2 * NET4: Implementation of BSD Unix domain sockets.
4 * Authors: Alan Cox, <alan@lxorguk.ukuu.org.uk>
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License
8 * as published by the Free Software Foundation; either version
9 * 2 of the License, or (at your option) any later version.
12 * Linus Torvalds : Assorted bug cures.
13 * Niibe Yutaka : async I/O support.
14 * Carsten Paeth : PF_UNIX check, address fixes.
15 * Alan Cox : Limit size of allocated blocks.
16 * Alan Cox : Fixed the stupid socketpair bug.
17 * Alan Cox : BSD compatibility fine tuning.
18 * Alan Cox : Fixed a bug in connect when interrupted.
19 * Alan Cox : Sorted out a proper draft version of
20 * file descriptor passing hacked up from
22 * Marty Leisner : Fixes to fd passing
23 * Nick Nevin : recvmsg bugfix.
24 * Alan Cox : Started proper garbage collector
25 * Heiko EiBfeldt : Missing verify_area check
26 * Alan Cox : Started POSIXisms
27 * Andreas Schwab : Replace inode by dentry for proper
29 * Kirk Petersen : Made this a module
30 * Christoph Rohland : Elegant non-blocking accept/connect algorithm.
32 * Alexey Kuznetosv : Repaired (I hope) bugs introduces
33 * by above two patches.
34 * Andrea Arcangeli : If possible we block in connect(2)
35 * if the max backlog of the listen socket
36 * is been reached. This won't break
37 * old apps and it will avoid huge amount
38 * of socks hashed (this for unix_gc()
39 * performances reasons).
40 * Security fix that limits the max
41 * number of socks to 2*max_files and
42 * the number of skb queueable in the
44 * Artur Skawina : Hash function optimizations
45 * Alexey Kuznetsov : Full scale SMP. Lot of bugs are introduced 8)
46 * Malcolm Beattie : Set peercred for socketpair
47 * Michal Ostrowski : Module initialization cleanup.
48 * Arnaldo C. Melo : Remove MOD_{INC,DEC}_USE_COUNT,
49 * the core infrastructure is doing that
50 * for all net proto families now (2.5.69+)
53 * Known differences from reference BSD that was tested:
56 * ECONNREFUSED is not returned from one end of a connected() socket to the
57 * other the moment one end closes.
58 * fstat() doesn't return st_dev=0, and give the blksize as high water mark
59 * and a fake inode identifier (nor the BSD first socket fstat twice bug).
61 * accept() returns a path name even if the connecting socket has closed
62 * in the meantime (BSD loses the path and gives up).
63 * accept() returns 0 length path for an unbound connector. BSD returns 16
64 * and a null first byte in the path (but not for gethost/peername - BSD bug ??)
65 * socketpair(...SOCK_RAW..) doesn't panic the kernel.
66 * BSD af_unix apparently has connect forgetting to block properly.
67 * (need to check this with the POSIX spec in detail)
69 * Differences from 2.0.0-11-... (ANK)
70 * Bug fixes and improvements.
71 * - client shutdown killed server socket.
72 * - removed all useless cli/sti pairs.
74 * Semantic changes/extensions.
75 * - generic control message passing.
76 * - SCM_CREDENTIALS control message.
77 * - "Abstract" (not FS based) socket bindings.
78 * Abstract names are sequences of bytes (not zero terminated)
79 * started by 0, so that this name space does not intersect
83 #include <linux/module.h>
84 #include <linux/kernel.h>
85 #include <linux/signal.h>
86 #include <linux/sched.h>
87 #include <linux/errno.h>
88 #include <linux/string.h>
89 #include <linux/stat.h>
90 #include <linux/dcache.h>
91 #include <linux/namei.h>
92 #include <linux/socket.h>
94 #include <linux/fcntl.h>
95 #include <linux/termios.h>
96 #include <linux/sockios.h>
97 #include <linux/net.h>
100 #include <linux/slab.h>
101 #include <asm/uaccess.h>
102 #include <linux/skbuff.h>
103 #include <linux/netdevice.h>
104 #include <net/net_namespace.h>
105 #include <net/sock.h>
106 #include <net/tcp_states.h>
107 #include <net/af_unix.h>
108 #include <linux/proc_fs.h>
109 #include <linux/seq_file.h>
111 #include <linux/init.h>
112 #include <linux/poll.h>
113 #include <linux/rtnetlink.h>
114 #include <linux/mount.h>
115 #include <net/checksum.h>
116 #include <linux/security.h>
117 #include <linux/freezer.h>
119 struct hlist_head unix_socket_table[2 * UNIX_HASH_SIZE];
120 EXPORT_SYMBOL_GPL(unix_socket_table);
121 DEFINE_SPINLOCK(unix_table_lock);
122 EXPORT_SYMBOL_GPL(unix_table_lock);
123 static atomic_long_t unix_nr_socks;
126 static struct hlist_head *unix_sockets_unbound(void *addr)
128 unsigned long hash = (unsigned long)addr;
132 hash %= UNIX_HASH_SIZE;
133 return &unix_socket_table[UNIX_HASH_SIZE + hash];
136 #define UNIX_ABSTRACT(sk) (unix_sk(sk)->addr->hash < UNIX_HASH_SIZE)
138 #ifdef CONFIG_SECURITY_NETWORK
139 static void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb)
141 memcpy(UNIXSID(skb), &scm->secid, sizeof(u32));
144 static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb)
146 scm->secid = *UNIXSID(skb);
149 static inline void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb)
152 static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb)
154 #endif /* CONFIG_SECURITY_NETWORK */
157 * SMP locking strategy:
158 * hash table is protected with spinlock unix_table_lock
159 * each socket state is protected by separate spin lock.
162 static inline unsigned int unix_hash_fold(__wsum n)
164 unsigned int hash = (__force unsigned int)csum_fold(n);
167 return hash&(UNIX_HASH_SIZE-1);
170 #define unix_peer(sk) (unix_sk(sk)->peer)
172 static inline int unix_our_peer(struct sock *sk, struct sock *osk)
174 return unix_peer(osk) == sk;
177 static inline int unix_may_send(struct sock *sk, struct sock *osk)
179 return unix_peer(osk) == NULL || unix_our_peer(sk, osk);
182 static inline int unix_recvq_full(struct sock const *sk)
184 return skb_queue_len(&sk->sk_receive_queue) > sk->sk_max_ack_backlog;
187 struct sock *unix_peer_get(struct sock *s)
195 unix_state_unlock(s);
198 EXPORT_SYMBOL_GPL(unix_peer_get);
200 static inline void unix_release_addr(struct unix_address *addr)
202 if (atomic_dec_and_test(&addr->refcnt))
207 * Check unix socket name:
208 * - should be not zero length.
209 * - if started by not zero, should be NULL terminated (FS object)
210 * - if started by zero, it is abstract name.
213 static int unix_mkname(struct sockaddr_un *sunaddr, int len, unsigned int *hashp)
215 if (len <= sizeof(short) || len > sizeof(*sunaddr))
217 if (!sunaddr || sunaddr->sun_family != AF_UNIX)
219 if (sunaddr->sun_path[0]) {
221 * This may look like an off by one error but it is a bit more
222 * subtle. 108 is the longest valid AF_UNIX path for a binding.
223 * sun_path[108] doesn't as such exist. However in kernel space
224 * we are guaranteed that it is a valid memory location in our
225 * kernel address buffer.
227 ((char *)sunaddr)[len] = 0;
228 len = strlen(sunaddr->sun_path)+1+sizeof(short);
232 *hashp = unix_hash_fold(csum_partial(sunaddr, len, 0));
236 static void __unix_remove_socket(struct sock *sk)
238 sk_del_node_init(sk);
241 static void __unix_insert_socket(struct hlist_head *list, struct sock *sk)
243 WARN_ON(!sk_unhashed(sk));
244 sk_add_node(sk, list);
247 static inline void unix_remove_socket(struct sock *sk)
249 spin_lock(&unix_table_lock);
250 __unix_remove_socket(sk);
251 spin_unlock(&unix_table_lock);
254 static inline void unix_insert_socket(struct hlist_head *list, struct sock *sk)
256 spin_lock(&unix_table_lock);
257 __unix_insert_socket(list, sk);
258 spin_unlock(&unix_table_lock);
261 static struct sock *__unix_find_socket_byname(struct net *net,
262 struct sockaddr_un *sunname,
263 int len, int type, unsigned int hash)
267 sk_for_each(s, &unix_socket_table[hash ^ type]) {
268 struct unix_sock *u = unix_sk(s);
270 if (!net_eq(sock_net(s), net))
273 if (u->addr->len == len &&
274 !memcmp(u->addr->name, sunname, len))
282 static inline struct sock *unix_find_socket_byname(struct net *net,
283 struct sockaddr_un *sunname,
289 spin_lock(&unix_table_lock);
290 s = __unix_find_socket_byname(net, sunname, len, type, hash);
293 spin_unlock(&unix_table_lock);
297 static struct sock *unix_find_socket_byinode(struct inode *i)
301 spin_lock(&unix_table_lock);
303 &unix_socket_table[i->i_ino & (UNIX_HASH_SIZE - 1)]) {
304 struct dentry *dentry = unix_sk(s)->path.dentry;
306 if (dentry && dentry->d_inode == i) {
313 spin_unlock(&unix_table_lock);
317 static inline int unix_writable(struct sock *sk)
319 return (atomic_read(&sk->sk_wmem_alloc) << 2) <= sk->sk_sndbuf;
322 static void unix_write_space(struct sock *sk)
324 struct socket_wq *wq;
327 if (unix_writable(sk)) {
328 wq = rcu_dereference(sk->sk_wq);
329 if (wq_has_sleeper(wq))
330 wake_up_interruptible_sync_poll(&wq->wait,
331 POLLOUT | POLLWRNORM | POLLWRBAND);
332 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
337 /* When dgram socket disconnects (or changes its peer), we clear its receive
338 * queue of packets arrived from previous peer. First, it allows to do
339 * flow control based only on wmem_alloc; second, sk connected to peer
340 * may receive messages only from that peer. */
341 static void unix_dgram_disconnected(struct sock *sk, struct sock *other)
343 if (!skb_queue_empty(&sk->sk_receive_queue)) {
344 skb_queue_purge(&sk->sk_receive_queue);
345 wake_up_interruptible_all(&unix_sk(sk)->peer_wait);
347 /* If one link of bidirectional dgram pipe is disconnected,
348 * we signal error. Messages are lost. Do not make this,
349 * when peer was not connected to us.
351 if (!sock_flag(other, SOCK_DEAD) && unix_peer(other) == sk) {
352 other->sk_err = ECONNRESET;
353 other->sk_error_report(other);
358 static void unix_sock_destructor(struct sock *sk)
360 struct unix_sock *u = unix_sk(sk);
362 skb_queue_purge(&sk->sk_receive_queue);
364 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
365 WARN_ON(!sk_unhashed(sk));
366 WARN_ON(sk->sk_socket);
367 if (!sock_flag(sk, SOCK_DEAD)) {
368 printk(KERN_INFO "Attempt to release alive unix socket: %p\n", sk);
373 unix_release_addr(u->addr);
375 atomic_long_dec(&unix_nr_socks);
377 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
379 #ifdef UNIX_REFCNT_DEBUG
380 printk(KERN_DEBUG "UNIX %p is destroyed, %ld are still alive.\n", sk,
381 atomic_long_read(&unix_nr_socks));
385 static void unix_release_sock(struct sock *sk, int embrion)
387 struct unix_sock *u = unix_sk(sk);
393 unix_remove_socket(sk);
398 sk->sk_shutdown = SHUTDOWN_MASK;
400 u->path.dentry = NULL;
402 state = sk->sk_state;
403 sk->sk_state = TCP_CLOSE;
404 unix_state_unlock(sk);
406 wake_up_interruptible_all(&u->peer_wait);
408 skpair = unix_peer(sk);
410 if (skpair != NULL) {
411 if (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) {
412 unix_state_lock(skpair);
414 skpair->sk_shutdown = SHUTDOWN_MASK;
415 if (!skb_queue_empty(&sk->sk_receive_queue) || embrion)
416 skpair->sk_err = ECONNRESET;
417 unix_state_unlock(skpair);
418 skpair->sk_state_change(skpair);
419 sk_wake_async(skpair, SOCK_WAKE_WAITD, POLL_HUP);
421 sock_put(skpair); /* It may now die */
422 unix_peer(sk) = NULL;
425 /* Try to flush out this socket. Throw out buffers at least */
427 while ((skb = skb_dequeue(&sk->sk_receive_queue)) != NULL) {
428 if (state == TCP_LISTEN)
429 unix_release_sock(skb->sk, 1);
430 /* passed fds are erased in the kfree_skb hook */
439 /* ---- Socket is dead now and most probably destroyed ---- */
442 * Fixme: BSD difference: In BSD all sockets connected to us get
443 * ECONNRESET and we die on the spot. In Linux we behave
444 * like files and pipes do and wait for the last
447 * Can't we simply set sock->err?
449 * What the above comment does talk about? --ANK(980817)
452 if (unix_tot_inflight)
453 unix_gc(); /* Garbage collect fds */
456 static void init_peercred(struct sock *sk)
458 put_pid(sk->sk_peer_pid);
459 if (sk->sk_peer_cred)
460 put_cred(sk->sk_peer_cred);
461 sk->sk_peer_pid = get_pid(task_tgid(current));
462 sk->sk_peer_cred = get_current_cred();
465 static void copy_peercred(struct sock *sk, struct sock *peersk)
467 put_pid(sk->sk_peer_pid);
468 if (sk->sk_peer_cred)
469 put_cred(sk->sk_peer_cred);
470 sk->sk_peer_pid = get_pid(peersk->sk_peer_pid);
471 sk->sk_peer_cred = get_cred(peersk->sk_peer_cred);
474 static int unix_listen(struct socket *sock, int backlog)
477 struct sock *sk = sock->sk;
478 struct unix_sock *u = unix_sk(sk);
479 struct pid *old_pid = NULL;
482 if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
483 goto out; /* Only stream/seqpacket sockets accept */
486 goto out; /* No listens on an unbound socket */
488 if (sk->sk_state != TCP_CLOSE && sk->sk_state != TCP_LISTEN)
490 if (backlog > sk->sk_max_ack_backlog)
491 wake_up_interruptible_all(&u->peer_wait);
492 sk->sk_max_ack_backlog = backlog;
493 sk->sk_state = TCP_LISTEN;
494 /* set credentials so connect can copy them */
499 unix_state_unlock(sk);
505 static int unix_release(struct socket *);
506 static int unix_bind(struct socket *, struct sockaddr *, int);
507 static int unix_stream_connect(struct socket *, struct sockaddr *,
508 int addr_len, int flags);
509 static int unix_socketpair(struct socket *, struct socket *);
510 static int unix_accept(struct socket *, struct socket *, int);
511 static int unix_getname(struct socket *, struct sockaddr *, int *, int);
512 static unsigned int unix_poll(struct file *, struct socket *, poll_table *);
513 static unsigned int unix_dgram_poll(struct file *, struct socket *,
515 static int unix_ioctl(struct socket *, unsigned int, unsigned long);
516 static int unix_shutdown(struct socket *, int);
517 static int unix_stream_sendmsg(struct kiocb *, struct socket *,
518 struct msghdr *, size_t);
519 static int unix_stream_recvmsg(struct kiocb *, struct socket *,
520 struct msghdr *, size_t, int);
521 static int unix_dgram_sendmsg(struct kiocb *, struct socket *,
522 struct msghdr *, size_t);
523 static int unix_dgram_recvmsg(struct kiocb *, struct socket *,
524 struct msghdr *, size_t, int);
525 static int unix_dgram_connect(struct socket *, struct sockaddr *,
527 static int unix_seqpacket_sendmsg(struct kiocb *, struct socket *,
528 struct msghdr *, size_t);
529 static int unix_seqpacket_recvmsg(struct kiocb *, struct socket *,
530 struct msghdr *, size_t, int);
532 static int unix_set_peek_off(struct sock *sk, int val)
534 struct unix_sock *u = unix_sk(sk);
536 if (mutex_lock_interruptible(&u->readlock))
539 sk->sk_peek_off = val;
540 mutex_unlock(&u->readlock);
546 static const struct proto_ops unix_stream_ops = {
548 .owner = THIS_MODULE,
549 .release = unix_release,
551 .connect = unix_stream_connect,
552 .socketpair = unix_socketpair,
553 .accept = unix_accept,
554 .getname = unix_getname,
557 .listen = unix_listen,
558 .shutdown = unix_shutdown,
559 .setsockopt = sock_no_setsockopt,
560 .getsockopt = sock_no_getsockopt,
561 .sendmsg = unix_stream_sendmsg,
562 .recvmsg = unix_stream_recvmsg,
563 .mmap = sock_no_mmap,
564 .sendpage = sock_no_sendpage,
565 .set_peek_off = unix_set_peek_off,
568 static const struct proto_ops unix_dgram_ops = {
570 .owner = THIS_MODULE,
571 .release = unix_release,
573 .connect = unix_dgram_connect,
574 .socketpair = unix_socketpair,
575 .accept = sock_no_accept,
576 .getname = unix_getname,
577 .poll = unix_dgram_poll,
579 .listen = sock_no_listen,
580 .shutdown = unix_shutdown,
581 .setsockopt = sock_no_setsockopt,
582 .getsockopt = sock_no_getsockopt,
583 .sendmsg = unix_dgram_sendmsg,
584 .recvmsg = unix_dgram_recvmsg,
585 .mmap = sock_no_mmap,
586 .sendpage = sock_no_sendpage,
587 .set_peek_off = unix_set_peek_off,
590 static const struct proto_ops unix_seqpacket_ops = {
592 .owner = THIS_MODULE,
593 .release = unix_release,
595 .connect = unix_stream_connect,
596 .socketpair = unix_socketpair,
597 .accept = unix_accept,
598 .getname = unix_getname,
599 .poll = unix_dgram_poll,
601 .listen = unix_listen,
602 .shutdown = unix_shutdown,
603 .setsockopt = sock_no_setsockopt,
604 .getsockopt = sock_no_getsockopt,
605 .sendmsg = unix_seqpacket_sendmsg,
606 .recvmsg = unix_seqpacket_recvmsg,
607 .mmap = sock_no_mmap,
608 .sendpage = sock_no_sendpage,
609 .set_peek_off = unix_set_peek_off,
612 static struct proto unix_proto = {
614 .owner = THIS_MODULE,
615 .obj_size = sizeof(struct unix_sock),
619 * AF_UNIX sockets do not interact with hardware, hence they
620 * dont trigger interrupts - so it's safe for them to have
621 * bh-unsafe locking for their sk_receive_queue.lock. Split off
622 * this special lock-class by reinitializing the spinlock key:
624 static struct lock_class_key af_unix_sk_receive_queue_lock_key;
626 static struct sock *unix_create1(struct net *net, struct socket *sock)
628 struct sock *sk = NULL;
631 atomic_long_inc(&unix_nr_socks);
632 if (atomic_long_read(&unix_nr_socks) > 2 * get_max_files())
635 sk = sk_alloc(net, PF_UNIX, GFP_KERNEL, &unix_proto);
639 sock_init_data(sock, sk);
640 lockdep_set_class(&sk->sk_receive_queue.lock,
641 &af_unix_sk_receive_queue_lock_key);
643 sk->sk_write_space = unix_write_space;
644 sk->sk_max_ack_backlog = net->unx.sysctl_max_dgram_qlen;
645 sk->sk_destruct = unix_sock_destructor;
647 u->path.dentry = NULL;
649 spin_lock_init(&u->lock);
650 atomic_long_set(&u->inflight, 0);
651 INIT_LIST_HEAD(&u->link);
652 mutex_init(&u->readlock); /* single task reading lock */
653 init_waitqueue_head(&u->peer_wait);
654 unix_insert_socket(unix_sockets_unbound(sk), sk);
657 atomic_long_dec(&unix_nr_socks);
660 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
666 static int unix_create(struct net *net, struct socket *sock, int protocol,
669 if (protocol && protocol != PF_UNIX)
670 return -EPROTONOSUPPORT;
672 sock->state = SS_UNCONNECTED;
674 switch (sock->type) {
676 sock->ops = &unix_stream_ops;
679 * Believe it or not BSD has AF_UNIX, SOCK_RAW though
683 sock->type = SOCK_DGRAM;
685 sock->ops = &unix_dgram_ops;
688 sock->ops = &unix_seqpacket_ops;
691 return -ESOCKTNOSUPPORT;
694 return unix_create1(net, sock) ? 0 : -ENOMEM;
697 static int unix_release(struct socket *sock)
699 struct sock *sk = sock->sk;
704 unix_release_sock(sk, 0);
710 static int unix_autobind(struct socket *sock)
712 struct sock *sk = sock->sk;
713 struct net *net = sock_net(sk);
714 struct unix_sock *u = unix_sk(sk);
715 static u32 ordernum = 1;
716 struct unix_address *addr;
718 unsigned int retries = 0;
720 err = mutex_lock_interruptible(&u->readlock);
729 addr = kzalloc(sizeof(*addr) + sizeof(short) + 16, GFP_KERNEL);
733 addr->name->sun_family = AF_UNIX;
734 atomic_set(&addr->refcnt, 1);
737 addr->len = sprintf(addr->name->sun_path+1, "%05x", ordernum) + 1 + sizeof(short);
738 addr->hash = unix_hash_fold(csum_partial(addr->name, addr->len, 0));
740 spin_lock(&unix_table_lock);
741 ordernum = (ordernum+1)&0xFFFFF;
743 if (__unix_find_socket_byname(net, addr->name, addr->len, sock->type,
745 spin_unlock(&unix_table_lock);
747 * __unix_find_socket_byname() may take long time if many names
748 * are already in use.
751 /* Give up if all names seems to be in use. */
752 if (retries++ == 0xFFFFF) {
759 addr->hash ^= sk->sk_type;
761 __unix_remove_socket(sk);
763 __unix_insert_socket(&unix_socket_table[addr->hash], sk);
764 spin_unlock(&unix_table_lock);
767 out: mutex_unlock(&u->readlock);
771 static struct sock *unix_find_other(struct net *net,
772 struct sockaddr_un *sunname, int len,
773 int type, unsigned int hash, int *error)
779 if (sunname->sun_path[0]) {
781 err = kern_path(sunname->sun_path, LOOKUP_FOLLOW, &path);
784 inode = path.dentry->d_inode;
785 err = inode_permission(inode, MAY_WRITE);
790 if (!S_ISSOCK(inode->i_mode))
792 u = unix_find_socket_byinode(inode);
796 if (u->sk_type == type)
802 if (u->sk_type != type) {
808 u = unix_find_socket_byname(net, sunname, len, type, hash);
810 struct dentry *dentry;
811 dentry = unix_sk(u)->path.dentry;
813 touch_atime(&unix_sk(u)->path);
826 static int unix_mknod(const char *sun_path, umode_t mode, struct path *res)
828 struct dentry *dentry;
832 * Get the parent directory, calculate the hash for last
835 dentry = kern_path_create(AT_FDCWD, sun_path, &path, 0);
836 err = PTR_ERR(dentry);
841 * All right, let's create it.
843 err = security_path_mknod(&path, dentry, mode, 0);
845 err = vfs_mknod(path.dentry->d_inode, dentry, mode, 0);
847 res->mnt = mntget(path.mnt);
848 res->dentry = dget(dentry);
851 done_path_create(&path, dentry);
855 static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
857 struct sock *sk = sock->sk;
858 struct net *net = sock_net(sk);
859 struct unix_sock *u = unix_sk(sk);
860 struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
861 char *sun_path = sunaddr->sun_path;
864 struct unix_address *addr;
865 struct hlist_head *list;
868 if (sunaddr->sun_family != AF_UNIX)
871 if (addr_len == sizeof(short)) {
872 err = unix_autobind(sock);
876 err = unix_mkname(sunaddr, addr_len, &hash);
881 err = mutex_lock_interruptible(&u->readlock);
890 addr = kmalloc(sizeof(*addr)+addr_len, GFP_KERNEL);
894 memcpy(addr->name, sunaddr, addr_len);
895 addr->len = addr_len;
896 addr->hash = hash ^ sk->sk_type;
897 atomic_set(&addr->refcnt, 1);
901 umode_t mode = S_IFSOCK |
902 (SOCK_INODE(sock)->i_mode & ~current_umask());
903 err = unix_mknod(sun_path, mode, &path);
907 unix_release_addr(addr);
910 addr->hash = UNIX_HASH_SIZE;
911 hash = path.dentry->d_inode->i_ino & (UNIX_HASH_SIZE-1);
912 spin_lock(&unix_table_lock);
914 list = &unix_socket_table[hash];
916 spin_lock(&unix_table_lock);
918 if (__unix_find_socket_byname(net, sunaddr, addr_len,
919 sk->sk_type, hash)) {
920 unix_release_addr(addr);
924 list = &unix_socket_table[addr->hash];
928 __unix_remove_socket(sk);
930 __unix_insert_socket(list, sk);
933 spin_unlock(&unix_table_lock);
935 mutex_unlock(&u->readlock);
940 static void unix_state_double_lock(struct sock *sk1, struct sock *sk2)
942 if (unlikely(sk1 == sk2) || !sk2) {
943 unix_state_lock(sk1);
947 unix_state_lock(sk1);
948 unix_state_lock_nested(sk2);
950 unix_state_lock(sk2);
951 unix_state_lock_nested(sk1);
955 static void unix_state_double_unlock(struct sock *sk1, struct sock *sk2)
957 if (unlikely(sk1 == sk2) || !sk2) {
958 unix_state_unlock(sk1);
961 unix_state_unlock(sk1);
962 unix_state_unlock(sk2);
965 static int unix_dgram_connect(struct socket *sock, struct sockaddr *addr,
968 struct sock *sk = sock->sk;
969 struct net *net = sock_net(sk);
970 struct sockaddr_un *sunaddr = (struct sockaddr_un *)addr;
975 if (addr->sa_family != AF_UNSPEC) {
976 err = unix_mkname(sunaddr, alen, &hash);
981 if (test_bit(SOCK_PASSCRED, &sock->flags) &&
982 !unix_sk(sk)->addr && (err = unix_autobind(sock)) != 0)
986 other = unix_find_other(net, sunaddr, alen, sock->type, hash, &err);
990 unix_state_double_lock(sk, other);
992 /* Apparently VFS overslept socket death. Retry. */
993 if (sock_flag(other, SOCK_DEAD)) {
994 unix_state_double_unlock(sk, other);
1000 if (!unix_may_send(sk, other))
1003 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1009 * 1003.1g breaking connected state with AF_UNSPEC
1012 unix_state_double_lock(sk, other);
1016 * If it was connected, reconnect.
1018 if (unix_peer(sk)) {
1019 struct sock *old_peer = unix_peer(sk);
1020 unix_peer(sk) = other;
1021 unix_state_double_unlock(sk, other);
1023 if (other != old_peer)
1024 unix_dgram_disconnected(sk, old_peer);
1027 unix_peer(sk) = other;
1028 unix_state_double_unlock(sk, other);
1033 unix_state_double_unlock(sk, other);
1039 static long unix_wait_for_peer(struct sock *other, long timeo)
1041 struct unix_sock *u = unix_sk(other);
1045 prepare_to_wait_exclusive(&u->peer_wait, &wait, TASK_INTERRUPTIBLE);
1047 sched = !sock_flag(other, SOCK_DEAD) &&
1048 !(other->sk_shutdown & RCV_SHUTDOWN) &&
1049 unix_recvq_full(other);
1051 unix_state_unlock(other);
1054 timeo = schedule_timeout(timeo);
1056 finish_wait(&u->peer_wait, &wait);
1060 static int unix_stream_connect(struct socket *sock, struct sockaddr *uaddr,
1061 int addr_len, int flags)
1063 struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
1064 struct sock *sk = sock->sk;
1065 struct net *net = sock_net(sk);
1066 struct unix_sock *u = unix_sk(sk), *newu, *otheru;
1067 struct sock *newsk = NULL;
1068 struct sock *other = NULL;
1069 struct sk_buff *skb = NULL;
1075 err = unix_mkname(sunaddr, addr_len, &hash);
1080 if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr &&
1081 (err = unix_autobind(sock)) != 0)
1084 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
1086 /* First of all allocate resources.
1087 If we will make it after state is locked,
1088 we will have to recheck all again in any case.
1093 /* create new sock for complete connection */
1094 newsk = unix_create1(sock_net(sk), NULL);
1098 /* Allocate skb for sending to listening sock */
1099 skb = sock_wmalloc(newsk, 1, 0, GFP_KERNEL);
1104 /* Find listening sock. */
1105 other = unix_find_other(net, sunaddr, addr_len, sk->sk_type, hash, &err);
1109 /* Latch state of peer */
1110 unix_state_lock(other);
1112 /* Apparently VFS overslept socket death. Retry. */
1113 if (sock_flag(other, SOCK_DEAD)) {
1114 unix_state_unlock(other);
1119 err = -ECONNREFUSED;
1120 if (other->sk_state != TCP_LISTEN)
1122 if (other->sk_shutdown & RCV_SHUTDOWN)
1125 if (unix_recvq_full(other)) {
1130 timeo = unix_wait_for_peer(other, timeo);
1132 err = sock_intr_errno(timeo);
1133 if (signal_pending(current))
1141 It is tricky place. We need to grab our state lock and cannot
1142 drop lock on peer. It is dangerous because deadlock is
1143 possible. Connect to self case and simultaneous
1144 attempt to connect are eliminated by checking socket
1145 state. other is TCP_LISTEN, if sk is TCP_LISTEN we
1146 check this before attempt to grab lock.
1148 Well, and we have to recheck the state after socket locked.
1154 /* This is ok... continue with connect */
1156 case TCP_ESTABLISHED:
1157 /* Socket is already connected */
1165 unix_state_lock_nested(sk);
1167 if (sk->sk_state != st) {
1168 unix_state_unlock(sk);
1169 unix_state_unlock(other);
1174 err = security_unix_stream_connect(sk, other, newsk);
1176 unix_state_unlock(sk);
1180 /* The way is open! Fastly set all the necessary fields... */
1183 unix_peer(newsk) = sk;
1184 newsk->sk_state = TCP_ESTABLISHED;
1185 newsk->sk_type = sk->sk_type;
1186 init_peercred(newsk);
1187 newu = unix_sk(newsk);
1188 RCU_INIT_POINTER(newsk->sk_wq, &newu->peer_wq);
1189 otheru = unix_sk(other);
1191 /* copy address information from listening to new sock*/
1193 atomic_inc(&otheru->addr->refcnt);
1194 newu->addr = otheru->addr;
1196 if (otheru->path.dentry) {
1197 path_get(&otheru->path);
1198 newu->path = otheru->path;
1201 /* Set credentials */
1202 copy_peercred(sk, other);
1204 sock->state = SS_CONNECTED;
1205 sk->sk_state = TCP_ESTABLISHED;
1208 smp_mb__after_atomic_inc(); /* sock_hold() does an atomic_inc() */
1209 unix_peer(sk) = newsk;
1211 unix_state_unlock(sk);
1213 /* take ten and and send info to listening sock */
1214 spin_lock(&other->sk_receive_queue.lock);
1215 __skb_queue_tail(&other->sk_receive_queue, skb);
1216 spin_unlock(&other->sk_receive_queue.lock);
1217 unix_state_unlock(other);
1218 other->sk_data_ready(other, 0);
1224 unix_state_unlock(other);
1229 unix_release_sock(newsk, 0);
1235 static int unix_socketpair(struct socket *socka, struct socket *sockb)
1237 struct sock *ska = socka->sk, *skb = sockb->sk;
1239 /* Join our sockets back to back */
1242 unix_peer(ska) = skb;
1243 unix_peer(skb) = ska;
1247 if (ska->sk_type != SOCK_DGRAM) {
1248 ska->sk_state = TCP_ESTABLISHED;
1249 skb->sk_state = TCP_ESTABLISHED;
1250 socka->state = SS_CONNECTED;
1251 sockb->state = SS_CONNECTED;
1256 static void unix_sock_inherit_flags(const struct socket *old,
1259 if (test_bit(SOCK_PASSCRED, &old->flags))
1260 set_bit(SOCK_PASSCRED, &new->flags);
1261 if (test_bit(SOCK_PASSSEC, &old->flags))
1262 set_bit(SOCK_PASSSEC, &new->flags);
1265 static int unix_accept(struct socket *sock, struct socket *newsock, int flags)
1267 struct sock *sk = sock->sk;
1269 struct sk_buff *skb;
1273 if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
1277 if (sk->sk_state != TCP_LISTEN)
1280 /* If socket state is TCP_LISTEN it cannot change (for now...),
1281 * so that no locks are necessary.
1284 skb = skb_recv_datagram(sk, 0, flags&O_NONBLOCK, &err);
1286 /* This means receive shutdown. */
1293 skb_free_datagram(sk, skb);
1294 wake_up_interruptible(&unix_sk(sk)->peer_wait);
1296 /* attach accepted sock to socket */
1297 unix_state_lock(tsk);
1298 newsock->state = SS_CONNECTED;
1299 unix_sock_inherit_flags(sock, newsock);
1300 sock_graft(tsk, newsock);
1301 unix_state_unlock(tsk);
1309 static int unix_getname(struct socket *sock, struct sockaddr *uaddr, int *uaddr_len, int peer)
1311 struct sock *sk = sock->sk;
1312 struct unix_sock *u;
1313 DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr, uaddr);
1317 sk = unix_peer_get(sk);
1328 unix_state_lock(sk);
1330 sunaddr->sun_family = AF_UNIX;
1331 sunaddr->sun_path[0] = 0;
1332 *uaddr_len = sizeof(short);
1334 struct unix_address *addr = u->addr;
1336 *uaddr_len = addr->len;
1337 memcpy(sunaddr, addr->name, *uaddr_len);
1339 unix_state_unlock(sk);
1345 static void unix_detach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1349 scm->fp = UNIXCB(skb).fp;
1350 UNIXCB(skb).fp = NULL;
1352 for (i = scm->fp->count-1; i >= 0; i--)
1353 unix_notinflight(scm->fp->fp[i]);
1356 static void unix_destruct_scm(struct sk_buff *skb)
1358 struct scm_cookie scm;
1359 memset(&scm, 0, sizeof(scm));
1360 scm.pid = UNIXCB(skb).pid;
1362 unix_detach_fds(&scm, skb);
1364 /* Alas, it calls VFS */
1365 /* So fscking what? fput() had been SMP-safe since the last Summer */
1370 #define MAX_RECURSION_LEVEL 4
1372 static int unix_attach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1375 unsigned char max_level = 0;
1376 int unix_sock_count = 0;
1378 for (i = scm->fp->count - 1; i >= 0; i--) {
1379 struct sock *sk = unix_get_socket(scm->fp->fp[i]);
1383 max_level = max(max_level,
1384 unix_sk(sk)->recursion_level);
1387 if (unlikely(max_level > MAX_RECURSION_LEVEL))
1388 return -ETOOMANYREFS;
1391 * Need to duplicate file references for the sake of garbage
1392 * collection. Otherwise a socket in the fps might become a
1393 * candidate for GC while the skb is not yet queued.
1395 UNIXCB(skb).fp = scm_fp_dup(scm->fp);
1396 if (!UNIXCB(skb).fp)
1399 if (unix_sock_count) {
1400 for (i = scm->fp->count - 1; i >= 0; i--)
1401 unix_inflight(scm->fp->fp[i]);
1406 static int unix_scm_to_skb(struct scm_cookie *scm, struct sk_buff *skb, bool send_fds)
1410 UNIXCB(skb).pid = get_pid(scm->pid);
1411 UNIXCB(skb).uid = scm->creds.uid;
1412 UNIXCB(skb).gid = scm->creds.gid;
1413 UNIXCB(skb).fp = NULL;
1414 if (scm->fp && send_fds)
1415 err = unix_attach_fds(scm, skb);
1417 skb->destructor = unix_destruct_scm;
1422 * Some apps rely on write() giving SCM_CREDENTIALS
1423 * We include credentials if source or destination socket
1424 * asserted SOCK_PASSCRED.
1426 static void maybe_add_creds(struct sk_buff *skb, const struct socket *sock,
1427 const struct sock *other)
1429 if (UNIXCB(skb).pid)
1431 if (test_bit(SOCK_PASSCRED, &sock->flags) ||
1432 !other->sk_socket ||
1433 test_bit(SOCK_PASSCRED, &other->sk_socket->flags)) {
1434 UNIXCB(skb).pid = get_pid(task_tgid(current));
1435 current_uid_gid(&UNIXCB(skb).uid, &UNIXCB(skb).gid);
1440 * Send AF_UNIX data.
1443 static int unix_dgram_sendmsg(struct kiocb *kiocb, struct socket *sock,
1444 struct msghdr *msg, size_t len)
1446 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1447 struct sock *sk = sock->sk;
1448 struct net *net = sock_net(sk);
1449 struct unix_sock *u = unix_sk(sk);
1450 struct sockaddr_un *sunaddr = msg->msg_name;
1451 struct sock *other = NULL;
1452 int namelen = 0; /* fake GCC */
1455 struct sk_buff *skb;
1457 struct scm_cookie tmp_scm;
1461 if (NULL == siocb->scm)
1462 siocb->scm = &tmp_scm;
1464 err = scm_send(sock, msg, siocb->scm, false);
1469 if (msg->msg_flags&MSG_OOB)
1472 if (msg->msg_namelen) {
1473 err = unix_mkname(sunaddr, msg->msg_namelen, &hash);
1480 other = unix_peer_get(sk);
1485 if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr
1486 && (err = unix_autobind(sock)) != 0)
1490 if (len > sk->sk_sndbuf - 32)
1493 if (len > SKB_MAX_ALLOC)
1494 data_len = min_t(size_t,
1495 len - SKB_MAX_ALLOC,
1496 MAX_SKB_FRAGS * PAGE_SIZE);
1498 skb = sock_alloc_send_pskb(sk, len - data_len, data_len,
1499 msg->msg_flags & MSG_DONTWAIT, &err);
1503 err = unix_scm_to_skb(siocb->scm, skb, true);
1506 max_level = err + 1;
1507 unix_get_secdata(siocb->scm, skb);
1509 skb_put(skb, len - data_len);
1510 skb->data_len = data_len;
1512 err = skb_copy_datagram_from_iovec(skb, 0, msg->msg_iov, 0, len);
1516 timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1521 if (sunaddr == NULL)
1524 other = unix_find_other(net, sunaddr, namelen, sk->sk_type,
1530 if (sk_filter(other, skb) < 0) {
1531 /* Toss the packet but do not return any error to the sender */
1536 unix_state_lock(other);
1538 if (!unix_may_send(sk, other))
1541 if (sock_flag(other, SOCK_DEAD)) {
1543 * Check with 1003.1g - what should
1546 unix_state_unlock(other);
1550 unix_state_lock(sk);
1551 if (unix_peer(sk) == other) {
1552 unix_peer(sk) = NULL;
1553 unix_state_unlock(sk);
1555 unix_dgram_disconnected(sk, other);
1557 err = -ECONNREFUSED;
1559 unix_state_unlock(sk);
1569 if (other->sk_shutdown & RCV_SHUTDOWN)
1572 if (sk->sk_type != SOCK_SEQPACKET) {
1573 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1578 if (unix_peer(other) != sk && unix_recvq_full(other)) {
1584 timeo = unix_wait_for_peer(other, timeo);
1586 err = sock_intr_errno(timeo);
1587 if (signal_pending(current))
1593 if (sock_flag(other, SOCK_RCVTSTAMP))
1594 __net_timestamp(skb);
1595 maybe_add_creds(skb, sock, other);
1596 skb_queue_tail(&other->sk_receive_queue, skb);
1597 if (max_level > unix_sk(other)->recursion_level)
1598 unix_sk(other)->recursion_level = max_level;
1599 unix_state_unlock(other);
1600 other->sk_data_ready(other, len);
1602 scm_destroy(siocb->scm);
1606 unix_state_unlock(other);
1612 scm_destroy(siocb->scm);
1617 static int unix_stream_sendmsg(struct kiocb *kiocb, struct socket *sock,
1618 struct msghdr *msg, size_t len)
1620 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1621 struct sock *sk = sock->sk;
1622 struct sock *other = NULL;
1624 struct sk_buff *skb;
1626 struct scm_cookie tmp_scm;
1627 bool fds_sent = false;
1630 if (NULL == siocb->scm)
1631 siocb->scm = &tmp_scm;
1633 err = scm_send(sock, msg, siocb->scm, false);
1638 if (msg->msg_flags&MSG_OOB)
1641 if (msg->msg_namelen) {
1642 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
1646 other = unix_peer(sk);
1651 if (sk->sk_shutdown & SEND_SHUTDOWN)
1654 while (sent < len) {
1656 * Optimisation for the fact that under 0.01% of X
1657 * messages typically need breaking up.
1662 /* Keep two messages in the pipe so it schedules better */
1663 if (size > ((sk->sk_sndbuf >> 1) - 64))
1664 size = (sk->sk_sndbuf >> 1) - 64;
1666 if (size > SKB_MAX_ALLOC)
1667 size = SKB_MAX_ALLOC;
1673 skb = sock_alloc_send_skb(sk, size, msg->msg_flags&MSG_DONTWAIT,
1680 * If you pass two values to the sock_alloc_send_skb
1681 * it tries to grab the large buffer with GFP_NOFS
1682 * (which can fail easily), and if it fails grab the
1683 * fallback size buffer which is under a page and will
1686 size = min_t(int, size, skb_tailroom(skb));
1689 /* Only send the fds in the first buffer */
1690 err = unix_scm_to_skb(siocb->scm, skb, !fds_sent);
1695 max_level = err + 1;
1698 err = memcpy_fromiovec(skb_put(skb, size), msg->msg_iov, size);
1704 unix_state_lock(other);
1706 if (sock_flag(other, SOCK_DEAD) ||
1707 (other->sk_shutdown & RCV_SHUTDOWN))
1710 maybe_add_creds(skb, sock, other);
1711 skb_queue_tail(&other->sk_receive_queue, skb);
1712 if (max_level > unix_sk(other)->recursion_level)
1713 unix_sk(other)->recursion_level = max_level;
1714 unix_state_unlock(other);
1715 other->sk_data_ready(other, size);
1719 scm_destroy(siocb->scm);
1725 unix_state_unlock(other);
1728 if (sent == 0 && !(msg->msg_flags&MSG_NOSIGNAL))
1729 send_sig(SIGPIPE, current, 0);
1732 scm_destroy(siocb->scm);
1734 return sent ? : err;
1737 static int unix_seqpacket_sendmsg(struct kiocb *kiocb, struct socket *sock,
1738 struct msghdr *msg, size_t len)
1741 struct sock *sk = sock->sk;
1743 err = sock_error(sk);
1747 if (sk->sk_state != TCP_ESTABLISHED)
1750 if (msg->msg_namelen)
1751 msg->msg_namelen = 0;
1753 return unix_dgram_sendmsg(kiocb, sock, msg, len);
1756 static int unix_seqpacket_recvmsg(struct kiocb *iocb, struct socket *sock,
1757 struct msghdr *msg, size_t size,
1760 struct sock *sk = sock->sk;
1762 if (sk->sk_state != TCP_ESTABLISHED)
1765 return unix_dgram_recvmsg(iocb, sock, msg, size, flags);
1768 static void unix_copy_addr(struct msghdr *msg, struct sock *sk)
1770 struct unix_sock *u = unix_sk(sk);
1773 msg->msg_namelen = u->addr->len;
1774 memcpy(msg->msg_name, u->addr->name, u->addr->len);
1778 static int unix_dgram_recvmsg(struct kiocb *iocb, struct socket *sock,
1779 struct msghdr *msg, size_t size,
1782 struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1783 struct scm_cookie tmp_scm;
1784 struct sock *sk = sock->sk;
1785 struct unix_sock *u = unix_sk(sk);
1786 int noblock = flags & MSG_DONTWAIT;
1787 struct sk_buff *skb;
1795 err = mutex_lock_interruptible(&u->readlock);
1796 if (unlikely(err)) {
1797 /* recvmsg() in non blocking mode is supposed to return -EAGAIN
1798 * sk_rcvtimeo is not honored by mutex_lock_interruptible()
1800 err = noblock ? -EAGAIN : -ERESTARTSYS;
1804 skip = sk_peek_offset(sk, flags);
1806 skb = __skb_recv_datagram(sk, flags, &peeked, &skip, &err);
1808 unix_state_lock(sk);
1809 /* Signal EOF on disconnected non-blocking SEQPACKET socket. */
1810 if (sk->sk_type == SOCK_SEQPACKET && err == -EAGAIN &&
1811 (sk->sk_shutdown & RCV_SHUTDOWN))
1813 unix_state_unlock(sk);
1817 wake_up_interruptible_sync_poll(&u->peer_wait,
1818 POLLOUT | POLLWRNORM | POLLWRBAND);
1821 unix_copy_addr(msg, skb->sk);
1823 if (size > skb->len - skip)
1824 size = skb->len - skip;
1825 else if (size < skb->len - skip)
1826 msg->msg_flags |= MSG_TRUNC;
1828 err = skb_copy_datagram_iovec(skb, skip, msg->msg_iov, size);
1832 if (sock_flag(sk, SOCK_RCVTSTAMP))
1833 __sock_recv_timestamp(msg, sk, skb);
1836 siocb->scm = &tmp_scm;
1837 memset(&tmp_scm, 0, sizeof(tmp_scm));
1839 scm_set_cred(siocb->scm, UNIXCB(skb).pid, UNIXCB(skb).uid, UNIXCB(skb).gid);
1840 unix_set_secdata(siocb->scm, skb);
1842 if (!(flags & MSG_PEEK)) {
1844 unix_detach_fds(siocb->scm, skb);
1846 sk_peek_offset_bwd(sk, skb->len);
1848 /* It is questionable: on PEEK we could:
1849 - do not return fds - good, but too simple 8)
1850 - return fds, and do not return them on read (old strategy,
1852 - clone fds (I chose it for now, it is the most universal
1855 POSIX 1003.1g does not actually define this clearly
1856 at all. POSIX 1003.1g doesn't define a lot of things
1861 sk_peek_offset_fwd(sk, size);
1864 siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1866 err = (flags & MSG_TRUNC) ? skb->len - skip : size;
1868 scm_recv(sock, msg, siocb->scm, flags);
1871 skb_free_datagram(sk, skb);
1873 mutex_unlock(&u->readlock);
1879 * Sleep until more data has arrived. But check for races..
1881 static long unix_stream_data_wait(struct sock *sk, long timeo,
1882 struct sk_buff *last)
1886 unix_state_lock(sk);
1889 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1891 if (skb_peek_tail(&sk->sk_receive_queue) != last ||
1893 (sk->sk_shutdown & RCV_SHUTDOWN) ||
1894 signal_pending(current) ||
1898 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1899 unix_state_unlock(sk);
1900 timeo = freezable_schedule_timeout(timeo);
1901 unix_state_lock(sk);
1902 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1905 finish_wait(sk_sleep(sk), &wait);
1906 unix_state_unlock(sk);
1910 static int unix_stream_recvmsg(struct kiocb *iocb, struct socket *sock,
1911 struct msghdr *msg, size_t size,
1914 struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1915 struct scm_cookie tmp_scm;
1916 struct sock *sk = sock->sk;
1917 struct unix_sock *u = unix_sk(sk);
1918 struct sockaddr_un *sunaddr = msg->msg_name;
1920 int noblock = flags & MSG_DONTWAIT;
1921 int check_creds = 0;
1928 if (sk->sk_state != TCP_ESTABLISHED)
1935 target = sock_rcvlowat(sk, flags&MSG_WAITALL, size);
1936 timeo = sock_rcvtimeo(sk, noblock);
1938 /* Lock the socket to prevent queue disordering
1939 * while sleeps in memcpy_tomsg
1943 siocb->scm = &tmp_scm;
1944 memset(&tmp_scm, 0, sizeof(tmp_scm));
1947 err = mutex_lock_interruptible(&u->readlock);
1948 if (unlikely(err)) {
1949 /* recvmsg() in non blocking mode is supposed to return -EAGAIN
1950 * sk_rcvtimeo is not honored by mutex_lock_interruptible()
1952 err = noblock ? -EAGAIN : -ERESTARTSYS;
1958 struct sk_buff *skb, *last;
1960 unix_state_lock(sk);
1961 last = skb = skb_peek(&sk->sk_receive_queue);
1964 unix_sk(sk)->recursion_level = 0;
1965 if (copied >= target)
1969 * POSIX 1003.1g mandates this order.
1972 err = sock_error(sk);
1975 if (sk->sk_shutdown & RCV_SHUTDOWN)
1978 unix_state_unlock(sk);
1982 mutex_unlock(&u->readlock);
1984 timeo = unix_stream_data_wait(sk, timeo, last);
1986 if (signal_pending(current)
1987 || mutex_lock_interruptible(&u->readlock)) {
1988 err = sock_intr_errno(timeo);
1994 unix_state_unlock(sk);
1998 skip = sk_peek_offset(sk, flags);
1999 while (skip >= skb->len) {
2002 skb = skb_peek_next(skb, &sk->sk_receive_queue);
2007 unix_state_unlock(sk);
2010 /* Never glue messages from different writers */
2011 if ((UNIXCB(skb).pid != siocb->scm->pid) ||
2012 !uid_eq(UNIXCB(skb).uid, siocb->scm->creds.uid) ||
2013 !gid_eq(UNIXCB(skb).gid, siocb->scm->creds.gid))
2015 } else if (test_bit(SOCK_PASSCRED, &sock->flags)) {
2016 /* Copy credentials */
2017 scm_set_cred(siocb->scm, UNIXCB(skb).pid, UNIXCB(skb).uid, UNIXCB(skb).gid);
2021 /* Copy address just once */
2023 unix_copy_addr(msg, skb->sk);
2027 chunk = min_t(unsigned int, skb->len - skip, size);
2028 if (memcpy_toiovec(msg->msg_iov, skb->data + skip, chunk)) {
2036 /* Mark read part of skb as used */
2037 if (!(flags & MSG_PEEK)) {
2038 skb_pull(skb, chunk);
2040 sk_peek_offset_bwd(sk, chunk);
2043 unix_detach_fds(siocb->scm, skb);
2048 skb_unlink(skb, &sk->sk_receive_queue);
2054 /* It is questionable, see note in unix_dgram_recvmsg.
2057 siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
2059 sk_peek_offset_fwd(sk, chunk);
2065 mutex_unlock(&u->readlock);
2066 scm_recv(sock, msg, siocb->scm, flags);
2068 return copied ? : err;
2071 static int unix_shutdown(struct socket *sock, int mode)
2073 struct sock *sk = sock->sk;
2076 if (mode < SHUT_RD || mode > SHUT_RDWR)
2079 * SHUT_RD (0) -> RCV_SHUTDOWN (1)
2080 * SHUT_WR (1) -> SEND_SHUTDOWN (2)
2081 * SHUT_RDWR (2) -> SHUTDOWN_MASK (3)
2085 unix_state_lock(sk);
2086 sk->sk_shutdown |= mode;
2087 other = unix_peer(sk);
2090 unix_state_unlock(sk);
2091 sk->sk_state_change(sk);
2094 (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET)) {
2098 if (mode&RCV_SHUTDOWN)
2099 peer_mode |= SEND_SHUTDOWN;
2100 if (mode&SEND_SHUTDOWN)
2101 peer_mode |= RCV_SHUTDOWN;
2102 unix_state_lock(other);
2103 other->sk_shutdown |= peer_mode;
2104 unix_state_unlock(other);
2105 other->sk_state_change(other);
2106 if (peer_mode == SHUTDOWN_MASK)
2107 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_HUP);
2108 else if (peer_mode & RCV_SHUTDOWN)
2109 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_IN);
2117 long unix_inq_len(struct sock *sk)
2119 struct sk_buff *skb;
2122 if (sk->sk_state == TCP_LISTEN)
2125 spin_lock(&sk->sk_receive_queue.lock);
2126 if (sk->sk_type == SOCK_STREAM ||
2127 sk->sk_type == SOCK_SEQPACKET) {
2128 skb_queue_walk(&sk->sk_receive_queue, skb)
2131 skb = skb_peek(&sk->sk_receive_queue);
2135 spin_unlock(&sk->sk_receive_queue.lock);
2139 EXPORT_SYMBOL_GPL(unix_inq_len);
2141 long unix_outq_len(struct sock *sk)
2143 return sk_wmem_alloc_get(sk);
2145 EXPORT_SYMBOL_GPL(unix_outq_len);
2147 static int unix_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2149 struct sock *sk = sock->sk;
2155 amount = unix_outq_len(sk);
2156 err = put_user(amount, (int __user *)arg);
2159 amount = unix_inq_len(sk);
2163 err = put_user(amount, (int __user *)arg);
2172 static unsigned int unix_poll(struct file *file, struct socket *sock, poll_table *wait)
2174 struct sock *sk = sock->sk;
2177 sock_poll_wait(file, sk_sleep(sk), wait);
2180 /* exceptional events? */
2183 if (sk->sk_shutdown == SHUTDOWN_MASK)
2185 if (sk->sk_shutdown & RCV_SHUTDOWN)
2186 mask |= POLLRDHUP | POLLIN | POLLRDNORM;
2189 if (!skb_queue_empty(&sk->sk_receive_queue))
2190 mask |= POLLIN | POLLRDNORM;
2192 /* Connection-based need to check for termination and startup */
2193 if ((sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) &&
2194 sk->sk_state == TCP_CLOSE)
2198 * we set writable also when the other side has shut down the
2199 * connection. This prevents stuck sockets.
2201 if (unix_writable(sk))
2202 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2207 static unsigned int unix_dgram_poll(struct file *file, struct socket *sock,
2210 struct sock *sk = sock->sk, *other;
2211 unsigned int mask, writable;
2213 sock_poll_wait(file, sk_sleep(sk), wait);
2216 /* exceptional events? */
2217 if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue))
2219 (sock_flag(sk, SOCK_SELECT_ERR_QUEUE) ? POLLPRI : 0);
2221 if (sk->sk_shutdown & RCV_SHUTDOWN)
2222 mask |= POLLRDHUP | POLLIN | POLLRDNORM;
2223 if (sk->sk_shutdown == SHUTDOWN_MASK)
2227 if (!skb_queue_empty(&sk->sk_receive_queue))
2228 mask |= POLLIN | POLLRDNORM;
2230 /* Connection-based need to check for termination and startup */
2231 if (sk->sk_type == SOCK_SEQPACKET) {
2232 if (sk->sk_state == TCP_CLOSE)
2234 /* connection hasn't started yet? */
2235 if (sk->sk_state == TCP_SYN_SENT)
2239 /* No write status requested, avoid expensive OUT tests. */
2240 if (!(poll_requested_events(wait) & (POLLWRBAND|POLLWRNORM|POLLOUT)))
2243 writable = unix_writable(sk);
2244 other = unix_peer_get(sk);
2246 if (unix_peer(other) != sk) {
2247 sock_poll_wait(file, &unix_sk(other)->peer_wait, wait);
2248 if (unix_recvq_full(other))
2255 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2257 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
2262 #ifdef CONFIG_PROC_FS
2264 #define BUCKET_SPACE (BITS_PER_LONG - (UNIX_HASH_BITS + 1) - 1)
2266 #define get_bucket(x) ((x) >> BUCKET_SPACE)
2267 #define get_offset(x) ((x) & ((1L << BUCKET_SPACE) - 1))
2268 #define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
2270 static struct sock *unix_from_bucket(struct seq_file *seq, loff_t *pos)
2272 unsigned long offset = get_offset(*pos);
2273 unsigned long bucket = get_bucket(*pos);
2275 unsigned long count = 0;
2277 for (sk = sk_head(&unix_socket_table[bucket]); sk; sk = sk_next(sk)) {
2278 if (sock_net(sk) != seq_file_net(seq))
2280 if (++count == offset)
2287 static struct sock *unix_next_socket(struct seq_file *seq,
2291 unsigned long bucket;
2293 while (sk > (struct sock *)SEQ_START_TOKEN) {
2297 if (sock_net(sk) == seq_file_net(seq))
2302 sk = unix_from_bucket(seq, pos);
2307 bucket = get_bucket(*pos) + 1;
2308 *pos = set_bucket_offset(bucket, 1);
2309 } while (bucket < ARRAY_SIZE(unix_socket_table));
2314 static void *unix_seq_start(struct seq_file *seq, loff_t *pos)
2315 __acquires(unix_table_lock)
2317 spin_lock(&unix_table_lock);
2320 return SEQ_START_TOKEN;
2322 if (get_bucket(*pos) >= ARRAY_SIZE(unix_socket_table))
2325 return unix_next_socket(seq, NULL, pos);
2328 static void *unix_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2331 return unix_next_socket(seq, v, pos);
2334 static void unix_seq_stop(struct seq_file *seq, void *v)
2335 __releases(unix_table_lock)
2337 spin_unlock(&unix_table_lock);
2340 static int unix_seq_show(struct seq_file *seq, void *v)
2343 if (v == SEQ_START_TOKEN)
2344 seq_puts(seq, "Num RefCount Protocol Flags Type St "
2348 struct unix_sock *u = unix_sk(s);
2351 seq_printf(seq, "%pK: %08X %08X %08X %04X %02X %5lu",
2353 atomic_read(&s->sk_refcnt),
2355 s->sk_state == TCP_LISTEN ? __SO_ACCEPTCON : 0,
2358 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTED : SS_UNCONNECTED) :
2359 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTING : SS_DISCONNECTING),
2367 len = u->addr->len - sizeof(short);
2368 if (!UNIX_ABSTRACT(s))
2374 for ( ; i < len; i++)
2375 seq_putc(seq, u->addr->name->sun_path[i]);
2377 unix_state_unlock(s);
2378 seq_putc(seq, '\n');
2384 static const struct seq_operations unix_seq_ops = {
2385 .start = unix_seq_start,
2386 .next = unix_seq_next,
2387 .stop = unix_seq_stop,
2388 .show = unix_seq_show,
2391 static int unix_seq_open(struct inode *inode, struct file *file)
2393 return seq_open_net(inode, file, &unix_seq_ops,
2394 sizeof(struct seq_net_private));
2397 static const struct file_operations unix_seq_fops = {
2398 .owner = THIS_MODULE,
2399 .open = unix_seq_open,
2401 .llseek = seq_lseek,
2402 .release = seq_release_net,
2407 static const struct net_proto_family unix_family_ops = {
2409 .create = unix_create,
2410 .owner = THIS_MODULE,
2414 static int __net_init unix_net_init(struct net *net)
2416 int error = -ENOMEM;
2418 net->unx.sysctl_max_dgram_qlen = 10;
2419 if (unix_sysctl_register(net))
2422 #ifdef CONFIG_PROC_FS
2423 if (!proc_create("unix", 0, net->proc_net, &unix_seq_fops)) {
2424 unix_sysctl_unregister(net);
2433 static void __net_exit unix_net_exit(struct net *net)
2435 unix_sysctl_unregister(net);
2436 remove_proc_entry("unix", net->proc_net);
2439 static struct pernet_operations unix_net_ops = {
2440 .init = unix_net_init,
2441 .exit = unix_net_exit,
2444 static int __init af_unix_init(void)
2448 BUILD_BUG_ON(sizeof(struct unix_skb_parms) > FIELD_SIZEOF(struct sk_buff, cb));
2450 rc = proto_register(&unix_proto, 1);
2452 printk(KERN_CRIT "%s: Cannot create unix_sock SLAB cache!\n",
2457 sock_register(&unix_family_ops);
2458 register_pernet_subsys(&unix_net_ops);
2463 static void __exit af_unix_exit(void)
2465 sock_unregister(PF_UNIX);
2466 proto_unregister(&unix_proto);
2467 unregister_pernet_subsys(&unix_net_ops);
2470 /* Earlier than device_initcall() so that other drivers invoking
2471 request_module() don't end up in a loop when modprobe tries
2472 to use a UNIX socket. But later than subsys_initcall() because
2473 we depend on stuff initialised there */
2474 fs_initcall(af_unix_init);
2475 module_exit(af_unix_exit);
2477 MODULE_LICENSE("GPL");
2478 MODULE_ALIAS_NETPROTO(PF_UNIX);