UNIX_HASH_SIZE
- 1)]) {
293 struct dentry
*dentry
= unix_sk(s
)->dentry
;
295 if (!net_eq(sock_net(s
), net
))
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(&wq
->wait
);
323 sk_wake_async(sk
, SOCK_WAKE_SPACE
, POLL_OUT
);
328 /* When dgram socket disconnects (or changes its peer), we clear its receive
329 * queue of packets arrived from previous peer. First, it allows to do
330 * flow control based only on wmem_alloc; second, sk connected to peer
331 * may receive messages only from that peer. */
332 static void unix_dgram_disconnected(struct sock
*sk
, struct sock
*other
)
334 if (!skb_queue_empty(&sk
->sk_receive_queue
)) {
335 skb_queue_purge(&sk
->sk_receive_queue
);
336 wake_up_interruptible_all(&unix_sk(sk
)->peer_wait
);
338 /* If one link of bidirectional dgram pipe is disconnected,
339 * we signal error. Messages are lost. Do not make this,
340 * when peer was not connected to us.
342 if (!sock_flag(other
, SOCK_DEAD
) && unix_peer(other
) == sk
) {
343 other
->sk_err
= ECONNRESET
;
344 other
->sk_error_report(other
);
349 static void unix_sock_destructor(struct sock
*sk
)
351 struct unix_sock
*u
= unix_sk(sk
);
353 skb_queue_purge(&sk
->sk_receive_queue
);
355 WARN_ON(atomic_read(&sk
->sk_wmem_alloc
));
356 WARN_ON(!sk_unhashed(sk
));
357 WARN_ON(sk
->sk_socket
);
358 if (!sock_flag(sk
, SOCK_DEAD
)) {
359 printk(KERN_INFO
"Attempt to release alive unix socket: %p\n", sk
);
364 unix_release_addr(u
->addr
);
366 atomic_dec(&unix_nr_socks
);
368 sock_prot_inuse_add(sock_net(sk
), sk
->sk_prot
, -1);
370 #ifdef UNIX_REFCNT_DEBUG
371 printk(KERN_DEBUG
"UNIX %p is destroyed, %d are still alive.\n", sk
,
372 atomic_read(&unix_nr_socks
));
376 static int unix_release_sock(struct sock
*sk
, int embrion
)
378 struct unix_sock
*u
= unix_sk(sk
);
379 struct dentry
*dentry
;
380 struct vfsmount
*mnt
;
385 unix_remove_socket(sk
);
390 sk
->sk_shutdown
= SHUTDOWN_MASK
;
395 state
= sk
->sk_state
;
396 sk
->sk_state
= TCP_CLOSE
;
397 unix_state_unlock(sk
);
399 wake_up_interruptible_all(&u
->peer_wait
);
401 skpair
= unix_peer(sk
);
403 if (skpair
!= NULL
) {
404 if (sk
->sk_type
== SOCK_STREAM
|| sk
->sk_type
== SOCK_SEQPACKET
) {
405 unix_state_lock(skpair
);
407 skpair
->sk_shutdown
= SHUTDOWN_MASK
;
408 if (!skb_queue_empty(&sk
->sk_receive_queue
) || embrion
)
409 skpair
->sk_err
= ECONNRESET
;
410 unix_state_unlock(skpair
);
411 skpair
->sk_state_change(skpair
);
412 sk_wake_async(skpair
, SOCK_WAKE_WAITD
, POLL_HUP
);
414 sock_put(skpair
); /* It may now die */
415 unix_peer(sk
) = NULL
;
418 /* Try to flush out this socket. Throw out buffers at least */
420 while ((skb
= skb_dequeue(&sk
->sk_receive_queue
)) != NULL
) {
421 if (state
== TCP_LISTEN
)
422 unix_release_sock(skb
->sk
, 1);
423 /* passed fds are erased in the kfree_skb hook */
434 /* ---- Socket is dead now and most probably destroyed ---- */
437 * Fixme: BSD difference: In BSD all sockets connected to use get
438 * ECONNRESET and we die on the spot. In Linux we behave
439 * like files and pipes do and wait for the last
442 * Can't we simply set sock->err?
444 * What the above comment does talk about? --ANK(980817)
447 if (unix_tot_inflight
)
448 unix_gc(); /* Garbage collect fds */
453 static void init_peercred(struct sock
*sk
)
455 put_pid(sk
->sk_peer_pid
);
456 if (sk
->sk_peer_cred
)
457 put_cred(sk
->sk_peer_cred
);
458 sk
->sk_peer_pid
= get_pid(task_tgid(current
));
459 sk
->sk_peer_cred
= get_current_cred();
462 static void copy_peercred(struct sock
*sk
, struct sock
*peersk
)
464 put_pid(sk
->sk_peer_pid
);
465 if (sk
->sk_peer_cred
)
466 put_cred(sk
->sk_peer_cred
);
467 sk
->sk_peer_pid
= get_pid(peersk
->sk_peer_pid
);
468 sk
->sk_peer_cred
= get_cred(peersk
->sk_peer_cred
);
471 static int unix_listen(struct socket
*sock
, int backlog
)
474 struct sock
*sk
= sock
->sk
;
475 struct unix_sock
*u
= unix_sk(sk
);
476 struct pid
*old_pid
= NULL
;
477 const struct cred
*old_cred
= NULL
;
480 if (sock
->type
!= SOCK_STREAM
&& sock
->type
!= SOCK_SEQPACKET
)
481 goto out
; /* Only stream/seqpacket sockets accept */
484 goto out
; /* No listens on an unbound socket */
486 if (sk
->sk_state
!= TCP_CLOSE
&& sk
->sk_state
!= TCP_LISTEN
)
488 if (backlog
> sk
->sk_max_ack_backlog
)
489 wake_up_interruptible_all(&u
->peer_wait
);
490 sk
->sk_max_ack_backlog
= backlog
;
491 sk
->sk_state
= TCP_LISTEN
;
492 /* set credentials so connect can copy them */
497 unix_state_unlock(sk
);
505 static int unix_release(struct socket
*);
506 static int unix_bind(struct socket
*, struct sockaddr
*, int);
507 static int unix_stream_connect(struct socket
*, struct sockaddr
*,
508 int addr_len
, int flags
);
509 static int unix_socketpair(struct socket
*, struct socket
*);
510 static int unix_accept(struct socket
*, struct socket
*, int);
511 static int unix_getname(struct socket
*, struct sockaddr
*, int *, int);
512 static unsigned int unix_poll(struct file
*, struct socket
*, poll_table
*);
513 static unsigned int unix_dgram_poll(struct file
*, struct socket
*,
515 static int unix_ioctl(struct socket
*, unsigned int, unsigned long);
516 static int unix_shutdown(struct socket
*, int);
517 static int unix_stream_sendmsg(struct kiocb
*, struct socket
*,
518 struct msghdr
*, size_t);
519 static int unix_stream_recvmsg(struct kiocb
*, struct socket
*,
520 struct msghdr
*, size_t, int);
521 static int unix_dgram_sendmsg(struct kiocb
*, struct socket
*,
522 struct msghdr
*, size_t);
523 static int unix_dgram_recvmsg(struct kiocb
*, struct socket
*,
524 struct msghdr
*, size_t, int);
525 static int unix_dgram_connect(struct socket
*, struct sockaddr
*,
527 static int unix_seqpacket_sendmsg(struct kiocb
*, struct socket
*,
528 struct msghdr
*, size_t);
530 static const struct proto_ops unix_stream_ops
= {
532 .owner
= THIS_MODULE
,
533 .release
= unix_release
,
535 .connect
= unix_stream_connect
,
536 .socketpair
= unix_socketpair
,
537 .accept
= unix_accept
,
538 .getname
= unix_getname
,
541 .listen
= unix_listen
,
542 .shutdown
= unix_shutdown
,
543 .setsockopt
= sock_no_setsockopt
,
544 .getsockopt
= sock_no_getsockopt
,
545 .sendmsg
= unix_stream_sendmsg
,
546 .recvmsg
= unix_stream_recvmsg
,
547 .mmap
= sock_no_mmap
,
548 .sendpage
= sock_no_sendpage
,
551 static const struct proto_ops unix_dgram_ops
= {
553 .owner
= THIS_MODULE
,
554 .release
= unix_release
,
556 .connect
= unix_dgram_connect
,
557 .socketpair
= unix_socketpair
,
558 .accept
= sock_no_accept
,
559 .getname
= unix_getname
,
560 .poll
= unix_dgram_poll
,
562 .listen
= sock_no_listen
,
563 .shutdown
= unix_shutdown
,
564 .setsockopt
= sock_no_setsockopt
,
565 .getsockopt
= sock_no_getsockopt
,
566 .sendmsg
= unix_dgram_sendmsg
,
567 .recvmsg
= unix_dgram_recvmsg
,
568 .mmap
= sock_no_mmap
,
569 .sendpage
= sock_no_sendpage
,
572 static const struct proto_ops unix_seqpacket_ops
= {
574 .owner
= THIS_MODULE
,
575 .release
= unix_release
,
577 .connect
= unix_stream_connect
,
578 .socketpair
= unix_socketpair
,
579 .accept
= unix_accept
,
580 .getname
= unix_getname
,
581 .poll
= unix_dgram_poll
,
583 .listen
= unix_listen
,
584 .shutdown
= unix_shutdown
,
585 .setsockopt
= sock_no_setsockopt
,
586 .getsockopt
= sock_no_getsockopt
,
587 .sendmsg
= unix_seqpacket_sendmsg
,
588 .recvmsg
= unix_dgram_recvmsg
,
589 .mmap
= sock_no_mmap
,
590 .sendpage
= sock_no_sendpage
,
593 static struct proto unix_proto
= {
595 .owner
= THIS_MODULE
,
596 .obj_size
= sizeof(struct unix_sock
),
600 * AF_UNIX sockets do not interact with hardware, hence they
601 * dont trigger interrupts - so it's safe for them to have
602 * bh-unsafe locking for their sk_receive_queue.lock. Split off
603 * this special lock-class by reinitializing the spinlock key:
605 static struct lock_class_key af_unix_sk_receive_queue_lock_key
;
607 static struct sock
*unix_create1(struct net
*net
, struct socket
*sock
)
609 struct sock
*sk
= NULL
;
612 atomic_inc(&unix_nr_socks
);
613 if (atomic_read(&unix_nr_socks
) > 2 * get_max_files())
616 sk
= sk_alloc(net
, PF_UNIX
, GFP_KERNEL
, &unix_proto
);
620 sock_init_data(sock
, sk
);
621 lockdep_set_class(&sk
->sk_receive_queue
.lock
,
622 &af_unix_sk_receive_queue_lock_key
);
624 sk
->sk_write_space
= unix_write_space
;
625 sk
->sk_max_ack_backlog
= net
->unx
.sysctl_max_dgram_qlen
;
626 sk
->sk_destruct
= unix_sock_destructor
;
630 spin_lock_init(&u
->lock
);
631 atomic_long_set(&u
->inflight
, 0);
632 INIT_LIST_HEAD(&u
->link
);
633 mutex_init(&u
->readlock
); /* single task reading lock */
634 init_waitqueue_head(&u
->peer_wait
);
635 unix_insert_socket(unix_sockets_unbound
, sk
);
638 atomic_dec(&unix_nr_socks
);
641 sock_prot_inuse_add(sock_net(sk
), sk
->sk_prot
, 1);
647 static int unix_create(struct net
*net
, struct socket
*sock
, int protocol
,
650 if (protocol
&& protocol
!= PF_UNIX
)
651 return -EPROTONOSUPPORT
;
653 sock
->state
= SS_UNCONNECTED
;
655 switch (sock
->type
) {
657 sock
->ops
= &unix_stream_ops
;
660 * Believe it or not BSD has AF_UNIX, SOCK_RAW though
664 sock
->type
= SOCK_DGRAM
;
666 sock
->ops
= &unix_dgram_ops
;
669 sock
->ops
= &unix_seqpacket_ops
;
672 return -ESOCKTNOSUPPORT
;
675 return unix_create1(net
, sock
) ? 0 : -ENOMEM
;
678 static int unix_release(struct socket
*sock
)
680 struct sock
*sk
= sock
->sk
;
687 return unix_release_sock(sk
, 0);
690 static int unix_autobind(struct socket
*sock
)
692 struct sock
*sk
= sock
->sk
;
693 struct net
*net
= sock_net(sk
);
694 struct unix_sock
*u
= unix_sk(sk
);
695 static u32 ordernum
= 1;
696 struct unix_address
*addr
;
699 mutex_lock(&u
->readlock
);
706 addr
= kzalloc(sizeof(*addr
) + sizeof(short) + 16, GFP_KERNEL
);
710 addr
->name
->sun_family
= AF_UNIX
;
711 atomic_set(&addr
->refcnt
, 1);
714 addr
->len
= sprintf(addr
->name
->sun_path
+1, "%05x", ordernum
) + 1 + sizeof(short);
715 addr
->hash
= unix_hash_fold(csum_partial(addr
->name
, addr
->len
, 0));
717 spin_lock(&unix_table_lock
);
718 ordernum
= (ordernum
+1)&0xFFFFF;
720 if (__unix_find_socket_byname(net
, addr
->name
, addr
->len
, sock
->type
,
722 spin_unlock(&unix_table_lock
);
723 /* Sanity yield. It is unusual case, but yet... */
724 if (!(ordernum
&0xFF))
728 addr
->hash
^= sk
->sk_type
;
730 __unix_remove_socket(sk
);
732 __unix_insert_socket(&unix_socket_table
[addr
->hash
], sk
);
733 spin_unlock(&unix_table_lock
);
736 out
: mutex_unlock(&u
->readlock
);
740 static struct sock
*unix_find_other(struct net
*net
,
741 struct sockaddr_un
*sunname
, int len
,
742 int type
, unsigned hash
, int *error
)
748 if (sunname
->sun_path
[0]) {
750 err
= kern_path(sunname
->sun_path
, LOOKUP_FOLLOW
, &path
);
753 inode
= path
.dentry
->d_inode
;
754 err
= inode_permission(inode
, MAY_WRITE
);
759 if (!S_ISSOCK(inode
->i_mode
))
761 u
= unix_find_socket_byinode(net
, inode
);
765 if (u
->sk_type
== type
)
766 touch_atime(path
.mnt
, path
.dentry
);
771 if (u
->sk_type
!= type
) {
777 u
= unix_find_socket_byname(net
, sunname
, len
, type
, hash
);
779 struct dentry
*dentry
;
780 dentry
= unix_sk(u
)->dentry
;
782 touch_atime(unix_sk(u
)->mnt
, dentry
);
796 static int unix_bind(struct socket
*sock
, struct sockaddr
*uaddr
, int addr_len
)
798 struct sock
*sk
= sock
->sk
;
799 struct net
*net
= sock_net(sk
);
800 struct unix_sock
*u
= unix_sk(sk
);
801 struct sockaddr_un
*sunaddr
= (struct sockaddr_un
*)uaddr
;
802 struct dentry
*dentry
= NULL
;
806 struct unix_address
*addr
;
807 struct hlist_head
*list
;
810 if (sunaddr
->sun_family
!= AF_UNIX
)
813 if (addr_len
== sizeof(short)) {
814 err
= unix_autobind(sock
);
818 err
= unix_mkname(sunaddr
, addr_len
, &hash
);
823 mutex_lock(&u
->readlock
);
830 addr
= kmalloc(sizeof(*addr
)+addr_len
, GFP_KERNEL
);
834 memcpy(addr
->name
, sunaddr
, addr_len
);
835 addr
->len
= addr_len
;
836 addr
->hash
= hash
^ sk
->sk_type
;
837 atomic_set(&addr
->refcnt
, 1);
839 if (sunaddr
->sun_path
[0]) {
843 * Get the parent directory, calculate the hash for last
846 err
= path_lookup(sunaddr
->sun_path
, LOOKUP_PARENT
, &nd
);
848 goto out_mknod_parent
;
850 dentry
= lookup_create(&nd
, 0);
851 err
= PTR_ERR(dentry
);
853 goto out_mknod_unlock
;
856 * All right, let's create it.
859 (SOCK_INODE(sock
)->i_mode
& ~current_umask());
860 err
= mnt_want_write(nd
.path
.mnt
);
863 err
= security_path_mknod(&nd
.path
, dentry
, mode
, 0);
865 goto out_mknod_drop_write
;
866 err
= vfs_mknod(nd
.path
.dentry
->d_inode
, dentry
, mode
, 0);
867 out_mknod_drop_write
:
868 mnt_drop_write(nd
.path
.mnt
);
871 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
872 dput(nd
.path
.dentry
);
873 nd
.path
.dentry
= dentry
;
875 addr
->hash
= UNIX_HASH_SIZE
;
878 spin_lock(&unix_table_lock
);
880 if (!sunaddr
->sun_path
[0]) {
882 if (__unix_find_socket_byname(net
, sunaddr
, addr_len
,
883 sk
->sk_type
, hash
)) {
884 unix_release_addr(addr
);
888 list
= &unix_socket_table
[addr
->hash
];
890 list
= &unix_socket_table
[dentry
->d_inode
->i_ino
& (UNIX_HASH_SIZE
-1)];
891 u
->dentry
= nd
.path
.dentry
;
892 u
->mnt
= nd
.path
.mnt
;
896 __unix_remove_socket(sk
);
898 __unix_insert_socket(list
, sk
);
901 spin_unlock(&unix_table_lock
);
903 mutex_unlock(&u
->readlock
);
910 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
915 unix_release_addr(addr
);
919 static void unix_state_double_lock(struct sock
*sk1
, struct sock
*sk2
)
921 if (unlikely(sk1
== sk2
) || !sk2
) {
922 unix_state_lock(sk1
);
926 unix_state_lock(sk1
);
927 unix_state_lock_nested(sk2
);
929 unix_state_lock(sk2
);
930 unix_state_lock_nested(sk1
);
934 static void unix_state_double_unlock(struct sock
*sk1
, struct sock
*sk2
)
936 if (unlikely(sk1
== sk2
) || !sk2
) {
937 unix_state_unlock(sk1
);
940 unix_state_unlock(sk1
);
941 unix_state_unlock(sk2
);
944 static int unix_dgram_connect(struct socket
*sock
, struct sockaddr
*addr
,
947 struct sock
*sk
= sock
->sk
;
948 struct net
*net
= sock_net(sk
);
949 struct sockaddr_un
*sunaddr
= (struct sockaddr_un
*)addr
;
954 if (addr
->sa_family
!= AF_UNSPEC
) {
955 err
= unix_mkname(sunaddr
, alen
, &hash
);
960 if (test_bit(SOCK_PASSCRED
, &sock
->flags
) &&
961 !unix_sk(sk
)->addr
&& (err
= unix_autobind(sock
)) != 0)
965 other
= unix_find_other(net
, sunaddr
, alen
, sock
->type
, hash
, &err
);
969 unix_state_double_lock(sk
, other
);
971 /* Apparently VFS overslept socket death. Retry. */
972 if (sock_flag(other
, SOCK_DEAD
)) {
973 unix_state_double_unlock(sk
, other
);
979 if (!unix_may_send(sk
, other
))
982 err
= security_unix_may_send(sk
->sk_socket
, other
->sk_socket
);
988 * 1003.1g breaking connected state with AF_UNSPEC
991 unix_state_double_lock(sk
, other
);
995 * If it was connected, reconnect.
998 struct sock
*old_peer
= unix_peer(sk
);
999 unix_peer(sk
) = other
;
1000 unix_state_double_unlock(sk
, other
);
1002 if (other
!= old_peer
)
1003 unix_dgram_disconnected(sk
, old_peer
);
1006 unix_peer(sk
) = other
;
1007 unix_state_double_unlock(sk
, other
);
1012 unix_state_double_unlock(sk
, other
);
1018 static long unix_wait_for_peer(struct sock
*other
, long timeo
)
1020 struct unix_sock
*u
= unix_sk(other
);
1024 prepare_to_wait_exclusive(&u
->peer_wait
, &wait
, TASK_INTERRUPTIBLE
);
1026 sched
= !sock_flag(other
, SOCK_DEAD
) &&
1027 !(other
->sk_shutdown
& RCV_SHUTDOWN
) &&
1028 unix_recvq_full(other
);
1030 unix_state_unlock(other
);
1033 timeo
= schedule_timeout(timeo
);
1035 finish_wait(&u
->peer_wait
, &wait
);
1039 static int unix_stream_connect(struct socket
*sock
, struct sockaddr
*uaddr
,
1040 int addr_len
, int flags
)
1042 struct sockaddr_un
*sunaddr
= (struct sockaddr_un
*)uaddr
;
1043 struct sock
*sk
= sock
->sk
;
1044 struct net
*net
= sock_net(sk
);
1045 struct unix_sock
*u
= unix_sk(sk
), *newu
, *otheru
;
1046 struct sock
*newsk
= NULL
;
1047 struct sock
*other
= NULL
;
1048 struct sk_buff
*skb
= NULL
;
1054 err
= unix_mkname(sunaddr
, addr_len
, &hash
);
1059 if (test_bit(SOCK_PASSCRED
, &sock
->flags
) && !u
->addr
&&
1060 (err
= unix_autobind(sock
)) != 0)
1063 timeo
= sock_sndtimeo(sk
, flags
& O_NONBLOCK
);
1065 /* First of all allocate resources.
1066 If we will make it after state is locked,
1067 we will have to recheck all again in any case.
1072 /* create new sock for complete connection */
1073 newsk
= unix_create1(sock_net(sk
), NULL
);
1077 /* Allocate skb for sending to listening sock */
1078 skb
= sock_wmalloc(newsk
, 1, 0, GFP_KERNEL
);
1083 /* Find listening sock. */
1084 other
= unix_find_other(net
, sunaddr
, addr_len
, sk
->sk_type
, hash
, &err
);
1088 /* Latch state of peer */
1089 unix_state_lock(other
);
1091 /* Apparently VFS overslept socket death. Retry. */
1092 if (sock_flag(other
, SOCK_DEAD
)) {
1093 unix_state_unlock(other
);
1098 err
= -ECONNREFUSED
;
1099 if (other
->sk_state
!= TCP_LISTEN
)
1101 if (other
->sk_shutdown
& RCV_SHUTDOWN
)
1104 if (unix_recvq_full(other
)) {
1109 timeo
= unix_wait_for_peer(other
, timeo
);
1111 err
= sock_intr_errno(timeo
);
1112 if (signal_pending(current
))
1120 It is tricky place. We need to grab write lock and cannot
1121 drop lock on peer. It is dangerous because deadlock is
1122 possible. Connect to self case and simultaneous
1123 attempt to connect are eliminated by checking socket
1124 state. other is TCP_LISTEN, if sk is TCP_LISTEN we
1125 check this before attempt to grab lock.
1127 Well, and we have to recheck the state after socket locked.
1133 /* This is ok... continue with connect */
1135 case TCP_ESTABLISHED
:
1136 /* Socket is already connected */
1144 unix_state_lock_nested(sk
);
1146 if (sk
->sk_state
!= st
) {
1147 unix_state_unlock(sk
);
1148 unix_state_unlock(other
);
1153 err
= security_unix_stream_connect(sock
, other
->sk_socket
, newsk
);
1155 unix_state_unlock(sk
);
1159 /* The way is open! Fastly set all the necessary fields... */
1162 unix_peer(newsk
) = sk
;
1163 newsk
->sk_state
= TCP_ESTABLISHED
;
1164 newsk
->sk_type
= sk
->sk_type
;
1165 init_peercred(newsk
);
1166 newu
= unix_sk(newsk
);
1167 newsk
->sk_wq
= &newu
->peer_wq
;
1168 otheru
= unix_sk(other
);
1170 /* copy address information from listening to new sock*/
1172 atomic_inc(&otheru
->addr
->refcnt
);
1173 newu
->addr
= otheru
->addr
;
1175 if (otheru
->dentry
) {
1176 newu
->dentry
= dget(otheru
->dentry
);
1177 newu
->mnt
= mntget(otheru
->mnt
);
1180 /* Set credentials */
1181 copy_peercred(sk
, other
);
1183 sock
->state
= SS_CONNECTED
;
1184 sk
->sk_state
= TCP_ESTABLISHED
;
1187 smp_mb__after_atomic_inc(); /* sock_hold() does an atomic_inc() */
1188 unix_peer(sk
) = newsk
;
1190 unix_state_unlock(sk
);
1192 /* take ten and and send info to listening sock */
1193 spin_lock(&other
->sk_receive_queue
.lock
);
1194 __skb_queue_tail(&other
->sk_receive_queue
, skb
);
1195 spin_unlock(&other
->sk_receive_queue
.lock
);
1196 unix_state_unlock(other
);
1197 other
->sk_data_ready(other
, 0);
1203 unix_state_unlock(other
);
1208 unix_release_sock(newsk
, 0);
1214 static int unix_socketpair(struct socket
*socka
, struct socket
*sockb
)
1216 struct sock
*ska
= socka
->sk
, *skb
= sockb
->sk
;
1218 /* Join our sockets back to back */
1221 unix_peer(ska
) = skb
;
1222 unix_peer(skb
) = ska
;
1226 if (ska
->sk_type
!= SOCK_DGRAM
) {
1227 ska
->sk_state
= TCP_ESTABLISHED
;
1228 skb
->sk_state
= TCP_ESTABLISHED
;
1229 socka
->state
= SS_CONNECTED
;
1230 sockb
->state
= SS_CONNECTED
;
1235 static int unix_accept(struct socket
*sock
, struct socket
*newsock
, int flags
)
1237 struct sock
*sk
= sock
->sk
;
1239 struct sk_buff
*skb
;
1243 if (sock
->type
!= SOCK_STREAM
&& sock
->type
!= SOCK_SEQPACKET
)
1247 if (sk
->sk_state
!= TCP_LISTEN
)
1250 /* If socket state is TCP_LISTEN it cannot change (for now...),
1251 * so that no locks are necessary.
1254 skb
= skb_recv_datagram(sk
, 0, flags
&O_NONBLOCK
, &err
);
1256 /* This means receive shutdown. */
1263 skb_free_datagram(sk
, skb
);
1264 wake_up_interruptible(&unix_sk(sk
)->peer_wait
);
1266 /* attach accepted sock to socket */
1267 unix_state_lock(tsk
);
1268 newsock
->state
= SS_CONNECTED
;
1269 sock_graft(tsk
, newsock
);
1270 unix_state_unlock(tsk
);
1278 static int unix_getname(struct socket
*sock
, struct sockaddr
*uaddr
, int *uaddr_len
, int peer
)
1280 struct sock
*sk
= sock
->sk
;
1281 struct unix_sock
*u
;
1282 DECLARE_SOCKADDR(struct sockaddr_un
*, sunaddr
, uaddr
);
1286 sk
= unix_peer_get(sk
);
1297 unix_state_lock(sk
);
1299 sunaddr
->sun_family
= AF_UNIX
;
1300 sunaddr
->sun_path
[0] = 0;
1301 *uaddr_len
= sizeof(short);
1303 struct unix_address
*addr
= u
->addr
;
1305 *uaddr_len
= addr
->len
;
1306 memcpy(sunaddr
, addr
->name
, *uaddr_len
);
1308 unix_state_unlock(sk
);
1314 static void unix_detach_fds(struct scm_cookie
*scm
, struct sk_buff
*skb
)
1318 scm
->fp
= UNIXCB(skb
).fp
;
1319 UNIXCB(skb
).fp
= NULL
;
1321 for (i
= scm
->fp
->count
-1; i
>= 0; i
--)
1322 unix_notinflight(scm
->fp
->fp
[i
]);
1325 static void unix_destruct_scm(struct sk_buff
*skb
)
1327 struct scm_cookie scm
;
1328 memset(&scm
, 0, sizeof(scm
));
1329 scm
.pid
= UNIXCB(skb
).pid
;
1330 scm
.cred
= UNIXCB(skb
).cred
;
1332 unix_detach_fds(&scm
, skb
);
1334 /* Alas, it calls VFS */
1335 /* So fscking what? fput() had been SMP-safe since the last Summer */
1340 static int unix_attach_fds(struct scm_cookie
*scm
, struct sk_buff
*skb
)
1345 * Need to duplicate file references for the sake of garbage
1346 * collection. Otherwise a socket in the fps might become a
1347 * candidate for GC while the skb is not yet queued.
1349 UNIXCB(skb
).fp
= scm_fp_dup(scm
->fp
);
1350 if (!UNIXCB(skb
).fp
)
1353 for (i
= scm
->fp
->count
-1; i
>= 0; i
--)
1354 unix_inflight(scm
->fp
->fp
[i
]);
1358 static int unix_scm_to_skb(struct scm_cookie
*scm
, struct sk_buff
*skb
, bool send_fds
)
1361 UNIXCB(skb
).pid
= get_pid(scm
->pid
);
1362 UNIXCB(skb
).cred
= get_cred(scm
->cred
);
1363 UNIXCB(skb
).fp
= NULL
;
1364 if (scm
->fp
&& send_fds
)
1365 err
= unix_attach_fds(scm
, skb
);
1367 skb
->destructor
= unix_destruct_scm
;
1372 * Send AF_UNIX data.
1375 static int unix_dgram_sendmsg(struct kiocb
*kiocb
, struct socket
*sock
,
1376 struct msghdr
*msg
, size_t len
)
1378 struct sock_iocb
*siocb
= kiocb_to_siocb(kiocb
);
1379 struct sock
*sk
= sock
->sk
;
1380 struct net
*net
= sock_net(sk
);
1381 struct unix_sock
*u
= unix_sk(sk
);
1382 struct sockaddr_un
*sunaddr
= msg
->msg_name
;
1383 struct sock
*other
= NULL
;
1384 int namelen
= 0; /* fake GCC */
1387 struct sk_buff
*skb
;
1389 struct scm_cookie tmp_scm
;
1391 if (NULL
== siocb
->scm
)
1392 siocb
->scm
= &tmp_scm
;
1394 err
= scm_send(sock
, msg
, siocb
->scm
);
1399 if (msg
->msg_flags
&MSG_OOB
)
1402 if (msg
->msg_namelen
) {
1403 err
= unix_mkname(sunaddr
, msg
->msg_namelen
, &hash
);
1410 other
= unix_peer_get(sk
);
1415 if (test_bit(SOCK_PASSCRED
, &sock
->flags
) && !u
->addr
1416 && (err
= unix_autobind(sock
)) != 0)
1420 if (len
> sk
->sk_sndbuf
- 32)
1423 skb
= sock_alloc_send_skb(sk
, len
, msg
->msg_flags
&MSG_DONTWAIT
, &err
);
1427 err
= unix_scm_to_skb(siocb
->scm
, skb
, true);
1430 unix_get_secdata(siocb
->scm
, skb
);
1432 skb_reset_transport_header(skb
);
1433 err
= memcpy_fromiovec(skb_put(skb
, len
), msg
->msg_iov
, len
);
1437 timeo
= sock_sndtimeo(sk
, msg
->msg_flags
& MSG_DONTWAIT
);
1442 if (sunaddr
== NULL
)
1445 other
= unix_find_other(net
, sunaddr
, namelen
, sk
->sk_type
,
1451 unix_state_lock(other
);
1453 if (!unix_may_send(sk
, other
))
1456 if (sock_flag(other
, SOCK_DEAD
)) {
1458 * Check with 1003.1g - what should
1461 unix_state_unlock(other
);
1465 unix_state_lock(sk
);
1466 if (unix_peer(sk
) == other
) {
1467 unix_peer(sk
) = NULL
;
1468 unix_state_unlock(sk
);
1470 unix_dgram_disconnected(sk
, other
);
1472 err
= -ECONNREFUSED
;
1474 unix_state_unlock(sk
);
1484 if (other
->sk_shutdown
& RCV_SHUTDOWN
)
1487 if (sk
->sk_type
!= SOCK_SEQPACKET
) {
1488 err
= security_unix_may_send(sk
->sk_socket
, other
->sk_socket
);
1493 if (unix_peer(other
) != sk
&& unix_recvq_full(other
)) {
1499 timeo
= unix_wait_for_peer(other
, timeo
);
1501 err
= sock_intr_errno(timeo
);
1502 if (signal_pending(current
))
1508 skb_queue_tail(&other
->sk_receive_queue
, skb
);
1509 unix_state_unlock(other
);
1510 other
->sk_data_ready(other
, len
);
1512 scm_destroy(siocb
->scm
);
1516 unix_state_unlock(other
);
1522 scm_destroy(siocb
->scm
);
1527 static int unix_stream_sendmsg(struct kiocb
*kiocb
, struct socket
*sock
,
1528 struct msghdr
*msg
, size_t len
)
1530 struct sock_iocb
*siocb
= kiocb_to_siocb(kiocb
);
1531 struct sock
*sk
= sock
->sk
;
1532 struct sock
*other
= NULL
;
1533 struct sockaddr_un
*sunaddr
= msg
->msg_name
;
1535 struct sk_buff
*skb
;
1537 struct scm_cookie tmp_scm
;
1538 bool fds_sent
= false;
1540 if (NULL
== siocb
->scm
)
1541 siocb
->scm
= &tmp_scm
;
1543 err
= scm_send(sock
, msg
, siocb
->scm
);
1548 if (msg
->msg_flags
&MSG_OOB
)
1551 if (msg
->msg_namelen
) {
1552 err
= sk
->sk_state
== TCP_ESTABLISHED
? -EISCONN
: -EOPNOTSUPP
;
1557 other
= unix_peer(sk
);
1562 if (sk
->sk_shutdown
& SEND_SHUTDOWN
)
1565 while (sent
< len
) {
1567 * Optimisation for the fact that under 0.01% of X
1568 * messages typically need breaking up.
1573 /* Keep two messages in the pipe so it schedules better */
1574 if (size
> ((sk
->sk_sndbuf
>> 1) - 64))
1575 size
= (sk
->sk_sndbuf
>> 1) - 64;
1577 if (size
> SKB_MAX_ALLOC
)
1578 size
= SKB_MAX_ALLOC
;
1584 skb
= sock_alloc_send_skb(sk
, size
, msg
->msg_flags
&MSG_DONTWAIT
,
1591 * If you pass two values to the sock_alloc_send_skb
1592 * it tries to grab the large buffer with GFP_NOFS
1593 * (which can fail easily), and if it fails grab the
1594 * fallback size buffer which is under a page and will
1597 size
= min_t(int, size
, skb_tailroom(skb
));
1600 /* Only send the fds in the first buffer */
1601 err
= unix_scm_to_skb(siocb
->scm
, skb
, !fds_sent
);
1608 err
= memcpy_fromiovec(skb_put(skb
, size
), msg
->msg_iov
, size
);
1614 unix_state_lock(other
);
1616 if (sock_flag(other
, SOCK_DEAD
) ||
1617 (other
->sk_shutdown
& RCV_SHUTDOWN
))
1620 skb_queue_tail(&other
->sk_receive_queue
, skb
);
1621 unix_state_unlock(other
);
1622 other
->sk_data_ready(other
, size
);
1626 scm_destroy(siocb
->scm
);
1632 unix_state_unlock(other
);
1635 if (sent
== 0 && !(msg
->msg_flags
&MSG_NOSIGNAL
))
1636 send_sig(SIGPIPE
, current
, 0);
1639 scm_destroy(siocb
->scm
);
1641 return sent
? : err
;
1644 static int unix_seqpacket_sendmsg(struct kiocb
*kiocb
, struct socket
*sock
,
1645 struct msghdr
*msg
, size_t len
)
1648 struct sock
*sk
= sock
->sk
;
1650 err
= sock_error(sk
);
1654 if (sk
->sk_state
!= TCP_ESTABLISHED
)
1657 if (msg
->msg_namelen
)
1658 msg
->msg_namelen
= 0;
1660 return unix_dgram_sendmsg(kiocb
, sock
, msg
, len
);
1663 static void unix_copy_addr(struct msghdr
*msg
, struct sock
*sk
)
1665 struct unix_sock
*u
= unix_sk(sk
);
1667 msg
->msg_namelen
= 0;
1669 msg
->msg_namelen
= u
->addr
->len
;
1670 memcpy(msg
->msg_name
, u
->addr
->name
, u
->addr
->len
);
1674 static int unix_dgram_recvmsg(struct kiocb
*iocb
, struct socket
*sock
,
1675 struct msghdr
*msg
, size_t size
,
1678 struct sock_iocb
*siocb
= kiocb_to_siocb(iocb
);
1679 struct scm_cookie tmp_scm
;
1680 struct sock
*sk
= sock
->sk
;
1681 struct unix_sock
*u
= unix_sk(sk
);
1682 int noblock
= flags
& MSG_DONTWAIT
;
1683 struct sk_buff
*skb
;
1690 msg
->msg_namelen
= 0;
1692 mutex_lock(&u
->readlock
);
1694 skb
= skb_recv_datagram(sk
, flags
, noblock
, &err
);
1696 unix_state_lock(sk
);
1697 /* Signal EOF on disconnected non-blocking SEQPACKET socket. */
1698 if (sk
->sk_type
== SOCK_SEQPACKET
&& err
== -EAGAIN
&&
1699 (sk
->sk_shutdown
& RCV_SHUTDOWN
))
1701 unix_state_unlock(sk
);
1705 wake_up_interruptible_sync(&u
->peer_wait
);
1708 unix_copy_addr(msg
, skb
->sk
);
1710 if (size
> skb
->len
)
1712 else if (size
< skb
->len
)
1713 msg
->msg_flags
|= MSG_TRUNC
;
1715 err
= skb_copy_datagram_iovec(skb
, 0, msg
->msg_iov
, size
);
1720 siocb
->scm
= &tmp_scm
;
1721 memset(&tmp_scm
, 0, sizeof(tmp_scm
));
1723 scm_set_cred(siocb
->scm
, UNIXCB(skb
).pid
, UNIXCB(skb
).cred
);
1724 unix_set_secdata(siocb
->scm
, skb
);
1726 if (!(flags
& MSG_PEEK
)) {
1728 unix_detach_fds(siocb
->scm
, skb
);
1730 /* It is questionable: on PEEK we could:
1731 - do not return fds - good, but too simple 8)
1732 - return fds, and do not return them on read (old strategy,
1734 - clone fds (I chose it for now, it is the most universal
1737 POSIX 1003.1g does not actually define this clearly
1738 at all. POSIX 1003.1g doesn't define a lot of things
1743 siocb
->scm
->fp
= scm_fp_dup(UNIXCB(skb
).fp
);
1747 scm_recv(sock
, msg
, siocb
->scm
, flags
);
1750 skb_free_datagram(sk
, skb
);
1752 mutex_unlock(&u
->readlock
);
1758 * Sleep until data has arrive. But check for races..
1761 static long unix_stream_data_wait(struct sock
*sk
, long timeo
)
1765 unix_state_lock(sk
);
1768 prepare_to_wait(sk_sleep(sk
), &wait
, TASK_INTERRUPTIBLE
);
1770 if (!skb_queue_empty(&sk
->sk_receive_queue
) ||
1772 (sk
->sk_shutdown
& RCV_SHUTDOWN
) ||
1773 signal_pending(current
) ||
1777 set_bit(SOCK_ASYNC_WAITDATA
, &sk
->sk_socket
->flags
);
1778 unix_state_unlock(sk
);
1779 timeo
= schedule_timeout(timeo
);
1780 unix_state_lock(sk
);
1781 clear_bit(SOCK_ASYNC_WAITDATA
, &sk
->sk_socket
->flags
);
1784 finish_wait(sk_sleep(sk
), &wait
);
1785 unix_state_unlock(sk
);
1791 static int unix_stream_recvmsg(struct kiocb
*iocb
, struct socket
*sock
,
1792 struct msghdr
*msg
, size_t size
,
1795 struct sock_iocb
*siocb
= kiocb_to_siocb(iocb
);
1796 struct scm_cookie tmp_scm
;
1797 struct sock
*sk
= sock
->sk
;
1798 struct unix_sock
*u
= unix_sk(sk
);
1799 struct sockaddr_un
*sunaddr
= msg
->msg_name
;
1801 int check_creds
= 0;
1807 if (sk
->sk_state
!= TCP_ESTABLISHED
)
1814 target
= sock_rcvlowat(sk
, flags
&MSG_WAITALL
, size
);
1815 timeo
= sock_rcvtimeo(sk
, flags
&MSG_DONTWAIT
);
1817 msg
->msg_namelen
= 0;
1819 /* Lock the socket to prevent queue disordering
1820 * while sleeps in memcpy_tomsg
1824 siocb
->scm
= &tmp_scm
;
1825 memset(&tmp_scm
, 0, sizeof(tmp_scm
));
1828 mutex_lock(&u
->readlock
);
1832 struct sk_buff
*skb
;
1834 unix_state_lock(sk
);
1835 skb
= skb_dequeue(&sk
->sk_receive_queue
);
1837 if (copied
>= target
)
1841 * POSIX 1003.1g mandates this order.
1844 err
= sock_error(sk
);
1847 if (sk
->sk_shutdown
& RCV_SHUTDOWN
)
1850 unix_state_unlock(sk
);
1854 mutex_unlock(&u
->readlock
);
1856 timeo
= unix_stream_data_wait(sk
, timeo
);
1858 if (signal_pending(current
)) {
1859 err
= sock_intr_errno(timeo
);
1862 mutex_lock(&u
->readlock
);
1865 unix_state_unlock(sk
);
1868 unix_state_unlock(sk
);
1871 /* Never glue messages from different writers */
1872 if ((UNIXCB(skb
).pid
!= siocb
->scm
->pid
) ||
1873 (UNIXCB(skb
).cred
!= siocb
->scm
->cred
)) {
1874 skb_queue_head(&sk
->sk_receive_queue
, skb
);
1878 /* Copy credentials */
1879 scm_set_cred(siocb
->scm
, UNIXCB(skb
).pid
, UNIXCB(skb
).cred
);
1883 /* Copy address just once */
1885 unix_copy_addr(msg
, skb
->sk
);
1889 chunk
= min_t(unsigned int, skb
->len
, size
);
1890 if (memcpy_toiovec(msg
->msg_iov
, skb
->data
, chunk
)) {
1891 skb_queue_head(&sk
->sk_receive_queue
, skb
);
1899 /* Mark read part of skb as used */
1900 if (!(flags
& MSG_PEEK
)) {
1901 skb_pull(skb
, chunk
);
1904 unix_detach_fds(siocb
->scm
, skb
);
1906 /* put the skb back if we didn't use it up.. */
1908 skb_queue_head(&sk
->sk_receive_queue
, skb
);
1917 /* It is questionable, see note in unix_dgram_recvmsg.
1920 siocb
->scm
->fp
= scm_fp_dup(UNIXCB(skb
).fp
);
1922 /* put message back and return */
1923 skb_queue_head(&sk
->sk_receive_queue
, skb
);
1928 mutex_unlock(&u
->readlock
);
1929 scm_recv(sock
, msg
, siocb
->scm
, flags
);
1931 return copied
? : err
;
1934 static int unix_shutdown(struct socket
*sock
, int mode
)
1936 struct sock
*sk
= sock
->sk
;
1939 mode
= (mode
+1)&(RCV_SHUTDOWN
|SEND_SHUTDOWN
);
1942 unix_state_lock(sk
);
1943 sk
->sk_shutdown
|= mode
;
1944 other
= unix_peer(sk
);
1947 unix_state_unlock(sk
);
1948 sk
->sk_state_change(sk
);
1951 (sk
->sk_type
== SOCK_STREAM
|| sk
->sk_type
== SOCK_SEQPACKET
)) {
1955 if (mode
&RCV_SHUTDOWN
)
1956 peer_mode
|= SEND_SHUTDOWN
;
1957 if (mode
&SEND_SHUTDOWN
)
1958 peer_mode
|= RCV_SHUTDOWN
;
1959 unix_state_lock(other
);
1960 other
->sk_shutdown
|= peer_mode
;
1961 unix_state_unlock(other
);
1962 other
->sk_state_change(other
);
1963 if (peer_mode
== SHUTDOWN_MASK
)
1964 sk_wake_async(other
, SOCK_WAKE_WAITD
, POLL_HUP
);
1965 else if (peer_mode
& RCV_SHUTDOWN
)
1966 sk_wake_async(other
, SOCK_WAKE_WAITD
, POLL_IN
);
1974 static int unix_ioctl(struct socket
*sock
, unsigned int cmd
, unsigned long arg
)
1976 struct sock
*sk
= sock
->sk
;
1982 amount
= sk_wmem_alloc_get(sk
);
1983 err
= put_user(amount
, (int __user
*)arg
);
1987 struct sk_buff
*skb
;
1989 if (sk
->sk_state
== TCP_LISTEN
) {
1994 spin_lock(&sk
->sk_receive_queue
.lock
);
1995 if (sk
->sk_type
== SOCK_STREAM
||
1996 sk
->sk_type
== SOCK_SEQPACKET
) {
1997 skb_queue_walk(&sk
->sk_receive_queue
, skb
)
2000 skb
= skb_peek(&sk
->sk_receive_queue
);
2004 spin_unlock(&sk
->sk_receive_queue
.lock
);
2005 err
= put_user(amount
, (int __user
*)arg
);
2016 static unsigned int unix_poll(struct file
*file
, struct socket
*sock
, poll_table
*wait
)
2018 struct sock
*sk
= sock
->sk
;
2021 sock_poll_wait(file
, sk_sleep(sk
), wait
);
2024 /* exceptional events? */
2027 if (sk
->sk_shutdown
== SHUTDOWN_MASK
)
2029 if (sk
->sk_shutdown
& RCV_SHUTDOWN
)
2033 if (!skb_queue_empty(&sk
->sk_receive_queue
) ||
2034 (sk
->sk_shutdown
& RCV_SHUTDOWN
))
2035 mask
|= POLLIN
| POLLRDNORM
;
2037 /* Connection-based need to check for termination and startup */
2038 if ((sk
->sk_type
== SOCK_STREAM
|| sk
->sk_type
== SOCK_SEQPACKET
) &&
2039 sk
->sk_state
== TCP_CLOSE
)
2043 * we set writable also when the other side has shut down the
2044 * connection. This prevents stuck sockets.
2046 if (unix_writable(sk
))
2047 mask
|= POLLOUT
| POLLWRNORM
| POLLWRBAND
;
2052 static unsigned int unix_dgram_poll(struct file
*file
, struct socket
*sock
,
2055 struct sock
*sk
= sock
->sk
, *other
;
2056 unsigned int mask
, writable
;
2058 sock_poll_wait(file
, sk_sleep(sk
), wait
);
2061 /* exceptional events? */
2062 if (sk
->sk_err
|| !skb_queue_empty(&sk
->sk_error_queue
))
2064 if (sk
->sk_shutdown
& RCV_SHUTDOWN
)
2066 if (sk
->sk_shutdown
== SHUTDOWN_MASK
)
2070 if (!skb_queue_empty(&sk
->sk_receive_queue
) ||
2071 (sk
->sk_shutdown
& RCV_SHUTDOWN
))
2072 mask
|= POLLIN
| POLLRDNORM
;
2074 /* Connection-based need to check for termination and startup */
2075 if (sk
->sk_type
== SOCK_SEQPACKET
) {
2076 if (sk
->sk_state
== TCP_CLOSE
)
2078 /* connection hasn't started yet? */
2079 if (sk
->sk_state
== TCP_SYN_SENT
)
2084 writable
= unix_writable(sk
);
2086 other
= unix_peer_get(sk
);
2088 if (unix_peer(other
) != sk
) {
2089 sock_poll_wait(file
, &unix_sk(other
)->peer_wait
,
2091 if (unix_recvq_full(other
))
2100 mask
|= POLLOUT
| POLLWRNORM
| POLLWRBAND
;
2102 set_bit(SOCK_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
);
2107 #ifdef CONFIG_PROC_FS
2108 static struct sock
*first_unix_socket(int *i
)
2110 for (*i
= 0; *i
<= UNIX_HASH_SIZE
; (*i
)++) {
2111 if (!hlist_empty(&unix_socket_table
[*i
]))
2112 return __sk_head(&unix_socket_table
[*i
]);
2117 static struct sock
*next_unix_socket(int *i
, struct sock
*s
)
2119 struct sock
*next
= sk_next(s
);
2120 /* More in this chain? */
2123 /* Look for next non-empty chain. */
2124 for ((*i
)++; *i
<= UNIX_HASH_SIZE
; (*i
)++) {
2125 if (!hlist_empty(&unix_socket_table
[*i
]))
2126 return __sk_head(&unix_socket_table
[*i
]);
2131 struct unix_iter_state
{
2132 struct seq_net_private p
;
2136 static struct sock
*unix_seq_idx(struct seq_file
*seq
, loff_t pos
)
2138 struct unix_iter_state
*iter
= seq
->private;
2142 for (s
= first_unix_socket(&iter
->i
); s
; s
= next_unix_socket(&iter
->i
, s
)) {
2143 if (sock_net(s
) != seq_file_net(seq
))
2152 static void *unix_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2153 __acquires(unix_table_lock
)
2155 spin_lock(&unix_table_lock
);
2156 return *pos
? unix_seq_idx(seq
, *pos
- 1) : SEQ_START_TOKEN
;
2159 static void *unix_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2161 struct unix_iter_state
*iter
= seq
->private;
2162 struct sock
*sk
= v
;
2165 if (v
== SEQ_START_TOKEN
)
2166 sk
= first_unix_socket(&iter
->i
);
2168 sk
= next_unix_socket(&iter
->i
, sk
);
2169 while (sk
&& (sock_net(sk
) != seq_file_net(seq
)))
2170 sk
= next_unix_socket(&iter
->i
, sk
);
2174 static void unix_seq_stop(struct seq_file
*seq
, void *v
)
2175 __releases(unix_table_lock
)
2177 spin_unlock(&unix_table_lock
);
2180 static int unix_seq_show(struct seq_file
*seq
, void *v
)
2183 if (v
== SEQ_START_TOKEN
)
2184 seq_puts(seq
, "Num RefCount Protocol Flags Type St "
2188 struct unix_sock
*u
= unix_sk(s
);
2191 seq_printf(seq
, "%p: %08X %08X %08X %04X %02X %5lu",
2193 atomic_read(&s
->sk_refcnt
),
2195 s
->sk_state
== TCP_LISTEN
? __SO_ACCEPTCON
: 0,
2198 (s
->sk_state
== TCP_ESTABLISHED
? SS_CONNECTED
: SS_UNCONNECTED
) :
2199 (s
->sk_state
== TCP_ESTABLISHED
? SS_CONNECTING
: SS_DISCONNECTING
),
2207 len
= u
->addr
->len
- sizeof(short);
2208 if (!UNIX_ABSTRACT(s
))
2214 for ( ; i
< len
; i
++)
2215 seq_putc(seq
, u
->addr
->name
->sun_path
[i
]);
2217 unix_state_unlock(s
);
2218 seq_putc(seq
, '\n');
2224 static const struct seq_operations unix_seq_ops
= {
2225 .start
= unix_seq_start
,
2226 .next
= unix_seq_next
,
2227 .stop
= unix_seq_stop
,
2228 .show
= unix_seq_show
,
2231 static int unix_seq_open(struct inode
*inode
, struct file
*file
)
2233 return seq_open_net(inode
, file
, &unix_seq_ops
,
2234 sizeof(struct unix_iter_state
));
2237 static const struct file_operations unix_seq_fops
= {
2238 .owner
= THIS_MODULE
,
2239 .open
= unix_seq_open
,
2241 .llseek
= seq_lseek
,
2242 .release
= seq_release_net
,
2247 static const struct net_proto_family unix_family_ops
= {
2249 .create
= unix_create
,
2250 .owner
= THIS_MODULE
,
2254 static int __net_init
unix_net_init(struct net
*net
)
2256 int error
= -ENOMEM
;
2258 net
->unx
.sysctl_max_dgram_qlen
= 10;
2259 if (unix_sysctl_register(net
))
2262 #ifdef CONFIG_PROC_FS
2263 if (!proc_net_fops_create(net
, "unix", 0, &unix_seq_fops
)) {
2264 unix_sysctl_unregister(net
);
2273 static void __net_exit
unix_net_exit(struct net
*net
)
2275 unix_sysctl_unregister(net
);
2276 proc_net_remove(net
, "unix");
2279 static struct pernet_operations unix_net_ops
= {
2280 .init
= unix_net_init
,
2281 .exit
= unix_net_exit
,
2284 static int __init
af_unix_init(void)
2287 struct sk_buff
*dummy_skb
;
2289 BUILD_BUG_ON(sizeof(struct unix_skb_parms
) > sizeof(dummy_skb
->cb
));
2291 rc
= proto_register(&unix_proto
, 1);
2293 printk(KERN_CRIT
"%s: Cannot create unix_sock SLAB cache!\n",
2298 sock_register(&unix_family_ops
);
2299 register_pernet_subsys(&unix_net_ops
);
2304 static void __exit
af_unix_exit(void)
2306 sock_unregister(PF_UNIX
);
2307 proto_unregister(&unix_proto
);
2308 unregister_pernet_subsys(&unix_net_ops
);
2311 /* Earlier than device_initcall() so that other drivers invoking
2312 request_module() don't end up in a loop when modprobe tries
2313 to use a UNIX socket. But later than subsys_initcall() because
2314 we depend on stuff initialised there */
2315 fs_initcall(af_unix_init
);
2316 module_exit(af_unix_exit
);
2318 MODULE_LICENSE("GPL");
2319 MODULE_ALIAS_NETPROTO(PF_UNIX
);