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 static struct hlist_head unix_socket_table[UNIX_HASH_SIZE + 1];
119 static DEFINE_SPINLOCK(unix_table_lock);
120 static atomic_t unix_nr_socks = ATOMIC_INIT(0);
122 #define unix_sockets_unbound (&unix_socket_table[UNIX_HASH_SIZE])
124 #define UNIX_ABSTRACT(sk) (unix_sk(sk)->addr->hash != UNIX_HASH_SIZE)
126 #ifdef CONFIG_SECURITY_NETWORK
127 static void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb)
129 memcpy(UNIXSID(skb), &scm->secid, sizeof(u32));
132 static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb)
134 scm->secid = *UNIXSID(skb);
137 static inline void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb)
140 static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb)
142 #endif /* CONFIG_SECURITY_NETWORK */
145 * SMP locking strategy:
146 * hash table is protected with spinlock unix_table_lock
147 * each socket state is protected by separate rwlock.
150 static inline unsigned unix_hash_fold(__wsum n)
152 unsigned hash = (__force unsigned)n;
155 return hash&(UNIX_HASH_SIZE-1);
158 #define unix_peer(sk) (unix_sk(sk)->peer)
160 static inline int unix_our_peer(struct sock *sk, struct sock *osk)
162 return unix_peer(osk) == sk;
165 static inline int unix_may_send(struct sock *sk, struct sock *osk)
167 return unix_peer(osk) == NULL || unix_our_peer(sk, osk);
170 static inline int unix_recvq_full(struct sock const *sk)
172 return skb_queue_len(&sk->sk_receive_queue) > sk->sk_max_ack_backlog;
175 static struct sock *unix_peer_get(struct sock *s)
183 unix_state_unlock(s);
187 static inline void unix_release_addr(struct unix_address *addr)
189 if (atomic_dec_and_test(&addr->refcnt))
194 * Check unix socket name:
195 * - should be not zero length.
196 * - if started by not zero, should be NULL terminated (FS object)
197 * - if started by zero, it is abstract name.
200 static int unix_mkname(struct sockaddr_un *sunaddr, int len, unsigned *hashp)
202 if (len <= sizeof(short) || len > sizeof(*sunaddr))
204 if (!sunaddr || sunaddr->sun_family != AF_UNIX)
206 if (sunaddr->sun_path[0]) {
208 * This may look like an off by one error but it is a bit more
209 * subtle. 108 is the longest valid AF_UNIX path for a binding.
210 * sun_path[108] doesnt as such exist. However in kernel space
211 * we are guaranteed that it is a valid memory location in our
212 * kernel address buffer.
214 ((char *)sunaddr)[len] = 0;
215 len = strlen(sunaddr->sun_path)+1+sizeof(short);
219 *hashp = unix_hash_fold(csum_partial(sunaddr, len, 0));
223 static void __unix_remove_socket(struct sock *sk)
225 sk_del_node_init(sk);
228 static void __unix_insert_socket(struct hlist_head *list, struct sock *sk)
230 WARN_ON(!sk_unhashed(sk));
231 sk_add_node(sk, list);
234 static inline void unix_remove_socket(struct sock *sk)
236 spin_lock(&unix_table_lock);
237 __unix_remove_socket(sk);
238 spin_unlock(&unix_table_lock);
241 static inline void unix_insert_socket(struct hlist_head *list, struct sock *sk)
243 spin_lock(&unix_table_lock);
244 __unix_insert_socket(list, sk);
245 spin_unlock(&unix_table_lock);
248 static struct sock *__unix_find_socket_byname(struct net *net,
249 struct sockaddr_un *sunname,
250 int len, int type, unsigned hash)
253 struct hlist_node *node;
255 sk_for_each(s, node, &unix_socket_table[hash ^ type]) {
256 struct unix_sock *u = unix_sk(s);
258 if (!net_eq(sock_net(s), net))
261 if (u->addr->len == len &&
262 !memcmp(u->addr->name, sunname, len))
270 static inline struct sock *unix_find_socket_byname(struct net *net,
271 struct sockaddr_un *sunname,
277 spin_lock(&unix_table_lock);
278 s = __unix_find_socket_byname(net, sunname, len, type, hash);
281 spin_unlock(&unix_table_lock);
285 static struct sock *unix_find_socket_byinode(struct net *net, struct inode *i)
288 struct hlist_node *node;
290 spin_lock(&unix_table_lock);
292 &unix_socket_table[i->i_ino & (UNIX_HASH_SIZE - 1)]) {
293 struct dentry *dentry = unix_sk(s)->dentry;
295 if (!net_eq(sock_net(s), net))
298 if (dentry && dentry->d_inode == i) {
305 spin_unlock(&unix_table_lock);
309 static inline int unix_writable(struct sock *sk)
311 return (atomic_read(&sk->sk_wmem_alloc) << 2) <= sk->sk_sndbuf;
314 static void unix_write_space(struct sock *sk)
316 read_lock(&sk->sk_callback_lock);
317 if (unix_writable(sk)) {
318 if (sk_has_sleeper(sk))
319 wake_up_interruptible_sync(sk->sk_sleep);
320 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
322 read_unlock(&sk->sk_callback_lock);
325 /* When dgram socket disconnects (or changes its peer), we clear its receive
326 * queue of packets arrived from previous peer. First, it allows to do
327 * flow control based only on wmem_alloc; second, sk connected to peer
328 * may receive messages only from that peer. */
329 static void unix_dgram_disconnected(struct sock *sk, struct sock *other)
331 if (!skb_queue_empty(&sk->sk_receive_queue)) {
332 skb_queue_purge(&sk->sk_receive_queue);
333 wake_up_interruptible_all(&unix_sk(sk)->peer_wait);
335 /* If one link of bidirectional dgram pipe is disconnected,
336 * we signal error. Messages are lost. Do not make this,
337 * when peer was not connected to us.
339 if (!sock_flag(other, SOCK_DEAD) && unix_peer(other) == sk) {
340 other->sk_err = ECONNRESET;
341 other->sk_error_report(other);
346 static void unix_sock_destructor(struct sock *sk)
348 struct unix_sock *u = unix_sk(sk);
350 skb_queue_purge(&sk->sk_receive_queue);
352 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
353 WARN_ON(!sk_unhashed(sk));
354 WARN_ON(sk->sk_socket);
355 if (!sock_flag(sk, SOCK_DEAD)) {
356 printk(KERN_INFO "Attempt to release alive unix socket: %p\n", sk);
361 unix_release_addr(u->addr);
363 atomic_dec(&unix_nr_socks);
365 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
367 #ifdef UNIX_REFCNT_DEBUG
368 printk(KERN_DEBUG "UNIX %p is destroyed, %d are still alive.\n", sk,
369 atomic_read(&unix_nr_socks));
373 static int unix_release_sock(struct sock *sk, int embrion)
375 struct unix_sock *u = unix_sk(sk);
376 struct dentry *dentry;
377 struct vfsmount *mnt;
382 unix_remove_socket(sk);
387 sk->sk_shutdown = SHUTDOWN_MASK;
392 state = sk->sk_state;
393 sk->sk_state = TCP_CLOSE;
394 unix_state_unlock(sk);
396 wake_up_interruptible_all(&u->peer_wait);
398 skpair = unix_peer(sk);
400 if (skpair != NULL) {
401 if (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) {
402 unix_state_lock(skpair);
404 skpair->sk_shutdown = SHUTDOWN_MASK;
405 if (!skb_queue_empty(&sk->sk_receive_queue) || embrion)
406 skpair->sk_err = ECONNRESET;
407 unix_state_unlock(skpair);
408 skpair->sk_state_change(skpair);
409 read_lock(&skpair->sk_callback_lock);
410 sk_wake_async(skpair, SOCK_WAKE_WAITD, POLL_HUP);
411 read_unlock(&skpair->sk_callback_lock);
413 sock_put(skpair); /* It may now die */
414 unix_peer(sk) = NULL;
417 /* Try to flush out this socket. Throw out buffers at least */
419 while ((skb = skb_dequeue(&sk->sk_receive_queue)) != NULL) {
420 if (state == TCP_LISTEN)
421 unix_release_sock(skb->sk, 1);
422 /* passed fds are erased in the kfree_skb hook */
433 /* ---- Socket is dead now and most probably destroyed ---- */
436 * Fixme: BSD difference: In BSD all sockets connected to use get
437 * ECONNRESET and we die on the spot. In Linux we behave
438 * like files and pipes do and wait for the last
441 * Can't we simply set sock->err?
443 * What the above comment does talk about? --ANK(980817)
446 if (unix_tot_inflight)
447 unix_gc(); /* Garbage collect fds */
452 static int unix_listen(struct socket *sock, int backlog)
455 struct sock *sk = sock->sk;
456 struct unix_sock *u = unix_sk(sk);
459 if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
460 goto out; /* Only stream/seqpacket sockets accept */
463 goto out; /* No listens on an unbound socket */
465 if (sk->sk_state != TCP_CLOSE && sk->sk_state != TCP_LISTEN)
467 if (backlog > sk->sk_max_ack_backlog)
468 wake_up_interruptible_all(&u->peer_wait);
469 sk->sk_max_ack_backlog = backlog;
470 sk->sk_state = TCP_LISTEN;
471 /* set credentials so connect can copy them */
472 sk->sk_peercred.pid = task_tgid_vnr(current);
473 current_euid_egid(&sk->sk_peercred.uid, &sk->sk_peercred.gid);
477 unix_state_unlock(sk);
482 static int unix_release(struct socket *);
483 static int unix_bind(struct socket *, struct sockaddr *, int);
484 static int unix_stream_connect(struct socket *, struct sockaddr *,
485 int addr_len, int flags);
486 static int unix_socketpair(struct socket *, struct socket *);
487 static int unix_accept(struct socket *, struct socket *, int);
488 static int unix_getname(struct socket *, struct sockaddr *, int *, int);
489 static unsigned int unix_poll(struct file *, struct socket *, poll_table *);
490 static unsigned int unix_dgram_poll(struct file *, struct socket *,
492 static int unix_ioctl(struct socket *, unsigned int, unsigned long);
493 static int unix_shutdown(struct socket *, int);
494 static int unix_stream_sendmsg(struct kiocb *, struct socket *,
495 struct msghdr *, size_t);
496 static int unix_stream_recvmsg(struct kiocb *, struct socket *,
497 struct msghdr *, size_t, int);
498 static int unix_dgram_sendmsg(struct kiocb *, struct socket *,
499 struct msghdr *, size_t);
500 static int unix_dgram_recvmsg(struct kiocb *, struct socket *,
501 struct msghdr *, size_t, int);
502 static int unix_dgram_connect(struct socket *, struct sockaddr *,
504 static int unix_seqpacket_sendmsg(struct kiocb *, struct socket *,
505 struct msghdr *, size_t);
507 static const struct proto_ops unix_stream_ops = {
509 .owner = THIS_MODULE,
510 .release = unix_release,
512 .connect = unix_stream_connect,
513 .socketpair = unix_socketpair,
514 .accept = unix_accept,
515 .getname = unix_getname,
518 .listen = unix_listen,
519 .shutdown = unix_shutdown,
520 .setsockopt = sock_no_setsockopt,
521 .getsockopt = sock_no_getsockopt,
522 .sendmsg = unix_stream_sendmsg,
523 .recvmsg = unix_stream_recvmsg,
524 .mmap = sock_no_mmap,
525 .sendpage = sock_no_sendpage,
528 static const struct proto_ops unix_dgram_ops = {
530 .owner = THIS_MODULE,
531 .release = unix_release,
533 .connect = unix_dgram_connect,
534 .socketpair = unix_socketpair,
535 .accept = sock_no_accept,
536 .getname = unix_getname,
537 .poll = unix_dgram_poll,
539 .listen = sock_no_listen,
540 .shutdown = unix_shutdown,
541 .setsockopt = sock_no_setsockopt,
542 .getsockopt = sock_no_getsockopt,
543 .sendmsg = unix_dgram_sendmsg,
544 .recvmsg = unix_dgram_recvmsg,
545 .mmap = sock_no_mmap,
546 .sendpage = sock_no_sendpage,
549 static const struct proto_ops unix_seqpacket_ops = {
551 .owner = THIS_MODULE,
552 .release = unix_release,
554 .connect = unix_stream_connect,
555 .socketpair = unix_socketpair,
556 .accept = unix_accept,
557 .getname = unix_getname,
558 .poll = unix_dgram_poll,
560 .listen = unix_listen,
561 .shutdown = unix_shutdown,
562 .setsockopt = sock_no_setsockopt,
563 .getsockopt = sock_no_getsockopt,
564 .sendmsg = unix_seqpacket_sendmsg,
565 .recvmsg = unix_dgram_recvmsg,
566 .mmap = sock_no_mmap,
567 .sendpage = sock_no_sendpage,
570 static struct proto unix_proto = {
572 .owner = THIS_MODULE,
573 .obj_size = sizeof(struct unix_sock),
577 * AF_UNIX sockets do not interact with hardware, hence they
578 * dont trigger interrupts - so it's safe for them to have
579 * bh-unsafe locking for their sk_receive_queue.lock. Split off
580 * this special lock-class by reinitializing the spinlock key:
582 static struct lock_class_key af_unix_sk_receive_queue_lock_key;
584 static struct sock *unix_create1(struct net *net, struct socket *sock)
586 struct sock *sk = NULL;
589 atomic_inc(&unix_nr_socks);
590 if (atomic_read(&unix_nr_socks) > 2 * get_max_files())
593 sk = sk_alloc(net, PF_UNIX, GFP_KERNEL, &unix_proto);
597 sock_init_data(sock, sk);
598 lockdep_set_class(&sk->sk_receive_queue.lock,
599 &af_unix_sk_receive_queue_lock_key);
601 sk->sk_write_space = unix_write_space;
602 sk->sk_max_ack_backlog = net->unx.sysctl_max_dgram_qlen;
603 sk->sk_destruct = unix_sock_destructor;
607 spin_lock_init(&u->lock);
608 atomic_long_set(&u->inflight, 0);
609 INIT_LIST_HEAD(&u->link);
610 mutex_init(&u->readlock); /* single task reading lock */
611 init_waitqueue_head(&u->peer_wait);
612 unix_insert_socket(unix_sockets_unbound, sk);
615 atomic_dec(&unix_nr_socks);
618 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
624 static int unix_create(struct net *net, struct socket *sock, int protocol)
626 if (protocol && protocol != PF_UNIX)
627 return -EPROTONOSUPPORT;
629 sock->state = SS_UNCONNECTED;
631 switch (sock->type) {
633 sock->ops = &unix_stream_ops;
636 * Believe it or not BSD has AF_UNIX, SOCK_RAW though
640 sock->type = SOCK_DGRAM;
642 sock->ops = &unix_dgram_ops;
645 sock->ops = &unix_seqpacket_ops;
648 return -ESOCKTNOSUPPORT;
651 return unix_create1(net, sock) ? 0 : -ENOMEM;
654 static int unix_release(struct socket *sock)
656 struct sock *sk = sock->sk;
663 return unix_release_sock(sk, 0);
666 static int unix_autobind(struct socket *sock)
668 struct sock *sk = sock->sk;
669 struct net *net = sock_net(sk);
670 struct unix_sock *u = unix_sk(sk);
671 static u32 ordernum = 1;
672 struct unix_address *addr;
674 unsigned int retries = 0;
676 mutex_lock(&u->readlock);
683 addr = kzalloc(sizeof(*addr) + sizeof(short) + 16, GFP_KERNEL);
687 addr->name->sun_family = AF_UNIX;
688 atomic_set(&addr->refcnt, 1);
691 addr->len = sprintf(addr->name->sun_path+1, "%05x", ordernum) + 1 + sizeof(short);
692 addr->hash = unix_hash_fold(csum_partial(addr->name, addr->len, 0));
694 spin_lock(&unix_table_lock);
695 ordernum = (ordernum+1)&0xFFFFF;
697 if (__unix_find_socket_byname(net, addr->name, addr->len, sock->type,
699 spin_unlock(&unix_table_lock);
701 * __unix_find_socket_byname() may take long time if many names
702 * are already in use.
705 /* Give up if all names seems to be in use. */
706 if (retries++ == 0xFFFFF) {
713 addr->hash ^= sk->sk_type;
715 __unix_remove_socket(sk);
717 __unix_insert_socket(&unix_socket_table[addr->hash], sk);
718 spin_unlock(&unix_table_lock);
721 out: mutex_unlock(&u->readlock);
725 static struct sock *unix_find_other(struct net *net,
726 struct sockaddr_un *sunname, int len,
727 int type, unsigned hash, int *error)
733 if (sunname->sun_path[0]) {
735 err = kern_path(sunname->sun_path, LOOKUP_FOLLOW, &path);
738 inode = path.dentry->d_inode;
739 err = inode_permission(inode, MAY_WRITE);
744 if (!S_ISSOCK(inode->i_mode))
746 u = unix_find_socket_byinode(net, inode);
750 if (u->sk_type == type)
751 touch_atime(path.mnt, path.dentry);
756 if (u->sk_type != type) {
762 u = unix_find_socket_byname(net, sunname, len, type, hash);
764 struct dentry *dentry;
765 dentry = unix_sk(u)->dentry;
767 touch_atime(unix_sk(u)->mnt, dentry);
781 static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
783 struct sock *sk = sock->sk;
784 struct net *net = sock_net(sk);
785 struct unix_sock *u = unix_sk(sk);
786 struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
787 struct dentry *dentry = NULL;
791 struct unix_address *addr;
792 struct hlist_head *list;
795 if (sunaddr->sun_family != AF_UNIX)
798 if (addr_len == sizeof(short)) {
799 err = unix_autobind(sock);
803 err = unix_mkname(sunaddr, addr_len, &hash);
808 mutex_lock(&u->readlock);
815 addr = kmalloc(sizeof(*addr)+addr_len, GFP_KERNEL);
819 memcpy(addr->name, sunaddr, addr_len);
820 addr->len = addr_len;
821 addr->hash = hash ^ sk->sk_type;
822 atomic_set(&addr->refcnt, 1);
824 if (sunaddr->sun_path[0]) {
828 * Get the parent directory, calculate the hash for last
831 err = path_lookup(sunaddr->sun_path, LOOKUP_PARENT, &nd);
833 goto out_mknod_parent;
835 dentry = lookup_create(&nd, 0);
836 err = PTR_ERR(dentry);
838 goto out_mknod_unlock;
841 * All right, let's create it.
844 (SOCK_INODE(sock)->i_mode & ~current_umask());
845 err = mnt_want_write(nd.path.mnt);
848 err = security_path_mknod(&nd.path, dentry, mode, 0);
850 goto out_mknod_drop_write;
851 err = vfs_mknod(nd.path.dentry->d_inode, dentry, mode, 0);
852 out_mknod_drop_write:
853 mnt_drop_write(nd.path.mnt);
856 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
857 dput(nd.path.dentry);
858 nd.path.dentry = dentry;
860 addr->hash = UNIX_HASH_SIZE;
863 spin_lock(&unix_table_lock);
865 if (!sunaddr->sun_path[0]) {
867 if (__unix_find_socket_byname(net, sunaddr, addr_len,
868 sk->sk_type, hash)) {
869 unix_release_addr(addr);
873 list = &unix_socket_table[addr->hash];
875 list = &unix_socket_table[dentry->d_inode->i_ino & (UNIX_HASH_SIZE-1)];
876 u->dentry = nd.path.dentry;
877 u->mnt = nd.path.mnt;
881 __unix_remove_socket(sk);
883 __unix_insert_socket(list, sk);
886 spin_unlock(&unix_table_lock);
888 mutex_unlock(&u->readlock);
895 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
900 unix_release_addr(addr);
904 static void unix_state_double_lock(struct sock *sk1, struct sock *sk2)
906 if (unlikely(sk1 == sk2) || !sk2) {
907 unix_state_lock(sk1);
911 unix_state_lock(sk1);
912 unix_state_lock_nested(sk2);
914 unix_state_lock(sk2);
915 unix_state_lock_nested(sk1);
919 static void unix_state_double_unlock(struct sock *sk1, struct sock *sk2)
921 if (unlikely(sk1 == sk2) || !sk2) {
922 unix_state_unlock(sk1);
925 unix_state_unlock(sk1);
926 unix_state_unlock(sk2);
929 static int unix_dgram_connect(struct socket *sock, struct sockaddr *addr,
932 struct sock *sk = sock->sk;
933 struct net *net = sock_net(sk);
934 struct sockaddr_un *sunaddr = (struct sockaddr_un *)addr;
939 if (addr->sa_family != AF_UNSPEC) {
940 err = unix_mkname(sunaddr, alen, &hash);
945 if (test_bit(SOCK_PASSCRED, &sock->flags) &&
946 !unix_sk(sk)->addr && (err = unix_autobind(sock)) != 0)
950 other = unix_find_other(net, sunaddr, alen, sock->type, hash, &err);
954 unix_state_double_lock(sk, other);
956 /* Apparently VFS overslept socket death. Retry. */
957 if (sock_flag(other, SOCK_DEAD)) {
958 unix_state_double_unlock(sk, other);
964 if (!unix_may_send(sk, other))
967 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
973 * 1003.1g breaking connected state with AF_UNSPEC
976 unix_state_double_lock(sk, other);
980 * If it was connected, reconnect.
983 struct sock *old_peer = unix_peer(sk);
984 unix_peer(sk) = other;
985 unix_state_double_unlock(sk, other);
987 if (other != old_peer)
988 unix_dgram_disconnected(sk, old_peer);
991 unix_peer(sk) = other;
992 unix_state_double_unlock(sk, other);
997 unix_state_double_unlock(sk, other);
1003 static long unix_wait_for_peer(struct sock *other, long timeo)
1005 struct unix_sock *u = unix_sk(other);
1009 prepare_to_wait_exclusive(&u->peer_wait, &wait, TASK_INTERRUPTIBLE);
1011 sched = !sock_flag(other, SOCK_DEAD) &&
1012 !(other->sk_shutdown & RCV_SHUTDOWN) &&
1013 unix_recvq_full(other);
1015 unix_state_unlock(other);
1018 timeo = schedule_timeout(timeo);
1020 finish_wait(&u->peer_wait, &wait);
1024 static int unix_stream_connect(struct socket *sock, struct sockaddr *uaddr,
1025 int addr_len, int flags)
1027 struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
1028 struct sock *sk = sock->sk;
1029 struct net *net = sock_net(sk);
1030 struct unix_sock *u = unix_sk(sk), *newu, *otheru;
1031 struct sock *newsk = NULL;
1032 struct sock *other = NULL;
1033 struct sk_buff *skb = NULL;
1039 err = unix_mkname(sunaddr, addr_len, &hash);
1044 if (test_bit(SOCK_PASSCRED, &sock->flags)
1045 && !u->addr && (err = unix_autobind(sock)) != 0)
1048 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
1050 /* First of all allocate resources.
1051 If we will make it after state is locked,
1052 we will have to recheck all again in any case.
1057 /* create new sock for complete connection */
1058 newsk = unix_create1(sock_net(sk), NULL);
1062 /* Allocate skb for sending to listening sock */
1063 skb = sock_wmalloc(newsk, 1, 0, GFP_KERNEL);
1068 /* Find listening sock. */
1069 other = unix_find_other(net, sunaddr, addr_len, sk->sk_type, hash, &err);
1073 /* Latch state of peer */
1074 unix_state_lock(other);
1076 /* Apparently VFS overslept socket death. Retry. */
1077 if (sock_flag(other, SOCK_DEAD)) {
1078 unix_state_unlock(other);
1083 err = -ECONNREFUSED;
1084 if (other->sk_state != TCP_LISTEN)
1086 if (other->sk_shutdown & RCV_SHUTDOWN)
1089 if (unix_recvq_full(other)) {
1094 timeo = unix_wait_for_peer(other, timeo);
1096 err = sock_intr_errno(timeo);
1097 if (signal_pending(current))
1105 It is tricky place. We need to grab write lock and cannot
1106 drop lock on peer. It is dangerous because deadlock is
1107 possible. Connect to self case and simultaneous
1108 attempt to connect are eliminated by checking socket
1109 state. other is TCP_LISTEN, if sk is TCP_LISTEN we
1110 check this before attempt to grab lock.
1112 Well, and we have to recheck the state after socket locked.
1118 /* This is ok... continue with connect */
1120 case TCP_ESTABLISHED:
1121 /* Socket is already connected */
1129 unix_state_lock_nested(sk);
1131 if (sk->sk_state != st) {
1132 unix_state_unlock(sk);
1133 unix_state_unlock(other);
1138 err = security_unix_stream_connect(sock, other->sk_socket, newsk);
1140 unix_state_unlock(sk);
1144 /* The way is open! Fastly set all the necessary fields... */
1147 unix_peer(newsk) = sk;
1148 newsk->sk_state = TCP_ESTABLISHED;
1149 newsk->sk_type = sk->sk_type;
1150 newsk->sk_peercred.pid = task_tgid_vnr(current);
1151 current_euid_egid(&newsk->sk_peercred.uid, &newsk->sk_peercred.gid);
1152 newu = unix_sk(newsk);
1153 newsk->sk_sleep = &newu->peer_wait;
1154 otheru = unix_sk(other);
1156 /* copy address information from listening to new sock*/
1158 atomic_inc(&otheru->addr->refcnt);
1159 newu->addr = otheru->addr;
1161 if (otheru->dentry) {
1162 newu->dentry = dget(otheru->dentry);
1163 newu->mnt = mntget(otheru->mnt);
1166 /* Set credentials */
1167 sk->sk_peercred = other->sk_peercred;
1169 sock->state = SS_CONNECTED;
1170 sk->sk_state = TCP_ESTABLISHED;
1173 smp_mb__after_atomic_inc(); /* sock_hold() does an atomic_inc() */
1174 unix_peer(sk) = newsk;
1176 unix_state_unlock(sk);
1178 /* take ten and and send info to listening sock */
1179 spin_lock(&other->sk_receive_queue.lock);
1180 __skb_queue_tail(&other->sk_receive_queue, skb);
1181 spin_unlock(&other->sk_receive_queue.lock);
1182 unix_state_unlock(other);
1183 other->sk_data_ready(other, 0);
1189 unix_state_unlock(other);
1194 unix_release_sock(newsk, 0);
1200 static int unix_socketpair(struct socket *socka, struct socket *sockb)
1202 struct sock *ska = socka->sk, *skb = sockb->sk;
1204 /* Join our sockets back to back */
1207 unix_peer(ska) = skb;
1208 unix_peer(skb) = ska;
1209 ska->sk_peercred.pid = skb->sk_peercred.pid = task_tgid_vnr(current);
1210 current_euid_egid(&skb->sk_peercred.uid, &skb->sk_peercred.gid);
1211 ska->sk_peercred.uid = skb->sk_peercred.uid;
1212 ska->sk_peercred.gid = skb->sk_peercred.gid;
1214 if (ska->sk_type != SOCK_DGRAM) {
1215 ska->sk_state = TCP_ESTABLISHED;
1216 skb->sk_state = TCP_ESTABLISHED;
1217 socka->state = SS_CONNECTED;
1218 sockb->state = SS_CONNECTED;
1223 static int unix_accept(struct socket *sock, struct socket *newsock, int flags)
1225 struct sock *sk = sock->sk;
1227 struct sk_buff *skb;
1231 if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
1235 if (sk->sk_state != TCP_LISTEN)
1238 /* If socket state is TCP_LISTEN it cannot change (for now...),
1239 * so that no locks are necessary.
1242 skb = skb_recv_datagram(sk, 0, flags&O_NONBLOCK, &err);
1244 /* This means receive shutdown. */
1251 skb_free_datagram(sk, skb);
1252 wake_up_interruptible(&unix_sk(sk)->peer_wait);
1254 /* attach accepted sock to socket */
1255 unix_state_lock(tsk);
1256 newsock->state = SS_CONNECTED;
1257 sock_graft(tsk, newsock);
1258 unix_state_unlock(tsk);
1266 static int unix_getname(struct socket *sock, struct sockaddr *uaddr, int *uaddr_len, int peer)
1268 struct sock *sk = sock->sk;
1269 struct unix_sock *u;
1270 struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
1274 sk = unix_peer_get(sk);
1285 unix_state_lock(sk);
1287 sunaddr->sun_family = AF_UNIX;
1288 sunaddr->sun_path[0] = 0;
1289 *uaddr_len = sizeof(short);
1291 struct unix_address *addr = u->addr;
1293 *uaddr_len = addr->len;
1294 memcpy(sunaddr, addr->name, *uaddr_len);
1296 unix_state_unlock(sk);
1302 static void unix_detach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1306 scm->fp = UNIXCB(skb).fp;
1307 skb->destructor = sock_wfree;
1308 UNIXCB(skb).fp = NULL;
1310 for (i = scm->fp->count-1; i >= 0; i--)
1311 unix_notinflight(scm->fp->fp[i]);
1314 static void unix_destruct_fds(struct sk_buff *skb)
1316 struct scm_cookie scm;
1317 memset(&scm, 0, sizeof(scm));
1318 unix_detach_fds(&scm, skb);
1320 /* Alas, it calls VFS */
1321 /* So fscking what? fput() had been SMP-safe since the last Summer */
1326 static int unix_attach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1331 * Need to duplicate file references for the sake of garbage
1332 * collection. Otherwise a socket in the fps might become a
1333 * candidate for GC while the skb is not yet queued.
1335 UNIXCB(skb).fp = scm_fp_dup(scm->fp);
1336 if (!UNIXCB(skb).fp)
1339 for (i = scm->fp->count-1; i >= 0; i--)
1340 unix_inflight(scm->fp->fp[i]);
1341 skb->destructor = unix_destruct_fds;
1346 * Send AF_UNIX data.
1349 static int unix_dgram_sendmsg(struct kiocb *kiocb, struct socket *sock,
1350 struct msghdr *msg, size_t len)
1352 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1353 struct sock *sk = sock->sk;
1354 struct net *net = sock_net(sk);
1355 struct unix_sock *u = unix_sk(sk);
1356 struct sockaddr_un *sunaddr = msg->msg_name;
1357 struct sock *other = NULL;
1358 int namelen = 0; /* fake GCC */
1361 struct sk_buff *skb;
1363 struct scm_cookie tmp_scm;
1365 if (NULL == siocb->scm)
1366 siocb->scm = &tmp_scm;
1368 err = scm_send(sock, msg, siocb->scm);
1373 if (msg->msg_flags&MSG_OOB)
1376 if (msg->msg_namelen) {
1377 err = unix_mkname(sunaddr, msg->msg_namelen, &hash);
1384 other = unix_peer_get(sk);
1389 if (test_bit(SOCK_PASSCRED, &sock->flags)
1390 && !u->addr && (err = unix_autobind(sock)) != 0)
1394 if (len > sk->sk_sndbuf - 32)
1397 skb = sock_alloc_send_skb(sk, len, msg->msg_flags&MSG_DONTWAIT, &err);
1401 memcpy(UNIXCREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1402 if (siocb->scm->fp) {
1403 err = unix_attach_fds(siocb->scm, skb);
1407 unix_get_secdata(siocb->scm, skb);
1409 skb_reset_transport_header(skb);
1410 err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
1414 timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1419 if (sunaddr == NULL)
1422 other = unix_find_other(net, sunaddr, namelen, sk->sk_type,
1428 unix_state_lock(other);
1430 if (!unix_may_send(sk, other))
1433 if (sock_flag(other, SOCK_DEAD)) {
1435 * Check with 1003.1g - what should
1438 unix_state_unlock(other);
1442 unix_state_lock(sk);
1443 if (unix_peer(sk) == other) {
1444 unix_peer(sk) = NULL;
1445 unix_state_unlock(sk);
1447 unix_dgram_disconnected(sk, other);
1449 err = -ECONNREFUSED;
1451 unix_state_unlock(sk);
1461 if (other->sk_shutdown & RCV_SHUTDOWN)
1464 if (sk->sk_type != SOCK_SEQPACKET) {
1465 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1470 if (unix_peer(other) != sk && unix_recvq_full(other)) {
1476 timeo = unix_wait_for_peer(other, timeo);
1478 err = sock_intr_errno(timeo);
1479 if (signal_pending(current))
1485 skb_queue_tail(&other->sk_receive_queue, skb);
1486 unix_state_unlock(other);
1487 other->sk_data_ready(other, len);
1489 scm_destroy(siocb->scm);
1493 unix_state_unlock(other);
1499 scm_destroy(siocb->scm);
1504 static int unix_stream_sendmsg(struct kiocb *kiocb, struct socket *sock,
1505 struct msghdr *msg, size_t len)
1507 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1508 struct sock *sk = sock->sk;
1509 struct sock *other = NULL;
1510 struct sockaddr_un *sunaddr = msg->msg_name;
1512 struct sk_buff *skb;
1514 struct scm_cookie tmp_scm;
1515 bool fds_sent = false;
1517 if (NULL == siocb->scm)
1518 siocb->scm = &tmp_scm;
1520 err = scm_send(sock, msg, siocb->scm);
1525 if (msg->msg_flags&MSG_OOB)
1528 if (msg->msg_namelen) {
1529 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
1534 other = unix_peer(sk);
1539 if (sk->sk_shutdown & SEND_SHUTDOWN)
1542 while (sent < len) {
1544 * Optimisation for the fact that under 0.01% of X
1545 * messages typically need breaking up.
1550 /* Keep two messages in the pipe so it schedules better */
1551 if (size > ((sk->sk_sndbuf >> 1) - 64))
1552 size = (sk->sk_sndbuf >> 1) - 64;
1554 if (size > SKB_MAX_ALLOC)
1555 size = SKB_MAX_ALLOC;
1561 skb = sock_alloc_send_skb(sk, size, msg->msg_flags&MSG_DONTWAIT,
1568 * If you pass two values to the sock_alloc_send_skb
1569 * it tries to grab the large buffer with GFP_NOFS
1570 * (which can fail easily), and if it fails grab the
1571 * fallback size buffer which is under a page and will
1574 size = min_t(int, size, skb_tailroom(skb));
1576 memcpy(UNIXCREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1577 /* Only send the fds in the first buffer */
1578 if (siocb->scm->fp && !fds_sent) {
1579 err = unix_attach_fds(siocb->scm, skb);
1587 err = memcpy_fromiovec(skb_put(skb, size), msg->msg_iov, size);
1593 unix_state_lock(other);
1595 if (sock_flag(other, SOCK_DEAD) ||
1596 (other->sk_shutdown & RCV_SHUTDOWN))
1599 skb_queue_tail(&other->sk_receive_queue, skb);
1600 unix_state_unlock(other);
1601 other->sk_data_ready(other, size);
1605 scm_destroy(siocb->scm);
1611 unix_state_unlock(other);
1614 if (sent == 0 && !(msg->msg_flags&MSG_NOSIGNAL))
1615 send_sig(SIGPIPE, current, 0);
1618 scm_destroy(siocb->scm);
1620 return sent ? : err;
1623 static int unix_seqpacket_sendmsg(struct kiocb *kiocb, struct socket *sock,
1624 struct msghdr *msg, size_t len)
1627 struct sock *sk = sock->sk;
1629 err = sock_error(sk);
1633 if (sk->sk_state != TCP_ESTABLISHED)
1636 if (msg->msg_namelen)
1637 msg->msg_namelen = 0;
1639 return unix_dgram_sendmsg(kiocb, sock, msg, len);
1642 static void unix_copy_addr(struct msghdr *msg, struct sock *sk)
1644 struct unix_sock *u = unix_sk(sk);
1646 msg->msg_namelen = 0;
1648 msg->msg_namelen = u->addr->len;
1649 memcpy(msg->msg_name, u->addr->name, u->addr->len);
1653 static int unix_dgram_recvmsg(struct kiocb *iocb, struct socket *sock,
1654 struct msghdr *msg, size_t size,
1657 struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1658 struct scm_cookie tmp_scm;
1659 struct sock *sk = sock->sk;
1660 struct unix_sock *u = unix_sk(sk);
1661 int noblock = flags & MSG_DONTWAIT;
1662 struct sk_buff *skb;
1669 msg->msg_namelen = 0;
1671 mutex_lock(&u->readlock);
1673 skb = skb_recv_datagram(sk, flags, noblock, &err);
1675 unix_state_lock(sk);
1676 /* Signal EOF on disconnected non-blocking SEQPACKET socket. */
1677 if (sk->sk_type == SOCK_SEQPACKET && err == -EAGAIN &&
1678 (sk->sk_shutdown & RCV_SHUTDOWN))
1680 unix_state_unlock(sk);
1684 wake_up_interruptible_sync(&u->peer_wait);
1687 unix_copy_addr(msg, skb->sk);
1689 if (size > skb->len)
1691 else if (size < skb->len)
1692 msg->msg_flags |= MSG_TRUNC;
1694 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, size);
1699 siocb->scm = &tmp_scm;
1700 memset(&tmp_scm, 0, sizeof(tmp_scm));
1702 siocb->scm->creds = *UNIXCREDS(skb);
1703 unix_set_secdata(siocb->scm, skb);
1705 if (!(flags & MSG_PEEK)) {
1707 unix_detach_fds(siocb->scm, skb);
1709 /* It is questionable: on PEEK we could:
1710 - do not return fds - good, but too simple 8)
1711 - return fds, and do not return them on read (old strategy,
1713 - clone fds (I chose it for now, it is the most universal
1716 POSIX 1003.1g does not actually define this clearly
1717 at all. POSIX 1003.1g doesn't define a lot of things
1722 siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1726 scm_recv(sock, msg, siocb->scm, flags);
1729 skb_free_datagram(sk, skb);
1731 mutex_unlock(&u->readlock);
1737 * Sleep until data has arrive. But check for races..
1740 static long unix_stream_data_wait(struct sock *sk, long timeo)
1744 unix_state_lock(sk);
1747 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1749 if (!skb_queue_empty(&sk->sk_receive_queue) ||
1751 (sk->sk_shutdown & RCV_SHUTDOWN) ||
1752 signal_pending(current) ||
1756 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1757 unix_state_unlock(sk);
1758 timeo = schedule_timeout(timeo);
1759 unix_state_lock(sk);
1760 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1763 finish_wait(sk->sk_sleep, &wait);
1764 unix_state_unlock(sk);
1770 static int unix_stream_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 struct sockaddr_un *sunaddr = msg->msg_name;
1780 int check_creds = 0;
1786 if (sk->sk_state != TCP_ESTABLISHED)
1793 target = sock_rcvlowat(sk, flags&MSG_WAITALL, size);
1794 timeo = sock_rcvtimeo(sk, flags&MSG_DONTWAIT);
1796 msg->msg_namelen = 0;
1798 /* Lock the socket to prevent queue disordering
1799 * while sleeps in memcpy_tomsg
1803 siocb->scm = &tmp_scm;
1804 memset(&tmp_scm, 0, sizeof(tmp_scm));
1807 mutex_lock(&u->readlock);
1811 struct sk_buff *skb;
1813 unix_state_lock(sk);
1814 skb = skb_dequeue(&sk->sk_receive_queue);
1816 if (copied >= target)
1820 * POSIX 1003.1g mandates this order.
1823 err = sock_error(sk);
1826 if (sk->sk_shutdown & RCV_SHUTDOWN)
1829 unix_state_unlock(sk);
1833 mutex_unlock(&u->readlock);
1835 timeo = unix_stream_data_wait(sk, timeo);
1837 if (signal_pending(current)) {
1838 err = sock_intr_errno(timeo);
1841 mutex_lock(&u->readlock);
1844 unix_state_unlock(sk);
1847 unix_state_unlock(sk);
1850 /* Never glue messages from different writers */
1851 if (memcmp(UNIXCREDS(skb), &siocb->scm->creds,
1852 sizeof(siocb->scm->creds)) != 0) {
1853 skb_queue_head(&sk->sk_receive_queue, skb);
1857 /* Copy credentials */
1858 siocb->scm->creds = *UNIXCREDS(skb);
1862 /* Copy address just once */
1864 unix_copy_addr(msg, skb->sk);
1868 chunk = min_t(unsigned int, skb->len, size);
1869 if (memcpy_toiovec(msg->msg_iov, skb->data, chunk)) {
1870 skb_queue_head(&sk->sk_receive_queue, skb);
1878 /* Mark read part of skb as used */
1879 if (!(flags & MSG_PEEK)) {
1880 skb_pull(skb, chunk);
1883 unix_detach_fds(siocb->scm, skb);
1885 /* put the skb back if we didn't use it up.. */
1887 skb_queue_head(&sk->sk_receive_queue, skb);
1896 /* It is questionable, see note in unix_dgram_recvmsg.
1899 siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1901 /* put message back and return */
1902 skb_queue_head(&sk->sk_receive_queue, skb);
1907 mutex_unlock(&u->readlock);
1908 scm_recv(sock, msg, siocb->scm, flags);
1910 return copied ? : err;
1913 static int unix_shutdown(struct socket *sock, int mode)
1915 struct sock *sk = sock->sk;
1918 mode = (mode+1)&(RCV_SHUTDOWN|SEND_SHUTDOWN);
1921 unix_state_lock(sk);
1922 sk->sk_shutdown |= mode;
1923 other = unix_peer(sk);
1926 unix_state_unlock(sk);
1927 sk->sk_state_change(sk);
1930 (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET)) {
1934 if (mode&RCV_SHUTDOWN)
1935 peer_mode |= SEND_SHUTDOWN;
1936 if (mode&SEND_SHUTDOWN)
1937 peer_mode |= RCV_SHUTDOWN;
1938 unix_state_lock(other);
1939 other->sk_shutdown |= peer_mode;
1940 unix_state_unlock(other);
1941 other->sk_state_change(other);
1942 read_lock(&other->sk_callback_lock);
1943 if (peer_mode == SHUTDOWN_MASK)
1944 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_HUP);
1945 else if (peer_mode & RCV_SHUTDOWN)
1946 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_IN);
1947 read_unlock(&other->sk_callback_lock);
1955 static int unix_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1957 struct sock *sk = sock->sk;
1963 amount = sk_wmem_alloc_get(sk);
1964 err = put_user(amount, (int __user *)arg);
1968 struct sk_buff *skb;
1970 if (sk->sk_state == TCP_LISTEN) {
1975 spin_lock(&sk->sk_receive_queue.lock);
1976 if (sk->sk_type == SOCK_STREAM ||
1977 sk->sk_type == SOCK_SEQPACKET) {
1978 skb_queue_walk(&sk->sk_receive_queue, skb)
1981 skb = skb_peek(&sk->sk_receive_queue);
1985 spin_unlock(&sk->sk_receive_queue.lock);
1986 err = put_user(amount, (int __user *)arg);
1997 static unsigned int unix_poll(struct file *file, struct socket *sock, poll_table *wait)
1999 struct sock *sk = sock->sk;
2002 sock_poll_wait(file, sk->sk_sleep, wait);
2005 /* exceptional events? */
2008 if (sk->sk_shutdown == SHUTDOWN_MASK)
2010 if (sk->sk_shutdown & RCV_SHUTDOWN)
2014 if (!skb_queue_empty(&sk->sk_receive_queue) ||
2015 (sk->sk_shutdown & RCV_SHUTDOWN))
2016 mask |= POLLIN | POLLRDNORM;
2018 /* Connection-based need to check for termination and startup */
2019 if ((sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) &&
2020 sk->sk_state == TCP_CLOSE)
2024 * we set writable also when the other side has shut down the
2025 * connection. This prevents stuck sockets.
2027 if (unix_writable(sk))
2028 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2033 static unsigned int unix_dgram_poll(struct file *file, struct socket *sock,
2036 struct sock *sk = sock->sk, *other;
2037 unsigned int mask, writable;
2039 sock_poll_wait(file, sk->sk_sleep, wait);
2042 /* exceptional events? */
2043 if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue))
2045 if (sk->sk_shutdown & RCV_SHUTDOWN)
2047 if (sk->sk_shutdown == SHUTDOWN_MASK)
2051 if (!skb_queue_empty(&sk->sk_receive_queue) ||
2052 (sk->sk_shutdown & RCV_SHUTDOWN))
2053 mask |= POLLIN | POLLRDNORM;
2055 /* Connection-based need to check for termination and startup */
2056 if (sk->sk_type == SOCK_SEQPACKET) {
2057 if (sk->sk_state == TCP_CLOSE)
2059 /* connection hasn't started yet? */
2060 if (sk->sk_state == TCP_SYN_SENT)
2065 writable = unix_writable(sk);
2067 other = unix_peer_get(sk);
2069 if (unix_peer(other) != sk) {
2070 sock_poll_wait(file, &unix_sk(other)->peer_wait,
2072 if (unix_recvq_full(other))
2081 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2083 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
2088 #ifdef CONFIG_PROC_FS
2089 static struct sock *first_unix_socket(int *i)
2091 for (*i = 0; *i <= UNIX_HASH_SIZE; (*i)++) {
2092 if (!hlist_empty(&unix_socket_table[*i]))
2093 return __sk_head(&unix_socket_table[*i]);
2098 static struct sock *next_unix_socket(int *i, struct sock *s)
2100 struct sock *next = sk_next(s);
2101 /* More in this chain? */
2104 /* Look for next non-empty chain. */
2105 for ((*i)++; *i <= UNIX_HASH_SIZE; (*i)++) {
2106 if (!hlist_empty(&unix_socket_table[*i]))
2107 return __sk_head(&unix_socket_table[*i]);
2112 struct unix_iter_state {
2113 struct seq_net_private p;
2117 static struct sock *unix_seq_idx(struct seq_file *seq, loff_t pos)
2119 struct unix_iter_state *iter = seq->private;
2123 for (s = first_unix_socket(&iter->i); s; s = next_unix_socket(&iter->i, s)) {
2124 if (sock_net(s) != seq_file_net(seq))
2133 static void *unix_seq_start(struct seq_file *seq, loff_t *pos)
2134 __acquires(unix_table_lock)
2136 spin_lock(&unix_table_lock);
2137 return *pos ? unix_seq_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2140 static void *unix_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2142 struct unix_iter_state *iter = seq->private;
2143 struct sock *sk = v;
2146 if (v == SEQ_START_TOKEN)
2147 sk = first_unix_socket(&iter->i);
2149 sk = next_unix_socket(&iter->i, sk);
2150 while (sk && (sock_net(sk) != seq_file_net(seq)))
2151 sk = next_unix_socket(&iter->i, sk);
2155 static void unix_seq_stop(struct seq_file *seq, void *v)
2156 __releases(unix_table_lock)
2158 spin_unlock(&unix_table_lock);
2161 static int unix_seq_show(struct seq_file *seq, void *v)
2164 if (v == SEQ_START_TOKEN)
2165 seq_puts(seq, "Num RefCount Protocol Flags Type St "
2169 struct unix_sock *u = unix_sk(s);
2172 seq_printf(seq, "%p: %08X %08X %08X %04X %02X %5lu",
2174 atomic_read(&s->sk_refcnt),
2176 s->sk_state == TCP_LISTEN ? __SO_ACCEPTCON : 0,
2179 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTED : SS_UNCONNECTED) :
2180 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTING : SS_DISCONNECTING),
2188 len = u->addr->len - sizeof(short);
2189 if (!UNIX_ABSTRACT(s))
2195 for ( ; i < len; i++)
2196 seq_putc(seq, u->addr->name->sun_path[i]);
2198 unix_state_unlock(s);
2199 seq_putc(seq, '\n');
2205 static const struct seq_operations unix_seq_ops = {
2206 .start = unix_seq_start,
2207 .next = unix_seq_next,
2208 .stop = unix_seq_stop,
2209 .show = unix_seq_show,
2212 static int unix_seq_open(struct inode *inode, struct file *file)
2214 return seq_open_net(inode, file, &unix_seq_ops,
2215 sizeof(struct unix_iter_state));
2218 static const struct file_operations unix_seq_fops = {
2219 .owner = THIS_MODULE,
2220 .open = unix_seq_open,
2222 .llseek = seq_lseek,
2223 .release = seq_release_net,
2228 static struct net_proto_family unix_family_ops = {
2230 .create = unix_create,
2231 .owner = THIS_MODULE,
2235 static int unix_net_init(struct net *net)
2237 int error = -ENOMEM;
2239 net->unx.sysctl_max_dgram_qlen = 10;
2240 if (unix_sysctl_register(net))
2243 #ifdef CONFIG_PROC_FS
2244 if (!proc_net_fops_create(net, "unix", 0, &unix_seq_fops)) {
2245 unix_sysctl_unregister(net);
2254 static void unix_net_exit(struct net *net)
2256 unix_sysctl_unregister(net);
2257 proc_net_remove(net, "unix");
2260 static struct pernet_operations unix_net_ops = {
2261 .init = unix_net_init,
2262 .exit = unix_net_exit,
2265 static int __init af_unix_init(void)
2268 struct sk_buff *dummy_skb;
2270 BUILD_BUG_ON(sizeof(struct unix_skb_parms) > sizeof(dummy_skb->cb));
2272 rc = proto_register(&unix_proto, 1);
2274 printk(KERN_CRIT "%s: Cannot create unix_sock SLAB cache!\n",
2279 sock_register(&unix_family_ops);
2280 register_pernet_subsys(&unix_net_ops);
2285 static void __exit af_unix_exit(void)
2287 sock_unregister(PF_UNIX);
2288 proto_unregister(&unix_proto);
2289 unregister_pernet_subsys(&unix_net_ops);
2292 /* Earlier than device_initcall() so that other drivers invoking
2293 request_module() don't end up in a loop when modprobe tries
2294 to use a UNIX socket. But later than subsys_initcall() because
2295 we depend on stuff initialised there */
2296 fs_initcall(af_unix_init);
2297 module_exit(af_unix_exit);
2299 MODULE_LICENSE("GPL");
2300 MODULE_ALIAS_NETPROTO(PF_UNIX);