2 * NET4: Implementation of BSD Unix domain sockets.
4 * Authors: Alan Cox, <alan@lxorguk.ukuu.org.uk>
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License
8 * as published by the Free Software Foundation; either version
9 * 2 of the License, or (at your option) any later version.
12 * Linus Torvalds : Assorted bug cures.
13 * Niibe Yutaka : async I/O support.
14 * Carsten Paeth : PF_UNIX check, address fixes.
15 * Alan Cox : Limit size of allocated blocks.
16 * Alan Cox : Fixed the stupid socketpair bug.
17 * Alan Cox : BSD compatibility fine tuning.
18 * Alan Cox : Fixed a bug in connect when interrupted.
19 * Alan Cox : Sorted out a proper draft version of
20 * file descriptor passing hacked up from
22 * Marty Leisner : Fixes to fd passing
23 * Nick Nevin : recvmsg bugfix.
24 * Alan Cox : Started proper garbage collector
25 * Heiko EiBfeldt : Missing verify_area check
26 * Alan Cox : Started POSIXisms
27 * Andreas Schwab : Replace inode by dentry for proper
29 * Kirk Petersen : Made this a module
30 * Christoph Rohland : Elegant non-blocking accept/connect algorithm.
32 * Alexey Kuznetosv : Repaired (I hope) bugs introduces
33 * by above two patches.
34 * Andrea Arcangeli : If possible we block in connect(2)
35 * if the max backlog of the listen socket
36 * is been reached. This won't break
37 * old apps and it will avoid huge amount
38 * of socks hashed (this for unix_gc()
39 * performances reasons).
40 * Security fix that limits the max
41 * number of socks to 2*max_files and
42 * the number of skb queueable in the
44 * Artur Skawina : Hash function optimizations
45 * Alexey Kuznetsov : Full scale SMP. Lot of bugs are introduced 8)
46 * Malcolm Beattie : Set peercred for socketpair
47 * Michal Ostrowski : Module initialization cleanup.
48 * Arnaldo C. Melo : Remove MOD_{INC,DEC}_USE_COUNT,
49 * the core infrastructure is doing that
50 * for all net proto families now (2.5.69+)
53 * Known differences from reference BSD that was tested:
56 * ECONNREFUSED is not returned from one end of a connected() socket to the
57 * other the moment one end closes.
58 * fstat() doesn't return st_dev=0, and give the blksize as high water mark
59 * and a fake inode identifier (nor the BSD first socket fstat twice bug).
61 * accept() returns a path name even if the connecting socket has closed
62 * in the meantime (BSD loses the path and gives up).
63 * accept() returns 0 length path for an unbound connector. BSD returns 16
64 * and a null first byte in the path (but not for gethost/peername - BSD bug ??)
65 * socketpair(...SOCK_RAW..) doesn't panic the kernel.
66 * BSD af_unix apparently has connect forgetting to block properly.
67 * (need to check this with the POSIX spec in detail)
69 * Differences from 2.0.0-11-... (ANK)
70 * Bug fixes and improvements.
71 * - client shutdown killed server socket.
72 * - removed all useless cli/sti pairs.
74 * Semantic changes/extensions.
75 * - generic control message passing.
76 * - SCM_CREDENTIALS control message.
77 * - "Abstract" (not FS based) socket bindings.
78 * Abstract names are sequences of bytes (not zero terminated)
79 * started by 0, so that this name space does not intersect
83 #include <linux/module.h>
84 #include <linux/kernel.h>
85 #include <linux/signal.h>
86 #include <linux/sched.h>
87 #include <linux/errno.h>
88 #include <linux/string.h>
89 #include <linux/stat.h>
90 #include <linux/dcache.h>
91 #include <linux/namei.h>
92 #include <linux/socket.h>
94 #include <linux/fcntl.h>
95 #include <linux/termios.h>
96 #include <linux/sockios.h>
97 #include <linux/net.h>
100 #include <linux/slab.h>
101 #include <asm/uaccess.h>
102 #include <linux/skbuff.h>
103 #include <linux/netdevice.h>
104 #include <net/net_namespace.h>
105 #include <net/sock.h>
106 #include <net/tcp_states.h>
107 #include <net/af_unix.h>
108 #include <linux/proc_fs.h>
109 #include <linux/seq_file.h>
111 #include <linux/init.h>
112 #include <linux/poll.h>
113 #include <linux/rtnetlink.h>
114 #include <linux/mount.h>
115 #include <net/checksum.h>
116 #include <linux/security.h>
118 struct hlist_head unix_socket_table[2 * UNIX_HASH_SIZE];
119 EXPORT_SYMBOL_GPL(unix_socket_table);
120 DEFINE_SPINLOCK(unix_table_lock);
121 EXPORT_SYMBOL_GPL(unix_table_lock);
122 static atomic_long_t unix_nr_socks;
125 static struct hlist_head *unix_sockets_unbound(void *addr)
127 unsigned long hash = (unsigned long)addr;
131 hash %= UNIX_HASH_SIZE;
132 return &unix_socket_table[UNIX_HASH_SIZE + hash];
135 #define UNIX_ABSTRACT(sk) (unix_sk(sk)->addr->hash < UNIX_HASH_SIZE)
137 #ifdef CONFIG_SECURITY_NETWORK
138 static void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb)
140 memcpy(UNIXSID(skb), &scm->secid, sizeof(u32));
143 static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb)
145 scm->secid = *UNIXSID(skb);
148 static inline void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb)
151 static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb)
153 #endif /* CONFIG_SECURITY_NETWORK */
156 * SMP locking strategy:
157 * hash table is protected with spinlock unix_table_lock
158 * each socket state is protected by separate spin lock.
161 static inline unsigned int unix_hash_fold(__wsum n)
163 unsigned int hash = (__force unsigned int)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 mutex_lock(&u->readlock);
727 addr = kzalloc(sizeof(*addr) + sizeof(short) + 16, GFP_KERNEL);
731 addr->name->sun_family = AF_UNIX;
732 atomic_set(&addr->refcnt, 1);
735 addr->len = sprintf(addr->name->sun_path+1, "%05x", ordernum) + 1 + sizeof(short);
736 addr->hash = unix_hash_fold(csum_partial(addr->name, addr->len, 0));
738 spin_lock(&unix_table_lock);
739 ordernum = (ordernum+1)&0xFFFFF;
741 if (__unix_find_socket_byname(net, addr->name, addr->len, sock->type,
743 spin_unlock(&unix_table_lock);
745 * __unix_find_socket_byname() may take long time if many names
746 * are already in use.
749 /* Give up if all names seems to be in use. */
750 if (retries++ == 0xFFFFF) {
757 addr->hash ^= sk->sk_type;
759 __unix_remove_socket(sk);
761 __unix_insert_socket(&unix_socket_table[addr->hash], sk);
762 spin_unlock(&unix_table_lock);
765 out: mutex_unlock(&u->readlock);
769 static struct sock *unix_find_other(struct net *net,
770 struct sockaddr_un *sunname, int len,
771 int type, unsigned int hash, int *error)
777 if (sunname->sun_path[0]) {
779 err = kern_path(sunname->sun_path, LOOKUP_FOLLOW, &path);
782 inode = path.dentry->d_inode;
783 err = inode_permission(inode, MAY_WRITE);
788 if (!S_ISSOCK(inode->i_mode))
790 u = unix_find_socket_byinode(inode);
794 if (u->sk_type == type)
800 if (u->sk_type != type) {
806 u = unix_find_socket_byname(net, sunname, len, type, hash);
808 struct dentry *dentry;
809 dentry = unix_sk(u)->path.dentry;
811 touch_atime(&unix_sk(u)->path);
824 static int unix_mknod(const char *sun_path, umode_t mode, struct path *res)
826 struct dentry *dentry;
830 * Get the parent directory, calculate the hash for last
833 dentry = kern_path_create(AT_FDCWD, sun_path, &path, 0);
834 err = PTR_ERR(dentry);
839 * All right, let's create it.
841 err = security_path_mknod(&path, dentry, mode, 0);
843 err = vfs_mknod(path.dentry->d_inode, dentry, mode, 0);
845 res->mnt = mntget(path.mnt);
846 res->dentry = dget(dentry);
849 done_path_create(&path, dentry);
853 static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
855 struct sock *sk = sock->sk;
856 struct net *net = sock_net(sk);
857 struct unix_sock *u = unix_sk(sk);
858 struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
859 char *sun_path = sunaddr->sun_path;
862 struct unix_address *addr;
863 struct hlist_head *list;
866 if (sunaddr->sun_family != AF_UNIX)
869 if (addr_len == sizeof(short)) {
870 err = unix_autobind(sock);
874 err = unix_mkname(sunaddr, addr_len, &hash);
879 mutex_lock(&u->readlock);
886 addr = kmalloc(sizeof(*addr)+addr_len, GFP_KERNEL);
890 memcpy(addr->name, sunaddr, addr_len);
891 addr->len = addr_len;
892 addr->hash = hash ^ sk->sk_type;
893 atomic_set(&addr->refcnt, 1);
897 umode_t mode = S_IFSOCK |
898 (SOCK_INODE(sock)->i_mode & ~current_umask());
899 err = unix_mknod(sun_path, mode, &path);
903 unix_release_addr(addr);
906 addr->hash = UNIX_HASH_SIZE;
907 hash = path.dentry->d_inode->i_ino & (UNIX_HASH_SIZE-1);
908 spin_lock(&unix_table_lock);
910 list = &unix_socket_table[hash];
912 spin_lock(&unix_table_lock);
914 if (__unix_find_socket_byname(net, sunaddr, addr_len,
915 sk->sk_type, hash)) {
916 unix_release_addr(addr);
920 list = &unix_socket_table[addr->hash];
924 __unix_remove_socket(sk);
926 __unix_insert_socket(list, sk);
929 spin_unlock(&unix_table_lock);
931 mutex_unlock(&u->readlock);
936 static void unix_state_double_lock(struct sock *sk1, struct sock *sk2)
938 if (unlikely(sk1 == sk2) || !sk2) {
939 unix_state_lock(sk1);
943 unix_state_lock(sk1);
944 unix_state_lock_nested(sk2);
946 unix_state_lock(sk2);
947 unix_state_lock_nested(sk1);
951 static void unix_state_double_unlock(struct sock *sk1, struct sock *sk2)
953 if (unlikely(sk1 == sk2) || !sk2) {
954 unix_state_unlock(sk1);
957 unix_state_unlock(sk1);
958 unix_state_unlock(sk2);
961 static int unix_dgram_connect(struct socket *sock, struct sockaddr *addr,
964 struct sock *sk = sock->sk;
965 struct net *net = sock_net(sk);
966 struct sockaddr_un *sunaddr = (struct sockaddr_un *)addr;
971 if (addr->sa_family != AF_UNSPEC) {
972 err = unix_mkname(sunaddr, alen, &hash);
977 if (test_bit(SOCK_PASSCRED, &sock->flags) &&
978 !unix_sk(sk)->addr && (err = unix_autobind(sock)) != 0)
982 other = unix_find_other(net, sunaddr, alen, sock->type, hash, &err);
986 unix_state_double_lock(sk, other);
988 /* Apparently VFS overslept socket death. Retry. */
989 if (sock_flag(other, SOCK_DEAD)) {
990 unix_state_double_unlock(sk, other);
996 if (!unix_may_send(sk, other))
999 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1005 * 1003.1g breaking connected state with AF_UNSPEC
1008 unix_state_double_lock(sk, other);
1012 * If it was connected, reconnect.
1014 if (unix_peer(sk)) {
1015 struct sock *old_peer = unix_peer(sk);
1016 unix_peer(sk) = other;
1017 unix_state_double_unlock(sk, other);
1019 if (other != old_peer)
1020 unix_dgram_disconnected(sk, old_peer);
1023 unix_peer(sk) = other;
1024 unix_state_double_unlock(sk, other);
1029 unix_state_double_unlock(sk, other);
1035 static long unix_wait_for_peer(struct sock *other, long timeo)
1037 struct unix_sock *u = unix_sk(other);
1041 prepare_to_wait_exclusive(&u->peer_wait, &wait, TASK_INTERRUPTIBLE);
1043 sched = !sock_flag(other, SOCK_DEAD) &&
1044 !(other->sk_shutdown & RCV_SHUTDOWN) &&
1045 unix_recvq_full(other);
1047 unix_state_unlock(other);
1050 timeo = schedule_timeout(timeo);
1052 finish_wait(&u->peer_wait, &wait);
1056 static int unix_stream_connect(struct socket *sock, struct sockaddr *uaddr,
1057 int addr_len, int flags)
1059 struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
1060 struct sock *sk = sock->sk;
1061 struct net *net = sock_net(sk);
1062 struct unix_sock *u = unix_sk(sk), *newu, *otheru;
1063 struct sock *newsk = NULL;
1064 struct sock *other = NULL;
1065 struct sk_buff *skb = NULL;
1071 err = unix_mkname(sunaddr, addr_len, &hash);
1076 if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr &&
1077 (err = unix_autobind(sock)) != 0)
1080 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
1082 /* First of all allocate resources.
1083 If we will make it after state is locked,
1084 we will have to recheck all again in any case.
1089 /* create new sock for complete connection */
1090 newsk = unix_create1(sock_net(sk), NULL);
1094 /* Allocate skb for sending to listening sock */
1095 skb = sock_wmalloc(newsk, 1, 0, GFP_KERNEL);
1100 /* Find listening sock. */
1101 other = unix_find_other(net, sunaddr, addr_len, sk->sk_type, hash, &err);
1105 /* Latch state of peer */
1106 unix_state_lock(other);
1108 /* Apparently VFS overslept socket death. Retry. */
1109 if (sock_flag(other, SOCK_DEAD)) {
1110 unix_state_unlock(other);
1115 err = -ECONNREFUSED;
1116 if (other->sk_state != TCP_LISTEN)
1118 if (other->sk_shutdown & RCV_SHUTDOWN)
1121 if (unix_recvq_full(other)) {
1126 timeo = unix_wait_for_peer(other, timeo);
1128 err = sock_intr_errno(timeo);
1129 if (signal_pending(current))
1137 It is tricky place. We need to grab our state lock and cannot
1138 drop lock on peer. It is dangerous because deadlock is
1139 possible. Connect to self case and simultaneous
1140 attempt to connect are eliminated by checking socket
1141 state. other is TCP_LISTEN, if sk is TCP_LISTEN we
1142 check this before attempt to grab lock.
1144 Well, and we have to recheck the state after socket locked.
1150 /* This is ok... continue with connect */
1152 case TCP_ESTABLISHED:
1153 /* Socket is already connected */
1161 unix_state_lock_nested(sk);
1163 if (sk->sk_state != st) {
1164 unix_state_unlock(sk);
1165 unix_state_unlock(other);
1170 err = security_unix_stream_connect(sk, other, newsk);
1172 unix_state_unlock(sk);
1176 /* The way is open! Fastly set all the necessary fields... */
1179 unix_peer(newsk) = sk;
1180 newsk->sk_state = TCP_ESTABLISHED;
1181 newsk->sk_type = sk->sk_type;
1182 init_peercred(newsk);
1183 newu = unix_sk(newsk);
1184 RCU_INIT_POINTER(newsk->sk_wq, &newu->peer_wq);
1185 otheru = unix_sk(other);
1187 /* copy address information from listening to new sock*/
1189 atomic_inc(&otheru->addr->refcnt);
1190 newu->addr = otheru->addr;
1192 if (otheru->path.dentry) {
1193 path_get(&otheru->path);
1194 newu->path = otheru->path;
1197 /* Set credentials */
1198 copy_peercred(sk, other);
1200 sock->state = SS_CONNECTED;
1201 sk->sk_state = TCP_ESTABLISHED;
1204 smp_mb__after_atomic_inc(); /* sock_hold() does an atomic_inc() */
1205 unix_peer(sk) = newsk;
1207 unix_state_unlock(sk);
1209 /* take ten and and send info to listening sock */
1210 spin_lock(&other->sk_receive_queue.lock);
1211 __skb_queue_tail(&other->sk_receive_queue, skb);
1212 spin_unlock(&other->sk_receive_queue.lock);
1213 unix_state_unlock(other);
1214 other->sk_data_ready(other, 0);
1220 unix_state_unlock(other);
1225 unix_release_sock(newsk, 0);
1231 static int unix_socketpair(struct socket *socka, struct socket *sockb)
1233 struct sock *ska = socka->sk, *skb = sockb->sk;
1235 /* Join our sockets back to back */
1238 unix_peer(ska) = skb;
1239 unix_peer(skb) = ska;
1243 if (ska->sk_type != SOCK_DGRAM) {
1244 ska->sk_state = TCP_ESTABLISHED;
1245 skb->sk_state = TCP_ESTABLISHED;
1246 socka->state = SS_CONNECTED;
1247 sockb->state = SS_CONNECTED;
1252 static void unix_sock_inherit_flags(const struct socket *old,
1255 if (test_bit(SOCK_PASSCRED, &old->flags))
1256 set_bit(SOCK_PASSCRED, &new->flags);
1257 if (test_bit(SOCK_PASSSEC, &old->flags))
1258 set_bit(SOCK_PASSSEC, &new->flags);
1261 static int unix_accept(struct socket *sock, struct socket *newsock, int flags)
1263 struct sock *sk = sock->sk;
1265 struct sk_buff *skb;
1269 if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
1273 if (sk->sk_state != TCP_LISTEN)
1276 /* If socket state is TCP_LISTEN it cannot change (for now...),
1277 * so that no locks are necessary.
1280 skb = skb_recv_datagram(sk, 0, flags&O_NONBLOCK, &err);
1282 /* This means receive shutdown. */
1289 skb_free_datagram(sk, skb);
1290 wake_up_interruptible(&unix_sk(sk)->peer_wait);
1292 /* attach accepted sock to socket */
1293 unix_state_lock(tsk);
1294 newsock->state = SS_CONNECTED;
1295 unix_sock_inherit_flags(sock, newsock);
1296 sock_graft(tsk, newsock);
1297 unix_state_unlock(tsk);
1305 static int unix_getname(struct socket *sock, struct sockaddr *uaddr, int *uaddr_len, int peer)
1307 struct sock *sk = sock->sk;
1308 struct unix_sock *u;
1309 DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr, uaddr);
1313 sk = unix_peer_get(sk);
1324 unix_state_lock(sk);
1326 sunaddr->sun_family = AF_UNIX;
1327 sunaddr->sun_path[0] = 0;
1328 *uaddr_len = sizeof(short);
1330 struct unix_address *addr = u->addr;
1332 *uaddr_len = addr->len;
1333 memcpy(sunaddr, addr->name, *uaddr_len);
1335 unix_state_unlock(sk);
1341 static void unix_detach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1345 scm->fp = UNIXCB(skb).fp;
1346 UNIXCB(skb).fp = NULL;
1348 for (i = scm->fp->count-1; i >= 0; i--)
1349 unix_notinflight(scm->fp->fp[i]);
1352 static void unix_destruct_scm(struct sk_buff *skb)
1354 struct scm_cookie scm;
1355 memset(&scm, 0, sizeof(scm));
1356 scm.pid = UNIXCB(skb).pid;
1358 unix_detach_fds(&scm, skb);
1360 /* Alas, it calls VFS */
1361 /* So fscking what? fput() had been SMP-safe since the last Summer */
1366 #define MAX_RECURSION_LEVEL 4
1368 static int unix_attach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1371 unsigned char max_level = 0;
1372 int unix_sock_count = 0;
1374 for (i = scm->fp->count - 1; i >= 0; i--) {
1375 struct sock *sk = unix_get_socket(scm->fp->fp[i]);
1379 max_level = max(max_level,
1380 unix_sk(sk)->recursion_level);
1383 if (unlikely(max_level > MAX_RECURSION_LEVEL))
1384 return -ETOOMANYREFS;
1387 * Need to duplicate file references for the sake of garbage
1388 * collection. Otherwise a socket in the fps might become a
1389 * candidate for GC while the skb is not yet queued.
1391 UNIXCB(skb).fp = scm_fp_dup(scm->fp);
1392 if (!UNIXCB(skb).fp)
1395 if (unix_sock_count) {
1396 for (i = scm->fp->count - 1; i >= 0; i--)
1397 unix_inflight(scm->fp->fp[i]);
1402 static int unix_scm_to_skb(struct scm_cookie *scm, struct sk_buff *skb, bool send_fds)
1406 UNIXCB(skb).pid = get_pid(scm->pid);
1407 UNIXCB(skb).uid = scm->creds.uid;
1408 UNIXCB(skb).gid = scm->creds.gid;
1409 UNIXCB(skb).fp = NULL;
1410 if (scm->fp && send_fds)
1411 err = unix_attach_fds(scm, skb);
1413 skb->destructor = unix_destruct_scm;
1418 * Some apps rely on write() giving SCM_CREDENTIALS
1419 * We include credentials if source or destination socket
1420 * asserted SOCK_PASSCRED.
1422 static void maybe_add_creds(struct sk_buff *skb, const struct socket *sock,
1423 const struct sock *other)
1425 if (UNIXCB(skb).pid)
1427 if (test_bit(SOCK_PASSCRED, &sock->flags) ||
1428 !other->sk_socket ||
1429 test_bit(SOCK_PASSCRED, &other->sk_socket->flags)) {
1430 UNIXCB(skb).pid = get_pid(task_tgid(current));
1431 current_uid_gid(&UNIXCB(skb).uid, &UNIXCB(skb).gid);
1436 * Send AF_UNIX data.
1439 static int unix_dgram_sendmsg(struct kiocb *kiocb, struct socket *sock,
1440 struct msghdr *msg, size_t len)
1442 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1443 struct sock *sk = sock->sk;
1444 struct net *net = sock_net(sk);
1445 struct unix_sock *u = unix_sk(sk);
1446 struct sockaddr_un *sunaddr = msg->msg_name;
1447 struct sock *other = NULL;
1448 int namelen = 0; /* fake GCC */
1451 struct sk_buff *skb;
1453 struct scm_cookie tmp_scm;
1457 if (NULL == siocb->scm)
1458 siocb->scm = &tmp_scm;
1460 err = scm_send(sock, msg, siocb->scm, false);
1465 if (msg->msg_flags&MSG_OOB)
1468 if (msg->msg_namelen) {
1469 err = unix_mkname(sunaddr, msg->msg_namelen, &hash);
1476 other = unix_peer_get(sk);
1481 if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr
1482 && (err = unix_autobind(sock)) != 0)
1486 if (len > sk->sk_sndbuf - 32)
1489 if (len > SKB_MAX_ALLOC)
1490 data_len = min_t(size_t,
1491 len - SKB_MAX_ALLOC,
1492 MAX_SKB_FRAGS * PAGE_SIZE);
1494 skb = sock_alloc_send_pskb(sk, len - data_len, data_len,
1495 msg->msg_flags & MSG_DONTWAIT, &err);
1499 err = unix_scm_to_skb(siocb->scm, skb, true);
1502 max_level = err + 1;
1503 unix_get_secdata(siocb->scm, skb);
1505 skb_put(skb, len - data_len);
1506 skb->data_len = data_len;
1508 err = skb_copy_datagram_from_iovec(skb, 0, msg->msg_iov, 0, len);
1512 timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1517 if (sunaddr == NULL)
1520 other = unix_find_other(net, sunaddr, namelen, sk->sk_type,
1526 if (sk_filter(other, skb) < 0) {
1527 /* Toss the packet but do not return any error to the sender */
1532 unix_state_lock(other);
1534 if (!unix_may_send(sk, other))
1537 if (sock_flag(other, SOCK_DEAD)) {
1539 * Check with 1003.1g - what should
1542 unix_state_unlock(other);
1546 unix_state_lock(sk);
1547 if (unix_peer(sk) == other) {
1548 unix_peer(sk) = NULL;
1549 unix_state_unlock(sk);
1551 unix_dgram_disconnected(sk, other);
1553 err = -ECONNREFUSED;
1555 unix_state_unlock(sk);
1565 if (other->sk_shutdown & RCV_SHUTDOWN)
1568 if (sk->sk_type != SOCK_SEQPACKET) {
1569 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1574 if (unix_peer(other) != sk && unix_recvq_full(other)) {
1580 timeo = unix_wait_for_peer(other, timeo);
1582 err = sock_intr_errno(timeo);
1583 if (signal_pending(current))
1589 if (sock_flag(other, SOCK_RCVTSTAMP))
1590 __net_timestamp(skb);
1591 maybe_add_creds(skb, sock, other);
1592 skb_queue_tail(&other->sk_receive_queue, skb);
1593 if (max_level > unix_sk(other)->recursion_level)
1594 unix_sk(other)->recursion_level = max_level;
1595 unix_state_unlock(other);
1596 other->sk_data_ready(other, len);
1598 scm_destroy(siocb->scm);
1602 unix_state_unlock(other);
1608 scm_destroy(siocb->scm);
1613 static int unix_stream_sendmsg(struct kiocb *kiocb, struct socket *sock,
1614 struct msghdr *msg, size_t len)
1616 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1617 struct sock *sk = sock->sk;
1618 struct sock *other = NULL;
1620 struct sk_buff *skb;
1622 struct scm_cookie tmp_scm;
1623 bool fds_sent = false;
1626 if (NULL == siocb->scm)
1627 siocb->scm = &tmp_scm;
1629 err = scm_send(sock, msg, siocb->scm, false);
1634 if (msg->msg_flags&MSG_OOB)
1637 if (msg->msg_namelen) {
1638 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
1642 other = unix_peer(sk);
1647 if (sk->sk_shutdown & SEND_SHUTDOWN)
1650 while (sent < len) {
1652 * Optimisation for the fact that under 0.01% of X
1653 * messages typically need breaking up.
1658 /* Keep two messages in the pipe so it schedules better */
1659 if (size > ((sk->sk_sndbuf >> 1) - 64))
1660 size = (sk->sk_sndbuf >> 1) - 64;
1662 if (size > SKB_MAX_ALLOC)
1663 size = SKB_MAX_ALLOC;
1669 skb = sock_alloc_send_skb(sk, size, msg->msg_flags&MSG_DONTWAIT,
1676 * If you pass two values to the sock_alloc_send_skb
1677 * it tries to grab the large buffer with GFP_NOFS
1678 * (which can fail easily), and if it fails grab the
1679 * fallback size buffer which is under a page and will
1682 size = min_t(int, size, skb_tailroom(skb));
1685 /* Only send the fds in the first buffer */
1686 err = unix_scm_to_skb(siocb->scm, skb, !fds_sent);
1691 max_level = err + 1;
1694 err = memcpy_fromiovec(skb_put(skb, size), msg->msg_iov, size);
1700 unix_state_lock(other);
1702 if (sock_flag(other, SOCK_DEAD) ||
1703 (other->sk_shutdown & RCV_SHUTDOWN))
1706 maybe_add_creds(skb, sock, other);
1707 skb_queue_tail(&other->sk_receive_queue, skb);
1708 if (max_level > unix_sk(other)->recursion_level)
1709 unix_sk(other)->recursion_level = max_level;
1710 unix_state_unlock(other);
1711 other->sk_data_ready(other, size);
1715 scm_destroy(siocb->scm);
1721 unix_state_unlock(other);
1724 if (sent == 0 && !(msg->msg_flags&MSG_NOSIGNAL))
1725 send_sig(SIGPIPE, current, 0);
1728 scm_destroy(siocb->scm);
1730 return sent ? : err;
1733 static int unix_seqpacket_sendmsg(struct kiocb *kiocb, struct socket *sock,
1734 struct msghdr *msg, size_t len)
1737 struct sock *sk = sock->sk;
1739 err = sock_error(sk);
1743 if (sk->sk_state != TCP_ESTABLISHED)
1746 if (msg->msg_namelen)
1747 msg->msg_namelen = 0;
1749 return unix_dgram_sendmsg(kiocb, sock, msg, len);
1752 static int unix_seqpacket_recvmsg(struct kiocb *iocb, struct socket *sock,
1753 struct msghdr *msg, size_t size,
1756 struct sock *sk = sock->sk;
1758 if (sk->sk_state != TCP_ESTABLISHED)
1761 return unix_dgram_recvmsg(iocb, sock, msg, size, flags);
1764 static void unix_copy_addr(struct msghdr *msg, struct sock *sk)
1766 struct unix_sock *u = unix_sk(sk);
1769 msg->msg_namelen = u->addr->len;
1770 memcpy(msg->msg_name, u->addr->name, u->addr->len);
1774 static int unix_dgram_recvmsg(struct kiocb *iocb, struct socket *sock,
1775 struct msghdr *msg, size_t size,
1778 struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1779 struct scm_cookie tmp_scm;
1780 struct sock *sk = sock->sk;
1781 struct unix_sock *u = unix_sk(sk);
1782 int noblock = flags & MSG_DONTWAIT;
1783 struct sk_buff *skb;
1791 err = mutex_lock_interruptible(&u->readlock);
1793 err = sock_intr_errno(sock_rcvtimeo(sk, noblock));
1797 skip = sk_peek_offset(sk, flags);
1799 skb = __skb_recv_datagram(sk, flags, &peeked, &skip, &err);
1801 unix_state_lock(sk);
1802 /* Signal EOF on disconnected non-blocking SEQPACKET socket. */
1803 if (sk->sk_type == SOCK_SEQPACKET && err == -EAGAIN &&
1804 (sk->sk_shutdown & RCV_SHUTDOWN))
1806 unix_state_unlock(sk);
1810 wake_up_interruptible_sync_poll(&u->peer_wait,
1811 POLLOUT | POLLWRNORM | POLLWRBAND);
1814 unix_copy_addr(msg, skb->sk);
1816 if (size > skb->len - skip)
1817 size = skb->len - skip;
1818 else if (size < skb->len - skip)
1819 msg->msg_flags |= MSG_TRUNC;
1821 err = skb_copy_datagram_iovec(skb, skip, msg->msg_iov, size);
1825 if (sock_flag(sk, SOCK_RCVTSTAMP))
1826 __sock_recv_timestamp(msg, sk, skb);
1829 siocb->scm = &tmp_scm;
1830 memset(&tmp_scm, 0, sizeof(tmp_scm));
1832 scm_set_cred(siocb->scm, UNIXCB(skb).pid, UNIXCB(skb).uid, UNIXCB(skb).gid);
1833 unix_set_secdata(siocb->scm, skb);
1835 if (!(flags & MSG_PEEK)) {
1837 unix_detach_fds(siocb->scm, skb);
1839 sk_peek_offset_bwd(sk, skb->len);
1841 /* It is questionable: on PEEK we could:
1842 - do not return fds - good, but too simple 8)
1843 - return fds, and do not return them on read (old strategy,
1845 - clone fds (I chose it for now, it is the most universal
1848 POSIX 1003.1g does not actually define this clearly
1849 at all. POSIX 1003.1g doesn't define a lot of things
1854 sk_peek_offset_fwd(sk, size);
1857 siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1859 err = (flags & MSG_TRUNC) ? skb->len - skip : size;
1861 scm_recv(sock, msg, siocb->scm, flags);
1864 skb_free_datagram(sk, skb);
1866 mutex_unlock(&u->readlock);
1872 * Sleep until more data has arrived. But check for races..
1874 static long unix_stream_data_wait(struct sock *sk, long timeo,
1875 struct sk_buff *last)
1879 unix_state_lock(sk);
1882 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1884 if (skb_peek_tail(&sk->sk_receive_queue) != last ||
1886 (sk->sk_shutdown & RCV_SHUTDOWN) ||
1887 signal_pending(current) ||
1891 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1892 unix_state_unlock(sk);
1893 timeo = schedule_timeout(timeo);
1894 unix_state_lock(sk);
1895 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1898 finish_wait(sk_sleep(sk), &wait);
1899 unix_state_unlock(sk);
1903 static int unix_stream_recvmsg(struct kiocb *iocb, struct socket *sock,
1904 struct msghdr *msg, size_t size,
1907 struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1908 struct scm_cookie tmp_scm;
1909 struct sock *sk = sock->sk;
1910 struct unix_sock *u = unix_sk(sk);
1911 struct sockaddr_un *sunaddr = msg->msg_name;
1913 int check_creds = 0;
1920 if (sk->sk_state != TCP_ESTABLISHED)
1927 target = sock_rcvlowat(sk, flags&MSG_WAITALL, size);
1928 timeo = sock_rcvtimeo(sk, flags&MSG_DONTWAIT);
1930 /* Lock the socket to prevent queue disordering
1931 * while sleeps in memcpy_tomsg
1935 siocb->scm = &tmp_scm;
1936 memset(&tmp_scm, 0, sizeof(tmp_scm));
1939 err = mutex_lock_interruptible(&u->readlock);
1941 err = sock_intr_errno(timeo);
1947 struct sk_buff *skb, *last;
1949 unix_state_lock(sk);
1950 last = skb = skb_peek(&sk->sk_receive_queue);
1953 unix_sk(sk)->recursion_level = 0;
1954 if (copied >= target)
1958 * POSIX 1003.1g mandates this order.
1961 err = sock_error(sk);
1964 if (sk->sk_shutdown & RCV_SHUTDOWN)
1967 unix_state_unlock(sk);
1971 mutex_unlock(&u->readlock);
1973 timeo = unix_stream_data_wait(sk, timeo, last);
1975 if (signal_pending(current)
1976 || mutex_lock_interruptible(&u->readlock)) {
1977 err = sock_intr_errno(timeo);
1983 unix_state_unlock(sk);
1987 skip = sk_peek_offset(sk, flags);
1988 while (skip >= skb->len) {
1991 skb = skb_peek_next(skb, &sk->sk_receive_queue);
1996 unix_state_unlock(sk);
1999 /* Never glue messages from different writers */
2000 if ((UNIXCB(skb).pid != siocb->scm->pid) ||
2001 !uid_eq(UNIXCB(skb).uid, siocb->scm->creds.uid) ||
2002 !gid_eq(UNIXCB(skb).gid, siocb->scm->creds.gid))
2004 } else if (test_bit(SOCK_PASSCRED, &sock->flags)) {
2005 /* Copy credentials */
2006 scm_set_cred(siocb->scm, UNIXCB(skb).pid, UNIXCB(skb).uid, UNIXCB(skb).gid);
2010 /* Copy address just once */
2012 unix_copy_addr(msg, skb->sk);
2016 chunk = min_t(unsigned int, skb->len - skip, size);
2017 if (memcpy_toiovec(msg->msg_iov, skb->data + skip, chunk)) {
2025 /* Mark read part of skb as used */
2026 if (!(flags & MSG_PEEK)) {
2027 skb_pull(skb, chunk);
2029 sk_peek_offset_bwd(sk, chunk);
2032 unix_detach_fds(siocb->scm, skb);
2037 skb_unlink(skb, &sk->sk_receive_queue);
2043 /* It is questionable, see note in unix_dgram_recvmsg.
2046 siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
2048 sk_peek_offset_fwd(sk, chunk);
2054 mutex_unlock(&u->readlock);
2055 scm_recv(sock, msg, siocb->scm, flags);
2057 return copied ? : err;
2060 static int unix_shutdown(struct socket *sock, int mode)
2062 struct sock *sk = sock->sk;
2065 if (mode < SHUT_RD || mode > SHUT_RDWR)
2068 * SHUT_RD (0) -> RCV_SHUTDOWN (1)
2069 * SHUT_WR (1) -> SEND_SHUTDOWN (2)
2070 * SHUT_RDWR (2) -> SHUTDOWN_MASK (3)
2074 unix_state_lock(sk);
2075 sk->sk_shutdown |= mode;
2076 other = unix_peer(sk);
2079 unix_state_unlock(sk);
2080 sk->sk_state_change(sk);
2083 (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET)) {
2087 if (mode&RCV_SHUTDOWN)
2088 peer_mode |= SEND_SHUTDOWN;
2089 if (mode&SEND_SHUTDOWN)
2090 peer_mode |= RCV_SHUTDOWN;
2091 unix_state_lock(other);
2092 other->sk_shutdown |= peer_mode;
2093 unix_state_unlock(other);
2094 other->sk_state_change(other);
2095 if (peer_mode == SHUTDOWN_MASK)
2096 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_HUP);
2097 else if (peer_mode & RCV_SHUTDOWN)
2098 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_IN);
2106 long unix_inq_len(struct sock *sk)
2108 struct sk_buff *skb;
2111 if (sk->sk_state == TCP_LISTEN)
2114 spin_lock(&sk->sk_receive_queue.lock);
2115 if (sk->sk_type == SOCK_STREAM ||
2116 sk->sk_type == SOCK_SEQPACKET) {
2117 skb_queue_walk(&sk->sk_receive_queue, skb)
2120 skb = skb_peek(&sk->sk_receive_queue);
2124 spin_unlock(&sk->sk_receive_queue.lock);
2128 EXPORT_SYMBOL_GPL(unix_inq_len);
2130 long unix_outq_len(struct sock *sk)
2132 return sk_wmem_alloc_get(sk);
2134 EXPORT_SYMBOL_GPL(unix_outq_len);
2136 static int unix_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2138 struct sock *sk = sock->sk;
2144 amount = unix_outq_len(sk);
2145 err = put_user(amount, (int __user *)arg);
2148 amount = unix_inq_len(sk);
2152 err = put_user(amount, (int __user *)arg);
2161 static unsigned int unix_poll(struct file *file, struct socket *sock, poll_table *wait)
2163 struct sock *sk = sock->sk;
2166 sock_poll_wait(file, sk_sleep(sk), wait);
2169 /* exceptional events? */
2172 if (sk->sk_shutdown == SHUTDOWN_MASK)
2174 if (sk->sk_shutdown & RCV_SHUTDOWN)
2175 mask |= POLLRDHUP | POLLIN | POLLRDNORM;
2178 if (!skb_queue_empty(&sk->sk_receive_queue))
2179 mask |= POLLIN | POLLRDNORM;
2181 /* Connection-based need to check for termination and startup */
2182 if ((sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) &&
2183 sk->sk_state == TCP_CLOSE)
2187 * we set writable also when the other side has shut down the
2188 * connection. This prevents stuck sockets.
2190 if (unix_writable(sk))
2191 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2196 static unsigned int unix_dgram_poll(struct file *file, struct socket *sock,
2199 struct sock *sk = sock->sk, *other;
2200 unsigned int mask, writable;
2202 sock_poll_wait(file, sk_sleep(sk), wait);
2205 /* exceptional events? */
2206 if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue))
2208 (sock_flag(sk, SOCK_SELECT_ERR_QUEUE) ? POLLPRI : 0);
2210 if (sk->sk_shutdown & RCV_SHUTDOWN)
2211 mask |= POLLRDHUP | POLLIN | POLLRDNORM;
2212 if (sk->sk_shutdown == SHUTDOWN_MASK)
2216 if (!skb_queue_empty(&sk->sk_receive_queue))
2217 mask |= POLLIN | POLLRDNORM;
2219 /* Connection-based need to check for termination and startup */
2220 if (sk->sk_type == SOCK_SEQPACKET) {
2221 if (sk->sk_state == TCP_CLOSE)
2223 /* connection hasn't started yet? */
2224 if (sk->sk_state == TCP_SYN_SENT)
2228 /* No write status requested, avoid expensive OUT tests. */
2229 if (!(poll_requested_events(wait) & (POLLWRBAND|POLLWRNORM|POLLOUT)))
2232 writable = unix_writable(sk);
2233 other = unix_peer_get(sk);
2235 if (unix_peer(other) != sk) {
2236 sock_poll_wait(file, &unix_sk(other)->peer_wait, wait);
2237 if (unix_recvq_full(other))
2244 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2246 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
2251 #ifdef CONFIG_PROC_FS
2253 #define BUCKET_SPACE (BITS_PER_LONG - (UNIX_HASH_BITS + 1) - 1)
2255 #define get_bucket(x) ((x) >> BUCKET_SPACE)
2256 #define get_offset(x) ((x) & ((1L << BUCKET_SPACE) - 1))
2257 #define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
2259 static struct sock *unix_from_bucket(struct seq_file *seq, loff_t *pos)
2261 unsigned long offset = get_offset(*pos);
2262 unsigned long bucket = get_bucket(*pos);
2264 unsigned long count = 0;
2266 for (sk = sk_head(&unix_socket_table[bucket]); sk; sk = sk_next(sk)) {
2267 if (sock_net(sk) != seq_file_net(seq))
2269 if (++count == offset)
2276 static struct sock *unix_next_socket(struct seq_file *seq,
2280 unsigned long bucket;
2282 while (sk > (struct sock *)SEQ_START_TOKEN) {
2286 if (sock_net(sk) == seq_file_net(seq))
2291 sk = unix_from_bucket(seq, pos);
2296 bucket = get_bucket(*pos) + 1;
2297 *pos = set_bucket_offset(bucket, 1);
2298 } while (bucket < ARRAY_SIZE(unix_socket_table));
2303 static void *unix_seq_start(struct seq_file *seq, loff_t *pos)
2304 __acquires(unix_table_lock)
2306 spin_lock(&unix_table_lock);
2309 return SEQ_START_TOKEN;
2311 if (get_bucket(*pos) >= ARRAY_SIZE(unix_socket_table))
2314 return unix_next_socket(seq, NULL, pos);
2317 static void *unix_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2320 return unix_next_socket(seq, v, pos);
2323 static void unix_seq_stop(struct seq_file *seq, void *v)
2324 __releases(unix_table_lock)
2326 spin_unlock(&unix_table_lock);
2329 static int unix_seq_show(struct seq_file *seq, void *v)
2332 if (v == SEQ_START_TOKEN)
2333 seq_puts(seq, "Num RefCount Protocol Flags Type St "
2337 struct unix_sock *u = unix_sk(s);
2340 seq_printf(seq, "%pK: %08X %08X %08X %04X %02X %5lu",
2342 atomic_read(&s->sk_refcnt),
2344 s->sk_state == TCP_LISTEN ? __SO_ACCEPTCON : 0,
2347 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTED : SS_UNCONNECTED) :
2348 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTING : SS_DISCONNECTING),
2356 len = u->addr->len - sizeof(short);
2357 if (!UNIX_ABSTRACT(s))
2363 for ( ; i < len; i++)
2364 seq_putc(seq, u->addr->name->sun_path[i]);
2366 unix_state_unlock(s);
2367 seq_putc(seq, '\n');
2373 static const struct seq_operations unix_seq_ops = {
2374 .start = unix_seq_start,
2375 .next = unix_seq_next,
2376 .stop = unix_seq_stop,
2377 .show = unix_seq_show,
2380 static int unix_seq_open(struct inode *inode, struct file *file)
2382 return seq_open_net(inode, file, &unix_seq_ops,
2383 sizeof(struct seq_net_private));
2386 static const struct file_operations unix_seq_fops = {
2387 .owner = THIS_MODULE,
2388 .open = unix_seq_open,
2390 .llseek = seq_lseek,
2391 .release = seq_release_net,
2396 static const struct net_proto_family unix_family_ops = {
2398 .create = unix_create,
2399 .owner = THIS_MODULE,
2403 static int __net_init unix_net_init(struct net *net)
2405 int error = -ENOMEM;
2407 net->unx.sysctl_max_dgram_qlen = 10;
2408 if (unix_sysctl_register(net))
2411 #ifdef CONFIG_PROC_FS
2412 if (!proc_create("unix", 0, net->proc_net, &unix_seq_fops)) {
2413 unix_sysctl_unregister(net);
2422 static void __net_exit unix_net_exit(struct net *net)
2424 unix_sysctl_unregister(net);
2425 remove_proc_entry("unix", net->proc_net);
2428 static struct pernet_operations unix_net_ops = {
2429 .init = unix_net_init,
2430 .exit = unix_net_exit,
2433 static int __init af_unix_init(void)
2437 BUILD_BUG_ON(sizeof(struct unix_skb_parms) > FIELD_SIZEOF(struct sk_buff, cb));
2439 rc = proto_register(&unix_proto, 1);
2441 printk(KERN_CRIT "%s: Cannot create unix_sock SLAB cache!\n",
2446 sock_register(&unix_family_ops);
2447 register_pernet_subsys(&unix_net_ops);
2452 static void __exit af_unix_exit(void)
2454 sock_unregister(PF_UNIX);
2455 proto_unregister(&unix_proto);
2456 unregister_pernet_subsys(&unix_net_ops);
2459 /* Earlier than device_initcall() so that other drivers invoking
2460 request_module() don't end up in a loop when modprobe tries
2461 to use a UNIX socket. But later than subsys_initcall() because
2462 we depend on stuff initialised there */
2463 fs_initcall(af_unix_init);
2464 module_exit(af_unix_exit);
2466 MODULE_LICENSE("GPL");
2467 MODULE_ALIAS_NETPROTO(PF_UNIX);