2 * NET4: Implementation of BSD Unix domain sockets.
4 * Authors: Alan Cox, <alan@lxorguk.ukuu.org.uk>
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License
8 * as published by the Free Software Foundation; either version
9 * 2 of the License, or (at your option) any later version.
12 * Linus Torvalds : Assorted bug cures.
13 * Niibe Yutaka : async I/O support.
14 * Carsten Paeth : PF_UNIX check, address fixes.
15 * Alan Cox : Limit size of allocated blocks.
16 * Alan Cox : Fixed the stupid socketpair bug.
17 * Alan Cox : BSD compatibility fine tuning.
18 * Alan Cox : Fixed a bug in connect when interrupted.
19 * Alan Cox : Sorted out a proper draft version of
20 * file descriptor passing hacked up from
22 * Marty Leisner : Fixes to fd passing
23 * Nick Nevin : recvmsg bugfix.
24 * Alan Cox : Started proper garbage collector
25 * Heiko EiBfeldt : Missing verify_area check
26 * Alan Cox : Started POSIXisms
27 * Andreas Schwab : Replace inode by dentry for proper
29 * Kirk Petersen : Made this a module
30 * Christoph Rohland : Elegant non-blocking accept/connect algorithm.
32 * Alexey Kuznetosv : Repaired (I hope) bugs introduces
33 * by above two patches.
34 * Andrea Arcangeli : If possible we block in connect(2)
35 * if the max backlog of the listen socket
36 * is been reached. This won't break
37 * old apps and it will avoid huge amount
38 * of socks hashed (this for unix_gc()
39 * performances reasons).
40 * Security fix that limits the max
41 * number of socks to 2*max_files and
42 * the number of skb queueable in the
44 * Artur Skawina : Hash function optimizations
45 * Alexey Kuznetsov : Full scale SMP. Lot of bugs are introduced 8)
46 * Malcolm Beattie : Set peercred for socketpair
47 * Michal Ostrowski : Module initialization cleanup.
48 * Arnaldo C. Melo : Remove MOD_{INC,DEC}_USE_COUNT,
49 * the core infrastructure is doing that
50 * for all net proto families now (2.5.69+)
53 * Known differences from reference BSD that was tested:
56 * ECONNREFUSED is not returned from one end of a connected() socket to the
57 * other the moment one end closes.
58 * fstat() doesn't return st_dev=0, and give the blksize as high water mark
59 * and a fake inode identifier (nor the BSD first socket fstat twice bug).
61 * accept() returns a path name even if the connecting socket has closed
62 * in the meantime (BSD loses the path and gives up).
63 * accept() returns 0 length path for an unbound connector. BSD returns 16
64 * and a null first byte in the path (but not for gethost/peername - BSD bug ??)
65 * socketpair(...SOCK_RAW..) doesn't panic the kernel.
66 * BSD af_unix apparently has connect forgetting to block properly.
67 * (need to check this with the POSIX spec in detail)
69 * Differences from 2.0.0-11-... (ANK)
70 * Bug fixes and improvements.
71 * - client shutdown killed server socket.
72 * - removed all useless cli/sti pairs.
74 * Semantic changes/extensions.
75 * - generic control message passing.
76 * - SCM_CREDENTIALS control message.
77 * - "Abstract" (not FS based) socket bindings.
78 * Abstract names are sequences of bytes (not zero terminated)
79 * started by 0, so that this name space does not intersect
83 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
85 #include <linux/module.h>
86 #include <linux/kernel.h>
87 #include <linux/signal.h>
88 #include <linux/sched.h>
89 #include <linux/errno.h>
90 #include <linux/string.h>
91 #include <linux/stat.h>
92 #include <linux/dcache.h>
93 #include <linux/namei.h>
94 #include <linux/socket.h>
96 #include <linux/fcntl.h>
97 #include <linux/termios.h>
98 #include <linux/sockios.h>
99 #include <linux/net.h>
100 #include <linux/in.h>
101 #include <linux/fs.h>
102 #include <linux/slab.h>
103 #include <asm/uaccess.h>
104 #include <linux/skbuff.h>
105 #include <linux/netdevice.h>
106 #include <net/net_namespace.h>
107 #include <net/sock.h>
108 #include <net/tcp_states.h>
109 #include <net/af_unix.h>
110 #include <linux/proc_fs.h>
111 #include <linux/seq_file.h>
113 #include <linux/init.h>
114 #include <linux/poll.h>
115 #include <linux/rtnetlink.h>
116 #include <linux/mount.h>
117 #include <net/checksum.h>
118 #include <linux/security.h>
119 #include <linux/freezer.h>
121 struct hlist_head unix_socket_table
[2 * UNIX_HASH_SIZE
];
122 EXPORT_SYMBOL_GPL(unix_socket_table
);
123 DEFINE_SPINLOCK(unix_table_lock
);
124 EXPORT_SYMBOL_GPL(unix_table_lock
);
125 static atomic_long_t unix_nr_socks
;
128 static struct hlist_head
*unix_sockets_unbound(void *addr
)
130 unsigned long hash
= (unsigned long)addr
;
134 hash
%= UNIX_HASH_SIZE
;
135 return &unix_socket_table
[UNIX_HASH_SIZE
+ hash
];
138 #define UNIX_ABSTRACT(sk) (unix_sk(sk)->addr->hash < UNIX_HASH_SIZE)
140 #ifdef CONFIG_SECURITY_NETWORK
141 static void unix_get_secdata(struct scm_cookie
*scm
, struct sk_buff
*skb
)
143 UNIXCB(skb
).secid
= scm
->secid
;
146 static inline void unix_set_secdata(struct scm_cookie
*scm
, struct sk_buff
*skb
)
148 scm
->secid
= UNIXCB(skb
).secid
;
151 static inline bool unix_secdata_eq(struct scm_cookie
*scm
, struct sk_buff
*skb
)
153 return (scm
->secid
== UNIXCB(skb
).secid
);
156 static inline void unix_get_secdata(struct scm_cookie
*scm
, struct sk_buff
*skb
)
159 static inline void unix_set_secdata(struct scm_cookie
*scm
, struct sk_buff
*skb
)
162 static inline bool unix_secdata_eq(struct scm_cookie
*scm
, struct sk_buff
*skb
)
166 #endif /* CONFIG_SECURITY_NETWORK */
169 * SMP locking strategy:
170 * hash table is protected with spinlock unix_table_lock
171 * each socket state is protected by separate spin lock.
174 static inline unsigned int unix_hash_fold(__wsum n
)
176 unsigned int hash
= (__force
unsigned int)csum_fold(n
);
179 return hash
&(UNIX_HASH_SIZE
-1);
182 #define unix_peer(sk) (unix_sk(sk)->peer)
184 static inline int unix_our_peer(struct sock
*sk
, struct sock
*osk
)
186 return unix_peer(osk
) == sk
;
189 static inline int unix_may_send(struct sock
*sk
, struct sock
*osk
)
191 return unix_peer(osk
) == NULL
|| unix_our_peer(sk
, osk
);
194 static inline int unix_recvq_full(struct sock
const *sk
)
196 return skb_queue_len(&sk
->sk_receive_queue
) > sk
->sk_max_ack_backlog
;
199 struct sock
*unix_peer_get(struct sock
*s
)
207 unix_state_unlock(s
);
210 EXPORT_SYMBOL_GPL(unix_peer_get
);
212 static inline void unix_release_addr(struct unix_address
*addr
)
214 if (atomic_dec_and_test(&addr
->refcnt
))
219 * Check unix socket name:
220 * - should be not zero length.
221 * - if started by not zero, should be NULL terminated (FS object)
222 * - if started by zero, it is abstract name.
225 static int unix_mkname(struct sockaddr_un
*sunaddr
, int len
, unsigned int *hashp
)
227 if (len
<= sizeof(short) || len
> sizeof(*sunaddr
))
229 if (!sunaddr
|| sunaddr
->sun_family
!= AF_UNIX
)
231 if (sunaddr
->sun_path
[0]) {
233 * This may look like an off by one error but it is a bit more
234 * subtle. 108 is the longest valid AF_UNIX path for a binding.
235 * sun_path[108] doesn't as such exist. However in kernel space
236 * we are guaranteed that it is a valid memory location in our
237 * kernel address buffer.
239 ((char *)sunaddr
)[len
] = 0;
240 len
= strlen(sunaddr
->sun_path
)+1+sizeof(short);
244 *hashp
= unix_hash_fold(csum_partial(sunaddr
, len
, 0));
248 static void __unix_remove_socket(struct sock
*sk
)
250 sk_del_node_init(sk
);
253 static void __unix_insert_socket(struct hlist_head
*list
, struct sock
*sk
)
255 WARN_ON(!sk_unhashed(sk
));
256 sk_add_node(sk
, list
);
259 static inline void unix_remove_socket(struct sock
*sk
)
261 spin_lock(&unix_table_lock
);
262 __unix_remove_socket(sk
);
263 spin_unlock(&unix_table_lock
);
266 static inline void unix_insert_socket(struct hlist_head
*list
, struct sock
*sk
)
268 spin_lock(&unix_table_lock
);
269 __unix_insert_socket(list
, sk
);
270 spin_unlock(&unix_table_lock
);
273 static struct sock
*__unix_find_socket_byname(struct net
*net
,
274 struct sockaddr_un
*sunname
,
275 int len
, int type
, unsigned int hash
)
279 sk_for_each(s
, &unix_socket_table
[hash
^ type
]) {
280 struct unix_sock
*u
= unix_sk(s
);
282 if (!net_eq(sock_net(s
), net
))
285 if (u
->addr
->len
== len
&&
286 !memcmp(u
->addr
->name
, sunname
, len
))
294 static inline struct sock
*unix_find_socket_byname(struct net
*net
,
295 struct sockaddr_un
*sunname
,
301 spin_lock(&unix_table_lock
);
302 s
= __unix_find_socket_byname(net
, sunname
, len
, type
, hash
);
305 spin_unlock(&unix_table_lock
);
309 static struct sock
*unix_find_socket_byinode(struct inode
*i
)
313 spin_lock(&unix_table_lock
);
315 &unix_socket_table
[i
->i_ino
& (UNIX_HASH_SIZE
- 1)]) {
316 struct dentry
*dentry
= unix_sk(s
)->path
.dentry
;
318 if (dentry
&& d_backing_inode(dentry
) == i
) {
325 spin_unlock(&unix_table_lock
);
329 /* Support code for asymmetrically connected dgram sockets
331 * If a datagram socket is connected to a socket not itself connected
332 * to the first socket (eg, /dev/log), clients may only enqueue more
333 * messages if the present receive queue of the server socket is not
334 * "too large". This means there's a second writeability condition
335 * poll and sendmsg need to test. The dgram recv code will do a wake
336 * up on the peer_wait wait queue of a socket upon reception of a
337 * datagram which needs to be propagated to sleeping would-be writers
338 * since these might not have sent anything so far. This can't be
339 * accomplished via poll_wait because the lifetime of the server
340 * socket might be less than that of its clients if these break their
341 * association with it or if the server socket is closed while clients
342 * are still connected to it and there's no way to inform "a polling
343 * implementation" that it should let go of a certain wait queue
345 * In order to propagate a wake up, a wait_queue_t of the client
346 * socket is enqueued on the peer_wait queue of the server socket
347 * whose wake function does a wake_up on the ordinary client socket
348 * wait queue. This connection is established whenever a write (or
349 * poll for write) hit the flow control condition and broken when the
350 * association to the server socket is dissolved or after a wake up
354 static int unix_dgram_peer_wake_relay(wait_queue_t
*q
, unsigned mode
, int flags
,
358 wait_queue_head_t
*u_sleep
;
360 u
= container_of(q
, struct unix_sock
, peer_wake
);
362 __remove_wait_queue(&unix_sk(u
->peer_wake
.private)->peer_wait
,
364 u
->peer_wake
.private = NULL
;
366 /* relaying can only happen while the wq still exists */
367 u_sleep
= sk_sleep(&u
->sk
);
369 wake_up_interruptible_poll(u_sleep
, key
);
374 static int unix_dgram_peer_wake_connect(struct sock
*sk
, struct sock
*other
)
376 struct unix_sock
*u
, *u_other
;
380 u_other
= unix_sk(other
);
382 spin_lock(&u_other
->peer_wait
.lock
);
384 if (!u
->peer_wake
.private) {
385 u
->peer_wake
.private = other
;
386 __add_wait_queue(&u_other
->peer_wait
, &u
->peer_wake
);
391 spin_unlock(&u_other
->peer_wait
.lock
);
395 static void unix_dgram_peer_wake_disconnect(struct sock
*sk
,
398 struct unix_sock
*u
, *u_other
;
401 u_other
= unix_sk(other
);
402 spin_lock(&u_other
->peer_wait
.lock
);
404 if (u
->peer_wake
.private == other
) {
405 __remove_wait_queue(&u_other
->peer_wait
, &u
->peer_wake
);
406 u
->peer_wake
.private = NULL
;
409 spin_unlock(&u_other
->peer_wait
.lock
);
412 static void unix_dgram_peer_wake_disconnect_wakeup(struct sock
*sk
,
415 unix_dgram_peer_wake_disconnect(sk
, other
);
416 wake_up_interruptible_poll(sk_sleep(sk
),
423 * - unix_peer(sk) == other
424 * - association is stable
426 static int unix_dgram_peer_wake_me(struct sock
*sk
, struct sock
*other
)
430 connected
= unix_dgram_peer_wake_connect(sk
, other
);
432 if (unix_recvq_full(other
))
436 unix_dgram_peer_wake_disconnect(sk
, other
);
441 static int unix_writable(const struct sock
*sk
)
443 return sk
->sk_state
!= TCP_LISTEN
&&
444 (atomic_read(&sk
->sk_wmem_alloc
) << 2) <= sk
->sk_sndbuf
;
447 static void unix_write_space(struct sock
*sk
)
449 struct socket_wq
*wq
;
452 if (unix_writable(sk
)) {
453 wq
= rcu_dereference(sk
->sk_wq
);
454 if (wq_has_sleeper(wq
))
455 wake_up_interruptible_sync_poll(&wq
->wait
,
456 POLLOUT
| POLLWRNORM
| POLLWRBAND
);
457 sk_wake_async(sk
, SOCK_WAKE_SPACE
, POLL_OUT
);
462 /* When dgram socket disconnects (or changes its peer), we clear its receive
463 * queue of packets arrived from previous peer. First, it allows to do
464 * flow control based only on wmem_alloc; second, sk connected to peer
465 * may receive messages only from that peer. */
466 static void unix_dgram_disconnected(struct sock
*sk
, struct sock
*other
)
468 if (!skb_queue_empty(&sk
->sk_receive_queue
)) {
469 skb_queue_purge(&sk
->sk_receive_queue
);
470 wake_up_interruptible_all(&unix_sk(sk
)->peer_wait
);
472 /* If one link of bidirectional dgram pipe is disconnected,
473 * we signal error. Messages are lost. Do not make this,
474 * when peer was not connected to us.
476 if (!sock_flag(other
, SOCK_DEAD
) && unix_peer(other
) == sk
) {
477 other
->sk_err
= ECONNRESET
;
478 other
->sk_error_report(other
);
483 static void unix_sock_destructor(struct sock
*sk
)
485 struct unix_sock
*u
= unix_sk(sk
);
487 skb_queue_purge(&sk
->sk_receive_queue
);
489 WARN_ON(atomic_read(&sk
->sk_wmem_alloc
));
490 WARN_ON(!sk_unhashed(sk
));
491 WARN_ON(sk
->sk_socket
);
492 if (!sock_flag(sk
, SOCK_DEAD
)) {
493 pr_info("Attempt to release alive unix socket: %p\n", sk
);
498 unix_release_addr(u
->addr
);
500 atomic_long_dec(&unix_nr_socks
);
502 sock_prot_inuse_add(sock_net(sk
), sk
->sk_prot
, -1);
504 #ifdef UNIX_REFCNT_DEBUG
505 pr_debug("UNIX %p is destroyed, %ld are still alive.\n", sk
,
506 atomic_long_read(&unix_nr_socks
));
510 static void unix_release_sock(struct sock
*sk
, int embrion
)
512 struct unix_sock
*u
= unix_sk(sk
);
518 unix_remove_socket(sk
);
523 sk
->sk_shutdown
= SHUTDOWN_MASK
;
525 u
->path
.dentry
= NULL
;
527 state
= sk
->sk_state
;
528 sk
->sk_state
= TCP_CLOSE
;
529 unix_state_unlock(sk
);
531 wake_up_interruptible_all(&u
->peer_wait
);
533 skpair
= unix_peer(sk
);
535 if (skpair
!= NULL
) {
536 if (sk
->sk_type
== SOCK_STREAM
|| sk
->sk_type
== SOCK_SEQPACKET
) {
537 unix_state_lock(skpair
);
539 skpair
->sk_shutdown
= SHUTDOWN_MASK
;
540 if (!skb_queue_empty(&sk
->sk_receive_queue
) || embrion
)
541 skpair
->sk_err
= ECONNRESET
;
542 unix_state_unlock(skpair
);
543 skpair
->sk_state_change(skpair
);
544 sk_wake_async(skpair
, SOCK_WAKE_WAITD
, POLL_HUP
);
547 unix_dgram_peer_wake_disconnect(sk
, skpair
);
548 sock_put(skpair
); /* It may now die */
549 unix_peer(sk
) = NULL
;
552 /* Try to flush out this socket. Throw out buffers at least */
554 while ((skb
= skb_dequeue(&sk
->sk_receive_queue
)) != NULL
) {
555 if (state
== TCP_LISTEN
)
556 unix_release_sock(skb
->sk
, 1);
557 /* passed fds are erased in the kfree_skb hook */
558 UNIXCB(skb
).consumed
= skb
->len
;
567 /* ---- Socket is dead now and most probably destroyed ---- */
570 * Fixme: BSD difference: In BSD all sockets connected to us get
571 * ECONNRESET and we die on the spot. In Linux we behave
572 * like files and pipes do and wait for the last
575 * Can't we simply set sock->err?
577 * What the above comment does talk about? --ANK(980817)
580 if (unix_tot_inflight
)
581 unix_gc(); /* Garbage collect fds */
584 static void init_peercred(struct sock
*sk
)
586 put_pid(sk
->sk_peer_pid
);
587 if (sk
->sk_peer_cred
)
588 put_cred(sk
->sk_peer_cred
);
589 sk
->sk_peer_pid
= get_pid(task_tgid(current
));
590 sk
->sk_peer_cred
= get_current_cred();
593 static void copy_peercred(struct sock
*sk
, struct sock
*peersk
)
595 put_pid(sk
->sk_peer_pid
);
596 if (sk
->sk_peer_cred
)
597 put_cred(sk
->sk_peer_cred
);
598 sk
->sk_peer_pid
= get_pid(peersk
->sk_peer_pid
);
599 sk
->sk_peer_cred
= get_cred(peersk
->sk_peer_cred
);
602 static int unix_listen(struct socket
*sock
, int backlog
)
605 struct sock
*sk
= sock
->sk
;
606 struct unix_sock
*u
= unix_sk(sk
);
607 struct pid
*old_pid
= NULL
;
610 if (sock
->type
!= SOCK_STREAM
&& sock
->type
!= SOCK_SEQPACKET
)
611 goto out
; /* Only stream/seqpacket sockets accept */
614 goto out
; /* No listens on an unbound socket */
616 if (sk
->sk_state
!= TCP_CLOSE
&& sk
->sk_state
!= TCP_LISTEN
)
618 if (backlog
> sk
->sk_max_ack_backlog
)
619 wake_up_interruptible_all(&u
->peer_wait
);
620 sk
->sk_max_ack_backlog
= backlog
;
621 sk
->sk_state
= TCP_LISTEN
;
622 /* set credentials so connect can copy them */
627 unix_state_unlock(sk
);
633 static int unix_release(struct socket
*);
634 static int unix_bind(struct socket
*, struct sockaddr
*, int);
635 static int unix_stream_connect(struct socket
*, struct sockaddr
*,
636 int addr_len
, int flags
);
637 static int unix_socketpair(struct socket
*, struct socket
*);
638 static int unix_accept(struct socket
*, struct socket
*, int);
639 static int unix_getname(struct socket
*, struct sockaddr
*, int *, int);
640 static unsigned int unix_poll(struct file
*, struct socket
*, poll_table
*);
641 static unsigned int unix_dgram_poll(struct file
*, struct socket
*,
643 static int unix_ioctl(struct socket
*, unsigned int, unsigned long);
644 static int unix_shutdown(struct socket
*, int);
645 static int unix_stream_sendmsg(struct socket
*, struct msghdr
*, size_t);
646 static int unix_stream_recvmsg(struct socket
*, struct msghdr
*, size_t, int);
647 static ssize_t
unix_stream_sendpage(struct socket
*, struct page
*, int offset
,
648 size_t size
, int flags
);
649 static ssize_t
unix_stream_splice_read(struct socket
*, loff_t
*ppos
,
650 struct pipe_inode_info
*, size_t size
,
652 static int unix_dgram_sendmsg(struct socket
*, struct msghdr
*, size_t);
653 static int unix_dgram_recvmsg(struct socket
*, struct msghdr
*, size_t, int);
654 static int unix_dgram_connect(struct socket
*, struct sockaddr
*,
656 static int unix_seqpacket_sendmsg(struct socket
*, struct msghdr
*, size_t);
657 static int unix_seqpacket_recvmsg(struct socket
*, struct msghdr
*, size_t,
660 static int unix_set_peek_off(struct sock
*sk
, int val
)
662 struct unix_sock
*u
= unix_sk(sk
);
664 if (mutex_lock_interruptible(&u
->readlock
))
667 sk
->sk_peek_off
= val
;
668 mutex_unlock(&u
->readlock
);
674 static const struct proto_ops unix_stream_ops
= {
676 .owner
= THIS_MODULE
,
677 .release
= unix_release
,
679 .connect
= unix_stream_connect
,
680 .socketpair
= unix_socketpair
,
681 .accept
= unix_accept
,
682 .getname
= unix_getname
,
685 .listen
= unix_listen
,
686 .shutdown
= unix_shutdown
,
687 .setsockopt
= sock_no_setsockopt
,
688 .getsockopt
= sock_no_getsockopt
,
689 .sendmsg
= unix_stream_sendmsg
,
690 .recvmsg
= unix_stream_recvmsg
,
691 .mmap
= sock_no_mmap
,
692 .sendpage
= unix_stream_sendpage
,
693 .splice_read
= unix_stream_splice_read
,
694 .set_peek_off
= unix_set_peek_off
,
697 static const struct proto_ops unix_dgram_ops
= {
699 .owner
= THIS_MODULE
,
700 .release
= unix_release
,
702 .connect
= unix_dgram_connect
,
703 .socketpair
= unix_socketpair
,
704 .accept
= sock_no_accept
,
705 .getname
= unix_getname
,
706 .poll
= unix_dgram_poll
,
708 .listen
= sock_no_listen
,
709 .shutdown
= unix_shutdown
,
710 .setsockopt
= sock_no_setsockopt
,
711 .getsockopt
= sock_no_getsockopt
,
712 .sendmsg
= unix_dgram_sendmsg
,
713 .recvmsg
= unix_dgram_recvmsg
,
714 .mmap
= sock_no_mmap
,
715 .sendpage
= sock_no_sendpage
,
716 .set_peek_off
= unix_set_peek_off
,
719 static const struct proto_ops unix_seqpacket_ops
= {
721 .owner
= THIS_MODULE
,
722 .release
= unix_release
,
724 .connect
= unix_stream_connect
,
725 .socketpair
= unix_socketpair
,
726 .accept
= unix_accept
,
727 .getname
= unix_getname
,
728 .poll
= unix_dgram_poll
,
730 .listen
= unix_listen
,
731 .shutdown
= unix_shutdown
,
732 .setsockopt
= sock_no_setsockopt
,
733 .getsockopt
= sock_no_getsockopt
,
734 .sendmsg
= unix_seqpacket_sendmsg
,
735 .recvmsg
= unix_seqpacket_recvmsg
,
736 .mmap
= sock_no_mmap
,
737 .sendpage
= sock_no_sendpage
,
738 .set_peek_off
= unix_set_peek_off
,
741 static struct proto unix_proto
= {
743 .owner
= THIS_MODULE
,
744 .obj_size
= sizeof(struct unix_sock
),
748 * AF_UNIX sockets do not interact with hardware, hence they
749 * dont trigger interrupts - so it's safe for them to have
750 * bh-unsafe locking for their sk_receive_queue.lock. Split off
751 * this special lock-class by reinitializing the spinlock key:
753 static struct lock_class_key af_unix_sk_receive_queue_lock_key
;
755 static struct sock
*unix_create1(struct net
*net
, struct socket
*sock
, int kern
)
757 struct sock
*sk
= NULL
;
760 atomic_long_inc(&unix_nr_socks
);
761 if (atomic_long_read(&unix_nr_socks
) > 2 * get_max_files())
764 sk
= sk_alloc(net
, PF_UNIX
, GFP_KERNEL
, &unix_proto
, kern
);
768 sock_init_data(sock
, sk
);
769 lockdep_set_class(&sk
->sk_receive_queue
.lock
,
770 &af_unix_sk_receive_queue_lock_key
);
772 sk
->sk_write_space
= unix_write_space
;
773 sk
->sk_max_ack_backlog
= net
->unx
.sysctl_max_dgram_qlen
;
774 sk
->sk_destruct
= unix_sock_destructor
;
776 u
->path
.dentry
= NULL
;
778 spin_lock_init(&u
->lock
);
779 atomic_long_set(&u
->inflight
, 0);
780 INIT_LIST_HEAD(&u
->link
);
781 mutex_init(&u
->readlock
); /* single task reading lock */
782 init_waitqueue_head(&u
->peer_wait
);
783 init_waitqueue_func_entry(&u
->peer_wake
, unix_dgram_peer_wake_relay
);
784 unix_insert_socket(unix_sockets_unbound(sk
), sk
);
787 atomic_long_dec(&unix_nr_socks
);
790 sock_prot_inuse_add(sock_net(sk
), sk
->sk_prot
, 1);
796 static int unix_create(struct net
*net
, struct socket
*sock
, int protocol
,
799 if (protocol
&& protocol
!= PF_UNIX
)
800 return -EPROTONOSUPPORT
;
802 sock
->state
= SS_UNCONNECTED
;
804 switch (sock
->type
) {
806 sock
->ops
= &unix_stream_ops
;
809 * Believe it or not BSD has AF_UNIX, SOCK_RAW though
813 sock
->type
= SOCK_DGRAM
;
815 sock
->ops
= &unix_dgram_ops
;
818 sock
->ops
= &unix_seqpacket_ops
;
821 return -ESOCKTNOSUPPORT
;
824 return unix_create1(net
, sock
, kern
) ? 0 : -ENOMEM
;
827 static int unix_release(struct socket
*sock
)
829 struct sock
*sk
= sock
->sk
;
834 unix_release_sock(sk
, 0);
840 static int unix_autobind(struct socket
*sock
)
842 struct sock
*sk
= sock
->sk
;
843 struct net
*net
= sock_net(sk
);
844 struct unix_sock
*u
= unix_sk(sk
);
845 static u32 ordernum
= 1;
846 struct unix_address
*addr
;
848 unsigned int retries
= 0;
850 err
= mutex_lock_interruptible(&u
->readlock
);
859 addr
= kzalloc(sizeof(*addr
) + sizeof(short) + 16, GFP_KERNEL
);
863 addr
->name
->sun_family
= AF_UNIX
;
864 atomic_set(&addr
->refcnt
, 1);
867 addr
->len
= sprintf(addr
->name
->sun_path
+1, "%05x", ordernum
) + 1 + sizeof(short);
868 addr
->hash
= unix_hash_fold(csum_partial(addr
->name
, addr
->len
, 0));
870 spin_lock(&unix_table_lock
);
871 ordernum
= (ordernum
+1)&0xFFFFF;
873 if (__unix_find_socket_byname(net
, addr
->name
, addr
->len
, sock
->type
,
875 spin_unlock(&unix_table_lock
);
877 * __unix_find_socket_byname() may take long time if many names
878 * are already in use.
881 /* Give up if all names seems to be in use. */
882 if (retries
++ == 0xFFFFF) {
889 addr
->hash
^= sk
->sk_type
;
891 __unix_remove_socket(sk
);
893 __unix_insert_socket(&unix_socket_table
[addr
->hash
], sk
);
894 spin_unlock(&unix_table_lock
);
897 out
: mutex_unlock(&u
->readlock
);
901 static struct sock
*unix_find_other(struct net
*net
,
902 struct sockaddr_un
*sunname
, int len
,
903 int type
, unsigned int hash
, int *error
)
909 if (sunname
->sun_path
[0]) {
911 err
= kern_path(sunname
->sun_path
, LOOKUP_FOLLOW
, &path
);
914 inode
= d_backing_inode(path
.dentry
);
915 err
= inode_permission(inode
, MAY_WRITE
);
920 if (!S_ISSOCK(inode
->i_mode
))
922 u
= unix_find_socket_byinode(inode
);
926 if (u
->sk_type
== type
)
932 if (u
->sk_type
!= type
) {
938 u
= unix_find_socket_byname(net
, sunname
, len
, type
, hash
);
940 struct dentry
*dentry
;
941 dentry
= unix_sk(u
)->path
.dentry
;
943 touch_atime(&unix_sk(u
)->path
);
956 static int unix_mknod(const char *sun_path
, umode_t mode
, struct path
*res
)
958 struct dentry
*dentry
;
962 * Get the parent directory, calculate the hash for last
965 dentry
= kern_path_create(AT_FDCWD
, sun_path
, &path
, 0);
966 err
= PTR_ERR(dentry
);
971 * All right, let's create it.
973 err
= security_path_mknod(&path
, dentry
, mode
, 0);
975 err
= vfs_mknod(d_inode(path
.dentry
), dentry
, mode
, 0);
977 res
->mnt
= mntget(path
.mnt
);
978 res
->dentry
= dget(dentry
);
981 done_path_create(&path
, dentry
);
985 static int unix_bind(struct socket
*sock
, struct sockaddr
*uaddr
, int addr_len
)
987 struct sock
*sk
= sock
->sk
;
988 struct net
*net
= sock_net(sk
);
989 struct unix_sock
*u
= unix_sk(sk
);
990 struct sockaddr_un
*sunaddr
= (struct sockaddr_un
*)uaddr
;
991 char *sun_path
= sunaddr
->sun_path
;
994 struct unix_address
*addr
;
995 struct hlist_head
*list
;
998 if (sunaddr
->sun_family
!= AF_UNIX
)
1001 if (addr_len
== sizeof(short)) {
1002 err
= unix_autobind(sock
);
1006 err
= unix_mkname(sunaddr
, addr_len
, &hash
);
1011 err
= mutex_lock_interruptible(&u
->readlock
);
1020 addr
= kmalloc(sizeof(*addr
)+addr_len
, GFP_KERNEL
);
1024 memcpy(addr
->name
, sunaddr
, addr_len
);
1025 addr
->len
= addr_len
;
1026 addr
->hash
= hash
^ sk
->sk_type
;
1027 atomic_set(&addr
->refcnt
, 1);
1031 umode_t mode
= S_IFSOCK
|
1032 (SOCK_INODE(sock
)->i_mode
& ~current_umask());
1033 err
= unix_mknod(sun_path
, mode
, &path
);
1037 unix_release_addr(addr
);
1040 addr
->hash
= UNIX_HASH_SIZE
;
1041 hash
= d_backing_inode(path
.dentry
)->i_ino
& (UNIX_HASH_SIZE
-1);
1042 spin_lock(&unix_table_lock
);
1044 list
= &unix_socket_table
[hash
];
1046 spin_lock(&unix_table_lock
);
1048 if (__unix_find_socket_byname(net
, sunaddr
, addr_len
,
1049 sk
->sk_type
, hash
)) {
1050 unix_release_addr(addr
);
1054 list
= &unix_socket_table
[addr
->hash
];
1058 __unix_remove_socket(sk
);
1060 __unix_insert_socket(list
, sk
);
1063 spin_unlock(&unix_table_lock
);
1065 mutex_unlock(&u
->readlock
);
1070 static void unix_state_double_lock(struct sock
*sk1
, struct sock
*sk2
)
1072 if (unlikely(sk1
== sk2
) || !sk2
) {
1073 unix_state_lock(sk1
);
1077 unix_state_lock(sk1
);
1078 unix_state_lock_nested(sk2
);
1080 unix_state_lock(sk2
);
1081 unix_state_lock_nested(sk1
);
1085 static void unix_state_double_unlock(struct sock
*sk1
, struct sock
*sk2
)
1087 if (unlikely(sk1
== sk2
) || !sk2
) {
1088 unix_state_unlock(sk1
);
1091 unix_state_unlock(sk1
);
1092 unix_state_unlock(sk2
);
1095 static int unix_dgram_connect(struct socket
*sock
, struct sockaddr
*addr
,
1096 int alen
, int flags
)
1098 struct sock
*sk
= sock
->sk
;
1099 struct net
*net
= sock_net(sk
);
1100 struct sockaddr_un
*sunaddr
= (struct sockaddr_un
*)addr
;
1105 if (addr
->sa_family
!= AF_UNSPEC
) {
1106 err
= unix_mkname(sunaddr
, alen
, &hash
);
1111 if (test_bit(SOCK_PASSCRED
, &sock
->flags
) &&
1112 !unix_sk(sk
)->addr
&& (err
= unix_autobind(sock
)) != 0)
1116 other
= unix_find_other(net
, sunaddr
, alen
, sock
->type
, hash
, &err
);
1120 unix_state_double_lock(sk
, other
);
1122 /* Apparently VFS overslept socket death. Retry. */
1123 if (sock_flag(other
, SOCK_DEAD
)) {
1124 unix_state_double_unlock(sk
, other
);
1130 if (!unix_may_send(sk
, other
))
1133 err
= security_unix_may_send(sk
->sk_socket
, other
->sk_socket
);
1139 * 1003.1g breaking connected state with AF_UNSPEC
1142 unix_state_double_lock(sk
, other
);
1146 * If it was connected, reconnect.
1148 if (unix_peer(sk
)) {
1149 struct sock
*old_peer
= unix_peer(sk
);
1150 unix_peer(sk
) = other
;
1151 unix_dgram_peer_wake_disconnect_wakeup(sk
, old_peer
);
1153 unix_state_double_unlock(sk
, other
);
1155 if (other
!= old_peer
)
1156 unix_dgram_disconnected(sk
, old_peer
);
1159 unix_peer(sk
) = other
;
1160 unix_state_double_unlock(sk
, other
);
1165 unix_state_double_unlock(sk
, other
);
1171 static long unix_wait_for_peer(struct sock
*other
, long timeo
)
1173 struct unix_sock
*u
= unix_sk(other
);
1177 prepare_to_wait_exclusive(&u
->peer_wait
, &wait
, TASK_INTERRUPTIBLE
);
1179 sched
= !sock_flag(other
, SOCK_DEAD
) &&
1180 !(other
->sk_shutdown
& RCV_SHUTDOWN
) &&
1181 unix_recvq_full(other
);
1183 unix_state_unlock(other
);
1186 timeo
= schedule_timeout(timeo
);
1188 finish_wait(&u
->peer_wait
, &wait
);
1192 static int unix_stream_connect(struct socket
*sock
, struct sockaddr
*uaddr
,
1193 int addr_len
, int flags
)
1195 struct sockaddr_un
*sunaddr
= (struct sockaddr_un
*)uaddr
;
1196 struct sock
*sk
= sock
->sk
;
1197 struct net
*net
= sock_net(sk
);
1198 struct unix_sock
*u
= unix_sk(sk
), *newu
, *otheru
;
1199 struct sock
*newsk
= NULL
;
1200 struct sock
*other
= NULL
;
1201 struct sk_buff
*skb
= NULL
;
1207 err
= unix_mkname(sunaddr
, addr_len
, &hash
);
1212 if (test_bit(SOCK_PASSCRED
, &sock
->flags
) && !u
->addr
&&
1213 (err
= unix_autobind(sock
)) != 0)
1216 timeo
= sock_sndtimeo(sk
, flags
& O_NONBLOCK
);
1218 /* First of all allocate resources.
1219 If we will make it after state is locked,
1220 we will have to recheck all again in any case.
1225 /* create new sock for complete connection */
1226 newsk
= unix_create1(sock_net(sk
), NULL
, 0);
1230 /* Allocate skb for sending to listening sock */
1231 skb
= sock_wmalloc(newsk
, 1, 0, GFP_KERNEL
);
1236 /* Find listening sock. */
1237 other
= unix_find_other(net
, sunaddr
, addr_len
, sk
->sk_type
, hash
, &err
);
1241 /* Latch state of peer */
1242 unix_state_lock(other
);
1244 /* Apparently VFS overslept socket death. Retry. */
1245 if (sock_flag(other
, SOCK_DEAD
)) {
1246 unix_state_unlock(other
);
1251 err
= -ECONNREFUSED
;
1252 if (other
->sk_state
!= TCP_LISTEN
)
1254 if (other
->sk_shutdown
& RCV_SHUTDOWN
)
1257 if (unix_recvq_full(other
)) {
1262 timeo
= unix_wait_for_peer(other
, timeo
);
1264 err
= sock_intr_errno(timeo
);
1265 if (signal_pending(current
))
1273 It is tricky place. We need to grab our state lock and cannot
1274 drop lock on peer. It is dangerous because deadlock is
1275 possible. Connect to self case and simultaneous
1276 attempt to connect are eliminated by checking socket
1277 state. other is TCP_LISTEN, if sk is TCP_LISTEN we
1278 check this before attempt to grab lock.
1280 Well, and we have to recheck the state after socket locked.
1286 /* This is ok... continue with connect */
1288 case TCP_ESTABLISHED
:
1289 /* Socket is already connected */
1297 unix_state_lock_nested(sk
);
1299 if (sk
->sk_state
!= st
) {
1300 unix_state_unlock(sk
);
1301 unix_state_unlock(other
);
1306 err
= security_unix_stream_connect(sk
, other
, newsk
);
1308 unix_state_unlock(sk
);
1312 /* The way is open! Fastly set all the necessary fields... */
1315 unix_peer(newsk
) = sk
;
1316 newsk
->sk_state
= TCP_ESTABLISHED
;
1317 newsk
->sk_type
= sk
->sk_type
;
1318 init_peercred(newsk
);
1319 newu
= unix_sk(newsk
);
1320 RCU_INIT_POINTER(newsk
->sk_wq
, &newu
->peer_wq
);
1321 otheru
= unix_sk(other
);
1323 /* copy address information from listening to new sock*/
1325 atomic_inc(&otheru
->addr
->refcnt
);
1326 newu
->addr
= otheru
->addr
;
1328 if (otheru
->path
.dentry
) {
1329 path_get(&otheru
->path
);
1330 newu
->path
= otheru
->path
;
1333 /* Set credentials */
1334 copy_peercred(sk
, other
);
1336 sock
->state
= SS_CONNECTED
;
1337 sk
->sk_state
= TCP_ESTABLISHED
;
1340 smp_mb__after_atomic(); /* sock_hold() does an atomic_inc() */
1341 unix_peer(sk
) = newsk
;
1343 unix_state_unlock(sk
);
1345 /* take ten and and send info to listening sock */
1346 spin_lock(&other
->sk_receive_queue
.lock
);
1347 __skb_queue_tail(&other
->sk_receive_queue
, skb
);
1348 spin_unlock(&other
->sk_receive_queue
.lock
);
1349 unix_state_unlock(other
);
1350 other
->sk_data_ready(other
);
1356 unix_state_unlock(other
);
1361 unix_release_sock(newsk
, 0);
1367 static int unix_socketpair(struct socket
*socka
, struct socket
*sockb
)
1369 struct sock
*ska
= socka
->sk
, *skb
= sockb
->sk
;
1371 /* Join our sockets back to back */
1374 unix_peer(ska
) = skb
;
1375 unix_peer(skb
) = ska
;
1379 if (ska
->sk_type
!= SOCK_DGRAM
) {
1380 ska
->sk_state
= TCP_ESTABLISHED
;
1381 skb
->sk_state
= TCP_ESTABLISHED
;
1382 socka
->state
= SS_CONNECTED
;
1383 sockb
->state
= SS_CONNECTED
;
1388 static void unix_sock_inherit_flags(const struct socket
*old
,
1391 if (test_bit(SOCK_PASSCRED
, &old
->flags
))
1392 set_bit(SOCK_PASSCRED
, &new->flags
);
1393 if (test_bit(SOCK_PASSSEC
, &old
->flags
))
1394 set_bit(SOCK_PASSSEC
, &new->flags
);
1397 static int unix_accept(struct socket
*sock
, struct socket
*newsock
, int flags
)
1399 struct sock
*sk
= sock
->sk
;
1401 struct sk_buff
*skb
;
1405 if (sock
->type
!= SOCK_STREAM
&& sock
->type
!= SOCK_SEQPACKET
)
1409 if (sk
->sk_state
!= TCP_LISTEN
)
1412 /* If socket state is TCP_LISTEN it cannot change (for now...),
1413 * so that no locks are necessary.
1416 skb
= skb_recv_datagram(sk
, 0, flags
&O_NONBLOCK
, &err
);
1418 /* This means receive shutdown. */
1425 skb_free_datagram(sk
, skb
);
1426 wake_up_interruptible(&unix_sk(sk
)->peer_wait
);
1428 /* attach accepted sock to socket */
1429 unix_state_lock(tsk
);
1430 newsock
->state
= SS_CONNECTED
;
1431 unix_sock_inherit_flags(sock
, newsock
);
1432 sock_graft(tsk
, newsock
);
1433 unix_state_unlock(tsk
);
1441 static int unix_getname(struct socket
*sock
, struct sockaddr
*uaddr
, int *uaddr_len
, int peer
)
1443 struct sock
*sk
= sock
->sk
;
1444 struct unix_sock
*u
;
1445 DECLARE_SOCKADDR(struct sockaddr_un
*, sunaddr
, uaddr
);
1449 sk
= unix_peer_get(sk
);
1460 unix_state_lock(sk
);
1462 sunaddr
->sun_family
= AF_UNIX
;
1463 sunaddr
->sun_path
[0] = 0;
1464 *uaddr_len
= sizeof(short);
1466 struct unix_address
*addr
= u
->addr
;
1468 *uaddr_len
= addr
->len
;
1469 memcpy(sunaddr
, addr
->name
, *uaddr_len
);
1471 unix_state_unlock(sk
);
1477 static void unix_detach_fds(struct scm_cookie
*scm
, struct sk_buff
*skb
)
1481 scm
->fp
= UNIXCB(skb
).fp
;
1482 UNIXCB(skb
).fp
= NULL
;
1484 for (i
= scm
->fp
->count
-1; i
>= 0; i
--)
1485 unix_notinflight(scm
->fp
->fp
[i
]);
1488 static void unix_destruct_scm(struct sk_buff
*skb
)
1490 struct scm_cookie scm
;
1491 memset(&scm
, 0, sizeof(scm
));
1492 scm
.pid
= UNIXCB(skb
).pid
;
1494 unix_detach_fds(&scm
, skb
);
1496 /* Alas, it calls VFS */
1497 /* So fscking what? fput() had been SMP-safe since the last Summer */
1502 #define MAX_RECURSION_LEVEL 4
1504 static int unix_attach_fds(struct scm_cookie
*scm
, struct sk_buff
*skb
)
1507 unsigned char max_level
= 0;
1508 int unix_sock_count
= 0;
1510 for (i
= scm
->fp
->count
- 1; i
>= 0; i
--) {
1511 struct sock
*sk
= unix_get_socket(scm
->fp
->fp
[i
]);
1515 max_level
= max(max_level
,
1516 unix_sk(sk
)->recursion_level
);
1519 if (unlikely(max_level
> MAX_RECURSION_LEVEL
))
1520 return -ETOOMANYREFS
;
1523 * Need to duplicate file references for the sake of garbage
1524 * collection. Otherwise a socket in the fps might become a
1525 * candidate for GC while the skb is not yet queued.
1527 UNIXCB(skb
).fp
= scm_fp_dup(scm
->fp
);
1528 if (!UNIXCB(skb
).fp
)
1531 if (unix_sock_count
) {
1532 for (i
= scm
->fp
->count
- 1; i
>= 0; i
--)
1533 unix_inflight(scm
->fp
->fp
[i
]);
1538 static int unix_scm_to_skb(struct scm_cookie
*scm
, struct sk_buff
*skb
, bool send_fds
)
1542 UNIXCB(skb
).pid
= get_pid(scm
->pid
);
1543 UNIXCB(skb
).uid
= scm
->creds
.uid
;
1544 UNIXCB(skb
).gid
= scm
->creds
.gid
;
1545 UNIXCB(skb
).fp
= NULL
;
1546 unix_get_secdata(scm
, skb
);
1547 if (scm
->fp
&& send_fds
)
1548 err
= unix_attach_fds(scm
, skb
);
1550 skb
->destructor
= unix_destruct_scm
;
1555 * Some apps rely on write() giving SCM_CREDENTIALS
1556 * We include credentials if source or destination socket
1557 * asserted SOCK_PASSCRED.
1559 static void maybe_add_creds(struct sk_buff
*skb
, const struct socket
*sock
,
1560 const struct sock
*other
)
1562 if (UNIXCB(skb
).pid
)
1564 if (test_bit(SOCK_PASSCRED
, &sock
->flags
) ||
1565 !other
->sk_socket
||
1566 test_bit(SOCK_PASSCRED
, &other
->sk_socket
->flags
)) {
1567 UNIXCB(skb
).pid
= get_pid(task_tgid(current
));
1568 current_uid_gid(&UNIXCB(skb
).uid
, &UNIXCB(skb
).gid
);
1573 * Send AF_UNIX data.
1576 static int unix_dgram_sendmsg(struct socket
*sock
, struct msghdr
*msg
,
1579 struct sock
*sk
= sock
->sk
;
1580 struct net
*net
= sock_net(sk
);
1581 struct unix_sock
*u
= unix_sk(sk
);
1582 DECLARE_SOCKADDR(struct sockaddr_un
*, sunaddr
, msg
->msg_name
);
1583 struct sock
*other
= NULL
;
1584 int namelen
= 0; /* fake GCC */
1587 struct sk_buff
*skb
;
1589 struct scm_cookie scm
;
1595 err
= scm_send(sock
, msg
, &scm
, false);
1600 if (msg
->msg_flags
&MSG_OOB
)
1603 if (msg
->msg_namelen
) {
1604 err
= unix_mkname(sunaddr
, msg
->msg_namelen
, &hash
);
1611 other
= unix_peer_get(sk
);
1616 if (test_bit(SOCK_PASSCRED
, &sock
->flags
) && !u
->addr
1617 && (err
= unix_autobind(sock
)) != 0)
1621 if (len
> sk
->sk_sndbuf
- 32)
1624 if (len
> SKB_MAX_ALLOC
) {
1625 data_len
= min_t(size_t,
1626 len
- SKB_MAX_ALLOC
,
1627 MAX_SKB_FRAGS
* PAGE_SIZE
);
1628 data_len
= PAGE_ALIGN(data_len
);
1630 BUILD_BUG_ON(SKB_MAX_ALLOC
< PAGE_SIZE
);
1633 skb
= sock_alloc_send_pskb(sk
, len
- data_len
, data_len
,
1634 msg
->msg_flags
& MSG_DONTWAIT
, &err
,
1635 PAGE_ALLOC_COSTLY_ORDER
);
1639 err
= unix_scm_to_skb(&scm
, skb
, true);
1642 max_level
= err
+ 1;
1644 skb_put(skb
, len
- data_len
);
1645 skb
->data_len
= data_len
;
1647 err
= skb_copy_datagram_from_iter(skb
, 0, &msg
->msg_iter
, len
);
1651 timeo
= sock_sndtimeo(sk
, msg
->msg_flags
& MSG_DONTWAIT
);
1656 if (sunaddr
== NULL
)
1659 other
= unix_find_other(net
, sunaddr
, namelen
, sk
->sk_type
,
1665 if (sk_filter(other
, skb
) < 0) {
1666 /* Toss the packet but do not return any error to the sender */
1672 unix_state_lock(other
);
1675 if (!unix_may_send(sk
, other
))
1678 if (unlikely(sock_flag(other
, SOCK_DEAD
))) {
1680 * Check with 1003.1g - what should
1683 unix_state_unlock(other
);
1687 unix_state_lock(sk
);
1690 if (unix_peer(sk
) == other
) {
1691 unix_peer(sk
) = NULL
;
1692 unix_dgram_peer_wake_disconnect_wakeup(sk
, other
);
1694 unix_state_unlock(sk
);
1696 unix_dgram_disconnected(sk
, other
);
1698 err
= -ECONNREFUSED
;
1700 unix_state_unlock(sk
);
1710 if (other
->sk_shutdown
& RCV_SHUTDOWN
)
1713 if (sk
->sk_type
!= SOCK_SEQPACKET
) {
1714 err
= security_unix_may_send(sk
->sk_socket
, other
->sk_socket
);
1719 if (unlikely(unix_peer(other
) != sk
&& unix_recvq_full(other
))) {
1721 timeo
= unix_wait_for_peer(other
, timeo
);
1723 err
= sock_intr_errno(timeo
);
1724 if (signal_pending(current
))
1731 unix_state_unlock(other
);
1732 unix_state_double_lock(sk
, other
);
1735 if (unix_peer(sk
) != other
||
1736 unix_dgram_peer_wake_me(sk
, other
)) {
1744 goto restart_locked
;
1748 if (unlikely(sk_locked
))
1749 unix_state_unlock(sk
);
1751 if (sock_flag(other
, SOCK_RCVTSTAMP
))
1752 __net_timestamp(skb
);
1753 maybe_add_creds(skb
, sock
, other
);
1754 skb_queue_tail(&other
->sk_receive_queue
, skb
);
1755 if (max_level
> unix_sk(other
)->recursion_level
)
1756 unix_sk(other
)->recursion_level
= max_level
;
1757 unix_state_unlock(other
);
1758 other
->sk_data_ready(other
);
1765 unix_state_unlock(sk
);
1766 unix_state_unlock(other
);
1776 /* We use paged skbs for stream sockets, and limit occupancy to 32768
1777 * bytes, and a minimun of a full page.
1779 #define UNIX_SKB_FRAGS_SZ (PAGE_SIZE << get_order(32768))
1781 static int unix_stream_sendmsg(struct socket
*sock
, struct msghdr
*msg
,
1784 struct sock
*sk
= sock
->sk
;
1785 struct sock
*other
= NULL
;
1787 struct sk_buff
*skb
;
1789 struct scm_cookie scm
;
1790 bool fds_sent
= false;
1795 err
= scm_send(sock
, msg
, &scm
, false);
1800 if (msg
->msg_flags
&MSG_OOB
)
1803 if (msg
->msg_namelen
) {
1804 err
= sk
->sk_state
== TCP_ESTABLISHED
? -EISCONN
: -EOPNOTSUPP
;
1808 other
= unix_peer(sk
);
1813 if (sk
->sk_shutdown
& SEND_SHUTDOWN
)
1816 while (sent
< len
) {
1819 /* Keep two messages in the pipe so it schedules better */
1820 size
= min_t(int, size
, (sk
->sk_sndbuf
>> 1) - 64);
1822 /* allow fallback to order-0 allocations */
1823 size
= min_t(int, size
, SKB_MAX_HEAD(0) + UNIX_SKB_FRAGS_SZ
);
1825 data_len
= max_t(int, 0, size
- SKB_MAX_HEAD(0));
1827 data_len
= min_t(size_t, size
, PAGE_ALIGN(data_len
));
1829 skb
= sock_alloc_send_pskb(sk
, size
- data_len
, data_len
,
1830 msg
->msg_flags
& MSG_DONTWAIT
, &err
,
1831 get_order(UNIX_SKB_FRAGS_SZ
));
1835 /* Only send the fds in the first buffer */
1836 err
= unix_scm_to_skb(&scm
, skb
, !fds_sent
);
1841 max_level
= err
+ 1;
1844 skb_put(skb
, size
- data_len
);
1845 skb
->data_len
= data_len
;
1847 err
= skb_copy_datagram_from_iter(skb
, 0, &msg
->msg_iter
, size
);
1853 unix_state_lock(other
);
1855 if (sock_flag(other
, SOCK_DEAD
) ||
1856 (other
->sk_shutdown
& RCV_SHUTDOWN
))
1859 maybe_add_creds(skb
, sock
, other
);
1860 skb_queue_tail(&other
->sk_receive_queue
, skb
);
1861 if (max_level
> unix_sk(other
)->recursion_level
)
1862 unix_sk(other
)->recursion_level
= max_level
;
1863 unix_state_unlock(other
);
1864 other
->sk_data_ready(other
);
1873 unix_state_unlock(other
);
1876 if (sent
== 0 && !(msg
->msg_flags
&MSG_NOSIGNAL
))
1877 send_sig(SIGPIPE
, current
, 0);
1881 return sent
? : err
;
1884 static ssize_t
unix_stream_sendpage(struct socket
*socket
, struct page
*page
,
1885 int offset
, size_t size
, int flags
)
1888 bool send_sigpipe
= true;
1889 struct sock
*other
, *sk
= socket
->sk
;
1890 struct sk_buff
*skb
, *newskb
= NULL
, *tail
= NULL
;
1892 if (flags
& MSG_OOB
)
1895 other
= unix_peer(sk
);
1896 if (!other
|| sk
->sk_state
!= TCP_ESTABLISHED
)
1901 unix_state_unlock(other
);
1902 mutex_unlock(&unix_sk(other
)->readlock
);
1903 newskb
= sock_alloc_send_pskb(sk
, 0, 0, flags
& MSG_DONTWAIT
,
1909 /* we must acquire readlock as we modify already present
1910 * skbs in the sk_receive_queue and mess with skb->len
1912 err
= mutex_lock_interruptible(&unix_sk(other
)->readlock
);
1914 err
= flags
& MSG_DONTWAIT
? -EAGAIN
: -ERESTARTSYS
;
1915 send_sigpipe
= false;
1919 if (sk
->sk_shutdown
& SEND_SHUTDOWN
) {
1924 unix_state_lock(other
);
1926 if (sock_flag(other
, SOCK_DEAD
) ||
1927 other
->sk_shutdown
& RCV_SHUTDOWN
) {
1929 goto err_state_unlock
;
1932 skb
= skb_peek_tail(&other
->sk_receive_queue
);
1933 if (tail
&& tail
== skb
) {
1940 } else if (newskb
) {
1941 /* this is fast path, we don't necessarily need to
1942 * call to kfree_skb even though with newskb == NULL
1943 * this - does no harm
1945 consume_skb(newskb
);
1949 if (skb_append_pagefrags(skb
, page
, offset
, size
)) {
1955 skb
->data_len
+= size
;
1956 skb
->truesize
+= size
;
1957 atomic_add(size
, &sk
->sk_wmem_alloc
);
1960 spin_lock(&other
->sk_receive_queue
.lock
);
1961 __skb_queue_tail(&other
->sk_receive_queue
, newskb
);
1962 spin_unlock(&other
->sk_receive_queue
.lock
);
1965 unix_state_unlock(other
);
1966 mutex_unlock(&unix_sk(other
)->readlock
);
1968 other
->sk_data_ready(other
);
1973 unix_state_unlock(other
);
1975 mutex_unlock(&unix_sk(other
)->readlock
);
1978 if (send_sigpipe
&& !(flags
& MSG_NOSIGNAL
))
1979 send_sig(SIGPIPE
, current
, 0);
1983 static int unix_seqpacket_sendmsg(struct socket
*sock
, struct msghdr
*msg
,
1987 struct sock
*sk
= sock
->sk
;
1989 err
= sock_error(sk
);
1993 if (sk
->sk_state
!= TCP_ESTABLISHED
)
1996 if (msg
->msg_namelen
)
1997 msg
->msg_namelen
= 0;
1999 return unix_dgram_sendmsg(sock
, msg
, len
);
2002 static int unix_seqpacket_recvmsg(struct socket
*sock
, struct msghdr
*msg
,
2003 size_t size
, int flags
)
2005 struct sock
*sk
= sock
->sk
;
2007 if (sk
->sk_state
!= TCP_ESTABLISHED
)
2010 return unix_dgram_recvmsg(sock
, msg
, size
, flags
);
2013 static void unix_copy_addr(struct msghdr
*msg
, struct sock
*sk
)
2015 struct unix_sock
*u
= unix_sk(sk
);
2018 msg
->msg_namelen
= u
->addr
->len
;
2019 memcpy(msg
->msg_name
, u
->addr
->name
, u
->addr
->len
);
2023 static int unix_dgram_recvmsg(struct socket
*sock
, struct msghdr
*msg
,
2024 size_t size
, int flags
)
2026 struct scm_cookie scm
;
2027 struct sock
*sk
= sock
->sk
;
2028 struct unix_sock
*u
= unix_sk(sk
);
2029 int noblock
= flags
& MSG_DONTWAIT
;
2030 struct sk_buff
*skb
;
2038 err
= mutex_lock_interruptible(&u
->readlock
);
2039 if (unlikely(err
)) {
2040 /* recvmsg() in non blocking mode is supposed to return -EAGAIN
2041 * sk_rcvtimeo is not honored by mutex_lock_interruptible()
2043 err
= noblock
? -EAGAIN
: -ERESTARTSYS
;
2047 skip
= sk_peek_offset(sk
, flags
);
2049 skb
= __skb_recv_datagram(sk
, flags
, &peeked
, &skip
, &err
);
2051 unix_state_lock(sk
);
2052 /* Signal EOF on disconnected non-blocking SEQPACKET socket. */
2053 if (sk
->sk_type
== SOCK_SEQPACKET
&& err
== -EAGAIN
&&
2054 (sk
->sk_shutdown
& RCV_SHUTDOWN
))
2056 unix_state_unlock(sk
);
2060 wake_up_interruptible_sync_poll(&u
->peer_wait
,
2061 POLLOUT
| POLLWRNORM
| POLLWRBAND
);
2064 unix_copy_addr(msg
, skb
->sk
);
2066 if (size
> skb
->len
- skip
)
2067 size
= skb
->len
- skip
;
2068 else if (size
< skb
->len
- skip
)
2069 msg
->msg_flags
|= MSG_TRUNC
;
2071 err
= skb_copy_datagram_msg(skb
, skip
, msg
, size
);
2075 if (sock_flag(sk
, SOCK_RCVTSTAMP
))
2076 __sock_recv_timestamp(msg
, sk
, skb
);
2078 memset(&scm
, 0, sizeof(scm
));
2080 scm_set_cred(&scm
, UNIXCB(skb
).pid
, UNIXCB(skb
).uid
, UNIXCB(skb
).gid
);
2081 unix_set_secdata(&scm
, skb
);
2083 if (!(flags
& MSG_PEEK
)) {
2085 unix_detach_fds(&scm
, skb
);
2087 sk_peek_offset_bwd(sk
, skb
->len
);
2089 /* It is questionable: on PEEK we could:
2090 - do not return fds - good, but too simple 8)
2091 - return fds, and do not return them on read (old strategy,
2093 - clone fds (I chose it for now, it is the most universal
2096 POSIX 1003.1g does not actually define this clearly
2097 at all. POSIX 1003.1g doesn't define a lot of things
2102 sk_peek_offset_fwd(sk
, size
);
2105 scm
.fp
= scm_fp_dup(UNIXCB(skb
).fp
);
2107 err
= (flags
& MSG_TRUNC
) ? skb
->len
- skip
: size
;
2109 scm_recv(sock
, msg
, &scm
, flags
);
2112 skb_free_datagram(sk
, skb
);
2114 mutex_unlock(&u
->readlock
);
2120 * Sleep until more data has arrived. But check for races..
2122 static long unix_stream_data_wait(struct sock
*sk
, long timeo
,
2123 struct sk_buff
*last
, unsigned int last_len
)
2125 struct sk_buff
*tail
;
2128 unix_state_lock(sk
);
2131 prepare_to_wait(sk_sleep(sk
), &wait
, TASK_INTERRUPTIBLE
);
2133 tail
= skb_peek_tail(&sk
->sk_receive_queue
);
2135 (tail
&& tail
->len
!= last_len
) ||
2137 (sk
->sk_shutdown
& RCV_SHUTDOWN
) ||
2138 signal_pending(current
) ||
2142 set_bit(SOCK_ASYNC_WAITDATA
, &sk
->sk_socket
->flags
);
2143 unix_state_unlock(sk
);
2144 timeo
= freezable_schedule_timeout(timeo
);
2145 unix_state_lock(sk
);
2147 if (sock_flag(sk
, SOCK_DEAD
))
2150 clear_bit(SOCK_ASYNC_WAITDATA
, &sk
->sk_socket
->flags
);
2153 finish_wait(sk_sleep(sk
), &wait
);
2154 unix_state_unlock(sk
);
2158 static unsigned int unix_skb_len(const struct sk_buff
*skb
)
2160 return skb
->len
- UNIXCB(skb
).consumed
;
2163 struct unix_stream_read_state
{
2164 int (*recv_actor
)(struct sk_buff
*, int, int,
2165 struct unix_stream_read_state
*);
2166 struct socket
*socket
;
2168 struct pipe_inode_info
*pipe
;
2171 unsigned int splice_flags
;
2174 static int unix_stream_read_generic(struct unix_stream_read_state
*state
)
2176 struct scm_cookie scm
;
2177 struct socket
*sock
= state
->socket
;
2178 struct sock
*sk
= sock
->sk
;
2179 struct unix_sock
*u
= unix_sk(sk
);
2181 int flags
= state
->flags
;
2182 int noblock
= flags
& MSG_DONTWAIT
;
2183 bool check_creds
= false;
2188 size_t size
= state
->size
;
2189 unsigned int last_len
;
2192 if (sk
->sk_state
!= TCP_ESTABLISHED
)
2196 if (flags
& MSG_OOB
)
2199 target
= sock_rcvlowat(sk
, flags
& MSG_WAITALL
, size
);
2200 timeo
= sock_rcvtimeo(sk
, noblock
);
2202 memset(&scm
, 0, sizeof(scm
));
2204 /* Lock the socket to prevent queue disordering
2205 * while sleeps in memcpy_tomsg
2207 err
= mutex_lock_interruptible(&u
->readlock
);
2208 if (unlikely(err
)) {
2209 /* recvmsg() in non blocking mode is supposed to return -EAGAIN
2210 * sk_rcvtimeo is not honored by mutex_lock_interruptible()
2212 err
= noblock
? -EAGAIN
: -ERESTARTSYS
;
2216 if (flags
& MSG_PEEK
)
2217 skip
= sk_peek_offset(sk
, flags
);
2224 struct sk_buff
*skb
, *last
;
2226 unix_state_lock(sk
);
2227 if (sock_flag(sk
, SOCK_DEAD
)) {
2231 last
= skb
= skb_peek(&sk
->sk_receive_queue
);
2232 last_len
= last
? last
->len
: 0;
2235 unix_sk(sk
)->recursion_level
= 0;
2236 if (copied
>= target
)
2240 * POSIX 1003.1g mandates this order.
2243 err
= sock_error(sk
);
2246 if (sk
->sk_shutdown
& RCV_SHUTDOWN
)
2249 unix_state_unlock(sk
);
2253 mutex_unlock(&u
->readlock
);
2255 timeo
= unix_stream_data_wait(sk
, timeo
, last
,
2258 if (signal_pending(current
) ||
2259 mutex_lock_interruptible(&u
->readlock
)) {
2260 err
= sock_intr_errno(timeo
);
2266 unix_state_unlock(sk
);
2270 while (skip
>= unix_skb_len(skb
)) {
2271 skip
-= unix_skb_len(skb
);
2273 last_len
= skb
->len
;
2274 skb
= skb_peek_next(skb
, &sk
->sk_receive_queue
);
2279 unix_state_unlock(sk
);
2282 /* Never glue messages from different writers */
2283 if ((UNIXCB(skb
).pid
!= scm
.pid
) ||
2284 !uid_eq(UNIXCB(skb
).uid
, scm
.creds
.uid
) ||
2285 !gid_eq(UNIXCB(skb
).gid
, scm
.creds
.gid
) ||
2286 !unix_secdata_eq(&scm
, skb
))
2288 } else if (test_bit(SOCK_PASSCRED
, &sock
->flags
)) {
2289 /* Copy credentials */
2290 scm_set_cred(&scm
, UNIXCB(skb
).pid
, UNIXCB(skb
).uid
, UNIXCB(skb
).gid
);
2291 unix_set_secdata(&scm
, skb
);
2295 /* Copy address just once */
2296 if (state
->msg
&& state
->msg
->msg_name
) {
2297 DECLARE_SOCKADDR(struct sockaddr_un
*, sunaddr
,
2298 state
->msg
->msg_name
);
2299 unix_copy_addr(state
->msg
, skb
->sk
);
2303 chunk
= min_t(unsigned int, unix_skb_len(skb
) - skip
, size
);
2305 chunk
= state
->recv_actor(skb
, skip
, chunk
, state
);
2306 drop_skb
= !unix_skb_len(skb
);
2307 /* skb is only safe to use if !drop_skb */
2318 /* the skb was touched by a concurrent reader;
2319 * we should not expect anything from this skb
2320 * anymore and assume it invalid - we can be
2321 * sure it was dropped from the socket queue
2323 * let's report a short read
2329 /* Mark read part of skb as used */
2330 if (!(flags
& MSG_PEEK
)) {
2331 UNIXCB(skb
).consumed
+= chunk
;
2333 sk_peek_offset_bwd(sk
, chunk
);
2336 unix_detach_fds(&scm
, skb
);
2338 if (unix_skb_len(skb
))
2341 skb_unlink(skb
, &sk
->sk_receive_queue
);
2347 /* It is questionable, see note in unix_dgram_recvmsg.
2350 scm
.fp
= scm_fp_dup(UNIXCB(skb
).fp
);
2352 sk_peek_offset_fwd(sk
, chunk
);
2359 last_len
= skb
->len
;
2360 unix_state_lock(sk
);
2361 skb
= skb_peek_next(skb
, &sk
->sk_receive_queue
);
2364 unix_state_unlock(sk
);
2369 mutex_unlock(&u
->readlock
);
2371 scm_recv(sock
, state
->msg
, &scm
, flags
);
2375 return copied
? : err
;
2378 static int unix_stream_read_actor(struct sk_buff
*skb
,
2379 int skip
, int chunk
,
2380 struct unix_stream_read_state
*state
)
2384 ret
= skb_copy_datagram_msg(skb
, UNIXCB(skb
).consumed
+ skip
,
2386 return ret
?: chunk
;
2389 static int unix_stream_recvmsg(struct socket
*sock
, struct msghdr
*msg
,
2390 size_t size
, int flags
)
2392 struct unix_stream_read_state state
= {
2393 .recv_actor
= unix_stream_read_actor
,
2400 return unix_stream_read_generic(&state
);
2403 static ssize_t
skb_unix_socket_splice(struct sock
*sk
,
2404 struct pipe_inode_info
*pipe
,
2405 struct splice_pipe_desc
*spd
)
2408 struct unix_sock
*u
= unix_sk(sk
);
2410 mutex_unlock(&u
->readlock
);
2411 ret
= splice_to_pipe(pipe
, spd
);
2412 mutex_lock(&u
->readlock
);
2417 static int unix_stream_splice_actor(struct sk_buff
*skb
,
2418 int skip
, int chunk
,
2419 struct unix_stream_read_state
*state
)
2421 return skb_splice_bits(skb
, state
->socket
->sk
,
2422 UNIXCB(skb
).consumed
+ skip
,
2423 state
->pipe
, chunk
, state
->splice_flags
,
2424 skb_unix_socket_splice
);
2427 static ssize_t
unix_stream_splice_read(struct socket
*sock
, loff_t
*ppos
,
2428 struct pipe_inode_info
*pipe
,
2429 size_t size
, unsigned int flags
)
2431 struct unix_stream_read_state state
= {
2432 .recv_actor
= unix_stream_splice_actor
,
2436 .splice_flags
= flags
,
2439 if (unlikely(*ppos
))
2442 if (sock
->file
->f_flags
& O_NONBLOCK
||
2443 flags
& SPLICE_F_NONBLOCK
)
2444 state
.flags
= MSG_DONTWAIT
;
2446 return unix_stream_read_generic(&state
);
2449 static int unix_shutdown(struct socket
*sock
, int mode
)
2451 struct sock
*sk
= sock
->sk
;
2454 if (mode
< SHUT_RD
|| mode
> SHUT_RDWR
)
2457 * SHUT_RD (0) -> RCV_SHUTDOWN (1)
2458 * SHUT_WR (1) -> SEND_SHUTDOWN (2)
2459 * SHUT_RDWR (2) -> SHUTDOWN_MASK (3)
2463 unix_state_lock(sk
);
2464 sk
->sk_shutdown
|= mode
;
2465 other
= unix_peer(sk
);
2468 unix_state_unlock(sk
);
2469 sk
->sk_state_change(sk
);
2472 (sk
->sk_type
== SOCK_STREAM
|| sk
->sk_type
== SOCK_SEQPACKET
)) {
2476 if (mode
&RCV_SHUTDOWN
)
2477 peer_mode
|= SEND_SHUTDOWN
;
2478 if (mode
&SEND_SHUTDOWN
)
2479 peer_mode
|= RCV_SHUTDOWN
;
2480 unix_state_lock(other
);
2481 other
->sk_shutdown
|= peer_mode
;
2482 unix_state_unlock(other
);
2483 other
->sk_state_change(other
);
2484 if (peer_mode
== SHUTDOWN_MASK
)
2485 sk_wake_async(other
, SOCK_WAKE_WAITD
, POLL_HUP
);
2486 else if (peer_mode
& RCV_SHUTDOWN
)
2487 sk_wake_async(other
, SOCK_WAKE_WAITD
, POLL_IN
);
2495 long unix_inq_len(struct sock
*sk
)
2497 struct sk_buff
*skb
;
2500 if (sk
->sk_state
== TCP_LISTEN
)
2503 spin_lock(&sk
->sk_receive_queue
.lock
);
2504 if (sk
->sk_type
== SOCK_STREAM
||
2505 sk
->sk_type
== SOCK_SEQPACKET
) {
2506 skb_queue_walk(&sk
->sk_receive_queue
, skb
)
2507 amount
+= unix_skb_len(skb
);
2509 skb
= skb_peek(&sk
->sk_receive_queue
);
2513 spin_unlock(&sk
->sk_receive_queue
.lock
);
2517 EXPORT_SYMBOL_GPL(unix_inq_len
);
2519 long unix_outq_len(struct sock
*sk
)
2521 return sk_wmem_alloc_get(sk
);
2523 EXPORT_SYMBOL_GPL(unix_outq_len
);
2525 static int unix_ioctl(struct socket
*sock
, unsigned int cmd
, unsigned long arg
)
2527 struct sock
*sk
= sock
->sk
;
2533 amount
= unix_outq_len(sk
);
2534 err
= put_user(amount
, (int __user
*)arg
);
2537 amount
= unix_inq_len(sk
);
2541 err
= put_user(amount
, (int __user
*)arg
);
2550 static unsigned int unix_poll(struct file
*file
, struct socket
*sock
, poll_table
*wait
)
2552 struct sock
*sk
= sock
->sk
;
2555 sock_poll_wait(file
, sk_sleep(sk
), wait
);
2558 /* exceptional events? */
2561 if (sk
->sk_shutdown
== SHUTDOWN_MASK
)
2563 if (sk
->sk_shutdown
& RCV_SHUTDOWN
)
2564 mask
|= POLLRDHUP
| POLLIN
| POLLRDNORM
;
2567 if (!skb_queue_empty(&sk
->sk_receive_queue
))
2568 mask
|= POLLIN
| POLLRDNORM
;
2570 /* Connection-based need to check for termination and startup */
2571 if ((sk
->sk_type
== SOCK_STREAM
|| sk
->sk_type
== SOCK_SEQPACKET
) &&
2572 sk
->sk_state
== TCP_CLOSE
)
2576 * we set writable also when the other side has shut down the
2577 * connection. This prevents stuck sockets.
2579 if (unix_writable(sk
))
2580 mask
|= POLLOUT
| POLLWRNORM
| POLLWRBAND
;
2585 static unsigned int unix_dgram_poll(struct file
*file
, struct socket
*sock
,
2588 struct sock
*sk
= sock
->sk
, *other
;
2589 unsigned int mask
, writable
;
2591 sock_poll_wait(file
, sk_sleep(sk
), wait
);
2594 /* exceptional events? */
2595 if (sk
->sk_err
|| !skb_queue_empty(&sk
->sk_error_queue
))
2597 (sock_flag(sk
, SOCK_SELECT_ERR_QUEUE
) ? POLLPRI
: 0);
2599 if (sk
->sk_shutdown
& RCV_SHUTDOWN
)
2600 mask
|= POLLRDHUP
| POLLIN
| POLLRDNORM
;
2601 if (sk
->sk_shutdown
== SHUTDOWN_MASK
)
2605 if (!skb_queue_empty(&sk
->sk_receive_queue
))
2606 mask
|= POLLIN
| POLLRDNORM
;
2608 /* Connection-based need to check for termination and startup */
2609 if (sk
->sk_type
== SOCK_SEQPACKET
) {
2610 if (sk
->sk_state
== TCP_CLOSE
)
2612 /* connection hasn't started yet? */
2613 if (sk
->sk_state
== TCP_SYN_SENT
)
2617 /* No write status requested, avoid expensive OUT tests. */
2618 if (!(poll_requested_events(wait
) & (POLLWRBAND
|POLLWRNORM
|POLLOUT
)))
2621 writable
= unix_writable(sk
);
2623 unix_state_lock(sk
);
2625 other
= unix_peer(sk
);
2626 if (other
&& unix_peer(other
) != sk
&&
2627 unix_recvq_full(other
) &&
2628 unix_dgram_peer_wake_me(sk
, other
))
2631 unix_state_unlock(sk
);
2635 mask
|= POLLOUT
| POLLWRNORM
| POLLWRBAND
;
2637 set_bit(SOCK_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
);
2642 #ifdef CONFIG_PROC_FS
2644 #define BUCKET_SPACE (BITS_PER_LONG - (UNIX_HASH_BITS + 1) - 1)
2646 #define get_bucket(x) ((x) >> BUCKET_SPACE)
2647 #define get_offset(x) ((x) & ((1L << BUCKET_SPACE) - 1))
2648 #define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
2650 static struct sock
*unix_from_bucket(struct seq_file
*seq
, loff_t
*pos
)
2652 unsigned long offset
= get_offset(*pos
);
2653 unsigned long bucket
= get_bucket(*pos
);
2655 unsigned long count
= 0;
2657 for (sk
= sk_head(&unix_socket_table
[bucket
]); sk
; sk
= sk_next(sk
)) {
2658 if (sock_net(sk
) != seq_file_net(seq
))
2660 if (++count
== offset
)
2667 static struct sock
*unix_next_socket(struct seq_file
*seq
,
2671 unsigned long bucket
;
2673 while (sk
> (struct sock
*)SEQ_START_TOKEN
) {
2677 if (sock_net(sk
) == seq_file_net(seq
))
2682 sk
= unix_from_bucket(seq
, pos
);
2687 bucket
= get_bucket(*pos
) + 1;
2688 *pos
= set_bucket_offset(bucket
, 1);
2689 } while (bucket
< ARRAY_SIZE(unix_socket_table
));
2694 static void *unix_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2695 __acquires(unix_table_lock
)
2697 spin_lock(&unix_table_lock
);
2700 return SEQ_START_TOKEN
;
2702 if (get_bucket(*pos
) >= ARRAY_SIZE(unix_socket_table
))
2705 return unix_next_socket(seq
, NULL
, pos
);
2708 static void *unix_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2711 return unix_next_socket(seq
, v
, pos
);
2714 static void unix_seq_stop(struct seq_file
*seq
, void *v
)
2715 __releases(unix_table_lock
)
2717 spin_unlock(&unix_table_lock
);
2720 static int unix_seq_show(struct seq_file
*seq
, void *v
)
2723 if (v
== SEQ_START_TOKEN
)
2724 seq_puts(seq
, "Num RefCount Protocol Flags Type St "
2728 struct unix_sock
*u
= unix_sk(s
);
2731 seq_printf(seq
, "%pK: %08X %08X %08X %04X %02X %5lu",
2733 atomic_read(&s
->sk_refcnt
),
2735 s
->sk_state
== TCP_LISTEN
? __SO_ACCEPTCON
: 0,
2738 (s
->sk_state
== TCP_ESTABLISHED
? SS_CONNECTED
: SS_UNCONNECTED
) :
2739 (s
->sk_state
== TCP_ESTABLISHED
? SS_CONNECTING
: SS_DISCONNECTING
),
2747 len
= u
->addr
->len
- sizeof(short);
2748 if (!UNIX_ABSTRACT(s
))
2754 for ( ; i
< len
; i
++)
2755 seq_putc(seq
, u
->addr
->name
->sun_path
[i
]);
2757 unix_state_unlock(s
);
2758 seq_putc(seq
, '\n');
2764 static const struct seq_operations unix_seq_ops
= {
2765 .start
= unix_seq_start
,
2766 .next
= unix_seq_next
,
2767 .stop
= unix_seq_stop
,
2768 .show
= unix_seq_show
,
2771 static int unix_seq_open(struct inode
*inode
, struct file
*file
)
2773 return seq_open_net(inode
, file
, &unix_seq_ops
,
2774 sizeof(struct seq_net_private
));
2777 static const struct file_operations unix_seq_fops
= {
2778 .owner
= THIS_MODULE
,
2779 .open
= unix_seq_open
,
2781 .llseek
= seq_lseek
,
2782 .release
= seq_release_net
,
2787 static const struct net_proto_family unix_family_ops
= {
2789 .create
= unix_create
,
2790 .owner
= THIS_MODULE
,
2794 static int __net_init
unix_net_init(struct net
*net
)
2796 int error
= -ENOMEM
;
2798 net
->unx
.sysctl_max_dgram_qlen
= 10;
2799 if (unix_sysctl_register(net
))
2802 #ifdef CONFIG_PROC_FS
2803 if (!proc_create("unix", 0, net
->proc_net
, &unix_seq_fops
)) {
2804 unix_sysctl_unregister(net
);
2813 static void __net_exit
unix_net_exit(struct net
*net
)
2815 unix_sysctl_unregister(net
);
2816 remove_proc_entry("unix", net
->proc_net
);
2819 static struct pernet_operations unix_net_ops
= {
2820 .init
= unix_net_init
,
2821 .exit
= unix_net_exit
,
2824 static int __init
af_unix_init(void)
2828 BUILD_BUG_ON(sizeof(struct unix_skb_parms
) > FIELD_SIZEOF(struct sk_buff
, cb
));
2830 rc
= proto_register(&unix_proto
, 1);
2832 pr_crit("%s: Cannot create unix_sock SLAB cache!\n", __func__
);
2836 sock_register(&unix_family_ops
);
2837 register_pernet_subsys(&unix_net_ops
);
2842 static void __exit
af_unix_exit(void)
2844 sock_unregister(PF_UNIX
);
2845 proto_unregister(&unix_proto
);
2846 unregister_pernet_subsys(&unix_net_ops
);
2849 /* Earlier than device_initcall() so that other drivers invoking
2850 request_module() don't end up in a loop when modprobe tries
2851 to use a UNIX socket. But later than subsys_initcall() because
2852 we depend on stuff initialised there */
2853 fs_initcall(af_unix_init
);
2854 module_exit(af_unix_exit
);
2856 MODULE_LICENSE("GPL");
2857 MODULE_ALIAS_NETPROTO(PF_UNIX
);