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