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 void unix_set_peek_off(struct sock *sk, int val)
535 struct unix_sock *u = unix_sk(sk);
537 mutex_lock(&u->readlock);
538 sk->sk_peek_off = val;
539 mutex_unlock(&u->readlock);
543 static const struct proto_ops unix_stream_ops = {
545 .owner = THIS_MODULE,
546 .release = unix_release,
548 .connect = unix_stream_connect,
549 .socketpair = unix_socketpair,
550 .accept = unix_accept,
551 .getname = unix_getname,
554 .listen = unix_listen,
555 .shutdown = unix_shutdown,
556 .setsockopt = sock_no_setsockopt,
557 .getsockopt = sock_no_getsockopt,
558 .sendmsg = unix_stream_sendmsg,
559 .recvmsg = unix_stream_recvmsg,
560 .mmap = sock_no_mmap,
561 .sendpage = sock_no_sendpage,
562 .set_peek_off = unix_set_peek_off,
565 static const struct proto_ops unix_dgram_ops = {
567 .owner = THIS_MODULE,
568 .release = unix_release,
570 .connect = unix_dgram_connect,
571 .socketpair = unix_socketpair,
572 .accept = sock_no_accept,
573 .getname = unix_getname,
574 .poll = unix_dgram_poll,
576 .listen = sock_no_listen,
577 .shutdown = unix_shutdown,
578 .setsockopt = sock_no_setsockopt,
579 .getsockopt = sock_no_getsockopt,
580 .sendmsg = unix_dgram_sendmsg,
581 .recvmsg = unix_dgram_recvmsg,
582 .mmap = sock_no_mmap,
583 .sendpage = sock_no_sendpage,
584 .set_peek_off = unix_set_peek_off,
587 static const struct proto_ops unix_seqpacket_ops = {
589 .owner = THIS_MODULE,
590 .release = unix_release,
592 .connect = unix_stream_connect,
593 .socketpair = unix_socketpair,
594 .accept = unix_accept,
595 .getname = unix_getname,
596 .poll = unix_dgram_poll,
598 .listen = unix_listen,
599 .shutdown = unix_shutdown,
600 .setsockopt = sock_no_setsockopt,
601 .getsockopt = sock_no_getsockopt,
602 .sendmsg = unix_seqpacket_sendmsg,
603 .recvmsg = unix_seqpacket_recvmsg,
604 .mmap = sock_no_mmap,
605 .sendpage = sock_no_sendpage,
606 .set_peek_off = unix_set_peek_off,
609 static struct proto unix_proto = {
611 .owner = THIS_MODULE,
612 .obj_size = sizeof(struct unix_sock),
616 * AF_UNIX sockets do not interact with hardware, hence they
617 * dont trigger interrupts - so it's safe for them to have
618 * bh-unsafe locking for their sk_receive_queue.lock. Split off
619 * this special lock-class by reinitializing the spinlock key:
621 static struct lock_class_key af_unix_sk_receive_queue_lock_key;
623 static struct sock *unix_create1(struct net *net, struct socket *sock)
625 struct sock *sk = NULL;
628 atomic_long_inc(&unix_nr_socks);
629 if (atomic_long_read(&unix_nr_socks) > 2 * get_max_files())
632 sk = sk_alloc(net, PF_UNIX, GFP_KERNEL, &unix_proto);
636 sock_init_data(sock, sk);
637 lockdep_set_class(&sk->sk_receive_queue.lock,
638 &af_unix_sk_receive_queue_lock_key);
640 sk->sk_write_space = unix_write_space;
641 sk->sk_max_ack_backlog = net->unx.sysctl_max_dgram_qlen;
642 sk->sk_destruct = unix_sock_destructor;
644 u->path.dentry = NULL;
646 spin_lock_init(&u->lock);
647 atomic_long_set(&u->inflight, 0);
648 INIT_LIST_HEAD(&u->link);
649 mutex_init(&u->readlock); /* single task reading lock */
650 init_waitqueue_head(&u->peer_wait);
651 unix_insert_socket(unix_sockets_unbound(sk), sk);
654 atomic_long_dec(&unix_nr_socks);
657 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
663 static int unix_create(struct net *net, struct socket *sock, int protocol,
666 if (protocol && protocol != PF_UNIX)
667 return -EPROTONOSUPPORT;
669 sock->state = SS_UNCONNECTED;
671 switch (sock->type) {
673 sock->ops = &unix_stream_ops;
676 * Believe it or not BSD has AF_UNIX, SOCK_RAW though
680 sock->type = SOCK_DGRAM;
682 sock->ops = &unix_dgram_ops;
685 sock->ops = &unix_seqpacket_ops;
688 return -ESOCKTNOSUPPORT;
691 return unix_create1(net, sock) ? 0 : -ENOMEM;
694 static int unix_release(struct socket *sock)
696 struct sock *sk = sock->sk;
701 unix_release_sock(sk, 0);
707 static int unix_autobind(struct socket *sock)
709 struct sock *sk = sock->sk;
710 struct net *net = sock_net(sk);
711 struct unix_sock *u = unix_sk(sk);
712 static u32 ordernum = 1;
713 struct unix_address *addr;
715 unsigned int retries = 0;
717 mutex_lock(&u->readlock);
724 addr = kzalloc(sizeof(*addr) + sizeof(short) + 16, GFP_KERNEL);
728 addr->name->sun_family = AF_UNIX;
729 atomic_set(&addr->refcnt, 1);
732 addr->len = sprintf(addr->name->sun_path+1, "%05x", ordernum) + 1 + sizeof(short);
733 addr->hash = unix_hash_fold(csum_partial(addr->name, addr->len, 0));
735 spin_lock(&unix_table_lock);
736 ordernum = (ordernum+1)&0xFFFFF;
738 if (__unix_find_socket_byname(net, addr->name, addr->len, sock->type,
740 spin_unlock(&unix_table_lock);
742 * __unix_find_socket_byname() may take long time if many names
743 * are already in use.
746 /* Give up if all names seems to be in use. */
747 if (retries++ == 0xFFFFF) {
754 addr->hash ^= sk->sk_type;
756 __unix_remove_socket(sk);
758 __unix_insert_socket(&unix_socket_table[addr->hash], sk);
759 spin_unlock(&unix_table_lock);
762 out: mutex_unlock(&u->readlock);
766 static struct sock *unix_find_other(struct net *net,
767 struct sockaddr_un *sunname, int len,
768 int type, unsigned int hash, int *error)
774 if (sunname->sun_path[0]) {
776 err = kern_path(sunname->sun_path, LOOKUP_FOLLOW, &path);
779 inode = path.dentry->d_inode;
780 err = inode_permission(inode, MAY_WRITE);
785 if (!S_ISSOCK(inode->i_mode))
787 u = unix_find_socket_byinode(inode);
791 if (u->sk_type == type)
797 if (u->sk_type != type) {
803 u = unix_find_socket_byname(net, sunname, len, type, hash);
805 struct dentry *dentry;
806 dentry = unix_sk(u)->path.dentry;
808 touch_atime(&unix_sk(u)->path);
821 static int unix_mknod(const char *sun_path, umode_t mode, struct path *res)
823 struct dentry *dentry;
827 * Get the parent directory, calculate the hash for last
830 dentry = kern_path_create(AT_FDCWD, sun_path, &path, 0);
831 err = PTR_ERR(dentry);
836 * All right, let's create it.
838 err = security_path_mknod(&path, dentry, mode, 0);
840 err = vfs_mknod(path.dentry->d_inode, dentry, mode, 0);
842 res->mnt = mntget(path.mnt);
843 res->dentry = dget(dentry);
846 done_path_create(&path, dentry);
850 static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
852 struct sock *sk = sock->sk;
853 struct net *net = sock_net(sk);
854 struct unix_sock *u = unix_sk(sk);
855 struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
856 char *sun_path = sunaddr->sun_path;
859 struct unix_address *addr;
860 struct hlist_head *list;
863 if (sunaddr->sun_family != AF_UNIX)
866 if (addr_len == sizeof(short)) {
867 err = unix_autobind(sock);
871 err = unix_mkname(sunaddr, addr_len, &hash);
876 mutex_lock(&u->readlock);
883 addr = kmalloc(sizeof(*addr)+addr_len, GFP_KERNEL);
887 memcpy(addr->name, sunaddr, addr_len);
888 addr->len = addr_len;
889 addr->hash = hash ^ sk->sk_type;
890 atomic_set(&addr->refcnt, 1);
894 umode_t mode = S_IFSOCK |
895 (SOCK_INODE(sock)->i_mode & ~current_umask());
896 err = unix_mknod(sun_path, mode, &path);
900 unix_release_addr(addr);
903 addr->hash = UNIX_HASH_SIZE;
904 hash = path.dentry->d_inode->i_ino & (UNIX_HASH_SIZE-1);
905 spin_lock(&unix_table_lock);
907 list = &unix_socket_table[hash];
909 spin_lock(&unix_table_lock);
911 if (__unix_find_socket_byname(net, sunaddr, addr_len,
912 sk->sk_type, hash)) {
913 unix_release_addr(addr);
917 list = &unix_socket_table[addr->hash];
921 __unix_remove_socket(sk);
923 __unix_insert_socket(list, sk);
926 spin_unlock(&unix_table_lock);
928 mutex_unlock(&u->readlock);
933 static void unix_state_double_lock(struct sock *sk1, struct sock *sk2)
935 if (unlikely(sk1 == sk2) || !sk2) {
936 unix_state_lock(sk1);
940 unix_state_lock(sk1);
941 unix_state_lock_nested(sk2);
943 unix_state_lock(sk2);
944 unix_state_lock_nested(sk1);
948 static void unix_state_double_unlock(struct sock *sk1, struct sock *sk2)
950 if (unlikely(sk1 == sk2) || !sk2) {
951 unix_state_unlock(sk1);
954 unix_state_unlock(sk1);
955 unix_state_unlock(sk2);
958 static int unix_dgram_connect(struct socket *sock, struct sockaddr *addr,
961 struct sock *sk = sock->sk;
962 struct net *net = sock_net(sk);
963 struct sockaddr_un *sunaddr = (struct sockaddr_un *)addr;
968 if (addr->sa_family != AF_UNSPEC) {
969 err = unix_mkname(sunaddr, alen, &hash);
974 if (test_bit(SOCK_PASSCRED, &sock->flags) &&
975 !unix_sk(sk)->addr && (err = unix_autobind(sock)) != 0)
979 other = unix_find_other(net, sunaddr, alen, sock->type, hash, &err);
983 unix_state_double_lock(sk, other);
985 /* Apparently VFS overslept socket death. Retry. */
986 if (sock_flag(other, SOCK_DEAD)) {
987 unix_state_double_unlock(sk, other);
993 if (!unix_may_send(sk, other))
996 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1002 * 1003.1g breaking connected state with AF_UNSPEC
1005 unix_state_double_lock(sk, other);
1009 * If it was connected, reconnect.
1011 if (unix_peer(sk)) {
1012 struct sock *old_peer = unix_peer(sk);
1013 unix_peer(sk) = other;
1014 unix_state_double_unlock(sk, other);
1016 if (other != old_peer)
1017 unix_dgram_disconnected(sk, old_peer);
1020 unix_peer(sk) = other;
1021 unix_state_double_unlock(sk, other);
1026 unix_state_double_unlock(sk, other);
1032 static long unix_wait_for_peer(struct sock *other, long timeo)
1034 struct unix_sock *u = unix_sk(other);
1038 prepare_to_wait_exclusive(&u->peer_wait, &wait, TASK_INTERRUPTIBLE);
1040 sched = !sock_flag(other, SOCK_DEAD) &&
1041 !(other->sk_shutdown & RCV_SHUTDOWN) &&
1042 unix_recvq_full(other);
1044 unix_state_unlock(other);
1047 timeo = schedule_timeout(timeo);
1049 finish_wait(&u->peer_wait, &wait);
1053 static int unix_stream_connect(struct socket *sock, struct sockaddr *uaddr,
1054 int addr_len, int flags)
1056 struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
1057 struct sock *sk = sock->sk;
1058 struct net *net = sock_net(sk);
1059 struct unix_sock *u = unix_sk(sk), *newu, *otheru;
1060 struct sock *newsk = NULL;
1061 struct sock *other = NULL;
1062 struct sk_buff *skb = NULL;
1068 err = unix_mkname(sunaddr, addr_len, &hash);
1073 if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr &&
1074 (err = unix_autobind(sock)) != 0)
1077 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
1079 /* First of all allocate resources.
1080 If we will make it after state is locked,
1081 we will have to recheck all again in any case.
1086 /* create new sock for complete connection */
1087 newsk = unix_create1(sock_net(sk), NULL);
1091 /* Allocate skb for sending to listening sock */
1092 skb = sock_wmalloc(newsk, 1, 0, GFP_KERNEL);
1097 /* Find listening sock. */
1098 other = unix_find_other(net, sunaddr, addr_len, sk->sk_type, hash, &err);
1102 /* Latch state of peer */
1103 unix_state_lock(other);
1105 /* Apparently VFS overslept socket death. Retry. */
1106 if (sock_flag(other, SOCK_DEAD)) {
1107 unix_state_unlock(other);
1112 err = -ECONNREFUSED;
1113 if (other->sk_state != TCP_LISTEN)
1115 if (other->sk_shutdown & RCV_SHUTDOWN)
1118 if (unix_recvq_full(other)) {
1123 timeo = unix_wait_for_peer(other, timeo);
1125 err = sock_intr_errno(timeo);
1126 if (signal_pending(current))
1134 It is tricky place. We need to grab our state lock and cannot
1135 drop lock on peer. It is dangerous because deadlock is
1136 possible. Connect to self case and simultaneous
1137 attempt to connect are eliminated by checking socket
1138 state. other is TCP_LISTEN, if sk is TCP_LISTEN we
1139 check this before attempt to grab lock.
1141 Well, and we have to recheck the state after socket locked.
1147 /* This is ok... continue with connect */
1149 case TCP_ESTABLISHED:
1150 /* Socket is already connected */
1158 unix_state_lock_nested(sk);
1160 if (sk->sk_state != st) {
1161 unix_state_unlock(sk);
1162 unix_state_unlock(other);
1167 err = security_unix_stream_connect(sk, other, newsk);
1169 unix_state_unlock(sk);
1173 /* The way is open! Fastly set all the necessary fields... */
1176 unix_peer(newsk) = sk;
1177 newsk->sk_state = TCP_ESTABLISHED;
1178 newsk->sk_type = sk->sk_type;
1179 init_peercred(newsk);
1180 newu = unix_sk(newsk);
1181 RCU_INIT_POINTER(newsk->sk_wq, &newu->peer_wq);
1182 otheru = unix_sk(other);
1184 /* copy address information from listening to new sock*/
1186 atomic_inc(&otheru->addr->refcnt);
1187 newu->addr = otheru->addr;
1189 if (otheru->path.dentry) {
1190 path_get(&otheru->path);
1191 newu->path = otheru->path;
1194 /* Set credentials */
1195 copy_peercred(sk, other);
1197 sock->state = SS_CONNECTED;
1198 sk->sk_state = TCP_ESTABLISHED;
1201 smp_mb__after_atomic_inc(); /* sock_hold() does an atomic_inc() */
1202 unix_peer(sk) = newsk;
1204 unix_state_unlock(sk);
1206 /* take ten and and send info to listening sock */
1207 spin_lock(&other->sk_receive_queue.lock);
1208 __skb_queue_tail(&other->sk_receive_queue, skb);
1209 spin_unlock(&other->sk_receive_queue.lock);
1210 unix_state_unlock(other);
1211 other->sk_data_ready(other, 0);
1217 unix_state_unlock(other);
1222 unix_release_sock(newsk, 0);
1228 static int unix_socketpair(struct socket *socka, struct socket *sockb)
1230 struct sock *ska = socka->sk, *skb = sockb->sk;
1232 /* Join our sockets back to back */
1235 unix_peer(ska) = skb;
1236 unix_peer(skb) = ska;
1240 if (ska->sk_type != SOCK_DGRAM) {
1241 ska->sk_state = TCP_ESTABLISHED;
1242 skb->sk_state = TCP_ESTABLISHED;
1243 socka->state = SS_CONNECTED;
1244 sockb->state = SS_CONNECTED;
1249 static void unix_sock_inherit_flags(const struct socket *old,
1252 if (test_bit(SOCK_PASSCRED, &old->flags))
1253 set_bit(SOCK_PASSCRED, &new->flags);
1254 if (test_bit(SOCK_PASSSEC, &old->flags))
1255 set_bit(SOCK_PASSSEC, &new->flags);
1258 static int unix_accept(struct socket *sock, struct socket *newsock, int flags)
1260 struct sock *sk = sock->sk;
1262 struct sk_buff *skb;
1266 if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
1270 if (sk->sk_state != TCP_LISTEN)
1273 /* If socket state is TCP_LISTEN it cannot change (for now...),
1274 * so that no locks are necessary.
1277 skb = skb_recv_datagram(sk, 0, flags&O_NONBLOCK, &err);
1279 /* This means receive shutdown. */
1286 skb_free_datagram(sk, skb);
1287 wake_up_interruptible(&unix_sk(sk)->peer_wait);
1289 /* attach accepted sock to socket */
1290 unix_state_lock(tsk);
1291 newsock->state = SS_CONNECTED;
1292 unix_sock_inherit_flags(sock, newsock);
1293 sock_graft(tsk, newsock);
1294 unix_state_unlock(tsk);
1302 static int unix_getname(struct socket *sock, struct sockaddr *uaddr, int *uaddr_len, int peer)
1304 struct sock *sk = sock->sk;
1305 struct unix_sock *u;
1306 DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr, uaddr);
1310 sk = unix_peer_get(sk);
1321 unix_state_lock(sk);
1323 sunaddr->sun_family = AF_UNIX;
1324 sunaddr->sun_path[0] = 0;
1325 *uaddr_len = sizeof(short);
1327 struct unix_address *addr = u->addr;
1329 *uaddr_len = addr->len;
1330 memcpy(sunaddr, addr->name, *uaddr_len);
1332 unix_state_unlock(sk);
1338 static void unix_detach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1342 scm->fp = UNIXCB(skb).fp;
1343 UNIXCB(skb).fp = NULL;
1345 for (i = scm->fp->count-1; i >= 0; i--)
1346 unix_notinflight(scm->fp->fp[i]);
1349 static void unix_destruct_scm(struct sk_buff *skb)
1351 struct scm_cookie scm;
1352 memset(&scm, 0, sizeof(scm));
1353 scm.pid = UNIXCB(skb).pid;
1355 unix_detach_fds(&scm, skb);
1357 /* Alas, it calls VFS */
1358 /* So fscking what? fput() had been SMP-safe since the last Summer */
1363 #define MAX_RECURSION_LEVEL 4
1365 static int unix_attach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1368 unsigned char max_level = 0;
1369 int unix_sock_count = 0;
1371 for (i = scm->fp->count - 1; i >= 0; i--) {
1372 struct sock *sk = unix_get_socket(scm->fp->fp[i]);
1376 max_level = max(max_level,
1377 unix_sk(sk)->recursion_level);
1380 if (unlikely(max_level > MAX_RECURSION_LEVEL))
1381 return -ETOOMANYREFS;
1384 * Need to duplicate file references for the sake of garbage
1385 * collection. Otherwise a socket in the fps might become a
1386 * candidate for GC while the skb is not yet queued.
1388 UNIXCB(skb).fp = scm_fp_dup(scm->fp);
1389 if (!UNIXCB(skb).fp)
1392 if (unix_sock_count) {
1393 for (i = scm->fp->count - 1; i >= 0; i--)
1394 unix_inflight(scm->fp->fp[i]);
1399 static int unix_scm_to_skb(struct scm_cookie *scm, struct sk_buff *skb, bool send_fds)
1403 UNIXCB(skb).pid = get_pid(scm->pid);
1404 UNIXCB(skb).uid = scm->creds.uid;
1405 UNIXCB(skb).gid = scm->creds.gid;
1406 UNIXCB(skb).fp = NULL;
1407 if (scm->fp && send_fds)
1408 err = unix_attach_fds(scm, skb);
1410 skb->destructor = unix_destruct_scm;
1415 * Some apps rely on write() giving SCM_CREDENTIALS
1416 * We include credentials if source or destination socket
1417 * asserted SOCK_PASSCRED.
1419 static void maybe_add_creds(struct sk_buff *skb, const struct socket *sock,
1420 const struct sock *other)
1422 if (UNIXCB(skb).pid)
1424 if (test_bit(SOCK_PASSCRED, &sock->flags) ||
1425 !other->sk_socket ||
1426 test_bit(SOCK_PASSCRED, &other->sk_socket->flags)) {
1427 UNIXCB(skb).pid = get_pid(task_tgid(current));
1428 current_uid_gid(&UNIXCB(skb).uid, &UNIXCB(skb).gid);
1433 * Send AF_UNIX data.
1436 static int unix_dgram_sendmsg(struct kiocb *kiocb, struct socket *sock,
1437 struct msghdr *msg, size_t len)
1439 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1440 struct sock *sk = sock->sk;
1441 struct net *net = sock_net(sk);
1442 struct unix_sock *u = unix_sk(sk);
1443 struct sockaddr_un *sunaddr = msg->msg_name;
1444 struct sock *other = NULL;
1445 int namelen = 0; /* fake GCC */
1448 struct sk_buff *skb;
1450 struct scm_cookie tmp_scm;
1454 if (NULL == siocb->scm)
1455 siocb->scm = &tmp_scm;
1457 err = scm_send(sock, msg, siocb->scm, false);
1462 if (msg->msg_flags&MSG_OOB)
1465 if (msg->msg_namelen) {
1466 err = unix_mkname(sunaddr, msg->msg_namelen, &hash);
1473 other = unix_peer_get(sk);
1478 if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr
1479 && (err = unix_autobind(sock)) != 0)
1483 if (len > sk->sk_sndbuf - 32)
1486 if (len > SKB_MAX_ALLOC)
1487 data_len = min_t(size_t,
1488 len - SKB_MAX_ALLOC,
1489 MAX_SKB_FRAGS * PAGE_SIZE);
1491 skb = sock_alloc_send_pskb(sk, len - data_len, data_len,
1492 msg->msg_flags & MSG_DONTWAIT, &err,
1493 PAGE_ALLOC_COSTLY_ORDER);
1497 err = unix_scm_to_skb(siocb->scm, skb, true);
1500 max_level = err + 1;
1501 unix_get_secdata(siocb->scm, skb);
1503 skb_put(skb, len - data_len);
1504 skb->data_len = data_len;
1506 err = skb_copy_datagram_from_iovec(skb, 0, msg->msg_iov, 0, len);
1510 timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1515 if (sunaddr == NULL)
1518 other = unix_find_other(net, sunaddr, namelen, sk->sk_type,
1524 if (sk_filter(other, skb) < 0) {
1525 /* Toss the packet but do not return any error to the sender */
1530 unix_state_lock(other);
1532 if (!unix_may_send(sk, other))
1535 if (sock_flag(other, SOCK_DEAD)) {
1537 * Check with 1003.1g - what should
1540 unix_state_unlock(other);
1544 unix_state_lock(sk);
1545 if (unix_peer(sk) == other) {
1546 unix_peer(sk) = NULL;
1547 unix_state_unlock(sk);
1549 unix_dgram_disconnected(sk, other);
1551 err = -ECONNREFUSED;
1553 unix_state_unlock(sk);
1563 if (other->sk_shutdown & RCV_SHUTDOWN)
1566 if (sk->sk_type != SOCK_SEQPACKET) {
1567 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1572 if (unix_peer(other) != sk && unix_recvq_full(other)) {
1578 timeo = unix_wait_for_peer(other, timeo);
1580 err = sock_intr_errno(timeo);
1581 if (signal_pending(current))
1587 if (sock_flag(other, SOCK_RCVTSTAMP))
1588 __net_timestamp(skb);
1589 maybe_add_creds(skb, sock, other);
1590 skb_queue_tail(&other->sk_receive_queue, skb);
1591 if (max_level > unix_sk(other)->recursion_level)
1592 unix_sk(other)->recursion_level = max_level;
1593 unix_state_unlock(other);
1594 other->sk_data_ready(other, len);
1596 scm_destroy(siocb->scm);
1600 unix_state_unlock(other);
1606 scm_destroy(siocb->scm);
1610 /* We use paged skbs for stream sockets, and limit occupancy to 32768
1611 * bytes, and a minimun of a full page.
1613 #define UNIX_SKB_FRAGS_SZ (PAGE_SIZE << get_order(32768))
1615 static int unix_stream_sendmsg(struct kiocb *kiocb, struct socket *sock,
1616 struct msghdr *msg, size_t len)
1618 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1619 struct sock *sk = sock->sk;
1620 struct sock *other = NULL;
1622 struct sk_buff *skb;
1624 struct scm_cookie tmp_scm;
1625 bool fds_sent = false;
1629 if (NULL == siocb->scm)
1630 siocb->scm = &tmp_scm;
1632 err = scm_send(sock, msg, siocb->scm, false);
1637 if (msg->msg_flags&MSG_OOB)
1640 if (msg->msg_namelen) {
1641 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
1645 other = unix_peer(sk);
1650 if (sk->sk_shutdown & SEND_SHUTDOWN)
1653 while (sent < len) {
1656 /* Keep two messages in the pipe so it schedules better */
1657 size = min_t(int, size, (sk->sk_sndbuf >> 1) - 64);
1659 /* allow fallback to order-0 allocations */
1660 size = min_t(int, size, SKB_MAX_HEAD(0) + UNIX_SKB_FRAGS_SZ);
1662 data_len = max_t(int, 0, size - SKB_MAX_HEAD(0));
1664 skb = sock_alloc_send_pskb(sk, size - data_len, data_len,
1665 msg->msg_flags & MSG_DONTWAIT, &err,
1666 get_order(UNIX_SKB_FRAGS_SZ));
1670 /* Only send the fds in the first buffer */
1671 err = unix_scm_to_skb(siocb->scm, skb, !fds_sent);
1676 max_level = err + 1;
1679 skb_put(skb, size - data_len);
1680 skb->data_len = data_len;
1682 err = skb_copy_datagram_from_iovec(skb, 0, msg->msg_iov,
1689 unix_state_lock(other);
1691 if (sock_flag(other, SOCK_DEAD) ||
1692 (other->sk_shutdown & RCV_SHUTDOWN))
1695 maybe_add_creds(skb, sock, other);
1696 skb_queue_tail(&other->sk_receive_queue, skb);
1697 if (max_level > unix_sk(other)->recursion_level)
1698 unix_sk(other)->recursion_level = max_level;
1699 unix_state_unlock(other);
1700 other->sk_data_ready(other, size);
1704 scm_destroy(siocb->scm);
1710 unix_state_unlock(other);
1713 if (sent == 0 && !(msg->msg_flags&MSG_NOSIGNAL))
1714 send_sig(SIGPIPE, current, 0);
1717 scm_destroy(siocb->scm);
1719 return sent ? : err;
1722 static int unix_seqpacket_sendmsg(struct kiocb *kiocb, struct socket *sock,
1723 struct msghdr *msg, size_t len)
1726 struct sock *sk = sock->sk;
1728 err = sock_error(sk);
1732 if (sk->sk_state != TCP_ESTABLISHED)
1735 if (msg->msg_namelen)
1736 msg->msg_namelen = 0;
1738 return unix_dgram_sendmsg(kiocb, sock, msg, len);
1741 static int unix_seqpacket_recvmsg(struct kiocb *iocb, struct socket *sock,
1742 struct msghdr *msg, size_t size,
1745 struct sock *sk = sock->sk;
1747 if (sk->sk_state != TCP_ESTABLISHED)
1750 return unix_dgram_recvmsg(iocb, sock, msg, size, flags);
1753 static void unix_copy_addr(struct msghdr *msg, struct sock *sk)
1755 struct unix_sock *u = unix_sk(sk);
1757 msg->msg_namelen = 0;
1759 msg->msg_namelen = u->addr->len;
1760 memcpy(msg->msg_name, u->addr->name, u->addr->len);
1764 static int unix_dgram_recvmsg(struct kiocb *iocb, struct socket *sock,
1765 struct msghdr *msg, size_t size,
1768 struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1769 struct scm_cookie tmp_scm;
1770 struct sock *sk = sock->sk;
1771 struct unix_sock *u = unix_sk(sk);
1772 int noblock = flags & MSG_DONTWAIT;
1773 struct sk_buff *skb;
1781 msg->msg_namelen = 0;
1783 err = mutex_lock_interruptible(&u->readlock);
1785 err = sock_intr_errno(sock_rcvtimeo(sk, noblock));
1789 skip = sk_peek_offset(sk, flags);
1791 skb = __skb_recv_datagram(sk, flags, &peeked, &skip, &err);
1793 unix_state_lock(sk);
1794 /* Signal EOF on disconnected non-blocking SEQPACKET socket. */
1795 if (sk->sk_type == SOCK_SEQPACKET && err == -EAGAIN &&
1796 (sk->sk_shutdown & RCV_SHUTDOWN))
1798 unix_state_unlock(sk);
1802 wake_up_interruptible_sync_poll(&u->peer_wait,
1803 POLLOUT | POLLWRNORM | POLLWRBAND);
1806 unix_copy_addr(msg, skb->sk);
1808 if (size > skb->len - skip)
1809 size = skb->len - skip;
1810 else if (size < skb->len - skip)
1811 msg->msg_flags |= MSG_TRUNC;
1813 err = skb_copy_datagram_iovec(skb, skip, msg->msg_iov, size);
1817 if (sock_flag(sk, SOCK_RCVTSTAMP))
1818 __sock_recv_timestamp(msg, sk, skb);
1821 siocb->scm = &tmp_scm;
1822 memset(&tmp_scm, 0, sizeof(tmp_scm));
1824 scm_set_cred(siocb->scm, UNIXCB(skb).pid, UNIXCB(skb).uid, UNIXCB(skb).gid);
1825 unix_set_secdata(siocb->scm, skb);
1827 if (!(flags & MSG_PEEK)) {
1829 unix_detach_fds(siocb->scm, skb);
1831 sk_peek_offset_bwd(sk, skb->len);
1833 /* It is questionable: on PEEK we could:
1834 - do not return fds - good, but too simple 8)
1835 - return fds, and do not return them on read (old strategy,
1837 - clone fds (I chose it for now, it is the most universal
1840 POSIX 1003.1g does not actually define this clearly
1841 at all. POSIX 1003.1g doesn't define a lot of things
1846 sk_peek_offset_fwd(sk, size);
1849 siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1851 err = (flags & MSG_TRUNC) ? skb->len - skip : size;
1853 scm_recv(sock, msg, siocb->scm, flags);
1856 skb_free_datagram(sk, skb);
1858 mutex_unlock(&u->readlock);
1864 * Sleep until more data has arrived. But check for races..
1866 static long unix_stream_data_wait(struct sock *sk, long timeo,
1867 struct sk_buff *last)
1871 unix_state_lock(sk);
1874 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1876 if (skb_peek_tail(&sk->sk_receive_queue) != last ||
1878 (sk->sk_shutdown & RCV_SHUTDOWN) ||
1879 signal_pending(current) ||
1883 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1884 unix_state_unlock(sk);
1885 timeo = freezable_schedule_timeout(timeo);
1886 unix_state_lock(sk);
1887 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1890 finish_wait(sk_sleep(sk), &wait);
1891 unix_state_unlock(sk);
1895 static unsigned int unix_skb_len(const struct sk_buff *skb)
1897 return skb->len - UNIXCB(skb).consumed;
1900 static int unix_stream_recvmsg(struct kiocb *iocb, struct socket *sock,
1901 struct msghdr *msg, size_t size,
1904 struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1905 struct scm_cookie tmp_scm;
1906 struct sock *sk = sock->sk;
1907 struct unix_sock *u = unix_sk(sk);
1908 struct sockaddr_un *sunaddr = msg->msg_name;
1910 int check_creds = 0;
1917 if (sk->sk_state != TCP_ESTABLISHED)
1924 target = sock_rcvlowat(sk, flags&MSG_WAITALL, size);
1925 timeo = sock_rcvtimeo(sk, flags&MSG_DONTWAIT);
1927 msg->msg_namelen = 0;
1929 /* Lock the socket to prevent queue disordering
1930 * while sleeps in memcpy_tomsg
1934 siocb->scm = &tmp_scm;
1935 memset(&tmp_scm, 0, sizeof(tmp_scm));
1938 err = mutex_lock_interruptible(&u->readlock);
1940 err = sock_intr_errno(timeo);
1946 struct sk_buff *skb, *last;
1948 unix_state_lock(sk);
1949 last = skb = skb_peek(&sk->sk_receive_queue);
1952 unix_sk(sk)->recursion_level = 0;
1953 if (copied >= target)
1957 * POSIX 1003.1g mandates this order.
1960 err = sock_error(sk);
1963 if (sk->sk_shutdown & RCV_SHUTDOWN)
1966 unix_state_unlock(sk);
1970 mutex_unlock(&u->readlock);
1972 timeo = unix_stream_data_wait(sk, timeo, last);
1974 if (signal_pending(current)
1975 || mutex_lock_interruptible(&u->readlock)) {
1976 err = sock_intr_errno(timeo);
1982 unix_state_unlock(sk);
1986 skip = sk_peek_offset(sk, flags);
1987 while (skip >= unix_skb_len(skb)) {
1988 skip -= unix_skb_len(skb);
1990 skb = skb_peek_next(skb, &sk->sk_receive_queue);
1995 unix_state_unlock(sk);
1998 /* Never glue messages from different writers */
1999 if ((UNIXCB(skb).pid != siocb->scm->pid) ||
2000 !uid_eq(UNIXCB(skb).uid, siocb->scm->creds.uid) ||
2001 !gid_eq(UNIXCB(skb).gid, siocb->scm->creds.gid))
2003 } else if (test_bit(SOCK_PASSCRED, &sock->flags)) {
2004 /* Copy credentials */
2005 scm_set_cred(siocb->scm, UNIXCB(skb).pid, UNIXCB(skb).uid, UNIXCB(skb).gid);
2009 /* Copy address just once */
2011 unix_copy_addr(msg, skb->sk);
2015 chunk = min_t(unsigned int, unix_skb_len(skb) - skip, size);
2016 if (skb_copy_datagram_iovec(skb, UNIXCB(skb).consumed + skip,
2017 msg->msg_iov, chunk)) {
2025 /* Mark read part of skb as used */
2026 if (!(flags & MSG_PEEK)) {
2027 UNIXCB(skb).consumed += chunk;
2029 sk_peek_offset_bwd(sk, chunk);
2032 unix_detach_fds(siocb->scm, skb);
2034 if (unix_skb_len(skb))
2037 skb_unlink(skb, &sk->sk_receive_queue);
2043 /* It is questionable, see note in unix_dgram_recvmsg.
2046 siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
2048 sk_peek_offset_fwd(sk, chunk);
2054 mutex_unlock(&u->readlock);
2055 scm_recv(sock, msg, siocb->scm, flags);
2057 return copied ? : err;
2060 static int unix_shutdown(struct socket *sock, int mode)
2062 struct sock *sk = sock->sk;
2065 if (mode < SHUT_RD || mode > SHUT_RDWR)
2068 * SHUT_RD (0) -> RCV_SHUTDOWN (1)
2069 * SHUT_WR (1) -> SEND_SHUTDOWN (2)
2070 * SHUT_RDWR (2) -> SHUTDOWN_MASK (3)
2074 unix_state_lock(sk);
2075 sk->sk_shutdown |= mode;
2076 other = unix_peer(sk);
2079 unix_state_unlock(sk);
2080 sk->sk_state_change(sk);
2083 (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET)) {
2087 if (mode&RCV_SHUTDOWN)
2088 peer_mode |= SEND_SHUTDOWN;
2089 if (mode&SEND_SHUTDOWN)
2090 peer_mode |= RCV_SHUTDOWN;
2091 unix_state_lock(other);
2092 other->sk_shutdown |= peer_mode;
2093 unix_state_unlock(other);
2094 other->sk_state_change(other);
2095 if (peer_mode == SHUTDOWN_MASK)
2096 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_HUP);
2097 else if (peer_mode & RCV_SHUTDOWN)
2098 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_IN);
2106 long unix_inq_len(struct sock *sk)
2108 struct sk_buff *skb;
2111 if (sk->sk_state == TCP_LISTEN)
2114 spin_lock(&sk->sk_receive_queue.lock);
2115 if (sk->sk_type == SOCK_STREAM ||
2116 sk->sk_type == SOCK_SEQPACKET) {
2117 skb_queue_walk(&sk->sk_receive_queue, skb)
2118 amount += unix_skb_len(skb);
2120 skb = skb_peek(&sk->sk_receive_queue);
2124 spin_unlock(&sk->sk_receive_queue.lock);
2128 EXPORT_SYMBOL_GPL(unix_inq_len);
2130 long unix_outq_len(struct sock *sk)
2132 return sk_wmem_alloc_get(sk);
2134 EXPORT_SYMBOL_GPL(unix_outq_len);
2136 static int unix_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2138 struct sock *sk = sock->sk;
2144 amount = unix_outq_len(sk);
2145 err = put_user(amount, (int __user *)arg);
2148 amount = unix_inq_len(sk);
2152 err = put_user(amount, (int __user *)arg);
2161 static unsigned int unix_poll(struct file *file, struct socket *sock, poll_table *wait)
2163 struct sock *sk = sock->sk;
2166 sock_poll_wait(file, sk_sleep(sk), wait);
2169 /* exceptional events? */
2172 if (sk->sk_shutdown == SHUTDOWN_MASK)
2174 if (sk->sk_shutdown & RCV_SHUTDOWN)
2175 mask |= POLLRDHUP | POLLIN | POLLRDNORM;
2178 if (!skb_queue_empty(&sk->sk_receive_queue))
2179 mask |= POLLIN | POLLRDNORM;
2181 /* Connection-based need to check for termination and startup */
2182 if ((sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) &&
2183 sk->sk_state == TCP_CLOSE)
2187 * we set writable also when the other side has shut down the
2188 * connection. This prevents stuck sockets.
2190 if (unix_writable(sk))
2191 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2196 static unsigned int unix_dgram_poll(struct file *file, struct socket *sock,
2199 struct sock *sk = sock->sk, *other;
2200 unsigned int mask, writable;
2202 sock_poll_wait(file, sk_sleep(sk), wait);
2205 /* exceptional events? */
2206 if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue))
2208 (sock_flag(sk, SOCK_SELECT_ERR_QUEUE) ? POLLPRI : 0);
2210 if (sk->sk_shutdown & RCV_SHUTDOWN)
2211 mask |= POLLRDHUP | POLLIN | POLLRDNORM;
2212 if (sk->sk_shutdown == SHUTDOWN_MASK)
2216 if (!skb_queue_empty(&sk->sk_receive_queue))
2217 mask |= POLLIN | POLLRDNORM;
2219 /* Connection-based need to check for termination and startup */
2220 if (sk->sk_type == SOCK_SEQPACKET) {
2221 if (sk->sk_state == TCP_CLOSE)
2223 /* connection hasn't started yet? */
2224 if (sk->sk_state == TCP_SYN_SENT)
2228 /* No write status requested, avoid expensive OUT tests. */
2229 if (!(poll_requested_events(wait) & (POLLWRBAND|POLLWRNORM|POLLOUT)))
2232 writable = unix_writable(sk);
2233 other = unix_peer_get(sk);
2235 if (unix_peer(other) != sk) {
2236 sock_poll_wait(file, &unix_sk(other)->peer_wait, wait);
2237 if (unix_recvq_full(other))
2244 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2246 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
2251 #ifdef CONFIG_PROC_FS
2253 #define BUCKET_SPACE (BITS_PER_LONG - (UNIX_HASH_BITS + 1) - 1)
2255 #define get_bucket(x) ((x) >> BUCKET_SPACE)
2256 #define get_offset(x) ((x) & ((1L << BUCKET_SPACE) - 1))
2257 #define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
2259 static struct sock *unix_from_bucket(struct seq_file *seq, loff_t *pos)
2261 unsigned long offset = get_offset(*pos);
2262 unsigned long bucket = get_bucket(*pos);
2264 unsigned long count = 0;
2266 for (sk = sk_head(&unix_socket_table[bucket]); sk; sk = sk_next(sk)) {
2267 if (sock_net(sk) != seq_file_net(seq))
2269 if (++count == offset)
2276 static struct sock *unix_next_socket(struct seq_file *seq,
2280 unsigned long bucket;
2282 while (sk > (struct sock *)SEQ_START_TOKEN) {
2286 if (sock_net(sk) == seq_file_net(seq))
2291 sk = unix_from_bucket(seq, pos);
2296 bucket = get_bucket(*pos) + 1;
2297 *pos = set_bucket_offset(bucket, 1);
2298 } while (bucket < ARRAY_SIZE(unix_socket_table));
2303 static void *unix_seq_start(struct seq_file *seq, loff_t *pos)
2304 __acquires(unix_table_lock)
2306 spin_lock(&unix_table_lock);
2309 return SEQ_START_TOKEN;
2311 if (get_bucket(*pos) >= ARRAY_SIZE(unix_socket_table))
2314 return unix_next_socket(seq, NULL, pos);
2317 static void *unix_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2320 return unix_next_socket(seq, v, pos);
2323 static void unix_seq_stop(struct seq_file *seq, void *v)
2324 __releases(unix_table_lock)
2326 spin_unlock(&unix_table_lock);
2329 static int unix_seq_show(struct seq_file *seq, void *v)
2332 if (v == SEQ_START_TOKEN)
2333 seq_puts(seq, "Num RefCount Protocol Flags Type St "
2337 struct unix_sock *u = unix_sk(s);
2340 seq_printf(seq, "%pK: %08X %08X %08X %04X %02X %5lu",
2342 atomic_read(&s->sk_refcnt),
2344 s->sk_state == TCP_LISTEN ? __SO_ACCEPTCON : 0,
2347 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTED : SS_UNCONNECTED) :
2348 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTING : SS_DISCONNECTING),
2356 len = u->addr->len - sizeof(short);
2357 if (!UNIX_ABSTRACT(s))
2363 for ( ; i < len; i++)
2364 seq_putc(seq, u->addr->name->sun_path[i]);
2366 unix_state_unlock(s);
2367 seq_putc(seq, '\n');
2373 static const struct seq_operations unix_seq_ops = {
2374 .start = unix_seq_start,
2375 .next = unix_seq_next,
2376 .stop = unix_seq_stop,
2377 .show = unix_seq_show,
2380 static int unix_seq_open(struct inode *inode, struct file *file)
2382 return seq_open_net(inode, file, &unix_seq_ops,
2383 sizeof(struct seq_net_private));
2386 static const struct file_operations unix_seq_fops = {
2387 .owner = THIS_MODULE,
2388 .open = unix_seq_open,
2390 .llseek = seq_lseek,
2391 .release = seq_release_net,
2396 static const struct net_proto_family unix_family_ops = {
2398 .create = unix_create,
2399 .owner = THIS_MODULE,
2403 static int __net_init unix_net_init(struct net *net)
2405 int error = -ENOMEM;
2407 net->unx.sysctl_max_dgram_qlen = 10;
2408 if (unix_sysctl_register(net))
2411 #ifdef CONFIG_PROC_FS
2412 if (!proc_create("unix", 0, net->proc_net, &unix_seq_fops)) {
2413 unix_sysctl_unregister(net);
2422 static void __net_exit unix_net_exit(struct net *net)
2424 unix_sysctl_unregister(net);
2425 remove_proc_entry("unix", net->proc_net);
2428 static struct pernet_operations unix_net_ops = {
2429 .init = unix_net_init,
2430 .exit = unix_net_exit,
2433 static int __init af_unix_init(void)
2437 BUILD_BUG_ON(sizeof(struct unix_skb_parms) > FIELD_SIZEOF(struct sk_buff, cb));
2439 rc = proto_register(&unix_proto, 1);
2441 printk(KERN_CRIT "%s: Cannot create unix_sock SLAB cache!\n",
2446 sock_register(&unix_family_ops);
2447 register_pernet_subsys(&unix_net_ops);
2452 static void __exit af_unix_exit(void)
2454 sock_unregister(PF_UNIX);
2455 proto_unregister(&unix_proto);
2456 unregister_pernet_subsys(&unix_net_ops);
2459 /* Earlier than device_initcall() so that other drivers invoking
2460 request_module() don't end up in a loop when modprobe tries
2461 to use a UNIX socket. But later than subsys_initcall() because
2462 we depend on stuff initialised there */
2463 fs_initcall(af_unix_init);
2464 module_exit(af_unix_exit);
2466 MODULE_LICENSE("GPL");
2467 MODULE_ALIAS_NETPROTO(PF_UNIX);