2 * NET4: Implementation of BSD Unix domain sockets.
4 * Authors: Alan Cox, <alan@lxorguk.ukuu.org.uk>
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License
8 * as published by the Free Software Foundation; either version
9 * 2 of the License, or (at your option) any later version.
12 * Linus Torvalds : Assorted bug cures.
13 * Niibe Yutaka : async I/O support.
14 * Carsten Paeth : PF_UNIX check, address fixes.
15 * Alan Cox : Limit size of allocated blocks.
16 * Alan Cox : Fixed the stupid socketpair bug.
17 * Alan Cox : BSD compatibility fine tuning.
18 * Alan Cox : Fixed a bug in connect when interrupted.
19 * Alan Cox : Sorted out a proper draft version of
20 * file descriptor passing hacked up from
22 * Marty Leisner : Fixes to fd passing
23 * Nick Nevin : recvmsg bugfix.
24 * Alan Cox : Started proper garbage collector
25 * Heiko EiBfeldt : Missing verify_area check
26 * Alan Cox : Started POSIXisms
27 * Andreas Schwab : Replace inode by dentry for proper
29 * Kirk Petersen : Made this a module
30 * Christoph Rohland : Elegant non-blocking accept/connect algorithm.
32 * Alexey Kuznetosv : Repaired (I hope) bugs introduces
33 * by above two patches.
34 * Andrea Arcangeli : If possible we block in connect(2)
35 * if the max backlog of the listen socket
36 * is been reached. This won't break
37 * old apps and it will avoid huge amount
38 * of socks hashed (this for unix_gc()
39 * performances reasons).
40 * Security fix that limits the max
41 * number of socks to 2*max_files and
42 * the number of skb queueable in the
44 * Artur Skawina : Hash function optimizations
45 * Alexey Kuznetsov : Full scale SMP. Lot of bugs are introduced 8)
46 * Malcolm Beattie : Set peercred for socketpair
47 * Michal Ostrowski : Module initialization cleanup.
48 * Arnaldo C. Melo : Remove MOD_{INC,DEC}_USE_COUNT,
49 * the core infrastructure is doing that
50 * for all net proto families now (2.5.69+)
53 * Known differences from reference BSD that was tested:
56 * ECONNREFUSED is not returned from one end of a connected() socket to the
57 * other the moment one end closes.
58 * fstat() doesn't return st_dev=0, and give the blksize as high water mark
59 * and a fake inode identifier (nor the BSD first socket fstat twice bug).
61 * accept() returns a path name even if the connecting socket has closed
62 * in the meantime (BSD loses the path and gives up).
63 * accept() returns 0 length path for an unbound connector. BSD returns 16
64 * and a null first byte in the path (but not for gethost/peername - BSD bug ??)
65 * socketpair(...SOCK_RAW..) doesn't panic the kernel.
66 * BSD af_unix apparently has connect forgetting to block properly.
67 * (need to check this with the POSIX spec in detail)
69 * Differences from 2.0.0-11-... (ANK)
70 * Bug fixes and improvements.
71 * - client shutdown killed server socket.
72 * - removed all useless cli/sti pairs.
74 * Semantic changes/extensions.
75 * - generic control message passing.
76 * - SCM_CREDENTIALS control message.
77 * - "Abstract" (not FS based) socket bindings.
78 * Abstract names are sequences of bytes (not zero terminated)
79 * started by 0, so that this name space does not intersect
83 #include <linux/module.h>
84 #include <linux/kernel.h>
85 #include <linux/signal.h>
86 #include <linux/sched.h>
87 #include <linux/errno.h>
88 #include <linux/string.h>
89 #include <linux/stat.h>
90 #include <linux/dcache.h>
91 #include <linux/namei.h>
92 #include <linux/socket.h>
94 #include <linux/fcntl.h>
95 #include <linux/termios.h>
96 #include <linux/sockios.h>
97 #include <linux/net.h>
100 #include <linux/slab.h>
101 #include <asm/uaccess.h>
102 #include <linux/skbuff.h>
103 #include <linux/netdevice.h>
104 #include <net/net_namespace.h>
105 #include <net/sock.h>
106 #include <net/tcp_states.h>
107 #include <net/af_unix.h>
108 #include <linux/proc_fs.h>
109 #include <linux/seq_file.h>
111 #include <linux/init.h>
112 #include <linux/poll.h>
113 #include <linux/rtnetlink.h>
114 #include <linux/mount.h>
115 #include <net/checksum.h>
116 #include <linux/security.h>
117 #include <linux/freezer.h>
119 struct hlist_head unix_socket_table[2 * UNIX_HASH_SIZE];
120 EXPORT_SYMBOL_GPL(unix_socket_table);
121 DEFINE_SPINLOCK(unix_table_lock);
122 EXPORT_SYMBOL_GPL(unix_table_lock);
123 static atomic_long_t unix_nr_socks;
126 static struct hlist_head *unix_sockets_unbound(void *addr)
128 unsigned long hash = (unsigned long)addr;
132 hash %= UNIX_HASH_SIZE;
133 return &unix_socket_table[UNIX_HASH_SIZE + hash];
136 #define UNIX_ABSTRACT(sk) (unix_sk(sk)->addr->hash < UNIX_HASH_SIZE)
138 #ifdef CONFIG_SECURITY_NETWORK
139 static void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb)
141 memcpy(UNIXSID(skb), &scm->secid, sizeof(u32));
144 static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb)
146 scm->secid = *UNIXSID(skb);
149 static inline void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb)
152 static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb)
154 #endif /* CONFIG_SECURITY_NETWORK */
157 * SMP locking strategy:
158 * hash table is protected with spinlock unix_table_lock
159 * each socket state is protected by separate spin lock.
162 static inline unsigned int unix_hash_fold(__wsum n)
164 unsigned int hash = (__force unsigned int)n;
168 return hash&(UNIX_HASH_SIZE-1);
171 #define unix_peer(sk) (unix_sk(sk)->peer)
173 static inline int unix_our_peer(struct sock *sk, struct sock *osk)
175 return unix_peer(osk) == sk;
178 static inline int unix_may_send(struct sock *sk, struct sock *osk)
180 return unix_peer(osk) == NULL || unix_our_peer(sk, osk);
183 static inline int unix_recvq_full(struct sock const *sk)
185 return skb_queue_len(&sk->sk_receive_queue) > sk->sk_max_ack_backlog;
188 struct sock *unix_peer_get(struct sock *s)
196 unix_state_unlock(s);
199 EXPORT_SYMBOL_GPL(unix_peer_get);
201 static inline void unix_release_addr(struct unix_address *addr)
203 if (atomic_dec_and_test(&addr->refcnt))
208 * Check unix socket name:
209 * - should be not zero length.
210 * - if started by not zero, should be NULL terminated (FS object)
211 * - if started by zero, it is abstract name.
214 static int unix_mkname(struct sockaddr_un *sunaddr, int len, unsigned int *hashp)
216 if (len <= sizeof(short) || len > sizeof(*sunaddr))
218 if (!sunaddr || sunaddr->sun_family != AF_UNIX)
220 if (sunaddr->sun_path[0]) {
222 * This may look like an off by one error but it is a bit more
223 * subtle. 108 is the longest valid AF_UNIX path for a binding.
224 * sun_path[108] doesn't as such exist. However in kernel space
225 * we are guaranteed that it is a valid memory location in our
226 * kernel address buffer.
228 ((char *)sunaddr)[len] = 0;
229 len = strlen(sunaddr->sun_path)+1+sizeof(short);
233 *hashp = unix_hash_fold(csum_partial(sunaddr, len, 0));
237 static void __unix_remove_socket(struct sock *sk)
239 sk_del_node_init(sk);
242 static void __unix_insert_socket(struct hlist_head *list, struct sock *sk)
244 WARN_ON(!sk_unhashed(sk));
245 sk_add_node(sk, list);
248 static inline void unix_remove_socket(struct sock *sk)
250 spin_lock(&unix_table_lock);
251 __unix_remove_socket(sk);
252 spin_unlock(&unix_table_lock);
255 static inline void unix_insert_socket(struct hlist_head *list, struct sock *sk)
257 spin_lock(&unix_table_lock);
258 __unix_insert_socket(list, sk);
259 spin_unlock(&unix_table_lock);
262 static struct sock *__unix_find_socket_byname(struct net *net,
263 struct sockaddr_un *sunname,
264 int len, int type, unsigned int hash)
268 sk_for_each(s, &unix_socket_table[hash ^ type]) {
269 struct unix_sock *u = unix_sk(s);
271 if (!net_eq(sock_net(s), net))
274 if (u->addr->len == len &&
275 !memcmp(u->addr->name, sunname, len))
283 static inline struct sock *unix_find_socket_byname(struct net *net,
284 struct sockaddr_un *sunname,
290 spin_lock(&unix_table_lock);
291 s = __unix_find_socket_byname(net, sunname, len, type, hash);
294 spin_unlock(&unix_table_lock);
298 static struct sock *unix_find_socket_byinode(struct inode *i)
302 spin_lock(&unix_table_lock);
304 &unix_socket_table[i->i_ino & (UNIX_HASH_SIZE - 1)]) {
305 struct dentry *dentry = unix_sk(s)->path.dentry;
307 if (dentry && dentry->d_inode == i) {
314 spin_unlock(&unix_table_lock);
318 static inline int unix_writable(struct sock *sk)
320 return (atomic_read(&sk->sk_wmem_alloc) << 2) <= sk->sk_sndbuf;
323 static void unix_write_space(struct sock *sk)
325 struct socket_wq *wq;
328 if (unix_writable(sk)) {
329 wq = rcu_dereference(sk->sk_wq);
330 if (wq_has_sleeper(wq))
331 wake_up_interruptible_sync_poll(&wq->wait,
332 POLLOUT | POLLWRNORM | POLLWRBAND);
333 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
338 /* When dgram socket disconnects (or changes its peer), we clear its receive
339 * queue of packets arrived from previous peer. First, it allows to do
340 * flow control based only on wmem_alloc; second, sk connected to peer
341 * may receive messages only from that peer. */
342 static void unix_dgram_disconnected(struct sock *sk, struct sock *other)
344 if (!skb_queue_empty(&sk->sk_receive_queue)) {
345 skb_queue_purge(&sk->sk_receive_queue);
346 wake_up_interruptible_all(&unix_sk(sk)->peer_wait);
348 /* If one link of bidirectional dgram pipe is disconnected,
349 * we signal error. Messages are lost. Do not make this,
350 * when peer was not connected to us.
352 if (!sock_flag(other, SOCK_DEAD) && unix_peer(other) == sk) {
353 other->sk_err = ECONNRESET;
354 other->sk_error_report(other);
359 static void unix_sock_destructor(struct sock *sk)
361 struct unix_sock *u = unix_sk(sk);
363 skb_queue_purge(&sk->sk_receive_queue);
365 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
366 WARN_ON(!sk_unhashed(sk));
367 WARN_ON(sk->sk_socket);
368 if (!sock_flag(sk, SOCK_DEAD)) {
369 printk(KERN_INFO "Attempt to release alive unix socket: %p\n", sk);
374 unix_release_addr(u->addr);
376 atomic_long_dec(&unix_nr_socks);
378 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
380 #ifdef UNIX_REFCNT_DEBUG
381 printk(KERN_DEBUG "UNIX %p is destroyed, %ld are still alive.\n", sk,
382 atomic_long_read(&unix_nr_socks));
386 static void unix_release_sock(struct sock *sk, int embrion)
388 struct unix_sock *u = unix_sk(sk);
394 unix_remove_socket(sk);
399 sk->sk_shutdown = SHUTDOWN_MASK;
401 u->path.dentry = NULL;
403 state = sk->sk_state;
404 sk->sk_state = TCP_CLOSE;
405 unix_state_unlock(sk);
407 wake_up_interruptible_all(&u->peer_wait);
409 skpair = unix_peer(sk);
411 if (skpair != NULL) {
412 if (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) {
413 unix_state_lock(skpair);
415 skpair->sk_shutdown = SHUTDOWN_MASK;
416 if (!skb_queue_empty(&sk->sk_receive_queue) || embrion)
417 skpair->sk_err = ECONNRESET;
418 unix_state_unlock(skpair);
419 skpair->sk_state_change(skpair);
420 sk_wake_async(skpair, SOCK_WAKE_WAITD, POLL_HUP);
422 sock_put(skpair); /* It may now die */
423 unix_peer(sk) = NULL;
426 /* Try to flush out this socket. Throw out buffers at least */
428 while ((skb = skb_dequeue(&sk->sk_receive_queue)) != NULL) {
429 if (state == TCP_LISTEN)
430 unix_release_sock(skb->sk, 1);
431 /* passed fds are erased in the kfree_skb hook */
440 /* ---- Socket is dead now and most probably destroyed ---- */
443 * Fixme: BSD difference: In BSD all sockets connected to us get
444 * ECONNRESET and we die on the spot. In Linux we behave
445 * like files and pipes do and wait for the last
448 * Can't we simply set sock->err?
450 * What the above comment does talk about? --ANK(980817)
453 if (unix_tot_inflight)
454 unix_gc(); /* Garbage collect fds */
457 static void init_peercred(struct sock *sk)
459 put_pid(sk->sk_peer_pid);
460 if (sk->sk_peer_cred)
461 put_cred(sk->sk_peer_cred);
462 sk->sk_peer_pid = get_pid(task_tgid(current));
463 sk->sk_peer_cred = get_current_cred();
466 static void copy_peercred(struct sock *sk, struct sock *peersk)
468 put_pid(sk->sk_peer_pid);
469 if (sk->sk_peer_cred)
470 put_cred(sk->sk_peer_cred);
471 sk->sk_peer_pid = get_pid(peersk->sk_peer_pid);
472 sk->sk_peer_cred = get_cred(peersk->sk_peer_cred);
475 static int unix_listen(struct socket *sock, int backlog)
478 struct sock *sk = sock->sk;
479 struct unix_sock *u = unix_sk(sk);
480 struct pid *old_pid = NULL;
483 if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
484 goto out; /* Only stream/seqpacket sockets accept */
487 goto out; /* No listens on an unbound socket */
489 if (sk->sk_state != TCP_CLOSE && sk->sk_state != TCP_LISTEN)
491 if (backlog > sk->sk_max_ack_backlog)
492 wake_up_interruptible_all(&u->peer_wait);
493 sk->sk_max_ack_backlog = backlog;
494 sk->sk_state = TCP_LISTEN;
495 /* set credentials so connect can copy them */
500 unix_state_unlock(sk);
506 static int unix_release(struct socket *);
507 static int unix_bind(struct socket *, struct sockaddr *, int);
508 static int unix_stream_connect(struct socket *, struct sockaddr *,
509 int addr_len, int flags);
510 static int unix_socketpair(struct socket *, struct socket *);
511 static int unix_accept(struct socket *, struct socket *, int);
512 static int unix_getname(struct socket *, struct sockaddr *, int *, int);
513 static unsigned int unix_poll(struct file *, struct socket *, poll_table *);
514 static unsigned int unix_dgram_poll(struct file *, struct socket *,
516 static int unix_ioctl(struct socket *, unsigned int, unsigned long);
517 static int unix_shutdown(struct socket *, int);
518 static int unix_stream_sendmsg(struct kiocb *, struct socket *,
519 struct msghdr *, size_t);
520 static int unix_stream_recvmsg(struct kiocb *, struct socket *,
521 struct msghdr *, size_t, int);
522 static int unix_dgram_sendmsg(struct kiocb *, struct socket *,
523 struct msghdr *, size_t);
524 static int unix_dgram_recvmsg(struct kiocb *, struct socket *,
525 struct msghdr *, size_t, int);
526 static int unix_dgram_connect(struct socket *, struct sockaddr *,
528 static int unix_seqpacket_sendmsg(struct kiocb *, struct socket *,
529 struct msghdr *, size_t);
530 static int unix_seqpacket_recvmsg(struct kiocb *, struct socket *,
531 struct msghdr *, size_t, int);
533 static int unix_set_peek_off(struct sock *sk, int val)
535 struct unix_sock *u = unix_sk(sk);
537 if (mutex_lock_interruptible(&u->readlock))
540 sk->sk_peek_off = val;
541 mutex_unlock(&u->readlock);
547 static const struct proto_ops unix_stream_ops = {
549 .owner = THIS_MODULE,
550 .release = unix_release,
552 .connect = unix_stream_connect,
553 .socketpair = unix_socketpair,
554 .accept = unix_accept,
555 .getname = unix_getname,
558 .listen = unix_listen,
559 .shutdown = unix_shutdown,
560 .setsockopt = sock_no_setsockopt,
561 .getsockopt = sock_no_getsockopt,
562 .sendmsg = unix_stream_sendmsg,
563 .recvmsg = unix_stream_recvmsg,
564 .mmap = sock_no_mmap,
565 .sendpage = sock_no_sendpage,
566 .set_peek_off = unix_set_peek_off,
569 static const struct proto_ops unix_dgram_ops = {
571 .owner = THIS_MODULE,
572 .release = unix_release,
574 .connect = unix_dgram_connect,
575 .socketpair = unix_socketpair,
576 .accept = sock_no_accept,
577 .getname = unix_getname,
578 .poll = unix_dgram_poll,
580 .listen = sock_no_listen,
581 .shutdown = unix_shutdown,
582 .setsockopt = sock_no_setsockopt,
583 .getsockopt = sock_no_getsockopt,
584 .sendmsg = unix_dgram_sendmsg,
585 .recvmsg = unix_dgram_recvmsg,
586 .mmap = sock_no_mmap,
587 .sendpage = sock_no_sendpage,
588 .set_peek_off = unix_set_peek_off,
591 static const struct proto_ops unix_seqpacket_ops = {
593 .owner = THIS_MODULE,
594 .release = unix_release,
596 .connect = unix_stream_connect,
597 .socketpair = unix_socketpair,
598 .accept = unix_accept,
599 .getname = unix_getname,
600 .poll = unix_dgram_poll,
602 .listen = unix_listen,
603 .shutdown = unix_shutdown,
604 .setsockopt = sock_no_setsockopt,
605 .getsockopt = sock_no_getsockopt,
606 .sendmsg = unix_seqpacket_sendmsg,
607 .recvmsg = unix_seqpacket_recvmsg,
608 .mmap = sock_no_mmap,
609 .sendpage = sock_no_sendpage,
610 .set_peek_off = unix_set_peek_off,
613 static struct proto unix_proto = {
615 .owner = THIS_MODULE,
616 .obj_size = sizeof(struct unix_sock),
620 * AF_UNIX sockets do not interact with hardware, hence they
621 * dont trigger interrupts - so it's safe for them to have
622 * bh-unsafe locking for their sk_receive_queue.lock. Split off
623 * this special lock-class by reinitializing the spinlock key:
625 static struct lock_class_key af_unix_sk_receive_queue_lock_key;
627 static struct sock *unix_create1(struct net *net, struct socket *sock)
629 struct sock *sk = NULL;
632 atomic_long_inc(&unix_nr_socks);
633 if (atomic_long_read(&unix_nr_socks) > 2 * get_max_files())
636 sk = sk_alloc(net, PF_UNIX, GFP_KERNEL, &unix_proto);
640 sock_init_data(sock, sk);
641 lockdep_set_class(&sk->sk_receive_queue.lock,
642 &af_unix_sk_receive_queue_lock_key);
644 sk->sk_write_space = unix_write_space;
645 sk->sk_max_ack_backlog = net->unx.sysctl_max_dgram_qlen;
646 sk->sk_destruct = unix_sock_destructor;
648 u->path.dentry = NULL;
650 spin_lock_init(&u->lock);
651 atomic_long_set(&u->inflight, 0);
652 INIT_LIST_HEAD(&u->link);
653 mutex_init(&u->readlock); /* single task reading lock */
654 init_waitqueue_head(&u->peer_wait);
655 unix_insert_socket(unix_sockets_unbound(sk), sk);
658 atomic_long_dec(&unix_nr_socks);
661 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
667 static int unix_create(struct net *net, struct socket *sock, int protocol,
670 if (protocol && protocol != PF_UNIX)
671 return -EPROTONOSUPPORT;
673 sock->state = SS_UNCONNECTED;
675 switch (sock->type) {
677 sock->ops = &unix_stream_ops;
680 * Believe it or not BSD has AF_UNIX, SOCK_RAW though
684 sock->type = SOCK_DGRAM;
686 sock->ops = &unix_dgram_ops;
689 sock->ops = &unix_seqpacket_ops;
692 return -ESOCKTNOSUPPORT;
695 return unix_create1(net, sock) ? 0 : -ENOMEM;
698 static int unix_release(struct socket *sock)
700 struct sock *sk = sock->sk;
705 unix_release_sock(sk, 0);
711 static int unix_autobind(struct socket *sock)
713 struct sock *sk = sock->sk;
714 struct net *net = sock_net(sk);
715 struct unix_sock *u = unix_sk(sk);
716 static u32 ordernum = 1;
717 struct unix_address *addr;
719 unsigned int retries = 0;
721 err = mutex_lock_interruptible(&u->readlock);
730 addr = kzalloc(sizeof(*addr) + sizeof(short) + 16, GFP_KERNEL);
734 addr->name->sun_family = AF_UNIX;
735 atomic_set(&addr->refcnt, 1);
738 addr->len = sprintf(addr->name->sun_path+1, "%05x", ordernum) + 1 + sizeof(short);
739 addr->hash = unix_hash_fold(csum_partial(addr->name, addr->len, 0));
741 spin_lock(&unix_table_lock);
742 ordernum = (ordernum+1)&0xFFFFF;
744 if (__unix_find_socket_byname(net, addr->name, addr->len, sock->type,
746 spin_unlock(&unix_table_lock);
748 * __unix_find_socket_byname() may take long time if many names
749 * are already in use.
752 /* Give up if all names seems to be in use. */
753 if (retries++ == 0xFFFFF) {
760 addr->hash ^= sk->sk_type;
762 __unix_remove_socket(sk);
764 __unix_insert_socket(&unix_socket_table[addr->hash], sk);
765 spin_unlock(&unix_table_lock);
768 out: mutex_unlock(&u->readlock);
772 static struct sock *unix_find_other(struct net *net,
773 struct sockaddr_un *sunname, int len,
774 int type, unsigned int hash, int *error)
780 if (sunname->sun_path[0]) {
782 err = kern_path(sunname->sun_path, LOOKUP_FOLLOW, &path);
785 inode = path.dentry->d_inode;
786 err = inode_permission(inode, MAY_WRITE);
791 if (!S_ISSOCK(inode->i_mode))
793 u = unix_find_socket_byinode(inode);
797 if (u->sk_type == type)
803 if (u->sk_type != type) {
809 u = unix_find_socket_byname(net, sunname, len, type, hash);
811 struct dentry *dentry;
812 dentry = unix_sk(u)->path.dentry;
814 touch_atime(&unix_sk(u)->path);
827 static int unix_mknod(const char *sun_path, umode_t mode, struct path *res)
829 struct dentry *dentry;
833 * Get the parent directory, calculate the hash for last
836 dentry = kern_path_create(AT_FDCWD, sun_path, &path, 0);
837 err = PTR_ERR(dentry);
842 * All right, let's create it.
844 err = security_path_mknod(&path, dentry, mode, 0);
846 err = vfs_mknod(path.dentry->d_inode, dentry, mode, 0);
848 res->mnt = mntget(path.mnt);
849 res->dentry = dget(dentry);
852 done_path_create(&path, dentry);
856 static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
858 struct sock *sk = sock->sk;
859 struct net *net = sock_net(sk);
860 struct unix_sock *u = unix_sk(sk);
861 struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
862 char *sun_path = sunaddr->sun_path;
865 struct unix_address *addr;
866 struct hlist_head *list;
869 if (sunaddr->sun_family != AF_UNIX)
872 if (addr_len == sizeof(short)) {
873 err = unix_autobind(sock);
877 err = unix_mkname(sunaddr, addr_len, &hash);
882 err = mutex_lock_interruptible(&u->readlock);
891 addr = kmalloc(sizeof(*addr)+addr_len, GFP_KERNEL);
895 memcpy(addr->name, sunaddr, addr_len);
896 addr->len = addr_len;
897 addr->hash = hash ^ sk->sk_type;
898 atomic_set(&addr->refcnt, 1);
902 umode_t mode = S_IFSOCK |
903 (SOCK_INODE(sock)->i_mode & ~current_umask());
904 err = unix_mknod(sun_path, mode, &path);
908 unix_release_addr(addr);
911 addr->hash = UNIX_HASH_SIZE;
912 hash = path.dentry->d_inode->i_ino & (UNIX_HASH_SIZE-1);
913 spin_lock(&unix_table_lock);
915 list = &unix_socket_table[hash];
917 spin_lock(&unix_table_lock);
919 if (__unix_find_socket_byname(net, sunaddr, addr_len,
920 sk->sk_type, hash)) {
921 unix_release_addr(addr);
925 list = &unix_socket_table[addr->hash];
929 __unix_remove_socket(sk);
931 __unix_insert_socket(list, sk);
934 spin_unlock(&unix_table_lock);
936 mutex_unlock(&u->readlock);
941 static void unix_state_double_lock(struct sock *sk1, struct sock *sk2)
943 if (unlikely(sk1 == sk2) || !sk2) {
944 unix_state_lock(sk1);
948 unix_state_lock(sk1);
949 unix_state_lock_nested(sk2);
951 unix_state_lock(sk2);
952 unix_state_lock_nested(sk1);
956 static void unix_state_double_unlock(struct sock *sk1, struct sock *sk2)
958 if (unlikely(sk1 == sk2) || !sk2) {
959 unix_state_unlock(sk1);
962 unix_state_unlock(sk1);
963 unix_state_unlock(sk2);
966 static int unix_dgram_connect(struct socket *sock, struct sockaddr *addr,
969 struct sock *sk = sock->sk;
970 struct net *net = sock_net(sk);
971 struct sockaddr_un *sunaddr = (struct sockaddr_un *)addr;
976 if (addr->sa_family != AF_UNSPEC) {
977 err = unix_mkname(sunaddr, alen, &hash);
982 if (test_bit(SOCK_PASSCRED, &sock->flags) &&
983 !unix_sk(sk)->addr && (err = unix_autobind(sock)) != 0)
987 other = unix_find_other(net, sunaddr, alen, sock->type, hash, &err);
991 unix_state_double_lock(sk, other);
993 /* Apparently VFS overslept socket death. Retry. */
994 if (sock_flag(other, SOCK_DEAD)) {
995 unix_state_double_unlock(sk, other);
1001 if (!unix_may_send(sk, other))
1004 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1010 * 1003.1g breaking connected state with AF_UNSPEC
1013 unix_state_double_lock(sk, other);
1017 * If it was connected, reconnect.
1019 if (unix_peer(sk)) {
1020 struct sock *old_peer = unix_peer(sk);
1021 unix_peer(sk) = other;
1022 unix_state_double_unlock(sk, other);
1024 if (other != old_peer)
1025 unix_dgram_disconnected(sk, old_peer);
1028 unix_peer(sk) = other;
1029 unix_state_double_unlock(sk, other);
1034 unix_state_double_unlock(sk, other);
1040 static long unix_wait_for_peer(struct sock *other, long timeo)
1042 struct unix_sock *u = unix_sk(other);
1046 prepare_to_wait_exclusive(&u->peer_wait, &wait, TASK_INTERRUPTIBLE);
1048 sched = !sock_flag(other, SOCK_DEAD) &&
1049 !(other->sk_shutdown & RCV_SHUTDOWN) &&
1050 unix_recvq_full(other);
1052 unix_state_unlock(other);
1055 timeo = schedule_timeout(timeo);
1057 finish_wait(&u->peer_wait, &wait);
1061 static int unix_stream_connect(struct socket *sock, struct sockaddr *uaddr,
1062 int addr_len, int flags)
1064 struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
1065 struct sock *sk = sock->sk;
1066 struct net *net = sock_net(sk);
1067 struct unix_sock *u = unix_sk(sk), *newu, *otheru;
1068 struct sock *newsk = NULL;
1069 struct sock *other = NULL;
1070 struct sk_buff *skb = NULL;
1076 err = unix_mkname(sunaddr, addr_len, &hash);
1081 if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr &&
1082 (err = unix_autobind(sock)) != 0)
1085 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
1087 /* First of all allocate resources.
1088 If we will make it after state is locked,
1089 we will have to recheck all again in any case.
1094 /* create new sock for complete connection */
1095 newsk = unix_create1(sock_net(sk), NULL);
1099 /* Allocate skb for sending to listening sock */
1100 skb = sock_wmalloc(newsk, 1, 0, GFP_KERNEL);
1105 /* Find listening sock. */
1106 other = unix_find_other(net, sunaddr, addr_len, sk->sk_type, hash, &err);
1110 /* Latch state of peer */
1111 unix_state_lock(other);
1113 /* Apparently VFS overslept socket death. Retry. */
1114 if (sock_flag(other, SOCK_DEAD)) {
1115 unix_state_unlock(other);
1120 err = -ECONNREFUSED;
1121 if (other->sk_state != TCP_LISTEN)
1123 if (other->sk_shutdown & RCV_SHUTDOWN)
1126 if (unix_recvq_full(other)) {
1131 timeo = unix_wait_for_peer(other, timeo);
1133 err = sock_intr_errno(timeo);
1134 if (signal_pending(current))
1142 It is tricky place. We need to grab our state lock and cannot
1143 drop lock on peer. It is dangerous because deadlock is
1144 possible. Connect to self case and simultaneous
1145 attempt to connect are eliminated by checking socket
1146 state. other is TCP_LISTEN, if sk is TCP_LISTEN we
1147 check this before attempt to grab lock.
1149 Well, and we have to recheck the state after socket locked.
1155 /* This is ok... continue with connect */
1157 case TCP_ESTABLISHED:
1158 /* Socket is already connected */
1166 unix_state_lock_nested(sk);
1168 if (sk->sk_state != st) {
1169 unix_state_unlock(sk);
1170 unix_state_unlock(other);
1175 err = security_unix_stream_connect(sk, other, newsk);
1177 unix_state_unlock(sk);
1181 /* The way is open! Fastly set all the necessary fields... */
1184 unix_peer(newsk) = sk;
1185 newsk->sk_state = TCP_ESTABLISHED;
1186 newsk->sk_type = sk->sk_type;
1187 init_peercred(newsk);
1188 newu = unix_sk(newsk);
1189 RCU_INIT_POINTER(newsk->sk_wq, &newu->peer_wq);
1190 otheru = unix_sk(other);
1192 /* copy address information from listening to new sock*/
1194 atomic_inc(&otheru->addr->refcnt);
1195 newu->addr = otheru->addr;
1197 if (otheru->path.dentry) {
1198 path_get(&otheru->path);
1199 newu->path = otheru->path;
1202 /* Set credentials */
1203 copy_peercred(sk, other);
1205 sock->state = SS_CONNECTED;
1206 sk->sk_state = TCP_ESTABLISHED;
1209 smp_mb__after_atomic_inc(); /* sock_hold() does an atomic_inc() */
1210 unix_peer(sk) = newsk;
1212 unix_state_unlock(sk);
1214 /* take ten and and send info to listening sock */
1215 spin_lock(&other->sk_receive_queue.lock);
1216 __skb_queue_tail(&other->sk_receive_queue, skb);
1217 spin_unlock(&other->sk_receive_queue.lock);
1218 unix_state_unlock(other);
1219 other->sk_data_ready(other, 0);
1225 unix_state_unlock(other);
1230 unix_release_sock(newsk, 0);
1236 static int unix_socketpair(struct socket *socka, struct socket *sockb)
1238 struct sock *ska = socka->sk, *skb = sockb->sk;
1240 /* Join our sockets back to back */
1243 unix_peer(ska) = skb;
1244 unix_peer(skb) = ska;
1248 if (ska->sk_type != SOCK_DGRAM) {
1249 ska->sk_state = TCP_ESTABLISHED;
1250 skb->sk_state = TCP_ESTABLISHED;
1251 socka->state = SS_CONNECTED;
1252 sockb->state = SS_CONNECTED;
1257 static void unix_sock_inherit_flags(const struct socket *old,
1260 if (test_bit(SOCK_PASSCRED, &old->flags))
1261 set_bit(SOCK_PASSCRED, &new->flags);
1262 if (test_bit(SOCK_PASSSEC, &old->flags))
1263 set_bit(SOCK_PASSSEC, &new->flags);
1266 static int unix_accept(struct socket *sock, struct socket *newsock, int flags)
1268 struct sock *sk = sock->sk;
1270 struct sk_buff *skb;
1274 if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
1278 if (sk->sk_state != TCP_LISTEN)
1281 /* If socket state is TCP_LISTEN it cannot change (for now...),
1282 * so that no locks are necessary.
1285 skb = skb_recv_datagram(sk, 0, flags&O_NONBLOCK, &err);
1287 /* This means receive shutdown. */
1294 skb_free_datagram(sk, skb);
1295 wake_up_interruptible(&unix_sk(sk)->peer_wait);
1297 /* attach accepted sock to socket */
1298 unix_state_lock(tsk);
1299 newsock->state = SS_CONNECTED;
1300 unix_sock_inherit_flags(sock, newsock);
1301 sock_graft(tsk, newsock);
1302 unix_state_unlock(tsk);
1310 static int unix_getname(struct socket *sock, struct sockaddr *uaddr, int *uaddr_len, int peer)
1312 struct sock *sk = sock->sk;
1313 struct unix_sock *u;
1314 DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr, uaddr);
1318 sk = unix_peer_get(sk);
1329 unix_state_lock(sk);
1331 sunaddr->sun_family = AF_UNIX;
1332 sunaddr->sun_path[0] = 0;
1333 *uaddr_len = sizeof(short);
1335 struct unix_address *addr = u->addr;
1337 *uaddr_len = addr->len;
1338 memcpy(sunaddr, addr->name, *uaddr_len);
1340 unix_state_unlock(sk);
1346 static void unix_detach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1350 scm->fp = UNIXCB(skb).fp;
1351 UNIXCB(skb).fp = NULL;
1353 for (i = scm->fp->count-1; i >= 0; i--)
1354 unix_notinflight(scm->fp->fp[i]);
1357 static void unix_destruct_scm(struct sk_buff *skb)
1359 struct scm_cookie scm;
1360 memset(&scm, 0, sizeof(scm));
1361 scm.pid = UNIXCB(skb).pid;
1363 unix_detach_fds(&scm, skb);
1365 /* Alas, it calls VFS */
1366 /* So fscking what? fput() had been SMP-safe since the last Summer */
1371 #define MAX_RECURSION_LEVEL 4
1373 static int unix_attach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1376 unsigned char max_level = 0;
1377 int unix_sock_count = 0;
1379 for (i = scm->fp->count - 1; i >= 0; i--) {
1380 struct sock *sk = unix_get_socket(scm->fp->fp[i]);
1384 max_level = max(max_level,
1385 unix_sk(sk)->recursion_level);
1388 if (unlikely(max_level > MAX_RECURSION_LEVEL))
1389 return -ETOOMANYREFS;
1392 * Need to duplicate file references for the sake of garbage
1393 * collection. Otherwise a socket in the fps might become a
1394 * candidate for GC while the skb is not yet queued.
1396 UNIXCB(skb).fp = scm_fp_dup(scm->fp);
1397 if (!UNIXCB(skb).fp)
1400 if (unix_sock_count) {
1401 for (i = scm->fp->count - 1; i >= 0; i--)
1402 unix_inflight(scm->fp->fp[i]);
1407 static int unix_scm_to_skb(struct scm_cookie *scm, struct sk_buff *skb, bool send_fds)
1411 UNIXCB(skb).pid = get_pid(scm->pid);
1412 UNIXCB(skb).uid = scm->creds.uid;
1413 UNIXCB(skb).gid = scm->creds.gid;
1414 UNIXCB(skb).fp = NULL;
1415 if (scm->fp && send_fds)
1416 err = unix_attach_fds(scm, skb);
1418 skb->destructor = unix_destruct_scm;
1423 * Some apps rely on write() giving SCM_CREDENTIALS
1424 * We include credentials if source or destination socket
1425 * asserted SOCK_PASSCRED.
1427 static void maybe_add_creds(struct sk_buff *skb, const struct socket *sock,
1428 const struct sock *other)
1430 if (UNIXCB(skb).pid)
1432 if (test_bit(SOCK_PASSCRED, &sock->flags) ||
1433 !other->sk_socket ||
1434 test_bit(SOCK_PASSCRED, &other->sk_socket->flags)) {
1435 UNIXCB(skb).pid = get_pid(task_tgid(current));
1436 current_uid_gid(&UNIXCB(skb).uid, &UNIXCB(skb).gid);
1441 * Send AF_UNIX data.
1444 static int unix_dgram_sendmsg(struct kiocb *kiocb, struct socket *sock,
1445 struct msghdr *msg, size_t len)
1447 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1448 struct sock *sk = sock->sk;
1449 struct net *net = sock_net(sk);
1450 struct unix_sock *u = unix_sk(sk);
1451 struct sockaddr_un *sunaddr = msg->msg_name;
1452 struct sock *other = NULL;
1453 int namelen = 0; /* fake GCC */
1456 struct sk_buff *skb;
1458 struct scm_cookie tmp_scm;
1462 if (NULL == siocb->scm)
1463 siocb->scm = &tmp_scm;
1465 err = scm_send(sock, msg, siocb->scm, false);
1470 if (msg->msg_flags&MSG_OOB)
1473 if (msg->msg_namelen) {
1474 err = unix_mkname(sunaddr, msg->msg_namelen, &hash);
1481 other = unix_peer_get(sk);
1486 if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr
1487 && (err = unix_autobind(sock)) != 0)
1491 if (len > sk->sk_sndbuf - 32)
1494 if (len > SKB_MAX_ALLOC)
1495 data_len = min_t(size_t,
1496 len - SKB_MAX_ALLOC,
1497 MAX_SKB_FRAGS * PAGE_SIZE);
1499 skb = sock_alloc_send_pskb(sk, len - data_len, data_len,
1500 msg->msg_flags & MSG_DONTWAIT, &err,
1501 PAGE_ALLOC_COSTLY_ORDER);
1505 err = unix_scm_to_skb(siocb->scm, skb, true);
1508 max_level = err + 1;
1509 unix_get_secdata(siocb->scm, skb);
1511 skb_put(skb, len - data_len);
1512 skb->data_len = data_len;
1514 err = skb_copy_datagram_from_iovec(skb, 0, msg->msg_iov, 0, len);
1518 timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1523 if (sunaddr == NULL)
1526 other = unix_find_other(net, sunaddr, namelen, sk->sk_type,
1532 if (sk_filter(other, skb) < 0) {
1533 /* Toss the packet but do not return any error to the sender */
1538 unix_state_lock(other);
1540 if (!unix_may_send(sk, other))
1543 if (sock_flag(other, SOCK_DEAD)) {
1545 * Check with 1003.1g - what should
1548 unix_state_unlock(other);
1552 unix_state_lock(sk);
1553 if (unix_peer(sk) == other) {
1554 unix_peer(sk) = NULL;
1555 unix_state_unlock(sk);
1557 unix_dgram_disconnected(sk, other);
1559 err = -ECONNREFUSED;
1561 unix_state_unlock(sk);
1571 if (other->sk_shutdown & RCV_SHUTDOWN)
1574 if (sk->sk_type != SOCK_SEQPACKET) {
1575 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1580 if (unix_peer(other) != sk && unix_recvq_full(other)) {
1586 timeo = unix_wait_for_peer(other, timeo);
1588 err = sock_intr_errno(timeo);
1589 if (signal_pending(current))
1595 if (sock_flag(other, SOCK_RCVTSTAMP))
1596 __net_timestamp(skb);
1597 maybe_add_creds(skb, sock, other);
1598 skb_queue_tail(&other->sk_receive_queue, skb);
1599 if (max_level > unix_sk(other)->recursion_level)
1600 unix_sk(other)->recursion_level = max_level;
1601 unix_state_unlock(other);
1602 other->sk_data_ready(other, len);
1604 scm_destroy(siocb->scm);
1608 unix_state_unlock(other);
1614 scm_destroy(siocb->scm);
1618 /* We use paged skbs for stream sockets, and limit occupancy to 32768
1619 * bytes, and a minimun of a full page.
1621 #define UNIX_SKB_FRAGS_SZ (PAGE_SIZE << get_order(32768))
1623 static int unix_stream_sendmsg(struct kiocb *kiocb, struct socket *sock,
1624 struct msghdr *msg, size_t len)
1626 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1627 struct sock *sk = sock->sk;
1628 struct sock *other = NULL;
1630 struct sk_buff *skb;
1632 struct scm_cookie tmp_scm;
1633 bool fds_sent = false;
1637 if (NULL == siocb->scm)
1638 siocb->scm = &tmp_scm;
1640 err = scm_send(sock, msg, siocb->scm, false);
1645 if (msg->msg_flags&MSG_OOB)
1648 if (msg->msg_namelen) {
1649 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
1653 other = unix_peer(sk);
1658 if (sk->sk_shutdown & SEND_SHUTDOWN)
1661 while (sent < len) {
1664 /* Keep two messages in the pipe so it schedules better */
1665 size = min_t(int, size, (sk->sk_sndbuf >> 1) - 64);
1667 /* allow fallback to order-0 allocations */
1668 size = min_t(int, size, SKB_MAX_HEAD(0) + UNIX_SKB_FRAGS_SZ);
1670 data_len = max_t(int, 0, size - SKB_MAX_HEAD(0));
1672 skb = sock_alloc_send_pskb(sk, size - data_len, data_len,
1673 msg->msg_flags & MSG_DONTWAIT, &err,
1674 get_order(UNIX_SKB_FRAGS_SZ));
1678 /* Only send the fds in the first buffer */
1679 err = unix_scm_to_skb(siocb->scm, skb, !fds_sent);
1684 max_level = err + 1;
1687 skb_put(skb, size - data_len);
1688 skb->data_len = data_len;
1690 err = skb_copy_datagram_from_iovec(skb, 0, msg->msg_iov,
1697 unix_state_lock(other);
1699 if (sock_flag(other, SOCK_DEAD) ||
1700 (other->sk_shutdown & RCV_SHUTDOWN))
1703 maybe_add_creds(skb, sock, other);
1704 skb_queue_tail(&other->sk_receive_queue, skb);
1705 if (max_level > unix_sk(other)->recursion_level)
1706 unix_sk(other)->recursion_level = max_level;
1707 unix_state_unlock(other);
1708 other->sk_data_ready(other, size);
1712 scm_destroy(siocb->scm);
1718 unix_state_unlock(other);
1721 if (sent == 0 && !(msg->msg_flags&MSG_NOSIGNAL))
1722 send_sig(SIGPIPE, current, 0);
1725 scm_destroy(siocb->scm);
1727 return sent ? : err;
1730 static int unix_seqpacket_sendmsg(struct kiocb *kiocb, struct socket *sock,
1731 struct msghdr *msg, size_t len)
1734 struct sock *sk = sock->sk;
1736 err = sock_error(sk);
1740 if (sk->sk_state != TCP_ESTABLISHED)
1743 if (msg->msg_namelen)
1744 msg->msg_namelen = 0;
1746 return unix_dgram_sendmsg(kiocb, sock, msg, len);
1749 static int unix_seqpacket_recvmsg(struct kiocb *iocb, struct socket *sock,
1750 struct msghdr *msg, size_t size,
1753 struct sock *sk = sock->sk;
1755 if (sk->sk_state != TCP_ESTABLISHED)
1758 return unix_dgram_recvmsg(iocb, sock, msg, size, flags);
1761 static void unix_copy_addr(struct msghdr *msg, struct sock *sk)
1763 struct unix_sock *u = unix_sk(sk);
1766 msg->msg_namelen = u->addr->len;
1767 memcpy(msg->msg_name, u->addr->name, u->addr->len);
1771 static int unix_dgram_recvmsg(struct kiocb *iocb, struct socket *sock,
1772 struct msghdr *msg, size_t size,
1775 struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1776 struct scm_cookie tmp_scm;
1777 struct sock *sk = sock->sk;
1778 struct unix_sock *u = unix_sk(sk);
1779 int noblock = flags & MSG_DONTWAIT;
1780 struct sk_buff *skb;
1788 err = mutex_lock_interruptible(&u->readlock);
1790 err = sock_intr_errno(sock_rcvtimeo(sk, noblock));
1794 skip = sk_peek_offset(sk, flags);
1796 skb = __skb_recv_datagram(sk, flags, &peeked, &skip, &err);
1798 unix_state_lock(sk);
1799 /* Signal EOF on disconnected non-blocking SEQPACKET socket. */
1800 if (sk->sk_type == SOCK_SEQPACKET && err == -EAGAIN &&
1801 (sk->sk_shutdown & RCV_SHUTDOWN))
1803 unix_state_unlock(sk);
1807 wake_up_interruptible_sync_poll(&u->peer_wait,
1808 POLLOUT | POLLWRNORM | POLLWRBAND);
1811 unix_copy_addr(msg, skb->sk);
1813 if (size > skb->len - skip)
1814 size = skb->len - skip;
1815 else if (size < skb->len - skip)
1816 msg->msg_flags |= MSG_TRUNC;
1818 err = skb_copy_datagram_iovec(skb, skip, msg->msg_iov, size);
1822 if (sock_flag(sk, SOCK_RCVTSTAMP))
1823 __sock_recv_timestamp(msg, sk, skb);
1826 siocb->scm = &tmp_scm;
1827 memset(&tmp_scm, 0, sizeof(tmp_scm));
1829 scm_set_cred(siocb->scm, UNIXCB(skb).pid, UNIXCB(skb).uid, UNIXCB(skb).gid);
1830 unix_set_secdata(siocb->scm, skb);
1832 if (!(flags & MSG_PEEK)) {
1834 unix_detach_fds(siocb->scm, skb);
1836 sk_peek_offset_bwd(sk, skb->len);
1838 /* It is questionable: on PEEK we could:
1839 - do not return fds - good, but too simple 8)
1840 - return fds, and do not return them on read (old strategy,
1842 - clone fds (I chose it for now, it is the most universal
1845 POSIX 1003.1g does not actually define this clearly
1846 at all. POSIX 1003.1g doesn't define a lot of things
1851 sk_peek_offset_fwd(sk, size);
1854 siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1856 err = (flags & MSG_TRUNC) ? skb->len - skip : size;
1858 scm_recv(sock, msg, siocb->scm, flags);
1861 skb_free_datagram(sk, skb);
1863 mutex_unlock(&u->readlock);
1869 * Sleep until more data has arrived. But check for races..
1871 static long unix_stream_data_wait(struct sock *sk, long timeo,
1872 struct sk_buff *last)
1876 unix_state_lock(sk);
1879 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1881 if (skb_peek_tail(&sk->sk_receive_queue) != last ||
1883 (sk->sk_shutdown & RCV_SHUTDOWN) ||
1884 signal_pending(current) ||
1888 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1889 unix_state_unlock(sk);
1890 timeo = freezable_schedule_timeout(timeo);
1891 unix_state_lock(sk);
1892 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1895 finish_wait(sk_sleep(sk), &wait);
1896 unix_state_unlock(sk);
1900 static unsigned int unix_skb_len(const struct sk_buff *skb)
1902 return skb->len - UNIXCB(skb).consumed;
1905 static int unix_stream_recvmsg(struct kiocb *iocb, struct socket *sock,
1906 struct msghdr *msg, size_t size,
1909 struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1910 struct scm_cookie tmp_scm;
1911 struct sock *sk = sock->sk;
1912 struct unix_sock *u = unix_sk(sk);
1913 struct sockaddr_un *sunaddr = msg->msg_name;
1915 int check_creds = 0;
1922 if (sk->sk_state != TCP_ESTABLISHED)
1929 target = sock_rcvlowat(sk, flags&MSG_WAITALL, size);
1930 timeo = sock_rcvtimeo(sk, flags&MSG_DONTWAIT);
1932 /* Lock the socket to prevent queue disordering
1933 * while sleeps in memcpy_tomsg
1937 siocb->scm = &tmp_scm;
1938 memset(&tmp_scm, 0, sizeof(tmp_scm));
1941 err = mutex_lock_interruptible(&u->readlock);
1943 err = sock_intr_errno(timeo);
1949 struct sk_buff *skb, *last;
1951 unix_state_lock(sk);
1952 last = skb = skb_peek(&sk->sk_receive_queue);
1955 unix_sk(sk)->recursion_level = 0;
1956 if (copied >= target)
1960 * POSIX 1003.1g mandates this order.
1963 err = sock_error(sk);
1966 if (sk->sk_shutdown & RCV_SHUTDOWN)
1969 unix_state_unlock(sk);
1973 mutex_unlock(&u->readlock);
1975 timeo = unix_stream_data_wait(sk, timeo, last);
1977 if (signal_pending(current)
1978 || mutex_lock_interruptible(&u->readlock)) {
1979 err = sock_intr_errno(timeo);
1985 unix_state_unlock(sk);
1989 skip = sk_peek_offset(sk, flags);
1990 while (skip >= unix_skb_len(skb)) {
1991 skip -= unix_skb_len(skb);
1993 skb = skb_peek_next(skb, &sk->sk_receive_queue);
1998 unix_state_unlock(sk);
2001 /* Never glue messages from different writers */
2002 if ((UNIXCB(skb).pid != siocb->scm->pid) ||
2003 !uid_eq(UNIXCB(skb).uid, siocb->scm->creds.uid) ||
2004 !gid_eq(UNIXCB(skb).gid, siocb->scm->creds.gid))
2006 } else if (test_bit(SOCK_PASSCRED, &sock->flags)) {
2007 /* Copy credentials */
2008 scm_set_cred(siocb->scm, UNIXCB(skb).pid, UNIXCB(skb).uid, UNIXCB(skb).gid);
2012 /* Copy address just once */
2014 unix_copy_addr(msg, skb->sk);
2018 chunk = min_t(unsigned int, unix_skb_len(skb) - skip, size);
2019 if (skb_copy_datagram_iovec(skb, UNIXCB(skb).consumed + skip,
2020 msg->msg_iov, chunk)) {
2028 /* Mark read part of skb as used */
2029 if (!(flags & MSG_PEEK)) {
2030 UNIXCB(skb).consumed += chunk;
2032 sk_peek_offset_bwd(sk, chunk);
2035 unix_detach_fds(siocb->scm, skb);
2037 if (unix_skb_len(skb))
2040 skb_unlink(skb, &sk->sk_receive_queue);
2046 /* It is questionable, see note in unix_dgram_recvmsg.
2049 siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
2051 sk_peek_offset_fwd(sk, chunk);
2057 mutex_unlock(&u->readlock);
2058 scm_recv(sock, msg, siocb->scm, flags);
2060 return copied ? : err;
2063 static int unix_shutdown(struct socket *sock, int mode)
2065 struct sock *sk = sock->sk;
2068 if (mode < SHUT_RD || mode > SHUT_RDWR)
2071 * SHUT_RD (0) -> RCV_SHUTDOWN (1)
2072 * SHUT_WR (1) -> SEND_SHUTDOWN (2)
2073 * SHUT_RDWR (2) -> SHUTDOWN_MASK (3)
2077 unix_state_lock(sk);
2078 sk->sk_shutdown |= mode;
2079 other = unix_peer(sk);
2082 unix_state_unlock(sk);
2083 sk->sk_state_change(sk);
2086 (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET)) {
2090 if (mode&RCV_SHUTDOWN)
2091 peer_mode |= SEND_SHUTDOWN;
2092 if (mode&SEND_SHUTDOWN)
2093 peer_mode |= RCV_SHUTDOWN;
2094 unix_state_lock(other);
2095 other->sk_shutdown |= peer_mode;
2096 unix_state_unlock(other);
2097 other->sk_state_change(other);
2098 if (peer_mode == SHUTDOWN_MASK)
2099 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_HUP);
2100 else if (peer_mode & RCV_SHUTDOWN)
2101 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_IN);
2109 long unix_inq_len(struct sock *sk)
2111 struct sk_buff *skb;
2114 if (sk->sk_state == TCP_LISTEN)
2117 spin_lock(&sk->sk_receive_queue.lock);
2118 if (sk->sk_type == SOCK_STREAM ||
2119 sk->sk_type == SOCK_SEQPACKET) {
2120 skb_queue_walk(&sk->sk_receive_queue, skb)
2121 amount += unix_skb_len(skb);
2123 skb = skb_peek(&sk->sk_receive_queue);
2127 spin_unlock(&sk->sk_receive_queue.lock);
2131 EXPORT_SYMBOL_GPL(unix_inq_len);
2133 long unix_outq_len(struct sock *sk)
2135 return sk_wmem_alloc_get(sk);
2137 EXPORT_SYMBOL_GPL(unix_outq_len);
2139 static int unix_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2141 struct sock *sk = sock->sk;
2147 amount = unix_outq_len(sk);
2148 err = put_user(amount, (int __user *)arg);
2151 amount = unix_inq_len(sk);
2155 err = put_user(amount, (int __user *)arg);
2164 static unsigned int unix_poll(struct file *file, struct socket *sock, poll_table *wait)
2166 struct sock *sk = sock->sk;
2169 sock_poll_wait(file, sk_sleep(sk), wait);
2172 /* exceptional events? */
2175 if (sk->sk_shutdown == SHUTDOWN_MASK)
2177 if (sk->sk_shutdown & RCV_SHUTDOWN)
2178 mask |= POLLRDHUP | POLLIN | POLLRDNORM;
2181 if (!skb_queue_empty(&sk->sk_receive_queue))
2182 mask |= POLLIN | POLLRDNORM;
2184 /* Connection-based need to check for termination and startup */
2185 if ((sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) &&
2186 sk->sk_state == TCP_CLOSE)
2190 * we set writable also when the other side has shut down the
2191 * connection. This prevents stuck sockets.
2193 if (unix_writable(sk))
2194 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2199 static unsigned int unix_dgram_poll(struct file *file, struct socket *sock,
2202 struct sock *sk = sock->sk, *other;
2203 unsigned int mask, writable;
2205 sock_poll_wait(file, sk_sleep(sk), wait);
2208 /* exceptional events? */
2209 if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue))
2211 (sock_flag(sk, SOCK_SELECT_ERR_QUEUE) ? POLLPRI : 0);
2213 if (sk->sk_shutdown & RCV_SHUTDOWN)
2214 mask |= POLLRDHUP | POLLIN | POLLRDNORM;
2215 if (sk->sk_shutdown == SHUTDOWN_MASK)
2219 if (!skb_queue_empty(&sk->sk_receive_queue))
2220 mask |= POLLIN | POLLRDNORM;
2222 /* Connection-based need to check for termination and startup */
2223 if (sk->sk_type == SOCK_SEQPACKET) {
2224 if (sk->sk_state == TCP_CLOSE)
2226 /* connection hasn't started yet? */
2227 if (sk->sk_state == TCP_SYN_SENT)
2231 /* No write status requested, avoid expensive OUT tests. */
2232 if (!(poll_requested_events(wait) & (POLLWRBAND|POLLWRNORM|POLLOUT)))
2235 writable = unix_writable(sk);
2236 other = unix_peer_get(sk);
2238 if (unix_peer(other) != sk) {
2239 sock_poll_wait(file, &unix_sk(other)->peer_wait, wait);
2240 if (unix_recvq_full(other))
2247 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2249 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
2254 #ifdef CONFIG_PROC_FS
2256 #define BUCKET_SPACE (BITS_PER_LONG - (UNIX_HASH_BITS + 1) - 1)
2258 #define get_bucket(x) ((x) >> BUCKET_SPACE)
2259 #define get_offset(x) ((x) & ((1L << BUCKET_SPACE) - 1))
2260 #define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
2262 static struct sock *unix_from_bucket(struct seq_file *seq, loff_t *pos)
2264 unsigned long offset = get_offset(*pos);
2265 unsigned long bucket = get_bucket(*pos);
2267 unsigned long count = 0;
2269 for (sk = sk_head(&unix_socket_table[bucket]); sk; sk = sk_next(sk)) {
2270 if (sock_net(sk) != seq_file_net(seq))
2272 if (++count == offset)
2279 static struct sock *unix_next_socket(struct seq_file *seq,
2283 unsigned long bucket;
2285 while (sk > (struct sock *)SEQ_START_TOKEN) {
2289 if (sock_net(sk) == seq_file_net(seq))
2294 sk = unix_from_bucket(seq, pos);
2299 bucket = get_bucket(*pos) + 1;
2300 *pos = set_bucket_offset(bucket, 1);
2301 } while (bucket < ARRAY_SIZE(unix_socket_table));
2306 static void *unix_seq_start(struct seq_file *seq, loff_t *pos)
2307 __acquires(unix_table_lock)
2309 spin_lock(&unix_table_lock);
2312 return SEQ_START_TOKEN;
2314 if (get_bucket(*pos) >= ARRAY_SIZE(unix_socket_table))
2317 return unix_next_socket(seq, NULL, pos);
2320 static void *unix_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2323 return unix_next_socket(seq, v, pos);
2326 static void unix_seq_stop(struct seq_file *seq, void *v)
2327 __releases(unix_table_lock)
2329 spin_unlock(&unix_table_lock);
2332 static int unix_seq_show(struct seq_file *seq, void *v)
2335 if (v == SEQ_START_TOKEN)
2336 seq_puts(seq, "Num RefCount Protocol Flags Type St "
2340 struct unix_sock *u = unix_sk(s);
2343 seq_printf(seq, "%pK: %08X %08X %08X %04X %02X %5lu",
2345 atomic_read(&s->sk_refcnt),
2347 s->sk_state == TCP_LISTEN ? __SO_ACCEPTCON : 0,
2350 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTED : SS_UNCONNECTED) :
2351 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTING : SS_DISCONNECTING),
2359 len = u->addr->len - sizeof(short);
2360 if (!UNIX_ABSTRACT(s))
2366 for ( ; i < len; i++)
2367 seq_putc(seq, u->addr->name->sun_path[i]);
2369 unix_state_unlock(s);
2370 seq_putc(seq, '\n');
2376 static const struct seq_operations unix_seq_ops = {
2377 .start = unix_seq_start,
2378 .next = unix_seq_next,
2379 .stop = unix_seq_stop,
2380 .show = unix_seq_show,
2383 static int unix_seq_open(struct inode *inode, struct file *file)
2385 return seq_open_net(inode, file, &unix_seq_ops,
2386 sizeof(struct seq_net_private));
2389 static const struct file_operations unix_seq_fops = {
2390 .owner = THIS_MODULE,
2391 .open = unix_seq_open,
2393 .llseek = seq_lseek,
2394 .release = seq_release_net,
2399 static const struct net_proto_family unix_family_ops = {
2401 .create = unix_create,
2402 .owner = THIS_MODULE,
2406 static int __net_init unix_net_init(struct net *net)
2408 int error = -ENOMEM;
2410 net->unx.sysctl_max_dgram_qlen = 10;
2411 if (unix_sysctl_register(net))
2414 #ifdef CONFIG_PROC_FS
2415 if (!proc_create("unix", 0, net->proc_net, &unix_seq_fops)) {
2416 unix_sysctl_unregister(net);
2425 static void __net_exit unix_net_exit(struct net *net)
2427 unix_sysctl_unregister(net);
2428 remove_proc_entry("unix", net->proc_net);
2431 static struct pernet_operations unix_net_ops = {
2432 .init = unix_net_init,
2433 .exit = unix_net_exit,
2436 static int __init af_unix_init(void)
2440 BUILD_BUG_ON(sizeof(struct unix_skb_parms) > FIELD_SIZEOF(struct sk_buff, cb));
2442 rc = proto_register(&unix_proto, 1);
2444 printk(KERN_CRIT "%s: Cannot create unix_sock SLAB cache!\n",
2449 sock_register(&unix_family_ops);
2450 register_pernet_subsys(&unix_net_ops);
2455 static void __exit af_unix_exit(void)
2457 sock_unregister(PF_UNIX);
2458 proto_unregister(&unix_proto);
2459 unregister_pernet_subsys(&unix_net_ops);
2462 /* Earlier than device_initcall() so that other drivers invoking
2463 request_module() don't end up in a loop when modprobe tries
2464 to use a UNIX socket. But later than subsys_initcall() because
2465 we depend on stuff initialised there */
2466 fs_initcall(af_unix_init);
2467 module_exit(af_unix_exit);
2469 MODULE_LICENSE("GPL");
2470 MODULE_ALIAS_NETPROTO(PF_UNIX);