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 memcpy(UNIXSID(skb), &scm->secid, sizeof(u32));
146 static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb)
148 scm->secid = *UNIXSID(skb);
151 static inline void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb)
154 static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb)
156 #endif /* CONFIG_SECURITY_NETWORK */
159 * SMP locking strategy:
160 * hash table is protected with spinlock unix_table_lock
161 * each socket state is protected by separate spin lock.
164 static inline unsigned int unix_hash_fold(__wsum n)
166 unsigned int hash = (__force unsigned int)csum_fold(n);
169 return hash&(UNIX_HASH_SIZE-1);
172 #define unix_peer(sk) (unix_sk(sk)->peer)
174 static inline int unix_our_peer(struct sock *sk, struct sock *osk)
176 return unix_peer(osk) == sk;
179 static inline int unix_may_send(struct sock *sk, struct sock *osk)
181 return unix_peer(osk) == NULL || unix_our_peer(sk, osk);
184 static inline int unix_recvq_full(struct sock const *sk)
186 return skb_queue_len(&sk->sk_receive_queue) > sk->sk_max_ack_backlog;
189 struct sock *unix_peer_get(struct sock *s)
197 unix_state_unlock(s);
200 EXPORT_SYMBOL_GPL(unix_peer_get);
202 static inline void unix_release_addr(struct unix_address *addr)
204 if (atomic_dec_and_test(&addr->refcnt))
209 * Check unix socket name:
210 * - should be not zero length.
211 * - if started by not zero, should be NULL terminated (FS object)
212 * - if started by zero, it is abstract name.
215 static int unix_mkname(struct sockaddr_un *sunaddr, int len, unsigned int *hashp)
217 if (len <= sizeof(short) || len > sizeof(*sunaddr))
219 if (!sunaddr || sunaddr->sun_family != AF_UNIX)
221 if (sunaddr->sun_path[0]) {
223 * This may look like an off by one error but it is a bit more
224 * subtle. 108 is the longest valid AF_UNIX path for a binding.
225 * sun_path[108] doesn't as such exist. However in kernel space
226 * we are guaranteed that it is a valid memory location in our
227 * kernel address buffer.
229 ((char *)sunaddr)[len] = 0;
230 len = strlen(sunaddr->sun_path)+1+sizeof(short);
234 *hashp = unix_hash_fold(csum_partial(sunaddr, len, 0));
238 static void __unix_remove_socket(struct sock *sk)
240 sk_del_node_init(sk);
243 static void __unix_insert_socket(struct hlist_head *list, struct sock *sk)
245 WARN_ON(!sk_unhashed(sk));
246 sk_add_node(sk, list);
249 static inline void unix_remove_socket(struct sock *sk)
251 spin_lock(&unix_table_lock);
252 __unix_remove_socket(sk);
253 spin_unlock(&unix_table_lock);
256 static inline void unix_insert_socket(struct hlist_head *list, struct sock *sk)
258 spin_lock(&unix_table_lock);
259 __unix_insert_socket(list, sk);
260 spin_unlock(&unix_table_lock);
263 static struct sock *__unix_find_socket_byname(struct net *net,
264 struct sockaddr_un *sunname,
265 int len, int type, unsigned int hash)
269 sk_for_each(s, &unix_socket_table[hash ^ type]) {
270 struct unix_sock *u = unix_sk(s);
272 if (!net_eq(sock_net(s), net))
275 if (u->addr->len == len &&
276 !memcmp(u->addr->name, sunname, len))
284 static inline struct sock *unix_find_socket_byname(struct net *net,
285 struct sockaddr_un *sunname,
291 spin_lock(&unix_table_lock);
292 s = __unix_find_socket_byname(net, sunname, len, type, hash);
295 spin_unlock(&unix_table_lock);
299 static struct sock *unix_find_socket_byinode(struct inode *i)
303 spin_lock(&unix_table_lock);
305 &unix_socket_table[i->i_ino & (UNIX_HASH_SIZE - 1)]) {
306 struct dentry *dentry = unix_sk(s)->path.dentry;
308 if (dentry && dentry->d_inode == i) {
315 spin_unlock(&unix_table_lock);
319 static inline int unix_writable(struct sock *sk)
321 return (atomic_read(&sk->sk_wmem_alloc) << 2) <= sk->sk_sndbuf;
324 static void unix_write_space(struct sock *sk)
326 struct socket_wq *wq;
329 if (unix_writable(sk)) {
330 wq = rcu_dereference(sk->sk_wq);
331 if (wq_has_sleeper(wq))
332 wake_up_interruptible_sync_poll(&wq->wait,
333 POLLOUT | POLLWRNORM | POLLWRBAND);
334 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
339 /* When dgram socket disconnects (or changes its peer), we clear its receive
340 * queue of packets arrived from previous peer. First, it allows to do
341 * flow control based only on wmem_alloc; second, sk connected to peer
342 * may receive messages only from that peer. */
343 static void unix_dgram_disconnected(struct sock *sk, struct sock *other)
345 if (!skb_queue_empty(&sk->sk_receive_queue)) {
346 skb_queue_purge(&sk->sk_receive_queue);
347 wake_up_interruptible_all(&unix_sk(sk)->peer_wait);
349 /* If one link of bidirectional dgram pipe is disconnected,
350 * we signal error. Messages are lost. Do not make this,
351 * when peer was not connected to us.
353 if (!sock_flag(other, SOCK_DEAD) && unix_peer(other) == sk) {
354 other->sk_err = ECONNRESET;
355 other->sk_error_report(other);
360 static void unix_sock_destructor(struct sock *sk)
362 struct unix_sock *u = unix_sk(sk);
364 skb_queue_purge(&sk->sk_receive_queue);
366 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
367 WARN_ON(!sk_unhashed(sk));
368 WARN_ON(sk->sk_socket);
369 if (!sock_flag(sk, SOCK_DEAD)) {
370 pr_info("Attempt to release alive unix socket: %p\n", sk);
375 unix_release_addr(u->addr);
377 atomic_long_dec(&unix_nr_socks);
379 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
381 #ifdef UNIX_REFCNT_DEBUG
382 pr_debug("UNIX %p is destroyed, %ld are still alive.\n", sk,
383 atomic_long_read(&unix_nr_socks));
387 static void unix_release_sock(struct sock *sk, int embrion)
389 struct unix_sock *u = unix_sk(sk);
395 unix_remove_socket(sk);
400 sk->sk_shutdown = SHUTDOWN_MASK;
402 u->path.dentry = NULL;
404 state = sk->sk_state;
405 sk->sk_state = TCP_CLOSE;
406 unix_state_unlock(sk);
408 wake_up_interruptible_all(&u->peer_wait);
410 skpair = unix_peer(sk);
412 if (skpair != NULL) {
413 if (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) {
414 unix_state_lock(skpair);
416 skpair->sk_shutdown = SHUTDOWN_MASK;
417 if (!skb_queue_empty(&sk->sk_receive_queue) || embrion)
418 skpair->sk_err = ECONNRESET;
419 unix_state_unlock(skpair);
420 skpair->sk_state_change(skpair);
421 sk_wake_async(skpair, SOCK_WAKE_WAITD, POLL_HUP);
423 sock_put(skpair); /* It may now die */
424 unix_peer(sk) = NULL;
427 /* Try to flush out this socket. Throw out buffers at least */
429 while ((skb = skb_dequeue(&sk->sk_receive_queue)) != NULL) {
430 if (state == TCP_LISTEN)
431 unix_release_sock(skb->sk, 1);
432 /* passed fds are erased in the kfree_skb hook */
441 /* ---- Socket is dead now and most probably destroyed ---- */
444 * Fixme: BSD difference: In BSD all sockets connected to us get
445 * ECONNRESET and we die on the spot. In Linux we behave
446 * like files and pipes do and wait for the last
449 * Can't we simply set sock->err?
451 * What the above comment does talk about? --ANK(980817)
454 if (unix_tot_inflight)
455 unix_gc(); /* Garbage collect fds */
458 static void init_peercred(struct sock *sk)
460 put_pid(sk->sk_peer_pid);
461 if (sk->sk_peer_cred)
462 put_cred(sk->sk_peer_cred);
463 sk->sk_peer_pid = get_pid(task_tgid(current));
464 sk->sk_peer_cred = get_current_cred();
467 static void copy_peercred(struct sock *sk, struct sock *peersk)
469 put_pid(sk->sk_peer_pid);
470 if (sk->sk_peer_cred)
471 put_cred(sk->sk_peer_cred);
472 sk->sk_peer_pid = get_pid(peersk->sk_peer_pid);
473 sk->sk_peer_cred = get_cred(peersk->sk_peer_cred);
476 static int unix_listen(struct socket *sock, int backlog)
479 struct sock *sk = sock->sk;
480 struct unix_sock *u = unix_sk(sk);
481 struct pid *old_pid = NULL;
484 if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
485 goto out; /* Only stream/seqpacket sockets accept */
488 goto out; /* No listens on an unbound socket */
490 if (sk->sk_state != TCP_CLOSE && sk->sk_state != TCP_LISTEN)
492 if (backlog > sk->sk_max_ack_backlog)
493 wake_up_interruptible_all(&u->peer_wait);
494 sk->sk_max_ack_backlog = backlog;
495 sk->sk_state = TCP_LISTEN;
496 /* set credentials so connect can copy them */
501 unix_state_unlock(sk);
507 static int unix_release(struct socket *);
508 static int unix_bind(struct socket *, struct sockaddr *, int);
509 static int unix_stream_connect(struct socket *, struct sockaddr *,
510 int addr_len, int flags);
511 static int unix_socketpair(struct socket *, struct socket *);
512 static int unix_accept(struct socket *, struct socket *, int);
513 static int unix_getname(struct socket *, struct sockaddr *, int *, int);
514 static unsigned int unix_poll(struct file *, struct socket *, poll_table *);
515 static unsigned int unix_dgram_poll(struct file *, struct socket *,
517 static int unix_ioctl(struct socket *, unsigned int, unsigned long);
518 static int unix_shutdown(struct socket *, int);
519 static int unix_stream_sendmsg(struct kiocb *, struct socket *,
520 struct msghdr *, size_t);
521 static int unix_stream_recvmsg(struct kiocb *, struct socket *,
522 struct msghdr *, size_t, int);
523 static int unix_dgram_sendmsg(struct kiocb *, struct socket *,
524 struct msghdr *, size_t);
525 static int unix_dgram_recvmsg(struct kiocb *, struct socket *,
526 struct msghdr *, size_t, int);
527 static int unix_dgram_connect(struct socket *, struct sockaddr *,
529 static int unix_seqpacket_sendmsg(struct kiocb *, struct socket *,
530 struct msghdr *, size_t);
531 static int unix_seqpacket_recvmsg(struct kiocb *, struct socket *,
532 struct msghdr *, size_t, int);
534 static int unix_set_peek_off(struct sock *sk, int val)
536 struct unix_sock *u = unix_sk(sk);
538 if (mutex_lock_interruptible(&u->readlock))
541 sk->sk_peek_off = val;
542 mutex_unlock(&u->readlock);
548 static const struct proto_ops unix_stream_ops = {
550 .owner = THIS_MODULE,
551 .release = unix_release,
553 .connect = unix_stream_connect,
554 .socketpair = unix_socketpair,
555 .accept = unix_accept,
556 .getname = unix_getname,
559 .listen = unix_listen,
560 .shutdown = unix_shutdown,
561 .setsockopt = sock_no_setsockopt,
562 .getsockopt = sock_no_getsockopt,
563 .sendmsg = unix_stream_sendmsg,
564 .recvmsg = unix_stream_recvmsg,
565 .mmap = sock_no_mmap,
566 .sendpage = sock_no_sendpage,
567 .set_peek_off = unix_set_peek_off,
570 static const struct proto_ops unix_dgram_ops = {
572 .owner = THIS_MODULE,
573 .release = unix_release,
575 .connect = unix_dgram_connect,
576 .socketpair = unix_socketpair,
577 .accept = sock_no_accept,
578 .getname = unix_getname,
579 .poll = unix_dgram_poll,
581 .listen = sock_no_listen,
582 .shutdown = unix_shutdown,
583 .setsockopt = sock_no_setsockopt,
584 .getsockopt = sock_no_getsockopt,
585 .sendmsg = unix_dgram_sendmsg,
586 .recvmsg = unix_dgram_recvmsg,
587 .mmap = sock_no_mmap,
588 .sendpage = sock_no_sendpage,
589 .set_peek_off = unix_set_peek_off,
592 static const struct proto_ops unix_seqpacket_ops = {
594 .owner = THIS_MODULE,
595 .release = unix_release,
597 .connect = unix_stream_connect,
598 .socketpair = unix_socketpair,
599 .accept = unix_accept,
600 .getname = unix_getname,
601 .poll = unix_dgram_poll,
603 .listen = unix_listen,
604 .shutdown = unix_shutdown,
605 .setsockopt = sock_no_setsockopt,
606 .getsockopt = sock_no_getsockopt,
607 .sendmsg = unix_seqpacket_sendmsg,
608 .recvmsg = unix_seqpacket_recvmsg,
609 .mmap = sock_no_mmap,
610 .sendpage = sock_no_sendpage,
611 .set_peek_off = unix_set_peek_off,
614 static struct proto unix_proto = {
616 .owner = THIS_MODULE,
617 .obj_size = sizeof(struct unix_sock),
621 * AF_UNIX sockets do not interact with hardware, hence they
622 * dont trigger interrupts - so it's safe for them to have
623 * bh-unsafe locking for their sk_receive_queue.lock. Split off
624 * this special lock-class by reinitializing the spinlock key:
626 static struct lock_class_key af_unix_sk_receive_queue_lock_key;
628 static struct sock *unix_create1(struct net *net, struct socket *sock)
630 struct sock *sk = NULL;
633 atomic_long_inc(&unix_nr_socks);
634 if (atomic_long_read(&unix_nr_socks) > 2 * get_max_files())
637 sk = sk_alloc(net, PF_UNIX, GFP_KERNEL, &unix_proto);
641 sock_init_data(sock, sk);
642 lockdep_set_class(&sk->sk_receive_queue.lock,
643 &af_unix_sk_receive_queue_lock_key);
645 sk->sk_write_space = unix_write_space;
646 sk->sk_max_ack_backlog = net->unx.sysctl_max_dgram_qlen;
647 sk->sk_destruct = unix_sock_destructor;
649 u->path.dentry = NULL;
651 spin_lock_init(&u->lock);
652 atomic_long_set(&u->inflight, 0);
653 INIT_LIST_HEAD(&u->link);
654 mutex_init(&u->readlock); /* single task reading lock */
655 init_waitqueue_head(&u->peer_wait);
656 unix_insert_socket(unix_sockets_unbound(sk), sk);
659 atomic_long_dec(&unix_nr_socks);
662 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
668 static int unix_create(struct net *net, struct socket *sock, int protocol,
671 if (protocol && protocol != PF_UNIX)
672 return -EPROTONOSUPPORT;
674 sock->state = SS_UNCONNECTED;
676 switch (sock->type) {
678 sock->ops = &unix_stream_ops;
681 * Believe it or not BSD has AF_UNIX, SOCK_RAW though
685 sock->type = SOCK_DGRAM;
687 sock->ops = &unix_dgram_ops;
690 sock->ops = &unix_seqpacket_ops;
693 return -ESOCKTNOSUPPORT;
696 return unix_create1(net, sock) ? 0 : -ENOMEM;
699 static int unix_release(struct socket *sock)
701 struct sock *sk = sock->sk;
706 unix_release_sock(sk, 0);
712 static int unix_autobind(struct socket *sock)
714 struct sock *sk = sock->sk;
715 struct net *net = sock_net(sk);
716 struct unix_sock *u = unix_sk(sk);
717 static u32 ordernum = 1;
718 struct unix_address *addr;
720 unsigned int retries = 0;
722 err = mutex_lock_interruptible(&u->readlock);
731 addr = kzalloc(sizeof(*addr) + sizeof(short) + 16, GFP_KERNEL);
735 addr->name->sun_family = AF_UNIX;
736 atomic_set(&addr->refcnt, 1);
739 addr->len = sprintf(addr->name->sun_path+1, "%05x", ordernum) + 1 + sizeof(short);
740 addr->hash = unix_hash_fold(csum_partial(addr->name, addr->len, 0));
742 spin_lock(&unix_table_lock);
743 ordernum = (ordernum+1)&0xFFFFF;
745 if (__unix_find_socket_byname(net, addr->name, addr->len, sock->type,
747 spin_unlock(&unix_table_lock);
749 * __unix_find_socket_byname() may take long time if many names
750 * are already in use.
753 /* Give up if all names seems to be in use. */
754 if (retries++ == 0xFFFFF) {
761 addr->hash ^= sk->sk_type;
763 __unix_remove_socket(sk);
765 __unix_insert_socket(&unix_socket_table[addr->hash], sk);
766 spin_unlock(&unix_table_lock);
769 out: mutex_unlock(&u->readlock);
773 static struct sock *unix_find_other(struct net *net,
774 struct sockaddr_un *sunname, int len,
775 int type, unsigned int hash, int *error)
781 if (sunname->sun_path[0]) {
783 err = kern_path(sunname->sun_path, LOOKUP_FOLLOW, &path);
786 inode = path.dentry->d_inode;
787 err = inode_permission(inode, MAY_WRITE);
792 if (!S_ISSOCK(inode->i_mode))
794 u = unix_find_socket_byinode(inode);
798 if (u->sk_type == type)
804 if (u->sk_type != type) {
810 u = unix_find_socket_byname(net, sunname, len, type, hash);
812 struct dentry *dentry;
813 dentry = unix_sk(u)->path.dentry;
815 touch_atime(&unix_sk(u)->path);
828 static int unix_mknod(const char *sun_path, umode_t mode, struct path *res)
830 struct dentry *dentry;
834 * Get the parent directory, calculate the hash for last
837 dentry = kern_path_create(AT_FDCWD, sun_path, &path, 0);
838 err = PTR_ERR(dentry);
843 * All right, let's create it.
845 err = security_path_mknod(&path, dentry, mode, 0);
847 err = vfs_mknod(path.dentry->d_inode, dentry, mode, 0);
849 res->mnt = mntget(path.mnt);
850 res->dentry = dget(dentry);
853 done_path_create(&path, dentry);
857 static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
859 struct sock *sk = sock->sk;
860 struct net *net = sock_net(sk);
861 struct unix_sock *u = unix_sk(sk);
862 struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
863 char *sun_path = sunaddr->sun_path;
866 struct unix_address *addr;
867 struct hlist_head *list;
870 if (sunaddr->sun_family != AF_UNIX)
873 if (addr_len == sizeof(short)) {
874 err = unix_autobind(sock);
878 err = unix_mkname(sunaddr, addr_len, &hash);
883 err = mutex_lock_interruptible(&u->readlock);
892 addr = kmalloc(sizeof(*addr)+addr_len, GFP_KERNEL);
896 memcpy(addr->name, sunaddr, addr_len);
897 addr->len = addr_len;
898 addr->hash = hash ^ sk->sk_type;
899 atomic_set(&addr->refcnt, 1);
903 umode_t mode = S_IFSOCK |
904 (SOCK_INODE(sock)->i_mode & ~current_umask());
905 err = unix_mknod(sun_path, mode, &path);
909 unix_release_addr(addr);
912 addr->hash = UNIX_HASH_SIZE;
913 hash = path.dentry->d_inode->i_ino & (UNIX_HASH_SIZE-1);
914 spin_lock(&unix_table_lock);
916 list = &unix_socket_table[hash];
918 spin_lock(&unix_table_lock);
920 if (__unix_find_socket_byname(net, sunaddr, addr_len,
921 sk->sk_type, hash)) {
922 unix_release_addr(addr);
926 list = &unix_socket_table[addr->hash];
930 __unix_remove_socket(sk);
932 __unix_insert_socket(list, sk);
935 spin_unlock(&unix_table_lock);
937 mutex_unlock(&u->readlock);
942 static void unix_state_double_lock(struct sock *sk1, struct sock *sk2)
944 if (unlikely(sk1 == sk2) || !sk2) {
945 unix_state_lock(sk1);
949 unix_state_lock(sk1);
950 unix_state_lock_nested(sk2);
952 unix_state_lock(sk2);
953 unix_state_lock_nested(sk1);
957 static void unix_state_double_unlock(struct sock *sk1, struct sock *sk2)
959 if (unlikely(sk1 == sk2) || !sk2) {
960 unix_state_unlock(sk1);
963 unix_state_unlock(sk1);
964 unix_state_unlock(sk2);
967 static int unix_dgram_connect(struct socket *sock, struct sockaddr *addr,
970 struct sock *sk = sock->sk;
971 struct net *net = sock_net(sk);
972 struct sockaddr_un *sunaddr = (struct sockaddr_un *)addr;
977 if (addr->sa_family != AF_UNSPEC) {
978 err = unix_mkname(sunaddr, alen, &hash);
983 if (test_bit(SOCK_PASSCRED, &sock->flags) &&
984 !unix_sk(sk)->addr && (err = unix_autobind(sock)) != 0)
988 other = unix_find_other(net, sunaddr, alen, sock->type, hash, &err);
992 unix_state_double_lock(sk, other);
994 /* Apparently VFS overslept socket death. Retry. */
995 if (sock_flag(other, SOCK_DEAD)) {
996 unix_state_double_unlock(sk, other);
1002 if (!unix_may_send(sk, other))
1005 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1011 * 1003.1g breaking connected state with AF_UNSPEC
1014 unix_state_double_lock(sk, other);
1018 * If it was connected, reconnect.
1020 if (unix_peer(sk)) {
1021 struct sock *old_peer = unix_peer(sk);
1022 unix_peer(sk) = other;
1023 unix_state_double_unlock(sk, other);
1025 if (other != old_peer)
1026 unix_dgram_disconnected(sk, old_peer);
1029 unix_peer(sk) = other;
1030 unix_state_double_unlock(sk, other);
1035 unix_state_double_unlock(sk, other);
1041 static long unix_wait_for_peer(struct sock *other, long timeo)
1043 struct unix_sock *u = unix_sk(other);
1047 prepare_to_wait_exclusive(&u->peer_wait, &wait, TASK_INTERRUPTIBLE);
1049 sched = !sock_flag(other, SOCK_DEAD) &&
1050 !(other->sk_shutdown & RCV_SHUTDOWN) &&
1051 unix_recvq_full(other);
1053 unix_state_unlock(other);
1056 timeo = schedule_timeout(timeo);
1058 finish_wait(&u->peer_wait, &wait);
1062 static int unix_stream_connect(struct socket *sock, struct sockaddr *uaddr,
1063 int addr_len, int flags)
1065 struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
1066 struct sock *sk = sock->sk;
1067 struct net *net = sock_net(sk);
1068 struct unix_sock *u = unix_sk(sk), *newu, *otheru;
1069 struct sock *newsk = NULL;
1070 struct sock *other = NULL;
1071 struct sk_buff *skb = NULL;
1077 err = unix_mkname(sunaddr, addr_len, &hash);
1082 if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr &&
1083 (err = unix_autobind(sock)) != 0)
1086 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
1088 /* First of all allocate resources.
1089 If we will make it after state is locked,
1090 we will have to recheck all again in any case.
1095 /* create new sock for complete connection */
1096 newsk = unix_create1(sock_net(sk), NULL);
1100 /* Allocate skb for sending to listening sock */
1101 skb = sock_wmalloc(newsk, 1, 0, GFP_KERNEL);
1106 /* Find listening sock. */
1107 other = unix_find_other(net, sunaddr, addr_len, sk->sk_type, hash, &err);
1111 /* Latch state of peer */
1112 unix_state_lock(other);
1114 /* Apparently VFS overslept socket death. Retry. */
1115 if (sock_flag(other, SOCK_DEAD)) {
1116 unix_state_unlock(other);
1121 err = -ECONNREFUSED;
1122 if (other->sk_state != TCP_LISTEN)
1124 if (other->sk_shutdown & RCV_SHUTDOWN)
1127 if (unix_recvq_full(other)) {
1132 timeo = unix_wait_for_peer(other, timeo);
1134 err = sock_intr_errno(timeo);
1135 if (signal_pending(current))
1143 It is tricky place. We need to grab our state lock and cannot
1144 drop lock on peer. It is dangerous because deadlock is
1145 possible. Connect to self case and simultaneous
1146 attempt to connect are eliminated by checking socket
1147 state. other is TCP_LISTEN, if sk is TCP_LISTEN we
1148 check this before attempt to grab lock.
1150 Well, and we have to recheck the state after socket locked.
1156 /* This is ok... continue with connect */
1158 case TCP_ESTABLISHED:
1159 /* Socket is already connected */
1167 unix_state_lock_nested(sk);
1169 if (sk->sk_state != st) {
1170 unix_state_unlock(sk);
1171 unix_state_unlock(other);
1176 err = security_unix_stream_connect(sk, other, newsk);
1178 unix_state_unlock(sk);
1182 /* The way is open! Fastly set all the necessary fields... */
1185 unix_peer(newsk) = sk;
1186 newsk->sk_state = TCP_ESTABLISHED;
1187 newsk->sk_type = sk->sk_type;
1188 init_peercred(newsk);
1189 newu = unix_sk(newsk);
1190 RCU_INIT_POINTER(newsk->sk_wq, &newu->peer_wq);
1191 otheru = unix_sk(other);
1193 /* copy address information from listening to new sock*/
1195 atomic_inc(&otheru->addr->refcnt);
1196 newu->addr = otheru->addr;
1198 if (otheru->path.dentry) {
1199 path_get(&otheru->path);
1200 newu->path = otheru->path;
1203 /* Set credentials */
1204 copy_peercred(sk, other);
1206 sock->state = SS_CONNECTED;
1207 sk->sk_state = TCP_ESTABLISHED;
1210 smp_mb__after_atomic_inc(); /* sock_hold() does an atomic_inc() */
1211 unix_peer(sk) = newsk;
1213 unix_state_unlock(sk);
1215 /* take ten and and send info to listening sock */
1216 spin_lock(&other->sk_receive_queue.lock);
1217 __skb_queue_tail(&other->sk_receive_queue, skb);
1218 spin_unlock(&other->sk_receive_queue.lock);
1219 unix_state_unlock(other);
1220 other->sk_data_ready(other, 0);
1226 unix_state_unlock(other);
1231 unix_release_sock(newsk, 0);
1237 static int unix_socketpair(struct socket *socka, struct socket *sockb)
1239 struct sock *ska = socka->sk, *skb = sockb->sk;
1241 /* Join our sockets back to back */
1244 unix_peer(ska) = skb;
1245 unix_peer(skb) = ska;
1249 if (ska->sk_type != SOCK_DGRAM) {
1250 ska->sk_state = TCP_ESTABLISHED;
1251 skb->sk_state = TCP_ESTABLISHED;
1252 socka->state = SS_CONNECTED;
1253 sockb->state = SS_CONNECTED;
1258 static void unix_sock_inherit_flags(const struct socket *old,
1261 if (test_bit(SOCK_PASSCRED, &old->flags))
1262 set_bit(SOCK_PASSCRED, &new->flags);
1263 if (test_bit(SOCK_PASSSEC, &old->flags))
1264 set_bit(SOCK_PASSSEC, &new->flags);
1267 static int unix_accept(struct socket *sock, struct socket *newsock, int flags)
1269 struct sock *sk = sock->sk;
1271 struct sk_buff *skb;
1275 if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
1279 if (sk->sk_state != TCP_LISTEN)
1282 /* If socket state is TCP_LISTEN it cannot change (for now...),
1283 * so that no locks are necessary.
1286 skb = skb_recv_datagram(sk, 0, flags&O_NONBLOCK, &err);
1288 /* This means receive shutdown. */
1295 skb_free_datagram(sk, skb);
1296 wake_up_interruptible(&unix_sk(sk)->peer_wait);
1298 /* attach accepted sock to socket */
1299 unix_state_lock(tsk);
1300 newsock->state = SS_CONNECTED;
1301 unix_sock_inherit_flags(sock, newsock);
1302 sock_graft(tsk, newsock);
1303 unix_state_unlock(tsk);
1311 static int unix_getname(struct socket *sock, struct sockaddr *uaddr, int *uaddr_len, int peer)
1313 struct sock *sk = sock->sk;
1314 struct unix_sock *u;
1315 DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr, uaddr);
1319 sk = unix_peer_get(sk);
1330 unix_state_lock(sk);
1332 sunaddr->sun_family = AF_UNIX;
1333 sunaddr->sun_path[0] = 0;
1334 *uaddr_len = sizeof(short);
1336 struct unix_address *addr = u->addr;
1338 *uaddr_len = addr->len;
1339 memcpy(sunaddr, addr->name, *uaddr_len);
1341 unix_state_unlock(sk);
1347 static void unix_detach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1351 scm->fp = UNIXCB(skb).fp;
1352 UNIXCB(skb).fp = NULL;
1354 for (i = scm->fp->count-1; i >= 0; i--)
1355 unix_notinflight(scm->fp->fp[i]);
1358 static void unix_destruct_scm(struct sk_buff *skb)
1360 struct scm_cookie scm;
1361 memset(&scm, 0, sizeof(scm));
1362 scm.pid = UNIXCB(skb).pid;
1364 unix_detach_fds(&scm, skb);
1366 /* Alas, it calls VFS */
1367 /* So fscking what? fput() had been SMP-safe since the last Summer */
1372 #define MAX_RECURSION_LEVEL 4
1374 static int unix_attach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1377 unsigned char max_level = 0;
1378 int unix_sock_count = 0;
1380 for (i = scm->fp->count - 1; i >= 0; i--) {
1381 struct sock *sk = unix_get_socket(scm->fp->fp[i]);
1385 max_level = max(max_level,
1386 unix_sk(sk)->recursion_level);
1389 if (unlikely(max_level > MAX_RECURSION_LEVEL))
1390 return -ETOOMANYREFS;
1393 * Need to duplicate file references for the sake of garbage
1394 * collection. Otherwise a socket in the fps might become a
1395 * candidate for GC while the skb is not yet queued.
1397 UNIXCB(skb).fp = scm_fp_dup(scm->fp);
1398 if (!UNIXCB(skb).fp)
1401 if (unix_sock_count) {
1402 for (i = scm->fp->count - 1; i >= 0; i--)
1403 unix_inflight(scm->fp->fp[i]);
1408 static int unix_scm_to_skb(struct scm_cookie *scm, struct sk_buff *skb, bool send_fds)
1412 UNIXCB(skb).pid = get_pid(scm->pid);
1413 UNIXCB(skb).uid = scm->creds.uid;
1414 UNIXCB(skb).gid = scm->creds.gid;
1415 UNIXCB(skb).fp = NULL;
1416 if (scm->fp && send_fds)
1417 err = unix_attach_fds(scm, skb);
1419 skb->destructor = unix_destruct_scm;
1424 * Some apps rely on write() giving SCM_CREDENTIALS
1425 * We include credentials if source or destination socket
1426 * asserted SOCK_PASSCRED.
1428 static void maybe_add_creds(struct sk_buff *skb, const struct socket *sock,
1429 const struct sock *other)
1431 if (UNIXCB(skb).pid)
1433 if (test_bit(SOCK_PASSCRED, &sock->flags) ||
1434 !other->sk_socket ||
1435 test_bit(SOCK_PASSCRED, &other->sk_socket->flags)) {
1436 UNIXCB(skb).pid = get_pid(task_tgid(current));
1437 current_uid_gid(&UNIXCB(skb).uid, &UNIXCB(skb).gid);
1442 * Send AF_UNIX data.
1445 static int unix_dgram_sendmsg(struct kiocb *kiocb, struct socket *sock,
1446 struct msghdr *msg, size_t len)
1448 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1449 struct sock *sk = sock->sk;
1450 struct net *net = sock_net(sk);
1451 struct unix_sock *u = unix_sk(sk);
1452 DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr, msg->msg_name);
1453 struct sock *other = NULL;
1454 int namelen = 0; /* fake GCC */
1457 struct sk_buff *skb;
1459 struct scm_cookie tmp_scm;
1463 if (NULL == siocb->scm)
1464 siocb->scm = &tmp_scm;
1466 err = scm_send(sock, msg, siocb->scm, false);
1471 if (msg->msg_flags&MSG_OOB)
1474 if (msg->msg_namelen) {
1475 err = unix_mkname(sunaddr, msg->msg_namelen, &hash);
1482 other = unix_peer_get(sk);
1487 if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr
1488 && (err = unix_autobind(sock)) != 0)
1492 if (len > sk->sk_sndbuf - 32)
1495 if (len > SKB_MAX_ALLOC)
1496 data_len = min_t(size_t,
1497 len - SKB_MAX_ALLOC,
1498 MAX_SKB_FRAGS * PAGE_SIZE);
1500 skb = sock_alloc_send_pskb(sk, len - data_len, data_len,
1501 msg->msg_flags & MSG_DONTWAIT, &err,
1502 PAGE_ALLOC_COSTLY_ORDER);
1506 err = unix_scm_to_skb(siocb->scm, skb, true);
1509 max_level = err + 1;
1510 unix_get_secdata(siocb->scm, skb);
1512 skb_put(skb, len - data_len);
1513 skb->data_len = data_len;
1515 err = skb_copy_datagram_from_iovec(skb, 0, msg->msg_iov, 0, len);
1519 timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1524 if (sunaddr == NULL)
1527 other = unix_find_other(net, sunaddr, namelen, sk->sk_type,
1533 if (sk_filter(other, skb) < 0) {
1534 /* Toss the packet but do not return any error to the sender */
1539 unix_state_lock(other);
1541 if (!unix_may_send(sk, other))
1544 if (sock_flag(other, SOCK_DEAD)) {
1546 * Check with 1003.1g - what should
1549 unix_state_unlock(other);
1553 unix_state_lock(sk);
1554 if (unix_peer(sk) == other) {
1555 unix_peer(sk) = NULL;
1556 unix_state_unlock(sk);
1558 unix_dgram_disconnected(sk, other);
1560 err = -ECONNREFUSED;
1562 unix_state_unlock(sk);
1572 if (other->sk_shutdown & RCV_SHUTDOWN)
1575 if (sk->sk_type != SOCK_SEQPACKET) {
1576 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1581 if (unix_peer(other) != sk && unix_recvq_full(other)) {
1587 timeo = unix_wait_for_peer(other, timeo);
1589 err = sock_intr_errno(timeo);
1590 if (signal_pending(current))
1596 if (sock_flag(other, SOCK_RCVTSTAMP))
1597 __net_timestamp(skb);
1598 maybe_add_creds(skb, sock, other);
1599 skb_queue_tail(&other->sk_receive_queue, skb);
1600 if (max_level > unix_sk(other)->recursion_level)
1601 unix_sk(other)->recursion_level = max_level;
1602 unix_state_unlock(other);
1603 other->sk_data_ready(other, len);
1605 scm_destroy(siocb->scm);
1609 unix_state_unlock(other);
1615 scm_destroy(siocb->scm);
1619 /* We use paged skbs for stream sockets, and limit occupancy to 32768
1620 * bytes, and a minimun of a full page.
1622 #define UNIX_SKB_FRAGS_SZ (PAGE_SIZE << get_order(32768))
1624 static int unix_stream_sendmsg(struct kiocb *kiocb, struct socket *sock,
1625 struct msghdr *msg, size_t len)
1627 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1628 struct sock *sk = sock->sk;
1629 struct sock *other = NULL;
1631 struct sk_buff *skb;
1633 struct scm_cookie tmp_scm;
1634 bool fds_sent = false;
1638 if (NULL == siocb->scm)
1639 siocb->scm = &tmp_scm;
1641 err = scm_send(sock, msg, siocb->scm, false);
1646 if (msg->msg_flags&MSG_OOB)
1649 if (msg->msg_namelen) {
1650 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
1654 other = unix_peer(sk);
1659 if (sk->sk_shutdown & SEND_SHUTDOWN)
1662 while (sent < len) {
1665 /* Keep two messages in the pipe so it schedules better */
1666 size = min_t(int, size, (sk->sk_sndbuf >> 1) - 64);
1668 /* allow fallback to order-0 allocations */
1669 size = min_t(int, size, SKB_MAX_HEAD(0) + UNIX_SKB_FRAGS_SZ);
1671 data_len = max_t(int, 0, size - SKB_MAX_HEAD(0));
1673 skb = sock_alloc_send_pskb(sk, size - data_len, data_len,
1674 msg->msg_flags & MSG_DONTWAIT, &err,
1675 get_order(UNIX_SKB_FRAGS_SZ));
1679 /* Only send the fds in the first buffer */
1680 err = unix_scm_to_skb(siocb->scm, skb, !fds_sent);
1685 max_level = err + 1;
1688 skb_put(skb, size - data_len);
1689 skb->data_len = data_len;
1691 err = skb_copy_datagram_from_iovec(skb, 0, msg->msg_iov,
1698 unix_state_lock(other);
1700 if (sock_flag(other, SOCK_DEAD) ||
1701 (other->sk_shutdown & RCV_SHUTDOWN))
1704 maybe_add_creds(skb, sock, other);
1705 skb_queue_tail(&other->sk_receive_queue, skb);
1706 if (max_level > unix_sk(other)->recursion_level)
1707 unix_sk(other)->recursion_level = max_level;
1708 unix_state_unlock(other);
1709 other->sk_data_ready(other, size);
1713 scm_destroy(siocb->scm);
1719 unix_state_unlock(other);
1722 if (sent == 0 && !(msg->msg_flags&MSG_NOSIGNAL))
1723 send_sig(SIGPIPE, current, 0);
1726 scm_destroy(siocb->scm);
1728 return sent ? : err;
1731 static int unix_seqpacket_sendmsg(struct kiocb *kiocb, struct socket *sock,
1732 struct msghdr *msg, size_t len)
1735 struct sock *sk = sock->sk;
1737 err = sock_error(sk);
1741 if (sk->sk_state != TCP_ESTABLISHED)
1744 if (msg->msg_namelen)
1745 msg->msg_namelen = 0;
1747 return unix_dgram_sendmsg(kiocb, sock, msg, len);
1750 static int unix_seqpacket_recvmsg(struct kiocb *iocb, struct socket *sock,
1751 struct msghdr *msg, size_t size,
1754 struct sock *sk = sock->sk;
1756 if (sk->sk_state != TCP_ESTABLISHED)
1759 return unix_dgram_recvmsg(iocb, sock, msg, size, flags);
1762 static void unix_copy_addr(struct msghdr *msg, struct sock *sk)
1764 struct unix_sock *u = unix_sk(sk);
1767 msg->msg_namelen = u->addr->len;
1768 memcpy(msg->msg_name, u->addr->name, u->addr->len);
1772 static int unix_dgram_recvmsg(struct kiocb *iocb, struct socket *sock,
1773 struct msghdr *msg, size_t size,
1776 struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1777 struct scm_cookie tmp_scm;
1778 struct sock *sk = sock->sk;
1779 struct unix_sock *u = unix_sk(sk);
1780 int noblock = flags & MSG_DONTWAIT;
1781 struct sk_buff *skb;
1789 err = mutex_lock_interruptible(&u->readlock);
1791 err = sock_intr_errno(sock_rcvtimeo(sk, noblock));
1795 skip = sk_peek_offset(sk, flags);
1797 skb = __skb_recv_datagram(sk, flags, &peeked, &skip, &err);
1799 unix_state_lock(sk);
1800 /* Signal EOF on disconnected non-blocking SEQPACKET socket. */
1801 if (sk->sk_type == SOCK_SEQPACKET && err == -EAGAIN &&
1802 (sk->sk_shutdown & RCV_SHUTDOWN))
1804 unix_state_unlock(sk);
1808 wake_up_interruptible_sync_poll(&u->peer_wait,
1809 POLLOUT | POLLWRNORM | POLLWRBAND);
1812 unix_copy_addr(msg, skb->sk);
1814 if (size > skb->len - skip)
1815 size = skb->len - skip;
1816 else if (size < skb->len - skip)
1817 msg->msg_flags |= MSG_TRUNC;
1819 err = skb_copy_datagram_iovec(skb, skip, msg->msg_iov, size);
1823 if (sock_flag(sk, SOCK_RCVTSTAMP))
1824 __sock_recv_timestamp(msg, sk, skb);
1827 siocb->scm = &tmp_scm;
1828 memset(&tmp_scm, 0, sizeof(tmp_scm));
1830 scm_set_cred(siocb->scm, UNIXCB(skb).pid, UNIXCB(skb).uid, UNIXCB(skb).gid);
1831 unix_set_secdata(siocb->scm, skb);
1833 if (!(flags & MSG_PEEK)) {
1835 unix_detach_fds(siocb->scm, skb);
1837 sk_peek_offset_bwd(sk, skb->len);
1839 /* It is questionable: on PEEK we could:
1840 - do not return fds - good, but too simple 8)
1841 - return fds, and do not return them on read (old strategy,
1843 - clone fds (I chose it for now, it is the most universal
1846 POSIX 1003.1g does not actually define this clearly
1847 at all. POSIX 1003.1g doesn't define a lot of things
1852 sk_peek_offset_fwd(sk, size);
1855 siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1857 err = (flags & MSG_TRUNC) ? skb->len - skip : size;
1859 scm_recv(sock, msg, siocb->scm, flags);
1862 skb_free_datagram(sk, skb);
1864 mutex_unlock(&u->readlock);
1870 * Sleep until more data has arrived. But check for races..
1872 static long unix_stream_data_wait(struct sock *sk, long timeo,
1873 struct sk_buff *last)
1877 unix_state_lock(sk);
1880 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1882 if (skb_peek_tail(&sk->sk_receive_queue) != last ||
1884 (sk->sk_shutdown & RCV_SHUTDOWN) ||
1885 signal_pending(current) ||
1889 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1890 unix_state_unlock(sk);
1891 timeo = freezable_schedule_timeout(timeo);
1892 unix_state_lock(sk);
1893 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1896 finish_wait(sk_sleep(sk), &wait);
1897 unix_state_unlock(sk);
1901 static unsigned int unix_skb_len(const struct sk_buff *skb)
1903 return skb->len - UNIXCB(skb).consumed;
1906 static int unix_stream_recvmsg(struct kiocb *iocb, struct socket *sock,
1907 struct msghdr *msg, size_t size,
1910 struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1911 struct scm_cookie tmp_scm;
1912 struct sock *sk = sock->sk;
1913 struct unix_sock *u = unix_sk(sk);
1914 DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr, msg->msg_name);
1916 int check_creds = 0;
1923 if (sk->sk_state != TCP_ESTABLISHED)
1930 target = sock_rcvlowat(sk, flags&MSG_WAITALL, size);
1931 timeo = sock_rcvtimeo(sk, flags&MSG_DONTWAIT);
1933 /* Lock the socket to prevent queue disordering
1934 * while sleeps in memcpy_tomsg
1938 siocb->scm = &tmp_scm;
1939 memset(&tmp_scm, 0, sizeof(tmp_scm));
1942 err = mutex_lock_interruptible(&u->readlock);
1944 err = sock_intr_errno(timeo);
1950 struct sk_buff *skb, *last;
1952 unix_state_lock(sk);
1953 last = skb = skb_peek(&sk->sk_receive_queue);
1956 unix_sk(sk)->recursion_level = 0;
1957 if (copied >= target)
1961 * POSIX 1003.1g mandates this order.
1964 err = sock_error(sk);
1967 if (sk->sk_shutdown & RCV_SHUTDOWN)
1970 unix_state_unlock(sk);
1974 mutex_unlock(&u->readlock);
1976 timeo = unix_stream_data_wait(sk, timeo, last);
1978 if (signal_pending(current)
1979 || mutex_lock_interruptible(&u->readlock)) {
1980 err = sock_intr_errno(timeo);
1986 unix_state_unlock(sk);
1990 skip = sk_peek_offset(sk, flags);
1991 while (skip >= unix_skb_len(skb)) {
1992 skip -= unix_skb_len(skb);
1994 skb = skb_peek_next(skb, &sk->sk_receive_queue);
1999 unix_state_unlock(sk);
2002 /* Never glue messages from different writers */
2003 if ((UNIXCB(skb).pid != siocb->scm->pid) ||
2004 !uid_eq(UNIXCB(skb).uid, siocb->scm->creds.uid) ||
2005 !gid_eq(UNIXCB(skb).gid, siocb->scm->creds.gid))
2007 } else if (test_bit(SOCK_PASSCRED, &sock->flags)) {
2008 /* Copy credentials */
2009 scm_set_cred(siocb->scm, UNIXCB(skb).pid, UNIXCB(skb).uid, UNIXCB(skb).gid);
2013 /* Copy address just once */
2015 unix_copy_addr(msg, skb->sk);
2019 chunk = min_t(unsigned int, unix_skb_len(skb) - skip, size);
2020 if (skb_copy_datagram_iovec(skb, UNIXCB(skb).consumed + skip,
2021 msg->msg_iov, chunk)) {
2029 /* Mark read part of skb as used */
2030 if (!(flags & MSG_PEEK)) {
2031 UNIXCB(skb).consumed += chunk;
2033 sk_peek_offset_bwd(sk, chunk);
2036 unix_detach_fds(siocb->scm, skb);
2038 if (unix_skb_len(skb))
2041 skb_unlink(skb, &sk->sk_receive_queue);
2047 /* It is questionable, see note in unix_dgram_recvmsg.
2050 siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
2052 sk_peek_offset_fwd(sk, chunk);
2058 mutex_unlock(&u->readlock);
2059 scm_recv(sock, msg, siocb->scm, flags);
2061 return copied ? : err;
2064 static int unix_shutdown(struct socket *sock, int mode)
2066 struct sock *sk = sock->sk;
2069 if (mode < SHUT_RD || mode > SHUT_RDWR)
2072 * SHUT_RD (0) -> RCV_SHUTDOWN (1)
2073 * SHUT_WR (1) -> SEND_SHUTDOWN (2)
2074 * SHUT_RDWR (2) -> SHUTDOWN_MASK (3)
2078 unix_state_lock(sk);
2079 sk->sk_shutdown |= mode;
2080 other = unix_peer(sk);
2083 unix_state_unlock(sk);
2084 sk->sk_state_change(sk);
2087 (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET)) {
2091 if (mode&RCV_SHUTDOWN)
2092 peer_mode |= SEND_SHUTDOWN;
2093 if (mode&SEND_SHUTDOWN)
2094 peer_mode |= RCV_SHUTDOWN;
2095 unix_state_lock(other);
2096 other->sk_shutdown |= peer_mode;
2097 unix_state_unlock(other);
2098 other->sk_state_change(other);
2099 if (peer_mode == SHUTDOWN_MASK)
2100 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_HUP);
2101 else if (peer_mode & RCV_SHUTDOWN)
2102 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_IN);
2110 long unix_inq_len(struct sock *sk)
2112 struct sk_buff *skb;
2115 if (sk->sk_state == TCP_LISTEN)
2118 spin_lock(&sk->sk_receive_queue.lock);
2119 if (sk->sk_type == SOCK_STREAM ||
2120 sk->sk_type == SOCK_SEQPACKET) {
2121 skb_queue_walk(&sk->sk_receive_queue, skb)
2122 amount += unix_skb_len(skb);
2124 skb = skb_peek(&sk->sk_receive_queue);
2128 spin_unlock(&sk->sk_receive_queue.lock);
2132 EXPORT_SYMBOL_GPL(unix_inq_len);
2134 long unix_outq_len(struct sock *sk)
2136 return sk_wmem_alloc_get(sk);
2138 EXPORT_SYMBOL_GPL(unix_outq_len);
2140 static int unix_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2142 struct sock *sk = sock->sk;
2148 amount = unix_outq_len(sk);
2149 err = put_user(amount, (int __user *)arg);
2152 amount = unix_inq_len(sk);
2156 err = put_user(amount, (int __user *)arg);
2165 static unsigned int unix_poll(struct file *file, struct socket *sock, poll_table *wait)
2167 struct sock *sk = sock->sk;
2170 sock_poll_wait(file, sk_sleep(sk), wait);
2173 /* exceptional events? */
2176 if (sk->sk_shutdown == SHUTDOWN_MASK)
2178 if (sk->sk_shutdown & RCV_SHUTDOWN)
2179 mask |= POLLRDHUP | POLLIN | POLLRDNORM;
2182 if (!skb_queue_empty(&sk->sk_receive_queue))
2183 mask |= POLLIN | POLLRDNORM;
2185 /* Connection-based need to check for termination and startup */
2186 if ((sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) &&
2187 sk->sk_state == TCP_CLOSE)
2191 * we set writable also when the other side has shut down the
2192 * connection. This prevents stuck sockets.
2194 if (unix_writable(sk))
2195 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2200 static unsigned int unix_dgram_poll(struct file *file, struct socket *sock,
2203 struct sock *sk = sock->sk, *other;
2204 unsigned int mask, writable;
2206 sock_poll_wait(file, sk_sleep(sk), wait);
2209 /* exceptional events? */
2210 if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue))
2212 (sock_flag(sk, SOCK_SELECT_ERR_QUEUE) ? POLLPRI : 0);
2214 if (sk->sk_shutdown & RCV_SHUTDOWN)
2215 mask |= POLLRDHUP | POLLIN | POLLRDNORM;
2216 if (sk->sk_shutdown == SHUTDOWN_MASK)
2220 if (!skb_queue_empty(&sk->sk_receive_queue))
2221 mask |= POLLIN | POLLRDNORM;
2223 /* Connection-based need to check for termination and startup */
2224 if (sk->sk_type == SOCK_SEQPACKET) {
2225 if (sk->sk_state == TCP_CLOSE)
2227 /* connection hasn't started yet? */
2228 if (sk->sk_state == TCP_SYN_SENT)
2232 /* No write status requested, avoid expensive OUT tests. */
2233 if (!(poll_requested_events(wait) & (POLLWRBAND|POLLWRNORM|POLLOUT)))
2236 writable = unix_writable(sk);
2237 other = unix_peer_get(sk);
2239 if (unix_peer(other) != sk) {
2240 sock_poll_wait(file, &unix_sk(other)->peer_wait, wait);
2241 if (unix_recvq_full(other))
2248 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2250 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
2255 #ifdef CONFIG_PROC_FS
2257 #define BUCKET_SPACE (BITS_PER_LONG - (UNIX_HASH_BITS + 1) - 1)
2259 #define get_bucket(x) ((x) >> BUCKET_SPACE)
2260 #define get_offset(x) ((x) & ((1L << BUCKET_SPACE) - 1))
2261 #define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
2263 static struct sock *unix_from_bucket(struct seq_file *seq, loff_t *pos)
2265 unsigned long offset = get_offset(*pos);
2266 unsigned long bucket = get_bucket(*pos);
2268 unsigned long count = 0;
2270 for (sk = sk_head(&unix_socket_table[bucket]); sk; sk = sk_next(sk)) {
2271 if (sock_net(sk) != seq_file_net(seq))
2273 if (++count == offset)
2280 static struct sock *unix_next_socket(struct seq_file *seq,
2284 unsigned long bucket;
2286 while (sk > (struct sock *)SEQ_START_TOKEN) {
2290 if (sock_net(sk) == seq_file_net(seq))
2295 sk = unix_from_bucket(seq, pos);
2300 bucket = get_bucket(*pos) + 1;
2301 *pos = set_bucket_offset(bucket, 1);
2302 } while (bucket < ARRAY_SIZE(unix_socket_table));
2307 static void *unix_seq_start(struct seq_file *seq, loff_t *pos)
2308 __acquires(unix_table_lock)
2310 spin_lock(&unix_table_lock);
2313 return SEQ_START_TOKEN;
2315 if (get_bucket(*pos) >= ARRAY_SIZE(unix_socket_table))
2318 return unix_next_socket(seq, NULL, pos);
2321 static void *unix_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2324 return unix_next_socket(seq, v, pos);
2327 static void unix_seq_stop(struct seq_file *seq, void *v)
2328 __releases(unix_table_lock)
2330 spin_unlock(&unix_table_lock);
2333 static int unix_seq_show(struct seq_file *seq, void *v)
2336 if (v == SEQ_START_TOKEN)
2337 seq_puts(seq, "Num RefCount Protocol Flags Type St "
2341 struct unix_sock *u = unix_sk(s);
2344 seq_printf(seq, "%pK: %08X %08X %08X %04X %02X %5lu",
2346 atomic_read(&s->sk_refcnt),
2348 s->sk_state == TCP_LISTEN ? __SO_ACCEPTCON : 0,
2351 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTED : SS_UNCONNECTED) :
2352 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTING : SS_DISCONNECTING),
2360 len = u->addr->len - sizeof(short);
2361 if (!UNIX_ABSTRACT(s))
2367 for ( ; i < len; i++)
2368 seq_putc(seq, u->addr->name->sun_path[i]);
2370 unix_state_unlock(s);
2371 seq_putc(seq, '\n');
2377 static const struct seq_operations unix_seq_ops = {
2378 .start = unix_seq_start,
2379 .next = unix_seq_next,
2380 .stop = unix_seq_stop,
2381 .show = unix_seq_show,
2384 static int unix_seq_open(struct inode *inode, struct file *file)
2386 return seq_open_net(inode, file, &unix_seq_ops,
2387 sizeof(struct seq_net_private));
2390 static const struct file_operations unix_seq_fops = {
2391 .owner = THIS_MODULE,
2392 .open = unix_seq_open,
2394 .llseek = seq_lseek,
2395 .release = seq_release_net,
2400 static const struct net_proto_family unix_family_ops = {
2402 .create = unix_create,
2403 .owner = THIS_MODULE,
2407 static int __net_init unix_net_init(struct net *net)
2409 int error = -ENOMEM;
2411 net->unx.sysctl_max_dgram_qlen = 10;
2412 if (unix_sysctl_register(net))
2415 #ifdef CONFIG_PROC_FS
2416 if (!proc_create("unix", 0, net->proc_net, &unix_seq_fops)) {
2417 unix_sysctl_unregister(net);
2426 static void __net_exit unix_net_exit(struct net *net)
2428 unix_sysctl_unregister(net);
2429 remove_proc_entry("unix", net->proc_net);
2432 static struct pernet_operations unix_net_ops = {
2433 .init = unix_net_init,
2434 .exit = unix_net_exit,
2437 static int __init af_unix_init(void)
2441 BUILD_BUG_ON(sizeof(struct unix_skb_parms) > FIELD_SIZEOF(struct sk_buff, cb));
2443 rc = proto_register(&unix_proto, 1);
2445 pr_crit("%s: Cannot create unix_sock SLAB cache!\n", __func__);
2449 sock_register(&unix_family_ops);
2450 register_pernet_subsys(&unix_net_ops);
2455 static void __exit af_unix_exit(void)
2457 sock_unregister(PF_UNIX);
2458 proto_unregister(&unix_proto);
2459 unregister_pernet_subsys(&unix_net_ops);
2462 /* Earlier than device_initcall() so that other drivers invoking
2463 request_module() don't end up in a loop when modprobe tries
2464 to use a UNIX socket. But later than subsys_initcall() because
2465 we depend on stuff initialised there */
2466 fs_initcall(af_unix_init);
2467 module_exit(af_unix_exit);
2469 MODULE_LICENSE("GPL");
2470 MODULE_ALIAS_NETPROTO(PF_UNIX);