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