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