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 void unix_set_peek_off(struct sock *sk, int val)
534 struct unix_sock *u = unix_sk(sk);
536 mutex_lock(&u->readlock);
537 sk->sk_peek_off = val;
538 mutex_unlock(&u->readlock);
542 static const struct proto_ops unix_stream_ops = {
544 .owner = THIS_MODULE,
545 .release = unix_release,
547 .connect = unix_stream_connect,
548 .socketpair = unix_socketpair,
549 .accept = unix_accept,
550 .getname = unix_getname,
553 .listen = unix_listen,
554 .shutdown = unix_shutdown,
555 .setsockopt = sock_no_setsockopt,
556 .getsockopt = sock_no_getsockopt,
557 .sendmsg = unix_stream_sendmsg,
558 .recvmsg = unix_stream_recvmsg,
559 .mmap = sock_no_mmap,
560 .sendpage = sock_no_sendpage,
561 .set_peek_off = unix_set_peek_off,
564 static const struct proto_ops unix_dgram_ops = {
566 .owner = THIS_MODULE,
567 .release = unix_release,
569 .connect = unix_dgram_connect,
570 .socketpair = unix_socketpair,
571 .accept = sock_no_accept,
572 .getname = unix_getname,
573 .poll = unix_dgram_poll,
575 .listen = sock_no_listen,
576 .shutdown = unix_shutdown,
577 .setsockopt = sock_no_setsockopt,
578 .getsockopt = sock_no_getsockopt,
579 .sendmsg = unix_dgram_sendmsg,
580 .recvmsg = unix_dgram_recvmsg,
581 .mmap = sock_no_mmap,
582 .sendpage = sock_no_sendpage,
583 .set_peek_off = unix_set_peek_off,
586 static const struct proto_ops unix_seqpacket_ops = {
588 .owner = THIS_MODULE,
589 .release = unix_release,
591 .connect = unix_stream_connect,
592 .socketpair = unix_socketpair,
593 .accept = unix_accept,
594 .getname = unix_getname,
595 .poll = unix_dgram_poll,
597 .listen = unix_listen,
598 .shutdown = unix_shutdown,
599 .setsockopt = sock_no_setsockopt,
600 .getsockopt = sock_no_getsockopt,
601 .sendmsg = unix_seqpacket_sendmsg,
602 .recvmsg = unix_seqpacket_recvmsg,
603 .mmap = sock_no_mmap,
604 .sendpage = sock_no_sendpage,
605 .set_peek_off = unix_set_peek_off,
608 static struct proto unix_proto = {
610 .owner = THIS_MODULE,
611 .obj_size = sizeof(struct unix_sock),
615 * AF_UNIX sockets do not interact with hardware, hence they
616 * dont trigger interrupts - so it's safe for them to have
617 * bh-unsafe locking for their sk_receive_queue.lock. Split off
618 * this special lock-class by reinitializing the spinlock key:
620 static struct lock_class_key af_unix_sk_receive_queue_lock_key;
622 static struct sock *unix_create1(struct net *net, struct socket *sock)
624 struct sock *sk = NULL;
627 atomic_long_inc(&unix_nr_socks);
628 if (atomic_long_read(&unix_nr_socks) > 2 * get_max_files())
631 sk = sk_alloc(net, PF_UNIX, GFP_KERNEL, &unix_proto);
635 sock_init_data(sock, sk);
636 lockdep_set_class(&sk->sk_receive_queue.lock,
637 &af_unix_sk_receive_queue_lock_key);
639 sk->sk_write_space = unix_write_space;
640 sk->sk_max_ack_backlog = net->unx.sysctl_max_dgram_qlen;
641 sk->sk_destruct = unix_sock_destructor;
643 u->path.dentry = NULL;
645 spin_lock_init(&u->lock);
646 atomic_long_set(&u->inflight, 0);
647 INIT_LIST_HEAD(&u->link);
648 mutex_init(&u->readlock); /* single task reading lock */
649 init_waitqueue_head(&u->peer_wait);
650 unix_insert_socket(unix_sockets_unbound(sk), sk);
653 atomic_long_dec(&unix_nr_socks);
656 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
662 static int unix_create(struct net *net, struct socket *sock, int protocol,
665 if (protocol && protocol != PF_UNIX)
666 return -EPROTONOSUPPORT;
668 sock->state = SS_UNCONNECTED;
670 switch (sock->type) {
672 sock->ops = &unix_stream_ops;
675 * Believe it or not BSD has AF_UNIX, SOCK_RAW though
679 sock->type = SOCK_DGRAM;
681 sock->ops = &unix_dgram_ops;
684 sock->ops = &unix_seqpacket_ops;
687 return -ESOCKTNOSUPPORT;
690 return unix_create1(net, sock) ? 0 : -ENOMEM;
693 static int unix_release(struct socket *sock)
695 struct sock *sk = sock->sk;
700 unix_release_sock(sk, 0);
706 static int unix_autobind(struct socket *sock)
708 struct sock *sk = sock->sk;
709 struct net *net = sock_net(sk);
710 struct unix_sock *u = unix_sk(sk);
711 static u32 ordernum = 1;
712 struct unix_address *addr;
714 unsigned int retries = 0;
716 mutex_lock(&u->readlock);
723 addr = kzalloc(sizeof(*addr) + sizeof(short) + 16, GFP_KERNEL);
727 addr->name->sun_family = AF_UNIX;
728 atomic_set(&addr->refcnt, 1);
731 addr->len = sprintf(addr->name->sun_path+1, "%05x", ordernum) + 1 + sizeof(short);
732 addr->hash = unix_hash_fold(csum_partial(addr->name, addr->len, 0));
734 spin_lock(&unix_table_lock);
735 ordernum = (ordernum+1)&0xFFFFF;
737 if (__unix_find_socket_byname(net, addr->name, addr->len, sock->type,
739 spin_unlock(&unix_table_lock);
741 * __unix_find_socket_byname() may take long time if many names
742 * are already in use.
745 /* Give up if all names seems to be in use. */
746 if (retries++ == 0xFFFFF) {
753 addr->hash ^= sk->sk_type;
755 __unix_remove_socket(sk);
757 __unix_insert_socket(&unix_socket_table[addr->hash], sk);
758 spin_unlock(&unix_table_lock);
761 out: mutex_unlock(&u->readlock);
765 static struct sock *unix_find_other(struct net *net,
766 struct sockaddr_un *sunname, int len,
767 int type, unsigned int hash, int *error)
773 if (sunname->sun_path[0]) {
775 err = kern_path(sunname->sun_path, LOOKUP_FOLLOW, &path);
778 inode = path.dentry->d_inode;
779 err = inode_permission(inode, MAY_WRITE);
784 if (!S_ISSOCK(inode->i_mode))
786 u = unix_find_socket_byinode(inode);
790 if (u->sk_type == type)
796 if (u->sk_type != type) {
802 u = unix_find_socket_byname(net, sunname, len, type, hash);
804 struct dentry *dentry;
805 dentry = unix_sk(u)->path.dentry;
807 touch_atime(&unix_sk(u)->path);
820 static int unix_mknod(const char *sun_path, umode_t mode, struct path *res)
822 struct dentry *dentry;
826 * Get the parent directory, calculate the hash for last
829 dentry = kern_path_create(AT_FDCWD, sun_path, &path, 0);
830 err = PTR_ERR(dentry);
835 * All right, let's create it.
837 err = security_path_mknod(&path, dentry, mode, 0);
839 err = vfs_mknod(path.dentry->d_inode, dentry, mode, 0);
841 res->mnt = mntget(path.mnt);
842 res->dentry = dget(dentry);
845 done_path_create(&path, dentry);
849 static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
851 struct sock *sk = sock->sk;
852 struct net *net = sock_net(sk);
853 struct unix_sock *u = unix_sk(sk);
854 struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
855 char *sun_path = sunaddr->sun_path;
858 struct unix_address *addr;
859 struct hlist_head *list;
862 if (sunaddr->sun_family != AF_UNIX)
865 if (addr_len == sizeof(short)) {
866 err = unix_autobind(sock);
870 err = unix_mkname(sunaddr, addr_len, &hash);
875 mutex_lock(&u->readlock);
882 addr = kmalloc(sizeof(*addr)+addr_len, GFP_KERNEL);
886 memcpy(addr->name, sunaddr, addr_len);
887 addr->len = addr_len;
888 addr->hash = hash ^ sk->sk_type;
889 atomic_set(&addr->refcnt, 1);
893 umode_t mode = S_IFSOCK |
894 (SOCK_INODE(sock)->i_mode & ~current_umask());
895 err = unix_mknod(sun_path, mode, &path);
899 unix_release_addr(addr);
902 addr->hash = UNIX_HASH_SIZE;
903 hash = path.dentry->d_inode->i_ino & (UNIX_HASH_SIZE-1);
904 spin_lock(&unix_table_lock);
906 list = &unix_socket_table[hash];
908 spin_lock(&unix_table_lock);
910 if (__unix_find_socket_byname(net, sunaddr, addr_len,
911 sk->sk_type, hash)) {
912 unix_release_addr(addr);
916 list = &unix_socket_table[addr->hash];
920 __unix_remove_socket(sk);
922 __unix_insert_socket(list, sk);
925 spin_unlock(&unix_table_lock);
927 mutex_unlock(&u->readlock);
932 static void unix_state_double_lock(struct sock *sk1, struct sock *sk2)
934 if (unlikely(sk1 == sk2) || !sk2) {
935 unix_state_lock(sk1);
939 unix_state_lock(sk1);
940 unix_state_lock_nested(sk2);
942 unix_state_lock(sk2);
943 unix_state_lock_nested(sk1);
947 static void unix_state_double_unlock(struct sock *sk1, struct sock *sk2)
949 if (unlikely(sk1 == sk2) || !sk2) {
950 unix_state_unlock(sk1);
953 unix_state_unlock(sk1);
954 unix_state_unlock(sk2);
957 static int unix_dgram_connect(struct socket *sock, struct sockaddr *addr,
960 struct sock *sk = sock->sk;
961 struct net *net = sock_net(sk);
962 struct sockaddr_un *sunaddr = (struct sockaddr_un *)addr;
967 if (addr->sa_family != AF_UNSPEC) {
968 err = unix_mkname(sunaddr, alen, &hash);
973 if (test_bit(SOCK_PASSCRED, &sock->flags) &&
974 !unix_sk(sk)->addr && (err = unix_autobind(sock)) != 0)
978 other = unix_find_other(net, sunaddr, alen, sock->type, hash, &err);
982 unix_state_double_lock(sk, other);
984 /* Apparently VFS overslept socket death. Retry. */
985 if (sock_flag(other, SOCK_DEAD)) {
986 unix_state_double_unlock(sk, other);
992 if (!unix_may_send(sk, other))
995 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1001 * 1003.1g breaking connected state with AF_UNSPEC
1004 unix_state_double_lock(sk, other);
1008 * If it was connected, reconnect.
1010 if (unix_peer(sk)) {
1011 struct sock *old_peer = unix_peer(sk);
1012 unix_peer(sk) = other;
1013 unix_state_double_unlock(sk, other);
1015 if (other != old_peer)
1016 unix_dgram_disconnected(sk, old_peer);
1019 unix_peer(sk) = other;
1020 unix_state_double_unlock(sk, other);
1025 unix_state_double_unlock(sk, other);
1031 static long unix_wait_for_peer(struct sock *other, long timeo)
1033 struct unix_sock *u = unix_sk(other);
1037 prepare_to_wait_exclusive(&u->peer_wait, &wait, TASK_INTERRUPTIBLE);
1039 sched = !sock_flag(other, SOCK_DEAD) &&
1040 !(other->sk_shutdown & RCV_SHUTDOWN) &&
1041 unix_recvq_full(other);
1043 unix_state_unlock(other);
1046 timeo = schedule_timeout(timeo);
1048 finish_wait(&u->peer_wait, &wait);
1052 static int unix_stream_connect(struct socket *sock, struct sockaddr *uaddr,
1053 int addr_len, int flags)
1055 struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
1056 struct sock *sk = sock->sk;
1057 struct net *net = sock_net(sk);
1058 struct unix_sock *u = unix_sk(sk), *newu, *otheru;
1059 struct sock *newsk = NULL;
1060 struct sock *other = NULL;
1061 struct sk_buff *skb = NULL;
1067 err = unix_mkname(sunaddr, addr_len, &hash);
1072 if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr &&
1073 (err = unix_autobind(sock)) != 0)
1076 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
1078 /* First of all allocate resources.
1079 If we will make it after state is locked,
1080 we will have to recheck all again in any case.
1085 /* create new sock for complete connection */
1086 newsk = unix_create1(sock_net(sk), NULL);
1090 /* Allocate skb for sending to listening sock */
1091 skb = sock_wmalloc(newsk, 1, 0, GFP_KERNEL);
1096 /* Find listening sock. */
1097 other = unix_find_other(net, sunaddr, addr_len, sk->sk_type, hash, &err);
1101 /* Latch state of peer */
1102 unix_state_lock(other);
1104 /* Apparently VFS overslept socket death. Retry. */
1105 if (sock_flag(other, SOCK_DEAD)) {
1106 unix_state_unlock(other);
1111 err = -ECONNREFUSED;
1112 if (other->sk_state != TCP_LISTEN)
1114 if (other->sk_shutdown & RCV_SHUTDOWN)
1117 if (unix_recvq_full(other)) {
1122 timeo = unix_wait_for_peer(other, timeo);
1124 err = sock_intr_errno(timeo);
1125 if (signal_pending(current))
1133 It is tricky place. We need to grab our state lock and cannot
1134 drop lock on peer. It is dangerous because deadlock is
1135 possible. Connect to self case and simultaneous
1136 attempt to connect are eliminated by checking socket
1137 state. other is TCP_LISTEN, if sk is TCP_LISTEN we
1138 check this before attempt to grab lock.
1140 Well, and we have to recheck the state after socket locked.
1146 /* This is ok... continue with connect */
1148 case TCP_ESTABLISHED:
1149 /* Socket is already connected */
1157 unix_state_lock_nested(sk);
1159 if (sk->sk_state != st) {
1160 unix_state_unlock(sk);
1161 unix_state_unlock(other);
1166 err = security_unix_stream_connect(sk, other, newsk);
1168 unix_state_unlock(sk);
1172 /* The way is open! Fastly set all the necessary fields... */
1175 unix_peer(newsk) = sk;
1176 newsk->sk_state = TCP_ESTABLISHED;
1177 newsk->sk_type = sk->sk_type;
1178 init_peercred(newsk);
1179 newu = unix_sk(newsk);
1180 RCU_INIT_POINTER(newsk->sk_wq, &newu->peer_wq);
1181 otheru = unix_sk(other);
1183 /* copy address information from listening to new sock*/
1185 atomic_inc(&otheru->addr->refcnt);
1186 newu->addr = otheru->addr;
1188 if (otheru->path.dentry) {
1189 path_get(&otheru->path);
1190 newu->path = otheru->path;
1193 /* Set credentials */
1194 copy_peercred(sk, other);
1196 sock->state = SS_CONNECTED;
1197 sk->sk_state = TCP_ESTABLISHED;
1200 smp_mb__after_atomic_inc(); /* sock_hold() does an atomic_inc() */
1201 unix_peer(sk) = newsk;
1203 unix_state_unlock(sk);
1205 /* take ten and and send info to listening sock */
1206 spin_lock(&other->sk_receive_queue.lock);
1207 __skb_queue_tail(&other->sk_receive_queue, skb);
1208 spin_unlock(&other->sk_receive_queue.lock);
1209 unix_state_unlock(other);
1210 other->sk_data_ready(other, 0);
1216 unix_state_unlock(other);
1221 unix_release_sock(newsk, 0);
1227 static int unix_socketpair(struct socket *socka, struct socket *sockb)
1229 struct sock *ska = socka->sk, *skb = sockb->sk;
1231 /* Join our sockets back to back */
1234 unix_peer(ska) = skb;
1235 unix_peer(skb) = ska;
1239 if (ska->sk_type != SOCK_DGRAM) {
1240 ska->sk_state = TCP_ESTABLISHED;
1241 skb->sk_state = TCP_ESTABLISHED;
1242 socka->state = SS_CONNECTED;
1243 sockb->state = SS_CONNECTED;
1248 static void unix_sock_inherit_flags(const struct socket *old,
1251 if (test_bit(SOCK_PASSCRED, &old->flags))
1252 set_bit(SOCK_PASSCRED, &new->flags);
1253 if (test_bit(SOCK_PASSSEC, &old->flags))
1254 set_bit(SOCK_PASSSEC, &new->flags);
1257 static int unix_accept(struct socket *sock, struct socket *newsock, int flags)
1259 struct sock *sk = sock->sk;
1261 struct sk_buff *skb;
1265 if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
1269 if (sk->sk_state != TCP_LISTEN)
1272 /* If socket state is TCP_LISTEN it cannot change (for now...),
1273 * so that no locks are necessary.
1276 skb = skb_recv_datagram(sk, 0, flags&O_NONBLOCK, &err);
1278 /* This means receive shutdown. */
1285 skb_free_datagram(sk, skb);
1286 wake_up_interruptible(&unix_sk(sk)->peer_wait);
1288 /* attach accepted sock to socket */
1289 unix_state_lock(tsk);
1290 newsock->state = SS_CONNECTED;
1291 unix_sock_inherit_flags(sock, newsock);
1292 sock_graft(tsk, newsock);
1293 unix_state_unlock(tsk);
1301 static int unix_getname(struct socket *sock, struct sockaddr *uaddr, int *uaddr_len, int peer)
1303 struct sock *sk = sock->sk;
1304 struct unix_sock *u;
1305 DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr, uaddr);
1309 sk = unix_peer_get(sk);
1320 unix_state_lock(sk);
1322 sunaddr->sun_family = AF_UNIX;
1323 sunaddr->sun_path[0] = 0;
1324 *uaddr_len = sizeof(short);
1326 struct unix_address *addr = u->addr;
1328 *uaddr_len = addr->len;
1329 memcpy(sunaddr, addr->name, *uaddr_len);
1331 unix_state_unlock(sk);
1337 static void unix_detach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1341 scm->fp = UNIXCB(skb).fp;
1342 UNIXCB(skb).fp = NULL;
1344 for (i = scm->fp->count-1; i >= 0; i--)
1345 unix_notinflight(scm->fp->fp[i]);
1348 static void unix_destruct_scm(struct sk_buff *skb)
1350 struct scm_cookie scm;
1351 memset(&scm, 0, sizeof(scm));
1352 scm.pid = UNIXCB(skb).pid;
1354 unix_detach_fds(&scm, skb);
1356 /* Alas, it calls VFS */
1357 /* So fscking what? fput() had been SMP-safe since the last Summer */
1362 #define MAX_RECURSION_LEVEL 4
1364 static int unix_attach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1367 unsigned char max_level = 0;
1368 int unix_sock_count = 0;
1370 for (i = scm->fp->count - 1; i >= 0; i--) {
1371 struct sock *sk = unix_get_socket(scm->fp->fp[i]);
1375 max_level = max(max_level,
1376 unix_sk(sk)->recursion_level);
1379 if (unlikely(max_level > MAX_RECURSION_LEVEL))
1380 return -ETOOMANYREFS;
1383 * Need to duplicate file references for the sake of garbage
1384 * collection. Otherwise a socket in the fps might become a
1385 * candidate for GC while the skb is not yet queued.
1387 UNIXCB(skb).fp = scm_fp_dup(scm->fp);
1388 if (!UNIXCB(skb).fp)
1391 if (unix_sock_count) {
1392 for (i = scm->fp->count - 1; i >= 0; i--)
1393 unix_inflight(scm->fp->fp[i]);
1398 static int unix_scm_to_skb(struct scm_cookie *scm, struct sk_buff *skb, bool send_fds)
1402 UNIXCB(skb).pid = get_pid(scm->pid);
1403 UNIXCB(skb).uid = scm->creds.uid;
1404 UNIXCB(skb).gid = scm->creds.gid;
1405 UNIXCB(skb).fp = NULL;
1406 if (scm->fp && send_fds)
1407 err = unix_attach_fds(scm, skb);
1409 skb->destructor = unix_destruct_scm;
1414 * Some apps rely on write() giving SCM_CREDENTIALS
1415 * We include credentials if source or destination socket
1416 * asserted SOCK_PASSCRED.
1418 static void maybe_add_creds(struct sk_buff *skb, const struct socket *sock,
1419 const struct sock *other)
1421 if (UNIXCB(skb).pid)
1423 if (test_bit(SOCK_PASSCRED, &sock->flags) ||
1424 !other->sk_socket ||
1425 test_bit(SOCK_PASSCRED, &other->sk_socket->flags)) {
1426 UNIXCB(skb).pid = get_pid(task_tgid(current));
1427 current_uid_gid(&UNIXCB(skb).uid, &UNIXCB(skb).gid);
1432 * Send AF_UNIX data.
1435 static int unix_dgram_sendmsg(struct kiocb *kiocb, struct socket *sock,
1436 struct msghdr *msg, size_t len)
1438 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1439 struct sock *sk = sock->sk;
1440 struct net *net = sock_net(sk);
1441 struct unix_sock *u = unix_sk(sk);
1442 struct sockaddr_un *sunaddr = msg->msg_name;
1443 struct sock *other = NULL;
1444 int namelen = 0; /* fake GCC */
1447 struct sk_buff *skb;
1449 struct scm_cookie tmp_scm;
1453 if (NULL == siocb->scm)
1454 siocb->scm = &tmp_scm;
1456 err = scm_send(sock, msg, siocb->scm, false);
1461 if (msg->msg_flags&MSG_OOB)
1464 if (msg->msg_namelen) {
1465 err = unix_mkname(sunaddr, msg->msg_namelen, &hash);
1472 other = unix_peer_get(sk);
1477 if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr
1478 && (err = unix_autobind(sock)) != 0)
1482 if (len > sk->sk_sndbuf - 32)
1485 if (len > SKB_MAX_ALLOC)
1486 data_len = min_t(size_t,
1487 len - SKB_MAX_ALLOC,
1488 MAX_SKB_FRAGS * PAGE_SIZE);
1490 skb = sock_alloc_send_pskb(sk, len - data_len, data_len,
1491 msg->msg_flags & MSG_DONTWAIT, &err);
1495 err = unix_scm_to_skb(siocb->scm, skb, true);
1498 max_level = err + 1;
1499 unix_get_secdata(siocb->scm, skb);
1501 skb_put(skb, len - data_len);
1502 skb->data_len = data_len;
1504 err = skb_copy_datagram_from_iovec(skb, 0, msg->msg_iov, 0, len);
1508 timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1513 if (sunaddr == NULL)
1516 other = unix_find_other(net, sunaddr, namelen, sk->sk_type,
1522 if (sk_filter(other, skb) < 0) {
1523 /* Toss the packet but do not return any error to the sender */
1528 unix_state_lock(other);
1530 if (!unix_may_send(sk, other))
1533 if (sock_flag(other, SOCK_DEAD)) {
1535 * Check with 1003.1g - what should
1538 unix_state_unlock(other);
1542 unix_state_lock(sk);
1543 if (unix_peer(sk) == other) {
1544 unix_peer(sk) = NULL;
1545 unix_state_unlock(sk);
1547 unix_dgram_disconnected(sk, other);
1549 err = -ECONNREFUSED;
1551 unix_state_unlock(sk);
1561 if (other->sk_shutdown & RCV_SHUTDOWN)
1564 if (sk->sk_type != SOCK_SEQPACKET) {
1565 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1570 if (unix_peer(other) != sk && unix_recvq_full(other)) {
1576 timeo = unix_wait_for_peer(other, timeo);
1578 err = sock_intr_errno(timeo);
1579 if (signal_pending(current))
1585 if (sock_flag(other, SOCK_RCVTSTAMP))
1586 __net_timestamp(skb);
1587 maybe_add_creds(skb, sock, other);
1588 skb_queue_tail(&other->sk_receive_queue, skb);
1589 if (max_level > unix_sk(other)->recursion_level)
1590 unix_sk(other)->recursion_level = max_level;
1591 unix_state_unlock(other);
1592 other->sk_data_ready(other, len);
1594 scm_destroy(siocb->scm);
1598 unix_state_unlock(other);
1604 scm_destroy(siocb->scm);
1609 static int unix_stream_sendmsg(struct kiocb *kiocb, struct socket *sock,
1610 struct msghdr *msg, size_t len)
1612 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1613 struct sock *sk = sock->sk;
1614 struct sock *other = NULL;
1616 struct sk_buff *skb;
1618 struct scm_cookie tmp_scm;
1619 bool fds_sent = false;
1622 if (NULL == siocb->scm)
1623 siocb->scm = &tmp_scm;
1625 err = scm_send(sock, msg, siocb->scm, false);
1630 if (msg->msg_flags&MSG_OOB)
1633 if (msg->msg_namelen) {
1634 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
1638 other = unix_peer(sk);
1643 if (sk->sk_shutdown & SEND_SHUTDOWN)
1646 while (sent < len) {
1648 * Optimisation for the fact that under 0.01% of X
1649 * messages typically need breaking up.
1654 /* Keep two messages in the pipe so it schedules better */
1655 if (size > ((sk->sk_sndbuf >> 1) - 64))
1656 size = (sk->sk_sndbuf >> 1) - 64;
1658 if (size > SKB_MAX_ALLOC)
1659 size = SKB_MAX_ALLOC;
1665 skb = sock_alloc_send_skb(sk, size, msg->msg_flags&MSG_DONTWAIT,
1672 * If you pass two values to the sock_alloc_send_skb
1673 * it tries to grab the large buffer with GFP_NOFS
1674 * (which can fail easily), and if it fails grab the
1675 * fallback size buffer which is under a page and will
1678 size = min_t(int, size, skb_tailroom(skb));
1681 /* Only send the fds in the first buffer */
1682 err = unix_scm_to_skb(siocb->scm, skb, !fds_sent);
1687 max_level = err + 1;
1690 err = memcpy_fromiovec(skb_put(skb, size), msg->msg_iov, size);
1696 unix_state_lock(other);
1698 if (sock_flag(other, SOCK_DEAD) ||
1699 (other->sk_shutdown & RCV_SHUTDOWN))
1702 maybe_add_creds(skb, sock, other);
1703 skb_queue_tail(&other->sk_receive_queue, skb);
1704 if (max_level > unix_sk(other)->recursion_level)
1705 unix_sk(other)->recursion_level = max_level;
1706 unix_state_unlock(other);
1707 other->sk_data_ready(other, size);
1711 scm_destroy(siocb->scm);
1717 unix_state_unlock(other);
1720 if (sent == 0 && !(msg->msg_flags&MSG_NOSIGNAL))
1721 send_sig(SIGPIPE, current, 0);
1724 scm_destroy(siocb->scm);
1726 return sent ? : err;
1729 static int unix_seqpacket_sendmsg(struct kiocb *kiocb, struct socket *sock,
1730 struct msghdr *msg, size_t len)
1733 struct sock *sk = sock->sk;
1735 err = sock_error(sk);
1739 if (sk->sk_state != TCP_ESTABLISHED)
1742 if (msg->msg_namelen)
1743 msg->msg_namelen = 0;
1745 return unix_dgram_sendmsg(kiocb, sock, msg, len);
1748 static int unix_seqpacket_recvmsg(struct kiocb *iocb, struct socket *sock,
1749 struct msghdr *msg, size_t size,
1752 struct sock *sk = sock->sk;
1754 if (sk->sk_state != TCP_ESTABLISHED)
1757 return unix_dgram_recvmsg(iocb, sock, msg, size, flags);
1760 static void unix_copy_addr(struct msghdr *msg, struct sock *sk)
1762 struct unix_sock *u = unix_sk(sk);
1765 msg->msg_namelen = u->addr->len;
1766 memcpy(msg->msg_name, u->addr->name, u->addr->len);
1770 static int unix_dgram_recvmsg(struct kiocb *iocb, struct socket *sock,
1771 struct msghdr *msg, size_t size,
1774 struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1775 struct scm_cookie tmp_scm;
1776 struct sock *sk = sock->sk;
1777 struct unix_sock *u = unix_sk(sk);
1778 int noblock = flags & MSG_DONTWAIT;
1779 struct sk_buff *skb;
1787 err = mutex_lock_interruptible(&u->readlock);
1789 err = sock_intr_errno(sock_rcvtimeo(sk, noblock));
1793 skip = sk_peek_offset(sk, flags);
1795 skb = __skb_recv_datagram(sk, flags, &peeked, &skip, &err);
1797 unix_state_lock(sk);
1798 /* Signal EOF on disconnected non-blocking SEQPACKET socket. */
1799 if (sk->sk_type == SOCK_SEQPACKET && err == -EAGAIN &&
1800 (sk->sk_shutdown & RCV_SHUTDOWN))
1802 unix_state_unlock(sk);
1806 wake_up_interruptible_sync_poll(&u->peer_wait,
1807 POLLOUT | POLLWRNORM | POLLWRBAND);
1810 unix_copy_addr(msg, skb->sk);
1812 if (size > skb->len - skip)
1813 size = skb->len - skip;
1814 else if (size < skb->len - skip)
1815 msg->msg_flags |= MSG_TRUNC;
1817 err = skb_copy_datagram_iovec(skb, skip, msg->msg_iov, size);
1821 if (sock_flag(sk, SOCK_RCVTSTAMP))
1822 __sock_recv_timestamp(msg, sk, skb);
1825 siocb->scm = &tmp_scm;
1826 memset(&tmp_scm, 0, sizeof(tmp_scm));
1828 scm_set_cred(siocb->scm, UNIXCB(skb).pid, UNIXCB(skb).uid, UNIXCB(skb).gid);
1829 unix_set_secdata(siocb->scm, skb);
1831 if (!(flags & MSG_PEEK)) {
1833 unix_detach_fds(siocb->scm, skb);
1835 sk_peek_offset_bwd(sk, skb->len);
1837 /* It is questionable: on PEEK we could:
1838 - do not return fds - good, but too simple 8)
1839 - return fds, and do not return them on read (old strategy,
1841 - clone fds (I chose it for now, it is the most universal
1844 POSIX 1003.1g does not actually define this clearly
1845 at all. POSIX 1003.1g doesn't define a lot of things
1850 sk_peek_offset_fwd(sk, size);
1853 siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1855 err = (flags & MSG_TRUNC) ? skb->len - skip : size;
1857 scm_recv(sock, msg, siocb->scm, flags);
1860 skb_free_datagram(sk, skb);
1862 mutex_unlock(&u->readlock);
1868 * Sleep until more data has arrived. But check for races..
1870 static long unix_stream_data_wait(struct sock *sk, long timeo,
1871 struct sk_buff *last)
1875 unix_state_lock(sk);
1878 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1880 if (skb_peek_tail(&sk->sk_receive_queue) != last ||
1882 (sk->sk_shutdown & RCV_SHUTDOWN) ||
1883 signal_pending(current) ||
1887 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1888 unix_state_unlock(sk);
1889 timeo = schedule_timeout(timeo);
1890 unix_state_lock(sk);
1891 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1894 finish_wait(sk_sleep(sk), &wait);
1895 unix_state_unlock(sk);
1899 static int unix_stream_recvmsg(struct kiocb *iocb, struct socket *sock,
1900 struct msghdr *msg, size_t size,
1903 struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1904 struct scm_cookie tmp_scm;
1905 struct sock *sk = sock->sk;
1906 struct unix_sock *u = unix_sk(sk);
1907 struct sockaddr_un *sunaddr = msg->msg_name;
1909 int check_creds = 0;
1916 if (sk->sk_state != TCP_ESTABLISHED)
1923 target = sock_rcvlowat(sk, flags&MSG_WAITALL, size);
1924 timeo = sock_rcvtimeo(sk, flags&MSG_DONTWAIT);
1926 /* Lock the socket to prevent queue disordering
1927 * while sleeps in memcpy_tomsg
1931 siocb->scm = &tmp_scm;
1932 memset(&tmp_scm, 0, sizeof(tmp_scm));
1935 err = mutex_lock_interruptible(&u->readlock);
1937 err = sock_intr_errno(timeo);
1943 struct sk_buff *skb, *last;
1945 unix_state_lock(sk);
1946 last = skb = skb_peek(&sk->sk_receive_queue);
1949 unix_sk(sk)->recursion_level = 0;
1950 if (copied >= target)
1954 * POSIX 1003.1g mandates this order.
1957 err = sock_error(sk);
1960 if (sk->sk_shutdown & RCV_SHUTDOWN)
1963 unix_state_unlock(sk);
1967 mutex_unlock(&u->readlock);
1969 timeo = unix_stream_data_wait(sk, timeo, last);
1971 if (signal_pending(current)
1972 || mutex_lock_interruptible(&u->readlock)) {
1973 err = sock_intr_errno(timeo);
1979 unix_state_unlock(sk);
1983 skip = sk_peek_offset(sk, flags);
1984 while (skip >= skb->len) {
1987 skb = skb_peek_next(skb, &sk->sk_receive_queue);
1992 unix_state_unlock(sk);
1995 /* Never glue messages from different writers */
1996 if ((UNIXCB(skb).pid != siocb->scm->pid) ||
1997 !uid_eq(UNIXCB(skb).uid, siocb->scm->creds.uid) ||
1998 !gid_eq(UNIXCB(skb).gid, siocb->scm->creds.gid))
2000 } else if (test_bit(SOCK_PASSCRED, &sock->flags)) {
2001 /* Copy credentials */
2002 scm_set_cred(siocb->scm, UNIXCB(skb).pid, UNIXCB(skb).uid, UNIXCB(skb).gid);
2006 /* Copy address just once */
2008 unix_copy_addr(msg, skb->sk);
2012 chunk = min_t(unsigned int, skb->len - skip, size);
2013 if (memcpy_toiovec(msg->msg_iov, skb->data + skip, chunk)) {
2021 /* Mark read part of skb as used */
2022 if (!(flags & MSG_PEEK)) {
2023 skb_pull(skb, chunk);
2025 sk_peek_offset_bwd(sk, chunk);
2028 unix_detach_fds(siocb->scm, skb);
2033 skb_unlink(skb, &sk->sk_receive_queue);
2039 /* It is questionable, see note in unix_dgram_recvmsg.
2042 siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
2044 sk_peek_offset_fwd(sk, chunk);
2050 mutex_unlock(&u->readlock);
2051 scm_recv(sock, msg, siocb->scm, flags);
2053 return copied ? : err;
2056 static int unix_shutdown(struct socket *sock, int mode)
2058 struct sock *sk = sock->sk;
2061 if (mode < SHUT_RD || mode > SHUT_RDWR)
2064 * SHUT_RD (0) -> RCV_SHUTDOWN (1)
2065 * SHUT_WR (1) -> SEND_SHUTDOWN (2)
2066 * SHUT_RDWR (2) -> SHUTDOWN_MASK (3)
2070 unix_state_lock(sk);
2071 sk->sk_shutdown |= mode;
2072 other = unix_peer(sk);
2075 unix_state_unlock(sk);
2076 sk->sk_state_change(sk);
2079 (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET)) {
2083 if (mode&RCV_SHUTDOWN)
2084 peer_mode |= SEND_SHUTDOWN;
2085 if (mode&SEND_SHUTDOWN)
2086 peer_mode |= RCV_SHUTDOWN;
2087 unix_state_lock(other);
2088 other->sk_shutdown |= peer_mode;
2089 unix_state_unlock(other);
2090 other->sk_state_change(other);
2091 if (peer_mode == SHUTDOWN_MASK)
2092 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_HUP);
2093 else if (peer_mode & RCV_SHUTDOWN)
2094 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_IN);
2102 long unix_inq_len(struct sock *sk)
2104 struct sk_buff *skb;
2107 if (sk->sk_state == TCP_LISTEN)
2110 spin_lock(&sk->sk_receive_queue.lock);
2111 if (sk->sk_type == SOCK_STREAM ||
2112 sk->sk_type == SOCK_SEQPACKET) {
2113 skb_queue_walk(&sk->sk_receive_queue, skb)
2116 skb = skb_peek(&sk->sk_receive_queue);
2120 spin_unlock(&sk->sk_receive_queue.lock);
2124 EXPORT_SYMBOL_GPL(unix_inq_len);
2126 long unix_outq_len(struct sock *sk)
2128 return sk_wmem_alloc_get(sk);
2130 EXPORT_SYMBOL_GPL(unix_outq_len);
2132 static int unix_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2134 struct sock *sk = sock->sk;
2140 amount = unix_outq_len(sk);
2141 err = put_user(amount, (int __user *)arg);
2144 amount = unix_inq_len(sk);
2148 err = put_user(amount, (int __user *)arg);
2157 static unsigned int unix_poll(struct file *file, struct socket *sock, poll_table *wait)
2159 struct sock *sk = sock->sk;
2162 sock_poll_wait(file, sk_sleep(sk), wait);
2165 /* exceptional events? */
2168 if (sk->sk_shutdown == SHUTDOWN_MASK)
2170 if (sk->sk_shutdown & RCV_SHUTDOWN)
2171 mask |= POLLRDHUP | POLLIN | POLLRDNORM;
2174 if (!skb_queue_empty(&sk->sk_receive_queue))
2175 mask |= POLLIN | POLLRDNORM;
2177 /* Connection-based need to check for termination and startup */
2178 if ((sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) &&
2179 sk->sk_state == TCP_CLOSE)
2183 * we set writable also when the other side has shut down the
2184 * connection. This prevents stuck sockets.
2186 if (unix_writable(sk))
2187 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2192 static unsigned int unix_dgram_poll(struct file *file, struct socket *sock,
2195 struct sock *sk = sock->sk, *other;
2196 unsigned int mask, writable;
2198 sock_poll_wait(file, sk_sleep(sk), wait);
2201 /* exceptional events? */
2202 if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue))
2204 (sock_flag(sk, SOCK_SELECT_ERR_QUEUE) ? POLLPRI : 0);
2206 if (sk->sk_shutdown & RCV_SHUTDOWN)
2207 mask |= POLLRDHUP | POLLIN | POLLRDNORM;
2208 if (sk->sk_shutdown == SHUTDOWN_MASK)
2212 if (!skb_queue_empty(&sk->sk_receive_queue))
2213 mask |= POLLIN | POLLRDNORM;
2215 /* Connection-based need to check for termination and startup */
2216 if (sk->sk_type == SOCK_SEQPACKET) {
2217 if (sk->sk_state == TCP_CLOSE)
2219 /* connection hasn't started yet? */
2220 if (sk->sk_state == TCP_SYN_SENT)
2224 /* No write status requested, avoid expensive OUT tests. */
2225 if (!(poll_requested_events(wait) & (POLLWRBAND|POLLWRNORM|POLLOUT)))
2228 writable = unix_writable(sk);
2229 other = unix_peer_get(sk);
2231 if (unix_peer(other) != sk) {
2232 sock_poll_wait(file, &unix_sk(other)->peer_wait, wait);
2233 if (unix_recvq_full(other))
2240 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2242 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
2247 #ifdef CONFIG_PROC_FS
2249 #define BUCKET_SPACE (BITS_PER_LONG - (UNIX_HASH_BITS + 1) - 1)
2251 #define get_bucket(x) ((x) >> BUCKET_SPACE)
2252 #define get_offset(x) ((x) & ((1L << BUCKET_SPACE) - 1))
2253 #define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
2255 static struct sock *unix_from_bucket(struct seq_file *seq, loff_t *pos)
2257 unsigned long offset = get_offset(*pos);
2258 unsigned long bucket = get_bucket(*pos);
2260 unsigned long count = 0;
2262 for (sk = sk_head(&unix_socket_table[bucket]); sk; sk = sk_next(sk)) {
2263 if (sock_net(sk) != seq_file_net(seq))
2265 if (++count == offset)
2272 static struct sock *unix_next_socket(struct seq_file *seq,
2276 unsigned long bucket;
2278 while (sk > (struct sock *)SEQ_START_TOKEN) {
2282 if (sock_net(sk) == seq_file_net(seq))
2287 sk = unix_from_bucket(seq, pos);
2292 bucket = get_bucket(*pos) + 1;
2293 *pos = set_bucket_offset(bucket, 1);
2294 } while (bucket < ARRAY_SIZE(unix_socket_table));
2299 static void *unix_seq_start(struct seq_file *seq, loff_t *pos)
2300 __acquires(unix_table_lock)
2302 spin_lock(&unix_table_lock);
2305 return SEQ_START_TOKEN;
2307 if (get_bucket(*pos) >= ARRAY_SIZE(unix_socket_table))
2310 return unix_next_socket(seq, NULL, pos);
2313 static void *unix_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2316 return unix_next_socket(seq, v, pos);
2319 static void unix_seq_stop(struct seq_file *seq, void *v)
2320 __releases(unix_table_lock)
2322 spin_unlock(&unix_table_lock);
2325 static int unix_seq_show(struct seq_file *seq, void *v)
2328 if (v == SEQ_START_TOKEN)
2329 seq_puts(seq, "Num RefCount Protocol Flags Type St "
2333 struct unix_sock *u = unix_sk(s);
2336 seq_printf(seq, "%pK: %08X %08X %08X %04X %02X %5lu",
2338 atomic_read(&s->sk_refcnt),
2340 s->sk_state == TCP_LISTEN ? __SO_ACCEPTCON : 0,
2343 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTED : SS_UNCONNECTED) :
2344 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTING : SS_DISCONNECTING),
2352 len = u->addr->len - sizeof(short);
2353 if (!UNIX_ABSTRACT(s))
2359 for ( ; i < len; i++)
2360 seq_putc(seq, u->addr->name->sun_path[i]);
2362 unix_state_unlock(s);
2363 seq_putc(seq, '\n');
2369 static const struct seq_operations unix_seq_ops = {
2370 .start = unix_seq_start,
2371 .next = unix_seq_next,
2372 .stop = unix_seq_stop,
2373 .show = unix_seq_show,
2376 static int unix_seq_open(struct inode *inode, struct file *file)
2378 return seq_open_net(inode, file, &unix_seq_ops,
2379 sizeof(struct seq_net_private));
2382 static const struct file_operations unix_seq_fops = {
2383 .owner = THIS_MODULE,
2384 .open = unix_seq_open,
2386 .llseek = seq_lseek,
2387 .release = seq_release_net,
2392 static const struct net_proto_family unix_family_ops = {
2394 .create = unix_create,
2395 .owner = THIS_MODULE,
2399 static int __net_init unix_net_init(struct net *net)
2401 int error = -ENOMEM;
2403 net->unx.sysctl_max_dgram_qlen = 10;
2404 if (unix_sysctl_register(net))
2407 #ifdef CONFIG_PROC_FS
2408 if (!proc_create("unix", 0, net->proc_net, &unix_seq_fops)) {
2409 unix_sysctl_unregister(net);
2418 static void __net_exit unix_net_exit(struct net *net)
2420 unix_sysctl_unregister(net);
2421 remove_proc_entry("unix", net->proc_net);
2424 static struct pernet_operations unix_net_ops = {
2425 .init = unix_net_init,
2426 .exit = unix_net_exit,
2429 static int __init af_unix_init(void)
2433 BUILD_BUG_ON(sizeof(struct unix_skb_parms) > FIELD_SIZEOF(struct sk_buff, cb));
2435 rc = proto_register(&unix_proto, 1);
2437 printk(KERN_CRIT "%s: Cannot create unix_sock SLAB cache!\n",
2442 sock_register(&unix_family_ops);
2443 register_pernet_subsys(&unix_net_ops);
2448 static void __exit af_unix_exit(void)
2450 sock_unregister(PF_UNIX);
2451 proto_unregister(&unix_proto);
2452 unregister_pernet_subsys(&unix_net_ops);
2455 /* Earlier than device_initcall() so that other drivers invoking
2456 request_module() don't end up in a loop when modprobe tries
2457 to use a UNIX socket. But later than subsys_initcall() because
2458 we depend on stuff initialised there */
2459 fs_initcall(af_unix_init);
2460 module_exit(af_unix_exit);
2462 MODULE_LICENSE("GPL");
2463 MODULE_ALIAS_NETPROTO(PF_UNIX);