2 * NET4: Implementation of BSD Unix domain sockets.
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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
85 #include <linux/module.h>
86 #include <linux/kernel.h>
87 #include <linux/signal.h>
88 #include <linux/sched/signal.h>
89 #include <linux/errno.h>
90 #include <linux/string.h>
91 #include <linux/stat.h>
92 #include <linux/dcache.h>
93 #include <linux/namei.h>
94 #include <linux/socket.h>
96 #include <linux/fcntl.h>
97 #include <linux/termios.h>
98 #include <linux/sockios.h>
99 #include <linux/net.h>
100 #include <linux/in.h>
101 #include <linux/fs.h>
102 #include <linux/slab.h>
103 #include <linux/uaccess.h>
104 #include <linux/skbuff.h>
105 #include <linux/netdevice.h>
106 #include <net/net_namespace.h>
107 #include <net/sock.h>
108 #include <net/tcp_states.h>
109 #include <net/af_unix.h>
110 #include <linux/proc_fs.h>
111 #include <linux/seq_file.h>
113 #include <linux/init.h>
114 #include <linux/poll.h>
115 #include <linux/rtnetlink.h>
116 #include <linux/mount.h>
117 #include <net/checksum.h>
118 #include <linux/security.h>
119 #include <linux/freezer.h>
120 #include <linux/file.h>
122 struct hlist_head unix_socket_table[2 * UNIX_HASH_SIZE];
123 EXPORT_SYMBOL_GPL(unix_socket_table);
124 DEFINE_SPINLOCK(unix_table_lock);
125 EXPORT_SYMBOL_GPL(unix_table_lock);
126 static atomic_long_t unix_nr_socks;
129 static struct hlist_head *unix_sockets_unbound(void *addr)
131 unsigned long hash = (unsigned long)addr;
135 hash %= UNIX_HASH_SIZE;
136 return &unix_socket_table[UNIX_HASH_SIZE + hash];
139 #define UNIX_ABSTRACT(sk) (unix_sk(sk)->addr->hash < UNIX_HASH_SIZE)
141 #ifdef CONFIG_SECURITY_NETWORK
142 static void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb)
144 UNIXCB(skb).secid = scm->secid;
147 static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb)
149 scm->secid = UNIXCB(skb).secid;
152 static inline bool unix_secdata_eq(struct scm_cookie *scm, struct sk_buff *skb)
154 return (scm->secid == UNIXCB(skb).secid);
157 static inline void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb)
160 static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb)
163 static inline bool unix_secdata_eq(struct scm_cookie *scm, struct sk_buff *skb)
167 #endif /* CONFIG_SECURITY_NETWORK */
170 * SMP locking strategy:
171 * hash table is protected with spinlock unix_table_lock
172 * each socket state is protected by separate spin lock.
175 static inline unsigned int unix_hash_fold(__wsum n)
177 unsigned int hash = (__force unsigned int)csum_fold(n);
180 return hash&(UNIX_HASH_SIZE-1);
183 #define unix_peer(sk) (unix_sk(sk)->peer)
185 static inline int unix_our_peer(struct sock *sk, struct sock *osk)
187 return unix_peer(osk) == sk;
190 static inline int unix_may_send(struct sock *sk, struct sock *osk)
192 return unix_peer(osk) == NULL || unix_our_peer(sk, osk);
195 static inline int unix_recvq_full(struct sock const *sk)
197 return skb_queue_len(&sk->sk_receive_queue) > sk->sk_max_ack_backlog;
200 struct sock *unix_peer_get(struct sock *s)
208 unix_state_unlock(s);
211 EXPORT_SYMBOL_GPL(unix_peer_get);
213 static inline void unix_release_addr(struct unix_address *addr)
215 if (refcount_dec_and_test(&addr->refcnt))
220 * Check unix socket name:
221 * - should be not zero length.
222 * - if started by not zero, should be NULL terminated (FS object)
223 * - if started by zero, it is abstract name.
226 static int unix_mkname(struct sockaddr_un *sunaddr, int len, unsigned int *hashp)
230 if (len <= sizeof(short) || len > sizeof(*sunaddr))
232 if (!sunaddr || sunaddr->sun_family != AF_UNIX)
234 if (sunaddr->sun_path[0]) {
236 * This may look like an off by one error but it is a bit more
237 * subtle. 108 is the longest valid AF_UNIX path for a binding.
238 * sun_path[108] doesn't as such exist. However in kernel space
239 * we are guaranteed that it is a valid memory location in our
240 * kernel address buffer.
242 ((char *)sunaddr)[len] = 0;
243 len = strlen(sunaddr->sun_path)+1+sizeof(short);
247 *hashp = unix_hash_fold(csum_partial(sunaddr, len, 0));
251 static void __unix_remove_socket(struct sock *sk)
253 sk_del_node_init(sk);
256 static void __unix_insert_socket(struct hlist_head *list, struct sock *sk)
258 WARN_ON(!sk_unhashed(sk));
259 sk_add_node(sk, list);
262 static inline void unix_remove_socket(struct sock *sk)
264 spin_lock(&unix_table_lock);
265 __unix_remove_socket(sk);
266 spin_unlock(&unix_table_lock);
269 static inline void unix_insert_socket(struct hlist_head *list, struct sock *sk)
271 spin_lock(&unix_table_lock);
272 __unix_insert_socket(list, sk);
273 spin_unlock(&unix_table_lock);
276 static struct sock *__unix_find_socket_byname(struct net *net,
277 struct sockaddr_un *sunname,
278 int len, int type, unsigned int hash)
282 sk_for_each(s, &unix_socket_table[hash ^ type]) {
283 struct unix_sock *u = unix_sk(s);
285 if (!net_eq(sock_net(s), net))
288 if (u->addr->len == len &&
289 !memcmp(u->addr->name, sunname, len))
297 static inline struct sock *unix_find_socket_byname(struct net *net,
298 struct sockaddr_un *sunname,
304 spin_lock(&unix_table_lock);
305 s = __unix_find_socket_byname(net, sunname, len, type, hash);
308 spin_unlock(&unix_table_lock);
312 static struct sock *unix_find_socket_byinode(struct inode *i)
316 spin_lock(&unix_table_lock);
318 &unix_socket_table[i->i_ino & (UNIX_HASH_SIZE - 1)]) {
319 struct dentry *dentry = unix_sk(s)->path.dentry;
321 if (dentry && d_backing_inode(dentry) == i) {
328 spin_unlock(&unix_table_lock);
332 /* Support code for asymmetrically connected dgram sockets
334 * If a datagram socket is connected to a socket not itself connected
335 * to the first socket (eg, /dev/log), clients may only enqueue more
336 * messages if the present receive queue of the server socket is not
337 * "too large". This means there's a second writeability condition
338 * poll and sendmsg need to test. The dgram recv code will do a wake
339 * up on the peer_wait wait queue of a socket upon reception of a
340 * datagram which needs to be propagated to sleeping would-be writers
341 * since these might not have sent anything so far. This can't be
342 * accomplished via poll_wait because the lifetime of the server
343 * socket might be less than that of its clients if these break their
344 * association with it or if the server socket is closed while clients
345 * are still connected to it and there's no way to inform "a polling
346 * implementation" that it should let go of a certain wait queue
348 * In order to propagate a wake up, a wait_queue_entry_t of the client
349 * socket is enqueued on the peer_wait queue of the server socket
350 * whose wake function does a wake_up on the ordinary client socket
351 * wait queue. This connection is established whenever a write (or
352 * poll for write) hit the flow control condition and broken when the
353 * association to the server socket is dissolved or after a wake up
357 static int unix_dgram_peer_wake_relay(wait_queue_entry_t *q, unsigned mode, int flags,
361 wait_queue_head_t *u_sleep;
363 u = container_of(q, struct unix_sock, peer_wake);
365 __remove_wait_queue(&unix_sk(u->peer_wake.private)->peer_wait,
367 u->peer_wake.private = NULL;
369 /* relaying can only happen while the wq still exists */
370 u_sleep = sk_sleep(&u->sk);
372 wake_up_interruptible_poll(u_sleep, key_to_poll(key));
377 static int unix_dgram_peer_wake_connect(struct sock *sk, struct sock *other)
379 struct unix_sock *u, *u_other;
383 u_other = unix_sk(other);
385 spin_lock(&u_other->peer_wait.lock);
387 if (!u->peer_wake.private) {
388 u->peer_wake.private = other;
389 __add_wait_queue(&u_other->peer_wait, &u->peer_wake);
394 spin_unlock(&u_other->peer_wait.lock);
398 static void unix_dgram_peer_wake_disconnect(struct sock *sk,
401 struct unix_sock *u, *u_other;
404 u_other = unix_sk(other);
405 spin_lock(&u_other->peer_wait.lock);
407 if (u->peer_wake.private == other) {
408 __remove_wait_queue(&u_other->peer_wait, &u->peer_wake);
409 u->peer_wake.private = NULL;
412 spin_unlock(&u_other->peer_wait.lock);
415 static void unix_dgram_peer_wake_disconnect_wakeup(struct sock *sk,
418 unix_dgram_peer_wake_disconnect(sk, other);
419 wake_up_interruptible_poll(sk_sleep(sk),
426 * - unix_peer(sk) == other
427 * - association is stable
429 static int unix_dgram_peer_wake_me(struct sock *sk, struct sock *other)
433 connected = unix_dgram_peer_wake_connect(sk, other);
435 /* If other is SOCK_DEAD, we want to make sure we signal
436 * POLLOUT, such that a subsequent write() can get a
437 * -ECONNREFUSED. Otherwise, if we haven't queued any skbs
438 * to other and its full, we will hang waiting for POLLOUT.
440 if (unix_recvq_full(other) && !sock_flag(other, SOCK_DEAD))
444 unix_dgram_peer_wake_disconnect(sk, other);
449 static int unix_writable(const struct sock *sk)
451 return sk->sk_state != TCP_LISTEN &&
452 (refcount_read(&sk->sk_wmem_alloc) << 2) <= sk->sk_sndbuf;
455 static void unix_write_space(struct sock *sk)
457 struct socket_wq *wq;
460 if (unix_writable(sk)) {
461 wq = rcu_dereference(sk->sk_wq);
462 if (skwq_has_sleeper(wq))
463 wake_up_interruptible_sync_poll(&wq->wait,
464 EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND);
465 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
470 /* When dgram socket disconnects (or changes its peer), we clear its receive
471 * queue of packets arrived from previous peer. First, it allows to do
472 * flow control based only on wmem_alloc; second, sk connected to peer
473 * may receive messages only from that peer. */
474 static void unix_dgram_disconnected(struct sock *sk, struct sock *other)
476 if (!skb_queue_empty(&sk->sk_receive_queue)) {
477 skb_queue_purge(&sk->sk_receive_queue);
478 wake_up_interruptible_all(&unix_sk(sk)->peer_wait);
480 /* If one link of bidirectional dgram pipe is disconnected,
481 * we signal error. Messages are lost. Do not make this,
482 * when peer was not connected to us.
484 if (!sock_flag(other, SOCK_DEAD) && unix_peer(other) == sk) {
485 other->sk_err = ECONNRESET;
486 other->sk_error_report(other);
491 static void unix_sock_destructor(struct sock *sk)
493 struct unix_sock *u = unix_sk(sk);
495 skb_queue_purge(&sk->sk_receive_queue);
497 WARN_ON(refcount_read(&sk->sk_wmem_alloc));
498 WARN_ON(!sk_unhashed(sk));
499 WARN_ON(sk->sk_socket);
500 if (!sock_flag(sk, SOCK_DEAD)) {
501 pr_info("Attempt to release alive unix socket: %p\n", sk);
506 unix_release_addr(u->addr);
508 atomic_long_dec(&unix_nr_socks);
510 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
512 #ifdef UNIX_REFCNT_DEBUG
513 pr_debug("UNIX %p is destroyed, %ld are still alive.\n", sk,
514 atomic_long_read(&unix_nr_socks));
518 static void unix_release_sock(struct sock *sk, int embrion)
520 struct unix_sock *u = unix_sk(sk);
526 unix_remove_socket(sk);
531 sk->sk_shutdown = SHUTDOWN_MASK;
533 u->path.dentry = NULL;
535 state = sk->sk_state;
536 sk->sk_state = TCP_CLOSE;
537 unix_state_unlock(sk);
539 wake_up_interruptible_all(&u->peer_wait);
541 skpair = unix_peer(sk);
543 if (skpair != NULL) {
544 if (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) {
545 unix_state_lock(skpair);
547 skpair->sk_shutdown = SHUTDOWN_MASK;
548 if (!skb_queue_empty(&sk->sk_receive_queue) || embrion)
549 skpair->sk_err = ECONNRESET;
550 unix_state_unlock(skpair);
551 skpair->sk_state_change(skpair);
552 sk_wake_async(skpair, SOCK_WAKE_WAITD, POLL_HUP);
555 unix_dgram_peer_wake_disconnect(sk, skpair);
556 sock_put(skpair); /* It may now die */
557 unix_peer(sk) = NULL;
560 /* Try to flush out this socket. Throw out buffers at least */
562 while ((skb = skb_dequeue(&sk->sk_receive_queue)) != NULL) {
563 if (state == TCP_LISTEN)
564 unix_release_sock(skb->sk, 1);
565 /* passed fds are erased in the kfree_skb hook */
566 UNIXCB(skb).consumed = skb->len;
575 /* ---- Socket is dead now and most probably destroyed ---- */
578 * Fixme: BSD difference: In BSD all sockets connected to us get
579 * ECONNRESET and we die on the spot. In Linux we behave
580 * like files and pipes do and wait for the last
583 * Can't we simply set sock->err?
585 * What the above comment does talk about? --ANK(980817)
588 if (unix_tot_inflight)
589 unix_gc(); /* Garbage collect fds */
592 static void init_peercred(struct sock *sk)
594 put_pid(sk->sk_peer_pid);
595 if (sk->sk_peer_cred)
596 put_cred(sk->sk_peer_cred);
597 sk->sk_peer_pid = get_pid(task_tgid(current));
598 sk->sk_peer_cred = get_current_cred();
601 static void copy_peercred(struct sock *sk, struct sock *peersk)
603 put_pid(sk->sk_peer_pid);
604 if (sk->sk_peer_cred)
605 put_cred(sk->sk_peer_cred);
606 sk->sk_peer_pid = get_pid(peersk->sk_peer_pid);
607 sk->sk_peer_cred = get_cred(peersk->sk_peer_cred);
610 static int unix_listen(struct socket *sock, int backlog)
613 struct sock *sk = sock->sk;
614 struct unix_sock *u = unix_sk(sk);
615 struct pid *old_pid = NULL;
618 if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
619 goto out; /* Only stream/seqpacket sockets accept */
622 goto out; /* No listens on an unbound socket */
624 if (sk->sk_state != TCP_CLOSE && sk->sk_state != TCP_LISTEN)
626 if (backlog > sk->sk_max_ack_backlog)
627 wake_up_interruptible_all(&u->peer_wait);
628 sk->sk_max_ack_backlog = backlog;
629 sk->sk_state = TCP_LISTEN;
630 /* set credentials so connect can copy them */
635 unix_state_unlock(sk);
641 static int unix_release(struct socket *);
642 static int unix_bind(struct socket *, struct sockaddr *, int);
643 static int unix_stream_connect(struct socket *, struct sockaddr *,
644 int addr_len, int flags);
645 static int unix_socketpair(struct socket *, struct socket *);
646 static int unix_accept(struct socket *, struct socket *, int, bool);
647 static int unix_getname(struct socket *, struct sockaddr *, int);
648 static __poll_t unix_poll(struct file *, struct socket *, poll_table *);
649 static __poll_t unix_dgram_poll(struct file *, struct socket *,
651 static int unix_ioctl(struct socket *, unsigned int, unsigned long);
652 static int unix_shutdown(struct socket *, int);
653 static int unix_stream_sendmsg(struct socket *, struct msghdr *, size_t);
654 static int unix_stream_recvmsg(struct socket *, struct msghdr *, size_t, int);
655 static ssize_t unix_stream_sendpage(struct socket *, struct page *, int offset,
656 size_t size, int flags);
657 static ssize_t unix_stream_splice_read(struct socket *, loff_t *ppos,
658 struct pipe_inode_info *, size_t size,
660 static int unix_dgram_sendmsg(struct socket *, struct msghdr *, size_t);
661 static int unix_dgram_recvmsg(struct socket *, struct msghdr *, size_t, int);
662 static int unix_dgram_connect(struct socket *, struct sockaddr *,
664 static int unix_seqpacket_sendmsg(struct socket *, struct msghdr *, size_t);
665 static int unix_seqpacket_recvmsg(struct socket *, struct msghdr *, size_t,
668 static int unix_set_peek_off(struct sock *sk, int val)
670 struct unix_sock *u = unix_sk(sk);
672 if (mutex_lock_interruptible(&u->iolock))
675 sk->sk_peek_off = val;
676 mutex_unlock(&u->iolock);
682 static const struct proto_ops unix_stream_ops = {
684 .owner = THIS_MODULE,
685 .release = unix_release,
687 .connect = unix_stream_connect,
688 .socketpair = unix_socketpair,
689 .accept = unix_accept,
690 .getname = unix_getname,
693 .listen = unix_listen,
694 .shutdown = unix_shutdown,
695 .setsockopt = sock_no_setsockopt,
696 .getsockopt = sock_no_getsockopt,
697 .sendmsg = unix_stream_sendmsg,
698 .recvmsg = unix_stream_recvmsg,
699 .mmap = sock_no_mmap,
700 .sendpage = unix_stream_sendpage,
701 .splice_read = unix_stream_splice_read,
702 .set_peek_off = unix_set_peek_off,
705 static const struct proto_ops unix_dgram_ops = {
707 .owner = THIS_MODULE,
708 .release = unix_release,
710 .connect = unix_dgram_connect,
711 .socketpair = unix_socketpair,
712 .accept = sock_no_accept,
713 .getname = unix_getname,
714 .poll = unix_dgram_poll,
716 .listen = sock_no_listen,
717 .shutdown = unix_shutdown,
718 .setsockopt = sock_no_setsockopt,
719 .getsockopt = sock_no_getsockopt,
720 .sendmsg = unix_dgram_sendmsg,
721 .recvmsg = unix_dgram_recvmsg,
722 .mmap = sock_no_mmap,
723 .sendpage = sock_no_sendpage,
724 .set_peek_off = unix_set_peek_off,
727 static const struct proto_ops unix_seqpacket_ops = {
729 .owner = THIS_MODULE,
730 .release = unix_release,
732 .connect = unix_stream_connect,
733 .socketpair = unix_socketpair,
734 .accept = unix_accept,
735 .getname = unix_getname,
736 .poll = unix_dgram_poll,
738 .listen = unix_listen,
739 .shutdown = unix_shutdown,
740 .setsockopt = sock_no_setsockopt,
741 .getsockopt = sock_no_getsockopt,
742 .sendmsg = unix_seqpacket_sendmsg,
743 .recvmsg = unix_seqpacket_recvmsg,
744 .mmap = sock_no_mmap,
745 .sendpage = sock_no_sendpage,
746 .set_peek_off = unix_set_peek_off,
749 static struct proto unix_proto = {
751 .owner = THIS_MODULE,
752 .obj_size = sizeof(struct unix_sock),
755 static struct sock *unix_create1(struct net *net, struct socket *sock, int kern)
757 struct sock *sk = NULL;
760 atomic_long_inc(&unix_nr_socks);
761 if (atomic_long_read(&unix_nr_socks) > 2 * get_max_files())
764 sk = sk_alloc(net, PF_UNIX, GFP_KERNEL, &unix_proto, kern);
768 sock_init_data(sock, sk);
770 sk->sk_allocation = GFP_KERNEL_ACCOUNT;
771 sk->sk_write_space = unix_write_space;
772 sk->sk_max_ack_backlog = net->unx.sysctl_max_dgram_qlen;
773 sk->sk_destruct = unix_sock_destructor;
775 u->path.dentry = NULL;
777 spin_lock_init(&u->lock);
778 atomic_long_set(&u->inflight, 0);
779 INIT_LIST_HEAD(&u->link);
780 mutex_init(&u->iolock); /* single task reading lock */
781 mutex_init(&u->bindlock); /* single task binding lock */
782 init_waitqueue_head(&u->peer_wait);
783 init_waitqueue_func_entry(&u->peer_wake, unix_dgram_peer_wake_relay);
784 unix_insert_socket(unix_sockets_unbound(sk), sk);
787 atomic_long_dec(&unix_nr_socks);
790 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
796 static int unix_create(struct net *net, struct socket *sock, int protocol,
799 if (protocol && protocol != PF_UNIX)
800 return -EPROTONOSUPPORT;
802 sock->state = SS_UNCONNECTED;
804 switch (sock->type) {
806 sock->ops = &unix_stream_ops;
809 * Believe it or not BSD has AF_UNIX, SOCK_RAW though
813 sock->type = SOCK_DGRAM;
816 sock->ops = &unix_dgram_ops;
819 sock->ops = &unix_seqpacket_ops;
822 return -ESOCKTNOSUPPORT;
825 return unix_create1(net, sock, kern) ? 0 : -ENOMEM;
828 static int unix_release(struct socket *sock)
830 struct sock *sk = sock->sk;
835 unix_release_sock(sk, 0);
841 static int unix_autobind(struct socket *sock)
843 struct sock *sk = sock->sk;
844 struct net *net = sock_net(sk);
845 struct unix_sock *u = unix_sk(sk);
846 static u32 ordernum = 1;
847 struct unix_address *addr;
849 unsigned int retries = 0;
851 err = mutex_lock_interruptible(&u->bindlock);
860 addr = kzalloc(sizeof(*addr) + sizeof(short) + 16, GFP_KERNEL);
864 addr->name->sun_family = AF_UNIX;
865 refcount_set(&addr->refcnt, 1);
868 addr->len = sprintf(addr->name->sun_path+1, "%05x", ordernum) + 1 + sizeof(short);
869 addr->hash = unix_hash_fold(csum_partial(addr->name, addr->len, 0));
871 spin_lock(&unix_table_lock);
872 ordernum = (ordernum+1)&0xFFFFF;
874 if (__unix_find_socket_byname(net, addr->name, addr->len, sock->type,
876 spin_unlock(&unix_table_lock);
878 * __unix_find_socket_byname() may take long time if many names
879 * are already in use.
882 /* Give up if all names seems to be in use. */
883 if (retries++ == 0xFFFFF) {
890 addr->hash ^= sk->sk_type;
892 __unix_remove_socket(sk);
894 __unix_insert_socket(&unix_socket_table[addr->hash], sk);
895 spin_unlock(&unix_table_lock);
898 out: mutex_unlock(&u->bindlock);
902 static struct sock *unix_find_other(struct net *net,
903 struct sockaddr_un *sunname, int len,
904 int type, unsigned int hash, int *error)
910 if (sunname->sun_path[0]) {
912 err = kern_path(sunname->sun_path, LOOKUP_FOLLOW, &path);
915 inode = d_backing_inode(path.dentry);
916 err = inode_permission(inode, MAY_WRITE);
921 if (!S_ISSOCK(inode->i_mode))
923 u = unix_find_socket_byinode(inode);
927 if (u->sk_type == type)
933 if (u->sk_type != type) {
939 u = unix_find_socket_byname(net, sunname, len, type, hash);
941 struct dentry *dentry;
942 dentry = unix_sk(u)->path.dentry;
944 touch_atime(&unix_sk(u)->path);
957 static int unix_mknod(const char *sun_path, umode_t mode, struct path *res)
959 struct dentry *dentry;
963 * Get the parent directory, calculate the hash for last
966 dentry = kern_path_create(AT_FDCWD, sun_path, &path, 0);
967 err = PTR_ERR(dentry);
972 * All right, let's create it.
974 err = security_path_mknod(&path, dentry, mode, 0);
976 err = vfs_mknod(d_inode(path.dentry), dentry, mode, 0);
978 res->mnt = mntget(path.mnt);
979 res->dentry = dget(dentry);
982 done_path_create(&path, dentry);
986 static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
988 struct sock *sk = sock->sk;
989 struct net *net = sock_net(sk);
990 struct unix_sock *u = unix_sk(sk);
991 struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
992 char *sun_path = sunaddr->sun_path;
995 struct unix_address *addr;
996 struct hlist_head *list;
997 struct path path = { };
1000 if (addr_len < offsetofend(struct sockaddr_un, sun_family) ||
1001 sunaddr->sun_family != AF_UNIX)
1004 if (addr_len == sizeof(short)) {
1005 err = unix_autobind(sock);
1009 err = unix_mkname(sunaddr, addr_len, &hash);
1015 umode_t mode = S_IFSOCK |
1016 (SOCK_INODE(sock)->i_mode & ~current_umask());
1017 err = unix_mknod(sun_path, mode, &path);
1025 err = mutex_lock_interruptible(&u->bindlock);
1034 addr = kmalloc(sizeof(*addr)+addr_len, GFP_KERNEL);
1038 memcpy(addr->name, sunaddr, addr_len);
1039 addr->len = addr_len;
1040 addr->hash = hash ^ sk->sk_type;
1041 refcount_set(&addr->refcnt, 1);
1044 addr->hash = UNIX_HASH_SIZE;
1045 hash = d_backing_inode(path.dentry)->i_ino & (UNIX_HASH_SIZE - 1);
1046 spin_lock(&unix_table_lock);
1048 list = &unix_socket_table[hash];
1050 spin_lock(&unix_table_lock);
1052 if (__unix_find_socket_byname(net, sunaddr, addr_len,
1053 sk->sk_type, hash)) {
1054 unix_release_addr(addr);
1058 list = &unix_socket_table[addr->hash];
1062 __unix_remove_socket(sk);
1064 __unix_insert_socket(list, sk);
1067 spin_unlock(&unix_table_lock);
1069 mutex_unlock(&u->bindlock);
1077 static void unix_state_double_lock(struct sock *sk1, struct sock *sk2)
1079 if (unlikely(sk1 == sk2) || !sk2) {
1080 unix_state_lock(sk1);
1084 unix_state_lock(sk1);
1085 unix_state_lock_nested(sk2);
1087 unix_state_lock(sk2);
1088 unix_state_lock_nested(sk1);
1092 static void unix_state_double_unlock(struct sock *sk1, struct sock *sk2)
1094 if (unlikely(sk1 == sk2) || !sk2) {
1095 unix_state_unlock(sk1);
1098 unix_state_unlock(sk1);
1099 unix_state_unlock(sk2);
1102 static int unix_dgram_connect(struct socket *sock, struct sockaddr *addr,
1103 int alen, int flags)
1105 struct sock *sk = sock->sk;
1106 struct net *net = sock_net(sk);
1107 struct sockaddr_un *sunaddr = (struct sockaddr_un *)addr;
1113 if (alen < offsetofend(struct sockaddr, sa_family))
1116 if (addr->sa_family != AF_UNSPEC) {
1117 err = unix_mkname(sunaddr, alen, &hash);
1122 if (test_bit(SOCK_PASSCRED, &sock->flags) &&
1123 !unix_sk(sk)->addr && (err = unix_autobind(sock)) != 0)
1127 other = unix_find_other(net, sunaddr, alen, sock->type, hash, &err);
1131 unix_state_double_lock(sk, other);
1133 /* Apparently VFS overslept socket death. Retry. */
1134 if (sock_flag(other, SOCK_DEAD)) {
1135 unix_state_double_unlock(sk, other);
1141 if (!unix_may_send(sk, other))
1144 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1150 * 1003.1g breaking connected state with AF_UNSPEC
1153 unix_state_double_lock(sk, other);
1157 * If it was connected, reconnect.
1159 if (unix_peer(sk)) {
1160 struct sock *old_peer = unix_peer(sk);
1161 unix_peer(sk) = other;
1162 unix_dgram_peer_wake_disconnect_wakeup(sk, old_peer);
1164 unix_state_double_unlock(sk, other);
1166 if (other != old_peer)
1167 unix_dgram_disconnected(sk, old_peer);
1170 unix_peer(sk) = other;
1171 unix_state_double_unlock(sk, other);
1176 unix_state_double_unlock(sk, other);
1182 static long unix_wait_for_peer(struct sock *other, long timeo)
1184 struct unix_sock *u = unix_sk(other);
1188 prepare_to_wait_exclusive(&u->peer_wait, &wait, TASK_INTERRUPTIBLE);
1190 sched = !sock_flag(other, SOCK_DEAD) &&
1191 !(other->sk_shutdown & RCV_SHUTDOWN) &&
1192 unix_recvq_full(other);
1194 unix_state_unlock(other);
1197 timeo = schedule_timeout(timeo);
1199 finish_wait(&u->peer_wait, &wait);
1203 static int unix_stream_connect(struct socket *sock, struct sockaddr *uaddr,
1204 int addr_len, int flags)
1206 struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
1207 struct sock *sk = sock->sk;
1208 struct net *net = sock_net(sk);
1209 struct unix_sock *u = unix_sk(sk), *newu, *otheru;
1210 struct sock *newsk = NULL;
1211 struct sock *other = NULL;
1212 struct sk_buff *skb = NULL;
1218 err = unix_mkname(sunaddr, addr_len, &hash);
1223 if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr &&
1224 (err = unix_autobind(sock)) != 0)
1227 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
1229 /* First of all allocate resources.
1230 If we will make it after state is locked,
1231 we will have to recheck all again in any case.
1236 /* create new sock for complete connection */
1237 newsk = unix_create1(sock_net(sk), NULL, 0);
1241 /* Allocate skb for sending to listening sock */
1242 skb = sock_wmalloc(newsk, 1, 0, GFP_KERNEL);
1247 /* Find listening sock. */
1248 other = unix_find_other(net, sunaddr, addr_len, sk->sk_type, hash, &err);
1252 /* Latch state of peer */
1253 unix_state_lock(other);
1255 /* Apparently VFS overslept socket death. Retry. */
1256 if (sock_flag(other, SOCK_DEAD)) {
1257 unix_state_unlock(other);
1262 err = -ECONNREFUSED;
1263 if (other->sk_state != TCP_LISTEN)
1265 if (other->sk_shutdown & RCV_SHUTDOWN)
1268 if (unix_recvq_full(other)) {
1273 timeo = unix_wait_for_peer(other, timeo);
1275 err = sock_intr_errno(timeo);
1276 if (signal_pending(current))
1284 It is tricky place. We need to grab our state lock and cannot
1285 drop lock on peer. It is dangerous because deadlock is
1286 possible. Connect to self case and simultaneous
1287 attempt to connect are eliminated by checking socket
1288 state. other is TCP_LISTEN, if sk is TCP_LISTEN we
1289 check this before attempt to grab lock.
1291 Well, and we have to recheck the state after socket locked.
1297 /* This is ok... continue with connect */
1299 case TCP_ESTABLISHED:
1300 /* Socket is already connected */
1308 unix_state_lock_nested(sk);
1310 if (sk->sk_state != st) {
1311 unix_state_unlock(sk);
1312 unix_state_unlock(other);
1317 err = security_unix_stream_connect(sk, other, newsk);
1319 unix_state_unlock(sk);
1323 /* The way is open! Fastly set all the necessary fields... */
1326 unix_peer(newsk) = sk;
1327 newsk->sk_state = TCP_ESTABLISHED;
1328 newsk->sk_type = sk->sk_type;
1329 init_peercred(newsk);
1330 newu = unix_sk(newsk);
1331 RCU_INIT_POINTER(newsk->sk_wq, &newu->peer_wq);
1332 otheru = unix_sk(other);
1334 /* copy address information from listening to new sock*/
1336 refcount_inc(&otheru->addr->refcnt);
1337 newu->addr = otheru->addr;
1339 if (otheru->path.dentry) {
1340 path_get(&otheru->path);
1341 newu->path = otheru->path;
1344 /* Set credentials */
1345 copy_peercred(sk, other);
1347 sock->state = SS_CONNECTED;
1348 sk->sk_state = TCP_ESTABLISHED;
1351 smp_mb__after_atomic(); /* sock_hold() does an atomic_inc() */
1352 unix_peer(sk) = newsk;
1354 unix_state_unlock(sk);
1356 /* take ten and and send info to listening sock */
1357 spin_lock(&other->sk_receive_queue.lock);
1358 __skb_queue_tail(&other->sk_receive_queue, skb);
1359 spin_unlock(&other->sk_receive_queue.lock);
1360 unix_state_unlock(other);
1361 other->sk_data_ready(other);
1367 unix_state_unlock(other);
1372 unix_release_sock(newsk, 0);
1378 static int unix_socketpair(struct socket *socka, struct socket *sockb)
1380 struct sock *ska = socka->sk, *skb = sockb->sk;
1382 /* Join our sockets back to back */
1385 unix_peer(ska) = skb;
1386 unix_peer(skb) = ska;
1390 if (ska->sk_type != SOCK_DGRAM) {
1391 ska->sk_state = TCP_ESTABLISHED;
1392 skb->sk_state = TCP_ESTABLISHED;
1393 socka->state = SS_CONNECTED;
1394 sockb->state = SS_CONNECTED;
1399 static void unix_sock_inherit_flags(const struct socket *old,
1402 if (test_bit(SOCK_PASSCRED, &old->flags))
1403 set_bit(SOCK_PASSCRED, &new->flags);
1404 if (test_bit(SOCK_PASSSEC, &old->flags))
1405 set_bit(SOCK_PASSSEC, &new->flags);
1408 static int unix_accept(struct socket *sock, struct socket *newsock, int flags,
1411 struct sock *sk = sock->sk;
1413 struct sk_buff *skb;
1417 if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
1421 if (sk->sk_state != TCP_LISTEN)
1424 /* If socket state is TCP_LISTEN it cannot change (for now...),
1425 * so that no locks are necessary.
1428 skb = skb_recv_datagram(sk, 0, flags&O_NONBLOCK, &err);
1430 /* This means receive shutdown. */
1437 skb_free_datagram(sk, skb);
1438 wake_up_interruptible(&unix_sk(sk)->peer_wait);
1440 /* attach accepted sock to socket */
1441 unix_state_lock(tsk);
1442 newsock->state = SS_CONNECTED;
1443 unix_sock_inherit_flags(sock, newsock);
1444 sock_graft(tsk, newsock);
1445 unix_state_unlock(tsk);
1453 static int unix_getname(struct socket *sock, struct sockaddr *uaddr, int peer)
1455 struct sock *sk = sock->sk;
1456 struct unix_sock *u;
1457 DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr, uaddr);
1461 sk = unix_peer_get(sk);
1472 unix_state_lock(sk);
1474 sunaddr->sun_family = AF_UNIX;
1475 sunaddr->sun_path[0] = 0;
1476 err = sizeof(short);
1478 struct unix_address *addr = u->addr;
1481 memcpy(sunaddr, addr->name, addr->len);
1483 unix_state_unlock(sk);
1489 static void unix_detach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1493 scm->fp = UNIXCB(skb).fp;
1494 UNIXCB(skb).fp = NULL;
1496 for (i = scm->fp->count-1; i >= 0; i--)
1497 unix_notinflight(scm->fp->user, scm->fp->fp[i]);
1500 static void unix_destruct_scm(struct sk_buff *skb)
1502 struct scm_cookie scm;
1503 memset(&scm, 0, sizeof(scm));
1504 scm.pid = UNIXCB(skb).pid;
1506 unix_detach_fds(&scm, skb);
1508 /* Alas, it calls VFS */
1509 /* So fscking what? fput() had been SMP-safe since the last Summer */
1515 * The "user->unix_inflight" variable is protected by the garbage
1516 * collection lock, and we just read it locklessly here. If you go
1517 * over the limit, there might be a tiny race in actually noticing
1518 * it across threads. Tough.
1520 static inline bool too_many_unix_fds(struct task_struct *p)
1522 struct user_struct *user = current_user();
1524 if (unlikely(user->unix_inflight > task_rlimit(p, RLIMIT_NOFILE)))
1525 return !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN);
1529 static int unix_attach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1533 if (too_many_unix_fds(current))
1534 return -ETOOMANYREFS;
1537 * Need to duplicate file references for the sake of garbage
1538 * collection. Otherwise a socket in the fps might become a
1539 * candidate for GC while the skb is not yet queued.
1541 UNIXCB(skb).fp = scm_fp_dup(scm->fp);
1542 if (!UNIXCB(skb).fp)
1545 for (i = scm->fp->count - 1; i >= 0; i--)
1546 unix_inflight(scm->fp->user, scm->fp->fp[i]);
1550 static int unix_scm_to_skb(struct scm_cookie *scm, struct sk_buff *skb, bool send_fds)
1554 UNIXCB(skb).pid = get_pid(scm->pid);
1555 UNIXCB(skb).uid = scm->creds.uid;
1556 UNIXCB(skb).gid = scm->creds.gid;
1557 UNIXCB(skb).fp = NULL;
1558 unix_get_secdata(scm, skb);
1559 if (scm->fp && send_fds)
1560 err = unix_attach_fds(scm, skb);
1562 skb->destructor = unix_destruct_scm;
1566 static bool unix_passcred_enabled(const struct socket *sock,
1567 const struct sock *other)
1569 return test_bit(SOCK_PASSCRED, &sock->flags) ||
1570 !other->sk_socket ||
1571 test_bit(SOCK_PASSCRED, &other->sk_socket->flags);
1575 * Some apps rely on write() giving SCM_CREDENTIALS
1576 * We include credentials if source or destination socket
1577 * asserted SOCK_PASSCRED.
1579 static void maybe_add_creds(struct sk_buff *skb, const struct socket *sock,
1580 const struct sock *other)
1582 if (UNIXCB(skb).pid)
1584 if (unix_passcred_enabled(sock, other)) {
1585 UNIXCB(skb).pid = get_pid(task_tgid(current));
1586 current_uid_gid(&UNIXCB(skb).uid, &UNIXCB(skb).gid);
1590 static int maybe_init_creds(struct scm_cookie *scm,
1591 struct socket *socket,
1592 const struct sock *other)
1595 struct msghdr msg = { .msg_controllen = 0 };
1597 err = scm_send(socket, &msg, scm, false);
1601 if (unix_passcred_enabled(socket, other)) {
1602 scm->pid = get_pid(task_tgid(current));
1603 current_uid_gid(&scm->creds.uid, &scm->creds.gid);
1608 static bool unix_skb_scm_eq(struct sk_buff *skb,
1609 struct scm_cookie *scm)
1611 const struct unix_skb_parms *u = &UNIXCB(skb);
1613 return u->pid == scm->pid &&
1614 uid_eq(u->uid, scm->creds.uid) &&
1615 gid_eq(u->gid, scm->creds.gid) &&
1616 unix_secdata_eq(scm, skb);
1620 * Send AF_UNIX data.
1623 static int unix_dgram_sendmsg(struct socket *sock, struct msghdr *msg,
1626 struct sock *sk = sock->sk;
1627 struct net *net = sock_net(sk);
1628 struct unix_sock *u = unix_sk(sk);
1629 DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr, msg->msg_name);
1630 struct sock *other = NULL;
1631 int namelen = 0; /* fake GCC */
1634 struct sk_buff *skb;
1636 struct scm_cookie scm;
1641 err = scm_send(sock, msg, &scm, false);
1646 if (msg->msg_flags&MSG_OOB)
1649 if (msg->msg_namelen) {
1650 err = unix_mkname(sunaddr, msg->msg_namelen, &hash);
1657 other = unix_peer_get(sk);
1662 if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr
1663 && (err = unix_autobind(sock)) != 0)
1667 if (len > sk->sk_sndbuf - 32)
1670 if (len > SKB_MAX_ALLOC) {
1671 data_len = min_t(size_t,
1672 len - SKB_MAX_ALLOC,
1673 MAX_SKB_FRAGS * PAGE_SIZE);
1674 data_len = PAGE_ALIGN(data_len);
1676 BUILD_BUG_ON(SKB_MAX_ALLOC < PAGE_SIZE);
1679 skb = sock_alloc_send_pskb(sk, len - data_len, data_len,
1680 msg->msg_flags & MSG_DONTWAIT, &err,
1681 PAGE_ALLOC_COSTLY_ORDER);
1685 err = unix_scm_to_skb(&scm, skb, true);
1689 skb_put(skb, len - data_len);
1690 skb->data_len = data_len;
1692 err = skb_copy_datagram_from_iter(skb, 0, &msg->msg_iter, len);
1696 timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1701 if (sunaddr == NULL)
1704 other = unix_find_other(net, sunaddr, namelen, sk->sk_type,
1710 if (sk_filter(other, skb) < 0) {
1711 /* Toss the packet but do not return any error to the sender */
1717 unix_state_lock(other);
1720 if (!unix_may_send(sk, other))
1723 if (unlikely(sock_flag(other, SOCK_DEAD))) {
1725 * Check with 1003.1g - what should
1728 unix_state_unlock(other);
1732 unix_state_lock(sk);
1735 if (unix_peer(sk) == other) {
1736 unix_peer(sk) = NULL;
1737 unix_dgram_peer_wake_disconnect_wakeup(sk, other);
1739 unix_state_unlock(sk);
1741 unix_dgram_disconnected(sk, other);
1743 err = -ECONNREFUSED;
1745 unix_state_unlock(sk);
1755 if (other->sk_shutdown & RCV_SHUTDOWN)
1758 if (sk->sk_type != SOCK_SEQPACKET) {
1759 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1764 /* other == sk && unix_peer(other) != sk if
1765 * - unix_peer(sk) == NULL, destination address bound to sk
1766 * - unix_peer(sk) == sk by time of get but disconnected before lock
1769 unlikely(unix_peer(other) != sk && unix_recvq_full(other))) {
1771 timeo = unix_wait_for_peer(other, timeo);
1773 err = sock_intr_errno(timeo);
1774 if (signal_pending(current))
1781 unix_state_unlock(other);
1782 unix_state_double_lock(sk, other);
1785 if (unix_peer(sk) != other ||
1786 unix_dgram_peer_wake_me(sk, other)) {
1794 goto restart_locked;
1798 if (unlikely(sk_locked))
1799 unix_state_unlock(sk);
1801 if (sock_flag(other, SOCK_RCVTSTAMP))
1802 __net_timestamp(skb);
1803 maybe_add_creds(skb, sock, other);
1804 skb_queue_tail(&other->sk_receive_queue, skb);
1805 unix_state_unlock(other);
1806 other->sk_data_ready(other);
1813 unix_state_unlock(sk);
1814 unix_state_unlock(other);
1824 /* We use paged skbs for stream sockets, and limit occupancy to 32768
1825 * bytes, and a minimum of a full page.
1827 #define UNIX_SKB_FRAGS_SZ (PAGE_SIZE << get_order(32768))
1829 static int unix_stream_sendmsg(struct socket *sock, struct msghdr *msg,
1832 struct sock *sk = sock->sk;
1833 struct sock *other = NULL;
1835 struct sk_buff *skb;
1837 struct scm_cookie scm;
1838 bool fds_sent = false;
1842 err = scm_send(sock, msg, &scm, false);
1847 if (msg->msg_flags&MSG_OOB)
1850 if (msg->msg_namelen) {
1851 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
1855 other = unix_peer(sk);
1860 if (sk->sk_shutdown & SEND_SHUTDOWN)
1863 while (sent < len) {
1866 /* Keep two messages in the pipe so it schedules better */
1867 size = min_t(int, size, (sk->sk_sndbuf >> 1) - 64);
1869 /* allow fallback to order-0 allocations */
1870 size = min_t(int, size, SKB_MAX_HEAD(0) + UNIX_SKB_FRAGS_SZ);
1872 data_len = max_t(int, 0, size - SKB_MAX_HEAD(0));
1874 data_len = min_t(size_t, size, PAGE_ALIGN(data_len));
1876 skb = sock_alloc_send_pskb(sk, size - data_len, data_len,
1877 msg->msg_flags & MSG_DONTWAIT, &err,
1878 get_order(UNIX_SKB_FRAGS_SZ));
1882 /* Only send the fds in the first buffer */
1883 err = unix_scm_to_skb(&scm, skb, !fds_sent);
1890 skb_put(skb, size - data_len);
1891 skb->data_len = data_len;
1893 err = skb_copy_datagram_from_iter(skb, 0, &msg->msg_iter, size);
1899 unix_state_lock(other);
1901 if (sock_flag(other, SOCK_DEAD) ||
1902 (other->sk_shutdown & RCV_SHUTDOWN))
1905 maybe_add_creds(skb, sock, other);
1906 skb_queue_tail(&other->sk_receive_queue, skb);
1907 unix_state_unlock(other);
1908 other->sk_data_ready(other);
1917 unix_state_unlock(other);
1920 if (sent == 0 && !(msg->msg_flags&MSG_NOSIGNAL))
1921 send_sig(SIGPIPE, current, 0);
1925 return sent ? : err;
1928 static ssize_t unix_stream_sendpage(struct socket *socket, struct page *page,
1929 int offset, size_t size, int flags)
1932 bool send_sigpipe = false;
1933 bool init_scm = true;
1934 struct scm_cookie scm;
1935 struct sock *other, *sk = socket->sk;
1936 struct sk_buff *skb, *newskb = NULL, *tail = NULL;
1938 if (flags & MSG_OOB)
1941 other = unix_peer(sk);
1942 if (!other || sk->sk_state != TCP_ESTABLISHED)
1947 unix_state_unlock(other);
1948 mutex_unlock(&unix_sk(other)->iolock);
1949 newskb = sock_alloc_send_pskb(sk, 0, 0, flags & MSG_DONTWAIT,
1955 /* we must acquire iolock as we modify already present
1956 * skbs in the sk_receive_queue and mess with skb->len
1958 err = mutex_lock_interruptible(&unix_sk(other)->iolock);
1960 err = flags & MSG_DONTWAIT ? -EAGAIN : -ERESTARTSYS;
1964 if (sk->sk_shutdown & SEND_SHUTDOWN) {
1966 send_sigpipe = true;
1970 unix_state_lock(other);
1972 if (sock_flag(other, SOCK_DEAD) ||
1973 other->sk_shutdown & RCV_SHUTDOWN) {
1975 send_sigpipe = true;
1976 goto err_state_unlock;
1980 err = maybe_init_creds(&scm, socket, other);
1982 goto err_state_unlock;
1986 skb = skb_peek_tail(&other->sk_receive_queue);
1987 if (tail && tail == skb) {
1989 } else if (!skb || !unix_skb_scm_eq(skb, &scm)) {
1996 } else if (newskb) {
1997 /* this is fast path, we don't necessarily need to
1998 * call to kfree_skb even though with newskb == NULL
1999 * this - does no harm
2001 consume_skb(newskb);
2005 if (skb_append_pagefrags(skb, page, offset, size)) {
2011 skb->data_len += size;
2012 skb->truesize += size;
2013 refcount_add(size, &sk->sk_wmem_alloc);
2016 err = unix_scm_to_skb(&scm, skb, false);
2018 goto err_state_unlock;
2019 spin_lock(&other->sk_receive_queue.lock);
2020 __skb_queue_tail(&other->sk_receive_queue, newskb);
2021 spin_unlock(&other->sk_receive_queue.lock);
2024 unix_state_unlock(other);
2025 mutex_unlock(&unix_sk(other)->iolock);
2027 other->sk_data_ready(other);
2032 unix_state_unlock(other);
2034 mutex_unlock(&unix_sk(other)->iolock);
2037 if (send_sigpipe && !(flags & MSG_NOSIGNAL))
2038 send_sig(SIGPIPE, current, 0);
2044 static int unix_seqpacket_sendmsg(struct socket *sock, struct msghdr *msg,
2048 struct sock *sk = sock->sk;
2050 err = sock_error(sk);
2054 if (sk->sk_state != TCP_ESTABLISHED)
2057 if (msg->msg_namelen)
2058 msg->msg_namelen = 0;
2060 return unix_dgram_sendmsg(sock, msg, len);
2063 static int unix_seqpacket_recvmsg(struct socket *sock, struct msghdr *msg,
2064 size_t size, int flags)
2066 struct sock *sk = sock->sk;
2068 if (sk->sk_state != TCP_ESTABLISHED)
2071 return unix_dgram_recvmsg(sock, msg, size, flags);
2074 static void unix_copy_addr(struct msghdr *msg, struct sock *sk)
2076 struct unix_sock *u = unix_sk(sk);
2079 msg->msg_namelen = u->addr->len;
2080 memcpy(msg->msg_name, u->addr->name, u->addr->len);
2084 static int unix_dgram_recvmsg(struct socket *sock, struct msghdr *msg,
2085 size_t size, int flags)
2087 struct scm_cookie scm;
2088 struct sock *sk = sock->sk;
2089 struct unix_sock *u = unix_sk(sk);
2090 struct sk_buff *skb, *last;
2099 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
2102 mutex_lock(&u->iolock);
2104 skip = sk_peek_offset(sk, flags);
2105 skb = __skb_try_recv_datagram(sk, flags, NULL, &peeked, &skip,
2110 mutex_unlock(&u->iolock);
2115 !__skb_wait_for_more_packets(sk, &err, &timeo, last));
2117 if (!skb) { /* implies iolock unlocked */
2118 unix_state_lock(sk);
2119 /* Signal EOF on disconnected non-blocking SEQPACKET socket. */
2120 if (sk->sk_type == SOCK_SEQPACKET && err == -EAGAIN &&
2121 (sk->sk_shutdown & RCV_SHUTDOWN))
2123 unix_state_unlock(sk);
2127 if (wq_has_sleeper(&u->peer_wait))
2128 wake_up_interruptible_sync_poll(&u->peer_wait,
2129 EPOLLOUT | EPOLLWRNORM |
2133 unix_copy_addr(msg, skb->sk);
2135 if (size > skb->len - skip)
2136 size = skb->len - skip;
2137 else if (size < skb->len - skip)
2138 msg->msg_flags |= MSG_TRUNC;
2140 err = skb_copy_datagram_msg(skb, skip, msg, size);
2144 if (sock_flag(sk, SOCK_RCVTSTAMP))
2145 __sock_recv_timestamp(msg, sk, skb);
2147 memset(&scm, 0, sizeof(scm));
2149 scm_set_cred(&scm, UNIXCB(skb).pid, UNIXCB(skb).uid, UNIXCB(skb).gid);
2150 unix_set_secdata(&scm, skb);
2152 if (!(flags & MSG_PEEK)) {
2154 unix_detach_fds(&scm, skb);
2156 sk_peek_offset_bwd(sk, skb->len);
2158 /* It is questionable: on PEEK we could:
2159 - do not return fds - good, but too simple 8)
2160 - return fds, and do not return them on read (old strategy,
2162 - clone fds (I chose it for now, it is the most universal
2165 POSIX 1003.1g does not actually define this clearly
2166 at all. POSIX 1003.1g doesn't define a lot of things
2171 sk_peek_offset_fwd(sk, size);
2174 scm.fp = scm_fp_dup(UNIXCB(skb).fp);
2176 err = (flags & MSG_TRUNC) ? skb->len - skip : size;
2178 scm_recv(sock, msg, &scm, flags);
2181 skb_free_datagram(sk, skb);
2182 mutex_unlock(&u->iolock);
2188 * Sleep until more data has arrived. But check for races..
2190 static long unix_stream_data_wait(struct sock *sk, long timeo,
2191 struct sk_buff *last, unsigned int last_len,
2194 struct sk_buff *tail;
2197 unix_state_lock(sk);
2200 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2202 tail = skb_peek_tail(&sk->sk_receive_queue);
2204 (tail && tail->len != last_len) ||
2206 (sk->sk_shutdown & RCV_SHUTDOWN) ||
2207 signal_pending(current) ||
2211 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2212 unix_state_unlock(sk);
2214 timeo = freezable_schedule_timeout(timeo);
2216 timeo = schedule_timeout(timeo);
2217 unix_state_lock(sk);
2219 if (sock_flag(sk, SOCK_DEAD))
2222 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2225 finish_wait(sk_sleep(sk), &wait);
2226 unix_state_unlock(sk);
2230 static unsigned int unix_skb_len(const struct sk_buff *skb)
2232 return skb->len - UNIXCB(skb).consumed;
2235 struct unix_stream_read_state {
2236 int (*recv_actor)(struct sk_buff *, int, int,
2237 struct unix_stream_read_state *);
2238 struct socket *socket;
2240 struct pipe_inode_info *pipe;
2243 unsigned int splice_flags;
2246 static int unix_stream_read_generic(struct unix_stream_read_state *state,
2249 struct scm_cookie scm;
2250 struct socket *sock = state->socket;
2251 struct sock *sk = sock->sk;
2252 struct unix_sock *u = unix_sk(sk);
2254 int flags = state->flags;
2255 int noblock = flags & MSG_DONTWAIT;
2256 bool check_creds = false;
2261 size_t size = state->size;
2262 unsigned int last_len;
2264 if (unlikely(sk->sk_state != TCP_ESTABLISHED)) {
2269 if (unlikely(flags & MSG_OOB)) {
2274 target = sock_rcvlowat(sk, flags & MSG_WAITALL, size);
2275 timeo = sock_rcvtimeo(sk, noblock);
2277 memset(&scm, 0, sizeof(scm));
2279 /* Lock the socket to prevent queue disordering
2280 * while sleeps in memcpy_tomsg
2282 mutex_lock(&u->iolock);
2284 skip = max(sk_peek_offset(sk, flags), 0);
2289 struct sk_buff *skb, *last;
2292 unix_state_lock(sk);
2293 if (sock_flag(sk, SOCK_DEAD)) {
2297 last = skb = skb_peek(&sk->sk_receive_queue);
2298 last_len = last ? last->len : 0;
2301 if (copied >= target)
2305 * POSIX 1003.1g mandates this order.
2308 err = sock_error(sk);
2311 if (sk->sk_shutdown & RCV_SHUTDOWN)
2314 unix_state_unlock(sk);
2320 mutex_unlock(&u->iolock);
2322 timeo = unix_stream_data_wait(sk, timeo, last,
2323 last_len, freezable);
2325 if (signal_pending(current)) {
2326 err = sock_intr_errno(timeo);
2331 mutex_lock(&u->iolock);
2334 unix_state_unlock(sk);
2338 while (skip >= unix_skb_len(skb)) {
2339 skip -= unix_skb_len(skb);
2341 last_len = skb->len;
2342 skb = skb_peek_next(skb, &sk->sk_receive_queue);
2347 unix_state_unlock(sk);
2350 /* Never glue messages from different writers */
2351 if (!unix_skb_scm_eq(skb, &scm))
2353 } else if (test_bit(SOCK_PASSCRED, &sock->flags)) {
2354 /* Copy credentials */
2355 scm_set_cred(&scm, UNIXCB(skb).pid, UNIXCB(skb).uid, UNIXCB(skb).gid);
2356 unix_set_secdata(&scm, skb);
2360 /* Copy address just once */
2361 if (state->msg && state->msg->msg_name) {
2362 DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr,
2363 state->msg->msg_name);
2364 unix_copy_addr(state->msg, skb->sk);
2368 chunk = min_t(unsigned int, unix_skb_len(skb) - skip, size);
2370 chunk = state->recv_actor(skb, skip, chunk, state);
2371 drop_skb = !unix_skb_len(skb);
2372 /* skb is only safe to use if !drop_skb */
2383 /* the skb was touched by a concurrent reader;
2384 * we should not expect anything from this skb
2385 * anymore and assume it invalid - we can be
2386 * sure it was dropped from the socket queue
2388 * let's report a short read
2394 /* Mark read part of skb as used */
2395 if (!(flags & MSG_PEEK)) {
2396 UNIXCB(skb).consumed += chunk;
2398 sk_peek_offset_bwd(sk, chunk);
2401 unix_detach_fds(&scm, skb);
2403 if (unix_skb_len(skb))
2406 skb_unlink(skb, &sk->sk_receive_queue);
2412 /* It is questionable, see note in unix_dgram_recvmsg.
2415 scm.fp = scm_fp_dup(UNIXCB(skb).fp);
2417 sk_peek_offset_fwd(sk, chunk);
2424 last_len = skb->len;
2425 unix_state_lock(sk);
2426 skb = skb_peek_next(skb, &sk->sk_receive_queue);
2429 unix_state_unlock(sk);
2434 mutex_unlock(&u->iolock);
2436 scm_recv(sock, state->msg, &scm, flags);
2440 return copied ? : err;
2443 static int unix_stream_read_actor(struct sk_buff *skb,
2444 int skip, int chunk,
2445 struct unix_stream_read_state *state)
2449 ret = skb_copy_datagram_msg(skb, UNIXCB(skb).consumed + skip,
2451 return ret ?: chunk;
2454 static int unix_stream_recvmsg(struct socket *sock, struct msghdr *msg,
2455 size_t size, int flags)
2457 struct unix_stream_read_state state = {
2458 .recv_actor = unix_stream_read_actor,
2465 return unix_stream_read_generic(&state, true);
2468 static int unix_stream_splice_actor(struct sk_buff *skb,
2469 int skip, int chunk,
2470 struct unix_stream_read_state *state)
2472 return skb_splice_bits(skb, state->socket->sk,
2473 UNIXCB(skb).consumed + skip,
2474 state->pipe, chunk, state->splice_flags);
2477 static ssize_t unix_stream_splice_read(struct socket *sock, loff_t *ppos,
2478 struct pipe_inode_info *pipe,
2479 size_t size, unsigned int flags)
2481 struct unix_stream_read_state state = {
2482 .recv_actor = unix_stream_splice_actor,
2486 .splice_flags = flags,
2489 if (unlikely(*ppos))
2492 if (sock->file->f_flags & O_NONBLOCK ||
2493 flags & SPLICE_F_NONBLOCK)
2494 state.flags = MSG_DONTWAIT;
2496 return unix_stream_read_generic(&state, false);
2499 static int unix_shutdown(struct socket *sock, int mode)
2501 struct sock *sk = sock->sk;
2504 if (mode < SHUT_RD || mode > SHUT_RDWR)
2507 * SHUT_RD (0) -> RCV_SHUTDOWN (1)
2508 * SHUT_WR (1) -> SEND_SHUTDOWN (2)
2509 * SHUT_RDWR (2) -> SHUTDOWN_MASK (3)
2513 unix_state_lock(sk);
2514 sk->sk_shutdown |= mode;
2515 other = unix_peer(sk);
2518 unix_state_unlock(sk);
2519 sk->sk_state_change(sk);
2522 (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET)) {
2526 if (mode&RCV_SHUTDOWN)
2527 peer_mode |= SEND_SHUTDOWN;
2528 if (mode&SEND_SHUTDOWN)
2529 peer_mode |= RCV_SHUTDOWN;
2530 unix_state_lock(other);
2531 other->sk_shutdown |= peer_mode;
2532 unix_state_unlock(other);
2533 other->sk_state_change(other);
2534 if (peer_mode == SHUTDOWN_MASK)
2535 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_HUP);
2536 else if (peer_mode & RCV_SHUTDOWN)
2537 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_IN);
2545 long unix_inq_len(struct sock *sk)
2547 struct sk_buff *skb;
2550 if (sk->sk_state == TCP_LISTEN)
2553 spin_lock(&sk->sk_receive_queue.lock);
2554 if (sk->sk_type == SOCK_STREAM ||
2555 sk->sk_type == SOCK_SEQPACKET) {
2556 skb_queue_walk(&sk->sk_receive_queue, skb)
2557 amount += unix_skb_len(skb);
2559 skb = skb_peek(&sk->sk_receive_queue);
2563 spin_unlock(&sk->sk_receive_queue.lock);
2567 EXPORT_SYMBOL_GPL(unix_inq_len);
2569 long unix_outq_len(struct sock *sk)
2571 return sk_wmem_alloc_get(sk);
2573 EXPORT_SYMBOL_GPL(unix_outq_len);
2575 static int unix_open_file(struct sock *sk)
2581 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
2584 unix_state_lock(sk);
2585 path = unix_sk(sk)->path;
2587 unix_state_unlock(sk);
2592 unix_state_unlock(sk);
2594 fd = get_unused_fd_flags(O_CLOEXEC);
2598 f = dentry_open(&path, O_PATH, current_cred());
2612 static int unix_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2614 struct sock *sk = sock->sk;
2620 amount = unix_outq_len(sk);
2621 err = put_user(amount, (int __user *)arg);
2624 amount = unix_inq_len(sk);
2628 err = put_user(amount, (int __user *)arg);
2631 err = unix_open_file(sk);
2640 static __poll_t unix_poll(struct file *file, struct socket *sock, poll_table *wait)
2642 struct sock *sk = sock->sk;
2645 sock_poll_wait(file, sock, wait);
2648 /* exceptional events? */
2651 if (sk->sk_shutdown == SHUTDOWN_MASK)
2653 if (sk->sk_shutdown & RCV_SHUTDOWN)
2654 mask |= EPOLLRDHUP | EPOLLIN | EPOLLRDNORM;
2657 if (!skb_queue_empty(&sk->sk_receive_queue))
2658 mask |= EPOLLIN | EPOLLRDNORM;
2660 /* Connection-based need to check for termination and startup */
2661 if ((sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) &&
2662 sk->sk_state == TCP_CLOSE)
2666 * we set writable also when the other side has shut down the
2667 * connection. This prevents stuck sockets.
2669 if (unix_writable(sk))
2670 mask |= EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND;
2675 static __poll_t unix_dgram_poll(struct file *file, struct socket *sock,
2678 struct sock *sk = sock->sk, *other;
2679 unsigned int writable;
2682 sock_poll_wait(file, sock, wait);
2685 /* exceptional events? */
2686 if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue))
2688 (sock_flag(sk, SOCK_SELECT_ERR_QUEUE) ? EPOLLPRI : 0);
2690 if (sk->sk_shutdown & RCV_SHUTDOWN)
2691 mask |= EPOLLRDHUP | EPOLLIN | EPOLLRDNORM;
2692 if (sk->sk_shutdown == SHUTDOWN_MASK)
2696 if (!skb_queue_empty(&sk->sk_receive_queue))
2697 mask |= EPOLLIN | EPOLLRDNORM;
2699 /* Connection-based need to check for termination and startup */
2700 if (sk->sk_type == SOCK_SEQPACKET) {
2701 if (sk->sk_state == TCP_CLOSE)
2703 /* connection hasn't started yet? */
2704 if (sk->sk_state == TCP_SYN_SENT)
2708 /* No write status requested, avoid expensive OUT tests. */
2709 if (!(poll_requested_events(wait) & (EPOLLWRBAND|EPOLLWRNORM|EPOLLOUT)))
2712 writable = unix_writable(sk);
2714 unix_state_lock(sk);
2716 other = unix_peer(sk);
2717 if (other && unix_peer(other) != sk &&
2718 unix_recvq_full(other) &&
2719 unix_dgram_peer_wake_me(sk, other))
2722 unix_state_unlock(sk);
2726 mask |= EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND;
2728 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2733 #ifdef CONFIG_PROC_FS
2735 #define BUCKET_SPACE (BITS_PER_LONG - (UNIX_HASH_BITS + 1) - 1)
2737 #define get_bucket(x) ((x) >> BUCKET_SPACE)
2738 #define get_offset(x) ((x) & ((1L << BUCKET_SPACE) - 1))
2739 #define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
2741 static struct sock *unix_from_bucket(struct seq_file *seq, loff_t *pos)
2743 unsigned long offset = get_offset(*pos);
2744 unsigned long bucket = get_bucket(*pos);
2746 unsigned long count = 0;
2748 for (sk = sk_head(&unix_socket_table[bucket]); sk; sk = sk_next(sk)) {
2749 if (sock_net(sk) != seq_file_net(seq))
2751 if (++count == offset)
2758 static struct sock *unix_next_socket(struct seq_file *seq,
2762 unsigned long bucket;
2764 while (sk > (struct sock *)SEQ_START_TOKEN) {
2768 if (sock_net(sk) == seq_file_net(seq))
2773 sk = unix_from_bucket(seq, pos);
2778 bucket = get_bucket(*pos) + 1;
2779 *pos = set_bucket_offset(bucket, 1);
2780 } while (bucket < ARRAY_SIZE(unix_socket_table));
2785 static void *unix_seq_start(struct seq_file *seq, loff_t *pos)
2786 __acquires(unix_table_lock)
2788 spin_lock(&unix_table_lock);
2791 return SEQ_START_TOKEN;
2793 if (get_bucket(*pos) >= ARRAY_SIZE(unix_socket_table))
2796 return unix_next_socket(seq, NULL, pos);
2799 static void *unix_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2802 return unix_next_socket(seq, v, pos);
2805 static void unix_seq_stop(struct seq_file *seq, void *v)
2806 __releases(unix_table_lock)
2808 spin_unlock(&unix_table_lock);
2811 static int unix_seq_show(struct seq_file *seq, void *v)
2814 if (v == SEQ_START_TOKEN)
2815 seq_puts(seq, "Num RefCount Protocol Flags Type St "
2819 struct unix_sock *u = unix_sk(s);
2822 seq_printf(seq, "%pK: %08X %08X %08X %04X %02X %5lu",
2824 refcount_read(&s->sk_refcnt),
2826 s->sk_state == TCP_LISTEN ? __SO_ACCEPTCON : 0,
2829 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTED : SS_UNCONNECTED) :
2830 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTING : SS_DISCONNECTING),
2838 len = u->addr->len - sizeof(short);
2839 if (!UNIX_ABSTRACT(s))
2845 for ( ; i < len; i++)
2846 seq_putc(seq, u->addr->name->sun_path[i] ?:
2849 unix_state_unlock(s);
2850 seq_putc(seq, '\n');
2856 static const struct seq_operations unix_seq_ops = {
2857 .start = unix_seq_start,
2858 .next = unix_seq_next,
2859 .stop = unix_seq_stop,
2860 .show = unix_seq_show,
2864 static const struct net_proto_family unix_family_ops = {
2866 .create = unix_create,
2867 .owner = THIS_MODULE,
2871 static int __net_init unix_net_init(struct net *net)
2873 int error = -ENOMEM;
2875 net->unx.sysctl_max_dgram_qlen = 10;
2876 if (unix_sysctl_register(net))
2879 #ifdef CONFIG_PROC_FS
2880 if (!proc_create_net("unix", 0, net->proc_net, &unix_seq_ops,
2881 sizeof(struct seq_net_private))) {
2882 unix_sysctl_unregister(net);
2891 static void __net_exit unix_net_exit(struct net *net)
2893 unix_sysctl_unregister(net);
2894 remove_proc_entry("unix", net->proc_net);
2897 static struct pernet_operations unix_net_ops = {
2898 .init = unix_net_init,
2899 .exit = unix_net_exit,
2902 static int __init af_unix_init(void)
2906 BUILD_BUG_ON(sizeof(struct unix_skb_parms) > FIELD_SIZEOF(struct sk_buff, cb));
2908 rc = proto_register(&unix_proto, 1);
2910 pr_crit("%s: Cannot create unix_sock SLAB cache!\n", __func__);
2914 sock_register(&unix_family_ops);
2915 register_pernet_subsys(&unix_net_ops);
2920 static void __exit af_unix_exit(void)
2922 sock_unregister(PF_UNIX);
2923 proto_unregister(&unix_proto);
2924 unregister_pernet_subsys(&unix_net_ops);
2927 /* Earlier than device_initcall() so that other drivers invoking
2928 request_module() don't end up in a loop when modprobe tries
2929 to use a UNIX socket. But later than subsys_initcall() because
2930 we depend on stuff initialised there */
2931 fs_initcall(af_unix_init);
2932 module_exit(af_unix_exit);
2934 MODULE_LICENSE("GPL");
2935 MODULE_ALIAS_NETPROTO(PF_UNIX);