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 (skwq_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(struct dentry
*dentry
, struct path
*path
, umode_t mode
,
961 err
= security_path_mknod(path
, dentry
, mode
, 0);
963 err
= vfs_mknod(d_inode(path
->dentry
), dentry
, mode
, 0);
965 res
->mnt
= mntget(path
->mnt
);
966 res
->dentry
= dget(dentry
);
973 static int unix_bind(struct socket
*sock
, struct sockaddr
*uaddr
, int addr_len
)
975 struct sock
*sk
= sock
->sk
;
976 struct net
*net
= sock_net(sk
);
977 struct unix_sock
*u
= unix_sk(sk
);
978 struct sockaddr_un
*sunaddr
= (struct sockaddr_un
*)uaddr
;
979 char *sun_path
= sunaddr
->sun_path
;
982 struct unix_address
*addr
;
983 struct hlist_head
*list
;
985 struct dentry
*dentry
;
988 if (sunaddr
->sun_family
!= AF_UNIX
)
991 if (addr_len
== sizeof(short)) {
992 err
= unix_autobind(sock
);
996 err
= unix_mkname(sunaddr
, addr_len
, &hash
);
1004 /* Get the parent directory, calculate the hash for last
1007 dentry
= kern_path_create(AT_FDCWD
, sun_path
, &path
, 0);
1009 if (IS_ERR(dentry
)) {
1010 /* delay report until after 'already bound' check */
1011 name_err
= PTR_ERR(dentry
);
1016 err
= mutex_lock_interruptible(&u
->readlock
);
1025 err
= name_err
== -EEXIST
? -EADDRINUSE
: name_err
;
1030 addr
= kmalloc(sizeof(*addr
)+addr_len
, GFP_KERNEL
);
1034 memcpy(addr
->name
, sunaddr
, addr_len
);
1035 addr
->len
= addr_len
;
1036 addr
->hash
= hash
^ sk
->sk_type
;
1037 atomic_set(&addr
->refcnt
, 1);
1041 umode_t mode
= S_IFSOCK
|
1042 (SOCK_INODE(sock
)->i_mode
& ~current_umask());
1043 err
= unix_mknod(dentry
, &path
, mode
, &u_path
);
1047 unix_release_addr(addr
);
1050 addr
->hash
= UNIX_HASH_SIZE
;
1051 hash
= d_backing_inode(dentry
)->i_ino
& (UNIX_HASH_SIZE
- 1);
1052 spin_lock(&unix_table_lock
);
1054 list
= &unix_socket_table
[hash
];
1056 spin_lock(&unix_table_lock
);
1058 if (__unix_find_socket_byname(net
, sunaddr
, addr_len
,
1059 sk
->sk_type
, hash
)) {
1060 unix_release_addr(addr
);
1064 list
= &unix_socket_table
[addr
->hash
];
1068 __unix_remove_socket(sk
);
1070 __unix_insert_socket(list
, sk
);
1073 spin_unlock(&unix_table_lock
);
1075 mutex_unlock(&u
->readlock
);
1078 done_path_create(&path
, dentry
);
1084 static void unix_state_double_lock(struct sock
*sk1
, struct sock
*sk2
)
1086 if (unlikely(sk1
== sk2
) || !sk2
) {
1087 unix_state_lock(sk1
);
1091 unix_state_lock(sk1
);
1092 unix_state_lock_nested(sk2
);
1094 unix_state_lock(sk2
);
1095 unix_state_lock_nested(sk1
);
1099 static void unix_state_double_unlock(struct sock
*sk1
, struct sock
*sk2
)
1101 if (unlikely(sk1
== sk2
) || !sk2
) {
1102 unix_state_unlock(sk1
);
1105 unix_state_unlock(sk1
);
1106 unix_state_unlock(sk2
);
1109 static int unix_dgram_connect(struct socket
*sock
, struct sockaddr
*addr
,
1110 int alen
, int flags
)
1112 struct sock
*sk
= sock
->sk
;
1113 struct net
*net
= sock_net(sk
);
1114 struct sockaddr_un
*sunaddr
= (struct sockaddr_un
*)addr
;
1119 if (addr
->sa_family
!= AF_UNSPEC
) {
1120 err
= unix_mkname(sunaddr
, alen
, &hash
);
1125 if (test_bit(SOCK_PASSCRED
, &sock
->flags
) &&
1126 !unix_sk(sk
)->addr
&& (err
= unix_autobind(sock
)) != 0)
1130 other
= unix_find_other(net
, sunaddr
, alen
, sock
->type
, hash
, &err
);
1134 unix_state_double_lock(sk
, other
);
1136 /* Apparently VFS overslept socket death. Retry. */
1137 if (sock_flag(other
, SOCK_DEAD
)) {
1138 unix_state_double_unlock(sk
, other
);
1144 if (!unix_may_send(sk
, other
))
1147 err
= security_unix_may_send(sk
->sk_socket
, other
->sk_socket
);
1153 * 1003.1g breaking connected state with AF_UNSPEC
1156 unix_state_double_lock(sk
, other
);
1160 * If it was connected, reconnect.
1162 if (unix_peer(sk
)) {
1163 struct sock
*old_peer
= unix_peer(sk
);
1164 unix_peer(sk
) = other
;
1165 unix_dgram_peer_wake_disconnect_wakeup(sk
, old_peer
);
1167 unix_state_double_unlock(sk
, other
);
1169 if (other
!= old_peer
)
1170 unix_dgram_disconnected(sk
, old_peer
);
1173 unix_peer(sk
) = other
;
1174 unix_state_double_unlock(sk
, other
);
1179 unix_state_double_unlock(sk
, other
);
1185 static long unix_wait_for_peer(struct sock
*other
, long timeo
)
1187 struct unix_sock
*u
= unix_sk(other
);
1191 prepare_to_wait_exclusive(&u
->peer_wait
, &wait
, TASK_INTERRUPTIBLE
);
1193 sched
= !sock_flag(other
, SOCK_DEAD
) &&
1194 !(other
->sk_shutdown
& RCV_SHUTDOWN
) &&
1195 unix_recvq_full(other
);
1197 unix_state_unlock(other
);
1200 timeo
= schedule_timeout(timeo
);
1202 finish_wait(&u
->peer_wait
, &wait
);
1206 static int unix_stream_connect(struct socket
*sock
, struct sockaddr
*uaddr
,
1207 int addr_len
, int flags
)
1209 struct sockaddr_un
*sunaddr
= (struct sockaddr_un
*)uaddr
;
1210 struct sock
*sk
= sock
->sk
;
1211 struct net
*net
= sock_net(sk
);
1212 struct unix_sock
*u
= unix_sk(sk
), *newu
, *otheru
;
1213 struct sock
*newsk
= NULL
;
1214 struct sock
*other
= NULL
;
1215 struct sk_buff
*skb
= NULL
;
1221 err
= unix_mkname(sunaddr
, addr_len
, &hash
);
1226 if (test_bit(SOCK_PASSCRED
, &sock
->flags
) && !u
->addr
&&
1227 (err
= unix_autobind(sock
)) != 0)
1230 timeo
= sock_sndtimeo(sk
, flags
& O_NONBLOCK
);
1232 /* First of all allocate resources.
1233 If we will make it after state is locked,
1234 we will have to recheck all again in any case.
1239 /* create new sock for complete connection */
1240 newsk
= unix_create1(sock_net(sk
), NULL
, 0);
1244 /* Allocate skb for sending to listening sock */
1245 skb
= sock_wmalloc(newsk
, 1, 0, GFP_KERNEL
);
1250 /* Find listening sock. */
1251 other
= unix_find_other(net
, sunaddr
, addr_len
, sk
->sk_type
, hash
, &err
);
1255 /* Latch state of peer */
1256 unix_state_lock(other
);
1258 /* Apparently VFS overslept socket death. Retry. */
1259 if (sock_flag(other
, SOCK_DEAD
)) {
1260 unix_state_unlock(other
);
1265 err
= -ECONNREFUSED
;
1266 if (other
->sk_state
!= TCP_LISTEN
)
1268 if (other
->sk_shutdown
& RCV_SHUTDOWN
)
1271 if (unix_recvq_full(other
)) {
1276 timeo
= unix_wait_for_peer(other
, timeo
);
1278 err
= sock_intr_errno(timeo
);
1279 if (signal_pending(current
))
1287 It is tricky place. We need to grab our state lock and cannot
1288 drop lock on peer. It is dangerous because deadlock is
1289 possible. Connect to self case and simultaneous
1290 attempt to connect are eliminated by checking socket
1291 state. other is TCP_LISTEN, if sk is TCP_LISTEN we
1292 check this before attempt to grab lock.
1294 Well, and we have to recheck the state after socket locked.
1300 /* This is ok... continue with connect */
1302 case TCP_ESTABLISHED
:
1303 /* Socket is already connected */
1311 unix_state_lock_nested(sk
);
1313 if (sk
->sk_state
!= st
) {
1314 unix_state_unlock(sk
);
1315 unix_state_unlock(other
);
1320 err
= security_unix_stream_connect(sk
, other
, newsk
);
1322 unix_state_unlock(sk
);
1326 /* The way is open! Fastly set all the necessary fields... */
1329 unix_peer(newsk
) = sk
;
1330 newsk
->sk_state
= TCP_ESTABLISHED
;
1331 newsk
->sk_type
= sk
->sk_type
;
1332 init_peercred(newsk
);
1333 newu
= unix_sk(newsk
);
1334 RCU_INIT_POINTER(newsk
->sk_wq
, &newu
->peer_wq
);
1335 otheru
= unix_sk(other
);
1337 /* copy address information from listening to new sock*/
1339 atomic_inc(&otheru
->addr
->refcnt
);
1340 newu
->addr
= otheru
->addr
;
1342 if (otheru
->path
.dentry
) {
1343 path_get(&otheru
->path
);
1344 newu
->path
= otheru
->path
;
1347 /* Set credentials */
1348 copy_peercred(sk
, other
);
1350 sock
->state
= SS_CONNECTED
;
1351 sk
->sk_state
= TCP_ESTABLISHED
;
1354 smp_mb__after_atomic(); /* sock_hold() does an atomic_inc() */
1355 unix_peer(sk
) = newsk
;
1357 unix_state_unlock(sk
);
1359 /* take ten and and send info to listening sock */
1360 spin_lock(&other
->sk_receive_queue
.lock
);
1361 __skb_queue_tail(&other
->sk_receive_queue
, skb
);
1362 spin_unlock(&other
->sk_receive_queue
.lock
);
1363 unix_state_unlock(other
);
1364 other
->sk_data_ready(other
);
1370 unix_state_unlock(other
);
1375 unix_release_sock(newsk
, 0);
1381 static int unix_socketpair(struct socket
*socka
, struct socket
*sockb
)
1383 struct sock
*ska
= socka
->sk
, *skb
= sockb
->sk
;
1385 /* Join our sockets back to back */
1388 unix_peer(ska
) = skb
;
1389 unix_peer(skb
) = ska
;
1393 if (ska
->sk_type
!= SOCK_DGRAM
) {
1394 ska
->sk_state
= TCP_ESTABLISHED
;
1395 skb
->sk_state
= TCP_ESTABLISHED
;
1396 socka
->state
= SS_CONNECTED
;
1397 sockb
->state
= SS_CONNECTED
;
1402 static void unix_sock_inherit_flags(const struct socket
*old
,
1405 if (test_bit(SOCK_PASSCRED
, &old
->flags
))
1406 set_bit(SOCK_PASSCRED
, &new->flags
);
1407 if (test_bit(SOCK_PASSSEC
, &old
->flags
))
1408 set_bit(SOCK_PASSSEC
, &new->flags
);
1411 static int unix_accept(struct socket
*sock
, struct socket
*newsock
, int flags
)
1413 struct sock
*sk
= sock
->sk
;
1415 struct sk_buff
*skb
;
1419 if (sock
->type
!= SOCK_STREAM
&& sock
->type
!= SOCK_SEQPACKET
)
1423 if (sk
->sk_state
!= TCP_LISTEN
)
1426 /* If socket state is TCP_LISTEN it cannot change (for now...),
1427 * so that no locks are necessary.
1430 skb
= skb_recv_datagram(sk
, 0, flags
&O_NONBLOCK
, &err
);
1432 /* This means receive shutdown. */
1439 skb_free_datagram(sk
, skb
);
1440 wake_up_interruptible(&unix_sk(sk
)->peer_wait
);
1442 /* attach accepted sock to socket */
1443 unix_state_lock(tsk
);
1444 newsock
->state
= SS_CONNECTED
;
1445 unix_sock_inherit_flags(sock
, newsock
);
1446 sock_graft(tsk
, newsock
);
1447 unix_state_unlock(tsk
);
1455 static int unix_getname(struct socket
*sock
, struct sockaddr
*uaddr
, int *uaddr_len
, int peer
)
1457 struct sock
*sk
= sock
->sk
;
1458 struct unix_sock
*u
;
1459 DECLARE_SOCKADDR(struct sockaddr_un
*, sunaddr
, uaddr
);
1463 sk
= unix_peer_get(sk
);
1474 unix_state_lock(sk
);
1476 sunaddr
->sun_family
= AF_UNIX
;
1477 sunaddr
->sun_path
[0] = 0;
1478 *uaddr_len
= sizeof(short);
1480 struct unix_address
*addr
= u
->addr
;
1482 *uaddr_len
= addr
->len
;
1483 memcpy(sunaddr
, addr
->name
, *uaddr_len
);
1485 unix_state_unlock(sk
);
1491 static void unix_detach_fds(struct scm_cookie
*scm
, struct sk_buff
*skb
)
1495 scm
->fp
= UNIXCB(skb
).fp
;
1496 UNIXCB(skb
).fp
= NULL
;
1498 for (i
= scm
->fp
->count
-1; i
>= 0; i
--)
1499 unix_notinflight(scm
->fp
->user
, scm
->fp
->fp
[i
]);
1502 static void unix_destruct_scm(struct sk_buff
*skb
)
1504 struct scm_cookie scm
;
1505 memset(&scm
, 0, sizeof(scm
));
1506 scm
.pid
= UNIXCB(skb
).pid
;
1508 unix_detach_fds(&scm
, skb
);
1510 /* Alas, it calls VFS */
1511 /* So fscking what? fput() had been SMP-safe since the last Summer */
1517 * The "user->unix_inflight" variable is protected by the garbage
1518 * collection lock, and we just read it locklessly here. If you go
1519 * over the limit, there might be a tiny race in actually noticing
1520 * it across threads. Tough.
1522 static inline bool too_many_unix_fds(struct task_struct
*p
)
1524 struct user_struct
*user
= current_user();
1526 if (unlikely(user
->unix_inflight
> task_rlimit(p
, RLIMIT_NOFILE
)))
1527 return !capable(CAP_SYS_RESOURCE
) && !capable(CAP_SYS_ADMIN
);
1531 #define MAX_RECURSION_LEVEL 4
1533 static int unix_attach_fds(struct scm_cookie
*scm
, struct sk_buff
*skb
)
1536 unsigned char max_level
= 0;
1537 int unix_sock_count
= 0;
1539 if (too_many_unix_fds(current
))
1540 return -ETOOMANYREFS
;
1542 for (i
= scm
->fp
->count
- 1; i
>= 0; i
--) {
1543 struct sock
*sk
= unix_get_socket(scm
->fp
->fp
[i
]);
1547 max_level
= max(max_level
,
1548 unix_sk(sk
)->recursion_level
);
1551 if (unlikely(max_level
> MAX_RECURSION_LEVEL
))
1552 return -ETOOMANYREFS
;
1555 * Need to duplicate file references for the sake of garbage
1556 * collection. Otherwise a socket in the fps might become a
1557 * candidate for GC while the skb is not yet queued.
1559 UNIXCB(skb
).fp
= scm_fp_dup(scm
->fp
);
1560 if (!UNIXCB(skb
).fp
)
1563 for (i
= scm
->fp
->count
- 1; i
>= 0; i
--)
1564 unix_inflight(scm
->fp
->user
, scm
->fp
->fp
[i
]);
1568 static int unix_scm_to_skb(struct scm_cookie
*scm
, struct sk_buff
*skb
, bool send_fds
)
1572 UNIXCB(skb
).pid
= get_pid(scm
->pid
);
1573 UNIXCB(skb
).uid
= scm
->creds
.uid
;
1574 UNIXCB(skb
).gid
= scm
->creds
.gid
;
1575 UNIXCB(skb
).fp
= NULL
;
1576 unix_get_secdata(scm
, skb
);
1577 if (scm
->fp
&& send_fds
)
1578 err
= unix_attach_fds(scm
, skb
);
1580 skb
->destructor
= unix_destruct_scm
;
1584 static bool unix_passcred_enabled(const struct socket
*sock
,
1585 const struct sock
*other
)
1587 return test_bit(SOCK_PASSCRED
, &sock
->flags
) ||
1588 !other
->sk_socket
||
1589 test_bit(SOCK_PASSCRED
, &other
->sk_socket
->flags
);
1593 * Some apps rely on write() giving SCM_CREDENTIALS
1594 * We include credentials if source or destination socket
1595 * asserted SOCK_PASSCRED.
1597 static void maybe_add_creds(struct sk_buff
*skb
, const struct socket
*sock
,
1598 const struct sock
*other
)
1600 if (UNIXCB(skb
).pid
)
1602 if (unix_passcred_enabled(sock
, other
)) {
1603 UNIXCB(skb
).pid
= get_pid(task_tgid(current
));
1604 current_uid_gid(&UNIXCB(skb
).uid
, &UNIXCB(skb
).gid
);
1608 static int maybe_init_creds(struct scm_cookie
*scm
,
1609 struct socket
*socket
,
1610 const struct sock
*other
)
1613 struct msghdr msg
= { .msg_controllen
= 0 };
1615 err
= scm_send(socket
, &msg
, scm
, false);
1619 if (unix_passcred_enabled(socket
, other
)) {
1620 scm
->pid
= get_pid(task_tgid(current
));
1621 current_uid_gid(&scm
->creds
.uid
, &scm
->creds
.gid
);
1626 static bool unix_skb_scm_eq(struct sk_buff
*skb
,
1627 struct scm_cookie
*scm
)
1629 const struct unix_skb_parms
*u
= &UNIXCB(skb
);
1631 return u
->pid
== scm
->pid
&&
1632 uid_eq(u
->uid
, scm
->creds
.uid
) &&
1633 gid_eq(u
->gid
, scm
->creds
.gid
) &&
1634 unix_secdata_eq(scm
, skb
);
1638 * Send AF_UNIX data.
1641 static int unix_dgram_sendmsg(struct socket
*sock
, struct msghdr
*msg
,
1644 struct sock
*sk
= sock
->sk
;
1645 struct net
*net
= sock_net(sk
);
1646 struct unix_sock
*u
= unix_sk(sk
);
1647 DECLARE_SOCKADDR(struct sockaddr_un
*, sunaddr
, msg
->msg_name
);
1648 struct sock
*other
= NULL
;
1649 int namelen
= 0; /* fake GCC */
1652 struct sk_buff
*skb
;
1654 struct scm_cookie scm
;
1660 err
= scm_send(sock
, msg
, &scm
, false);
1665 if (msg
->msg_flags
&MSG_OOB
)
1668 if (msg
->msg_namelen
) {
1669 err
= unix_mkname(sunaddr
, msg
->msg_namelen
, &hash
);
1676 other
= unix_peer_get(sk
);
1681 if (test_bit(SOCK_PASSCRED
, &sock
->flags
) && !u
->addr
1682 && (err
= unix_autobind(sock
)) != 0)
1686 if (len
> sk
->sk_sndbuf
- 32)
1689 if (len
> SKB_MAX_ALLOC
) {
1690 data_len
= min_t(size_t,
1691 len
- SKB_MAX_ALLOC
,
1692 MAX_SKB_FRAGS
* PAGE_SIZE
);
1693 data_len
= PAGE_ALIGN(data_len
);
1695 BUILD_BUG_ON(SKB_MAX_ALLOC
< PAGE_SIZE
);
1698 skb
= sock_alloc_send_pskb(sk
, len
- data_len
, data_len
,
1699 msg
->msg_flags
& MSG_DONTWAIT
, &err
,
1700 PAGE_ALLOC_COSTLY_ORDER
);
1704 err
= unix_scm_to_skb(&scm
, skb
, true);
1707 max_level
= err
+ 1;
1709 skb_put(skb
, len
- data_len
);
1710 skb
->data_len
= data_len
;
1712 err
= skb_copy_datagram_from_iter(skb
, 0, &msg
->msg_iter
, len
);
1716 timeo
= sock_sndtimeo(sk
, msg
->msg_flags
& MSG_DONTWAIT
);
1721 if (sunaddr
== NULL
)
1724 other
= unix_find_other(net
, sunaddr
, namelen
, sk
->sk_type
,
1730 if (sk_filter(other
, skb
) < 0) {
1731 /* Toss the packet but do not return any error to the sender */
1737 unix_state_lock(other
);
1740 if (!unix_may_send(sk
, other
))
1743 if (unlikely(sock_flag(other
, SOCK_DEAD
))) {
1745 * Check with 1003.1g - what should
1748 unix_state_unlock(other
);
1752 unix_state_lock(sk
);
1755 if (unix_peer(sk
) == other
) {
1756 unix_peer(sk
) = NULL
;
1757 unix_dgram_peer_wake_disconnect_wakeup(sk
, other
);
1759 unix_state_unlock(sk
);
1761 unix_dgram_disconnected(sk
, other
);
1763 err
= -ECONNREFUSED
;
1765 unix_state_unlock(sk
);
1775 if (other
->sk_shutdown
& RCV_SHUTDOWN
)
1778 if (sk
->sk_type
!= SOCK_SEQPACKET
) {
1779 err
= security_unix_may_send(sk
->sk_socket
, other
->sk_socket
);
1784 /* other == sk && unix_peer(other) != sk if
1785 * - unix_peer(sk) == NULL, destination address bound to sk
1786 * - unix_peer(sk) == sk by time of get but disconnected before lock
1789 unlikely(unix_peer(other
) != sk
&& unix_recvq_full(other
))) {
1791 timeo
= unix_wait_for_peer(other
, timeo
);
1793 err
= sock_intr_errno(timeo
);
1794 if (signal_pending(current
))
1801 unix_state_unlock(other
);
1802 unix_state_double_lock(sk
, other
);
1805 if (unix_peer(sk
) != other
||
1806 unix_dgram_peer_wake_me(sk
, other
)) {
1814 goto restart_locked
;
1818 if (unlikely(sk_locked
))
1819 unix_state_unlock(sk
);
1821 if (sock_flag(other
, SOCK_RCVTSTAMP
))
1822 __net_timestamp(skb
);
1823 maybe_add_creds(skb
, sock
, other
);
1824 skb_queue_tail(&other
->sk_receive_queue
, skb
);
1825 if (max_level
> unix_sk(other
)->recursion_level
)
1826 unix_sk(other
)->recursion_level
= max_level
;
1827 unix_state_unlock(other
);
1828 other
->sk_data_ready(other
);
1835 unix_state_unlock(sk
);
1836 unix_state_unlock(other
);
1846 /* We use paged skbs for stream sockets, and limit occupancy to 32768
1847 * bytes, and a minimun of a full page.
1849 #define UNIX_SKB_FRAGS_SZ (PAGE_SIZE << get_order(32768))
1851 static int unix_stream_sendmsg(struct socket
*sock
, struct msghdr
*msg
,
1854 struct sock
*sk
= sock
->sk
;
1855 struct sock
*other
= NULL
;
1857 struct sk_buff
*skb
;
1859 struct scm_cookie scm
;
1860 bool fds_sent
= false;
1865 err
= scm_send(sock
, msg
, &scm
, false);
1870 if (msg
->msg_flags
&MSG_OOB
)
1873 if (msg
->msg_namelen
) {
1874 err
= sk
->sk_state
== TCP_ESTABLISHED
? -EISCONN
: -EOPNOTSUPP
;
1878 other
= unix_peer(sk
);
1883 if (sk
->sk_shutdown
& SEND_SHUTDOWN
)
1886 while (sent
< len
) {
1889 /* Keep two messages in the pipe so it schedules better */
1890 size
= min_t(int, size
, (sk
->sk_sndbuf
>> 1) - 64);
1892 /* allow fallback to order-0 allocations */
1893 size
= min_t(int, size
, SKB_MAX_HEAD(0) + UNIX_SKB_FRAGS_SZ
);
1895 data_len
= max_t(int, 0, size
- SKB_MAX_HEAD(0));
1897 data_len
= min_t(size_t, size
, PAGE_ALIGN(data_len
));
1899 skb
= sock_alloc_send_pskb(sk
, size
- data_len
, data_len
,
1900 msg
->msg_flags
& MSG_DONTWAIT
, &err
,
1901 get_order(UNIX_SKB_FRAGS_SZ
));
1905 /* Only send the fds in the first buffer */
1906 err
= unix_scm_to_skb(&scm
, skb
, !fds_sent
);
1911 max_level
= err
+ 1;
1914 skb_put(skb
, size
- data_len
);
1915 skb
->data_len
= data_len
;
1917 err
= skb_copy_datagram_from_iter(skb
, 0, &msg
->msg_iter
, size
);
1923 unix_state_lock(other
);
1925 if (sock_flag(other
, SOCK_DEAD
) ||
1926 (other
->sk_shutdown
& RCV_SHUTDOWN
))
1929 maybe_add_creds(skb
, sock
, other
);
1930 skb_queue_tail(&other
->sk_receive_queue
, skb
);
1931 if (max_level
> unix_sk(other
)->recursion_level
)
1932 unix_sk(other
)->recursion_level
= max_level
;
1933 unix_state_unlock(other
);
1934 other
->sk_data_ready(other
);
1943 unix_state_unlock(other
);
1946 if (sent
== 0 && !(msg
->msg_flags
&MSG_NOSIGNAL
))
1947 send_sig(SIGPIPE
, current
, 0);
1951 return sent
? : err
;
1954 static ssize_t
unix_stream_sendpage(struct socket
*socket
, struct page
*page
,
1955 int offset
, size_t size
, int flags
)
1958 bool send_sigpipe
= false;
1959 bool init_scm
= true;
1960 struct scm_cookie scm
;
1961 struct sock
*other
, *sk
= socket
->sk
;
1962 struct sk_buff
*skb
, *newskb
= NULL
, *tail
= NULL
;
1964 if (flags
& MSG_OOB
)
1967 other
= unix_peer(sk
);
1968 if (!other
|| sk
->sk_state
!= TCP_ESTABLISHED
)
1973 unix_state_unlock(other
);
1974 mutex_unlock(&unix_sk(other
)->readlock
);
1975 newskb
= sock_alloc_send_pskb(sk
, 0, 0, flags
& MSG_DONTWAIT
,
1981 /* we must acquire readlock as we modify already present
1982 * skbs in the sk_receive_queue and mess with skb->len
1984 err
= mutex_lock_interruptible(&unix_sk(other
)->readlock
);
1986 err
= flags
& MSG_DONTWAIT
? -EAGAIN
: -ERESTARTSYS
;
1990 if (sk
->sk_shutdown
& SEND_SHUTDOWN
) {
1992 send_sigpipe
= true;
1996 unix_state_lock(other
);
1998 if (sock_flag(other
, SOCK_DEAD
) ||
1999 other
->sk_shutdown
& RCV_SHUTDOWN
) {
2001 send_sigpipe
= true;
2002 goto err_state_unlock
;
2006 err
= maybe_init_creds(&scm
, socket
, other
);
2008 goto err_state_unlock
;
2012 skb
= skb_peek_tail(&other
->sk_receive_queue
);
2013 if (tail
&& tail
== skb
) {
2015 } else if (!skb
|| !unix_skb_scm_eq(skb
, &scm
)) {
2022 } else if (newskb
) {
2023 /* this is fast path, we don't necessarily need to
2024 * call to kfree_skb even though with newskb == NULL
2025 * this - does no harm
2027 consume_skb(newskb
);
2031 if (skb_append_pagefrags(skb
, page
, offset
, size
)) {
2037 skb
->data_len
+= size
;
2038 skb
->truesize
+= size
;
2039 atomic_add(size
, &sk
->sk_wmem_alloc
);
2042 err
= unix_scm_to_skb(&scm
, skb
, false);
2044 goto err_state_unlock
;
2045 spin_lock(&other
->sk_receive_queue
.lock
);
2046 __skb_queue_tail(&other
->sk_receive_queue
, newskb
);
2047 spin_unlock(&other
->sk_receive_queue
.lock
);
2050 unix_state_unlock(other
);
2051 mutex_unlock(&unix_sk(other
)->readlock
);
2053 other
->sk_data_ready(other
);
2058 unix_state_unlock(other
);
2060 mutex_unlock(&unix_sk(other
)->readlock
);
2063 if (send_sigpipe
&& !(flags
& MSG_NOSIGNAL
))
2064 send_sig(SIGPIPE
, current
, 0);
2070 static int unix_seqpacket_sendmsg(struct socket
*sock
, struct msghdr
*msg
,
2074 struct sock
*sk
= sock
->sk
;
2076 err
= sock_error(sk
);
2080 if (sk
->sk_state
!= TCP_ESTABLISHED
)
2083 if (msg
->msg_namelen
)
2084 msg
->msg_namelen
= 0;
2086 return unix_dgram_sendmsg(sock
, msg
, len
);
2089 static int unix_seqpacket_recvmsg(struct socket
*sock
, struct msghdr
*msg
,
2090 size_t size
, int flags
)
2092 struct sock
*sk
= sock
->sk
;
2094 if (sk
->sk_state
!= TCP_ESTABLISHED
)
2097 return unix_dgram_recvmsg(sock
, msg
, size
, flags
);
2100 static void unix_copy_addr(struct msghdr
*msg
, struct sock
*sk
)
2102 struct unix_sock
*u
= unix_sk(sk
);
2105 msg
->msg_namelen
= u
->addr
->len
;
2106 memcpy(msg
->msg_name
, u
->addr
->name
, u
->addr
->len
);
2110 static int unix_dgram_recvmsg(struct socket
*sock
, struct msghdr
*msg
,
2111 size_t size
, int flags
)
2113 struct scm_cookie scm
;
2114 struct sock
*sk
= sock
->sk
;
2115 struct unix_sock
*u
= unix_sk(sk
);
2116 struct sk_buff
*skb
, *last
;
2125 timeo
= sock_rcvtimeo(sk
, flags
& MSG_DONTWAIT
);
2128 mutex_lock(&u
->readlock
);
2130 skip
= sk_peek_offset(sk
, flags
);
2131 skb
= __skb_try_recv_datagram(sk
, flags
, &peeked
, &skip
, &err
,
2136 mutex_unlock(&u
->readlock
);
2141 !__skb_wait_for_more_packets(sk
, &err
, &timeo
, last
));
2143 if (!skb
) { /* implies readlock unlocked */
2144 unix_state_lock(sk
);
2145 /* Signal EOF on disconnected non-blocking SEQPACKET socket. */
2146 if (sk
->sk_type
== SOCK_SEQPACKET
&& err
== -EAGAIN
&&
2147 (sk
->sk_shutdown
& RCV_SHUTDOWN
))
2149 unix_state_unlock(sk
);
2153 if (wq_has_sleeper(&u
->peer_wait
))
2154 wake_up_interruptible_sync_poll(&u
->peer_wait
,
2155 POLLOUT
| POLLWRNORM
|
2159 unix_copy_addr(msg
, skb
->sk
);
2161 if (size
> skb
->len
- skip
)
2162 size
= skb
->len
- skip
;
2163 else if (size
< skb
->len
- skip
)
2164 msg
->msg_flags
|= MSG_TRUNC
;
2166 err
= skb_copy_datagram_msg(skb
, skip
, msg
, size
);
2170 if (sock_flag(sk
, SOCK_RCVTSTAMP
))
2171 __sock_recv_timestamp(msg
, sk
, skb
);
2173 memset(&scm
, 0, sizeof(scm
));
2175 scm_set_cred(&scm
, UNIXCB(skb
).pid
, UNIXCB(skb
).uid
, UNIXCB(skb
).gid
);
2176 unix_set_secdata(&scm
, skb
);
2178 if (!(flags
& MSG_PEEK
)) {
2180 unix_detach_fds(&scm
, skb
);
2182 sk_peek_offset_bwd(sk
, skb
->len
);
2184 /* It is questionable: on PEEK we could:
2185 - do not return fds - good, but too simple 8)
2186 - return fds, and do not return them on read (old strategy,
2188 - clone fds (I chose it for now, it is the most universal
2191 POSIX 1003.1g does not actually define this clearly
2192 at all. POSIX 1003.1g doesn't define a lot of things
2197 sk_peek_offset_fwd(sk
, size
);
2200 scm
.fp
= scm_fp_dup(UNIXCB(skb
).fp
);
2202 err
= (flags
& MSG_TRUNC
) ? skb
->len
- skip
: size
;
2204 scm_recv(sock
, msg
, &scm
, flags
);
2207 skb_free_datagram(sk
, skb
);
2208 mutex_unlock(&u
->readlock
);
2214 * Sleep until more data has arrived. But check for races..
2216 static long unix_stream_data_wait(struct sock
*sk
, long timeo
,
2217 struct sk_buff
*last
, unsigned int last_len
)
2219 struct sk_buff
*tail
;
2222 unix_state_lock(sk
);
2225 prepare_to_wait(sk_sleep(sk
), &wait
, TASK_INTERRUPTIBLE
);
2227 tail
= skb_peek_tail(&sk
->sk_receive_queue
);
2229 (tail
&& tail
->len
!= last_len
) ||
2231 (sk
->sk_shutdown
& RCV_SHUTDOWN
) ||
2232 signal_pending(current
) ||
2236 sk_set_bit(SOCKWQ_ASYNC_WAITDATA
, sk
);
2237 unix_state_unlock(sk
);
2238 timeo
= freezable_schedule_timeout(timeo
);
2239 unix_state_lock(sk
);
2241 if (sock_flag(sk
, SOCK_DEAD
))
2244 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA
, sk
);
2247 finish_wait(sk_sleep(sk
), &wait
);
2248 unix_state_unlock(sk
);
2252 static unsigned int unix_skb_len(const struct sk_buff
*skb
)
2254 return skb
->len
- UNIXCB(skb
).consumed
;
2257 struct unix_stream_read_state
{
2258 int (*recv_actor
)(struct sk_buff
*, int, int,
2259 struct unix_stream_read_state
*);
2260 struct socket
*socket
;
2262 struct pipe_inode_info
*pipe
;
2265 unsigned int splice_flags
;
2268 static int unix_stream_read_generic(struct unix_stream_read_state
*state
)
2270 struct scm_cookie scm
;
2271 struct socket
*sock
= state
->socket
;
2272 struct sock
*sk
= sock
->sk
;
2273 struct unix_sock
*u
= unix_sk(sk
);
2275 int flags
= state
->flags
;
2276 int noblock
= flags
& MSG_DONTWAIT
;
2277 bool check_creds
= false;
2282 size_t size
= state
->size
;
2283 unsigned int last_len
;
2285 if (unlikely(sk
->sk_state
!= TCP_ESTABLISHED
)) {
2290 if (unlikely(flags
& MSG_OOB
)) {
2295 target
= sock_rcvlowat(sk
, flags
& MSG_WAITALL
, size
);
2296 timeo
= sock_rcvtimeo(sk
, noblock
);
2298 memset(&scm
, 0, sizeof(scm
));
2300 /* Lock the socket to prevent queue disordering
2301 * while sleeps in memcpy_tomsg
2303 mutex_lock(&u
->readlock
);
2305 if (flags
& MSG_PEEK
)
2306 skip
= sk_peek_offset(sk
, flags
);
2313 struct sk_buff
*skb
, *last
;
2316 unix_state_lock(sk
);
2317 if (sock_flag(sk
, SOCK_DEAD
)) {
2321 last
= skb
= skb_peek(&sk
->sk_receive_queue
);
2322 last_len
= last
? last
->len
: 0;
2325 unix_sk(sk
)->recursion_level
= 0;
2326 if (copied
>= target
)
2330 * POSIX 1003.1g mandates this order.
2333 err
= sock_error(sk
);
2336 if (sk
->sk_shutdown
& RCV_SHUTDOWN
)
2339 unix_state_unlock(sk
);
2345 mutex_unlock(&u
->readlock
);
2347 timeo
= unix_stream_data_wait(sk
, timeo
, last
,
2350 if (signal_pending(current
)) {
2351 err
= sock_intr_errno(timeo
);
2356 mutex_lock(&u
->readlock
);
2359 unix_state_unlock(sk
);
2363 while (skip
>= unix_skb_len(skb
)) {
2364 skip
-= unix_skb_len(skb
);
2366 last_len
= skb
->len
;
2367 skb
= skb_peek_next(skb
, &sk
->sk_receive_queue
);
2372 unix_state_unlock(sk
);
2375 /* Never glue messages from different writers */
2376 if (!unix_skb_scm_eq(skb
, &scm
))
2378 } else if (test_bit(SOCK_PASSCRED
, &sock
->flags
)) {
2379 /* Copy credentials */
2380 scm_set_cred(&scm
, UNIXCB(skb
).pid
, UNIXCB(skb
).uid
, UNIXCB(skb
).gid
);
2381 unix_set_secdata(&scm
, skb
);
2385 /* Copy address just once */
2386 if (state
->msg
&& state
->msg
->msg_name
) {
2387 DECLARE_SOCKADDR(struct sockaddr_un
*, sunaddr
,
2388 state
->msg
->msg_name
);
2389 unix_copy_addr(state
->msg
, skb
->sk
);
2393 chunk
= min_t(unsigned int, unix_skb_len(skb
) - skip
, size
);
2395 chunk
= state
->recv_actor(skb
, skip
, chunk
, state
);
2396 drop_skb
= !unix_skb_len(skb
);
2397 /* skb is only safe to use if !drop_skb */
2408 /* the skb was touched by a concurrent reader;
2409 * we should not expect anything from this skb
2410 * anymore and assume it invalid - we can be
2411 * sure it was dropped from the socket queue
2413 * let's report a short read
2419 /* Mark read part of skb as used */
2420 if (!(flags
& MSG_PEEK
)) {
2421 UNIXCB(skb
).consumed
+= chunk
;
2423 sk_peek_offset_bwd(sk
, chunk
);
2426 unix_detach_fds(&scm
, skb
);
2428 if (unix_skb_len(skb
))
2431 skb_unlink(skb
, &sk
->sk_receive_queue
);
2437 /* It is questionable, see note in unix_dgram_recvmsg.
2440 scm
.fp
= scm_fp_dup(UNIXCB(skb
).fp
);
2442 sk_peek_offset_fwd(sk
, chunk
);
2449 last_len
= skb
->len
;
2450 unix_state_lock(sk
);
2451 skb
= skb_peek_next(skb
, &sk
->sk_receive_queue
);
2454 unix_state_unlock(sk
);
2459 mutex_unlock(&u
->readlock
);
2461 scm_recv(sock
, state
->msg
, &scm
, flags
);
2465 return copied
? : err
;
2468 static int unix_stream_read_actor(struct sk_buff
*skb
,
2469 int skip
, int chunk
,
2470 struct unix_stream_read_state
*state
)
2474 ret
= skb_copy_datagram_msg(skb
, UNIXCB(skb
).consumed
+ skip
,
2476 return ret
?: chunk
;
2479 static int unix_stream_recvmsg(struct socket
*sock
, struct msghdr
*msg
,
2480 size_t size
, int flags
)
2482 struct unix_stream_read_state state
= {
2483 .recv_actor
= unix_stream_read_actor
,
2490 return unix_stream_read_generic(&state
);
2493 static ssize_t
skb_unix_socket_splice(struct sock
*sk
,
2494 struct pipe_inode_info
*pipe
,
2495 struct splice_pipe_desc
*spd
)
2498 struct unix_sock
*u
= unix_sk(sk
);
2500 mutex_unlock(&u
->readlock
);
2501 ret
= splice_to_pipe(pipe
, spd
);
2502 mutex_lock(&u
->readlock
);
2507 static int unix_stream_splice_actor(struct sk_buff
*skb
,
2508 int skip
, int chunk
,
2509 struct unix_stream_read_state
*state
)
2511 return skb_splice_bits(skb
, state
->socket
->sk
,
2512 UNIXCB(skb
).consumed
+ skip
,
2513 state
->pipe
, chunk
, state
->splice_flags
,
2514 skb_unix_socket_splice
);
2517 static ssize_t
unix_stream_splice_read(struct socket
*sock
, loff_t
*ppos
,
2518 struct pipe_inode_info
*pipe
,
2519 size_t size
, unsigned int flags
)
2521 struct unix_stream_read_state state
= {
2522 .recv_actor
= unix_stream_splice_actor
,
2526 .splice_flags
= flags
,
2529 if (unlikely(*ppos
))
2532 if (sock
->file
->f_flags
& O_NONBLOCK
||
2533 flags
& SPLICE_F_NONBLOCK
)
2534 state
.flags
= MSG_DONTWAIT
;
2536 return unix_stream_read_generic(&state
);
2539 static int unix_shutdown(struct socket
*sock
, int mode
)
2541 struct sock
*sk
= sock
->sk
;
2544 if (mode
< SHUT_RD
|| mode
> SHUT_RDWR
)
2547 * SHUT_RD (0) -> RCV_SHUTDOWN (1)
2548 * SHUT_WR (1) -> SEND_SHUTDOWN (2)
2549 * SHUT_RDWR (2) -> SHUTDOWN_MASK (3)
2553 unix_state_lock(sk
);
2554 sk
->sk_shutdown
|= mode
;
2555 other
= unix_peer(sk
);
2558 unix_state_unlock(sk
);
2559 sk
->sk_state_change(sk
);
2562 (sk
->sk_type
== SOCK_STREAM
|| sk
->sk_type
== SOCK_SEQPACKET
)) {
2566 if (mode
&RCV_SHUTDOWN
)
2567 peer_mode
|= SEND_SHUTDOWN
;
2568 if (mode
&SEND_SHUTDOWN
)
2569 peer_mode
|= RCV_SHUTDOWN
;
2570 unix_state_lock(other
);
2571 other
->sk_shutdown
|= peer_mode
;
2572 unix_state_unlock(other
);
2573 other
->sk_state_change(other
);
2574 if (peer_mode
== SHUTDOWN_MASK
)
2575 sk_wake_async(other
, SOCK_WAKE_WAITD
, POLL_HUP
);
2576 else if (peer_mode
& RCV_SHUTDOWN
)
2577 sk_wake_async(other
, SOCK_WAKE_WAITD
, POLL_IN
);
2585 long unix_inq_len(struct sock
*sk
)
2587 struct sk_buff
*skb
;
2590 if (sk
->sk_state
== TCP_LISTEN
)
2593 spin_lock(&sk
->sk_receive_queue
.lock
);
2594 if (sk
->sk_type
== SOCK_STREAM
||
2595 sk
->sk_type
== SOCK_SEQPACKET
) {
2596 skb_queue_walk(&sk
->sk_receive_queue
, skb
)
2597 amount
+= unix_skb_len(skb
);
2599 skb
= skb_peek(&sk
->sk_receive_queue
);
2603 spin_unlock(&sk
->sk_receive_queue
.lock
);
2607 EXPORT_SYMBOL_GPL(unix_inq_len
);
2609 long unix_outq_len(struct sock
*sk
)
2611 return sk_wmem_alloc_get(sk
);
2613 EXPORT_SYMBOL_GPL(unix_outq_len
);
2615 static int unix_ioctl(struct socket
*sock
, unsigned int cmd
, unsigned long arg
)
2617 struct sock
*sk
= sock
->sk
;
2623 amount
= unix_outq_len(sk
);
2624 err
= put_user(amount
, (int __user
*)arg
);
2627 amount
= unix_inq_len(sk
);
2631 err
= put_user(amount
, (int __user
*)arg
);
2640 static unsigned int unix_poll(struct file
*file
, struct socket
*sock
, poll_table
*wait
)
2642 struct sock
*sk
= sock
->sk
;
2645 sock_poll_wait(file
, sk_sleep(sk
), wait
);
2648 /* exceptional events? */
2651 if (sk
->sk_shutdown
== SHUTDOWN_MASK
)
2653 if (sk
->sk_shutdown
& RCV_SHUTDOWN
)
2654 mask
|= POLLRDHUP
| POLLIN
| POLLRDNORM
;
2657 if (!skb_queue_empty(&sk
->sk_receive_queue
))
2658 mask
|= POLLIN
| POLLRDNORM
;
2660 /* Connection-based need to check for termination and startup */
2661 if ((sk
->sk_type
== SOCK_STREAM
|| sk
->sk_type
== SOCK_SEQPACKET
) &&
2662 sk
->sk_state
== TCP_CLOSE
)
2666 * we set writable also when the other side has shut down the
2667 * connection. This prevents stuck sockets.
2669 if (unix_writable(sk
))
2670 mask
|= POLLOUT
| POLLWRNORM
| POLLWRBAND
;
2675 static unsigned int unix_dgram_poll(struct file
*file
, struct socket
*sock
,
2678 struct sock
*sk
= sock
->sk
, *other
;
2679 unsigned int mask
, writable
;
2681 sock_poll_wait(file
, sk_sleep(sk
), wait
);
2684 /* exceptional events? */
2685 if (sk
->sk_err
|| !skb_queue_empty(&sk
->sk_error_queue
))
2687 (sock_flag(sk
, SOCK_SELECT_ERR_QUEUE
) ? POLLPRI
: 0);
2689 if (sk
->sk_shutdown
& RCV_SHUTDOWN
)
2690 mask
|= POLLRDHUP
| POLLIN
| POLLRDNORM
;
2691 if (sk
->sk_shutdown
== SHUTDOWN_MASK
)
2695 if (!skb_queue_empty(&sk
->sk_receive_queue
))
2696 mask
|= POLLIN
| POLLRDNORM
;
2698 /* Connection-based need to check for termination and startup */
2699 if (sk
->sk_type
== SOCK_SEQPACKET
) {
2700 if (sk
->sk_state
== TCP_CLOSE
)
2702 /* connection hasn't started yet? */
2703 if (sk
->sk_state
== TCP_SYN_SENT
)
2707 /* No write status requested, avoid expensive OUT tests. */
2708 if (!(poll_requested_events(wait
) & (POLLWRBAND
|POLLWRNORM
|POLLOUT
)))
2711 writable
= unix_writable(sk
);
2713 unix_state_lock(sk
);
2715 other
= unix_peer(sk
);
2716 if (other
&& unix_peer(other
) != sk
&&
2717 unix_recvq_full(other
) &&
2718 unix_dgram_peer_wake_me(sk
, other
))
2721 unix_state_unlock(sk
);
2725 mask
|= POLLOUT
| POLLWRNORM
| POLLWRBAND
;
2727 sk_set_bit(SOCKWQ_ASYNC_NOSPACE
, sk
);
2732 #ifdef CONFIG_PROC_FS
2734 #define BUCKET_SPACE (BITS_PER_LONG - (UNIX_HASH_BITS + 1) - 1)
2736 #define get_bucket(x) ((x) >> BUCKET_SPACE)
2737 #define get_offset(x) ((x) & ((1L << BUCKET_SPACE) - 1))
2738 #define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
2740 static struct sock
*unix_from_bucket(struct seq_file
*seq
, loff_t
*pos
)
2742 unsigned long offset
= get_offset(*pos
);
2743 unsigned long bucket
= get_bucket(*pos
);
2745 unsigned long count
= 0;
2747 for (sk
= sk_head(&unix_socket_table
[bucket
]); sk
; sk
= sk_next(sk
)) {
2748 if (sock_net(sk
) != seq_file_net(seq
))
2750 if (++count
== offset
)
2757 static struct sock
*unix_next_socket(struct seq_file
*seq
,
2761 unsigned long bucket
;
2763 while (sk
> (struct sock
*)SEQ_START_TOKEN
) {
2767 if (sock_net(sk
) == seq_file_net(seq
))
2772 sk
= unix_from_bucket(seq
, pos
);
2777 bucket
= get_bucket(*pos
) + 1;
2778 *pos
= set_bucket_offset(bucket
, 1);
2779 } while (bucket
< ARRAY_SIZE(unix_socket_table
));
2784 static void *unix_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2785 __acquires(unix_table_lock
)
2787 spin_lock(&unix_table_lock
);
2790 return SEQ_START_TOKEN
;
2792 if (get_bucket(*pos
) >= ARRAY_SIZE(unix_socket_table
))
2795 return unix_next_socket(seq
, NULL
, pos
);
2798 static void *unix_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2801 return unix_next_socket(seq
, v
, pos
);
2804 static void unix_seq_stop(struct seq_file
*seq
, void *v
)
2805 __releases(unix_table_lock
)
2807 spin_unlock(&unix_table_lock
);
2810 static int unix_seq_show(struct seq_file
*seq
, void *v
)
2813 if (v
== SEQ_START_TOKEN
)
2814 seq_puts(seq
, "Num RefCount Protocol Flags Type St "
2818 struct unix_sock
*u
= unix_sk(s
);
2821 seq_printf(seq
, "%pK: %08X %08X %08X %04X %02X %5lu",
2823 atomic_read(&s
->sk_refcnt
),
2825 s
->sk_state
== TCP_LISTEN
? __SO_ACCEPTCON
: 0,
2828 (s
->sk_state
== TCP_ESTABLISHED
? SS_CONNECTED
: SS_UNCONNECTED
) :
2829 (s
->sk_state
== TCP_ESTABLISHED
? SS_CONNECTING
: SS_DISCONNECTING
),
2837 len
= u
->addr
->len
- sizeof(short);
2838 if (!UNIX_ABSTRACT(s
))
2844 for ( ; i
< len
; i
++)
2845 seq_putc(seq
, u
->addr
->name
->sun_path
[i
]);
2847 unix_state_unlock(s
);
2848 seq_putc(seq
, '\n');
2854 static const struct seq_operations unix_seq_ops
= {
2855 .start
= unix_seq_start
,
2856 .next
= unix_seq_next
,
2857 .stop
= unix_seq_stop
,
2858 .show
= unix_seq_show
,
2861 static int unix_seq_open(struct inode
*inode
, struct file
*file
)
2863 return seq_open_net(inode
, file
, &unix_seq_ops
,
2864 sizeof(struct seq_net_private
));
2867 static const struct file_operations unix_seq_fops
= {
2868 .owner
= THIS_MODULE
,
2869 .open
= unix_seq_open
,
2871 .llseek
= seq_lseek
,
2872 .release
= seq_release_net
,
2877 static const struct net_proto_family unix_family_ops
= {
2879 .create
= unix_create
,
2880 .owner
= THIS_MODULE
,
2884 static int __net_init
unix_net_init(struct net
*net
)
2886 int error
= -ENOMEM
;
2888 net
->unx
.sysctl_max_dgram_qlen
= 10;
2889 if (unix_sysctl_register(net
))
2892 #ifdef CONFIG_PROC_FS
2893 if (!proc_create("unix", 0, net
->proc_net
, &unix_seq_fops
)) {
2894 unix_sysctl_unregister(net
);
2903 static void __net_exit
unix_net_exit(struct net
*net
)
2905 unix_sysctl_unregister(net
);
2906 remove_proc_entry("unix", net
->proc_net
);
2909 static struct pernet_operations unix_net_ops
= {
2910 .init
= unix_net_init
,
2911 .exit
= unix_net_exit
,
2914 static int __init
af_unix_init(void)
2918 BUILD_BUG_ON(sizeof(struct unix_skb_parms
) > FIELD_SIZEOF(struct sk_buff
, cb
));
2920 rc
= proto_register(&unix_proto
, 1);
2922 pr_crit("%s: Cannot create unix_sock SLAB cache!\n", __func__
);
2926 sock_register(&unix_family_ops
);
2927 register_pernet_subsys(&unix_net_ops
);
2932 static void __exit
af_unix_exit(void)
2934 sock_unregister(PF_UNIX
);
2935 proto_unregister(&unix_proto
);
2936 unregister_pernet_subsys(&unix_net_ops
);
2939 /* Earlier than device_initcall() so that other drivers invoking
2940 request_module() don't end up in a loop when modprobe tries
2941 to use a UNIX socket. But later than subsys_initcall() because
2942 we depend on stuff initialised there */
2943 fs_initcall(af_unix_init
);
2944 module_exit(af_unix_exit
);
2946 MODULE_LICENSE("GPL");
2947 MODULE_ALIAS_NETPROTO(PF_UNIX
);