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 mutex_lock(&u->readlock);
728 addr = kzalloc(sizeof(*addr) + sizeof(short) + 16, GFP_KERNEL);
732 addr->name->sun_family = AF_UNIX;
733 atomic_set(&addr->refcnt, 1);
736 addr->len = sprintf(addr->name->sun_path+1, "%05x", ordernum) + 1 + sizeof(short);
737 addr->hash = unix_hash_fold(csum_partial(addr->name, addr->len, 0));
739 spin_lock(&unix_table_lock);
740 ordernum = (ordernum+1)&0xFFFFF;
742 if (__unix_find_socket_byname(net, addr->name, addr->len, sock->type,
744 spin_unlock(&unix_table_lock);
746 * __unix_find_socket_byname() may take long time if many names
747 * are already in use.
750 /* Give up if all names seems to be in use. */
751 if (retries++ == 0xFFFFF) {
758 addr->hash ^= sk->sk_type;
760 __unix_remove_socket(sk);
762 __unix_insert_socket(&unix_socket_table[addr->hash], sk);
763 spin_unlock(&unix_table_lock);
766 out: mutex_unlock(&u->readlock);
770 static struct sock *unix_find_other(struct net *net,
771 struct sockaddr_un *sunname, int len,
772 int type, unsigned int hash, int *error)
778 if (sunname->sun_path[0]) {
780 err = kern_path(sunname->sun_path, LOOKUP_FOLLOW, &path);
783 inode = path.dentry->d_inode;
784 err = inode_permission(inode, MAY_WRITE);
789 if (!S_ISSOCK(inode->i_mode))
791 u = unix_find_socket_byinode(inode);
795 if (u->sk_type == type)
801 if (u->sk_type != type) {
807 u = unix_find_socket_byname(net, sunname, len, type, hash);
809 struct dentry *dentry;
810 dentry = unix_sk(u)->path.dentry;
812 touch_atime(&unix_sk(u)->path);
825 static int unix_mknod(const char *sun_path, umode_t mode, struct path *res)
827 struct dentry *dentry;
831 * Get the parent directory, calculate the hash for last
834 dentry = kern_path_create(AT_FDCWD, sun_path, &path, 0);
835 err = PTR_ERR(dentry);
840 * All right, let's create it.
842 err = security_path_mknod(&path, dentry, mode, 0);
844 err = vfs_mknod(path.dentry->d_inode, dentry, mode, 0);
846 res->mnt = mntget(path.mnt);
847 res->dentry = dget(dentry);
850 done_path_create(&path, dentry);
854 static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
856 struct sock *sk = sock->sk;
857 struct net *net = sock_net(sk);
858 struct unix_sock *u = unix_sk(sk);
859 struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
860 char *sun_path = sunaddr->sun_path;
863 struct unix_address *addr;
864 struct hlist_head *list;
867 if (sunaddr->sun_family != AF_UNIX)
870 if (addr_len == sizeof(short)) {
871 err = unix_autobind(sock);
875 err = unix_mkname(sunaddr, addr_len, &hash);
880 mutex_lock(&u->readlock);
887 addr = kmalloc(sizeof(*addr)+addr_len, GFP_KERNEL);
891 memcpy(addr->name, sunaddr, addr_len);
892 addr->len = addr_len;
893 addr->hash = hash ^ sk->sk_type;
894 atomic_set(&addr->refcnt, 1);
898 umode_t mode = S_IFSOCK |
899 (SOCK_INODE(sock)->i_mode & ~current_umask());
900 err = unix_mknod(sun_path, mode, &path);
904 unix_release_addr(addr);
907 addr->hash = UNIX_HASH_SIZE;
908 hash = path.dentry->d_inode->i_ino & (UNIX_HASH_SIZE-1);
909 spin_lock(&unix_table_lock);
911 list = &unix_socket_table[hash];
913 spin_lock(&unix_table_lock);
915 if (__unix_find_socket_byname(net, sunaddr, addr_len,
916 sk->sk_type, hash)) {
917 unix_release_addr(addr);
921 list = &unix_socket_table[addr->hash];
925 __unix_remove_socket(sk);
927 __unix_insert_socket(list, sk);
930 spin_unlock(&unix_table_lock);
932 mutex_unlock(&u->readlock);
937 static void unix_state_double_lock(struct sock *sk1, struct sock *sk2)
939 if (unlikely(sk1 == sk2) || !sk2) {
940 unix_state_lock(sk1);
944 unix_state_lock(sk1);
945 unix_state_lock_nested(sk2);
947 unix_state_lock(sk2);
948 unix_state_lock_nested(sk1);
952 static void unix_state_double_unlock(struct sock *sk1, struct sock *sk2)
954 if (unlikely(sk1 == sk2) || !sk2) {
955 unix_state_unlock(sk1);
958 unix_state_unlock(sk1);
959 unix_state_unlock(sk2);
962 static int unix_dgram_connect(struct socket *sock, struct sockaddr *addr,
965 struct sock *sk = sock->sk;
966 struct net *net = sock_net(sk);
967 struct sockaddr_un *sunaddr = (struct sockaddr_un *)addr;
972 if (addr->sa_family != AF_UNSPEC) {
973 err = unix_mkname(sunaddr, alen, &hash);
978 if (test_bit(SOCK_PASSCRED, &sock->flags) &&
979 !unix_sk(sk)->addr && (err = unix_autobind(sock)) != 0)
983 other = unix_find_other(net, sunaddr, alen, sock->type, hash, &err);
987 unix_state_double_lock(sk, other);
989 /* Apparently VFS overslept socket death. Retry. */
990 if (sock_flag(other, SOCK_DEAD)) {
991 unix_state_double_unlock(sk, other);
997 if (!unix_may_send(sk, other))
1000 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1006 * 1003.1g breaking connected state with AF_UNSPEC
1009 unix_state_double_lock(sk, other);
1013 * If it was connected, reconnect.
1015 if (unix_peer(sk)) {
1016 struct sock *old_peer = unix_peer(sk);
1017 unix_peer(sk) = other;
1018 unix_state_double_unlock(sk, other);
1020 if (other != old_peer)
1021 unix_dgram_disconnected(sk, old_peer);
1024 unix_peer(sk) = other;
1025 unix_state_double_unlock(sk, other);
1030 unix_state_double_unlock(sk, other);
1036 static long unix_wait_for_peer(struct sock *other, long timeo)
1038 struct unix_sock *u = unix_sk(other);
1042 prepare_to_wait_exclusive(&u->peer_wait, &wait, TASK_INTERRUPTIBLE);
1044 sched = !sock_flag(other, SOCK_DEAD) &&
1045 !(other->sk_shutdown & RCV_SHUTDOWN) &&
1046 unix_recvq_full(other);
1048 unix_state_unlock(other);
1051 timeo = schedule_timeout(timeo);
1053 finish_wait(&u->peer_wait, &wait);
1057 static int unix_stream_connect(struct socket *sock, struct sockaddr *uaddr,
1058 int addr_len, int flags)
1060 struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
1061 struct sock *sk = sock->sk;
1062 struct net *net = sock_net(sk);
1063 struct unix_sock *u = unix_sk(sk), *newu, *otheru;
1064 struct sock *newsk = NULL;
1065 struct sock *other = NULL;
1066 struct sk_buff *skb = NULL;
1072 err = unix_mkname(sunaddr, addr_len, &hash);
1077 if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr &&
1078 (err = unix_autobind(sock)) != 0)
1081 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
1083 /* First of all allocate resources.
1084 If we will make it after state is locked,
1085 we will have to recheck all again in any case.
1090 /* create new sock for complete connection */
1091 newsk = unix_create1(sock_net(sk), NULL);
1095 /* Allocate skb for sending to listening sock */
1096 skb = sock_wmalloc(newsk, 1, 0, GFP_KERNEL);
1101 /* Find listening sock. */
1102 other = unix_find_other(net, sunaddr, addr_len, sk->sk_type, hash, &err);
1106 /* Latch state of peer */
1107 unix_state_lock(other);
1109 /* Apparently VFS overslept socket death. Retry. */
1110 if (sock_flag(other, SOCK_DEAD)) {
1111 unix_state_unlock(other);
1116 err = -ECONNREFUSED;
1117 if (other->sk_state != TCP_LISTEN)
1119 if (other->sk_shutdown & RCV_SHUTDOWN)
1122 if (unix_recvq_full(other)) {
1127 timeo = unix_wait_for_peer(other, timeo);
1129 err = sock_intr_errno(timeo);
1130 if (signal_pending(current))
1138 It is tricky place. We need to grab our state lock and cannot
1139 drop lock on peer. It is dangerous because deadlock is
1140 possible. Connect to self case and simultaneous
1141 attempt to connect are eliminated by checking socket
1142 state. other is TCP_LISTEN, if sk is TCP_LISTEN we
1143 check this before attempt to grab lock.
1145 Well, and we have to recheck the state after socket locked.
1151 /* This is ok... continue with connect */
1153 case TCP_ESTABLISHED:
1154 /* Socket is already connected */
1162 unix_state_lock_nested(sk);
1164 if (sk->sk_state != st) {
1165 unix_state_unlock(sk);
1166 unix_state_unlock(other);
1171 err = security_unix_stream_connect(sk, other, newsk);
1173 unix_state_unlock(sk);
1177 /* The way is open! Fastly set all the necessary fields... */
1180 unix_peer(newsk) = sk;
1181 newsk->sk_state = TCP_ESTABLISHED;
1182 newsk->sk_type = sk->sk_type;
1183 init_peercred(newsk);
1184 newu = unix_sk(newsk);
1185 RCU_INIT_POINTER(newsk->sk_wq, &newu->peer_wq);
1186 otheru = unix_sk(other);
1188 /* copy address information from listening to new sock*/
1190 atomic_inc(&otheru->addr->refcnt);
1191 newu->addr = otheru->addr;
1193 if (otheru->path.dentry) {
1194 path_get(&otheru->path);
1195 newu->path = otheru->path;
1198 /* Set credentials */
1199 copy_peercred(sk, other);
1201 sock->state = SS_CONNECTED;
1202 sk->sk_state = TCP_ESTABLISHED;
1205 smp_mb__after_atomic_inc(); /* sock_hold() does an atomic_inc() */
1206 unix_peer(sk) = newsk;
1208 unix_state_unlock(sk);
1210 /* take ten and and send info to listening sock */
1211 spin_lock(&other->sk_receive_queue.lock);
1212 __skb_queue_tail(&other->sk_receive_queue, skb);
1213 spin_unlock(&other->sk_receive_queue.lock);
1214 unix_state_unlock(other);
1215 other->sk_data_ready(other, 0);
1221 unix_state_unlock(other);
1226 unix_release_sock(newsk, 0);
1232 static int unix_socketpair(struct socket *socka, struct socket *sockb)
1234 struct sock *ska = socka->sk, *skb = sockb->sk;
1236 /* Join our sockets back to back */
1239 unix_peer(ska) = skb;
1240 unix_peer(skb) = ska;
1244 if (ska->sk_type != SOCK_DGRAM) {
1245 ska->sk_state = TCP_ESTABLISHED;
1246 skb->sk_state = TCP_ESTABLISHED;
1247 socka->state = SS_CONNECTED;
1248 sockb->state = SS_CONNECTED;
1253 static void unix_sock_inherit_flags(const struct socket *old,
1256 if (test_bit(SOCK_PASSCRED, &old->flags))
1257 set_bit(SOCK_PASSCRED, &new->flags);
1258 if (test_bit(SOCK_PASSSEC, &old->flags))
1259 set_bit(SOCK_PASSSEC, &new->flags);
1262 static int unix_accept(struct socket *sock, struct socket *newsock, int flags)
1264 struct sock *sk = sock->sk;
1266 struct sk_buff *skb;
1270 if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
1274 if (sk->sk_state != TCP_LISTEN)
1277 /* If socket state is TCP_LISTEN it cannot change (for now...),
1278 * so that no locks are necessary.
1281 skb = skb_recv_datagram(sk, 0, flags&O_NONBLOCK, &err);
1283 /* This means receive shutdown. */
1290 skb_free_datagram(sk, skb);
1291 wake_up_interruptible(&unix_sk(sk)->peer_wait);
1293 /* attach accepted sock to socket */
1294 unix_state_lock(tsk);
1295 newsock->state = SS_CONNECTED;
1296 unix_sock_inherit_flags(sock, newsock);
1297 sock_graft(tsk, newsock);
1298 unix_state_unlock(tsk);
1306 static int unix_getname(struct socket *sock, struct sockaddr *uaddr, int *uaddr_len, int peer)
1308 struct sock *sk = sock->sk;
1309 struct unix_sock *u;
1310 DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr, uaddr);
1314 sk = unix_peer_get(sk);
1325 unix_state_lock(sk);
1327 sunaddr->sun_family = AF_UNIX;
1328 sunaddr->sun_path[0] = 0;
1329 *uaddr_len = sizeof(short);
1331 struct unix_address *addr = u->addr;
1333 *uaddr_len = addr->len;
1334 memcpy(sunaddr, addr->name, *uaddr_len);
1336 unix_state_unlock(sk);
1342 static void unix_detach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1346 scm->fp = UNIXCB(skb).fp;
1347 UNIXCB(skb).fp = NULL;
1349 for (i = scm->fp->count-1; i >= 0; i--)
1350 unix_notinflight(scm->fp->fp[i]);
1353 static void unix_destruct_scm(struct sk_buff *skb)
1355 struct scm_cookie scm;
1356 memset(&scm, 0, sizeof(scm));
1357 scm.pid = UNIXCB(skb).pid;
1359 unix_detach_fds(&scm, skb);
1361 /* Alas, it calls VFS */
1362 /* So fscking what? fput() had been SMP-safe since the last Summer */
1367 #define MAX_RECURSION_LEVEL 4
1369 static int unix_attach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1372 unsigned char max_level = 0;
1373 int unix_sock_count = 0;
1375 for (i = scm->fp->count - 1; i >= 0; i--) {
1376 struct sock *sk = unix_get_socket(scm->fp->fp[i]);
1380 max_level = max(max_level,
1381 unix_sk(sk)->recursion_level);
1384 if (unlikely(max_level > MAX_RECURSION_LEVEL))
1385 return -ETOOMANYREFS;
1388 * Need to duplicate file references for the sake of garbage
1389 * collection. Otherwise a socket in the fps might become a
1390 * candidate for GC while the skb is not yet queued.
1392 UNIXCB(skb).fp = scm_fp_dup(scm->fp);
1393 if (!UNIXCB(skb).fp)
1396 if (unix_sock_count) {
1397 for (i = scm->fp->count - 1; i >= 0; i--)
1398 unix_inflight(scm->fp->fp[i]);
1403 static int unix_scm_to_skb(struct scm_cookie *scm, struct sk_buff *skb, bool send_fds)
1407 UNIXCB(skb).pid = get_pid(scm->pid);
1408 UNIXCB(skb).uid = scm->creds.uid;
1409 UNIXCB(skb).gid = scm->creds.gid;
1410 UNIXCB(skb).fp = NULL;
1411 if (scm->fp && send_fds)
1412 err = unix_attach_fds(scm, skb);
1414 skb->destructor = unix_destruct_scm;
1419 * Some apps rely on write() giving SCM_CREDENTIALS
1420 * We include credentials if source or destination socket
1421 * asserted SOCK_PASSCRED.
1423 static void maybe_add_creds(struct sk_buff *skb, const struct socket *sock,
1424 const struct sock *other)
1426 if (UNIXCB(skb).pid)
1428 if (test_bit(SOCK_PASSCRED, &sock->flags) ||
1429 !other->sk_socket ||
1430 test_bit(SOCK_PASSCRED, &other->sk_socket->flags)) {
1431 UNIXCB(skb).pid = get_pid(task_tgid(current));
1432 current_uid_gid(&UNIXCB(skb).uid, &UNIXCB(skb).gid);
1437 * Send AF_UNIX data.
1440 static int unix_dgram_sendmsg(struct kiocb *kiocb, struct socket *sock,
1441 struct msghdr *msg, size_t len)
1443 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1444 struct sock *sk = sock->sk;
1445 struct net *net = sock_net(sk);
1446 struct unix_sock *u = unix_sk(sk);
1447 struct sockaddr_un *sunaddr = msg->msg_name;
1448 struct sock *other = NULL;
1449 int namelen = 0; /* fake GCC */
1452 struct sk_buff *skb;
1454 struct scm_cookie tmp_scm;
1458 if (NULL == siocb->scm)
1459 siocb->scm = &tmp_scm;
1461 err = scm_send(sock, msg, siocb->scm, false);
1466 if (msg->msg_flags&MSG_OOB)
1469 if (msg->msg_namelen) {
1470 err = unix_mkname(sunaddr, msg->msg_namelen, &hash);
1477 other = unix_peer_get(sk);
1482 if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr
1483 && (err = unix_autobind(sock)) != 0)
1487 if (len > sk->sk_sndbuf - 32)
1490 if (len > SKB_MAX_ALLOC)
1491 data_len = min_t(size_t,
1492 len - SKB_MAX_ALLOC,
1493 MAX_SKB_FRAGS * PAGE_SIZE);
1495 skb = sock_alloc_send_pskb(sk, len - data_len, data_len,
1496 msg->msg_flags & MSG_DONTWAIT, &err,
1497 PAGE_ALLOC_COSTLY_ORDER);
1501 err = unix_scm_to_skb(siocb->scm, skb, true);
1504 max_level = err + 1;
1505 unix_get_secdata(siocb->scm, skb);
1507 skb_put(skb, len - data_len);
1508 skb->data_len = data_len;
1510 err = skb_copy_datagram_from_iovec(skb, 0, msg->msg_iov, 0, len);
1514 timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1519 if (sunaddr == NULL)
1522 other = unix_find_other(net, sunaddr, namelen, sk->sk_type,
1528 if (sk_filter(other, skb) < 0) {
1529 /* Toss the packet but do not return any error to the sender */
1534 unix_state_lock(other);
1536 if (!unix_may_send(sk, other))
1539 if (sock_flag(other, SOCK_DEAD)) {
1541 * Check with 1003.1g - what should
1544 unix_state_unlock(other);
1548 unix_state_lock(sk);
1549 if (unix_peer(sk) == other) {
1550 unix_peer(sk) = NULL;
1551 unix_state_unlock(sk);
1553 unix_dgram_disconnected(sk, other);
1555 err = -ECONNREFUSED;
1557 unix_state_unlock(sk);
1567 if (other->sk_shutdown & RCV_SHUTDOWN)
1570 if (sk->sk_type != SOCK_SEQPACKET) {
1571 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1576 if (unix_peer(other) != sk && unix_recvq_full(other)) {
1582 timeo = unix_wait_for_peer(other, timeo);
1584 err = sock_intr_errno(timeo);
1585 if (signal_pending(current))
1591 if (sock_flag(other, SOCK_RCVTSTAMP))
1592 __net_timestamp(skb);
1593 maybe_add_creds(skb, sock, other);
1594 skb_queue_tail(&other->sk_receive_queue, skb);
1595 if (max_level > unix_sk(other)->recursion_level)
1596 unix_sk(other)->recursion_level = max_level;
1597 unix_state_unlock(other);
1598 other->sk_data_ready(other, len);
1600 scm_destroy(siocb->scm);
1604 unix_state_unlock(other);
1610 scm_destroy(siocb->scm);
1614 /* We use paged skbs for stream sockets, and limit occupancy to 32768
1615 * bytes, and a minimun of a full page.
1617 #define UNIX_SKB_FRAGS_SZ (PAGE_SIZE << get_order(32768))
1619 static int unix_stream_sendmsg(struct kiocb *kiocb, struct socket *sock,
1620 struct msghdr *msg, size_t len)
1622 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1623 struct sock *sk = sock->sk;
1624 struct sock *other = NULL;
1626 struct sk_buff *skb;
1628 struct scm_cookie tmp_scm;
1629 bool fds_sent = false;
1633 if (NULL == siocb->scm)
1634 siocb->scm = &tmp_scm;
1636 err = scm_send(sock, msg, siocb->scm, false);
1641 if (msg->msg_flags&MSG_OOB)
1644 if (msg->msg_namelen) {
1645 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
1649 other = unix_peer(sk);
1654 if (sk->sk_shutdown & SEND_SHUTDOWN)
1657 while (sent < len) {
1660 /* Keep two messages in the pipe so it schedules better */
1661 size = min_t(int, size, (sk->sk_sndbuf >> 1) - 64);
1663 /* allow fallback to order-0 allocations */
1664 size = min_t(int, size, SKB_MAX_HEAD(0) + UNIX_SKB_FRAGS_SZ);
1666 data_len = max_t(int, 0, size - SKB_MAX_HEAD(0));
1668 skb = sock_alloc_send_pskb(sk, size - data_len, data_len,
1669 msg->msg_flags & MSG_DONTWAIT, &err,
1670 get_order(UNIX_SKB_FRAGS_SZ));
1674 /* Only send the fds in the first buffer */
1675 err = unix_scm_to_skb(siocb->scm, skb, !fds_sent);
1680 max_level = err + 1;
1683 skb_put(skb, size - data_len);
1684 skb->data_len = data_len;
1686 err = skb_copy_datagram_from_iovec(skb, 0, msg->msg_iov,
1693 unix_state_lock(other);
1695 if (sock_flag(other, SOCK_DEAD) ||
1696 (other->sk_shutdown & RCV_SHUTDOWN))
1699 maybe_add_creds(skb, sock, other);
1700 skb_queue_tail(&other->sk_receive_queue, skb);
1701 if (max_level > unix_sk(other)->recursion_level)
1702 unix_sk(other)->recursion_level = max_level;
1703 unix_state_unlock(other);
1704 other->sk_data_ready(other, size);
1708 scm_destroy(siocb->scm);
1714 unix_state_unlock(other);
1717 if (sent == 0 && !(msg->msg_flags&MSG_NOSIGNAL))
1718 send_sig(SIGPIPE, current, 0);
1721 scm_destroy(siocb->scm);
1723 return sent ? : err;
1726 static int unix_seqpacket_sendmsg(struct kiocb *kiocb, struct socket *sock,
1727 struct msghdr *msg, size_t len)
1730 struct sock *sk = sock->sk;
1732 err = sock_error(sk);
1736 if (sk->sk_state != TCP_ESTABLISHED)
1739 if (msg->msg_namelen)
1740 msg->msg_namelen = 0;
1742 return unix_dgram_sendmsg(kiocb, sock, msg, len);
1745 static int unix_seqpacket_recvmsg(struct kiocb *iocb, struct socket *sock,
1746 struct msghdr *msg, size_t size,
1749 struct sock *sk = sock->sk;
1751 if (sk->sk_state != TCP_ESTABLISHED)
1754 return unix_dgram_recvmsg(iocb, sock, msg, size, flags);
1757 static void unix_copy_addr(struct msghdr *msg, struct sock *sk)
1759 struct unix_sock *u = unix_sk(sk);
1762 msg->msg_namelen = u->addr->len;
1763 memcpy(msg->msg_name, u->addr->name, u->addr->len);
1767 static int unix_dgram_recvmsg(struct kiocb *iocb, struct socket *sock,
1768 struct msghdr *msg, size_t size,
1771 struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1772 struct scm_cookie tmp_scm;
1773 struct sock *sk = sock->sk;
1774 struct unix_sock *u = unix_sk(sk);
1775 int noblock = flags & MSG_DONTWAIT;
1776 struct sk_buff *skb;
1784 err = mutex_lock_interruptible(&u->readlock);
1786 err = sock_intr_errno(sock_rcvtimeo(sk, noblock));
1790 skip = sk_peek_offset(sk, flags);
1792 skb = __skb_recv_datagram(sk, flags, &peeked, &skip, &err);
1794 unix_state_lock(sk);
1795 /* Signal EOF on disconnected non-blocking SEQPACKET socket. */
1796 if (sk->sk_type == SOCK_SEQPACKET && err == -EAGAIN &&
1797 (sk->sk_shutdown & RCV_SHUTDOWN))
1799 unix_state_unlock(sk);
1803 wake_up_interruptible_sync_poll(&u->peer_wait,
1804 POLLOUT | POLLWRNORM | POLLWRBAND);
1807 unix_copy_addr(msg, skb->sk);
1809 if (size > skb->len - skip)
1810 size = skb->len - skip;
1811 else if (size < skb->len - skip)
1812 msg->msg_flags |= MSG_TRUNC;
1814 err = skb_copy_datagram_iovec(skb, skip, msg->msg_iov, size);
1818 if (sock_flag(sk, SOCK_RCVTSTAMP))
1819 __sock_recv_timestamp(msg, sk, skb);
1822 siocb->scm = &tmp_scm;
1823 memset(&tmp_scm, 0, sizeof(tmp_scm));
1825 scm_set_cred(siocb->scm, UNIXCB(skb).pid, UNIXCB(skb).uid, UNIXCB(skb).gid);
1826 unix_set_secdata(siocb->scm, skb);
1828 if (!(flags & MSG_PEEK)) {
1830 unix_detach_fds(siocb->scm, skb);
1832 sk_peek_offset_bwd(sk, skb->len);
1834 /* It is questionable: on PEEK we could:
1835 - do not return fds - good, but too simple 8)
1836 - return fds, and do not return them on read (old strategy,
1838 - clone fds (I chose it for now, it is the most universal
1841 POSIX 1003.1g does not actually define this clearly
1842 at all. POSIX 1003.1g doesn't define a lot of things
1847 sk_peek_offset_fwd(sk, size);
1850 siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1852 err = (flags & MSG_TRUNC) ? skb->len - skip : size;
1854 scm_recv(sock, msg, siocb->scm, flags);
1857 skb_free_datagram(sk, skb);
1859 mutex_unlock(&u->readlock);
1865 * Sleep until more data has arrived. But check for races..
1867 static long unix_stream_data_wait(struct sock *sk, long timeo,
1868 struct sk_buff *last)
1872 unix_state_lock(sk);
1875 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1877 if (skb_peek_tail(&sk->sk_receive_queue) != last ||
1879 (sk->sk_shutdown & RCV_SHUTDOWN) ||
1880 signal_pending(current) ||
1884 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1885 unix_state_unlock(sk);
1886 timeo = freezable_schedule_timeout(timeo);
1887 unix_state_lock(sk);
1888 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1891 finish_wait(sk_sleep(sk), &wait);
1892 unix_state_unlock(sk);
1896 static unsigned int unix_skb_len(const struct sk_buff *skb)
1898 return skb->len - UNIXCB(skb).consumed;
1901 static int unix_stream_recvmsg(struct kiocb *iocb, struct socket *sock,
1902 struct msghdr *msg, size_t size,
1905 struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1906 struct scm_cookie tmp_scm;
1907 struct sock *sk = sock->sk;
1908 struct unix_sock *u = unix_sk(sk);
1909 struct sockaddr_un *sunaddr = msg->msg_name;
1911 int check_creds = 0;
1918 if (sk->sk_state != TCP_ESTABLISHED)
1925 target = sock_rcvlowat(sk, flags&MSG_WAITALL, size);
1926 timeo = sock_rcvtimeo(sk, flags&MSG_DONTWAIT);
1928 /* Lock the socket to prevent queue disordering
1929 * while sleeps in memcpy_tomsg
1933 siocb->scm = &tmp_scm;
1934 memset(&tmp_scm, 0, sizeof(tmp_scm));
1937 err = mutex_lock_interruptible(&u->readlock);
1939 err = sock_intr_errno(timeo);
1945 struct sk_buff *skb, *last;
1947 unix_state_lock(sk);
1948 last = skb = skb_peek(&sk->sk_receive_queue);
1951 unix_sk(sk)->recursion_level = 0;
1952 if (copied >= target)
1956 * POSIX 1003.1g mandates this order.
1959 err = sock_error(sk);
1962 if (sk->sk_shutdown & RCV_SHUTDOWN)
1965 unix_state_unlock(sk);
1969 mutex_unlock(&u->readlock);
1971 timeo = unix_stream_data_wait(sk, timeo, last);
1973 if (signal_pending(current)
1974 || mutex_lock_interruptible(&u->readlock)) {
1975 err = sock_intr_errno(timeo);
1981 unix_state_unlock(sk);
1985 skip = sk_peek_offset(sk, flags);
1986 while (skip >= unix_skb_len(skb)) {
1987 skip -= unix_skb_len(skb);
1989 skb = skb_peek_next(skb, &sk->sk_receive_queue);
1994 unix_state_unlock(sk);
1997 /* Never glue messages from different writers */
1998 if ((UNIXCB(skb).pid != siocb->scm->pid) ||
1999 !uid_eq(UNIXCB(skb).uid, siocb->scm->creds.uid) ||
2000 !gid_eq(UNIXCB(skb).gid, siocb->scm->creds.gid))
2002 } else if (test_bit(SOCK_PASSCRED, &sock->flags)) {
2003 /* Copy credentials */
2004 scm_set_cred(siocb->scm, UNIXCB(skb).pid, UNIXCB(skb).uid, UNIXCB(skb).gid);
2008 /* Copy address just once */
2010 unix_copy_addr(msg, skb->sk);
2014 chunk = min_t(unsigned int, unix_skb_len(skb) - skip, size);
2015 if (skb_copy_datagram_iovec(skb, UNIXCB(skb).consumed + skip,
2016 msg->msg_iov, chunk)) {
2024 /* Mark read part of skb as used */
2025 if (!(flags & MSG_PEEK)) {
2026 UNIXCB(skb).consumed += chunk;
2028 sk_peek_offset_bwd(sk, chunk);
2031 unix_detach_fds(siocb->scm, skb);
2033 if (unix_skb_len(skb))
2036 skb_unlink(skb, &sk->sk_receive_queue);
2042 /* It is questionable, see note in unix_dgram_recvmsg.
2045 siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
2047 sk_peek_offset_fwd(sk, chunk);
2053 mutex_unlock(&u->readlock);
2054 scm_recv(sock, msg, siocb->scm, flags);
2056 return copied ? : err;
2059 static int unix_shutdown(struct socket *sock, int mode)
2061 struct sock *sk = sock->sk;
2064 if (mode < SHUT_RD || mode > SHUT_RDWR)
2067 * SHUT_RD (0) -> RCV_SHUTDOWN (1)
2068 * SHUT_WR (1) -> SEND_SHUTDOWN (2)
2069 * SHUT_RDWR (2) -> SHUTDOWN_MASK (3)
2073 unix_state_lock(sk);
2074 sk->sk_shutdown |= mode;
2075 other = unix_peer(sk);
2078 unix_state_unlock(sk);
2079 sk->sk_state_change(sk);
2082 (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET)) {
2086 if (mode&RCV_SHUTDOWN)
2087 peer_mode |= SEND_SHUTDOWN;
2088 if (mode&SEND_SHUTDOWN)
2089 peer_mode |= RCV_SHUTDOWN;
2090 unix_state_lock(other);
2091 other->sk_shutdown |= peer_mode;
2092 unix_state_unlock(other);
2093 other->sk_state_change(other);
2094 if (peer_mode == SHUTDOWN_MASK)
2095 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_HUP);
2096 else if (peer_mode & RCV_SHUTDOWN)
2097 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_IN);
2105 long unix_inq_len(struct sock *sk)
2107 struct sk_buff *skb;
2110 if (sk->sk_state == TCP_LISTEN)
2113 spin_lock(&sk->sk_receive_queue.lock);
2114 if (sk->sk_type == SOCK_STREAM ||
2115 sk->sk_type == SOCK_SEQPACKET) {
2116 skb_queue_walk(&sk->sk_receive_queue, skb)
2117 amount += unix_skb_len(skb);
2119 skb = skb_peek(&sk->sk_receive_queue);
2123 spin_unlock(&sk->sk_receive_queue.lock);
2127 EXPORT_SYMBOL_GPL(unix_inq_len);
2129 long unix_outq_len(struct sock *sk)
2131 return sk_wmem_alloc_get(sk);
2133 EXPORT_SYMBOL_GPL(unix_outq_len);
2135 static int unix_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2137 struct sock *sk = sock->sk;
2143 amount = unix_outq_len(sk);
2144 err = put_user(amount, (int __user *)arg);
2147 amount = unix_inq_len(sk);
2151 err = put_user(amount, (int __user *)arg);
2160 static unsigned int unix_poll(struct file *file, struct socket *sock, poll_table *wait)
2162 struct sock *sk = sock->sk;
2165 sock_poll_wait(file, sk_sleep(sk), wait);
2168 /* exceptional events? */
2171 if (sk->sk_shutdown == SHUTDOWN_MASK)
2173 if (sk->sk_shutdown & RCV_SHUTDOWN)
2174 mask |= POLLRDHUP | POLLIN | POLLRDNORM;
2177 if (!skb_queue_empty(&sk->sk_receive_queue))
2178 mask |= POLLIN | POLLRDNORM;
2180 /* Connection-based need to check for termination and startup */
2181 if ((sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) &&
2182 sk->sk_state == TCP_CLOSE)
2186 * we set writable also when the other side has shut down the
2187 * connection. This prevents stuck sockets.
2189 if (unix_writable(sk))
2190 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2195 static unsigned int unix_dgram_poll(struct file *file, struct socket *sock,
2198 struct sock *sk = sock->sk, *other;
2199 unsigned int mask, writable;
2201 sock_poll_wait(file, sk_sleep(sk), wait);
2204 /* exceptional events? */
2205 if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue))
2207 (sock_flag(sk, SOCK_SELECT_ERR_QUEUE) ? POLLPRI : 0);
2209 if (sk->sk_shutdown & RCV_SHUTDOWN)
2210 mask |= POLLRDHUP | POLLIN | POLLRDNORM;
2211 if (sk->sk_shutdown == SHUTDOWN_MASK)
2215 if (!skb_queue_empty(&sk->sk_receive_queue))
2216 mask |= POLLIN | POLLRDNORM;
2218 /* Connection-based need to check for termination and startup */
2219 if (sk->sk_type == SOCK_SEQPACKET) {
2220 if (sk->sk_state == TCP_CLOSE)
2222 /* connection hasn't started yet? */
2223 if (sk->sk_state == TCP_SYN_SENT)
2227 /* No write status requested, avoid expensive OUT tests. */
2228 if (!(poll_requested_events(wait) & (POLLWRBAND|POLLWRNORM|POLLOUT)))
2231 writable = unix_writable(sk);
2232 other = unix_peer_get(sk);
2234 if (unix_peer(other) != sk) {
2235 sock_poll_wait(file, &unix_sk(other)->peer_wait, wait);
2236 if (unix_recvq_full(other))
2243 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2245 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
2250 #ifdef CONFIG_PROC_FS
2252 #define BUCKET_SPACE (BITS_PER_LONG - (UNIX_HASH_BITS + 1) - 1)
2254 #define get_bucket(x) ((x) >> BUCKET_SPACE)
2255 #define get_offset(x) ((x) & ((1L << BUCKET_SPACE) - 1))
2256 #define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
2258 static struct sock *unix_from_bucket(struct seq_file *seq, loff_t *pos)
2260 unsigned long offset = get_offset(*pos);
2261 unsigned long bucket = get_bucket(*pos);
2263 unsigned long count = 0;
2265 for (sk = sk_head(&unix_socket_table[bucket]); sk; sk = sk_next(sk)) {
2266 if (sock_net(sk) != seq_file_net(seq))
2268 if (++count == offset)
2275 static struct sock *unix_next_socket(struct seq_file *seq,
2279 unsigned long bucket;
2281 while (sk > (struct sock *)SEQ_START_TOKEN) {
2285 if (sock_net(sk) == seq_file_net(seq))
2290 sk = unix_from_bucket(seq, pos);
2295 bucket = get_bucket(*pos) + 1;
2296 *pos = set_bucket_offset(bucket, 1);
2297 } while (bucket < ARRAY_SIZE(unix_socket_table));
2302 static void *unix_seq_start(struct seq_file *seq, loff_t *pos)
2303 __acquires(unix_table_lock)
2305 spin_lock(&unix_table_lock);
2308 return SEQ_START_TOKEN;
2310 if (get_bucket(*pos) >= ARRAY_SIZE(unix_socket_table))
2313 return unix_next_socket(seq, NULL, pos);
2316 static void *unix_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2319 return unix_next_socket(seq, v, pos);
2322 static void unix_seq_stop(struct seq_file *seq, void *v)
2323 __releases(unix_table_lock)
2325 spin_unlock(&unix_table_lock);
2328 static int unix_seq_show(struct seq_file *seq, void *v)
2331 if (v == SEQ_START_TOKEN)
2332 seq_puts(seq, "Num RefCount Protocol Flags Type St "
2336 struct unix_sock *u = unix_sk(s);
2339 seq_printf(seq, "%pK: %08X %08X %08X %04X %02X %5lu",
2341 atomic_read(&s->sk_refcnt),
2343 s->sk_state == TCP_LISTEN ? __SO_ACCEPTCON : 0,
2346 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTED : SS_UNCONNECTED) :
2347 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTING : SS_DISCONNECTING),
2355 len = u->addr->len - sizeof(short);
2356 if (!UNIX_ABSTRACT(s))
2362 for ( ; i < len; i++)
2363 seq_putc(seq, u->addr->name->sun_path[i]);
2365 unix_state_unlock(s);
2366 seq_putc(seq, '\n');
2372 static const struct seq_operations unix_seq_ops = {
2373 .start = unix_seq_start,
2374 .next = unix_seq_next,
2375 .stop = unix_seq_stop,
2376 .show = unix_seq_show,
2379 static int unix_seq_open(struct inode *inode, struct file *file)
2381 return seq_open_net(inode, file, &unix_seq_ops,
2382 sizeof(struct seq_net_private));
2385 static const struct file_operations unix_seq_fops = {
2386 .owner = THIS_MODULE,
2387 .open = unix_seq_open,
2389 .llseek = seq_lseek,
2390 .release = seq_release_net,
2395 static const struct net_proto_family unix_family_ops = {
2397 .create = unix_create,
2398 .owner = THIS_MODULE,
2402 static int __net_init unix_net_init(struct net *net)
2404 int error = -ENOMEM;
2406 net->unx.sysctl_max_dgram_qlen = 10;
2407 if (unix_sysctl_register(net))
2410 #ifdef CONFIG_PROC_FS
2411 if (!proc_create("unix", 0, net->proc_net, &unix_seq_fops)) {
2412 unix_sysctl_unregister(net);
2421 static void __net_exit unix_net_exit(struct net *net)
2423 unix_sysctl_unregister(net);
2424 remove_proc_entry("unix", net->proc_net);
2427 static struct pernet_operations unix_net_ops = {
2428 .init = unix_net_init,
2429 .exit = unix_net_exit,
2432 static int __init af_unix_init(void)
2436 BUILD_BUG_ON(sizeof(struct unix_skb_parms) > FIELD_SIZEOF(struct sk_buff, cb));
2438 rc = proto_register(&unix_proto, 1);
2440 printk(KERN_CRIT "%s: Cannot create unix_sock SLAB cache!\n",
2445 sock_register(&unix_family_ops);
2446 register_pernet_subsys(&unix_net_ops);
2451 static void __exit af_unix_exit(void)
2453 sock_unregister(PF_UNIX);
2454 proto_unregister(&unix_proto);
2455 unregister_pernet_subsys(&unix_net_ops);
2458 /* Earlier than device_initcall() so that other drivers invoking
2459 request_module() don't end up in a loop when modprobe tries
2460 to use a UNIX socket. But later than subsys_initcall() because
2461 we depend on stuff initialised there */
2462 fs_initcall(af_unix_init);
2463 module_exit(af_unix_exit);
2465 MODULE_LICENSE("GPL");
2466 MODULE_ALIAS_NETPROTO(PF_UNIX);