Loading...
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * NET4: Implementation of BSD Unix domain sockets.
4 *
5 * Authors: Alan Cox, <alan@lxorguk.ukuu.org.uk>
6 *
7 * Fixes:
8 * Linus Torvalds : Assorted bug cures.
9 * Niibe Yutaka : async I/O support.
10 * Carsten Paeth : PF_UNIX check, address fixes.
11 * Alan Cox : Limit size of allocated blocks.
12 * Alan Cox : Fixed the stupid socketpair bug.
13 * Alan Cox : BSD compatibility fine tuning.
14 * Alan Cox : Fixed a bug in connect when interrupted.
15 * Alan Cox : Sorted out a proper draft version of
16 * file descriptor passing hacked up from
17 * Mike Shaver's work.
18 * Marty Leisner : Fixes to fd passing
19 * Nick Nevin : recvmsg bugfix.
20 * Alan Cox : Started proper garbage collector
21 * Heiko EiBfeldt : Missing verify_area check
22 * Alan Cox : Started POSIXisms
23 * Andreas Schwab : Replace inode by dentry for proper
24 * reference counting
25 * Kirk Petersen : Made this a module
26 * Christoph Rohland : Elegant non-blocking accept/connect algorithm.
27 * Lots of bug fixes.
28 * Alexey Kuznetosv : Repaired (I hope) bugs introduces
29 * by above two patches.
30 * Andrea Arcangeli : If possible we block in connect(2)
31 * if the max backlog of the listen socket
32 * is been reached. This won't break
33 * old apps and it will avoid huge amount
34 * of socks hashed (this for unix_gc()
35 * performances reasons).
36 * Security fix that limits the max
37 * number of socks to 2*max_files and
38 * the number of skb queueable in the
39 * dgram receiver.
40 * Artur Skawina : Hash function optimizations
41 * Alexey Kuznetsov : Full scale SMP. Lot of bugs are introduced 8)
42 * Malcolm Beattie : Set peercred for socketpair
43 * Michal Ostrowski : Module initialization cleanup.
44 * Arnaldo C. Melo : Remove MOD_{INC,DEC}_USE_COUNT,
45 * the core infrastructure is doing that
46 * for all net proto families now (2.5.69+)
47 *
48 * Known differences from reference BSD that was tested:
49 *
50 * [TO FIX]
51 * ECONNREFUSED is not returned from one end of a connected() socket to the
52 * other the moment one end closes.
53 * fstat() doesn't return st_dev=0, and give the blksize as high water mark
54 * and a fake inode identifier (nor the BSD first socket fstat twice bug).
55 * [NOT TO FIX]
56 * accept() returns a path name even if the connecting socket has closed
57 * in the meantime (BSD loses the path and gives up).
58 * accept() returns 0 length path for an unbound connector. BSD returns 16
59 * and a null first byte in the path (but not for gethost/peername - BSD bug ??)
60 * socketpair(...SOCK_RAW..) doesn't panic the kernel.
61 * BSD af_unix apparently has connect forgetting to block properly.
62 * (need to check this with the POSIX spec in detail)
63 *
64 * Differences from 2.0.0-11-... (ANK)
65 * Bug fixes and improvements.
66 * - client shutdown killed server socket.
67 * - removed all useless cli/sti pairs.
68 *
69 * Semantic changes/extensions.
70 * - generic control message passing.
71 * - SCM_CREDENTIALS control message.
72 * - "Abstract" (not FS based) socket bindings.
73 * Abstract names are sequences of bytes (not zero terminated)
74 * started by 0, so that this name space does not intersect
75 * with BSD names.
76 */
77
78#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
79
80#include <linux/module.h>
81#include <linux/kernel.h>
82#include <linux/signal.h>
83#include <linux/sched/signal.h>
84#include <linux/errno.h>
85#include <linux/string.h>
86#include <linux/stat.h>
87#include <linux/dcache.h>
88#include <linux/namei.h>
89#include <linux/socket.h>
90#include <linux/un.h>
91#include <linux/fcntl.h>
92#include <linux/termios.h>
93#include <linux/sockios.h>
94#include <linux/net.h>
95#include <linux/in.h>
96#include <linux/fs.h>
97#include <linux/slab.h>
98#include <linux/uaccess.h>
99#include <linux/skbuff.h>
100#include <linux/netdevice.h>
101#include <net/net_namespace.h>
102#include <net/sock.h>
103#include <net/tcp_states.h>
104#include <net/af_unix.h>
105#include <linux/proc_fs.h>
106#include <linux/seq_file.h>
107#include <net/scm.h>
108#include <linux/init.h>
109#include <linux/poll.h>
110#include <linux/rtnetlink.h>
111#include <linux/mount.h>
112#include <net/checksum.h>
113#include <linux/security.h>
114#include <linux/freezer.h>
115#include <linux/file.h>
116
117#include "scm.h"
118
119struct hlist_head unix_socket_table[2 * UNIX_HASH_SIZE];
120EXPORT_SYMBOL_GPL(unix_socket_table);
121DEFINE_SPINLOCK(unix_table_lock);
122EXPORT_SYMBOL_GPL(unix_table_lock);
123static atomic_long_t unix_nr_socks;
124
125
126static struct hlist_head *unix_sockets_unbound(void *addr)
127{
128 unsigned long hash = (unsigned long)addr;
129
130 hash ^= hash >> 16;
131 hash ^= hash >> 8;
132 hash %= UNIX_HASH_SIZE;
133 return &unix_socket_table[UNIX_HASH_SIZE + hash];
134}
135
136#define UNIX_ABSTRACT(sk) (unix_sk(sk)->addr->hash < UNIX_HASH_SIZE)
137
138#ifdef CONFIG_SECURITY_NETWORK
139static void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb)
140{
141 UNIXCB(skb).secid = scm->secid;
142}
143
144static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb)
145{
146 scm->secid = UNIXCB(skb).secid;
147}
148
149static inline bool unix_secdata_eq(struct scm_cookie *scm, struct sk_buff *skb)
150{
151 return (scm->secid == UNIXCB(skb).secid);
152}
153#else
154static inline void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb)
155{ }
156
157static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb)
158{ }
159
160static inline bool unix_secdata_eq(struct scm_cookie *scm, struct sk_buff *skb)
161{
162 return true;
163}
164#endif /* CONFIG_SECURITY_NETWORK */
165
166/*
167 * SMP locking strategy:
168 * hash table is protected with spinlock unix_table_lock
169 * each socket state is protected by separate spin lock.
170 */
171
172static inline unsigned int unix_hash_fold(__wsum n)
173{
174 unsigned int hash = (__force unsigned int)csum_fold(n);
175
176 hash ^= hash>>8;
177 return hash&(UNIX_HASH_SIZE-1);
178}
179
180#define unix_peer(sk) (unix_sk(sk)->peer)
181
182static inline int unix_our_peer(struct sock *sk, struct sock *osk)
183{
184 return unix_peer(osk) == sk;
185}
186
187static inline int unix_may_send(struct sock *sk, struct sock *osk)
188{
189 return unix_peer(osk) == NULL || unix_our_peer(sk, osk);
190}
191
192static inline int unix_recvq_full(const struct sock *sk)
193{
194 return skb_queue_len(&sk->sk_receive_queue) > sk->sk_max_ack_backlog;
195}
196
197static inline int unix_recvq_full_lockless(const struct sock *sk)
198{
199 return skb_queue_len_lockless(&sk->sk_receive_queue) >
200 READ_ONCE(sk->sk_max_ack_backlog);
201}
202
203struct sock *unix_peer_get(struct sock *s)
204{
205 struct sock *peer;
206
207 unix_state_lock(s);
208 peer = unix_peer(s);
209 if (peer)
210 sock_hold(peer);
211 unix_state_unlock(s);
212 return peer;
213}
214EXPORT_SYMBOL_GPL(unix_peer_get);
215
216static inline void unix_release_addr(struct unix_address *addr)
217{
218 if (refcount_dec_and_test(&addr->refcnt))
219 kfree(addr);
220}
221
222/*
223 * Check unix socket name:
224 * - should be not zero length.
225 * - if started by not zero, should be NULL terminated (FS object)
226 * - if started by zero, it is abstract name.
227 */
228
229static int unix_mkname(struct sockaddr_un *sunaddr, int len, unsigned int *hashp)
230{
231 *hashp = 0;
232
233 if (len <= sizeof(short) || len > sizeof(*sunaddr))
234 return -EINVAL;
235 if (!sunaddr || sunaddr->sun_family != AF_UNIX)
236 return -EINVAL;
237 if (sunaddr->sun_path[0]) {
238 /*
239 * This may look like an off by one error but it is a bit more
240 * subtle. 108 is the longest valid AF_UNIX path for a binding.
241 * sun_path[108] doesn't as such exist. However in kernel space
242 * we are guaranteed that it is a valid memory location in our
243 * kernel address buffer.
244 */
245 ((char *)sunaddr)[len] = 0;
246 len = strlen(sunaddr->sun_path)+1+sizeof(short);
247 return len;
248 }
249
250 *hashp = unix_hash_fold(csum_partial(sunaddr, len, 0));
251 return len;
252}
253
254static void __unix_remove_socket(struct sock *sk)
255{
256 sk_del_node_init(sk);
257}
258
259static void __unix_insert_socket(struct hlist_head *list, struct sock *sk)
260{
261 WARN_ON(!sk_unhashed(sk));
262 sk_add_node(sk, list);
263}
264
265static void __unix_set_addr(struct sock *sk, struct unix_address *addr,
266 unsigned hash)
267{
268 __unix_remove_socket(sk);
269 smp_store_release(&unix_sk(sk)->addr, addr);
270 __unix_insert_socket(&unix_socket_table[hash], sk);
271}
272
273static inline void unix_remove_socket(struct sock *sk)
274{
275 spin_lock(&unix_table_lock);
276 __unix_remove_socket(sk);
277 spin_unlock(&unix_table_lock);
278}
279
280static inline void unix_insert_socket(struct hlist_head *list, struct sock *sk)
281{
282 spin_lock(&unix_table_lock);
283 __unix_insert_socket(list, sk);
284 spin_unlock(&unix_table_lock);
285}
286
287static struct sock *__unix_find_socket_byname(struct net *net,
288 struct sockaddr_un *sunname,
289 int len, unsigned int hash)
290{
291 struct sock *s;
292
293 sk_for_each(s, &unix_socket_table[hash]) {
294 struct unix_sock *u = unix_sk(s);
295
296 if (!net_eq(sock_net(s), net))
297 continue;
298
299 if (u->addr->len == len &&
300 !memcmp(u->addr->name, sunname, len))
301 return s;
302 }
303 return NULL;
304}
305
306static inline struct sock *unix_find_socket_byname(struct net *net,
307 struct sockaddr_un *sunname,
308 int len, unsigned int hash)
309{
310 struct sock *s;
311
312 spin_lock(&unix_table_lock);
313 s = __unix_find_socket_byname(net, sunname, len, hash);
314 if (s)
315 sock_hold(s);
316 spin_unlock(&unix_table_lock);
317 return s;
318}
319
320static struct sock *unix_find_socket_byinode(struct inode *i)
321{
322 struct sock *s;
323
324 spin_lock(&unix_table_lock);
325 sk_for_each(s,
326 &unix_socket_table[i->i_ino & (UNIX_HASH_SIZE - 1)]) {
327 struct dentry *dentry = unix_sk(s)->path.dentry;
328
329 if (dentry && d_backing_inode(dentry) == i) {
330 sock_hold(s);
331 goto found;
332 }
333 }
334 s = NULL;
335found:
336 spin_unlock(&unix_table_lock);
337 return s;
338}
339
340/* Support code for asymmetrically connected dgram sockets
341 *
342 * If a datagram socket is connected to a socket not itself connected
343 * to the first socket (eg, /dev/log), clients may only enqueue more
344 * messages if the present receive queue of the server socket is not
345 * "too large". This means there's a second writeability condition
346 * poll and sendmsg need to test. The dgram recv code will do a wake
347 * up on the peer_wait wait queue of a socket upon reception of a
348 * datagram which needs to be propagated to sleeping would-be writers
349 * since these might not have sent anything so far. This can't be
350 * accomplished via poll_wait because the lifetime of the server
351 * socket might be less than that of its clients if these break their
352 * association with it or if the server socket is closed while clients
353 * are still connected to it and there's no way to inform "a polling
354 * implementation" that it should let go of a certain wait queue
355 *
356 * In order to propagate a wake up, a wait_queue_entry_t of the client
357 * socket is enqueued on the peer_wait queue of the server socket
358 * whose wake function does a wake_up on the ordinary client socket
359 * wait queue. This connection is established whenever a write (or
360 * poll for write) hit the flow control condition and broken when the
361 * association to the server socket is dissolved or after a wake up
362 * was relayed.
363 */
364
365static int unix_dgram_peer_wake_relay(wait_queue_entry_t *q, unsigned mode, int flags,
366 void *key)
367{
368 struct unix_sock *u;
369 wait_queue_head_t *u_sleep;
370
371 u = container_of(q, struct unix_sock, peer_wake);
372
373 __remove_wait_queue(&unix_sk(u->peer_wake.private)->peer_wait,
374 q);
375 u->peer_wake.private = NULL;
376
377 /* relaying can only happen while the wq still exists */
378 u_sleep = sk_sleep(&u->sk);
379 if (u_sleep)
380 wake_up_interruptible_poll(u_sleep, key_to_poll(key));
381
382 return 0;
383}
384
385static int unix_dgram_peer_wake_connect(struct sock *sk, struct sock *other)
386{
387 struct unix_sock *u, *u_other;
388 int rc;
389
390 u = unix_sk(sk);
391 u_other = unix_sk(other);
392 rc = 0;
393 spin_lock(&u_other->peer_wait.lock);
394
395 if (!u->peer_wake.private) {
396 u->peer_wake.private = other;
397 __add_wait_queue(&u_other->peer_wait, &u->peer_wake);
398
399 rc = 1;
400 }
401
402 spin_unlock(&u_other->peer_wait.lock);
403 return rc;
404}
405
406static void unix_dgram_peer_wake_disconnect(struct sock *sk,
407 struct sock *other)
408{
409 struct unix_sock *u, *u_other;
410
411 u = unix_sk(sk);
412 u_other = unix_sk(other);
413 spin_lock(&u_other->peer_wait.lock);
414
415 if (u->peer_wake.private == other) {
416 __remove_wait_queue(&u_other->peer_wait, &u->peer_wake);
417 u->peer_wake.private = NULL;
418 }
419
420 spin_unlock(&u_other->peer_wait.lock);
421}
422
423static void unix_dgram_peer_wake_disconnect_wakeup(struct sock *sk,
424 struct sock *other)
425{
426 unix_dgram_peer_wake_disconnect(sk, other);
427 wake_up_interruptible_poll(sk_sleep(sk),
428 EPOLLOUT |
429 EPOLLWRNORM |
430 EPOLLWRBAND);
431}
432
433/* preconditions:
434 * - unix_peer(sk) == other
435 * - association is stable
436 */
437static int unix_dgram_peer_wake_me(struct sock *sk, struct sock *other)
438{
439 int connected;
440
441 connected = unix_dgram_peer_wake_connect(sk, other);
442
443 /* If other is SOCK_DEAD, we want to make sure we signal
444 * POLLOUT, such that a subsequent write() can get a
445 * -ECONNREFUSED. Otherwise, if we haven't queued any skbs
446 * to other and its full, we will hang waiting for POLLOUT.
447 */
448 if (unix_recvq_full(other) && !sock_flag(other, SOCK_DEAD))
449 return 1;
450
451 if (connected)
452 unix_dgram_peer_wake_disconnect(sk, other);
453
454 return 0;
455}
456
457static int unix_writable(const struct sock *sk)
458{
459 return sk->sk_state != TCP_LISTEN &&
460 (refcount_read(&sk->sk_wmem_alloc) << 2) <= sk->sk_sndbuf;
461}
462
463static void unix_write_space(struct sock *sk)
464{
465 struct socket_wq *wq;
466
467 rcu_read_lock();
468 if (unix_writable(sk)) {
469 wq = rcu_dereference(sk->sk_wq);
470 if (skwq_has_sleeper(wq))
471 wake_up_interruptible_sync_poll(&wq->wait,
472 EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND);
473 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
474 }
475 rcu_read_unlock();
476}
477
478/* When dgram socket disconnects (or changes its peer), we clear its receive
479 * queue of packets arrived from previous peer. First, it allows to do
480 * flow control based only on wmem_alloc; second, sk connected to peer
481 * may receive messages only from that peer. */
482static void unix_dgram_disconnected(struct sock *sk, struct sock *other)
483{
484 if (!skb_queue_empty(&sk->sk_receive_queue)) {
485 skb_queue_purge(&sk->sk_receive_queue);
486 wake_up_interruptible_all(&unix_sk(sk)->peer_wait);
487
488 /* If one link of bidirectional dgram pipe is disconnected,
489 * we signal error. Messages are lost. Do not make this,
490 * when peer was not connected to us.
491 */
492 if (!sock_flag(other, SOCK_DEAD) && unix_peer(other) == sk) {
493 other->sk_err = ECONNRESET;
494 sk_error_report(other);
495 }
496 }
497}
498
499static void unix_sock_destructor(struct sock *sk)
500{
501 struct unix_sock *u = unix_sk(sk);
502
503 skb_queue_purge(&sk->sk_receive_queue);
504
505 WARN_ON(refcount_read(&sk->sk_wmem_alloc));
506 WARN_ON(!sk_unhashed(sk));
507 WARN_ON(sk->sk_socket);
508 if (!sock_flag(sk, SOCK_DEAD)) {
509 pr_info("Attempt to release alive unix socket: %p\n", sk);
510 return;
511 }
512
513 if (u->addr)
514 unix_release_addr(u->addr);
515
516 atomic_long_dec(&unix_nr_socks);
517 local_bh_disable();
518 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
519 local_bh_enable();
520#ifdef UNIX_REFCNT_DEBUG
521 pr_debug("UNIX %p is destroyed, %ld are still alive.\n", sk,
522 atomic_long_read(&unix_nr_socks));
523#endif
524}
525
526static void unix_release_sock(struct sock *sk, int embrion)
527{
528 struct unix_sock *u = unix_sk(sk);
529 struct path path;
530 struct sock *skpair;
531 struct sk_buff *skb;
532 int state;
533
534 unix_remove_socket(sk);
535
536 /* Clear state */
537 unix_state_lock(sk);
538 sock_orphan(sk);
539 sk->sk_shutdown = SHUTDOWN_MASK;
540 path = u->path;
541 u->path.dentry = NULL;
542 u->path.mnt = NULL;
543 state = sk->sk_state;
544 sk->sk_state = TCP_CLOSE;
545
546 skpair = unix_peer(sk);
547 unix_peer(sk) = NULL;
548
549 unix_state_unlock(sk);
550
551 wake_up_interruptible_all(&u->peer_wait);
552
553 if (skpair != NULL) {
554 if (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) {
555 unix_state_lock(skpair);
556 /* No more writes */
557 skpair->sk_shutdown = SHUTDOWN_MASK;
558 if (!skb_queue_empty(&sk->sk_receive_queue) || embrion)
559 skpair->sk_err = ECONNRESET;
560 unix_state_unlock(skpair);
561 skpair->sk_state_change(skpair);
562 sk_wake_async(skpair, SOCK_WAKE_WAITD, POLL_HUP);
563 }
564
565 unix_dgram_peer_wake_disconnect(sk, skpair);
566 sock_put(skpair); /* It may now die */
567 }
568
569 /* Try to flush out this socket. Throw out buffers at least */
570
571 while ((skb = skb_dequeue(&sk->sk_receive_queue)) != NULL) {
572 if (state == TCP_LISTEN)
573 unix_release_sock(skb->sk, 1);
574 /* passed fds are erased in the kfree_skb hook */
575 UNIXCB(skb).consumed = skb->len;
576 kfree_skb(skb);
577 }
578
579 if (path.dentry)
580 path_put(&path);
581
582 sock_put(sk);
583
584 /* ---- Socket is dead now and most probably destroyed ---- */
585
586 /*
587 * Fixme: BSD difference: In BSD all sockets connected to us get
588 * ECONNRESET and we die on the spot. In Linux we behave
589 * like files and pipes do and wait for the last
590 * dereference.
591 *
592 * Can't we simply set sock->err?
593 *
594 * What the above comment does talk about? --ANK(980817)
595 */
596
597 if (unix_tot_inflight)
598 unix_gc(); /* Garbage collect fds */
599}
600
601static void init_peercred(struct sock *sk)
602{
603 const struct cred *old_cred;
604 struct pid *old_pid;
605
606 spin_lock(&sk->sk_peer_lock);
607 old_pid = sk->sk_peer_pid;
608 old_cred = sk->sk_peer_cred;
609 sk->sk_peer_pid = get_pid(task_tgid(current));
610 sk->sk_peer_cred = get_current_cred();
611 spin_unlock(&sk->sk_peer_lock);
612
613 put_pid(old_pid);
614 put_cred(old_cred);
615}
616
617static void copy_peercred(struct sock *sk, struct sock *peersk)
618{
619 const struct cred *old_cred;
620 struct pid *old_pid;
621
622 if (sk < peersk) {
623 spin_lock(&sk->sk_peer_lock);
624 spin_lock_nested(&peersk->sk_peer_lock, SINGLE_DEPTH_NESTING);
625 } else {
626 spin_lock(&peersk->sk_peer_lock);
627 spin_lock_nested(&sk->sk_peer_lock, SINGLE_DEPTH_NESTING);
628 }
629 old_pid = sk->sk_peer_pid;
630 old_cred = sk->sk_peer_cred;
631 sk->sk_peer_pid = get_pid(peersk->sk_peer_pid);
632 sk->sk_peer_cred = get_cred(peersk->sk_peer_cred);
633
634 spin_unlock(&sk->sk_peer_lock);
635 spin_unlock(&peersk->sk_peer_lock);
636
637 put_pid(old_pid);
638 put_cred(old_cred);
639}
640
641static int unix_listen(struct socket *sock, int backlog)
642{
643 int err;
644 struct sock *sk = sock->sk;
645 struct unix_sock *u = unix_sk(sk);
646
647 err = -EOPNOTSUPP;
648 if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
649 goto out; /* Only stream/seqpacket sockets accept */
650 err = -EINVAL;
651 if (!u->addr)
652 goto out; /* No listens on an unbound socket */
653 unix_state_lock(sk);
654 if (sk->sk_state != TCP_CLOSE && sk->sk_state != TCP_LISTEN)
655 goto out_unlock;
656 if (backlog > sk->sk_max_ack_backlog)
657 wake_up_interruptible_all(&u->peer_wait);
658 sk->sk_max_ack_backlog = backlog;
659 sk->sk_state = TCP_LISTEN;
660 /* set credentials so connect can copy them */
661 init_peercred(sk);
662 err = 0;
663
664out_unlock:
665 unix_state_unlock(sk);
666out:
667 return err;
668}
669
670static int unix_release(struct socket *);
671static int unix_bind(struct socket *, struct sockaddr *, int);
672static int unix_stream_connect(struct socket *, struct sockaddr *,
673 int addr_len, int flags);
674static int unix_socketpair(struct socket *, struct socket *);
675static int unix_accept(struct socket *, struct socket *, int, bool);
676static int unix_getname(struct socket *, struct sockaddr *, int);
677static __poll_t unix_poll(struct file *, struct socket *, poll_table *);
678static __poll_t unix_dgram_poll(struct file *, struct socket *,
679 poll_table *);
680static int unix_ioctl(struct socket *, unsigned int, unsigned long);
681#ifdef CONFIG_COMPAT
682static int unix_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg);
683#endif
684static int unix_shutdown(struct socket *, int);
685static int unix_stream_sendmsg(struct socket *, struct msghdr *, size_t);
686static int unix_stream_recvmsg(struct socket *, struct msghdr *, size_t, int);
687static ssize_t unix_stream_sendpage(struct socket *, struct page *, int offset,
688 size_t size, int flags);
689static ssize_t unix_stream_splice_read(struct socket *, loff_t *ppos,
690 struct pipe_inode_info *, size_t size,
691 unsigned int flags);
692static int unix_dgram_sendmsg(struct socket *, struct msghdr *, size_t);
693static int unix_dgram_recvmsg(struct socket *, struct msghdr *, size_t, int);
694static int unix_dgram_connect(struct socket *, struct sockaddr *,
695 int, int);
696static int unix_seqpacket_sendmsg(struct socket *, struct msghdr *, size_t);
697static int unix_seqpacket_recvmsg(struct socket *, struct msghdr *, size_t,
698 int);
699
700static int unix_set_peek_off(struct sock *sk, int val)
701{
702 struct unix_sock *u = unix_sk(sk);
703
704 if (mutex_lock_interruptible(&u->iolock))
705 return -EINTR;
706
707 sk->sk_peek_off = val;
708 mutex_unlock(&u->iolock);
709
710 return 0;
711}
712
713#ifdef CONFIG_PROC_FS
714static void unix_show_fdinfo(struct seq_file *m, struct socket *sock)
715{
716 struct sock *sk = sock->sk;
717 struct unix_sock *u;
718
719 if (sk) {
720 u = unix_sk(sock->sk);
721 seq_printf(m, "scm_fds: %u\n",
722 atomic_read(&u->scm_stat.nr_fds));
723 }
724}
725#else
726#define unix_show_fdinfo NULL
727#endif
728
729static const struct proto_ops unix_stream_ops = {
730 .family = PF_UNIX,
731 .owner = THIS_MODULE,
732 .release = unix_release,
733 .bind = unix_bind,
734 .connect = unix_stream_connect,
735 .socketpair = unix_socketpair,
736 .accept = unix_accept,
737 .getname = unix_getname,
738 .poll = unix_poll,
739 .ioctl = unix_ioctl,
740#ifdef CONFIG_COMPAT
741 .compat_ioctl = unix_compat_ioctl,
742#endif
743 .listen = unix_listen,
744 .shutdown = unix_shutdown,
745 .sendmsg = unix_stream_sendmsg,
746 .recvmsg = unix_stream_recvmsg,
747 .mmap = sock_no_mmap,
748 .sendpage = unix_stream_sendpage,
749 .splice_read = unix_stream_splice_read,
750 .set_peek_off = unix_set_peek_off,
751 .show_fdinfo = unix_show_fdinfo,
752};
753
754static const struct proto_ops unix_dgram_ops = {
755 .family = PF_UNIX,
756 .owner = THIS_MODULE,
757 .release = unix_release,
758 .bind = unix_bind,
759 .connect = unix_dgram_connect,
760 .socketpair = unix_socketpair,
761 .accept = sock_no_accept,
762 .getname = unix_getname,
763 .poll = unix_dgram_poll,
764 .ioctl = unix_ioctl,
765#ifdef CONFIG_COMPAT
766 .compat_ioctl = unix_compat_ioctl,
767#endif
768 .listen = sock_no_listen,
769 .shutdown = unix_shutdown,
770 .sendmsg = unix_dgram_sendmsg,
771 .recvmsg = unix_dgram_recvmsg,
772 .mmap = sock_no_mmap,
773 .sendpage = sock_no_sendpage,
774 .set_peek_off = unix_set_peek_off,
775 .show_fdinfo = unix_show_fdinfo,
776};
777
778static const struct proto_ops unix_seqpacket_ops = {
779 .family = PF_UNIX,
780 .owner = THIS_MODULE,
781 .release = unix_release,
782 .bind = unix_bind,
783 .connect = unix_stream_connect,
784 .socketpair = unix_socketpair,
785 .accept = unix_accept,
786 .getname = unix_getname,
787 .poll = unix_dgram_poll,
788 .ioctl = unix_ioctl,
789#ifdef CONFIG_COMPAT
790 .compat_ioctl = unix_compat_ioctl,
791#endif
792 .listen = unix_listen,
793 .shutdown = unix_shutdown,
794 .sendmsg = unix_seqpacket_sendmsg,
795 .recvmsg = unix_seqpacket_recvmsg,
796 .mmap = sock_no_mmap,
797 .sendpage = sock_no_sendpage,
798 .set_peek_off = unix_set_peek_off,
799 .show_fdinfo = unix_show_fdinfo,
800};
801
802static struct proto unix_proto = {
803 .name = "UNIX",
804 .owner = THIS_MODULE,
805 .obj_size = sizeof(struct unix_sock),
806};
807
808static struct sock *unix_create1(struct net *net, struct socket *sock, int kern)
809{
810 struct sock *sk = NULL;
811 struct unix_sock *u;
812
813 atomic_long_inc(&unix_nr_socks);
814 if (atomic_long_read(&unix_nr_socks) > 2 * get_max_files())
815 goto out;
816
817 sk = sk_alloc(net, PF_UNIX, GFP_KERNEL, &unix_proto, kern);
818 if (!sk)
819 goto out;
820
821 sock_init_data(sock, sk);
822
823 sk->sk_allocation = GFP_KERNEL_ACCOUNT;
824 sk->sk_write_space = unix_write_space;
825 sk->sk_max_ack_backlog = net->unx.sysctl_max_dgram_qlen;
826 sk->sk_destruct = unix_sock_destructor;
827 u = unix_sk(sk);
828 u->path.dentry = NULL;
829 u->path.mnt = NULL;
830 spin_lock_init(&u->lock);
831 atomic_long_set(&u->inflight, 0);
832 INIT_LIST_HEAD(&u->link);
833 mutex_init(&u->iolock); /* single task reading lock */
834 mutex_init(&u->bindlock); /* single task binding lock */
835 init_waitqueue_head(&u->peer_wait);
836 init_waitqueue_func_entry(&u->peer_wake, unix_dgram_peer_wake_relay);
837 memset(&u->scm_stat, 0, sizeof(struct scm_stat));
838 unix_insert_socket(unix_sockets_unbound(sk), sk);
839out:
840 if (sk == NULL)
841 atomic_long_dec(&unix_nr_socks);
842 else {
843 local_bh_disable();
844 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
845 local_bh_enable();
846 }
847 return sk;
848}
849
850static int unix_create(struct net *net, struct socket *sock, int protocol,
851 int kern)
852{
853 if (protocol && protocol != PF_UNIX)
854 return -EPROTONOSUPPORT;
855
856 sock->state = SS_UNCONNECTED;
857
858 switch (sock->type) {
859 case SOCK_STREAM:
860 sock->ops = &unix_stream_ops;
861 break;
862 /*
863 * Believe it or not BSD has AF_UNIX, SOCK_RAW though
864 * nothing uses it.
865 */
866 case SOCK_RAW:
867 sock->type = SOCK_DGRAM;
868 fallthrough;
869 case SOCK_DGRAM:
870 sock->ops = &unix_dgram_ops;
871 break;
872 case SOCK_SEQPACKET:
873 sock->ops = &unix_seqpacket_ops;
874 break;
875 default:
876 return -ESOCKTNOSUPPORT;
877 }
878
879 return unix_create1(net, sock, kern) ? 0 : -ENOMEM;
880}
881
882static int unix_release(struct socket *sock)
883{
884 struct sock *sk = sock->sk;
885
886 if (!sk)
887 return 0;
888
889 unix_release_sock(sk, 0);
890 sock->sk = NULL;
891
892 return 0;
893}
894
895static int unix_autobind(struct socket *sock)
896{
897 struct sock *sk = sock->sk;
898 struct net *net = sock_net(sk);
899 struct unix_sock *u = unix_sk(sk);
900 static u32 ordernum = 1;
901 struct unix_address *addr;
902 int err;
903 unsigned int retries = 0;
904
905 err = mutex_lock_interruptible(&u->bindlock);
906 if (err)
907 return err;
908
909 if (u->addr)
910 goto out;
911
912 err = -ENOMEM;
913 addr = kzalloc(sizeof(*addr) + sizeof(short) + 16, GFP_KERNEL);
914 if (!addr)
915 goto out;
916
917 addr->name->sun_family = AF_UNIX;
918 refcount_set(&addr->refcnt, 1);
919
920retry:
921 addr->len = sprintf(addr->name->sun_path+1, "%05x", ordernum) + 1 + sizeof(short);
922 addr->hash = unix_hash_fold(csum_partial(addr->name, addr->len, 0));
923 addr->hash ^= sk->sk_type;
924
925 spin_lock(&unix_table_lock);
926 ordernum = (ordernum+1)&0xFFFFF;
927
928 if (__unix_find_socket_byname(net, addr->name, addr->len, addr->hash)) {
929 spin_unlock(&unix_table_lock);
930 /*
931 * __unix_find_socket_byname() may take long time if many names
932 * are already in use.
933 */
934 cond_resched();
935 /* Give up if all names seems to be in use. */
936 if (retries++ == 0xFFFFF) {
937 err = -ENOSPC;
938 kfree(addr);
939 goto out;
940 }
941 goto retry;
942 }
943
944 __unix_set_addr(sk, addr, addr->hash);
945 spin_unlock(&unix_table_lock);
946 err = 0;
947
948out: mutex_unlock(&u->bindlock);
949 return err;
950}
951
952static struct sock *unix_find_other(struct net *net,
953 struct sockaddr_un *sunname, int len,
954 int type, unsigned int hash, int *error)
955{
956 struct sock *u;
957 struct path path;
958 int err = 0;
959
960 if (sunname->sun_path[0]) {
961 struct inode *inode;
962 err = kern_path(sunname->sun_path, LOOKUP_FOLLOW, &path);
963 if (err)
964 goto fail;
965 inode = d_backing_inode(path.dentry);
966 err = path_permission(&path, MAY_WRITE);
967 if (err)
968 goto put_fail;
969
970 err = -ECONNREFUSED;
971 if (!S_ISSOCK(inode->i_mode))
972 goto put_fail;
973 u = unix_find_socket_byinode(inode);
974 if (!u)
975 goto put_fail;
976
977 if (u->sk_type == type)
978 touch_atime(&path);
979
980 path_put(&path);
981
982 err = -EPROTOTYPE;
983 if (u->sk_type != type) {
984 sock_put(u);
985 goto fail;
986 }
987 } else {
988 err = -ECONNREFUSED;
989 u = unix_find_socket_byname(net, sunname, len, type ^ hash);
990 if (u) {
991 struct dentry *dentry;
992 dentry = unix_sk(u)->path.dentry;
993 if (dentry)
994 touch_atime(&unix_sk(u)->path);
995 } else
996 goto fail;
997 }
998 return u;
999
1000put_fail:
1001 path_put(&path);
1002fail:
1003 *error = err;
1004 return NULL;
1005}
1006
1007static int unix_bind_bsd(struct sock *sk, struct unix_address *addr)
1008{
1009 struct unix_sock *u = unix_sk(sk);
1010 umode_t mode = S_IFSOCK |
1011 (SOCK_INODE(sk->sk_socket)->i_mode & ~current_umask());
1012 struct user_namespace *ns; // barf...
1013 struct path parent;
1014 struct dentry *dentry;
1015 unsigned int hash;
1016 int err;
1017
1018 /*
1019 * Get the parent directory, calculate the hash for last
1020 * component.
1021 */
1022 dentry = kern_path_create(AT_FDCWD, addr->name->sun_path, &parent, 0);
1023 if (IS_ERR(dentry))
1024 return PTR_ERR(dentry);
1025 ns = mnt_user_ns(parent.mnt);
1026
1027 /*
1028 * All right, let's create it.
1029 */
1030 err = security_path_mknod(&parent, dentry, mode, 0);
1031 if (!err)
1032 err = vfs_mknod(ns, d_inode(parent.dentry), dentry, mode, 0);
1033 if (err)
1034 goto out;
1035 err = mutex_lock_interruptible(&u->bindlock);
1036 if (err)
1037 goto out_unlink;
1038 if (u->addr)
1039 goto out_unlock;
1040
1041 addr->hash = UNIX_HASH_SIZE;
1042 hash = d_backing_inode(dentry)->i_ino & (UNIX_HASH_SIZE - 1);
1043 spin_lock(&unix_table_lock);
1044 u->path.mnt = mntget(parent.mnt);
1045 u->path.dentry = dget(dentry);
1046 __unix_set_addr(sk, addr, hash);
1047 spin_unlock(&unix_table_lock);
1048 mutex_unlock(&u->bindlock);
1049 done_path_create(&parent, dentry);
1050 return 0;
1051
1052out_unlock:
1053 mutex_unlock(&u->bindlock);
1054 err = -EINVAL;
1055out_unlink:
1056 /* failed after successful mknod? unlink what we'd created... */
1057 vfs_unlink(ns, d_inode(parent.dentry), dentry, NULL);
1058out:
1059 done_path_create(&parent, dentry);
1060 return err;
1061}
1062
1063static int unix_bind_abstract(struct sock *sk, struct unix_address *addr)
1064{
1065 struct unix_sock *u = unix_sk(sk);
1066 int err;
1067
1068 err = mutex_lock_interruptible(&u->bindlock);
1069 if (err)
1070 return err;
1071
1072 if (u->addr) {
1073 mutex_unlock(&u->bindlock);
1074 return -EINVAL;
1075 }
1076
1077 spin_lock(&unix_table_lock);
1078 if (__unix_find_socket_byname(sock_net(sk), addr->name, addr->len,
1079 addr->hash)) {
1080 spin_unlock(&unix_table_lock);
1081 mutex_unlock(&u->bindlock);
1082 return -EADDRINUSE;
1083 }
1084 __unix_set_addr(sk, addr, addr->hash);
1085 spin_unlock(&unix_table_lock);
1086 mutex_unlock(&u->bindlock);
1087 return 0;
1088}
1089
1090static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
1091{
1092 struct sock *sk = sock->sk;
1093 struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
1094 char *sun_path = sunaddr->sun_path;
1095 int err;
1096 unsigned int hash;
1097 struct unix_address *addr;
1098
1099 if (addr_len < offsetofend(struct sockaddr_un, sun_family) ||
1100 sunaddr->sun_family != AF_UNIX)
1101 return -EINVAL;
1102
1103 if (addr_len == sizeof(short))
1104 return unix_autobind(sock);
1105
1106 err = unix_mkname(sunaddr, addr_len, &hash);
1107 if (err < 0)
1108 return err;
1109 addr_len = err;
1110 addr = kmalloc(sizeof(*addr)+addr_len, GFP_KERNEL);
1111 if (!addr)
1112 return -ENOMEM;
1113
1114 memcpy(addr->name, sunaddr, addr_len);
1115 addr->len = addr_len;
1116 addr->hash = hash ^ sk->sk_type;
1117 refcount_set(&addr->refcnt, 1);
1118
1119 if (sun_path[0])
1120 err = unix_bind_bsd(sk, addr);
1121 else
1122 err = unix_bind_abstract(sk, addr);
1123 if (err)
1124 unix_release_addr(addr);
1125 return err == -EEXIST ? -EADDRINUSE : err;
1126}
1127
1128static void unix_state_double_lock(struct sock *sk1, struct sock *sk2)
1129{
1130 if (unlikely(sk1 == sk2) || !sk2) {
1131 unix_state_lock(sk1);
1132 return;
1133 }
1134 if (sk1 < sk2) {
1135 unix_state_lock(sk1);
1136 unix_state_lock_nested(sk2);
1137 } else {
1138 unix_state_lock(sk2);
1139 unix_state_lock_nested(sk1);
1140 }
1141}
1142
1143static void unix_state_double_unlock(struct sock *sk1, struct sock *sk2)
1144{
1145 if (unlikely(sk1 == sk2) || !sk2) {
1146 unix_state_unlock(sk1);
1147 return;
1148 }
1149 unix_state_unlock(sk1);
1150 unix_state_unlock(sk2);
1151}
1152
1153static int unix_dgram_connect(struct socket *sock, struct sockaddr *addr,
1154 int alen, int flags)
1155{
1156 struct sock *sk = sock->sk;
1157 struct net *net = sock_net(sk);
1158 struct sockaddr_un *sunaddr = (struct sockaddr_un *)addr;
1159 struct sock *other;
1160 unsigned int hash;
1161 int err;
1162
1163 err = -EINVAL;
1164 if (alen < offsetofend(struct sockaddr, sa_family))
1165 goto out;
1166
1167 if (addr->sa_family != AF_UNSPEC) {
1168 err = unix_mkname(sunaddr, alen, &hash);
1169 if (err < 0)
1170 goto out;
1171 alen = err;
1172
1173 if (test_bit(SOCK_PASSCRED, &sock->flags) &&
1174 !unix_sk(sk)->addr && (err = unix_autobind(sock)) != 0)
1175 goto out;
1176
1177restart:
1178 other = unix_find_other(net, sunaddr, alen, sock->type, hash, &err);
1179 if (!other)
1180 goto out;
1181
1182 unix_state_double_lock(sk, other);
1183
1184 /* Apparently VFS overslept socket death. Retry. */
1185 if (sock_flag(other, SOCK_DEAD)) {
1186 unix_state_double_unlock(sk, other);
1187 sock_put(other);
1188 goto restart;
1189 }
1190
1191 err = -EPERM;
1192 if (!unix_may_send(sk, other))
1193 goto out_unlock;
1194
1195 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1196 if (err)
1197 goto out_unlock;
1198
1199 } else {
1200 /*
1201 * 1003.1g breaking connected state with AF_UNSPEC
1202 */
1203 other = NULL;
1204 unix_state_double_lock(sk, other);
1205 }
1206
1207 /*
1208 * If it was connected, reconnect.
1209 */
1210 if (unix_peer(sk)) {
1211 struct sock *old_peer = unix_peer(sk);
1212 unix_peer(sk) = other;
1213 unix_dgram_peer_wake_disconnect_wakeup(sk, old_peer);
1214
1215 unix_state_double_unlock(sk, other);
1216
1217 if (other != old_peer)
1218 unix_dgram_disconnected(sk, old_peer);
1219 sock_put(old_peer);
1220 } else {
1221 unix_peer(sk) = other;
1222 unix_state_double_unlock(sk, other);
1223 }
1224 return 0;
1225
1226out_unlock:
1227 unix_state_double_unlock(sk, other);
1228 sock_put(other);
1229out:
1230 return err;
1231}
1232
1233static long unix_wait_for_peer(struct sock *other, long timeo)
1234 __releases(&unix_sk(other)->lock)
1235{
1236 struct unix_sock *u = unix_sk(other);
1237 int sched;
1238 DEFINE_WAIT(wait);
1239
1240 prepare_to_wait_exclusive(&u->peer_wait, &wait, TASK_INTERRUPTIBLE);
1241
1242 sched = !sock_flag(other, SOCK_DEAD) &&
1243 !(other->sk_shutdown & RCV_SHUTDOWN) &&
1244 unix_recvq_full(other);
1245
1246 unix_state_unlock(other);
1247
1248 if (sched)
1249 timeo = schedule_timeout(timeo);
1250
1251 finish_wait(&u->peer_wait, &wait);
1252 return timeo;
1253}
1254
1255static int unix_stream_connect(struct socket *sock, struct sockaddr *uaddr,
1256 int addr_len, int flags)
1257{
1258 struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
1259 struct sock *sk = sock->sk;
1260 struct net *net = sock_net(sk);
1261 struct unix_sock *u = unix_sk(sk), *newu, *otheru;
1262 struct sock *newsk = NULL;
1263 struct sock *other = NULL;
1264 struct sk_buff *skb = NULL;
1265 unsigned int hash;
1266 int st;
1267 int err;
1268 long timeo;
1269
1270 err = unix_mkname(sunaddr, addr_len, &hash);
1271 if (err < 0)
1272 goto out;
1273 addr_len = err;
1274
1275 if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr &&
1276 (err = unix_autobind(sock)) != 0)
1277 goto out;
1278
1279 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
1280
1281 /* First of all allocate resources.
1282 If we will make it after state is locked,
1283 we will have to recheck all again in any case.
1284 */
1285
1286 err = -ENOMEM;
1287
1288 /* create new sock for complete connection */
1289 newsk = unix_create1(sock_net(sk), NULL, 0);
1290 if (newsk == NULL)
1291 goto out;
1292
1293 /* Allocate skb for sending to listening sock */
1294 skb = sock_wmalloc(newsk, 1, 0, GFP_KERNEL);
1295 if (skb == NULL)
1296 goto out;
1297
1298restart:
1299 /* Find listening sock. */
1300 other = unix_find_other(net, sunaddr, addr_len, sk->sk_type, hash, &err);
1301 if (!other)
1302 goto out;
1303
1304 /* Latch state of peer */
1305 unix_state_lock(other);
1306
1307 /* Apparently VFS overslept socket death. Retry. */
1308 if (sock_flag(other, SOCK_DEAD)) {
1309 unix_state_unlock(other);
1310 sock_put(other);
1311 goto restart;
1312 }
1313
1314 err = -ECONNREFUSED;
1315 if (other->sk_state != TCP_LISTEN)
1316 goto out_unlock;
1317 if (other->sk_shutdown & RCV_SHUTDOWN)
1318 goto out_unlock;
1319
1320 if (unix_recvq_full(other)) {
1321 err = -EAGAIN;
1322 if (!timeo)
1323 goto out_unlock;
1324
1325 timeo = unix_wait_for_peer(other, timeo);
1326
1327 err = sock_intr_errno(timeo);
1328 if (signal_pending(current))
1329 goto out;
1330 sock_put(other);
1331 goto restart;
1332 }
1333
1334 /* Latch our state.
1335
1336 It is tricky place. We need to grab our state lock and cannot
1337 drop lock on peer. It is dangerous because deadlock is
1338 possible. Connect to self case and simultaneous
1339 attempt to connect are eliminated by checking socket
1340 state. other is TCP_LISTEN, if sk is TCP_LISTEN we
1341 check this before attempt to grab lock.
1342
1343 Well, and we have to recheck the state after socket locked.
1344 */
1345 st = sk->sk_state;
1346
1347 switch (st) {
1348 case TCP_CLOSE:
1349 /* This is ok... continue with connect */
1350 break;
1351 case TCP_ESTABLISHED:
1352 /* Socket is already connected */
1353 err = -EISCONN;
1354 goto out_unlock;
1355 default:
1356 err = -EINVAL;
1357 goto out_unlock;
1358 }
1359
1360 unix_state_lock_nested(sk);
1361
1362 if (sk->sk_state != st) {
1363 unix_state_unlock(sk);
1364 unix_state_unlock(other);
1365 sock_put(other);
1366 goto restart;
1367 }
1368
1369 err = security_unix_stream_connect(sk, other, newsk);
1370 if (err) {
1371 unix_state_unlock(sk);
1372 goto out_unlock;
1373 }
1374
1375 /* The way is open! Fastly set all the necessary fields... */
1376
1377 sock_hold(sk);
1378 unix_peer(newsk) = sk;
1379 newsk->sk_state = TCP_ESTABLISHED;
1380 newsk->sk_type = sk->sk_type;
1381 init_peercred(newsk);
1382 newu = unix_sk(newsk);
1383 RCU_INIT_POINTER(newsk->sk_wq, &newu->peer_wq);
1384 otheru = unix_sk(other);
1385
1386 /* copy address information from listening to new sock
1387 *
1388 * The contents of *(otheru->addr) and otheru->path
1389 * are seen fully set up here, since we have found
1390 * otheru in hash under unix_table_lock. Insertion
1391 * into the hash chain we'd found it in had been done
1392 * in an earlier critical area protected by unix_table_lock,
1393 * the same one where we'd set *(otheru->addr) contents,
1394 * as well as otheru->path and otheru->addr itself.
1395 *
1396 * Using smp_store_release() here to set newu->addr
1397 * is enough to make those stores, as well as stores
1398 * to newu->path visible to anyone who gets newu->addr
1399 * by smp_load_acquire(). IOW, the same warranties
1400 * as for unix_sock instances bound in unix_bind() or
1401 * in unix_autobind().
1402 */
1403 if (otheru->path.dentry) {
1404 path_get(&otheru->path);
1405 newu->path = otheru->path;
1406 }
1407 refcount_inc(&otheru->addr->refcnt);
1408 smp_store_release(&newu->addr, otheru->addr);
1409
1410 /* Set credentials */
1411 copy_peercred(sk, other);
1412
1413 sock->state = SS_CONNECTED;
1414 sk->sk_state = TCP_ESTABLISHED;
1415 sock_hold(newsk);
1416
1417 smp_mb__after_atomic(); /* sock_hold() does an atomic_inc() */
1418 unix_peer(sk) = newsk;
1419
1420 unix_state_unlock(sk);
1421
1422 /* take ten and send info to listening sock */
1423 spin_lock(&other->sk_receive_queue.lock);
1424 __skb_queue_tail(&other->sk_receive_queue, skb);
1425 spin_unlock(&other->sk_receive_queue.lock);
1426 unix_state_unlock(other);
1427 other->sk_data_ready(other);
1428 sock_put(other);
1429 return 0;
1430
1431out_unlock:
1432 if (other)
1433 unix_state_unlock(other);
1434
1435out:
1436 kfree_skb(skb);
1437 if (newsk)
1438 unix_release_sock(newsk, 0);
1439 if (other)
1440 sock_put(other);
1441 return err;
1442}
1443
1444static int unix_socketpair(struct socket *socka, struct socket *sockb)
1445{
1446 struct sock *ska = socka->sk, *skb = sockb->sk;
1447
1448 /* Join our sockets back to back */
1449 sock_hold(ska);
1450 sock_hold(skb);
1451 unix_peer(ska) = skb;
1452 unix_peer(skb) = ska;
1453 init_peercred(ska);
1454 init_peercred(skb);
1455
1456 if (ska->sk_type != SOCK_DGRAM) {
1457 ska->sk_state = TCP_ESTABLISHED;
1458 skb->sk_state = TCP_ESTABLISHED;
1459 socka->state = SS_CONNECTED;
1460 sockb->state = SS_CONNECTED;
1461 }
1462 return 0;
1463}
1464
1465static void unix_sock_inherit_flags(const struct socket *old,
1466 struct socket *new)
1467{
1468 if (test_bit(SOCK_PASSCRED, &old->flags))
1469 set_bit(SOCK_PASSCRED, &new->flags);
1470 if (test_bit(SOCK_PASSSEC, &old->flags))
1471 set_bit(SOCK_PASSSEC, &new->flags);
1472}
1473
1474static int unix_accept(struct socket *sock, struct socket *newsock, int flags,
1475 bool kern)
1476{
1477 struct sock *sk = sock->sk;
1478 struct sock *tsk;
1479 struct sk_buff *skb;
1480 int err;
1481
1482 err = -EOPNOTSUPP;
1483 if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
1484 goto out;
1485
1486 err = -EINVAL;
1487 if (sk->sk_state != TCP_LISTEN)
1488 goto out;
1489
1490 /* If socket state is TCP_LISTEN it cannot change (for now...),
1491 * so that no locks are necessary.
1492 */
1493
1494 skb = skb_recv_datagram(sk, 0, flags&O_NONBLOCK, &err);
1495 if (!skb) {
1496 /* This means receive shutdown. */
1497 if (err == 0)
1498 err = -EINVAL;
1499 goto out;
1500 }
1501
1502 tsk = skb->sk;
1503 skb_free_datagram(sk, skb);
1504 wake_up_interruptible(&unix_sk(sk)->peer_wait);
1505
1506 /* attach accepted sock to socket */
1507 unix_state_lock(tsk);
1508 newsock->state = SS_CONNECTED;
1509 unix_sock_inherit_flags(sock, newsock);
1510 sock_graft(tsk, newsock);
1511 unix_state_unlock(tsk);
1512 return 0;
1513
1514out:
1515 return err;
1516}
1517
1518
1519static int unix_getname(struct socket *sock, struct sockaddr *uaddr, int peer)
1520{
1521 struct sock *sk = sock->sk;
1522 struct unix_address *addr;
1523 DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr, uaddr);
1524 int err = 0;
1525
1526 if (peer) {
1527 sk = unix_peer_get(sk);
1528
1529 err = -ENOTCONN;
1530 if (!sk)
1531 goto out;
1532 err = 0;
1533 } else {
1534 sock_hold(sk);
1535 }
1536
1537 addr = smp_load_acquire(&unix_sk(sk)->addr);
1538 if (!addr) {
1539 sunaddr->sun_family = AF_UNIX;
1540 sunaddr->sun_path[0] = 0;
1541 err = sizeof(short);
1542 } else {
1543 err = addr->len;
1544 memcpy(sunaddr, addr->name, addr->len);
1545 }
1546 sock_put(sk);
1547out:
1548 return err;
1549}
1550
1551static void unix_peek_fds(struct scm_cookie *scm, struct sk_buff *skb)
1552{
1553 scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1554
1555 /*
1556 * Garbage collection of unix sockets starts by selecting a set of
1557 * candidate sockets which have reference only from being in flight
1558 * (total_refs == inflight_refs). This condition is checked once during
1559 * the candidate collection phase, and candidates are marked as such, so
1560 * that non-candidates can later be ignored. While inflight_refs is
1561 * protected by unix_gc_lock, total_refs (file count) is not, hence this
1562 * is an instantaneous decision.
1563 *
1564 * Once a candidate, however, the socket must not be reinstalled into a
1565 * file descriptor while the garbage collection is in progress.
1566 *
1567 * If the above conditions are met, then the directed graph of
1568 * candidates (*) does not change while unix_gc_lock is held.
1569 *
1570 * Any operations that changes the file count through file descriptors
1571 * (dup, close, sendmsg) does not change the graph since candidates are
1572 * not installed in fds.
1573 *
1574 * Dequeing a candidate via recvmsg would install it into an fd, but
1575 * that takes unix_gc_lock to decrement the inflight count, so it's
1576 * serialized with garbage collection.
1577 *
1578 * MSG_PEEK is special in that it does not change the inflight count,
1579 * yet does install the socket into an fd. The following lock/unlock
1580 * pair is to ensure serialization with garbage collection. It must be
1581 * done between incrementing the file count and installing the file into
1582 * an fd.
1583 *
1584 * If garbage collection starts after the barrier provided by the
1585 * lock/unlock, then it will see the elevated refcount and not mark this
1586 * as a candidate. If a garbage collection is already in progress
1587 * before the file count was incremented, then the lock/unlock pair will
1588 * ensure that garbage collection is finished before progressing to
1589 * installing the fd.
1590 *
1591 * (*) A -> B where B is on the queue of A or B is on the queue of C
1592 * which is on the queue of listening socket A.
1593 */
1594 spin_lock(&unix_gc_lock);
1595 spin_unlock(&unix_gc_lock);
1596}
1597
1598static int unix_scm_to_skb(struct scm_cookie *scm, struct sk_buff *skb, bool send_fds)
1599{
1600 int err = 0;
1601
1602 UNIXCB(skb).pid = get_pid(scm->pid);
1603 UNIXCB(skb).uid = scm->creds.uid;
1604 UNIXCB(skb).gid = scm->creds.gid;
1605 UNIXCB(skb).fp = NULL;
1606 unix_get_secdata(scm, skb);
1607 if (scm->fp && send_fds)
1608 err = unix_attach_fds(scm, skb);
1609
1610 skb->destructor = unix_destruct_scm;
1611 return err;
1612}
1613
1614static bool unix_passcred_enabled(const struct socket *sock,
1615 const struct sock *other)
1616{
1617 return test_bit(SOCK_PASSCRED, &sock->flags) ||
1618 !other->sk_socket ||
1619 test_bit(SOCK_PASSCRED, &other->sk_socket->flags);
1620}
1621
1622/*
1623 * Some apps rely on write() giving SCM_CREDENTIALS
1624 * We include credentials if source or destination socket
1625 * asserted SOCK_PASSCRED.
1626 */
1627static void maybe_add_creds(struct sk_buff *skb, const struct socket *sock,
1628 const struct sock *other)
1629{
1630 if (UNIXCB(skb).pid)
1631 return;
1632 if (unix_passcred_enabled(sock, other)) {
1633 UNIXCB(skb).pid = get_pid(task_tgid(current));
1634 current_uid_gid(&UNIXCB(skb).uid, &UNIXCB(skb).gid);
1635 }
1636}
1637
1638static int maybe_init_creds(struct scm_cookie *scm,
1639 struct socket *socket,
1640 const struct sock *other)
1641{
1642 int err;
1643 struct msghdr msg = { .msg_controllen = 0 };
1644
1645 err = scm_send(socket, &msg, scm, false);
1646 if (err)
1647 return err;
1648
1649 if (unix_passcred_enabled(socket, other)) {
1650 scm->pid = get_pid(task_tgid(current));
1651 current_uid_gid(&scm->creds.uid, &scm->creds.gid);
1652 }
1653 return err;
1654}
1655
1656static bool unix_skb_scm_eq(struct sk_buff *skb,
1657 struct scm_cookie *scm)
1658{
1659 const struct unix_skb_parms *u = &UNIXCB(skb);
1660
1661 return u->pid == scm->pid &&
1662 uid_eq(u->uid, scm->creds.uid) &&
1663 gid_eq(u->gid, scm->creds.gid) &&
1664 unix_secdata_eq(scm, skb);
1665}
1666
1667static void scm_stat_add(struct sock *sk, struct sk_buff *skb)
1668{
1669 struct scm_fp_list *fp = UNIXCB(skb).fp;
1670 struct unix_sock *u = unix_sk(sk);
1671
1672 if (unlikely(fp && fp->count))
1673 atomic_add(fp->count, &u->scm_stat.nr_fds);
1674}
1675
1676static void scm_stat_del(struct sock *sk, struct sk_buff *skb)
1677{
1678 struct scm_fp_list *fp = UNIXCB(skb).fp;
1679 struct unix_sock *u = unix_sk(sk);
1680
1681 if (unlikely(fp && fp->count))
1682 atomic_sub(fp->count, &u->scm_stat.nr_fds);
1683}
1684
1685/*
1686 * Send AF_UNIX data.
1687 */
1688
1689static int unix_dgram_sendmsg(struct socket *sock, struct msghdr *msg,
1690 size_t len)
1691{
1692 struct sock *sk = sock->sk;
1693 struct net *net = sock_net(sk);
1694 struct unix_sock *u = unix_sk(sk);
1695 DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr, msg->msg_name);
1696 struct sock *other = NULL;
1697 int namelen = 0; /* fake GCC */
1698 int err;
1699 unsigned int hash;
1700 struct sk_buff *skb;
1701 long timeo;
1702 struct scm_cookie scm;
1703 int data_len = 0;
1704 int sk_locked;
1705
1706 wait_for_unix_gc();
1707 err = scm_send(sock, msg, &scm, false);
1708 if (err < 0)
1709 return err;
1710
1711 err = -EOPNOTSUPP;
1712 if (msg->msg_flags&MSG_OOB)
1713 goto out;
1714
1715 if (msg->msg_namelen) {
1716 err = unix_mkname(sunaddr, msg->msg_namelen, &hash);
1717 if (err < 0)
1718 goto out;
1719 namelen = err;
1720 } else {
1721 sunaddr = NULL;
1722 err = -ENOTCONN;
1723 other = unix_peer_get(sk);
1724 if (!other)
1725 goto out;
1726 }
1727
1728 if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr
1729 && (err = unix_autobind(sock)) != 0)
1730 goto out;
1731
1732 err = -EMSGSIZE;
1733 if (len > sk->sk_sndbuf - 32)
1734 goto out;
1735
1736 if (len > SKB_MAX_ALLOC) {
1737 data_len = min_t(size_t,
1738 len - SKB_MAX_ALLOC,
1739 MAX_SKB_FRAGS * PAGE_SIZE);
1740 data_len = PAGE_ALIGN(data_len);
1741
1742 BUILD_BUG_ON(SKB_MAX_ALLOC < PAGE_SIZE);
1743 }
1744
1745 skb = sock_alloc_send_pskb(sk, len - data_len, data_len,
1746 msg->msg_flags & MSG_DONTWAIT, &err,
1747 PAGE_ALLOC_COSTLY_ORDER);
1748 if (skb == NULL)
1749 goto out;
1750
1751 err = unix_scm_to_skb(&scm, skb, true);
1752 if (err < 0)
1753 goto out_free;
1754
1755 skb_put(skb, len - data_len);
1756 skb->data_len = data_len;
1757 skb->len = len;
1758 err = skb_copy_datagram_from_iter(skb, 0, &msg->msg_iter, len);
1759 if (err)
1760 goto out_free;
1761
1762 timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1763
1764restart:
1765 if (!other) {
1766 err = -ECONNRESET;
1767 if (sunaddr == NULL)
1768 goto out_free;
1769
1770 other = unix_find_other(net, sunaddr, namelen, sk->sk_type,
1771 hash, &err);
1772 if (other == NULL)
1773 goto out_free;
1774 }
1775
1776 if (sk_filter(other, skb) < 0) {
1777 /* Toss the packet but do not return any error to the sender */
1778 err = len;
1779 goto out_free;
1780 }
1781
1782 sk_locked = 0;
1783 unix_state_lock(other);
1784restart_locked:
1785 err = -EPERM;
1786 if (!unix_may_send(sk, other))
1787 goto out_unlock;
1788
1789 if (unlikely(sock_flag(other, SOCK_DEAD))) {
1790 /*
1791 * Check with 1003.1g - what should
1792 * datagram error
1793 */
1794 unix_state_unlock(other);
1795 sock_put(other);
1796
1797 if (!sk_locked)
1798 unix_state_lock(sk);
1799
1800 err = 0;
1801 if (unix_peer(sk) == other) {
1802 unix_peer(sk) = NULL;
1803 unix_dgram_peer_wake_disconnect_wakeup(sk, other);
1804
1805 unix_state_unlock(sk);
1806
1807 unix_dgram_disconnected(sk, other);
1808 sock_put(other);
1809 err = -ECONNREFUSED;
1810 } else {
1811 unix_state_unlock(sk);
1812 }
1813
1814 other = NULL;
1815 if (err)
1816 goto out_free;
1817 goto restart;
1818 }
1819
1820 err = -EPIPE;
1821 if (other->sk_shutdown & RCV_SHUTDOWN)
1822 goto out_unlock;
1823
1824 if (sk->sk_type != SOCK_SEQPACKET) {
1825 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1826 if (err)
1827 goto out_unlock;
1828 }
1829
1830 /* other == sk && unix_peer(other) != sk if
1831 * - unix_peer(sk) == NULL, destination address bound to sk
1832 * - unix_peer(sk) == sk by time of get but disconnected before lock
1833 */
1834 if (other != sk &&
1835 unlikely(unix_peer(other) != sk &&
1836 unix_recvq_full_lockless(other))) {
1837 if (timeo) {
1838 timeo = unix_wait_for_peer(other, timeo);
1839
1840 err = sock_intr_errno(timeo);
1841 if (signal_pending(current))
1842 goto out_free;
1843
1844 goto restart;
1845 }
1846
1847 if (!sk_locked) {
1848 unix_state_unlock(other);
1849 unix_state_double_lock(sk, other);
1850 }
1851
1852 if (unix_peer(sk) != other ||
1853 unix_dgram_peer_wake_me(sk, other)) {
1854 err = -EAGAIN;
1855 sk_locked = 1;
1856 goto out_unlock;
1857 }
1858
1859 if (!sk_locked) {
1860 sk_locked = 1;
1861 goto restart_locked;
1862 }
1863 }
1864
1865 if (unlikely(sk_locked))
1866 unix_state_unlock(sk);
1867
1868 if (sock_flag(other, SOCK_RCVTSTAMP))
1869 __net_timestamp(skb);
1870 maybe_add_creds(skb, sock, other);
1871 scm_stat_add(other, skb);
1872 skb_queue_tail(&other->sk_receive_queue, skb);
1873 unix_state_unlock(other);
1874 other->sk_data_ready(other);
1875 sock_put(other);
1876 scm_destroy(&scm);
1877 return len;
1878
1879out_unlock:
1880 if (sk_locked)
1881 unix_state_unlock(sk);
1882 unix_state_unlock(other);
1883out_free:
1884 kfree_skb(skb);
1885out:
1886 if (other)
1887 sock_put(other);
1888 scm_destroy(&scm);
1889 return err;
1890}
1891
1892/* We use paged skbs for stream sockets, and limit occupancy to 32768
1893 * bytes, and a minimum of a full page.
1894 */
1895#define UNIX_SKB_FRAGS_SZ (PAGE_SIZE << get_order(32768))
1896
1897static int unix_stream_sendmsg(struct socket *sock, struct msghdr *msg,
1898 size_t len)
1899{
1900 struct sock *sk = sock->sk;
1901 struct sock *other = NULL;
1902 int err, size;
1903 struct sk_buff *skb;
1904 int sent = 0;
1905 struct scm_cookie scm;
1906 bool fds_sent = false;
1907 int data_len;
1908
1909 wait_for_unix_gc();
1910 err = scm_send(sock, msg, &scm, false);
1911 if (err < 0)
1912 return err;
1913
1914 err = -EOPNOTSUPP;
1915 if (msg->msg_flags&MSG_OOB)
1916 goto out_err;
1917
1918 if (msg->msg_namelen) {
1919 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
1920 goto out_err;
1921 } else {
1922 err = -ENOTCONN;
1923 other = unix_peer(sk);
1924 if (!other)
1925 goto out_err;
1926 }
1927
1928 if (sk->sk_shutdown & SEND_SHUTDOWN)
1929 goto pipe_err;
1930
1931 while (sent < len) {
1932 size = len - sent;
1933
1934 /* Keep two messages in the pipe so it schedules better */
1935 size = min_t(int, size, (sk->sk_sndbuf >> 1) - 64);
1936
1937 /* allow fallback to order-0 allocations */
1938 size = min_t(int, size, SKB_MAX_HEAD(0) + UNIX_SKB_FRAGS_SZ);
1939
1940 data_len = max_t(int, 0, size - SKB_MAX_HEAD(0));
1941
1942 data_len = min_t(size_t, size, PAGE_ALIGN(data_len));
1943
1944 skb = sock_alloc_send_pskb(sk, size - data_len, data_len,
1945 msg->msg_flags & MSG_DONTWAIT, &err,
1946 get_order(UNIX_SKB_FRAGS_SZ));
1947 if (!skb)
1948 goto out_err;
1949
1950 /* Only send the fds in the first buffer */
1951 err = unix_scm_to_skb(&scm, skb, !fds_sent);
1952 if (err < 0) {
1953 kfree_skb(skb);
1954 goto out_err;
1955 }
1956 fds_sent = true;
1957
1958 skb_put(skb, size - data_len);
1959 skb->data_len = data_len;
1960 skb->len = size;
1961 err = skb_copy_datagram_from_iter(skb, 0, &msg->msg_iter, size);
1962 if (err) {
1963 kfree_skb(skb);
1964 goto out_err;
1965 }
1966
1967 unix_state_lock(other);
1968
1969 if (sock_flag(other, SOCK_DEAD) ||
1970 (other->sk_shutdown & RCV_SHUTDOWN))
1971 goto pipe_err_free;
1972
1973 maybe_add_creds(skb, sock, other);
1974 scm_stat_add(other, skb);
1975 skb_queue_tail(&other->sk_receive_queue, skb);
1976 unix_state_unlock(other);
1977 other->sk_data_ready(other);
1978 sent += size;
1979 }
1980
1981 scm_destroy(&scm);
1982
1983 return sent;
1984
1985pipe_err_free:
1986 unix_state_unlock(other);
1987 kfree_skb(skb);
1988pipe_err:
1989 if (sent == 0 && !(msg->msg_flags&MSG_NOSIGNAL))
1990 send_sig(SIGPIPE, current, 0);
1991 err = -EPIPE;
1992out_err:
1993 scm_destroy(&scm);
1994 return sent ? : err;
1995}
1996
1997static ssize_t unix_stream_sendpage(struct socket *socket, struct page *page,
1998 int offset, size_t size, int flags)
1999{
2000 int err;
2001 bool send_sigpipe = false;
2002 bool init_scm = true;
2003 struct scm_cookie scm;
2004 struct sock *other, *sk = socket->sk;
2005 struct sk_buff *skb, *newskb = NULL, *tail = NULL;
2006
2007 if (flags & MSG_OOB)
2008 return -EOPNOTSUPP;
2009
2010 other = unix_peer(sk);
2011 if (!other || sk->sk_state != TCP_ESTABLISHED)
2012 return -ENOTCONN;
2013
2014 if (false) {
2015alloc_skb:
2016 unix_state_unlock(other);
2017 mutex_unlock(&unix_sk(other)->iolock);
2018 newskb = sock_alloc_send_pskb(sk, 0, 0, flags & MSG_DONTWAIT,
2019 &err, 0);
2020 if (!newskb)
2021 goto err;
2022 }
2023
2024 /* we must acquire iolock as we modify already present
2025 * skbs in the sk_receive_queue and mess with skb->len
2026 */
2027 err = mutex_lock_interruptible(&unix_sk(other)->iolock);
2028 if (err) {
2029 err = flags & MSG_DONTWAIT ? -EAGAIN : -ERESTARTSYS;
2030 goto err;
2031 }
2032
2033 if (sk->sk_shutdown & SEND_SHUTDOWN) {
2034 err = -EPIPE;
2035 send_sigpipe = true;
2036 goto err_unlock;
2037 }
2038
2039 unix_state_lock(other);
2040
2041 if (sock_flag(other, SOCK_DEAD) ||
2042 other->sk_shutdown & RCV_SHUTDOWN) {
2043 err = -EPIPE;
2044 send_sigpipe = true;
2045 goto err_state_unlock;
2046 }
2047
2048 if (init_scm) {
2049 err = maybe_init_creds(&scm, socket, other);
2050 if (err)
2051 goto err_state_unlock;
2052 init_scm = false;
2053 }
2054
2055 skb = skb_peek_tail(&other->sk_receive_queue);
2056 if (tail && tail == skb) {
2057 skb = newskb;
2058 } else if (!skb || !unix_skb_scm_eq(skb, &scm)) {
2059 if (newskb) {
2060 skb = newskb;
2061 } else {
2062 tail = skb;
2063 goto alloc_skb;
2064 }
2065 } else if (newskb) {
2066 /* this is fast path, we don't necessarily need to
2067 * call to kfree_skb even though with newskb == NULL
2068 * this - does no harm
2069 */
2070 consume_skb(newskb);
2071 newskb = NULL;
2072 }
2073
2074 if (skb_append_pagefrags(skb, page, offset, size)) {
2075 tail = skb;
2076 goto alloc_skb;
2077 }
2078
2079 skb->len += size;
2080 skb->data_len += size;
2081 skb->truesize += size;
2082 refcount_add(size, &sk->sk_wmem_alloc);
2083
2084 if (newskb) {
2085 err = unix_scm_to_skb(&scm, skb, false);
2086 if (err)
2087 goto err_state_unlock;
2088 spin_lock(&other->sk_receive_queue.lock);
2089 __skb_queue_tail(&other->sk_receive_queue, newskb);
2090 spin_unlock(&other->sk_receive_queue.lock);
2091 }
2092
2093 unix_state_unlock(other);
2094 mutex_unlock(&unix_sk(other)->iolock);
2095
2096 other->sk_data_ready(other);
2097 scm_destroy(&scm);
2098 return size;
2099
2100err_state_unlock:
2101 unix_state_unlock(other);
2102err_unlock:
2103 mutex_unlock(&unix_sk(other)->iolock);
2104err:
2105 kfree_skb(newskb);
2106 if (send_sigpipe && !(flags & MSG_NOSIGNAL))
2107 send_sig(SIGPIPE, current, 0);
2108 if (!init_scm)
2109 scm_destroy(&scm);
2110 return err;
2111}
2112
2113static int unix_seqpacket_sendmsg(struct socket *sock, struct msghdr *msg,
2114 size_t len)
2115{
2116 int err;
2117 struct sock *sk = sock->sk;
2118
2119 err = sock_error(sk);
2120 if (err)
2121 return err;
2122
2123 if (sk->sk_state != TCP_ESTABLISHED)
2124 return -ENOTCONN;
2125
2126 if (msg->msg_namelen)
2127 msg->msg_namelen = 0;
2128
2129 return unix_dgram_sendmsg(sock, msg, len);
2130}
2131
2132static int unix_seqpacket_recvmsg(struct socket *sock, struct msghdr *msg,
2133 size_t size, int flags)
2134{
2135 struct sock *sk = sock->sk;
2136
2137 if (sk->sk_state != TCP_ESTABLISHED)
2138 return -ENOTCONN;
2139
2140 return unix_dgram_recvmsg(sock, msg, size, flags);
2141}
2142
2143static void unix_copy_addr(struct msghdr *msg, struct sock *sk)
2144{
2145 struct unix_address *addr = smp_load_acquire(&unix_sk(sk)->addr);
2146
2147 if (addr) {
2148 msg->msg_namelen = addr->len;
2149 memcpy(msg->msg_name, addr->name, addr->len);
2150 }
2151}
2152
2153static int unix_dgram_recvmsg(struct socket *sock, struct msghdr *msg,
2154 size_t size, int flags)
2155{
2156 struct scm_cookie scm;
2157 struct sock *sk = sock->sk;
2158 struct unix_sock *u = unix_sk(sk);
2159 struct sk_buff *skb, *last;
2160 long timeo;
2161 int skip;
2162 int err;
2163
2164 err = -EOPNOTSUPP;
2165 if (flags&MSG_OOB)
2166 goto out;
2167
2168 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
2169
2170 do {
2171 mutex_lock(&u->iolock);
2172
2173 skip = sk_peek_offset(sk, flags);
2174 skb = __skb_try_recv_datagram(sk, &sk->sk_receive_queue, flags,
2175 &skip, &err, &last);
2176 if (skb) {
2177 if (!(flags & MSG_PEEK))
2178 scm_stat_del(sk, skb);
2179 break;
2180 }
2181
2182 mutex_unlock(&u->iolock);
2183
2184 if (err != -EAGAIN)
2185 break;
2186 } while (timeo &&
2187 !__skb_wait_for_more_packets(sk, &sk->sk_receive_queue,
2188 &err, &timeo, last));
2189
2190 if (!skb) { /* implies iolock unlocked */
2191 unix_state_lock(sk);
2192 /* Signal EOF on disconnected non-blocking SEQPACKET socket. */
2193 if (sk->sk_type == SOCK_SEQPACKET && err == -EAGAIN &&
2194 (sk->sk_shutdown & RCV_SHUTDOWN))
2195 err = 0;
2196 unix_state_unlock(sk);
2197 goto out;
2198 }
2199
2200 if (wq_has_sleeper(&u->peer_wait))
2201 wake_up_interruptible_sync_poll(&u->peer_wait,
2202 EPOLLOUT | EPOLLWRNORM |
2203 EPOLLWRBAND);
2204
2205 if (msg->msg_name)
2206 unix_copy_addr(msg, skb->sk);
2207
2208 if (size > skb->len - skip)
2209 size = skb->len - skip;
2210 else if (size < skb->len - skip)
2211 msg->msg_flags |= MSG_TRUNC;
2212
2213 err = skb_copy_datagram_msg(skb, skip, msg, size);
2214 if (err)
2215 goto out_free;
2216
2217 if (sock_flag(sk, SOCK_RCVTSTAMP))
2218 __sock_recv_timestamp(msg, sk, skb);
2219
2220 memset(&scm, 0, sizeof(scm));
2221
2222 scm_set_cred(&scm, UNIXCB(skb).pid, UNIXCB(skb).uid, UNIXCB(skb).gid);
2223 unix_set_secdata(&scm, skb);
2224
2225 if (!(flags & MSG_PEEK)) {
2226 if (UNIXCB(skb).fp)
2227 unix_detach_fds(&scm, skb);
2228
2229 sk_peek_offset_bwd(sk, skb->len);
2230 } else {
2231 /* It is questionable: on PEEK we could:
2232 - do not return fds - good, but too simple 8)
2233 - return fds, and do not return them on read (old strategy,
2234 apparently wrong)
2235 - clone fds (I chose it for now, it is the most universal
2236 solution)
2237
2238 POSIX 1003.1g does not actually define this clearly
2239 at all. POSIX 1003.1g doesn't define a lot of things
2240 clearly however!
2241
2242 */
2243
2244 sk_peek_offset_fwd(sk, size);
2245
2246 if (UNIXCB(skb).fp)
2247 unix_peek_fds(&scm, skb);
2248 }
2249 err = (flags & MSG_TRUNC) ? skb->len - skip : size;
2250
2251 scm_recv(sock, msg, &scm, flags);
2252
2253out_free:
2254 skb_free_datagram(sk, skb);
2255 mutex_unlock(&u->iolock);
2256out:
2257 return err;
2258}
2259
2260/*
2261 * Sleep until more data has arrived. But check for races..
2262 */
2263static long unix_stream_data_wait(struct sock *sk, long timeo,
2264 struct sk_buff *last, unsigned int last_len,
2265 bool freezable)
2266{
2267 struct sk_buff *tail;
2268 DEFINE_WAIT(wait);
2269
2270 unix_state_lock(sk);
2271
2272 for (;;) {
2273 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2274
2275 tail = skb_peek_tail(&sk->sk_receive_queue);
2276 if (tail != last ||
2277 (tail && tail->len != last_len) ||
2278 sk->sk_err ||
2279 (sk->sk_shutdown & RCV_SHUTDOWN) ||
2280 signal_pending(current) ||
2281 !timeo)
2282 break;
2283
2284 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2285 unix_state_unlock(sk);
2286 if (freezable)
2287 timeo = freezable_schedule_timeout(timeo);
2288 else
2289 timeo = schedule_timeout(timeo);
2290 unix_state_lock(sk);
2291
2292 if (sock_flag(sk, SOCK_DEAD))
2293 break;
2294
2295 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2296 }
2297
2298 finish_wait(sk_sleep(sk), &wait);
2299 unix_state_unlock(sk);
2300 return timeo;
2301}
2302
2303static unsigned int unix_skb_len(const struct sk_buff *skb)
2304{
2305 return skb->len - UNIXCB(skb).consumed;
2306}
2307
2308struct unix_stream_read_state {
2309 int (*recv_actor)(struct sk_buff *, int, int,
2310 struct unix_stream_read_state *);
2311 struct socket *socket;
2312 struct msghdr *msg;
2313 struct pipe_inode_info *pipe;
2314 size_t size;
2315 int flags;
2316 unsigned int splice_flags;
2317};
2318
2319static int unix_stream_read_generic(struct unix_stream_read_state *state,
2320 bool freezable)
2321{
2322 struct scm_cookie scm;
2323 struct socket *sock = state->socket;
2324 struct sock *sk = sock->sk;
2325 struct unix_sock *u = unix_sk(sk);
2326 int copied = 0;
2327 int flags = state->flags;
2328 int noblock = flags & MSG_DONTWAIT;
2329 bool check_creds = false;
2330 int target;
2331 int err = 0;
2332 long timeo;
2333 int skip;
2334 size_t size = state->size;
2335 unsigned int last_len;
2336
2337 if (unlikely(sk->sk_state != TCP_ESTABLISHED)) {
2338 err = -EINVAL;
2339 goto out;
2340 }
2341
2342 if (unlikely(flags & MSG_OOB)) {
2343 err = -EOPNOTSUPP;
2344 goto out;
2345 }
2346
2347 target = sock_rcvlowat(sk, flags & MSG_WAITALL, size);
2348 timeo = sock_rcvtimeo(sk, noblock);
2349
2350 memset(&scm, 0, sizeof(scm));
2351
2352 /* Lock the socket to prevent queue disordering
2353 * while sleeps in memcpy_tomsg
2354 */
2355 mutex_lock(&u->iolock);
2356
2357 skip = max(sk_peek_offset(sk, flags), 0);
2358
2359 do {
2360 int chunk;
2361 bool drop_skb;
2362 struct sk_buff *skb, *last;
2363
2364redo:
2365 unix_state_lock(sk);
2366 if (sock_flag(sk, SOCK_DEAD)) {
2367 err = -ECONNRESET;
2368 goto unlock;
2369 }
2370 last = skb = skb_peek(&sk->sk_receive_queue);
2371 last_len = last ? last->len : 0;
2372again:
2373 if (skb == NULL) {
2374 if (copied >= target)
2375 goto unlock;
2376
2377 /*
2378 * POSIX 1003.1g mandates this order.
2379 */
2380
2381 err = sock_error(sk);
2382 if (err)
2383 goto unlock;
2384 if (sk->sk_shutdown & RCV_SHUTDOWN)
2385 goto unlock;
2386
2387 unix_state_unlock(sk);
2388 if (!timeo) {
2389 err = -EAGAIN;
2390 break;
2391 }
2392
2393 mutex_unlock(&u->iolock);
2394
2395 timeo = unix_stream_data_wait(sk, timeo, last,
2396 last_len, freezable);
2397
2398 if (signal_pending(current)) {
2399 err = sock_intr_errno(timeo);
2400 scm_destroy(&scm);
2401 goto out;
2402 }
2403
2404 mutex_lock(&u->iolock);
2405 goto redo;
2406unlock:
2407 unix_state_unlock(sk);
2408 break;
2409 }
2410
2411 while (skip >= unix_skb_len(skb)) {
2412 skip -= unix_skb_len(skb);
2413 last = skb;
2414 last_len = skb->len;
2415 skb = skb_peek_next(skb, &sk->sk_receive_queue);
2416 if (!skb)
2417 goto again;
2418 }
2419
2420 unix_state_unlock(sk);
2421
2422 if (check_creds) {
2423 /* Never glue messages from different writers */
2424 if (!unix_skb_scm_eq(skb, &scm))
2425 break;
2426 } else if (test_bit(SOCK_PASSCRED, &sock->flags)) {
2427 /* Copy credentials */
2428 scm_set_cred(&scm, UNIXCB(skb).pid, UNIXCB(skb).uid, UNIXCB(skb).gid);
2429 unix_set_secdata(&scm, skb);
2430 check_creds = true;
2431 }
2432
2433 /* Copy address just once */
2434 if (state->msg && state->msg->msg_name) {
2435 DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr,
2436 state->msg->msg_name);
2437 unix_copy_addr(state->msg, skb->sk);
2438 sunaddr = NULL;
2439 }
2440
2441 chunk = min_t(unsigned int, unix_skb_len(skb) - skip, size);
2442 skb_get(skb);
2443 chunk = state->recv_actor(skb, skip, chunk, state);
2444 drop_skb = !unix_skb_len(skb);
2445 /* skb is only safe to use if !drop_skb */
2446 consume_skb(skb);
2447 if (chunk < 0) {
2448 if (copied == 0)
2449 copied = -EFAULT;
2450 break;
2451 }
2452 copied += chunk;
2453 size -= chunk;
2454
2455 if (drop_skb) {
2456 /* the skb was touched by a concurrent reader;
2457 * we should not expect anything from this skb
2458 * anymore and assume it invalid - we can be
2459 * sure it was dropped from the socket queue
2460 *
2461 * let's report a short read
2462 */
2463 err = 0;
2464 break;
2465 }
2466
2467 /* Mark read part of skb as used */
2468 if (!(flags & MSG_PEEK)) {
2469 UNIXCB(skb).consumed += chunk;
2470
2471 sk_peek_offset_bwd(sk, chunk);
2472
2473 if (UNIXCB(skb).fp) {
2474 scm_stat_del(sk, skb);
2475 unix_detach_fds(&scm, skb);
2476 }
2477
2478 if (unix_skb_len(skb))
2479 break;
2480
2481 skb_unlink(skb, &sk->sk_receive_queue);
2482 consume_skb(skb);
2483
2484 if (scm.fp)
2485 break;
2486 } else {
2487 /* It is questionable, see note in unix_dgram_recvmsg.
2488 */
2489 if (UNIXCB(skb).fp)
2490 unix_peek_fds(&scm, skb);
2491
2492 sk_peek_offset_fwd(sk, chunk);
2493
2494 if (UNIXCB(skb).fp)
2495 break;
2496
2497 skip = 0;
2498 last = skb;
2499 last_len = skb->len;
2500 unix_state_lock(sk);
2501 skb = skb_peek_next(skb, &sk->sk_receive_queue);
2502 if (skb)
2503 goto again;
2504 unix_state_unlock(sk);
2505 break;
2506 }
2507 } while (size);
2508
2509 mutex_unlock(&u->iolock);
2510 if (state->msg)
2511 scm_recv(sock, state->msg, &scm, flags);
2512 else
2513 scm_destroy(&scm);
2514out:
2515 return copied ? : err;
2516}
2517
2518static int unix_stream_read_actor(struct sk_buff *skb,
2519 int skip, int chunk,
2520 struct unix_stream_read_state *state)
2521{
2522 int ret;
2523
2524 ret = skb_copy_datagram_msg(skb, UNIXCB(skb).consumed + skip,
2525 state->msg, chunk);
2526 return ret ?: chunk;
2527}
2528
2529static int unix_stream_recvmsg(struct socket *sock, struct msghdr *msg,
2530 size_t size, int flags)
2531{
2532 struct unix_stream_read_state state = {
2533 .recv_actor = unix_stream_read_actor,
2534 .socket = sock,
2535 .msg = msg,
2536 .size = size,
2537 .flags = flags
2538 };
2539
2540 return unix_stream_read_generic(&state, true);
2541}
2542
2543static int unix_stream_splice_actor(struct sk_buff *skb,
2544 int skip, int chunk,
2545 struct unix_stream_read_state *state)
2546{
2547 return skb_splice_bits(skb, state->socket->sk,
2548 UNIXCB(skb).consumed + skip,
2549 state->pipe, chunk, state->splice_flags);
2550}
2551
2552static ssize_t unix_stream_splice_read(struct socket *sock, loff_t *ppos,
2553 struct pipe_inode_info *pipe,
2554 size_t size, unsigned int flags)
2555{
2556 struct unix_stream_read_state state = {
2557 .recv_actor = unix_stream_splice_actor,
2558 .socket = sock,
2559 .pipe = pipe,
2560 .size = size,
2561 .splice_flags = flags,
2562 };
2563
2564 if (unlikely(*ppos))
2565 return -ESPIPE;
2566
2567 if (sock->file->f_flags & O_NONBLOCK ||
2568 flags & SPLICE_F_NONBLOCK)
2569 state.flags = MSG_DONTWAIT;
2570
2571 return unix_stream_read_generic(&state, false);
2572}
2573
2574static int unix_shutdown(struct socket *sock, int mode)
2575{
2576 struct sock *sk = sock->sk;
2577 struct sock *other;
2578
2579 if (mode < SHUT_RD || mode > SHUT_RDWR)
2580 return -EINVAL;
2581 /* This maps:
2582 * SHUT_RD (0) -> RCV_SHUTDOWN (1)
2583 * SHUT_WR (1) -> SEND_SHUTDOWN (2)
2584 * SHUT_RDWR (2) -> SHUTDOWN_MASK (3)
2585 */
2586 ++mode;
2587
2588 unix_state_lock(sk);
2589 sk->sk_shutdown |= mode;
2590 other = unix_peer(sk);
2591 if (other)
2592 sock_hold(other);
2593 unix_state_unlock(sk);
2594 sk->sk_state_change(sk);
2595
2596 if (other &&
2597 (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET)) {
2598
2599 int peer_mode = 0;
2600
2601 if (mode&RCV_SHUTDOWN)
2602 peer_mode |= SEND_SHUTDOWN;
2603 if (mode&SEND_SHUTDOWN)
2604 peer_mode |= RCV_SHUTDOWN;
2605 unix_state_lock(other);
2606 other->sk_shutdown |= peer_mode;
2607 unix_state_unlock(other);
2608 other->sk_state_change(other);
2609 if (peer_mode == SHUTDOWN_MASK)
2610 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_HUP);
2611 else if (peer_mode & RCV_SHUTDOWN)
2612 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_IN);
2613 }
2614 if (other)
2615 sock_put(other);
2616
2617 return 0;
2618}
2619
2620long unix_inq_len(struct sock *sk)
2621{
2622 struct sk_buff *skb;
2623 long amount = 0;
2624
2625 if (sk->sk_state == TCP_LISTEN)
2626 return -EINVAL;
2627
2628 spin_lock(&sk->sk_receive_queue.lock);
2629 if (sk->sk_type == SOCK_STREAM ||
2630 sk->sk_type == SOCK_SEQPACKET) {
2631 skb_queue_walk(&sk->sk_receive_queue, skb)
2632 amount += unix_skb_len(skb);
2633 } else {
2634 skb = skb_peek(&sk->sk_receive_queue);
2635 if (skb)
2636 amount = skb->len;
2637 }
2638 spin_unlock(&sk->sk_receive_queue.lock);
2639
2640 return amount;
2641}
2642EXPORT_SYMBOL_GPL(unix_inq_len);
2643
2644long unix_outq_len(struct sock *sk)
2645{
2646 return sk_wmem_alloc_get(sk);
2647}
2648EXPORT_SYMBOL_GPL(unix_outq_len);
2649
2650static int unix_open_file(struct sock *sk)
2651{
2652 struct path path;
2653 struct file *f;
2654 int fd;
2655
2656 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
2657 return -EPERM;
2658
2659 if (!smp_load_acquire(&unix_sk(sk)->addr))
2660 return -ENOENT;
2661
2662 path = unix_sk(sk)->path;
2663 if (!path.dentry)
2664 return -ENOENT;
2665
2666 path_get(&path);
2667
2668 fd = get_unused_fd_flags(O_CLOEXEC);
2669 if (fd < 0)
2670 goto out;
2671
2672 f = dentry_open(&path, O_PATH, current_cred());
2673 if (IS_ERR(f)) {
2674 put_unused_fd(fd);
2675 fd = PTR_ERR(f);
2676 goto out;
2677 }
2678
2679 fd_install(fd, f);
2680out:
2681 path_put(&path);
2682
2683 return fd;
2684}
2685
2686static int unix_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2687{
2688 struct sock *sk = sock->sk;
2689 long amount = 0;
2690 int err;
2691
2692 switch (cmd) {
2693 case SIOCOUTQ:
2694 amount = unix_outq_len(sk);
2695 err = put_user(amount, (int __user *)arg);
2696 break;
2697 case SIOCINQ:
2698 amount = unix_inq_len(sk);
2699 if (amount < 0)
2700 err = amount;
2701 else
2702 err = put_user(amount, (int __user *)arg);
2703 break;
2704 case SIOCUNIXFILE:
2705 err = unix_open_file(sk);
2706 break;
2707 default:
2708 err = -ENOIOCTLCMD;
2709 break;
2710 }
2711 return err;
2712}
2713
2714#ifdef CONFIG_COMPAT
2715static int unix_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2716{
2717 return unix_ioctl(sock, cmd, (unsigned long)compat_ptr(arg));
2718}
2719#endif
2720
2721static __poll_t unix_poll(struct file *file, struct socket *sock, poll_table *wait)
2722{
2723 struct sock *sk = sock->sk;
2724 __poll_t mask;
2725
2726 sock_poll_wait(file, sock, wait);
2727 mask = 0;
2728
2729 /* exceptional events? */
2730 if (sk->sk_err)
2731 mask |= EPOLLERR;
2732 if (sk->sk_shutdown == SHUTDOWN_MASK)
2733 mask |= EPOLLHUP;
2734 if (sk->sk_shutdown & RCV_SHUTDOWN)
2735 mask |= EPOLLRDHUP | EPOLLIN | EPOLLRDNORM;
2736
2737 /* readable? */
2738 if (!skb_queue_empty_lockless(&sk->sk_receive_queue))
2739 mask |= EPOLLIN | EPOLLRDNORM;
2740
2741 /* Connection-based need to check for termination and startup */
2742 if ((sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) &&
2743 sk->sk_state == TCP_CLOSE)
2744 mask |= EPOLLHUP;
2745
2746 /*
2747 * we set writable also when the other side has shut down the
2748 * connection. This prevents stuck sockets.
2749 */
2750 if (unix_writable(sk))
2751 mask |= EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND;
2752
2753 return mask;
2754}
2755
2756static __poll_t unix_dgram_poll(struct file *file, struct socket *sock,
2757 poll_table *wait)
2758{
2759 struct sock *sk = sock->sk, *other;
2760 unsigned int writable;
2761 __poll_t mask;
2762
2763 sock_poll_wait(file, sock, wait);
2764 mask = 0;
2765
2766 /* exceptional events? */
2767 if (sk->sk_err || !skb_queue_empty_lockless(&sk->sk_error_queue))
2768 mask |= EPOLLERR |
2769 (sock_flag(sk, SOCK_SELECT_ERR_QUEUE) ? EPOLLPRI : 0);
2770
2771 if (sk->sk_shutdown & RCV_SHUTDOWN)
2772 mask |= EPOLLRDHUP | EPOLLIN | EPOLLRDNORM;
2773 if (sk->sk_shutdown == SHUTDOWN_MASK)
2774 mask |= EPOLLHUP;
2775
2776 /* readable? */
2777 if (!skb_queue_empty_lockless(&sk->sk_receive_queue))
2778 mask |= EPOLLIN | EPOLLRDNORM;
2779
2780 /* Connection-based need to check for termination and startup */
2781 if (sk->sk_type == SOCK_SEQPACKET) {
2782 if (sk->sk_state == TCP_CLOSE)
2783 mask |= EPOLLHUP;
2784 /* connection hasn't started yet? */
2785 if (sk->sk_state == TCP_SYN_SENT)
2786 return mask;
2787 }
2788
2789 /* No write status requested, avoid expensive OUT tests. */
2790 if (!(poll_requested_events(wait) & (EPOLLWRBAND|EPOLLWRNORM|EPOLLOUT)))
2791 return mask;
2792
2793 writable = unix_writable(sk);
2794 if (writable) {
2795 unix_state_lock(sk);
2796
2797 other = unix_peer(sk);
2798 if (other && unix_peer(other) != sk &&
2799 unix_recvq_full_lockless(other) &&
2800 unix_dgram_peer_wake_me(sk, other))
2801 writable = 0;
2802
2803 unix_state_unlock(sk);
2804 }
2805
2806 if (writable)
2807 mask |= EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND;
2808 else
2809 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2810
2811 return mask;
2812}
2813
2814#ifdef CONFIG_PROC_FS
2815
2816#define BUCKET_SPACE (BITS_PER_LONG - (UNIX_HASH_BITS + 1) - 1)
2817
2818#define get_bucket(x) ((x) >> BUCKET_SPACE)
2819#define get_offset(x) ((x) & ((1L << BUCKET_SPACE) - 1))
2820#define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
2821
2822static struct sock *unix_from_bucket(struct seq_file *seq, loff_t *pos)
2823{
2824 unsigned long offset = get_offset(*pos);
2825 unsigned long bucket = get_bucket(*pos);
2826 struct sock *sk;
2827 unsigned long count = 0;
2828
2829 for (sk = sk_head(&unix_socket_table[bucket]); sk; sk = sk_next(sk)) {
2830 if (sock_net(sk) != seq_file_net(seq))
2831 continue;
2832 if (++count == offset)
2833 break;
2834 }
2835
2836 return sk;
2837}
2838
2839static struct sock *unix_next_socket(struct seq_file *seq,
2840 struct sock *sk,
2841 loff_t *pos)
2842{
2843 unsigned long bucket;
2844
2845 while (sk > (struct sock *)SEQ_START_TOKEN) {
2846 sk = sk_next(sk);
2847 if (!sk)
2848 goto next_bucket;
2849 if (sock_net(sk) == seq_file_net(seq))
2850 return sk;
2851 }
2852
2853 do {
2854 sk = unix_from_bucket(seq, pos);
2855 if (sk)
2856 return sk;
2857
2858next_bucket:
2859 bucket = get_bucket(*pos) + 1;
2860 *pos = set_bucket_offset(bucket, 1);
2861 } while (bucket < ARRAY_SIZE(unix_socket_table));
2862
2863 return NULL;
2864}
2865
2866static void *unix_seq_start(struct seq_file *seq, loff_t *pos)
2867 __acquires(unix_table_lock)
2868{
2869 spin_lock(&unix_table_lock);
2870
2871 if (!*pos)
2872 return SEQ_START_TOKEN;
2873
2874 if (get_bucket(*pos) >= ARRAY_SIZE(unix_socket_table))
2875 return NULL;
2876
2877 return unix_next_socket(seq, NULL, pos);
2878}
2879
2880static void *unix_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2881{
2882 ++*pos;
2883 return unix_next_socket(seq, v, pos);
2884}
2885
2886static void unix_seq_stop(struct seq_file *seq, void *v)
2887 __releases(unix_table_lock)
2888{
2889 spin_unlock(&unix_table_lock);
2890}
2891
2892static int unix_seq_show(struct seq_file *seq, void *v)
2893{
2894
2895 if (v == SEQ_START_TOKEN)
2896 seq_puts(seq, "Num RefCount Protocol Flags Type St "
2897 "Inode Path\n");
2898 else {
2899 struct sock *s = v;
2900 struct unix_sock *u = unix_sk(s);
2901 unix_state_lock(s);
2902
2903 seq_printf(seq, "%pK: %08X %08X %08X %04X %02X %5lu",
2904 s,
2905 refcount_read(&s->sk_refcnt),
2906 0,
2907 s->sk_state == TCP_LISTEN ? __SO_ACCEPTCON : 0,
2908 s->sk_type,
2909 s->sk_socket ?
2910 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTED : SS_UNCONNECTED) :
2911 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTING : SS_DISCONNECTING),
2912 sock_i_ino(s));
2913
2914 if (u->addr) { // under unix_table_lock here
2915 int i, len;
2916 seq_putc(seq, ' ');
2917
2918 i = 0;
2919 len = u->addr->len - sizeof(short);
2920 if (!UNIX_ABSTRACT(s))
2921 len--;
2922 else {
2923 seq_putc(seq, '@');
2924 i++;
2925 }
2926 for ( ; i < len; i++)
2927 seq_putc(seq, u->addr->name->sun_path[i] ?:
2928 '@');
2929 }
2930 unix_state_unlock(s);
2931 seq_putc(seq, '\n');
2932 }
2933
2934 return 0;
2935}
2936
2937static const struct seq_operations unix_seq_ops = {
2938 .start = unix_seq_start,
2939 .next = unix_seq_next,
2940 .stop = unix_seq_stop,
2941 .show = unix_seq_show,
2942};
2943#endif
2944
2945static const struct net_proto_family unix_family_ops = {
2946 .family = PF_UNIX,
2947 .create = unix_create,
2948 .owner = THIS_MODULE,
2949};
2950
2951
2952static int __net_init unix_net_init(struct net *net)
2953{
2954 int error = -ENOMEM;
2955
2956 net->unx.sysctl_max_dgram_qlen = 10;
2957 if (unix_sysctl_register(net))
2958 goto out;
2959
2960#ifdef CONFIG_PROC_FS
2961 if (!proc_create_net("unix", 0, net->proc_net, &unix_seq_ops,
2962 sizeof(struct seq_net_private))) {
2963 unix_sysctl_unregister(net);
2964 goto out;
2965 }
2966#endif
2967 error = 0;
2968out:
2969 return error;
2970}
2971
2972static void __net_exit unix_net_exit(struct net *net)
2973{
2974 unix_sysctl_unregister(net);
2975 remove_proc_entry("unix", net->proc_net);
2976}
2977
2978static struct pernet_operations unix_net_ops = {
2979 .init = unix_net_init,
2980 .exit = unix_net_exit,
2981};
2982
2983static int __init af_unix_init(void)
2984{
2985 int rc = -1;
2986
2987 BUILD_BUG_ON(sizeof(struct unix_skb_parms) > sizeof_field(struct sk_buff, cb));
2988
2989 rc = proto_register(&unix_proto, 1);
2990 if (rc != 0) {
2991 pr_crit("%s: Cannot create unix_sock SLAB cache!\n", __func__);
2992 goto out;
2993 }
2994
2995 sock_register(&unix_family_ops);
2996 register_pernet_subsys(&unix_net_ops);
2997out:
2998 return rc;
2999}
3000
3001static void __exit af_unix_exit(void)
3002{
3003 sock_unregister(PF_UNIX);
3004 proto_unregister(&unix_proto);
3005 unregister_pernet_subsys(&unix_net_ops);
3006}
3007
3008/* Earlier than device_initcall() so that other drivers invoking
3009 request_module() don't end up in a loop when modprobe tries
3010 to use a UNIX socket. But later than subsys_initcall() because
3011 we depend on stuff initialised there */
3012fs_initcall(af_unix_init);
3013module_exit(af_unix_exit);
3014
3015MODULE_LICENSE("GPL");
3016MODULE_ALIAS_NETPROTO(PF_UNIX);
1/*
2 * NET4: Implementation of BSD Unix domain sockets.
3 *
4 * Authors: Alan Cox, <alan@lxorguk.ukuu.org.uk>
5 *
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.
10 *
11 * Fixes:
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
21 * Mike Shaver's work.
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
28 * reference counting
29 * Kirk Petersen : Made this a module
30 * Christoph Rohland : Elegant non-blocking accept/connect algorithm.
31 * Lots of bug fixes.
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
43 * dgram receiver.
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+)
51 *
52 *
53 * Known differences from reference BSD that was tested:
54 *
55 * [TO FIX]
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).
60 * [NOT TO FIX]
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)
68 *
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.
73 *
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
80 * with BSD names.
81 */
82
83#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
84
85#include <linux/module.h>
86#include <linux/kernel.h>
87#include <linux/signal.h>
88#include <linux/sched.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>
95#include <linux/un.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 <asm/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>
112#include <net/scm.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
121struct hlist_head unix_socket_table[2 * UNIX_HASH_SIZE];
122EXPORT_SYMBOL_GPL(unix_socket_table);
123DEFINE_SPINLOCK(unix_table_lock);
124EXPORT_SYMBOL_GPL(unix_table_lock);
125static atomic_long_t unix_nr_socks;
126
127
128static struct hlist_head *unix_sockets_unbound(void *addr)
129{
130 unsigned long hash = (unsigned long)addr;
131
132 hash ^= hash >> 16;
133 hash ^= hash >> 8;
134 hash %= UNIX_HASH_SIZE;
135 return &unix_socket_table[UNIX_HASH_SIZE + hash];
136}
137
138#define UNIX_ABSTRACT(sk) (unix_sk(sk)->addr->hash < UNIX_HASH_SIZE)
139
140#ifdef CONFIG_SECURITY_NETWORK
141static void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb)
142{
143 memcpy(UNIXSID(skb), &scm->secid, sizeof(u32));
144}
145
146static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb)
147{
148 scm->secid = *UNIXSID(skb);
149}
150#else
151static inline void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb)
152{ }
153
154static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb)
155{ }
156#endif /* CONFIG_SECURITY_NETWORK */
157
158/*
159 * SMP locking strategy:
160 * hash table is protected with spinlock unix_table_lock
161 * each socket state is protected by separate spin lock.
162 */
163
164static inline unsigned int unix_hash_fold(__wsum n)
165{
166 unsigned int hash = (__force unsigned int)csum_fold(n);
167
168 hash ^= hash>>8;
169 return hash&(UNIX_HASH_SIZE-1);
170}
171
172#define unix_peer(sk) (unix_sk(sk)->peer)
173
174static inline int unix_our_peer(struct sock *sk, struct sock *osk)
175{
176 return unix_peer(osk) == sk;
177}
178
179static inline int unix_may_send(struct sock *sk, struct sock *osk)
180{
181 return unix_peer(osk) == NULL || unix_our_peer(sk, osk);
182}
183
184static inline int unix_recvq_full(struct sock const *sk)
185{
186 return skb_queue_len(&sk->sk_receive_queue) > sk->sk_max_ack_backlog;
187}
188
189struct sock *unix_peer_get(struct sock *s)
190{
191 struct sock *peer;
192
193 unix_state_lock(s);
194 peer = unix_peer(s);
195 if (peer)
196 sock_hold(peer);
197 unix_state_unlock(s);
198 return peer;
199}
200EXPORT_SYMBOL_GPL(unix_peer_get);
201
202static inline void unix_release_addr(struct unix_address *addr)
203{
204 if (atomic_dec_and_test(&addr->refcnt))
205 kfree(addr);
206}
207
208/*
209 * Check unix socket name:
210 * - should be not zero length.
211 * - if started by not zero, should be NULL terminated (FS object)
212 * - if started by zero, it is abstract name.
213 */
214
215static int unix_mkname(struct sockaddr_un *sunaddr, int len, unsigned int *hashp)
216{
217 if (len <= sizeof(short) || len > sizeof(*sunaddr))
218 return -EINVAL;
219 if (!sunaddr || sunaddr->sun_family != AF_UNIX)
220 return -EINVAL;
221 if (sunaddr->sun_path[0]) {
222 /*
223 * This may look like an off by one error but it is a bit more
224 * subtle. 108 is the longest valid AF_UNIX path for a binding.
225 * sun_path[108] doesn't as such exist. However in kernel space
226 * we are guaranteed that it is a valid memory location in our
227 * kernel address buffer.
228 */
229 ((char *)sunaddr)[len] = 0;
230 len = strlen(sunaddr->sun_path)+1+sizeof(short);
231 return len;
232 }
233
234 *hashp = unix_hash_fold(csum_partial(sunaddr, len, 0));
235 return len;
236}
237
238static void __unix_remove_socket(struct sock *sk)
239{
240 sk_del_node_init(sk);
241}
242
243static void __unix_insert_socket(struct hlist_head *list, struct sock *sk)
244{
245 WARN_ON(!sk_unhashed(sk));
246 sk_add_node(sk, list);
247}
248
249static inline void unix_remove_socket(struct sock *sk)
250{
251 spin_lock(&unix_table_lock);
252 __unix_remove_socket(sk);
253 spin_unlock(&unix_table_lock);
254}
255
256static inline void unix_insert_socket(struct hlist_head *list, struct sock *sk)
257{
258 spin_lock(&unix_table_lock);
259 __unix_insert_socket(list, sk);
260 spin_unlock(&unix_table_lock);
261}
262
263static struct sock *__unix_find_socket_byname(struct net *net,
264 struct sockaddr_un *sunname,
265 int len, int type, unsigned int hash)
266{
267 struct sock *s;
268
269 sk_for_each(s, &unix_socket_table[hash ^ type]) {
270 struct unix_sock *u = unix_sk(s);
271
272 if (!net_eq(sock_net(s), net))
273 continue;
274
275 if (u->addr->len == len &&
276 !memcmp(u->addr->name, sunname, len))
277 goto found;
278 }
279 s = NULL;
280found:
281 return s;
282}
283
284static inline struct sock *unix_find_socket_byname(struct net *net,
285 struct sockaddr_un *sunname,
286 int len, int type,
287 unsigned int hash)
288{
289 struct sock *s;
290
291 spin_lock(&unix_table_lock);
292 s = __unix_find_socket_byname(net, sunname, len, type, hash);
293 if (s)
294 sock_hold(s);
295 spin_unlock(&unix_table_lock);
296 return s;
297}
298
299static struct sock *unix_find_socket_byinode(struct inode *i)
300{
301 struct sock *s;
302
303 spin_lock(&unix_table_lock);
304 sk_for_each(s,
305 &unix_socket_table[i->i_ino & (UNIX_HASH_SIZE - 1)]) {
306 struct dentry *dentry = unix_sk(s)->path.dentry;
307
308 if (dentry && dentry->d_inode == i) {
309 sock_hold(s);
310 goto found;
311 }
312 }
313 s = NULL;
314found:
315 spin_unlock(&unix_table_lock);
316 return s;
317}
318
319static inline int unix_writable(struct sock *sk)
320{
321 return (atomic_read(&sk->sk_wmem_alloc) << 2) <= sk->sk_sndbuf;
322}
323
324static void unix_write_space(struct sock *sk)
325{
326 struct socket_wq *wq;
327
328 rcu_read_lock();
329 if (unix_writable(sk)) {
330 wq = rcu_dereference(sk->sk_wq);
331 if (wq_has_sleeper(wq))
332 wake_up_interruptible_sync_poll(&wq->wait,
333 POLLOUT | POLLWRNORM | POLLWRBAND);
334 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
335 }
336 rcu_read_unlock();
337}
338
339/* When dgram socket disconnects (or changes its peer), we clear its receive
340 * queue of packets arrived from previous peer. First, it allows to do
341 * flow control based only on wmem_alloc; second, sk connected to peer
342 * may receive messages only from that peer. */
343static void unix_dgram_disconnected(struct sock *sk, struct sock *other)
344{
345 if (!skb_queue_empty(&sk->sk_receive_queue)) {
346 skb_queue_purge(&sk->sk_receive_queue);
347 wake_up_interruptible_all(&unix_sk(sk)->peer_wait);
348
349 /* If one link of bidirectional dgram pipe is disconnected,
350 * we signal error. Messages are lost. Do not make this,
351 * when peer was not connected to us.
352 */
353 if (!sock_flag(other, SOCK_DEAD) && unix_peer(other) == sk) {
354 other->sk_err = ECONNRESET;
355 other->sk_error_report(other);
356 }
357 }
358}
359
360static void unix_sock_destructor(struct sock *sk)
361{
362 struct unix_sock *u = unix_sk(sk);
363
364 skb_queue_purge(&sk->sk_receive_queue);
365
366 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
367 WARN_ON(!sk_unhashed(sk));
368 WARN_ON(sk->sk_socket);
369 if (!sock_flag(sk, SOCK_DEAD)) {
370 pr_info("Attempt to release alive unix socket: %p\n", sk);
371 return;
372 }
373
374 if (u->addr)
375 unix_release_addr(u->addr);
376
377 atomic_long_dec(&unix_nr_socks);
378 local_bh_disable();
379 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
380 local_bh_enable();
381#ifdef UNIX_REFCNT_DEBUG
382 pr_debug("UNIX %p is destroyed, %ld are still alive.\n", sk,
383 atomic_long_read(&unix_nr_socks));
384#endif
385}
386
387static void unix_release_sock(struct sock *sk, int embrion)
388{
389 struct unix_sock *u = unix_sk(sk);
390 struct path path;
391 struct sock *skpair;
392 struct sk_buff *skb;
393 int state;
394
395 unix_remove_socket(sk);
396
397 /* Clear state */
398 unix_state_lock(sk);
399 sock_orphan(sk);
400 sk->sk_shutdown = SHUTDOWN_MASK;
401 path = u->path;
402 u->path.dentry = NULL;
403 u->path.mnt = NULL;
404 state = sk->sk_state;
405 sk->sk_state = TCP_CLOSE;
406 unix_state_unlock(sk);
407
408 wake_up_interruptible_all(&u->peer_wait);
409
410 skpair = unix_peer(sk);
411
412 if (skpair != NULL) {
413 if (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) {
414 unix_state_lock(skpair);
415 /* No more writes */
416 skpair->sk_shutdown = SHUTDOWN_MASK;
417 if (!skb_queue_empty(&sk->sk_receive_queue) || embrion)
418 skpair->sk_err = ECONNRESET;
419 unix_state_unlock(skpair);
420 skpair->sk_state_change(skpair);
421 sk_wake_async(skpair, SOCK_WAKE_WAITD, POLL_HUP);
422 }
423 sock_put(skpair); /* It may now die */
424 unix_peer(sk) = NULL;
425 }
426
427 /* Try to flush out this socket. Throw out buffers at least */
428
429 while ((skb = skb_dequeue(&sk->sk_receive_queue)) != NULL) {
430 if (state == TCP_LISTEN)
431 unix_release_sock(skb->sk, 1);
432 /* passed fds are erased in the kfree_skb hook */
433 kfree_skb(skb);
434 }
435
436 if (path.dentry)
437 path_put(&path);
438
439 sock_put(sk);
440
441 /* ---- Socket is dead now and most probably destroyed ---- */
442
443 /*
444 * Fixme: BSD difference: In BSD all sockets connected to us get
445 * ECONNRESET and we die on the spot. In Linux we behave
446 * like files and pipes do and wait for the last
447 * dereference.
448 *
449 * Can't we simply set sock->err?
450 *
451 * What the above comment does talk about? --ANK(980817)
452 */
453
454 if (unix_tot_inflight)
455 unix_gc(); /* Garbage collect fds */
456}
457
458static void init_peercred(struct sock *sk)
459{
460 put_pid(sk->sk_peer_pid);
461 if (sk->sk_peer_cred)
462 put_cred(sk->sk_peer_cred);
463 sk->sk_peer_pid = get_pid(task_tgid(current));
464 sk->sk_peer_cred = get_current_cred();
465}
466
467static void copy_peercred(struct sock *sk, struct sock *peersk)
468{
469 put_pid(sk->sk_peer_pid);
470 if (sk->sk_peer_cred)
471 put_cred(sk->sk_peer_cred);
472 sk->sk_peer_pid = get_pid(peersk->sk_peer_pid);
473 sk->sk_peer_cred = get_cred(peersk->sk_peer_cred);
474}
475
476static int unix_listen(struct socket *sock, int backlog)
477{
478 int err;
479 struct sock *sk = sock->sk;
480 struct unix_sock *u = unix_sk(sk);
481 struct pid *old_pid = NULL;
482
483 err = -EOPNOTSUPP;
484 if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
485 goto out; /* Only stream/seqpacket sockets accept */
486 err = -EINVAL;
487 if (!u->addr)
488 goto out; /* No listens on an unbound socket */
489 unix_state_lock(sk);
490 if (sk->sk_state != TCP_CLOSE && sk->sk_state != TCP_LISTEN)
491 goto out_unlock;
492 if (backlog > sk->sk_max_ack_backlog)
493 wake_up_interruptible_all(&u->peer_wait);
494 sk->sk_max_ack_backlog = backlog;
495 sk->sk_state = TCP_LISTEN;
496 /* set credentials so connect can copy them */
497 init_peercred(sk);
498 err = 0;
499
500out_unlock:
501 unix_state_unlock(sk);
502 put_pid(old_pid);
503out:
504 return err;
505}
506
507static int unix_release(struct socket *);
508static int unix_bind(struct socket *, struct sockaddr *, int);
509static int unix_stream_connect(struct socket *, struct sockaddr *,
510 int addr_len, int flags);
511static int unix_socketpair(struct socket *, struct socket *);
512static int unix_accept(struct socket *, struct socket *, int);
513static int unix_getname(struct socket *, struct sockaddr *, int *, int);
514static unsigned int unix_poll(struct file *, struct socket *, poll_table *);
515static unsigned int unix_dgram_poll(struct file *, struct socket *,
516 poll_table *);
517static int unix_ioctl(struct socket *, unsigned int, unsigned long);
518static int unix_shutdown(struct socket *, int);
519static int unix_stream_sendmsg(struct kiocb *, struct socket *,
520 struct msghdr *, size_t);
521static int unix_stream_recvmsg(struct kiocb *, struct socket *,
522 struct msghdr *, size_t, int);
523static int unix_dgram_sendmsg(struct kiocb *, struct socket *,
524 struct msghdr *, size_t);
525static int unix_dgram_recvmsg(struct kiocb *, struct socket *,
526 struct msghdr *, size_t, int);
527static int unix_dgram_connect(struct socket *, struct sockaddr *,
528 int, int);
529static int unix_seqpacket_sendmsg(struct kiocb *, struct socket *,
530 struct msghdr *, size_t);
531static int unix_seqpacket_recvmsg(struct kiocb *, struct socket *,
532 struct msghdr *, size_t, int);
533
534static int unix_set_peek_off(struct sock *sk, int val)
535{
536 struct unix_sock *u = unix_sk(sk);
537
538 if (mutex_lock_interruptible(&u->readlock))
539 return -EINTR;
540
541 sk->sk_peek_off = val;
542 mutex_unlock(&u->readlock);
543
544 return 0;
545}
546
547
548static const struct proto_ops unix_stream_ops = {
549 .family = PF_UNIX,
550 .owner = THIS_MODULE,
551 .release = unix_release,
552 .bind = unix_bind,
553 .connect = unix_stream_connect,
554 .socketpair = unix_socketpair,
555 .accept = unix_accept,
556 .getname = unix_getname,
557 .poll = unix_poll,
558 .ioctl = unix_ioctl,
559 .listen = unix_listen,
560 .shutdown = unix_shutdown,
561 .setsockopt = sock_no_setsockopt,
562 .getsockopt = sock_no_getsockopt,
563 .sendmsg = unix_stream_sendmsg,
564 .recvmsg = unix_stream_recvmsg,
565 .mmap = sock_no_mmap,
566 .sendpage = sock_no_sendpage,
567 .set_peek_off = unix_set_peek_off,
568};
569
570static const struct proto_ops unix_dgram_ops = {
571 .family = PF_UNIX,
572 .owner = THIS_MODULE,
573 .release = unix_release,
574 .bind = unix_bind,
575 .connect = unix_dgram_connect,
576 .socketpair = unix_socketpair,
577 .accept = sock_no_accept,
578 .getname = unix_getname,
579 .poll = unix_dgram_poll,
580 .ioctl = unix_ioctl,
581 .listen = sock_no_listen,
582 .shutdown = unix_shutdown,
583 .setsockopt = sock_no_setsockopt,
584 .getsockopt = sock_no_getsockopt,
585 .sendmsg = unix_dgram_sendmsg,
586 .recvmsg = unix_dgram_recvmsg,
587 .mmap = sock_no_mmap,
588 .sendpage = sock_no_sendpage,
589 .set_peek_off = unix_set_peek_off,
590};
591
592static const struct proto_ops unix_seqpacket_ops = {
593 .family = PF_UNIX,
594 .owner = THIS_MODULE,
595 .release = unix_release,
596 .bind = unix_bind,
597 .connect = unix_stream_connect,
598 .socketpair = unix_socketpair,
599 .accept = unix_accept,
600 .getname = unix_getname,
601 .poll = unix_dgram_poll,
602 .ioctl = unix_ioctl,
603 .listen = unix_listen,
604 .shutdown = unix_shutdown,
605 .setsockopt = sock_no_setsockopt,
606 .getsockopt = sock_no_getsockopt,
607 .sendmsg = unix_seqpacket_sendmsg,
608 .recvmsg = unix_seqpacket_recvmsg,
609 .mmap = sock_no_mmap,
610 .sendpage = sock_no_sendpage,
611 .set_peek_off = unix_set_peek_off,
612};
613
614static struct proto unix_proto = {
615 .name = "UNIX",
616 .owner = THIS_MODULE,
617 .obj_size = sizeof(struct unix_sock),
618};
619
620/*
621 * AF_UNIX sockets do not interact with hardware, hence they
622 * dont trigger interrupts - so it's safe for them to have
623 * bh-unsafe locking for their sk_receive_queue.lock. Split off
624 * this special lock-class by reinitializing the spinlock key:
625 */
626static struct lock_class_key af_unix_sk_receive_queue_lock_key;
627
628static struct sock *unix_create1(struct net *net, struct socket *sock)
629{
630 struct sock *sk = NULL;
631 struct unix_sock *u;
632
633 atomic_long_inc(&unix_nr_socks);
634 if (atomic_long_read(&unix_nr_socks) > 2 * get_max_files())
635 goto out;
636
637 sk = sk_alloc(net, PF_UNIX, GFP_KERNEL, &unix_proto);
638 if (!sk)
639 goto out;
640
641 sock_init_data(sock, sk);
642 lockdep_set_class(&sk->sk_receive_queue.lock,
643 &af_unix_sk_receive_queue_lock_key);
644
645 sk->sk_write_space = unix_write_space;
646 sk->sk_max_ack_backlog = net->unx.sysctl_max_dgram_qlen;
647 sk->sk_destruct = unix_sock_destructor;
648 u = unix_sk(sk);
649 u->path.dentry = NULL;
650 u->path.mnt = NULL;
651 spin_lock_init(&u->lock);
652 atomic_long_set(&u->inflight, 0);
653 INIT_LIST_HEAD(&u->link);
654 mutex_init(&u->readlock); /* single task reading lock */
655 init_waitqueue_head(&u->peer_wait);
656 unix_insert_socket(unix_sockets_unbound(sk), sk);
657out:
658 if (sk == NULL)
659 atomic_long_dec(&unix_nr_socks);
660 else {
661 local_bh_disable();
662 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
663 local_bh_enable();
664 }
665 return sk;
666}
667
668static int unix_create(struct net *net, struct socket *sock, int protocol,
669 int kern)
670{
671 if (protocol && protocol != PF_UNIX)
672 return -EPROTONOSUPPORT;
673
674 sock->state = SS_UNCONNECTED;
675
676 switch (sock->type) {
677 case SOCK_STREAM:
678 sock->ops = &unix_stream_ops;
679 break;
680 /*
681 * Believe it or not BSD has AF_UNIX, SOCK_RAW though
682 * nothing uses it.
683 */
684 case SOCK_RAW:
685 sock->type = SOCK_DGRAM;
686 case SOCK_DGRAM:
687 sock->ops = &unix_dgram_ops;
688 break;
689 case SOCK_SEQPACKET:
690 sock->ops = &unix_seqpacket_ops;
691 break;
692 default:
693 return -ESOCKTNOSUPPORT;
694 }
695
696 return unix_create1(net, sock) ? 0 : -ENOMEM;
697}
698
699static int unix_release(struct socket *sock)
700{
701 struct sock *sk = sock->sk;
702
703 if (!sk)
704 return 0;
705
706 unix_release_sock(sk, 0);
707 sock->sk = NULL;
708
709 return 0;
710}
711
712static int unix_autobind(struct socket *sock)
713{
714 struct sock *sk = sock->sk;
715 struct net *net = sock_net(sk);
716 struct unix_sock *u = unix_sk(sk);
717 static u32 ordernum = 1;
718 struct unix_address *addr;
719 int err;
720 unsigned int retries = 0;
721
722 err = mutex_lock_interruptible(&u->readlock);
723 if (err)
724 return err;
725
726 err = 0;
727 if (u->addr)
728 goto out;
729
730 err = -ENOMEM;
731 addr = kzalloc(sizeof(*addr) + sizeof(short) + 16, GFP_KERNEL);
732 if (!addr)
733 goto out;
734
735 addr->name->sun_family = AF_UNIX;
736 atomic_set(&addr->refcnt, 1);
737
738retry:
739 addr->len = sprintf(addr->name->sun_path+1, "%05x", ordernum) + 1 + sizeof(short);
740 addr->hash = unix_hash_fold(csum_partial(addr->name, addr->len, 0));
741
742 spin_lock(&unix_table_lock);
743 ordernum = (ordernum+1)&0xFFFFF;
744
745 if (__unix_find_socket_byname(net, addr->name, addr->len, sock->type,
746 addr->hash)) {
747 spin_unlock(&unix_table_lock);
748 /*
749 * __unix_find_socket_byname() may take long time if many names
750 * are already in use.
751 */
752 cond_resched();
753 /* Give up if all names seems to be in use. */
754 if (retries++ == 0xFFFFF) {
755 err = -ENOSPC;
756 kfree(addr);
757 goto out;
758 }
759 goto retry;
760 }
761 addr->hash ^= sk->sk_type;
762
763 __unix_remove_socket(sk);
764 u->addr = addr;
765 __unix_insert_socket(&unix_socket_table[addr->hash], sk);
766 spin_unlock(&unix_table_lock);
767 err = 0;
768
769out: mutex_unlock(&u->readlock);
770 return err;
771}
772
773static struct sock *unix_find_other(struct net *net,
774 struct sockaddr_un *sunname, int len,
775 int type, unsigned int hash, int *error)
776{
777 struct sock *u;
778 struct path path;
779 int err = 0;
780
781 if (sunname->sun_path[0]) {
782 struct inode *inode;
783 err = kern_path(sunname->sun_path, LOOKUP_FOLLOW, &path);
784 if (err)
785 goto fail;
786 inode = path.dentry->d_inode;
787 err = inode_permission(inode, MAY_WRITE);
788 if (err)
789 goto put_fail;
790
791 err = -ECONNREFUSED;
792 if (!S_ISSOCK(inode->i_mode))
793 goto put_fail;
794 u = unix_find_socket_byinode(inode);
795 if (!u)
796 goto put_fail;
797
798 if (u->sk_type == type)
799 touch_atime(&path);
800
801 path_put(&path);
802
803 err = -EPROTOTYPE;
804 if (u->sk_type != type) {
805 sock_put(u);
806 goto fail;
807 }
808 } else {
809 err = -ECONNREFUSED;
810 u = unix_find_socket_byname(net, sunname, len, type, hash);
811 if (u) {
812 struct dentry *dentry;
813 dentry = unix_sk(u)->path.dentry;
814 if (dentry)
815 touch_atime(&unix_sk(u)->path);
816 } else
817 goto fail;
818 }
819 return u;
820
821put_fail:
822 path_put(&path);
823fail:
824 *error = err;
825 return NULL;
826}
827
828static int unix_mknod(const char *sun_path, umode_t mode, struct path *res)
829{
830 struct dentry *dentry;
831 struct path path;
832 int err = 0;
833 /*
834 * Get the parent directory, calculate the hash for last
835 * component.
836 */
837 dentry = kern_path_create(AT_FDCWD, sun_path, &path, 0);
838 err = PTR_ERR(dentry);
839 if (IS_ERR(dentry))
840 return err;
841
842 /*
843 * All right, let's create it.
844 */
845 err = security_path_mknod(&path, dentry, mode, 0);
846 if (!err) {
847 err = vfs_mknod(path.dentry->d_inode, dentry, mode, 0);
848 if (!err) {
849 res->mnt = mntget(path.mnt);
850 res->dentry = dget(dentry);
851 }
852 }
853 done_path_create(&path, dentry);
854 return err;
855}
856
857static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
858{
859 struct sock *sk = sock->sk;
860 struct net *net = sock_net(sk);
861 struct unix_sock *u = unix_sk(sk);
862 struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
863 char *sun_path = sunaddr->sun_path;
864 int err;
865 unsigned int hash;
866 struct unix_address *addr;
867 struct hlist_head *list;
868
869 err = -EINVAL;
870 if (sunaddr->sun_family != AF_UNIX)
871 goto out;
872
873 if (addr_len == sizeof(short)) {
874 err = unix_autobind(sock);
875 goto out;
876 }
877
878 err = unix_mkname(sunaddr, addr_len, &hash);
879 if (err < 0)
880 goto out;
881 addr_len = err;
882
883 err = mutex_lock_interruptible(&u->readlock);
884 if (err)
885 goto out;
886
887 err = -EINVAL;
888 if (u->addr)
889 goto out_up;
890
891 err = -ENOMEM;
892 addr = kmalloc(sizeof(*addr)+addr_len, GFP_KERNEL);
893 if (!addr)
894 goto out_up;
895
896 memcpy(addr->name, sunaddr, addr_len);
897 addr->len = addr_len;
898 addr->hash = hash ^ sk->sk_type;
899 atomic_set(&addr->refcnt, 1);
900
901 if (sun_path[0]) {
902 struct path path;
903 umode_t mode = S_IFSOCK |
904 (SOCK_INODE(sock)->i_mode & ~current_umask());
905 err = unix_mknod(sun_path, mode, &path);
906 if (err) {
907 if (err == -EEXIST)
908 err = -EADDRINUSE;
909 unix_release_addr(addr);
910 goto out_up;
911 }
912 addr->hash = UNIX_HASH_SIZE;
913 hash = path.dentry->d_inode->i_ino & (UNIX_HASH_SIZE-1);
914 spin_lock(&unix_table_lock);
915 u->path = path;
916 list = &unix_socket_table[hash];
917 } else {
918 spin_lock(&unix_table_lock);
919 err = -EADDRINUSE;
920 if (__unix_find_socket_byname(net, sunaddr, addr_len,
921 sk->sk_type, hash)) {
922 unix_release_addr(addr);
923 goto out_unlock;
924 }
925
926 list = &unix_socket_table[addr->hash];
927 }
928
929 err = 0;
930 __unix_remove_socket(sk);
931 u->addr = addr;
932 __unix_insert_socket(list, sk);
933
934out_unlock:
935 spin_unlock(&unix_table_lock);
936out_up:
937 mutex_unlock(&u->readlock);
938out:
939 return err;
940}
941
942static void unix_state_double_lock(struct sock *sk1, struct sock *sk2)
943{
944 if (unlikely(sk1 == sk2) || !sk2) {
945 unix_state_lock(sk1);
946 return;
947 }
948 if (sk1 < sk2) {
949 unix_state_lock(sk1);
950 unix_state_lock_nested(sk2);
951 } else {
952 unix_state_lock(sk2);
953 unix_state_lock_nested(sk1);
954 }
955}
956
957static void unix_state_double_unlock(struct sock *sk1, struct sock *sk2)
958{
959 if (unlikely(sk1 == sk2) || !sk2) {
960 unix_state_unlock(sk1);
961 return;
962 }
963 unix_state_unlock(sk1);
964 unix_state_unlock(sk2);
965}
966
967static int unix_dgram_connect(struct socket *sock, struct sockaddr *addr,
968 int alen, int flags)
969{
970 struct sock *sk = sock->sk;
971 struct net *net = sock_net(sk);
972 struct sockaddr_un *sunaddr = (struct sockaddr_un *)addr;
973 struct sock *other;
974 unsigned int hash;
975 int err;
976
977 if (addr->sa_family != AF_UNSPEC) {
978 err = unix_mkname(sunaddr, alen, &hash);
979 if (err < 0)
980 goto out;
981 alen = err;
982
983 if (test_bit(SOCK_PASSCRED, &sock->flags) &&
984 !unix_sk(sk)->addr && (err = unix_autobind(sock)) != 0)
985 goto out;
986
987restart:
988 other = unix_find_other(net, sunaddr, alen, sock->type, hash, &err);
989 if (!other)
990 goto out;
991
992 unix_state_double_lock(sk, other);
993
994 /* Apparently VFS overslept socket death. Retry. */
995 if (sock_flag(other, SOCK_DEAD)) {
996 unix_state_double_unlock(sk, other);
997 sock_put(other);
998 goto restart;
999 }
1000
1001 err = -EPERM;
1002 if (!unix_may_send(sk, other))
1003 goto out_unlock;
1004
1005 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1006 if (err)
1007 goto out_unlock;
1008
1009 } else {
1010 /*
1011 * 1003.1g breaking connected state with AF_UNSPEC
1012 */
1013 other = NULL;
1014 unix_state_double_lock(sk, other);
1015 }
1016
1017 /*
1018 * If it was connected, reconnect.
1019 */
1020 if (unix_peer(sk)) {
1021 struct sock *old_peer = unix_peer(sk);
1022 unix_peer(sk) = other;
1023 unix_state_double_unlock(sk, other);
1024
1025 if (other != old_peer)
1026 unix_dgram_disconnected(sk, old_peer);
1027 sock_put(old_peer);
1028 } else {
1029 unix_peer(sk) = other;
1030 unix_state_double_unlock(sk, other);
1031 }
1032 return 0;
1033
1034out_unlock:
1035 unix_state_double_unlock(sk, other);
1036 sock_put(other);
1037out:
1038 return err;
1039}
1040
1041static long unix_wait_for_peer(struct sock *other, long timeo)
1042{
1043 struct unix_sock *u = unix_sk(other);
1044 int sched;
1045 DEFINE_WAIT(wait);
1046
1047 prepare_to_wait_exclusive(&u->peer_wait, &wait, TASK_INTERRUPTIBLE);
1048
1049 sched = !sock_flag(other, SOCK_DEAD) &&
1050 !(other->sk_shutdown & RCV_SHUTDOWN) &&
1051 unix_recvq_full(other);
1052
1053 unix_state_unlock(other);
1054
1055 if (sched)
1056 timeo = schedule_timeout(timeo);
1057
1058 finish_wait(&u->peer_wait, &wait);
1059 return timeo;
1060}
1061
1062static int unix_stream_connect(struct socket *sock, struct sockaddr *uaddr,
1063 int addr_len, int flags)
1064{
1065 struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
1066 struct sock *sk = sock->sk;
1067 struct net *net = sock_net(sk);
1068 struct unix_sock *u = unix_sk(sk), *newu, *otheru;
1069 struct sock *newsk = NULL;
1070 struct sock *other = NULL;
1071 struct sk_buff *skb = NULL;
1072 unsigned int hash;
1073 int st;
1074 int err;
1075 long timeo;
1076
1077 err = unix_mkname(sunaddr, addr_len, &hash);
1078 if (err < 0)
1079 goto out;
1080 addr_len = err;
1081
1082 if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr &&
1083 (err = unix_autobind(sock)) != 0)
1084 goto out;
1085
1086 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
1087
1088 /* First of all allocate resources.
1089 If we will make it after state is locked,
1090 we will have to recheck all again in any case.
1091 */
1092
1093 err = -ENOMEM;
1094
1095 /* create new sock for complete connection */
1096 newsk = unix_create1(sock_net(sk), NULL);
1097 if (newsk == NULL)
1098 goto out;
1099
1100 /* Allocate skb for sending to listening sock */
1101 skb = sock_wmalloc(newsk, 1, 0, GFP_KERNEL);
1102 if (skb == NULL)
1103 goto out;
1104
1105restart:
1106 /* Find listening sock. */
1107 other = unix_find_other(net, sunaddr, addr_len, sk->sk_type, hash, &err);
1108 if (!other)
1109 goto out;
1110
1111 /* Latch state of peer */
1112 unix_state_lock(other);
1113
1114 /* Apparently VFS overslept socket death. Retry. */
1115 if (sock_flag(other, SOCK_DEAD)) {
1116 unix_state_unlock(other);
1117 sock_put(other);
1118 goto restart;
1119 }
1120
1121 err = -ECONNREFUSED;
1122 if (other->sk_state != TCP_LISTEN)
1123 goto out_unlock;
1124 if (other->sk_shutdown & RCV_SHUTDOWN)
1125 goto out_unlock;
1126
1127 if (unix_recvq_full(other)) {
1128 err = -EAGAIN;
1129 if (!timeo)
1130 goto out_unlock;
1131
1132 timeo = unix_wait_for_peer(other, timeo);
1133
1134 err = sock_intr_errno(timeo);
1135 if (signal_pending(current))
1136 goto out;
1137 sock_put(other);
1138 goto restart;
1139 }
1140
1141 /* Latch our state.
1142
1143 It is tricky place. We need to grab our state lock and cannot
1144 drop lock on peer. It is dangerous because deadlock is
1145 possible. Connect to self case and simultaneous
1146 attempt to connect are eliminated by checking socket
1147 state. other is TCP_LISTEN, if sk is TCP_LISTEN we
1148 check this before attempt to grab lock.
1149
1150 Well, and we have to recheck the state after socket locked.
1151 */
1152 st = sk->sk_state;
1153
1154 switch (st) {
1155 case TCP_CLOSE:
1156 /* This is ok... continue with connect */
1157 break;
1158 case TCP_ESTABLISHED:
1159 /* Socket is already connected */
1160 err = -EISCONN;
1161 goto out_unlock;
1162 default:
1163 err = -EINVAL;
1164 goto out_unlock;
1165 }
1166
1167 unix_state_lock_nested(sk);
1168
1169 if (sk->sk_state != st) {
1170 unix_state_unlock(sk);
1171 unix_state_unlock(other);
1172 sock_put(other);
1173 goto restart;
1174 }
1175
1176 err = security_unix_stream_connect(sk, other, newsk);
1177 if (err) {
1178 unix_state_unlock(sk);
1179 goto out_unlock;
1180 }
1181
1182 /* The way is open! Fastly set all the necessary fields... */
1183
1184 sock_hold(sk);
1185 unix_peer(newsk) = sk;
1186 newsk->sk_state = TCP_ESTABLISHED;
1187 newsk->sk_type = sk->sk_type;
1188 init_peercred(newsk);
1189 newu = unix_sk(newsk);
1190 RCU_INIT_POINTER(newsk->sk_wq, &newu->peer_wq);
1191 otheru = unix_sk(other);
1192
1193 /* copy address information from listening to new sock*/
1194 if (otheru->addr) {
1195 atomic_inc(&otheru->addr->refcnt);
1196 newu->addr = otheru->addr;
1197 }
1198 if (otheru->path.dentry) {
1199 path_get(&otheru->path);
1200 newu->path = otheru->path;
1201 }
1202
1203 /* Set credentials */
1204 copy_peercred(sk, other);
1205
1206 sock->state = SS_CONNECTED;
1207 sk->sk_state = TCP_ESTABLISHED;
1208 sock_hold(newsk);
1209
1210 smp_mb__after_atomic_inc(); /* sock_hold() does an atomic_inc() */
1211 unix_peer(sk) = newsk;
1212
1213 unix_state_unlock(sk);
1214
1215 /* take ten and and send info to listening sock */
1216 spin_lock(&other->sk_receive_queue.lock);
1217 __skb_queue_tail(&other->sk_receive_queue, skb);
1218 spin_unlock(&other->sk_receive_queue.lock);
1219 unix_state_unlock(other);
1220 other->sk_data_ready(other);
1221 sock_put(other);
1222 return 0;
1223
1224out_unlock:
1225 if (other)
1226 unix_state_unlock(other);
1227
1228out:
1229 kfree_skb(skb);
1230 if (newsk)
1231 unix_release_sock(newsk, 0);
1232 if (other)
1233 sock_put(other);
1234 return err;
1235}
1236
1237static int unix_socketpair(struct socket *socka, struct socket *sockb)
1238{
1239 struct sock *ska = socka->sk, *skb = sockb->sk;
1240
1241 /* Join our sockets back to back */
1242 sock_hold(ska);
1243 sock_hold(skb);
1244 unix_peer(ska) = skb;
1245 unix_peer(skb) = ska;
1246 init_peercred(ska);
1247 init_peercred(skb);
1248
1249 if (ska->sk_type != SOCK_DGRAM) {
1250 ska->sk_state = TCP_ESTABLISHED;
1251 skb->sk_state = TCP_ESTABLISHED;
1252 socka->state = SS_CONNECTED;
1253 sockb->state = SS_CONNECTED;
1254 }
1255 return 0;
1256}
1257
1258static void unix_sock_inherit_flags(const struct socket *old,
1259 struct socket *new)
1260{
1261 if (test_bit(SOCK_PASSCRED, &old->flags))
1262 set_bit(SOCK_PASSCRED, &new->flags);
1263 if (test_bit(SOCK_PASSSEC, &old->flags))
1264 set_bit(SOCK_PASSSEC, &new->flags);
1265}
1266
1267static int unix_accept(struct socket *sock, struct socket *newsock, int flags)
1268{
1269 struct sock *sk = sock->sk;
1270 struct sock *tsk;
1271 struct sk_buff *skb;
1272 int err;
1273
1274 err = -EOPNOTSUPP;
1275 if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
1276 goto out;
1277
1278 err = -EINVAL;
1279 if (sk->sk_state != TCP_LISTEN)
1280 goto out;
1281
1282 /* If socket state is TCP_LISTEN it cannot change (for now...),
1283 * so that no locks are necessary.
1284 */
1285
1286 skb = skb_recv_datagram(sk, 0, flags&O_NONBLOCK, &err);
1287 if (!skb) {
1288 /* This means receive shutdown. */
1289 if (err == 0)
1290 err = -EINVAL;
1291 goto out;
1292 }
1293
1294 tsk = skb->sk;
1295 skb_free_datagram(sk, skb);
1296 wake_up_interruptible(&unix_sk(sk)->peer_wait);
1297
1298 /* attach accepted sock to socket */
1299 unix_state_lock(tsk);
1300 newsock->state = SS_CONNECTED;
1301 unix_sock_inherit_flags(sock, newsock);
1302 sock_graft(tsk, newsock);
1303 unix_state_unlock(tsk);
1304 return 0;
1305
1306out:
1307 return err;
1308}
1309
1310
1311static int unix_getname(struct socket *sock, struct sockaddr *uaddr, int *uaddr_len, int peer)
1312{
1313 struct sock *sk = sock->sk;
1314 struct unix_sock *u;
1315 DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr, uaddr);
1316 int err = 0;
1317
1318 if (peer) {
1319 sk = unix_peer_get(sk);
1320
1321 err = -ENOTCONN;
1322 if (!sk)
1323 goto out;
1324 err = 0;
1325 } else {
1326 sock_hold(sk);
1327 }
1328
1329 u = unix_sk(sk);
1330 unix_state_lock(sk);
1331 if (!u->addr) {
1332 sunaddr->sun_family = AF_UNIX;
1333 sunaddr->sun_path[0] = 0;
1334 *uaddr_len = sizeof(short);
1335 } else {
1336 struct unix_address *addr = u->addr;
1337
1338 *uaddr_len = addr->len;
1339 memcpy(sunaddr, addr->name, *uaddr_len);
1340 }
1341 unix_state_unlock(sk);
1342 sock_put(sk);
1343out:
1344 return err;
1345}
1346
1347static void unix_detach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1348{
1349 int i;
1350
1351 scm->fp = UNIXCB(skb).fp;
1352 UNIXCB(skb).fp = NULL;
1353
1354 for (i = scm->fp->count-1; i >= 0; i--)
1355 unix_notinflight(scm->fp->fp[i]);
1356}
1357
1358static void unix_destruct_scm(struct sk_buff *skb)
1359{
1360 struct scm_cookie scm;
1361 memset(&scm, 0, sizeof(scm));
1362 scm.pid = UNIXCB(skb).pid;
1363 if (UNIXCB(skb).fp)
1364 unix_detach_fds(&scm, skb);
1365
1366 /* Alas, it calls VFS */
1367 /* So fscking what? fput() had been SMP-safe since the last Summer */
1368 scm_destroy(&scm);
1369 sock_wfree(skb);
1370}
1371
1372#define MAX_RECURSION_LEVEL 4
1373
1374static int unix_attach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1375{
1376 int i;
1377 unsigned char max_level = 0;
1378 int unix_sock_count = 0;
1379
1380 for (i = scm->fp->count - 1; i >= 0; i--) {
1381 struct sock *sk = unix_get_socket(scm->fp->fp[i]);
1382
1383 if (sk) {
1384 unix_sock_count++;
1385 max_level = max(max_level,
1386 unix_sk(sk)->recursion_level);
1387 }
1388 }
1389 if (unlikely(max_level > MAX_RECURSION_LEVEL))
1390 return -ETOOMANYREFS;
1391
1392 /*
1393 * Need to duplicate file references for the sake of garbage
1394 * collection. Otherwise a socket in the fps might become a
1395 * candidate for GC while the skb is not yet queued.
1396 */
1397 UNIXCB(skb).fp = scm_fp_dup(scm->fp);
1398 if (!UNIXCB(skb).fp)
1399 return -ENOMEM;
1400
1401 if (unix_sock_count) {
1402 for (i = scm->fp->count - 1; i >= 0; i--)
1403 unix_inflight(scm->fp->fp[i]);
1404 }
1405 return max_level;
1406}
1407
1408static int unix_scm_to_skb(struct scm_cookie *scm, struct sk_buff *skb, bool send_fds)
1409{
1410 int err = 0;
1411
1412 UNIXCB(skb).pid = get_pid(scm->pid);
1413 UNIXCB(skb).uid = scm->creds.uid;
1414 UNIXCB(skb).gid = scm->creds.gid;
1415 UNIXCB(skb).fp = NULL;
1416 if (scm->fp && send_fds)
1417 err = unix_attach_fds(scm, skb);
1418
1419 skb->destructor = unix_destruct_scm;
1420 return err;
1421}
1422
1423/*
1424 * Some apps rely on write() giving SCM_CREDENTIALS
1425 * We include credentials if source or destination socket
1426 * asserted SOCK_PASSCRED.
1427 */
1428static void maybe_add_creds(struct sk_buff *skb, const struct socket *sock,
1429 const struct sock *other)
1430{
1431 if (UNIXCB(skb).pid)
1432 return;
1433 if (test_bit(SOCK_PASSCRED, &sock->flags) ||
1434 !other->sk_socket ||
1435 test_bit(SOCK_PASSCRED, &other->sk_socket->flags)) {
1436 UNIXCB(skb).pid = get_pid(task_tgid(current));
1437 current_uid_gid(&UNIXCB(skb).uid, &UNIXCB(skb).gid);
1438 }
1439}
1440
1441/*
1442 * Send AF_UNIX data.
1443 */
1444
1445static int unix_dgram_sendmsg(struct kiocb *kiocb, struct socket *sock,
1446 struct msghdr *msg, size_t len)
1447{
1448 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1449 struct sock *sk = sock->sk;
1450 struct net *net = sock_net(sk);
1451 struct unix_sock *u = unix_sk(sk);
1452 DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr, msg->msg_name);
1453 struct sock *other = NULL;
1454 int namelen = 0; /* fake GCC */
1455 int err;
1456 unsigned int hash;
1457 struct sk_buff *skb;
1458 long timeo;
1459 struct scm_cookie tmp_scm;
1460 int max_level;
1461 int data_len = 0;
1462
1463 if (NULL == siocb->scm)
1464 siocb->scm = &tmp_scm;
1465 wait_for_unix_gc();
1466 err = scm_send(sock, msg, siocb->scm, false);
1467 if (err < 0)
1468 return err;
1469
1470 err = -EOPNOTSUPP;
1471 if (msg->msg_flags&MSG_OOB)
1472 goto out;
1473
1474 if (msg->msg_namelen) {
1475 err = unix_mkname(sunaddr, msg->msg_namelen, &hash);
1476 if (err < 0)
1477 goto out;
1478 namelen = err;
1479 } else {
1480 sunaddr = NULL;
1481 err = -ENOTCONN;
1482 other = unix_peer_get(sk);
1483 if (!other)
1484 goto out;
1485 }
1486
1487 if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr
1488 && (err = unix_autobind(sock)) != 0)
1489 goto out;
1490
1491 err = -EMSGSIZE;
1492 if (len > sk->sk_sndbuf - 32)
1493 goto out;
1494
1495 if (len > SKB_MAX_ALLOC)
1496 data_len = min_t(size_t,
1497 len - SKB_MAX_ALLOC,
1498 MAX_SKB_FRAGS * PAGE_SIZE);
1499
1500 skb = sock_alloc_send_pskb(sk, len - data_len, data_len,
1501 msg->msg_flags & MSG_DONTWAIT, &err,
1502 PAGE_ALLOC_COSTLY_ORDER);
1503 if (skb == NULL)
1504 goto out;
1505
1506 err = unix_scm_to_skb(siocb->scm, skb, true);
1507 if (err < 0)
1508 goto out_free;
1509 max_level = err + 1;
1510 unix_get_secdata(siocb->scm, skb);
1511
1512 skb_put(skb, len - data_len);
1513 skb->data_len = data_len;
1514 skb->len = len;
1515 err = skb_copy_datagram_from_iovec(skb, 0, msg->msg_iov, 0, len);
1516 if (err)
1517 goto out_free;
1518
1519 timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1520
1521restart:
1522 if (!other) {
1523 err = -ECONNRESET;
1524 if (sunaddr == NULL)
1525 goto out_free;
1526
1527 other = unix_find_other(net, sunaddr, namelen, sk->sk_type,
1528 hash, &err);
1529 if (other == NULL)
1530 goto out_free;
1531 }
1532
1533 if (sk_filter(other, skb) < 0) {
1534 /* Toss the packet but do not return any error to the sender */
1535 err = len;
1536 goto out_free;
1537 }
1538
1539 unix_state_lock(other);
1540 err = -EPERM;
1541 if (!unix_may_send(sk, other))
1542 goto out_unlock;
1543
1544 if (sock_flag(other, SOCK_DEAD)) {
1545 /*
1546 * Check with 1003.1g - what should
1547 * datagram error
1548 */
1549 unix_state_unlock(other);
1550 sock_put(other);
1551
1552 err = 0;
1553 unix_state_lock(sk);
1554 if (unix_peer(sk) == other) {
1555 unix_peer(sk) = NULL;
1556 unix_state_unlock(sk);
1557
1558 unix_dgram_disconnected(sk, other);
1559 sock_put(other);
1560 err = -ECONNREFUSED;
1561 } else {
1562 unix_state_unlock(sk);
1563 }
1564
1565 other = NULL;
1566 if (err)
1567 goto out_free;
1568 goto restart;
1569 }
1570
1571 err = -EPIPE;
1572 if (other->sk_shutdown & RCV_SHUTDOWN)
1573 goto out_unlock;
1574
1575 if (sk->sk_type != SOCK_SEQPACKET) {
1576 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1577 if (err)
1578 goto out_unlock;
1579 }
1580
1581 if (unix_peer(other) != sk && unix_recvq_full(other)) {
1582 if (!timeo) {
1583 err = -EAGAIN;
1584 goto out_unlock;
1585 }
1586
1587 timeo = unix_wait_for_peer(other, timeo);
1588
1589 err = sock_intr_errno(timeo);
1590 if (signal_pending(current))
1591 goto out_free;
1592
1593 goto restart;
1594 }
1595
1596 if (sock_flag(other, SOCK_RCVTSTAMP))
1597 __net_timestamp(skb);
1598 maybe_add_creds(skb, sock, other);
1599 skb_queue_tail(&other->sk_receive_queue, skb);
1600 if (max_level > unix_sk(other)->recursion_level)
1601 unix_sk(other)->recursion_level = max_level;
1602 unix_state_unlock(other);
1603 other->sk_data_ready(other);
1604 sock_put(other);
1605 scm_destroy(siocb->scm);
1606 return len;
1607
1608out_unlock:
1609 unix_state_unlock(other);
1610out_free:
1611 kfree_skb(skb);
1612out:
1613 if (other)
1614 sock_put(other);
1615 scm_destroy(siocb->scm);
1616 return err;
1617}
1618
1619/* We use paged skbs for stream sockets, and limit occupancy to 32768
1620 * bytes, and a minimun of a full page.
1621 */
1622#define UNIX_SKB_FRAGS_SZ (PAGE_SIZE << get_order(32768))
1623
1624static int unix_stream_sendmsg(struct kiocb *kiocb, struct socket *sock,
1625 struct msghdr *msg, size_t len)
1626{
1627 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1628 struct sock *sk = sock->sk;
1629 struct sock *other = NULL;
1630 int err, size;
1631 struct sk_buff *skb;
1632 int sent = 0;
1633 struct scm_cookie tmp_scm;
1634 bool fds_sent = false;
1635 int max_level;
1636 int data_len;
1637
1638 if (NULL == siocb->scm)
1639 siocb->scm = &tmp_scm;
1640 wait_for_unix_gc();
1641 err = scm_send(sock, msg, siocb->scm, false);
1642 if (err < 0)
1643 return err;
1644
1645 err = -EOPNOTSUPP;
1646 if (msg->msg_flags&MSG_OOB)
1647 goto out_err;
1648
1649 if (msg->msg_namelen) {
1650 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
1651 goto out_err;
1652 } else {
1653 err = -ENOTCONN;
1654 other = unix_peer(sk);
1655 if (!other)
1656 goto out_err;
1657 }
1658
1659 if (sk->sk_shutdown & SEND_SHUTDOWN)
1660 goto pipe_err;
1661
1662 while (sent < len) {
1663 size = len - sent;
1664
1665 /* Keep two messages in the pipe so it schedules better */
1666 size = min_t(int, size, (sk->sk_sndbuf >> 1) - 64);
1667
1668 /* allow fallback to order-0 allocations */
1669 size = min_t(int, size, SKB_MAX_HEAD(0) + UNIX_SKB_FRAGS_SZ);
1670
1671 data_len = max_t(int, 0, size - SKB_MAX_HEAD(0));
1672
1673 skb = sock_alloc_send_pskb(sk, size - data_len, data_len,
1674 msg->msg_flags & MSG_DONTWAIT, &err,
1675 get_order(UNIX_SKB_FRAGS_SZ));
1676 if (!skb)
1677 goto out_err;
1678
1679 /* Only send the fds in the first buffer */
1680 err = unix_scm_to_skb(siocb->scm, skb, !fds_sent);
1681 if (err < 0) {
1682 kfree_skb(skb);
1683 goto out_err;
1684 }
1685 max_level = err + 1;
1686 fds_sent = true;
1687
1688 skb_put(skb, size - data_len);
1689 skb->data_len = data_len;
1690 skb->len = size;
1691 err = skb_copy_datagram_from_iovec(skb, 0, msg->msg_iov,
1692 sent, size);
1693 if (err) {
1694 kfree_skb(skb);
1695 goto out_err;
1696 }
1697
1698 unix_state_lock(other);
1699
1700 if (sock_flag(other, SOCK_DEAD) ||
1701 (other->sk_shutdown & RCV_SHUTDOWN))
1702 goto pipe_err_free;
1703
1704 maybe_add_creds(skb, sock, other);
1705 skb_queue_tail(&other->sk_receive_queue, skb);
1706 if (max_level > unix_sk(other)->recursion_level)
1707 unix_sk(other)->recursion_level = max_level;
1708 unix_state_unlock(other);
1709 other->sk_data_ready(other);
1710 sent += size;
1711 }
1712
1713 scm_destroy(siocb->scm);
1714 siocb->scm = NULL;
1715
1716 return sent;
1717
1718pipe_err_free:
1719 unix_state_unlock(other);
1720 kfree_skb(skb);
1721pipe_err:
1722 if (sent == 0 && !(msg->msg_flags&MSG_NOSIGNAL))
1723 send_sig(SIGPIPE, current, 0);
1724 err = -EPIPE;
1725out_err:
1726 scm_destroy(siocb->scm);
1727 siocb->scm = NULL;
1728 return sent ? : err;
1729}
1730
1731static int unix_seqpacket_sendmsg(struct kiocb *kiocb, struct socket *sock,
1732 struct msghdr *msg, size_t len)
1733{
1734 int err;
1735 struct sock *sk = sock->sk;
1736
1737 err = sock_error(sk);
1738 if (err)
1739 return err;
1740
1741 if (sk->sk_state != TCP_ESTABLISHED)
1742 return -ENOTCONN;
1743
1744 if (msg->msg_namelen)
1745 msg->msg_namelen = 0;
1746
1747 return unix_dgram_sendmsg(kiocb, sock, msg, len);
1748}
1749
1750static int unix_seqpacket_recvmsg(struct kiocb *iocb, struct socket *sock,
1751 struct msghdr *msg, size_t size,
1752 int flags)
1753{
1754 struct sock *sk = sock->sk;
1755
1756 if (sk->sk_state != TCP_ESTABLISHED)
1757 return -ENOTCONN;
1758
1759 return unix_dgram_recvmsg(iocb, sock, msg, size, flags);
1760}
1761
1762static void unix_copy_addr(struct msghdr *msg, struct sock *sk)
1763{
1764 struct unix_sock *u = unix_sk(sk);
1765
1766 if (u->addr) {
1767 msg->msg_namelen = u->addr->len;
1768 memcpy(msg->msg_name, u->addr->name, u->addr->len);
1769 }
1770}
1771
1772static int unix_dgram_recvmsg(struct kiocb *iocb, struct socket *sock,
1773 struct msghdr *msg, size_t size,
1774 int flags)
1775{
1776 struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1777 struct scm_cookie tmp_scm;
1778 struct sock *sk = sock->sk;
1779 struct unix_sock *u = unix_sk(sk);
1780 int noblock = flags & MSG_DONTWAIT;
1781 struct sk_buff *skb;
1782 int err;
1783 int peeked, skip;
1784
1785 err = -EOPNOTSUPP;
1786 if (flags&MSG_OOB)
1787 goto out;
1788
1789 err = mutex_lock_interruptible(&u->readlock);
1790 if (unlikely(err)) {
1791 /* recvmsg() in non blocking mode is supposed to return -EAGAIN
1792 * sk_rcvtimeo is not honored by mutex_lock_interruptible()
1793 */
1794 err = noblock ? -EAGAIN : -ERESTARTSYS;
1795 goto out;
1796 }
1797
1798 skip = sk_peek_offset(sk, flags);
1799
1800 skb = __skb_recv_datagram(sk, flags, &peeked, &skip, &err);
1801 if (!skb) {
1802 unix_state_lock(sk);
1803 /* Signal EOF on disconnected non-blocking SEQPACKET socket. */
1804 if (sk->sk_type == SOCK_SEQPACKET && err == -EAGAIN &&
1805 (sk->sk_shutdown & RCV_SHUTDOWN))
1806 err = 0;
1807 unix_state_unlock(sk);
1808 goto out_unlock;
1809 }
1810
1811 wake_up_interruptible_sync_poll(&u->peer_wait,
1812 POLLOUT | POLLWRNORM | POLLWRBAND);
1813
1814 if (msg->msg_name)
1815 unix_copy_addr(msg, skb->sk);
1816
1817 if (size > skb->len - skip)
1818 size = skb->len - skip;
1819 else if (size < skb->len - skip)
1820 msg->msg_flags |= MSG_TRUNC;
1821
1822 err = skb_copy_datagram_iovec(skb, skip, msg->msg_iov, size);
1823 if (err)
1824 goto out_free;
1825
1826 if (sock_flag(sk, SOCK_RCVTSTAMP))
1827 __sock_recv_timestamp(msg, sk, skb);
1828
1829 if (!siocb->scm) {
1830 siocb->scm = &tmp_scm;
1831 memset(&tmp_scm, 0, sizeof(tmp_scm));
1832 }
1833 scm_set_cred(siocb->scm, UNIXCB(skb).pid, UNIXCB(skb).uid, UNIXCB(skb).gid);
1834 unix_set_secdata(siocb->scm, skb);
1835
1836 if (!(flags & MSG_PEEK)) {
1837 if (UNIXCB(skb).fp)
1838 unix_detach_fds(siocb->scm, skb);
1839
1840 sk_peek_offset_bwd(sk, skb->len);
1841 } else {
1842 /* It is questionable: on PEEK we could:
1843 - do not return fds - good, but too simple 8)
1844 - return fds, and do not return them on read (old strategy,
1845 apparently wrong)
1846 - clone fds (I chose it for now, it is the most universal
1847 solution)
1848
1849 POSIX 1003.1g does not actually define this clearly
1850 at all. POSIX 1003.1g doesn't define a lot of things
1851 clearly however!
1852
1853 */
1854
1855 sk_peek_offset_fwd(sk, size);
1856
1857 if (UNIXCB(skb).fp)
1858 siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1859 }
1860 err = (flags & MSG_TRUNC) ? skb->len - skip : size;
1861
1862 scm_recv(sock, msg, siocb->scm, flags);
1863
1864out_free:
1865 skb_free_datagram(sk, skb);
1866out_unlock:
1867 mutex_unlock(&u->readlock);
1868out:
1869 return err;
1870}
1871
1872/*
1873 * Sleep until more data has arrived. But check for races..
1874 */
1875static long unix_stream_data_wait(struct sock *sk, long timeo,
1876 struct sk_buff *last)
1877{
1878 DEFINE_WAIT(wait);
1879
1880 unix_state_lock(sk);
1881
1882 for (;;) {
1883 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1884
1885 if (skb_peek_tail(&sk->sk_receive_queue) != last ||
1886 sk->sk_err ||
1887 (sk->sk_shutdown & RCV_SHUTDOWN) ||
1888 signal_pending(current) ||
1889 !timeo)
1890 break;
1891
1892 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1893 unix_state_unlock(sk);
1894 timeo = freezable_schedule_timeout(timeo);
1895 unix_state_lock(sk);
1896 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1897 }
1898
1899 finish_wait(sk_sleep(sk), &wait);
1900 unix_state_unlock(sk);
1901 return timeo;
1902}
1903
1904static unsigned int unix_skb_len(const struct sk_buff *skb)
1905{
1906 return skb->len - UNIXCB(skb).consumed;
1907}
1908
1909static int unix_stream_recvmsg(struct kiocb *iocb, struct socket *sock,
1910 struct msghdr *msg, size_t size,
1911 int flags)
1912{
1913 struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1914 struct scm_cookie tmp_scm;
1915 struct sock *sk = sock->sk;
1916 struct unix_sock *u = unix_sk(sk);
1917 DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr, msg->msg_name);
1918 int copied = 0;
1919 int noblock = flags & MSG_DONTWAIT;
1920 int check_creds = 0;
1921 int target;
1922 int err = 0;
1923 long timeo;
1924 int skip;
1925
1926 err = -EINVAL;
1927 if (sk->sk_state != TCP_ESTABLISHED)
1928 goto out;
1929
1930 err = -EOPNOTSUPP;
1931 if (flags&MSG_OOB)
1932 goto out;
1933
1934 target = sock_rcvlowat(sk, flags&MSG_WAITALL, size);
1935 timeo = sock_rcvtimeo(sk, noblock);
1936
1937 /* Lock the socket to prevent queue disordering
1938 * while sleeps in memcpy_tomsg
1939 */
1940
1941 if (!siocb->scm) {
1942 siocb->scm = &tmp_scm;
1943 memset(&tmp_scm, 0, sizeof(tmp_scm));
1944 }
1945
1946 err = mutex_lock_interruptible(&u->readlock);
1947 if (unlikely(err)) {
1948 /* recvmsg() in non blocking mode is supposed to return -EAGAIN
1949 * sk_rcvtimeo is not honored by mutex_lock_interruptible()
1950 */
1951 err = noblock ? -EAGAIN : -ERESTARTSYS;
1952 goto out;
1953 }
1954
1955 do {
1956 int chunk;
1957 struct sk_buff *skb, *last;
1958
1959 unix_state_lock(sk);
1960 last = skb = skb_peek(&sk->sk_receive_queue);
1961again:
1962 if (skb == NULL) {
1963 unix_sk(sk)->recursion_level = 0;
1964 if (copied >= target)
1965 goto unlock;
1966
1967 /*
1968 * POSIX 1003.1g mandates this order.
1969 */
1970
1971 err = sock_error(sk);
1972 if (err)
1973 goto unlock;
1974 if (sk->sk_shutdown & RCV_SHUTDOWN)
1975 goto unlock;
1976
1977 unix_state_unlock(sk);
1978 err = -EAGAIN;
1979 if (!timeo)
1980 break;
1981 mutex_unlock(&u->readlock);
1982
1983 timeo = unix_stream_data_wait(sk, timeo, last);
1984
1985 if (signal_pending(current)
1986 || mutex_lock_interruptible(&u->readlock)) {
1987 err = sock_intr_errno(timeo);
1988 goto out;
1989 }
1990
1991 continue;
1992 unlock:
1993 unix_state_unlock(sk);
1994 break;
1995 }
1996
1997 skip = sk_peek_offset(sk, flags);
1998 while (skip >= unix_skb_len(skb)) {
1999 skip -= unix_skb_len(skb);
2000 last = skb;
2001 skb = skb_peek_next(skb, &sk->sk_receive_queue);
2002 if (!skb)
2003 goto again;
2004 }
2005
2006 unix_state_unlock(sk);
2007
2008 if (check_creds) {
2009 /* Never glue messages from different writers */
2010 if ((UNIXCB(skb).pid != siocb->scm->pid) ||
2011 !uid_eq(UNIXCB(skb).uid, siocb->scm->creds.uid) ||
2012 !gid_eq(UNIXCB(skb).gid, siocb->scm->creds.gid))
2013 break;
2014 } else if (test_bit(SOCK_PASSCRED, &sock->flags)) {
2015 /* Copy credentials */
2016 scm_set_cred(siocb->scm, UNIXCB(skb).pid, UNIXCB(skb).uid, UNIXCB(skb).gid);
2017 check_creds = 1;
2018 }
2019
2020 /* Copy address just once */
2021 if (sunaddr) {
2022 unix_copy_addr(msg, skb->sk);
2023 sunaddr = NULL;
2024 }
2025
2026 chunk = min_t(unsigned int, unix_skb_len(skb) - skip, size);
2027 if (skb_copy_datagram_iovec(skb, UNIXCB(skb).consumed + skip,
2028 msg->msg_iov, chunk)) {
2029 if (copied == 0)
2030 copied = -EFAULT;
2031 break;
2032 }
2033 copied += chunk;
2034 size -= chunk;
2035
2036 /* Mark read part of skb as used */
2037 if (!(flags & MSG_PEEK)) {
2038 UNIXCB(skb).consumed += chunk;
2039
2040 sk_peek_offset_bwd(sk, chunk);
2041
2042 if (UNIXCB(skb).fp)
2043 unix_detach_fds(siocb->scm, skb);
2044
2045 if (unix_skb_len(skb))
2046 break;
2047
2048 skb_unlink(skb, &sk->sk_receive_queue);
2049 consume_skb(skb);
2050
2051 if (siocb->scm->fp)
2052 break;
2053 } else {
2054 /* It is questionable, see note in unix_dgram_recvmsg.
2055 */
2056 if (UNIXCB(skb).fp)
2057 siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
2058
2059 sk_peek_offset_fwd(sk, chunk);
2060
2061 break;
2062 }
2063 } while (size);
2064
2065 mutex_unlock(&u->readlock);
2066 scm_recv(sock, msg, siocb->scm, flags);
2067out:
2068 return copied ? : err;
2069}
2070
2071static int unix_shutdown(struct socket *sock, int mode)
2072{
2073 struct sock *sk = sock->sk;
2074 struct sock *other;
2075
2076 if (mode < SHUT_RD || mode > SHUT_RDWR)
2077 return -EINVAL;
2078 /* This maps:
2079 * SHUT_RD (0) -> RCV_SHUTDOWN (1)
2080 * SHUT_WR (1) -> SEND_SHUTDOWN (2)
2081 * SHUT_RDWR (2) -> SHUTDOWN_MASK (3)
2082 */
2083 ++mode;
2084
2085 unix_state_lock(sk);
2086 sk->sk_shutdown |= mode;
2087 other = unix_peer(sk);
2088 if (other)
2089 sock_hold(other);
2090 unix_state_unlock(sk);
2091 sk->sk_state_change(sk);
2092
2093 if (other &&
2094 (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET)) {
2095
2096 int peer_mode = 0;
2097
2098 if (mode&RCV_SHUTDOWN)
2099 peer_mode |= SEND_SHUTDOWN;
2100 if (mode&SEND_SHUTDOWN)
2101 peer_mode |= RCV_SHUTDOWN;
2102 unix_state_lock(other);
2103 other->sk_shutdown |= peer_mode;
2104 unix_state_unlock(other);
2105 other->sk_state_change(other);
2106 if (peer_mode == SHUTDOWN_MASK)
2107 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_HUP);
2108 else if (peer_mode & RCV_SHUTDOWN)
2109 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_IN);
2110 }
2111 if (other)
2112 sock_put(other);
2113
2114 return 0;
2115}
2116
2117long unix_inq_len(struct sock *sk)
2118{
2119 struct sk_buff *skb;
2120 long amount = 0;
2121
2122 if (sk->sk_state == TCP_LISTEN)
2123 return -EINVAL;
2124
2125 spin_lock(&sk->sk_receive_queue.lock);
2126 if (sk->sk_type == SOCK_STREAM ||
2127 sk->sk_type == SOCK_SEQPACKET) {
2128 skb_queue_walk(&sk->sk_receive_queue, skb)
2129 amount += unix_skb_len(skb);
2130 } else {
2131 skb = skb_peek(&sk->sk_receive_queue);
2132 if (skb)
2133 amount = skb->len;
2134 }
2135 spin_unlock(&sk->sk_receive_queue.lock);
2136
2137 return amount;
2138}
2139EXPORT_SYMBOL_GPL(unix_inq_len);
2140
2141long unix_outq_len(struct sock *sk)
2142{
2143 return sk_wmem_alloc_get(sk);
2144}
2145EXPORT_SYMBOL_GPL(unix_outq_len);
2146
2147static int unix_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2148{
2149 struct sock *sk = sock->sk;
2150 long amount = 0;
2151 int err;
2152
2153 switch (cmd) {
2154 case SIOCOUTQ:
2155 amount = unix_outq_len(sk);
2156 err = put_user(amount, (int __user *)arg);
2157 break;
2158 case SIOCINQ:
2159 amount = unix_inq_len(sk);
2160 if (amount < 0)
2161 err = amount;
2162 else
2163 err = put_user(amount, (int __user *)arg);
2164 break;
2165 default:
2166 err = -ENOIOCTLCMD;
2167 break;
2168 }
2169 return err;
2170}
2171
2172static unsigned int unix_poll(struct file *file, struct socket *sock, poll_table *wait)
2173{
2174 struct sock *sk = sock->sk;
2175 unsigned int mask;
2176
2177 sock_poll_wait(file, sk_sleep(sk), wait);
2178 mask = 0;
2179
2180 /* exceptional events? */
2181 if (sk->sk_err)
2182 mask |= POLLERR;
2183 if (sk->sk_shutdown == SHUTDOWN_MASK)
2184 mask |= POLLHUP;
2185 if (sk->sk_shutdown & RCV_SHUTDOWN)
2186 mask |= POLLRDHUP | POLLIN | POLLRDNORM;
2187
2188 /* readable? */
2189 if (!skb_queue_empty(&sk->sk_receive_queue))
2190 mask |= POLLIN | POLLRDNORM;
2191
2192 /* Connection-based need to check for termination and startup */
2193 if ((sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) &&
2194 sk->sk_state == TCP_CLOSE)
2195 mask |= POLLHUP;
2196
2197 /*
2198 * we set writable also when the other side has shut down the
2199 * connection. This prevents stuck sockets.
2200 */
2201 if (unix_writable(sk))
2202 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2203
2204 return mask;
2205}
2206
2207static unsigned int unix_dgram_poll(struct file *file, struct socket *sock,
2208 poll_table *wait)
2209{
2210 struct sock *sk = sock->sk, *other;
2211 unsigned int mask, writable;
2212
2213 sock_poll_wait(file, sk_sleep(sk), wait);
2214 mask = 0;
2215
2216 /* exceptional events? */
2217 if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue))
2218 mask |= POLLERR |
2219 (sock_flag(sk, SOCK_SELECT_ERR_QUEUE) ? POLLPRI : 0);
2220
2221 if (sk->sk_shutdown & RCV_SHUTDOWN)
2222 mask |= POLLRDHUP | POLLIN | POLLRDNORM;
2223 if (sk->sk_shutdown == SHUTDOWN_MASK)
2224 mask |= POLLHUP;
2225
2226 /* readable? */
2227 if (!skb_queue_empty(&sk->sk_receive_queue))
2228 mask |= POLLIN | POLLRDNORM;
2229
2230 /* Connection-based need to check for termination and startup */
2231 if (sk->sk_type == SOCK_SEQPACKET) {
2232 if (sk->sk_state == TCP_CLOSE)
2233 mask |= POLLHUP;
2234 /* connection hasn't started yet? */
2235 if (sk->sk_state == TCP_SYN_SENT)
2236 return mask;
2237 }
2238
2239 /* No write status requested, avoid expensive OUT tests. */
2240 if (!(poll_requested_events(wait) & (POLLWRBAND|POLLWRNORM|POLLOUT)))
2241 return mask;
2242
2243 writable = unix_writable(sk);
2244 other = unix_peer_get(sk);
2245 if (other) {
2246 if (unix_peer(other) != sk) {
2247 sock_poll_wait(file, &unix_sk(other)->peer_wait, wait);
2248 if (unix_recvq_full(other))
2249 writable = 0;
2250 }
2251 sock_put(other);
2252 }
2253
2254 if (writable)
2255 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2256 else
2257 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
2258
2259 return mask;
2260}
2261
2262#ifdef CONFIG_PROC_FS
2263
2264#define BUCKET_SPACE (BITS_PER_LONG - (UNIX_HASH_BITS + 1) - 1)
2265
2266#define get_bucket(x) ((x) >> BUCKET_SPACE)
2267#define get_offset(x) ((x) & ((1L << BUCKET_SPACE) - 1))
2268#define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
2269
2270static struct sock *unix_from_bucket(struct seq_file *seq, loff_t *pos)
2271{
2272 unsigned long offset = get_offset(*pos);
2273 unsigned long bucket = get_bucket(*pos);
2274 struct sock *sk;
2275 unsigned long count = 0;
2276
2277 for (sk = sk_head(&unix_socket_table[bucket]); sk; sk = sk_next(sk)) {
2278 if (sock_net(sk) != seq_file_net(seq))
2279 continue;
2280 if (++count == offset)
2281 break;
2282 }
2283
2284 return sk;
2285}
2286
2287static struct sock *unix_next_socket(struct seq_file *seq,
2288 struct sock *sk,
2289 loff_t *pos)
2290{
2291 unsigned long bucket;
2292
2293 while (sk > (struct sock *)SEQ_START_TOKEN) {
2294 sk = sk_next(sk);
2295 if (!sk)
2296 goto next_bucket;
2297 if (sock_net(sk) == seq_file_net(seq))
2298 return sk;
2299 }
2300
2301 do {
2302 sk = unix_from_bucket(seq, pos);
2303 if (sk)
2304 return sk;
2305
2306next_bucket:
2307 bucket = get_bucket(*pos) + 1;
2308 *pos = set_bucket_offset(bucket, 1);
2309 } while (bucket < ARRAY_SIZE(unix_socket_table));
2310
2311 return NULL;
2312}
2313
2314static void *unix_seq_start(struct seq_file *seq, loff_t *pos)
2315 __acquires(unix_table_lock)
2316{
2317 spin_lock(&unix_table_lock);
2318
2319 if (!*pos)
2320 return SEQ_START_TOKEN;
2321
2322 if (get_bucket(*pos) >= ARRAY_SIZE(unix_socket_table))
2323 return NULL;
2324
2325 return unix_next_socket(seq, NULL, pos);
2326}
2327
2328static void *unix_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2329{
2330 ++*pos;
2331 return unix_next_socket(seq, v, pos);
2332}
2333
2334static void unix_seq_stop(struct seq_file *seq, void *v)
2335 __releases(unix_table_lock)
2336{
2337 spin_unlock(&unix_table_lock);
2338}
2339
2340static int unix_seq_show(struct seq_file *seq, void *v)
2341{
2342
2343 if (v == SEQ_START_TOKEN)
2344 seq_puts(seq, "Num RefCount Protocol Flags Type St "
2345 "Inode Path\n");
2346 else {
2347 struct sock *s = v;
2348 struct unix_sock *u = unix_sk(s);
2349 unix_state_lock(s);
2350
2351 seq_printf(seq, "%pK: %08X %08X %08X %04X %02X %5lu",
2352 s,
2353 atomic_read(&s->sk_refcnt),
2354 0,
2355 s->sk_state == TCP_LISTEN ? __SO_ACCEPTCON : 0,
2356 s->sk_type,
2357 s->sk_socket ?
2358 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTED : SS_UNCONNECTED) :
2359 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTING : SS_DISCONNECTING),
2360 sock_i_ino(s));
2361
2362 if (u->addr) {
2363 int i, len;
2364 seq_putc(seq, ' ');
2365
2366 i = 0;
2367 len = u->addr->len - sizeof(short);
2368 if (!UNIX_ABSTRACT(s))
2369 len--;
2370 else {
2371 seq_putc(seq, '@');
2372 i++;
2373 }
2374 for ( ; i < len; i++)
2375 seq_putc(seq, u->addr->name->sun_path[i]);
2376 }
2377 unix_state_unlock(s);
2378 seq_putc(seq, '\n');
2379 }
2380
2381 return 0;
2382}
2383
2384static const struct seq_operations unix_seq_ops = {
2385 .start = unix_seq_start,
2386 .next = unix_seq_next,
2387 .stop = unix_seq_stop,
2388 .show = unix_seq_show,
2389};
2390
2391static int unix_seq_open(struct inode *inode, struct file *file)
2392{
2393 return seq_open_net(inode, file, &unix_seq_ops,
2394 sizeof(struct seq_net_private));
2395}
2396
2397static const struct file_operations unix_seq_fops = {
2398 .owner = THIS_MODULE,
2399 .open = unix_seq_open,
2400 .read = seq_read,
2401 .llseek = seq_lseek,
2402 .release = seq_release_net,
2403};
2404
2405#endif
2406
2407static const struct net_proto_family unix_family_ops = {
2408 .family = PF_UNIX,
2409 .create = unix_create,
2410 .owner = THIS_MODULE,
2411};
2412
2413
2414static int __net_init unix_net_init(struct net *net)
2415{
2416 int error = -ENOMEM;
2417
2418 net->unx.sysctl_max_dgram_qlen = 10;
2419 if (unix_sysctl_register(net))
2420 goto out;
2421
2422#ifdef CONFIG_PROC_FS
2423 if (!proc_create("unix", 0, net->proc_net, &unix_seq_fops)) {
2424 unix_sysctl_unregister(net);
2425 goto out;
2426 }
2427#endif
2428 error = 0;
2429out:
2430 return error;
2431}
2432
2433static void __net_exit unix_net_exit(struct net *net)
2434{
2435 unix_sysctl_unregister(net);
2436 remove_proc_entry("unix", net->proc_net);
2437}
2438
2439static struct pernet_operations unix_net_ops = {
2440 .init = unix_net_init,
2441 .exit = unix_net_exit,
2442};
2443
2444static int __init af_unix_init(void)
2445{
2446 int rc = -1;
2447
2448 BUILD_BUG_ON(sizeof(struct unix_skb_parms) > FIELD_SIZEOF(struct sk_buff, cb));
2449
2450 rc = proto_register(&unix_proto, 1);
2451 if (rc != 0) {
2452 pr_crit("%s: Cannot create unix_sock SLAB cache!\n", __func__);
2453 goto out;
2454 }
2455
2456 sock_register(&unix_family_ops);
2457 register_pernet_subsys(&unix_net_ops);
2458out:
2459 return rc;
2460}
2461
2462static void __exit af_unix_exit(void)
2463{
2464 sock_unregister(PF_UNIX);
2465 proto_unregister(&unix_proto);
2466 unregister_pernet_subsys(&unix_net_ops);
2467}
2468
2469/* Earlier than device_initcall() so that other drivers invoking
2470 request_module() don't end up in a loop when modprobe tries
2471 to use a UNIX socket. But later than subsys_initcall() because
2472 we depend on stuff initialised there */
2473fs_initcall(af_unix_init);
2474module_exit(af_unix_exit);
2475
2476MODULE_LICENSE("GPL");
2477MODULE_ALIAS_NETPROTO(PF_UNIX);