blob: c5bf5ef2bf89476b6d7ab766b8591e6af2e869fd [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * NET4: Implementation of BSD Unix domain sockets.
3 *
Alan Cox113aa832008-10-13 19:01:08 -07004 * Authors: Alan Cox, <alan@lxorguk.ukuu.org.uk>
Linus Torvalds1da177e2005-04-16 15:20:36 -07005 *
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 *
Linus Torvalds1da177e2005-04-16 15:20:36 -070011 * 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
wangweidong5cc208b2013-12-06 18:03:36 +080083#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
84
Linus Torvalds1da177e2005-04-16 15:20:36 -070085#include <linux/module.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070086#include <linux/kernel.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070087#include <linux/signal.h>
88#include <linux/sched.h>
89#include <linux/errno.h>
90#include <linux/string.h>
91#include <linux/stat.h>
92#include <linux/dcache.h>
93#include <linux/namei.h>
94#include <linux/socket.h>
95#include <linux/un.h>
96#include <linux/fcntl.h>
97#include <linux/termios.h>
98#include <linux/sockios.h>
99#include <linux/net.h>
100#include <linux/in.h>
101#include <linux/fs.h>
102#include <linux/slab.h>
103#include <asm/uaccess.h>
104#include <linux/skbuff.h>
105#include <linux/netdevice.h>
Eric W. Biederman457c4cb2007-09-12 12:01:34 +0200106#include <net/net_namespace.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107#include <net/sock.h>
Arnaldo Carvalho de Meloc752f072005-08-09 20:08:28 -0700108#include <net/tcp_states.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109#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>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115#include <linux/rtnetlink.h>
116#include <linux/mount.h>
117#include <net/checksum.h>
118#include <linux/security.h>
Colin Cross2b15af62013-05-06 23:50:21 +0000119#include <linux/freezer.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120
Eric Dumazet7123aaa2012-06-08 05:03:21 +0000121struct hlist_head unix_socket_table[2 * UNIX_HASH_SIZE];
Pavel Emelyanovfa7ff562011-12-15 02:44:03 +0000122EXPORT_SYMBOL_GPL(unix_socket_table);
123DEFINE_SPINLOCK(unix_table_lock);
124EXPORT_SYMBOL_GPL(unix_table_lock);
Eric Dumazet518de9b2010-10-26 14:22:44 -0700125static atomic_long_t unix_nr_socks;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127
Eric Dumazet7123aaa2012-06-08 05:03:21 +0000128static 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)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139
Catherine Zhang877ce7c2006-06-29 12:27:47 -0700140#ifdef CONFIG_SECURITY_NETWORK
Catherine Zhangdc49c1f2006-08-02 14:12:06 -0700141static void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb)
Catherine Zhang877ce7c2006-06-29 12:27:47 -0700142{
Stephen Smalley37a9a8d2015-06-10 08:44:59 -0400143 UNIXCB(skb).secid = scm->secid;
Catherine Zhang877ce7c2006-06-29 12:27:47 -0700144}
145
146static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb)
147{
Stephen Smalley37a9a8d2015-06-10 08:44:59 -0400148 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);
Catherine Zhang877ce7c2006-06-29 12:27:47 -0700154}
155#else
Catherine Zhangdc49c1f2006-08-02 14:12:06 -0700156static inline void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb)
Catherine Zhang877ce7c2006-06-29 12:27:47 -0700157{ }
158
159static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb)
160{ }
Stephen Smalley37a9a8d2015-06-10 08:44:59 -0400161
162static inline bool unix_secdata_eq(struct scm_cookie *scm, struct sk_buff *skb)
163{
164 return true;
165}
Catherine Zhang877ce7c2006-06-29 12:27:47 -0700166#endif /* CONFIG_SECURITY_NETWORK */
167
Linus Torvalds1da177e2005-04-16 15:20:36 -0700168/*
169 * SMP locking strategy:
David S. Millerfbe9cc42005-12-13 23:26:29 -0800170 * hash table is protected with spinlock unix_table_lock
Stephen Hemminger663717f2010-02-18 14:12:06 -0800171 * each socket state is protected by separate spin lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172 */
173
Eric Dumazet95c96172012-04-15 05:58:06 +0000174static inline unsigned int unix_hash_fold(__wsum n)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175{
Anton Blanchard0a134042014-03-05 14:29:58 +1100176 unsigned int hash = (__force unsigned int)csum_fold(n);
Eric Dumazet95c96172012-04-15 05:58:06 +0000177
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178 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{
Eric Dumazet6eba6a32008-11-16 22:58:44 -0800191 return unix_peer(osk) == NULL || unix_our_peer(sk, osk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192}
193
Rainer Weikusat3c734192008-06-17 22:28:05 -0700194static 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
Pavel Emelyanovfa7ff562011-12-15 02:44:03 +0000199struct sock *unix_peer_get(struct sock *s)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200{
201 struct sock *peer;
202
David S. Miller1c92b4e2007-05-31 13:24:26 -0700203 unix_state_lock(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204 peer = unix_peer(s);
205 if (peer)
206 sock_hold(peer);
David S. Miller1c92b4e2007-05-31 13:24:26 -0700207 unix_state_unlock(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208 return peer;
209}
Pavel Emelyanovfa7ff562011-12-15 02:44:03 +0000210EXPORT_SYMBOL_GPL(unix_peer_get);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211
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 */
YOSHIFUJI Hideakiac7bfa62007-02-09 23:25:23 +0900224
Eric Dumazet95c96172012-04-15 05:58:06 +0000225static int unix_mkname(struct sockaddr_un *sunaddr, int len, unsigned int *hashp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226{
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.
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300235 * sun_path[108] doesn't as such exist. However in kernel space
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236 * we are guaranteed that it is a valid memory location in our
237 * kernel address buffer.
238 */
Jianjun Konge27dfce2008-11-01 21:38:31 -0700239 ((char *)sunaddr)[len] = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240 len = strlen(sunaddr->sun_path)+1+sizeof(short);
241 return len;
242 }
243
Joe Perches07f07572008-11-19 15:44:53 -0800244 *hashp = unix_hash_fold(csum_partial(sunaddr, len, 0));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245 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{
Ilpo Järvinen547b7922008-07-25 21:43:18 -0700255 WARN_ON(!sk_unhashed(sk));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256 sk_add_node(sk, list);
257}
258
259static inline void unix_remove_socket(struct sock *sk)
260{
David S. Millerfbe9cc42005-12-13 23:26:29 -0800261 spin_lock(&unix_table_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262 __unix_remove_socket(sk);
David S. Millerfbe9cc42005-12-13 23:26:29 -0800263 spin_unlock(&unix_table_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264}
265
266static inline void unix_insert_socket(struct hlist_head *list, struct sock *sk)
267{
David S. Millerfbe9cc42005-12-13 23:26:29 -0800268 spin_lock(&unix_table_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269 __unix_insert_socket(list, sk);
David S. Millerfbe9cc42005-12-13 23:26:29 -0800270 spin_unlock(&unix_table_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271}
272
Denis V. Lunev097e66c2007-11-19 22:29:30 -0800273static struct sock *__unix_find_socket_byname(struct net *net,
274 struct sockaddr_un *sunname,
Eric Dumazet95c96172012-04-15 05:58:06 +0000275 int len, int type, unsigned int hash)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276{
277 struct sock *s;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278
Sasha Levinb67bfe02013-02-27 17:06:00 -0800279 sk_for_each(s, &unix_socket_table[hash ^ type]) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280 struct unix_sock *u = unix_sk(s);
281
YOSHIFUJI Hideaki878628f2008-03-26 03:57:35 +0900282 if (!net_eq(sock_net(s), net))
Denis V. Lunev097e66c2007-11-19 22:29:30 -0800283 continue;
284
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285 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
Denis V. Lunev097e66c2007-11-19 22:29:30 -0800294static inline struct sock *unix_find_socket_byname(struct net *net,
295 struct sockaddr_un *sunname,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296 int len, int type,
Eric Dumazet95c96172012-04-15 05:58:06 +0000297 unsigned int hash)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298{
299 struct sock *s;
300
David S. Millerfbe9cc42005-12-13 23:26:29 -0800301 spin_lock(&unix_table_lock);
Denis V. Lunev097e66c2007-11-19 22:29:30 -0800302 s = __unix_find_socket_byname(net, sunname, len, type, hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303 if (s)
304 sock_hold(s);
David S. Millerfbe9cc42005-12-13 23:26:29 -0800305 spin_unlock(&unix_table_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306 return s;
307}
308
Eric W. Biederman6616f782010-06-13 03:35:48 +0000309static struct sock *unix_find_socket_byinode(struct inode *i)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310{
311 struct sock *s;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312
David S. Millerfbe9cc42005-12-13 23:26:29 -0800313 spin_lock(&unix_table_lock);
Sasha Levinb67bfe02013-02-27 17:06:00 -0800314 sk_for_each(s,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315 &unix_socket_table[i->i_ino & (UNIX_HASH_SIZE - 1)]) {
Al Viro40ffe672012-03-14 21:54:32 -0400316 struct dentry *dentry = unix_sk(s)->path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317
David Howellsa25b3762015-03-17 22:26:21 +0000318 if (dentry && d_backing_inode(dentry) == i) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319 sock_hold(s);
320 goto found;
321 }
322 }
323 s = NULL;
324found:
David S. Millerfbe9cc42005-12-13 23:26:29 -0800325 spin_unlock(&unix_table_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326 return s;
327}
328
Rainer Weikusat7d267272015-11-20 22:07:23 +0000329/* 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
Eric Dumazet1586a582015-10-23 10:59:16 -0700441static int unix_writable(const struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442{
Eric Dumazet1586a582015-10-23 10:59:16 -0700443 return sk->sk_state != TCP_LISTEN &&
444 (atomic_read(&sk->sk_wmem_alloc) << 2) <= sk->sk_sndbuf;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445}
446
447static void unix_write_space(struct sock *sk)
448{
Eric Dumazet43815482010-04-29 11:01:49 +0000449 struct socket_wq *wq;
450
451 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452 if (unix_writable(sk)) {
Eric Dumazet43815482010-04-29 11:01:49 +0000453 wq = rcu_dereference(sk->sk_wq);
Herbert Xu1ce0bf52015-11-26 13:55:39 +0800454 if (skwq_has_sleeper(wq))
Eric Dumazet67426b72010-10-29 20:44:44 +0000455 wake_up_interruptible_sync_poll(&wq->wait,
456 POLLOUT | POLLWRNORM | POLLWRBAND);
Pavel Emelyanov8d8ad9d2007-11-26 20:10:50 +0800457 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458 }
Eric Dumazet43815482010-04-29 11:01:49 +0000459 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460}
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{
David S. Millerb03efcf2005-07-08 14:57:23 -0700468 if (!skb_queue_empty(&sk->sk_receive_queue)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469 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
Ilpo Järvinen547b7922008-07-25 21:43:18 -0700489 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
490 WARN_ON(!sk_unhashed(sk));
491 WARN_ON(sk->sk_socket);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492 if (!sock_flag(sk, SOCK_DEAD)) {
wangweidong5cc208b2013-12-06 18:03:36 +0800493 pr_info("Attempt to release alive unix socket: %p\n", sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494 return;
495 }
496
497 if (u->addr)
498 unix_release_addr(u->addr);
499
Eric Dumazet518de9b2010-10-26 14:22:44 -0700500 atomic_long_dec(&unix_nr_socks);
David S. Miller6f756a82008-11-23 17:34:03 -0800501 local_bh_disable();
Eric Dumazeta8076d82008-11-17 02:38:49 -0800502 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
David S. Miller6f756a82008-11-23 17:34:03 -0800503 local_bh_enable();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504#ifdef UNIX_REFCNT_DEBUG
wangweidong5cc208b2013-12-06 18:03:36 +0800505 pr_debug("UNIX %p is destroyed, %ld are still alive.\n", sk,
Eric Dumazet518de9b2010-10-26 14:22:44 -0700506 atomic_long_read(&unix_nr_socks));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507#endif
508}
509
Paul Mooreded34e02013-03-25 03:18:33 +0000510static void unix_release_sock(struct sock *sk, int embrion)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511{
512 struct unix_sock *u = unix_sk(sk);
Al Viro40ffe672012-03-14 21:54:32 -0400513 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514 struct sock *skpair;
515 struct sk_buff *skb;
516 int state;
517
518 unix_remove_socket(sk);
519
520 /* Clear state */
David S. Miller1c92b4e2007-05-31 13:24:26 -0700521 unix_state_lock(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522 sock_orphan(sk);
523 sk->sk_shutdown = SHUTDOWN_MASK;
Al Viro40ffe672012-03-14 21:54:32 -0400524 path = u->path;
525 u->path.dentry = NULL;
526 u->path.mnt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527 state = sk->sk_state;
528 sk->sk_state = TCP_CLOSE;
David S. Miller1c92b4e2007-05-31 13:24:26 -0700529 unix_state_unlock(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530
531 wake_up_interruptible_all(&u->peer_wait);
532
Jianjun Konge27dfce2008-11-01 21:38:31 -0700533 skpair = unix_peer(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534
Jianjun Konge27dfce2008-11-01 21:38:31 -0700535 if (skpair != NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536 if (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) {
David S. Miller1c92b4e2007-05-31 13:24:26 -0700537 unix_state_lock(skpair);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538 /* No more writes */
539 skpair->sk_shutdown = SHUTDOWN_MASK;
540 if (!skb_queue_empty(&sk->sk_receive_queue) || embrion)
541 skpair->sk_err = ECONNRESET;
David S. Miller1c92b4e2007-05-31 13:24:26 -0700542 unix_state_unlock(skpair);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543 skpair->sk_state_change(skpair);
Pavel Emelyanov8d8ad9d2007-11-26 20:10:50 +0800544 sk_wake_async(skpair, SOCK_WAKE_WAITD, POLL_HUP);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545 }
Rainer Weikusat7d267272015-11-20 22:07:23 +0000546
547 unix_dgram_peer_wake_disconnect(sk, skpair);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548 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) {
Jianjun Konge27dfce2008-11-01 21:38:31 -0700555 if (state == TCP_LISTEN)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556 unix_release_sock(skb->sk, 1);
557 /* passed fds are erased in the kfree_skb hook */
Hannes Frederic Sowa73ed5d22015-11-10 16:23:15 +0100558 UNIXCB(skb).consumed = skb->len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559 kfree_skb(skb);
560 }
561
Al Viro40ffe672012-03-14 21:54:32 -0400562 if (path.dentry)
563 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564
565 sock_put(sk);
566
567 /* ---- Socket is dead now and most probably destroyed ---- */
568
569 /*
Alan Coxe04dae82012-09-17 00:52:41 +0000570 * Fixme: BSD difference: In BSD all sockets connected to us get
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571 * 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
Pavel Emelyanov9305cfa2007-11-10 22:06:01 -0800580 if (unix_tot_inflight)
YOSHIFUJI Hideakiac7bfa62007-02-09 23:25:23 +0900581 unix_gc(); /* Garbage collect fds */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582}
583
Eric W. Biederman109f6e32010-06-13 03:30:14 +0000584static 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
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602static 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);
Eric W. Biederman109f6e32010-06-13 03:30:14 +0000607 struct pid *old_pid = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608
609 err = -EOPNOTSUPP;
Eric Dumazet6eba6a32008-11-16 22:58:44 -0800610 if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
611 goto out; /* Only stream/seqpacket sockets accept */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612 err = -EINVAL;
613 if (!u->addr)
Eric Dumazet6eba6a32008-11-16 22:58:44 -0800614 goto out; /* No listens on an unbound socket */
David S. Miller1c92b4e2007-05-31 13:24:26 -0700615 unix_state_lock(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616 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 */
Eric W. Biederman109f6e32010-06-13 03:30:14 +0000623 init_peercred(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 err = 0;
625
626out_unlock:
David S. Miller1c92b4e2007-05-31 13:24:26 -0700627 unix_state_unlock(sk);
Eric W. Biederman109f6e32010-06-13 03:30:14 +0000628 put_pid(old_pid);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629out:
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 *);
Rainer Weikusatec0d2152008-06-27 19:34:18 -0700641static unsigned int unix_dgram_poll(struct file *, struct socket *,
642 poll_table *);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643static int unix_ioctl(struct socket *, unsigned int, unsigned long);
644static int unix_shutdown(struct socket *, int);
Ying Xue1b784142015-03-02 15:37:48 +0800645static int unix_stream_sendmsg(struct socket *, struct msghdr *, size_t);
646static int unix_stream_recvmsg(struct socket *, struct msghdr *, size_t, int);
Hannes Frederic Sowa869e7c62015-05-21 16:59:59 +0200647static ssize_t unix_stream_sendpage(struct socket *, struct page *, int offset,
648 size_t size, int flags);
Hannes Frederic Sowa2b514572015-05-21 17:00:01 +0200649static ssize_t unix_stream_splice_read(struct socket *, loff_t *ppos,
650 struct pipe_inode_info *, size_t size,
651 unsigned int flags);
Ying Xue1b784142015-03-02 15:37:48 +0800652static int unix_dgram_sendmsg(struct socket *, struct msghdr *, size_t);
653static int unix_dgram_recvmsg(struct socket *, struct msghdr *, size_t, int);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654static int unix_dgram_connect(struct socket *, struct sockaddr *,
655 int, int);
Ying Xue1b784142015-03-02 15:37:48 +0800656static int unix_seqpacket_sendmsg(struct socket *, struct msghdr *, size_t);
657static int unix_seqpacket_recvmsg(struct socket *, struct msghdr *, size_t,
658 int);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659
Sasha Levin12663bf2013-12-07 17:26:27 -0500660static int unix_set_peek_off(struct sock *sk, int val)
Pavel Emelyanovf55bb7f2012-02-21 07:31:51 +0000661{
662 struct unix_sock *u = unix_sk(sk);
663
Sasha Levin12663bf2013-12-07 17:26:27 -0500664 if (mutex_lock_interruptible(&u->readlock))
665 return -EINTR;
666
Pavel Emelyanovf55bb7f2012-02-21 07:31:51 +0000667 sk->sk_peek_off = val;
668 mutex_unlock(&u->readlock);
Sasha Levin12663bf2013-12-07 17:26:27 -0500669
670 return 0;
Pavel Emelyanovf55bb7f2012-02-21 07:31:51 +0000671}
672
673
Eric Dumazet90ddc4f2005-12-22 12:49:22 -0800674static const struct proto_ops unix_stream_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675 .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,
Hannes Frederic Sowa869e7c62015-05-21 16:59:59 +0200692 .sendpage = unix_stream_sendpage,
Hannes Frederic Sowa2b514572015-05-21 17:00:01 +0200693 .splice_read = unix_stream_splice_read,
Pavel Emelyanovfc0d7532012-02-21 07:32:06 +0000694 .set_peek_off = unix_set_peek_off,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695};
696
Eric Dumazet90ddc4f2005-12-22 12:49:22 -0800697static const struct proto_ops unix_dgram_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698 .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,
Rainer Weikusatec0d2152008-06-27 19:34:18 -0700706 .poll = unix_dgram_poll,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707 .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,
Pavel Emelyanovf55bb7f2012-02-21 07:31:51 +0000716 .set_peek_off = unix_set_peek_off,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717};
718
Eric Dumazet90ddc4f2005-12-22 12:49:22 -0800719static const struct proto_ops unix_seqpacket_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720 .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,
Rainer Weikusatec0d2152008-06-27 19:34:18 -0700728 .poll = unix_dgram_poll,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729 .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,
Eric W. Biedermana05d2ad2011-04-24 01:54:57 +0000735 .recvmsg = unix_seqpacket_recvmsg,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736 .mmap = sock_no_mmap,
737 .sendpage = sock_no_sendpage,
Pavel Emelyanovf55bb7f2012-02-21 07:31:51 +0000738 .set_peek_off = unix_set_peek_off,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739};
740
741static struct proto unix_proto = {
Eric Dumazet248969a2008-11-17 00:00:30 -0800742 .name = "UNIX",
743 .owner = THIS_MODULE,
Eric Dumazet248969a2008-11-17 00:00:30 -0800744 .obj_size = sizeof(struct unix_sock),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745};
746
Ingo Molnara09785a2006-07-03 00:25:12 -0700747/*
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
Eric W. Biederman11aa9c22015-05-08 21:09:13 -0500755static struct sock *unix_create1(struct net *net, struct socket *sock, int kern)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756{
757 struct sock *sk = NULL;
758 struct unix_sock *u;
759
Eric Dumazet518de9b2010-10-26 14:22:44 -0700760 atomic_long_inc(&unix_nr_socks);
761 if (atomic_long_read(&unix_nr_socks) > 2 * get_max_files())
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762 goto out;
763
Eric W. Biederman11aa9c22015-05-08 21:09:13 -0500764 sk = sk_alloc(net, PF_UNIX, GFP_KERNEL, &unix_proto, kern);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765 if (!sk)
766 goto out;
767
Eric Dumazet6eba6a32008-11-16 22:58:44 -0800768 sock_init_data(sock, sk);
Ingo Molnara09785a2006-07-03 00:25:12 -0700769 lockdep_set_class(&sk->sk_receive_queue.lock,
770 &af_unix_sk_receive_queue_lock_key);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700771
772 sk->sk_write_space = unix_write_space;
Denis V. Luneva0a53c82007-12-11 04:19:17 -0800773 sk->sk_max_ack_backlog = net->unx.sysctl_max_dgram_qlen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774 sk->sk_destruct = unix_sock_destructor;
775 u = unix_sk(sk);
Al Viro40ffe672012-03-14 21:54:32 -0400776 u->path.dentry = NULL;
777 u->path.mnt = NULL;
Benjamin LaHaisefd19f322006-01-03 14:10:46 -0800778 spin_lock_init(&u->lock);
Al Viro516e0cc2008-07-26 00:39:17 -0400779 atomic_long_set(&u->inflight, 0);
Miklos Szeredi1fd05ba2007-07-11 14:22:39 -0700780 INIT_LIST_HEAD(&u->link);
Ingo Molnar57b47a52006-03-20 22:35:41 -0800781 mutex_init(&u->readlock); /* single task reading lock */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782 init_waitqueue_head(&u->peer_wait);
Rainer Weikusat7d267272015-11-20 22:07:23 +0000783 init_waitqueue_func_entry(&u->peer_wake, unix_dgram_peer_wake_relay);
Eric Dumazet7123aaa2012-06-08 05:03:21 +0000784 unix_insert_socket(unix_sockets_unbound(sk), sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785out:
Pavel Emelyanov284b3272007-11-10 22:08:30 -0800786 if (sk == NULL)
Eric Dumazet518de9b2010-10-26 14:22:44 -0700787 atomic_long_dec(&unix_nr_socks);
Eric Dumazet920de802008-11-24 00:09:29 -0800788 else {
789 local_bh_disable();
Eric Dumazeta8076d82008-11-17 02:38:49 -0800790 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
Eric Dumazet920de802008-11-24 00:09:29 -0800791 local_bh_enable();
792 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793 return sk;
794}
795
Eric Paris3f378b62009-11-05 22:18:14 -0800796static int unix_create(struct net *net, struct socket *sock, int protocol,
797 int kern)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798{
799 if (protocol && protocol != PF_UNIX)
800 return -EPROTONOSUPPORT;
801
802 sock->state = SS_UNCONNECTED;
803
804 switch (sock->type) {
805 case SOCK_STREAM:
806 sock->ops = &unix_stream_ops;
807 break;
808 /*
809 * Believe it or not BSD has AF_UNIX, SOCK_RAW though
810 * nothing uses it.
811 */
812 case SOCK_RAW:
Jianjun Konge27dfce2008-11-01 21:38:31 -0700813 sock->type = SOCK_DGRAM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814 case SOCK_DGRAM:
815 sock->ops = &unix_dgram_ops;
816 break;
817 case SOCK_SEQPACKET:
818 sock->ops = &unix_seqpacket_ops;
819 break;
820 default:
821 return -ESOCKTNOSUPPORT;
822 }
823
Eric W. Biederman11aa9c22015-05-08 21:09:13 -0500824 return unix_create1(net, sock, kern) ? 0 : -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825}
826
827static int unix_release(struct socket *sock)
828{
829 struct sock *sk = sock->sk;
830
831 if (!sk)
832 return 0;
833
Paul Mooreded34e02013-03-25 03:18:33 +0000834 unix_release_sock(sk, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835 sock->sk = NULL;
836
Paul Mooreded34e02013-03-25 03:18:33 +0000837 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838}
839
840static int unix_autobind(struct socket *sock)
841{
842 struct sock *sk = sock->sk;
YOSHIFUJI Hideaki3b1e0a62008-03-26 02:26:21 +0900843 struct net *net = sock_net(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844 struct unix_sock *u = unix_sk(sk);
845 static u32 ordernum = 1;
Eric Dumazet6eba6a32008-11-16 22:58:44 -0800846 struct unix_address *addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847 int err;
Tetsuo Handa8df73ff2010-09-04 01:34:28 +0000848 unsigned int retries = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849
Sasha Levin37ab4fa2013-12-13 10:54:22 -0500850 err = mutex_lock_interruptible(&u->readlock);
851 if (err)
852 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853
854 err = 0;
855 if (u->addr)
856 goto out;
857
858 err = -ENOMEM;
Panagiotis Issaris0da974f2006-07-21 14:51:30 -0700859 addr = kzalloc(sizeof(*addr) + sizeof(short) + 16, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860 if (!addr)
861 goto out;
862
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863 addr->name->sun_family = AF_UNIX;
864 atomic_set(&addr->refcnt, 1);
865
866retry:
867 addr->len = sprintf(addr->name->sun_path+1, "%05x", ordernum) + 1 + sizeof(short);
Joe Perches07f07572008-11-19 15:44:53 -0800868 addr->hash = unix_hash_fold(csum_partial(addr->name, addr->len, 0));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869
David S. Millerfbe9cc42005-12-13 23:26:29 -0800870 spin_lock(&unix_table_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871 ordernum = (ordernum+1)&0xFFFFF;
872
Denis V. Lunev097e66c2007-11-19 22:29:30 -0800873 if (__unix_find_socket_byname(net, addr->name, addr->len, sock->type,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874 addr->hash)) {
David S. Millerfbe9cc42005-12-13 23:26:29 -0800875 spin_unlock(&unix_table_lock);
Tetsuo Handa8df73ff2010-09-04 01:34:28 +0000876 /*
877 * __unix_find_socket_byname() may take long time if many names
878 * are already in use.
879 */
880 cond_resched();
881 /* Give up if all names seems to be in use. */
882 if (retries++ == 0xFFFFF) {
883 err = -ENOSPC;
884 kfree(addr);
885 goto out;
886 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887 goto retry;
888 }
889 addr->hash ^= sk->sk_type;
890
891 __unix_remove_socket(sk);
892 u->addr = addr;
893 __unix_insert_socket(&unix_socket_table[addr->hash], sk);
David S. Millerfbe9cc42005-12-13 23:26:29 -0800894 spin_unlock(&unix_table_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895 err = 0;
896
Ingo Molnar57b47a52006-03-20 22:35:41 -0800897out: mutex_unlock(&u->readlock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898 return err;
899}
900
Denis V. Lunev097e66c2007-11-19 22:29:30 -0800901static struct sock *unix_find_other(struct net *net,
902 struct sockaddr_un *sunname, int len,
Eric Dumazet95c96172012-04-15 05:58:06 +0000903 int type, unsigned int hash, int *error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904{
905 struct sock *u;
Al Viro421748e2008-08-02 01:04:36 -0400906 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907 int err = 0;
YOSHIFUJI Hideakiac7bfa62007-02-09 23:25:23 +0900908
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909 if (sunname->sun_path[0]) {
Al Viro421748e2008-08-02 01:04:36 -0400910 struct inode *inode;
911 err = kern_path(sunname->sun_path, LOOKUP_FOLLOW, &path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912 if (err)
913 goto fail;
David Howellsa25b3762015-03-17 22:26:21 +0000914 inode = d_backing_inode(path.dentry);
Al Viro421748e2008-08-02 01:04:36 -0400915 err = inode_permission(inode, MAY_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916 if (err)
917 goto put_fail;
918
919 err = -ECONNREFUSED;
Al Viro421748e2008-08-02 01:04:36 -0400920 if (!S_ISSOCK(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921 goto put_fail;
Eric W. Biederman6616f782010-06-13 03:35:48 +0000922 u = unix_find_socket_byinode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923 if (!u)
924 goto put_fail;
925
926 if (u->sk_type == type)
Al Viro68ac1232012-03-15 08:21:57 -0400927 touch_atime(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928
Al Viro421748e2008-08-02 01:04:36 -0400929 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930
Jianjun Konge27dfce2008-11-01 21:38:31 -0700931 err = -EPROTOTYPE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932 if (u->sk_type != type) {
933 sock_put(u);
934 goto fail;
935 }
936 } else {
937 err = -ECONNREFUSED;
Jianjun Konge27dfce2008-11-01 21:38:31 -0700938 u = unix_find_socket_byname(net, sunname, len, type, hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939 if (u) {
940 struct dentry *dentry;
Al Viro40ffe672012-03-14 21:54:32 -0400941 dentry = unix_sk(u)->path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942 if (dentry)
Al Viro68ac1232012-03-15 08:21:57 -0400943 touch_atime(&unix_sk(u)->path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944 } else
945 goto fail;
946 }
947 return u;
948
949put_fail:
Al Viro421748e2008-08-02 01:04:36 -0400950 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951fail:
Jianjun Konge27dfce2008-11-01 21:38:31 -0700952 *error = err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953 return NULL;
954}
955
Rainer Weikusatc845acb2016-01-03 18:56:38 +0000956static int unix_mknod(struct dentry *dentry, struct path *path, umode_t mode,
957 struct path *res)
Al Virofaf02012012-07-20 02:37:29 +0400958{
Rainer Weikusatc845acb2016-01-03 18:56:38 +0000959 int err;
Al Virofaf02012012-07-20 02:37:29 +0400960
Rainer Weikusatc845acb2016-01-03 18:56:38 +0000961 err = security_path_mknod(path, dentry, mode, 0);
Al Virofaf02012012-07-20 02:37:29 +0400962 if (!err) {
Rainer Weikusatc845acb2016-01-03 18:56:38 +0000963 err = vfs_mknod(d_inode(path->dentry), dentry, mode, 0);
Al Virofaf02012012-07-20 02:37:29 +0400964 if (!err) {
Rainer Weikusatc845acb2016-01-03 18:56:38 +0000965 res->mnt = mntget(path->mnt);
Al Virofaf02012012-07-20 02:37:29 +0400966 res->dentry = dget(dentry);
967 }
968 }
Rainer Weikusatc845acb2016-01-03 18:56:38 +0000969
Al Virofaf02012012-07-20 02:37:29 +0400970 return err;
971}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972
973static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
974{
975 struct sock *sk = sock->sk;
YOSHIFUJI Hideaki3b1e0a62008-03-26 02:26:21 +0900976 struct net *net = sock_net(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977 struct unix_sock *u = unix_sk(sk);
Jianjun Konge27dfce2008-11-01 21:38:31 -0700978 struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
Al Virodae6ad82011-06-26 11:50:15 -0400979 char *sun_path = sunaddr->sun_path;
Rainer Weikusatc845acb2016-01-03 18:56:38 +0000980 int err, name_err;
Eric Dumazet95c96172012-04-15 05:58:06 +0000981 unsigned int hash;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982 struct unix_address *addr;
983 struct hlist_head *list;
Rainer Weikusatc845acb2016-01-03 18:56:38 +0000984 struct path path;
985 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986
987 err = -EINVAL;
988 if (sunaddr->sun_family != AF_UNIX)
989 goto out;
990
Jianjun Konge27dfce2008-11-01 21:38:31 -0700991 if (addr_len == sizeof(short)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992 err = unix_autobind(sock);
993 goto out;
994 }
995
996 err = unix_mkname(sunaddr, addr_len, &hash);
997 if (err < 0)
998 goto out;
999 addr_len = err;
1000
Rainer Weikusatc845acb2016-01-03 18:56:38 +00001001 name_err = 0;
1002 dentry = NULL;
1003 if (sun_path[0]) {
1004 /* Get the parent directory, calculate the hash for last
1005 * component.
1006 */
1007 dentry = kern_path_create(AT_FDCWD, sun_path, &path, 0);
1008
1009 if (IS_ERR(dentry)) {
1010 /* delay report until after 'already bound' check */
1011 name_err = PTR_ERR(dentry);
1012 dentry = NULL;
1013 }
1014 }
1015
Sasha Levin37ab4fa2013-12-13 10:54:22 -05001016 err = mutex_lock_interruptible(&u->readlock);
1017 if (err)
Rainer Weikusatc845acb2016-01-03 18:56:38 +00001018 goto out_path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019
1020 err = -EINVAL;
1021 if (u->addr)
1022 goto out_up;
1023
Rainer Weikusatc845acb2016-01-03 18:56:38 +00001024 if (name_err) {
1025 err = name_err == -EEXIST ? -EADDRINUSE : name_err;
1026 goto out_up;
1027 }
1028
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029 err = -ENOMEM;
1030 addr = kmalloc(sizeof(*addr)+addr_len, GFP_KERNEL);
1031 if (!addr)
1032 goto out_up;
1033
1034 memcpy(addr->name, sunaddr, addr_len);
1035 addr->len = addr_len;
1036 addr->hash = hash ^ sk->sk_type;
1037 atomic_set(&addr->refcnt, 1);
1038
Rainer Weikusatc845acb2016-01-03 18:56:38 +00001039 if (dentry) {
1040 struct path u_path;
Al Virofaf02012012-07-20 02:37:29 +04001041 umode_t mode = S_IFSOCK |
Al Viroce3b0f82009-03-29 19:08:22 -04001042 (SOCK_INODE(sock)->i_mode & ~current_umask());
Rainer Weikusatc845acb2016-01-03 18:56:38 +00001043 err = unix_mknod(dentry, &path, mode, &u_path);
Al Virofaf02012012-07-20 02:37:29 +04001044 if (err) {
1045 if (err == -EEXIST)
1046 err = -EADDRINUSE;
1047 unix_release_addr(addr);
1048 goto out_up;
1049 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050 addr->hash = UNIX_HASH_SIZE;
Rainer Weikusatc845acb2016-01-03 18:56:38 +00001051 hash = d_backing_inode(dentry)->i_ino & (UNIX_HASH_SIZE - 1);
Al Virofaf02012012-07-20 02:37:29 +04001052 spin_lock(&unix_table_lock);
Rainer Weikusatc845acb2016-01-03 18:56:38 +00001053 u->path = u_path;
Al Virofaf02012012-07-20 02:37:29 +04001054 list = &unix_socket_table[hash];
1055 } else {
1056 spin_lock(&unix_table_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057 err = -EADDRINUSE;
Denis V. Lunev097e66c2007-11-19 22:29:30 -08001058 if (__unix_find_socket_byname(net, sunaddr, addr_len,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059 sk->sk_type, hash)) {
1060 unix_release_addr(addr);
1061 goto out_unlock;
1062 }
1063
1064 list = &unix_socket_table[addr->hash];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065 }
1066
1067 err = 0;
1068 __unix_remove_socket(sk);
1069 u->addr = addr;
1070 __unix_insert_socket(list, sk);
1071
1072out_unlock:
David S. Millerfbe9cc42005-12-13 23:26:29 -08001073 spin_unlock(&unix_table_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074out_up:
Ingo Molnar57b47a52006-03-20 22:35:41 -08001075 mutex_unlock(&u->readlock);
Rainer Weikusatc845acb2016-01-03 18:56:38 +00001076out_path:
1077 if (dentry)
1078 done_path_create(&path, dentry);
1079
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080out:
1081 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082}
1083
David S. Miller278a3de2007-05-31 15:19:20 -07001084static void unix_state_double_lock(struct sock *sk1, struct sock *sk2)
1085{
1086 if (unlikely(sk1 == sk2) || !sk2) {
1087 unix_state_lock(sk1);
1088 return;
1089 }
1090 if (sk1 < sk2) {
1091 unix_state_lock(sk1);
1092 unix_state_lock_nested(sk2);
1093 } else {
1094 unix_state_lock(sk2);
1095 unix_state_lock_nested(sk1);
1096 }
1097}
1098
1099static void unix_state_double_unlock(struct sock *sk1, struct sock *sk2)
1100{
1101 if (unlikely(sk1 == sk2) || !sk2) {
1102 unix_state_unlock(sk1);
1103 return;
1104 }
1105 unix_state_unlock(sk1);
1106 unix_state_unlock(sk2);
1107}
1108
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109static int unix_dgram_connect(struct socket *sock, struct sockaddr *addr,
1110 int alen, int flags)
1111{
1112 struct sock *sk = sock->sk;
YOSHIFUJI Hideaki3b1e0a62008-03-26 02:26:21 +09001113 struct net *net = sock_net(sk);
Jianjun Konge27dfce2008-11-01 21:38:31 -07001114 struct sockaddr_un *sunaddr = (struct sockaddr_un *)addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115 struct sock *other;
Eric Dumazet95c96172012-04-15 05:58:06 +00001116 unsigned int hash;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117 int err;
1118
1119 if (addr->sa_family != AF_UNSPEC) {
1120 err = unix_mkname(sunaddr, alen, &hash);
1121 if (err < 0)
1122 goto out;
1123 alen = err;
1124
1125 if (test_bit(SOCK_PASSCRED, &sock->flags) &&
1126 !unix_sk(sk)->addr && (err = unix_autobind(sock)) != 0)
1127 goto out;
1128
David S. Miller278a3de2007-05-31 15:19:20 -07001129restart:
Jianjun Konge27dfce2008-11-01 21:38:31 -07001130 other = unix_find_other(net, sunaddr, alen, sock->type, hash, &err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131 if (!other)
1132 goto out;
1133
David S. Miller278a3de2007-05-31 15:19:20 -07001134 unix_state_double_lock(sk, other);
1135
1136 /* Apparently VFS overslept socket death. Retry. */
1137 if (sock_flag(other, SOCK_DEAD)) {
1138 unix_state_double_unlock(sk, other);
1139 sock_put(other);
1140 goto restart;
1141 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001142
1143 err = -EPERM;
1144 if (!unix_may_send(sk, other))
1145 goto out_unlock;
1146
1147 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1148 if (err)
1149 goto out_unlock;
1150
1151 } else {
1152 /*
1153 * 1003.1g breaking connected state with AF_UNSPEC
1154 */
1155 other = NULL;
David S. Miller278a3de2007-05-31 15:19:20 -07001156 unix_state_double_lock(sk, other);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001157 }
1158
1159 /*
1160 * If it was connected, reconnect.
1161 */
1162 if (unix_peer(sk)) {
1163 struct sock *old_peer = unix_peer(sk);
Jianjun Konge27dfce2008-11-01 21:38:31 -07001164 unix_peer(sk) = other;
Rainer Weikusat7d267272015-11-20 22:07:23 +00001165 unix_dgram_peer_wake_disconnect_wakeup(sk, old_peer);
1166
David S. Miller278a3de2007-05-31 15:19:20 -07001167 unix_state_double_unlock(sk, other);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168
1169 if (other != old_peer)
1170 unix_dgram_disconnected(sk, old_peer);
1171 sock_put(old_peer);
1172 } else {
Jianjun Konge27dfce2008-11-01 21:38:31 -07001173 unix_peer(sk) = other;
David S. Miller278a3de2007-05-31 15:19:20 -07001174 unix_state_double_unlock(sk, other);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175 }
YOSHIFUJI Hideakiac7bfa62007-02-09 23:25:23 +09001176 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177
1178out_unlock:
David S. Miller278a3de2007-05-31 15:19:20 -07001179 unix_state_double_unlock(sk, other);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001180 sock_put(other);
1181out:
1182 return err;
1183}
1184
1185static long unix_wait_for_peer(struct sock *other, long timeo)
1186{
1187 struct unix_sock *u = unix_sk(other);
1188 int sched;
1189 DEFINE_WAIT(wait);
1190
1191 prepare_to_wait_exclusive(&u->peer_wait, &wait, TASK_INTERRUPTIBLE);
1192
1193 sched = !sock_flag(other, SOCK_DEAD) &&
1194 !(other->sk_shutdown & RCV_SHUTDOWN) &&
Rainer Weikusat3c734192008-06-17 22:28:05 -07001195 unix_recvq_full(other);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001196
David S. Miller1c92b4e2007-05-31 13:24:26 -07001197 unix_state_unlock(other);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198
1199 if (sched)
1200 timeo = schedule_timeout(timeo);
1201
1202 finish_wait(&u->peer_wait, &wait);
1203 return timeo;
1204}
1205
1206static int unix_stream_connect(struct socket *sock, struct sockaddr *uaddr,
1207 int addr_len, int flags)
1208{
Jianjun Konge27dfce2008-11-01 21:38:31 -07001209 struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210 struct sock *sk = sock->sk;
YOSHIFUJI Hideaki3b1e0a62008-03-26 02:26:21 +09001211 struct net *net = sock_net(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212 struct unix_sock *u = unix_sk(sk), *newu, *otheru;
1213 struct sock *newsk = NULL;
1214 struct sock *other = NULL;
1215 struct sk_buff *skb = NULL;
Eric Dumazet95c96172012-04-15 05:58:06 +00001216 unsigned int hash;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217 int st;
1218 int err;
1219 long timeo;
1220
1221 err = unix_mkname(sunaddr, addr_len, &hash);
1222 if (err < 0)
1223 goto out;
1224 addr_len = err;
1225
Joe Perchesf64f9e72009-11-29 16:55:45 -08001226 if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr &&
1227 (err = unix_autobind(sock)) != 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228 goto out;
1229
1230 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
1231
1232 /* First of all allocate resources.
1233 If we will make it after state is locked,
1234 we will have to recheck all again in any case.
1235 */
1236
1237 err = -ENOMEM;
1238
1239 /* create new sock for complete connection */
Eric W. Biederman11aa9c22015-05-08 21:09:13 -05001240 newsk = unix_create1(sock_net(sk), NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241 if (newsk == NULL)
1242 goto out;
1243
1244 /* Allocate skb for sending to listening sock */
1245 skb = sock_wmalloc(newsk, 1, 0, GFP_KERNEL);
1246 if (skb == NULL)
1247 goto out;
1248
1249restart:
1250 /* Find listening sock. */
Denis V. Lunev097e66c2007-11-19 22:29:30 -08001251 other = unix_find_other(net, sunaddr, addr_len, sk->sk_type, hash, &err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252 if (!other)
1253 goto out;
1254
1255 /* Latch state of peer */
David S. Miller1c92b4e2007-05-31 13:24:26 -07001256 unix_state_lock(other);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257
1258 /* Apparently VFS overslept socket death. Retry. */
1259 if (sock_flag(other, SOCK_DEAD)) {
David S. Miller1c92b4e2007-05-31 13:24:26 -07001260 unix_state_unlock(other);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261 sock_put(other);
1262 goto restart;
1263 }
1264
1265 err = -ECONNREFUSED;
1266 if (other->sk_state != TCP_LISTEN)
1267 goto out_unlock;
Tomoki Sekiyama77238f22009-10-18 23:17:37 -07001268 if (other->sk_shutdown & RCV_SHUTDOWN)
1269 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270
Rainer Weikusat3c734192008-06-17 22:28:05 -07001271 if (unix_recvq_full(other)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272 err = -EAGAIN;
1273 if (!timeo)
1274 goto out_unlock;
1275
1276 timeo = unix_wait_for_peer(other, timeo);
1277
1278 err = sock_intr_errno(timeo);
1279 if (signal_pending(current))
1280 goto out;
1281 sock_put(other);
1282 goto restart;
YOSHIFUJI Hideakiac7bfa62007-02-09 23:25:23 +09001283 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284
1285 /* Latch our state.
1286
Daniel Balutae5537bf2011-03-14 15:25:33 -07001287 It is tricky place. We need to grab our state lock and cannot
Linus Torvalds1da177e2005-04-16 15:20:36 -07001288 drop lock on peer. It is dangerous because deadlock is
1289 possible. Connect to self case and simultaneous
1290 attempt to connect are eliminated by checking socket
1291 state. other is TCP_LISTEN, if sk is TCP_LISTEN we
1292 check this before attempt to grab lock.
1293
1294 Well, and we have to recheck the state after socket locked.
1295 */
1296 st = sk->sk_state;
1297
1298 switch (st) {
1299 case TCP_CLOSE:
1300 /* This is ok... continue with connect */
1301 break;
1302 case TCP_ESTABLISHED:
1303 /* Socket is already connected */
1304 err = -EISCONN;
1305 goto out_unlock;
1306 default:
1307 err = -EINVAL;
1308 goto out_unlock;
1309 }
1310
David S. Miller1c92b4e2007-05-31 13:24:26 -07001311 unix_state_lock_nested(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001312
1313 if (sk->sk_state != st) {
David S. Miller1c92b4e2007-05-31 13:24:26 -07001314 unix_state_unlock(sk);
1315 unix_state_unlock(other);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316 sock_put(other);
1317 goto restart;
1318 }
1319
David S. Miller3610cda2011-01-05 15:38:53 -08001320 err = security_unix_stream_connect(sk, other, newsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321 if (err) {
David S. Miller1c92b4e2007-05-31 13:24:26 -07001322 unix_state_unlock(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001323 goto out_unlock;
1324 }
1325
1326 /* The way is open! Fastly set all the necessary fields... */
1327
1328 sock_hold(sk);
1329 unix_peer(newsk) = sk;
1330 newsk->sk_state = TCP_ESTABLISHED;
1331 newsk->sk_type = sk->sk_type;
Eric W. Biederman109f6e32010-06-13 03:30:14 +00001332 init_peercred(newsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333 newu = unix_sk(newsk);
Eric Dumazeteaefd112011-02-18 03:26:36 +00001334 RCU_INIT_POINTER(newsk->sk_wq, &newu->peer_wq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335 otheru = unix_sk(other);
1336
1337 /* copy address information from listening to new sock*/
1338 if (otheru->addr) {
1339 atomic_inc(&otheru->addr->refcnt);
1340 newu->addr = otheru->addr;
1341 }
Al Viro40ffe672012-03-14 21:54:32 -04001342 if (otheru->path.dentry) {
1343 path_get(&otheru->path);
1344 newu->path = otheru->path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345 }
1346
1347 /* Set credentials */
Eric W. Biederman109f6e32010-06-13 03:30:14 +00001348 copy_peercred(sk, other);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349
Linus Torvalds1da177e2005-04-16 15:20:36 -07001350 sock->state = SS_CONNECTED;
1351 sk->sk_state = TCP_ESTABLISHED;
Benjamin LaHaise830a1e52005-12-13 23:22:32 -08001352 sock_hold(newsk);
1353
Peter Zijlstra4e857c52014-03-17 18:06:10 +01001354 smp_mb__after_atomic(); /* sock_hold() does an atomic_inc() */
Benjamin LaHaise830a1e52005-12-13 23:22:32 -08001355 unix_peer(sk) = newsk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356
David S. Miller1c92b4e2007-05-31 13:24:26 -07001357 unix_state_unlock(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358
1359 /* take ten and and send info to listening sock */
1360 spin_lock(&other->sk_receive_queue.lock);
1361 __skb_queue_tail(&other->sk_receive_queue, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362 spin_unlock(&other->sk_receive_queue.lock);
David S. Miller1c92b4e2007-05-31 13:24:26 -07001363 unix_state_unlock(other);
David S. Miller676d2362014-04-11 16:15:36 -04001364 other->sk_data_ready(other);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365 sock_put(other);
1366 return 0;
1367
1368out_unlock:
1369 if (other)
David S. Miller1c92b4e2007-05-31 13:24:26 -07001370 unix_state_unlock(other);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371
1372out:
Wei Yongjun40d44442009-02-25 00:32:45 +00001373 kfree_skb(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374 if (newsk)
1375 unix_release_sock(newsk, 0);
1376 if (other)
1377 sock_put(other);
1378 return err;
1379}
1380
1381static int unix_socketpair(struct socket *socka, struct socket *sockb)
1382{
Jianjun Konge27dfce2008-11-01 21:38:31 -07001383 struct sock *ska = socka->sk, *skb = sockb->sk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384
1385 /* Join our sockets back to back */
1386 sock_hold(ska);
1387 sock_hold(skb);
Jianjun Konge27dfce2008-11-01 21:38:31 -07001388 unix_peer(ska) = skb;
1389 unix_peer(skb) = ska;
Eric W. Biederman109f6e32010-06-13 03:30:14 +00001390 init_peercred(ska);
1391 init_peercred(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392
1393 if (ska->sk_type != SOCK_DGRAM) {
1394 ska->sk_state = TCP_ESTABLISHED;
1395 skb->sk_state = TCP_ESTABLISHED;
1396 socka->state = SS_CONNECTED;
1397 sockb->state = SS_CONNECTED;
1398 }
1399 return 0;
1400}
1401
Daniel Borkmann90c6bd32013-10-17 22:51:31 +02001402static void unix_sock_inherit_flags(const struct socket *old,
1403 struct socket *new)
1404{
1405 if (test_bit(SOCK_PASSCRED, &old->flags))
1406 set_bit(SOCK_PASSCRED, &new->flags);
1407 if (test_bit(SOCK_PASSSEC, &old->flags))
1408 set_bit(SOCK_PASSSEC, &new->flags);
1409}
1410
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411static int unix_accept(struct socket *sock, struct socket *newsock, int flags)
1412{
1413 struct sock *sk = sock->sk;
1414 struct sock *tsk;
1415 struct sk_buff *skb;
1416 int err;
1417
1418 err = -EOPNOTSUPP;
Eric Dumazet6eba6a32008-11-16 22:58:44 -08001419 if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001420 goto out;
1421
1422 err = -EINVAL;
1423 if (sk->sk_state != TCP_LISTEN)
1424 goto out;
1425
1426 /* If socket state is TCP_LISTEN it cannot change (for now...),
1427 * so that no locks are necessary.
1428 */
1429
1430 skb = skb_recv_datagram(sk, 0, flags&O_NONBLOCK, &err);
1431 if (!skb) {
1432 /* This means receive shutdown. */
1433 if (err == 0)
1434 err = -EINVAL;
1435 goto out;
1436 }
1437
1438 tsk = skb->sk;
1439 skb_free_datagram(sk, skb);
1440 wake_up_interruptible(&unix_sk(sk)->peer_wait);
1441
1442 /* attach accepted sock to socket */
David S. Miller1c92b4e2007-05-31 13:24:26 -07001443 unix_state_lock(tsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001444 newsock->state = SS_CONNECTED;
Daniel Borkmann90c6bd32013-10-17 22:51:31 +02001445 unix_sock_inherit_flags(sock, newsock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446 sock_graft(tsk, newsock);
David S. Miller1c92b4e2007-05-31 13:24:26 -07001447 unix_state_unlock(tsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001448 return 0;
1449
1450out:
1451 return err;
1452}
1453
1454
1455static int unix_getname(struct socket *sock, struct sockaddr *uaddr, int *uaddr_len, int peer)
1456{
1457 struct sock *sk = sock->sk;
1458 struct unix_sock *u;
Cyrill Gorcunov13cfa972009-11-08 05:51:19 +00001459 DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr, uaddr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460 int err = 0;
1461
1462 if (peer) {
1463 sk = unix_peer_get(sk);
1464
1465 err = -ENOTCONN;
1466 if (!sk)
1467 goto out;
1468 err = 0;
1469 } else {
1470 sock_hold(sk);
1471 }
1472
1473 u = unix_sk(sk);
David S. Miller1c92b4e2007-05-31 13:24:26 -07001474 unix_state_lock(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475 if (!u->addr) {
1476 sunaddr->sun_family = AF_UNIX;
1477 sunaddr->sun_path[0] = 0;
1478 *uaddr_len = sizeof(short);
1479 } else {
1480 struct unix_address *addr = u->addr;
1481
1482 *uaddr_len = addr->len;
1483 memcpy(sunaddr, addr->name, *uaddr_len);
1484 }
David S. Miller1c92b4e2007-05-31 13:24:26 -07001485 unix_state_unlock(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486 sock_put(sk);
1487out:
1488 return err;
1489}
1490
1491static void unix_detach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1492{
1493 int i;
1494
1495 scm->fp = UNIXCB(skb).fp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496 UNIXCB(skb).fp = NULL;
1497
Eric Dumazet6eba6a32008-11-16 22:58:44 -08001498 for (i = scm->fp->count-1; i >= 0; i--)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499 unix_notinflight(scm->fp->fp[i]);
1500}
1501
Eric W. Biederman7361c362010-06-13 03:34:33 +00001502static void unix_destruct_scm(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503{
1504 struct scm_cookie scm;
1505 memset(&scm, 0, sizeof(scm));
Eric W. Biederman7361c362010-06-13 03:34:33 +00001506 scm.pid = UNIXCB(skb).pid;
Eric W. Biederman7361c362010-06-13 03:34:33 +00001507 if (UNIXCB(skb).fp)
1508 unix_detach_fds(&scm, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509
1510 /* Alas, it calls VFS */
1511 /* So fscking what? fput() had been SMP-safe since the last Summer */
1512 scm_destroy(&scm);
1513 sock_wfree(skb);
1514}
1515
willy tarreau712f4aa2016-01-10 07:54:56 +01001516/*
1517 * The "user->unix_inflight" variable is protected by the garbage
1518 * collection lock, and we just read it locklessly here. If you go
1519 * over the limit, there might be a tiny race in actually noticing
1520 * it across threads. Tough.
1521 */
1522static inline bool too_many_unix_fds(struct task_struct *p)
1523{
1524 struct user_struct *user = current_user();
1525
1526 if (unlikely(user->unix_inflight > task_rlimit(p, RLIMIT_NOFILE)))
1527 return !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN);
1528 return false;
1529}
1530
Eric Dumazet25888e32010-11-25 04:11:39 +00001531#define MAX_RECURSION_LEVEL 4
1532
Miklos Szeredi62093442008-11-09 15:23:57 +01001533static int unix_attach_fds(struct scm_cookie *scm, struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001534{
1535 int i;
Eric Dumazet25888e32010-11-25 04:11:39 +00001536 unsigned char max_level = 0;
1537 int unix_sock_count = 0;
1538
willy tarreau712f4aa2016-01-10 07:54:56 +01001539 if (too_many_unix_fds(current))
1540 return -ETOOMANYREFS;
1541
Eric Dumazet25888e32010-11-25 04:11:39 +00001542 for (i = scm->fp->count - 1; i >= 0; i--) {
1543 struct sock *sk = unix_get_socket(scm->fp->fp[i]);
1544
1545 if (sk) {
1546 unix_sock_count++;
1547 max_level = max(max_level,
1548 unix_sk(sk)->recursion_level);
1549 }
1550 }
1551 if (unlikely(max_level > MAX_RECURSION_LEVEL))
1552 return -ETOOMANYREFS;
Miklos Szeredi62093442008-11-09 15:23:57 +01001553
1554 /*
1555 * Need to duplicate file references for the sake of garbage
1556 * collection. Otherwise a socket in the fps might become a
1557 * candidate for GC while the skb is not yet queued.
1558 */
1559 UNIXCB(skb).fp = scm_fp_dup(scm->fp);
1560 if (!UNIXCB(skb).fp)
1561 return -ENOMEM;
1562
willy tarreau712f4aa2016-01-10 07:54:56 +01001563 for (i = scm->fp->count - 1; i >= 0; i--)
1564 unix_inflight(scm->fp->fp[i]);
Eric Dumazet25888e32010-11-25 04:11:39 +00001565 return max_level;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566}
1567
David S. Millerf78a5fd2011-09-16 19:34:00 -04001568static int unix_scm_to_skb(struct scm_cookie *scm, struct sk_buff *skb, bool send_fds)
Eric W. Biederman7361c362010-06-13 03:34:33 +00001569{
1570 int err = 0;
Eric Dumazet16e57262011-09-19 05:52:27 +00001571
David S. Millerf78a5fd2011-09-16 19:34:00 -04001572 UNIXCB(skb).pid = get_pid(scm->pid);
Eric W. Biederman6b0ee8c02013-04-03 17:28:16 +00001573 UNIXCB(skb).uid = scm->creds.uid;
1574 UNIXCB(skb).gid = scm->creds.gid;
Eric W. Biederman7361c362010-06-13 03:34:33 +00001575 UNIXCB(skb).fp = NULL;
Stephen Smalley37a9a8d2015-06-10 08:44:59 -04001576 unix_get_secdata(scm, skb);
Eric W. Biederman7361c362010-06-13 03:34:33 +00001577 if (scm->fp && send_fds)
1578 err = unix_attach_fds(scm, skb);
1579
1580 skb->destructor = unix_destruct_scm;
1581 return err;
1582}
1583
Hannes Frederic Sowa9490f882015-11-26 12:08:18 +01001584static bool unix_passcred_enabled(const struct socket *sock,
1585 const struct sock *other)
1586{
1587 return test_bit(SOCK_PASSCRED, &sock->flags) ||
1588 !other->sk_socket ||
1589 test_bit(SOCK_PASSCRED, &other->sk_socket->flags);
1590}
1591
Linus Torvalds1da177e2005-04-16 15:20:36 -07001592/*
Eric Dumazet16e57262011-09-19 05:52:27 +00001593 * Some apps rely on write() giving SCM_CREDENTIALS
1594 * We include credentials if source or destination socket
1595 * asserted SOCK_PASSCRED.
1596 */
1597static void maybe_add_creds(struct sk_buff *skb, const struct socket *sock,
1598 const struct sock *other)
1599{
Eric W. Biederman6b0ee8c02013-04-03 17:28:16 +00001600 if (UNIXCB(skb).pid)
Eric Dumazet16e57262011-09-19 05:52:27 +00001601 return;
Hannes Frederic Sowa9490f882015-11-26 12:08:18 +01001602 if (unix_passcred_enabled(sock, other)) {
Eric Dumazet16e57262011-09-19 05:52:27 +00001603 UNIXCB(skb).pid = get_pid(task_tgid(current));
David S. Miller6e0895c2013-04-22 20:32:51 -04001604 current_uid_gid(&UNIXCB(skb).uid, &UNIXCB(skb).gid);
Eric Dumazet16e57262011-09-19 05:52:27 +00001605 }
1606}
1607
Hannes Frederic Sowa9490f882015-11-26 12:08:18 +01001608static int maybe_init_creds(struct scm_cookie *scm,
1609 struct socket *socket,
1610 const struct sock *other)
1611{
1612 int err;
1613 struct msghdr msg = { .msg_controllen = 0 };
1614
1615 err = scm_send(socket, &msg, scm, false);
1616 if (err)
1617 return err;
1618
1619 if (unix_passcred_enabled(socket, other)) {
1620 scm->pid = get_pid(task_tgid(current));
1621 current_uid_gid(&scm->creds.uid, &scm->creds.gid);
1622 }
1623 return err;
1624}
1625
1626static bool unix_skb_scm_eq(struct sk_buff *skb,
1627 struct scm_cookie *scm)
1628{
1629 const struct unix_skb_parms *u = &UNIXCB(skb);
1630
1631 return u->pid == scm->pid &&
1632 uid_eq(u->uid, scm->creds.uid) &&
1633 gid_eq(u->gid, scm->creds.gid) &&
1634 unix_secdata_eq(scm, skb);
1635}
1636
Eric Dumazet16e57262011-09-19 05:52:27 +00001637/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001638 * Send AF_UNIX data.
1639 */
1640
Ying Xue1b784142015-03-02 15:37:48 +08001641static int unix_dgram_sendmsg(struct socket *sock, struct msghdr *msg,
1642 size_t len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001643{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001644 struct sock *sk = sock->sk;
YOSHIFUJI Hideaki3b1e0a62008-03-26 02:26:21 +09001645 struct net *net = sock_net(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001646 struct unix_sock *u = unix_sk(sk);
Steffen Hurrle342dfc32014-01-17 22:53:15 +01001647 DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr, msg->msg_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001648 struct sock *other = NULL;
1649 int namelen = 0; /* fake GCC */
1650 int err;
Eric Dumazet95c96172012-04-15 05:58:06 +00001651 unsigned int hash;
David S. Millerf78a5fd2011-09-16 19:34:00 -04001652 struct sk_buff *skb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001653 long timeo;
Christoph Hellwig7cc05662015-01-28 18:04:53 +01001654 struct scm_cookie scm;
Eric Dumazet25888e32010-11-25 04:11:39 +00001655 int max_level;
Eric Dumazeteb6a2482012-04-03 05:28:28 +00001656 int data_len = 0;
Rainer Weikusat7d267272015-11-20 22:07:23 +00001657 int sk_locked;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658
dann frazier5f23b732008-11-26 15:32:27 -08001659 wait_for_unix_gc();
Christoph Hellwig7cc05662015-01-28 18:04:53 +01001660 err = scm_send(sock, msg, &scm, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661 if (err < 0)
1662 return err;
1663
1664 err = -EOPNOTSUPP;
1665 if (msg->msg_flags&MSG_OOB)
1666 goto out;
1667
1668 if (msg->msg_namelen) {
1669 err = unix_mkname(sunaddr, msg->msg_namelen, &hash);
1670 if (err < 0)
1671 goto out;
1672 namelen = err;
1673 } else {
1674 sunaddr = NULL;
1675 err = -ENOTCONN;
1676 other = unix_peer_get(sk);
1677 if (!other)
1678 goto out;
1679 }
1680
Joe Perchesf64f9e72009-11-29 16:55:45 -08001681 if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr
1682 && (err = unix_autobind(sock)) != 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001683 goto out;
1684
1685 err = -EMSGSIZE;
1686 if (len > sk->sk_sndbuf - 32)
1687 goto out;
1688
Kirill Tkhai31ff6aa2014-05-15 19:56:28 +04001689 if (len > SKB_MAX_ALLOC) {
Eric Dumazeteb6a2482012-04-03 05:28:28 +00001690 data_len = min_t(size_t,
1691 len - SKB_MAX_ALLOC,
1692 MAX_SKB_FRAGS * PAGE_SIZE);
Kirill Tkhai31ff6aa2014-05-15 19:56:28 +04001693 data_len = PAGE_ALIGN(data_len);
1694
1695 BUILD_BUG_ON(SKB_MAX_ALLOC < PAGE_SIZE);
1696 }
Eric Dumazeteb6a2482012-04-03 05:28:28 +00001697
1698 skb = sock_alloc_send_pskb(sk, len - data_len, data_len,
Eric Dumazet28d64272013-08-08 14:38:47 -07001699 msg->msg_flags & MSG_DONTWAIT, &err,
1700 PAGE_ALLOC_COSTLY_ORDER);
Jianjun Konge27dfce2008-11-01 21:38:31 -07001701 if (skb == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001702 goto out;
1703
Christoph Hellwig7cc05662015-01-28 18:04:53 +01001704 err = unix_scm_to_skb(&scm, skb, true);
Eric Dumazet25888e32010-11-25 04:11:39 +00001705 if (err < 0)
Eric W. Biederman7361c362010-06-13 03:34:33 +00001706 goto out_free;
Eric Dumazet25888e32010-11-25 04:11:39 +00001707 max_level = err + 1;
Catherine Zhang877ce7c2006-06-29 12:27:47 -07001708
Eric Dumazeteb6a2482012-04-03 05:28:28 +00001709 skb_put(skb, len - data_len);
1710 skb->data_len = data_len;
1711 skb->len = len;
Al Viroc0371da2014-11-24 10:42:55 -05001712 err = skb_copy_datagram_from_iter(skb, 0, &msg->msg_iter, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001713 if (err)
1714 goto out_free;
1715
1716 timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1717
1718restart:
1719 if (!other) {
1720 err = -ECONNRESET;
1721 if (sunaddr == NULL)
1722 goto out_free;
1723
Denis V. Lunev097e66c2007-11-19 22:29:30 -08001724 other = unix_find_other(net, sunaddr, namelen, sk->sk_type,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001725 hash, &err);
Jianjun Konge27dfce2008-11-01 21:38:31 -07001726 if (other == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001727 goto out_free;
1728 }
1729
Alban Crequyd6ae3ba2011-01-18 06:39:15 +00001730 if (sk_filter(other, skb) < 0) {
1731 /* Toss the packet but do not return any error to the sender */
1732 err = len;
1733 goto out_free;
1734 }
1735
Rainer Weikusat7d267272015-11-20 22:07:23 +00001736 sk_locked = 0;
David S. Miller1c92b4e2007-05-31 13:24:26 -07001737 unix_state_lock(other);
Rainer Weikusat7d267272015-11-20 22:07:23 +00001738restart_locked:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001739 err = -EPERM;
1740 if (!unix_may_send(sk, other))
1741 goto out_unlock;
1742
Rainer Weikusat7d267272015-11-20 22:07:23 +00001743 if (unlikely(sock_flag(other, SOCK_DEAD))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001744 /*
1745 * Check with 1003.1g - what should
1746 * datagram error
1747 */
David S. Miller1c92b4e2007-05-31 13:24:26 -07001748 unix_state_unlock(other);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001749 sock_put(other);
1750
Rainer Weikusat7d267272015-11-20 22:07:23 +00001751 if (!sk_locked)
1752 unix_state_lock(sk);
1753
Linus Torvalds1da177e2005-04-16 15:20:36 -07001754 err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001755 if (unix_peer(sk) == other) {
Jianjun Konge27dfce2008-11-01 21:38:31 -07001756 unix_peer(sk) = NULL;
Rainer Weikusat7d267272015-11-20 22:07:23 +00001757 unix_dgram_peer_wake_disconnect_wakeup(sk, other);
1758
David S. Miller1c92b4e2007-05-31 13:24:26 -07001759 unix_state_unlock(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001760
1761 unix_dgram_disconnected(sk, other);
1762 sock_put(other);
1763 err = -ECONNREFUSED;
1764 } else {
David S. Miller1c92b4e2007-05-31 13:24:26 -07001765 unix_state_unlock(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766 }
1767
1768 other = NULL;
1769 if (err)
1770 goto out_free;
1771 goto restart;
1772 }
1773
1774 err = -EPIPE;
1775 if (other->sk_shutdown & RCV_SHUTDOWN)
1776 goto out_unlock;
1777
1778 if (sk->sk_type != SOCK_SEQPACKET) {
1779 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1780 if (err)
1781 goto out_unlock;
1782 }
1783
Rainer Weikusat7d267272015-11-20 22:07:23 +00001784 if (unlikely(unix_peer(other) != sk && unix_recvq_full(other))) {
1785 if (timeo) {
1786 timeo = unix_wait_for_peer(other, timeo);
1787
1788 err = sock_intr_errno(timeo);
1789 if (signal_pending(current))
1790 goto out_free;
1791
1792 goto restart;
1793 }
1794
1795 if (!sk_locked) {
1796 unix_state_unlock(other);
1797 unix_state_double_lock(sk, other);
1798 }
1799
1800 if (unix_peer(sk) != other ||
1801 unix_dgram_peer_wake_me(sk, other)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001802 err = -EAGAIN;
Rainer Weikusat7d267272015-11-20 22:07:23 +00001803 sk_locked = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001804 goto out_unlock;
1805 }
1806
Rainer Weikusat7d267272015-11-20 22:07:23 +00001807 if (!sk_locked) {
1808 sk_locked = 1;
1809 goto restart_locked;
1810 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001811 }
1812
Rainer Weikusat7d267272015-11-20 22:07:23 +00001813 if (unlikely(sk_locked))
1814 unix_state_unlock(sk);
1815
Alban Crequy3f661162010-10-04 08:48:28 +00001816 if (sock_flag(other, SOCK_RCVTSTAMP))
1817 __net_timestamp(skb);
Eric Dumazet16e57262011-09-19 05:52:27 +00001818 maybe_add_creds(skb, sock, other);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001819 skb_queue_tail(&other->sk_receive_queue, skb);
Eric Dumazet25888e32010-11-25 04:11:39 +00001820 if (max_level > unix_sk(other)->recursion_level)
1821 unix_sk(other)->recursion_level = max_level;
David S. Miller1c92b4e2007-05-31 13:24:26 -07001822 unix_state_unlock(other);
David S. Miller676d2362014-04-11 16:15:36 -04001823 other->sk_data_ready(other);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001824 sock_put(other);
Christoph Hellwig7cc05662015-01-28 18:04:53 +01001825 scm_destroy(&scm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001826 return len;
1827
1828out_unlock:
Rainer Weikusat7d267272015-11-20 22:07:23 +00001829 if (sk_locked)
1830 unix_state_unlock(sk);
David S. Miller1c92b4e2007-05-31 13:24:26 -07001831 unix_state_unlock(other);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001832out_free:
1833 kfree_skb(skb);
1834out:
1835 if (other)
1836 sock_put(other);
Christoph Hellwig7cc05662015-01-28 18:04:53 +01001837 scm_destroy(&scm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001838 return err;
1839}
1840
Eric Dumazete370a722013-08-08 14:37:32 -07001841/* We use paged skbs for stream sockets, and limit occupancy to 32768
1842 * bytes, and a minimun of a full page.
1843 */
1844#define UNIX_SKB_FRAGS_SZ (PAGE_SIZE << get_order(32768))
YOSHIFUJI Hideakiac7bfa62007-02-09 23:25:23 +09001845
Ying Xue1b784142015-03-02 15:37:48 +08001846static int unix_stream_sendmsg(struct socket *sock, struct msghdr *msg,
1847 size_t len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001848{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001849 struct sock *sk = sock->sk;
1850 struct sock *other = NULL;
Eric Dumazet6eba6a32008-11-16 22:58:44 -08001851 int err, size;
David S. Millerf78a5fd2011-09-16 19:34:00 -04001852 struct sk_buff *skb;
Jianjun Konge27dfce2008-11-01 21:38:31 -07001853 int sent = 0;
Christoph Hellwig7cc05662015-01-28 18:04:53 +01001854 struct scm_cookie scm;
Miklos Szeredi8ba69ba2009-09-11 11:31:45 -07001855 bool fds_sent = false;
Eric Dumazet25888e32010-11-25 04:11:39 +00001856 int max_level;
Eric Dumazete370a722013-08-08 14:37:32 -07001857 int data_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001858
dann frazier5f23b732008-11-26 15:32:27 -08001859 wait_for_unix_gc();
Christoph Hellwig7cc05662015-01-28 18:04:53 +01001860 err = scm_send(sock, msg, &scm, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001861 if (err < 0)
1862 return err;
1863
1864 err = -EOPNOTSUPP;
1865 if (msg->msg_flags&MSG_OOB)
1866 goto out_err;
1867
1868 if (msg->msg_namelen) {
1869 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
1870 goto out_err;
1871 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001872 err = -ENOTCONN;
Benjamin LaHaise830a1e52005-12-13 23:22:32 -08001873 other = unix_peer(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001874 if (!other)
1875 goto out_err;
1876 }
1877
1878 if (sk->sk_shutdown & SEND_SHUTDOWN)
1879 goto pipe_err;
1880
Eric Dumazet6eba6a32008-11-16 22:58:44 -08001881 while (sent < len) {
Eric Dumazete370a722013-08-08 14:37:32 -07001882 size = len - sent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001883
1884 /* Keep two messages in the pipe so it schedules better */
Eric Dumazete370a722013-08-08 14:37:32 -07001885 size = min_t(int, size, (sk->sk_sndbuf >> 1) - 64);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001886
Eric Dumazete370a722013-08-08 14:37:32 -07001887 /* allow fallback to order-0 allocations */
1888 size = min_t(int, size, SKB_MAX_HEAD(0) + UNIX_SKB_FRAGS_SZ);
YOSHIFUJI Hideakiac7bfa62007-02-09 23:25:23 +09001889
Eric Dumazete370a722013-08-08 14:37:32 -07001890 data_len = max_t(int, 0, size - SKB_MAX_HEAD(0));
YOSHIFUJI Hideakiac7bfa62007-02-09 23:25:23 +09001891
Kirill Tkhai31ff6aa2014-05-15 19:56:28 +04001892 data_len = min_t(size_t, size, PAGE_ALIGN(data_len));
1893
Eric Dumazete370a722013-08-08 14:37:32 -07001894 skb = sock_alloc_send_pskb(sk, size - data_len, data_len,
Eric Dumazet28d64272013-08-08 14:38:47 -07001895 msg->msg_flags & MSG_DONTWAIT, &err,
1896 get_order(UNIX_SKB_FRAGS_SZ));
Eric Dumazete370a722013-08-08 14:37:32 -07001897 if (!skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001898 goto out_err;
1899
David S. Millerf78a5fd2011-09-16 19:34:00 -04001900 /* Only send the fds in the first buffer */
Christoph Hellwig7cc05662015-01-28 18:04:53 +01001901 err = unix_scm_to_skb(&scm, skb, !fds_sent);
Eric Dumazet25888e32010-11-25 04:11:39 +00001902 if (err < 0) {
Eric W. Biederman7361c362010-06-13 03:34:33 +00001903 kfree_skb(skb);
David S. Millerf78a5fd2011-09-16 19:34:00 -04001904 goto out_err;
Miklos Szeredi62093442008-11-09 15:23:57 +01001905 }
Eric Dumazet25888e32010-11-25 04:11:39 +00001906 max_level = err + 1;
Eric W. Biederman7361c362010-06-13 03:34:33 +00001907 fds_sent = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001908
Eric Dumazete370a722013-08-08 14:37:32 -07001909 skb_put(skb, size - data_len);
1910 skb->data_len = data_len;
1911 skb->len = size;
Al Viroc0371da2014-11-24 10:42:55 -05001912 err = skb_copy_datagram_from_iter(skb, 0, &msg->msg_iter, size);
Eric Dumazet6eba6a32008-11-16 22:58:44 -08001913 if (err) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001914 kfree_skb(skb);
David S. Millerf78a5fd2011-09-16 19:34:00 -04001915 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001916 }
1917
David S. Miller1c92b4e2007-05-31 13:24:26 -07001918 unix_state_lock(other);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001919
1920 if (sock_flag(other, SOCK_DEAD) ||
1921 (other->sk_shutdown & RCV_SHUTDOWN))
1922 goto pipe_err_free;
1923
Eric Dumazet16e57262011-09-19 05:52:27 +00001924 maybe_add_creds(skb, sock, other);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001925 skb_queue_tail(&other->sk_receive_queue, skb);
Eric Dumazet25888e32010-11-25 04:11:39 +00001926 if (max_level > unix_sk(other)->recursion_level)
1927 unix_sk(other)->recursion_level = max_level;
David S. Miller1c92b4e2007-05-31 13:24:26 -07001928 unix_state_unlock(other);
David S. Miller676d2362014-04-11 16:15:36 -04001929 other->sk_data_ready(other);
Jianjun Konge27dfce2008-11-01 21:38:31 -07001930 sent += size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001931 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001932
Christoph Hellwig7cc05662015-01-28 18:04:53 +01001933 scm_destroy(&scm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001934
1935 return sent;
1936
1937pipe_err_free:
David S. Miller1c92b4e2007-05-31 13:24:26 -07001938 unix_state_unlock(other);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001939 kfree_skb(skb);
1940pipe_err:
Eric Dumazet6eba6a32008-11-16 22:58:44 -08001941 if (sent == 0 && !(msg->msg_flags&MSG_NOSIGNAL))
1942 send_sig(SIGPIPE, current, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001943 err = -EPIPE;
1944out_err:
Christoph Hellwig7cc05662015-01-28 18:04:53 +01001945 scm_destroy(&scm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001946 return sent ? : err;
1947}
1948
Hannes Frederic Sowa869e7c62015-05-21 16:59:59 +02001949static ssize_t unix_stream_sendpage(struct socket *socket, struct page *page,
1950 int offset, size_t size, int flags)
1951{
Hannes Frederic Sowa9490f882015-11-26 12:08:18 +01001952 int err;
1953 bool send_sigpipe = false;
1954 bool init_scm = true;
1955 struct scm_cookie scm;
Hannes Frederic Sowa869e7c62015-05-21 16:59:59 +02001956 struct sock *other, *sk = socket->sk;
1957 struct sk_buff *skb, *newskb = NULL, *tail = NULL;
1958
1959 if (flags & MSG_OOB)
1960 return -EOPNOTSUPP;
1961
1962 other = unix_peer(sk);
1963 if (!other || sk->sk_state != TCP_ESTABLISHED)
1964 return -ENOTCONN;
1965
1966 if (false) {
1967alloc_skb:
1968 unix_state_unlock(other);
1969 mutex_unlock(&unix_sk(other)->readlock);
1970 newskb = sock_alloc_send_pskb(sk, 0, 0, flags & MSG_DONTWAIT,
1971 &err, 0);
1972 if (!newskb)
Hannes Frederic Sowa9490f882015-11-26 12:08:18 +01001973 goto err;
Hannes Frederic Sowa869e7c62015-05-21 16:59:59 +02001974 }
1975
1976 /* we must acquire readlock as we modify already present
1977 * skbs in the sk_receive_queue and mess with skb->len
1978 */
1979 err = mutex_lock_interruptible(&unix_sk(other)->readlock);
1980 if (err) {
1981 err = flags & MSG_DONTWAIT ? -EAGAIN : -ERESTARTSYS;
Hannes Frederic Sowa869e7c62015-05-21 16:59:59 +02001982 goto err;
1983 }
1984
1985 if (sk->sk_shutdown & SEND_SHUTDOWN) {
1986 err = -EPIPE;
Hannes Frederic Sowa9490f882015-11-26 12:08:18 +01001987 send_sigpipe = true;
Hannes Frederic Sowa869e7c62015-05-21 16:59:59 +02001988 goto err_unlock;
1989 }
1990
1991 unix_state_lock(other);
1992
1993 if (sock_flag(other, SOCK_DEAD) ||
1994 other->sk_shutdown & RCV_SHUTDOWN) {
1995 err = -EPIPE;
Hannes Frederic Sowa9490f882015-11-26 12:08:18 +01001996 send_sigpipe = true;
Hannes Frederic Sowa869e7c62015-05-21 16:59:59 +02001997 goto err_state_unlock;
1998 }
1999
Hannes Frederic Sowa9490f882015-11-26 12:08:18 +01002000 if (init_scm) {
2001 err = maybe_init_creds(&scm, socket, other);
2002 if (err)
2003 goto err_state_unlock;
2004 init_scm = false;
2005 }
2006
Hannes Frederic Sowa869e7c62015-05-21 16:59:59 +02002007 skb = skb_peek_tail(&other->sk_receive_queue);
2008 if (tail && tail == skb) {
2009 skb = newskb;
Hannes Frederic Sowa9490f882015-11-26 12:08:18 +01002010 } else if (!skb || !unix_skb_scm_eq(skb, &scm)) {
2011 if (newskb) {
Hannes Frederic Sowa869e7c62015-05-21 16:59:59 +02002012 skb = newskb;
Hannes Frederic Sowa9490f882015-11-26 12:08:18 +01002013 } else {
2014 tail = skb;
Hannes Frederic Sowa869e7c62015-05-21 16:59:59 +02002015 goto alloc_skb;
Hannes Frederic Sowa9490f882015-11-26 12:08:18 +01002016 }
Hannes Frederic Sowa869e7c62015-05-21 16:59:59 +02002017 } else if (newskb) {
2018 /* this is fast path, we don't necessarily need to
2019 * call to kfree_skb even though with newskb == NULL
2020 * this - does no harm
2021 */
2022 consume_skb(newskb);
Hannes Frederic Sowa8844f972015-11-16 16:25:56 +01002023 newskb = NULL;
Hannes Frederic Sowa869e7c62015-05-21 16:59:59 +02002024 }
2025
2026 if (skb_append_pagefrags(skb, page, offset, size)) {
2027 tail = skb;
2028 goto alloc_skb;
2029 }
2030
2031 skb->len += size;
2032 skb->data_len += size;
2033 skb->truesize += size;
2034 atomic_add(size, &sk->sk_wmem_alloc);
2035
Hannes Frederic Sowaa3a116e2015-11-17 15:10:59 +01002036 if (newskb) {
Hannes Frederic Sowa9490f882015-11-26 12:08:18 +01002037 err = unix_scm_to_skb(&scm, skb, false);
2038 if (err)
2039 goto err_state_unlock;
Hannes Frederic Sowaa3a116e2015-11-17 15:10:59 +01002040 spin_lock(&other->sk_receive_queue.lock);
Hannes Frederic Sowa869e7c62015-05-21 16:59:59 +02002041 __skb_queue_tail(&other->sk_receive_queue, newskb);
Hannes Frederic Sowaa3a116e2015-11-17 15:10:59 +01002042 spin_unlock(&other->sk_receive_queue.lock);
2043 }
Hannes Frederic Sowa869e7c62015-05-21 16:59:59 +02002044
2045 unix_state_unlock(other);
2046 mutex_unlock(&unix_sk(other)->readlock);
2047
2048 other->sk_data_ready(other);
Hannes Frederic Sowa9490f882015-11-26 12:08:18 +01002049 scm_destroy(&scm);
Hannes Frederic Sowa869e7c62015-05-21 16:59:59 +02002050 return size;
2051
2052err_state_unlock:
2053 unix_state_unlock(other);
2054err_unlock:
2055 mutex_unlock(&unix_sk(other)->readlock);
2056err:
2057 kfree_skb(newskb);
2058 if (send_sigpipe && !(flags & MSG_NOSIGNAL))
2059 send_sig(SIGPIPE, current, 0);
Hannes Frederic Sowa9490f882015-11-26 12:08:18 +01002060 if (!init_scm)
2061 scm_destroy(&scm);
Hannes Frederic Sowa869e7c62015-05-21 16:59:59 +02002062 return err;
2063}
2064
Ying Xue1b784142015-03-02 15:37:48 +08002065static int unix_seqpacket_sendmsg(struct socket *sock, struct msghdr *msg,
2066 size_t len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002067{
2068 int err;
2069 struct sock *sk = sock->sk;
YOSHIFUJI Hideakiac7bfa62007-02-09 23:25:23 +09002070
Linus Torvalds1da177e2005-04-16 15:20:36 -07002071 err = sock_error(sk);
2072 if (err)
2073 return err;
2074
2075 if (sk->sk_state != TCP_ESTABLISHED)
2076 return -ENOTCONN;
2077
2078 if (msg->msg_namelen)
2079 msg->msg_namelen = 0;
2080
Ying Xue1b784142015-03-02 15:37:48 +08002081 return unix_dgram_sendmsg(sock, msg, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002082}
YOSHIFUJI Hideakiac7bfa62007-02-09 23:25:23 +09002083
Ying Xue1b784142015-03-02 15:37:48 +08002084static int unix_seqpacket_recvmsg(struct socket *sock, struct msghdr *msg,
2085 size_t size, int flags)
Eric W. Biedermana05d2ad2011-04-24 01:54:57 +00002086{
2087 struct sock *sk = sock->sk;
2088
2089 if (sk->sk_state != TCP_ESTABLISHED)
2090 return -ENOTCONN;
2091
Ying Xue1b784142015-03-02 15:37:48 +08002092 return unix_dgram_recvmsg(sock, msg, size, flags);
Eric W. Biedermana05d2ad2011-04-24 01:54:57 +00002093}
2094
Linus Torvalds1da177e2005-04-16 15:20:36 -07002095static void unix_copy_addr(struct msghdr *msg, struct sock *sk)
2096{
2097 struct unix_sock *u = unix_sk(sk);
2098
Linus Torvalds1da177e2005-04-16 15:20:36 -07002099 if (u->addr) {
2100 msg->msg_namelen = u->addr->len;
2101 memcpy(msg->msg_name, u->addr->name, u->addr->len);
2102 }
2103}
2104
Ying Xue1b784142015-03-02 15:37:48 +08002105static int unix_dgram_recvmsg(struct socket *sock, struct msghdr *msg,
2106 size_t size, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002107{
Christoph Hellwig7cc05662015-01-28 18:04:53 +01002108 struct scm_cookie scm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002109 struct sock *sk = sock->sk;
2110 struct unix_sock *u = unix_sk(sk);
Rainer Weikusat64874282015-12-06 21:11:38 +00002111 struct sk_buff *skb, *last;
2112 long timeo;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002113 int err;
Pavel Emelyanovf55bb7f2012-02-21 07:31:51 +00002114 int peeked, skip;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002115
2116 err = -EOPNOTSUPP;
2117 if (flags&MSG_OOB)
2118 goto out;
2119
Rainer Weikusat64874282015-12-06 21:11:38 +00002120 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002121
Rainer Weikusat64874282015-12-06 21:11:38 +00002122 do {
2123 mutex_lock(&u->readlock);
Pavel Emelyanovf55bb7f2012-02-21 07:31:51 +00002124
Rainer Weikusat64874282015-12-06 21:11:38 +00002125 skip = sk_peek_offset(sk, flags);
2126 skb = __skb_try_recv_datagram(sk, flags, &peeked, &skip, &err,
2127 &last);
2128 if (skb)
2129 break;
2130
2131 mutex_unlock(&u->readlock);
2132
2133 if (err != -EAGAIN)
2134 break;
2135 } while (timeo &&
2136 !__skb_wait_for_more_packets(sk, &err, &timeo, last));
2137
2138 if (!skb) { /* implies readlock unlocked */
Florian Zumbiehl0a112252007-11-29 23:19:23 +11002139 unix_state_lock(sk);
2140 /* Signal EOF on disconnected non-blocking SEQPACKET socket. */
2141 if (sk->sk_type == SOCK_SEQPACKET && err == -EAGAIN &&
2142 (sk->sk_shutdown & RCV_SHUTDOWN))
2143 err = 0;
2144 unix_state_unlock(sk);
Rainer Weikusat64874282015-12-06 21:11:38 +00002145 goto out;
Florian Zumbiehl0a112252007-11-29 23:19:23 +11002146 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002147
Rainer Weikusat77b75f42015-11-26 19:23:15 +00002148 if (wq_has_sleeper(&u->peer_wait))
2149 wake_up_interruptible_sync_poll(&u->peer_wait,
2150 POLLOUT | POLLWRNORM |
2151 POLLWRBAND);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002152
2153 if (msg->msg_name)
2154 unix_copy_addr(msg, skb->sk);
2155
Pavel Emelyanovf55bb7f2012-02-21 07:31:51 +00002156 if (size > skb->len - skip)
2157 size = skb->len - skip;
2158 else if (size < skb->len - skip)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002159 msg->msg_flags |= MSG_TRUNC;
2160
David S. Miller51f3d022014-11-05 16:46:40 -05002161 err = skb_copy_datagram_msg(skb, skip, msg, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002162 if (err)
2163 goto out_free;
2164
Alban Crequy3f661162010-10-04 08:48:28 +00002165 if (sock_flag(sk, SOCK_RCVTSTAMP))
2166 __sock_recv_timestamp(msg, sk, skb);
2167
Christoph Hellwig7cc05662015-01-28 18:04:53 +01002168 memset(&scm, 0, sizeof(scm));
2169
2170 scm_set_cred(&scm, UNIXCB(skb).pid, UNIXCB(skb).uid, UNIXCB(skb).gid);
2171 unix_set_secdata(&scm, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002172
Eric Dumazet6eba6a32008-11-16 22:58:44 -08002173 if (!(flags & MSG_PEEK)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002174 if (UNIXCB(skb).fp)
Christoph Hellwig7cc05662015-01-28 18:04:53 +01002175 unix_detach_fds(&scm, skb);
Pavel Emelyanovf55bb7f2012-02-21 07:31:51 +00002176
2177 sk_peek_offset_bwd(sk, skb->len);
Eric Dumazet6eba6a32008-11-16 22:58:44 -08002178 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002179 /* It is questionable: on PEEK we could:
2180 - do not return fds - good, but too simple 8)
2181 - return fds, and do not return them on read (old strategy,
2182 apparently wrong)
2183 - clone fds (I chose it for now, it is the most universal
2184 solution)
YOSHIFUJI Hideakiac7bfa62007-02-09 23:25:23 +09002185
2186 POSIX 1003.1g does not actually define this clearly
2187 at all. POSIX 1003.1g doesn't define a lot of things
2188 clearly however!
2189
Linus Torvalds1da177e2005-04-16 15:20:36 -07002190 */
Pavel Emelyanovf55bb7f2012-02-21 07:31:51 +00002191
2192 sk_peek_offset_fwd(sk, size);
2193
Linus Torvalds1da177e2005-04-16 15:20:36 -07002194 if (UNIXCB(skb).fp)
Christoph Hellwig7cc05662015-01-28 18:04:53 +01002195 scm.fp = scm_fp_dup(UNIXCB(skb).fp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002196 }
Eric Dumazet9f6f9af2012-02-21 23:24:55 +00002197 err = (flags & MSG_TRUNC) ? skb->len - skip : size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002198
Christoph Hellwig7cc05662015-01-28 18:04:53 +01002199 scm_recv(sock, msg, &scm, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002200
2201out_free:
Eric Dumazet6eba6a32008-11-16 22:58:44 -08002202 skb_free_datagram(sk, skb);
Ingo Molnar57b47a52006-03-20 22:35:41 -08002203 mutex_unlock(&u->readlock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002204out:
2205 return err;
2206}
2207
2208/*
Benjamin Poirier79f632c2013-04-29 11:42:14 +00002209 * Sleep until more data has arrived. But check for races..
Linus Torvalds1da177e2005-04-16 15:20:36 -07002210 */
Benjamin Poirier79f632c2013-04-29 11:42:14 +00002211static long unix_stream_data_wait(struct sock *sk, long timeo,
Hannes Frederic Sowa2b514572015-05-21 17:00:01 +02002212 struct sk_buff *last, unsigned int last_len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002213{
Hannes Frederic Sowa2b514572015-05-21 17:00:01 +02002214 struct sk_buff *tail;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002215 DEFINE_WAIT(wait);
2216
David S. Miller1c92b4e2007-05-31 13:24:26 -07002217 unix_state_lock(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002218
2219 for (;;) {
Eric Dumazetaa395142010-04-20 13:03:51 +00002220 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002221
Hannes Frederic Sowa2b514572015-05-21 17:00:01 +02002222 tail = skb_peek_tail(&sk->sk_receive_queue);
2223 if (tail != last ||
2224 (tail && tail->len != last_len) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07002225 sk->sk_err ||
2226 (sk->sk_shutdown & RCV_SHUTDOWN) ||
2227 signal_pending(current) ||
2228 !timeo)
2229 break;
2230
Eric Dumazet9cd3e072015-11-29 20:03:10 -08002231 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
David S. Miller1c92b4e2007-05-31 13:24:26 -07002232 unix_state_unlock(sk);
Colin Cross2b15af62013-05-06 23:50:21 +00002233 timeo = freezable_schedule_timeout(timeo);
David S. Miller1c92b4e2007-05-31 13:24:26 -07002234 unix_state_lock(sk);
Mark Salyzynb48732e2015-05-26 08:22:19 -07002235
2236 if (sock_flag(sk, SOCK_DEAD))
2237 break;
2238
Eric Dumazet9cd3e072015-11-29 20:03:10 -08002239 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002240 }
2241
Eric Dumazetaa395142010-04-20 13:03:51 +00002242 finish_wait(sk_sleep(sk), &wait);
David S. Miller1c92b4e2007-05-31 13:24:26 -07002243 unix_state_unlock(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002244 return timeo;
2245}
2246
Eric Dumazete370a722013-08-08 14:37:32 -07002247static unsigned int unix_skb_len(const struct sk_buff *skb)
2248{
2249 return skb->len - UNIXCB(skb).consumed;
2250}
2251
Hannes Frederic Sowa2b514572015-05-21 17:00:01 +02002252struct unix_stream_read_state {
2253 int (*recv_actor)(struct sk_buff *, int, int,
2254 struct unix_stream_read_state *);
2255 struct socket *socket;
2256 struct msghdr *msg;
2257 struct pipe_inode_info *pipe;
2258 size_t size;
2259 int flags;
2260 unsigned int splice_flags;
2261};
2262
2263static int unix_stream_read_generic(struct unix_stream_read_state *state)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002264{
Christoph Hellwig7cc05662015-01-28 18:04:53 +01002265 struct scm_cookie scm;
Hannes Frederic Sowa2b514572015-05-21 17:00:01 +02002266 struct socket *sock = state->socket;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002267 struct sock *sk = sock->sk;
2268 struct unix_sock *u = unix_sk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002269 int copied = 0;
Hannes Frederic Sowa2b514572015-05-21 17:00:01 +02002270 int flags = state->flags;
Eric Dumazetde144392014-03-25 18:42:27 -07002271 int noblock = flags & MSG_DONTWAIT;
Hannes Frederic Sowa2b514572015-05-21 17:00:01 +02002272 bool check_creds = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002273 int target;
2274 int err = 0;
2275 long timeo;
Pavel Emelyanovfc0d7532012-02-21 07:32:06 +00002276 int skip;
Hannes Frederic Sowa2b514572015-05-21 17:00:01 +02002277 size_t size = state->size;
2278 unsigned int last_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002279
2280 err = -EINVAL;
2281 if (sk->sk_state != TCP_ESTABLISHED)
2282 goto out;
2283
2284 err = -EOPNOTSUPP;
Hannes Frederic Sowa2b514572015-05-21 17:00:01 +02002285 if (flags & MSG_OOB)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002286 goto out;
2287
Hannes Frederic Sowa2b514572015-05-21 17:00:01 +02002288 target = sock_rcvlowat(sk, flags & MSG_WAITALL, size);
Eric Dumazetde144392014-03-25 18:42:27 -07002289 timeo = sock_rcvtimeo(sk, noblock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002290
Hannes Frederic Sowa2b514572015-05-21 17:00:01 +02002291 memset(&scm, 0, sizeof(scm));
2292
Linus Torvalds1da177e2005-04-16 15:20:36 -07002293 /* Lock the socket to prevent queue disordering
2294 * while sleeps in memcpy_tomsg
2295 */
Rainer Weikusat3822b5c2015-12-16 20:09:25 +00002296 mutex_lock(&u->readlock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002297
Andrey Vagine9193d62015-10-02 00:05:36 +03002298 if (flags & MSG_PEEK)
2299 skip = sk_peek_offset(sk, flags);
2300 else
2301 skip = 0;
2302
Eric Dumazet6eba6a32008-11-16 22:58:44 -08002303 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002304 int chunk;
Hannes Frederic Sowa73ed5d22015-11-10 16:23:15 +01002305 bool drop_skb;
Benjamin Poirier79f632c2013-04-29 11:42:14 +00002306 struct sk_buff *skb, *last;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002307
Miklos Szeredi3c0d2f32007-06-05 13:10:29 -07002308 unix_state_lock(sk);
Mark Salyzynb48732e2015-05-26 08:22:19 -07002309 if (sock_flag(sk, SOCK_DEAD)) {
2310 err = -ECONNRESET;
2311 goto unlock;
2312 }
Benjamin Poirier79f632c2013-04-29 11:42:14 +00002313 last = skb = skb_peek(&sk->sk_receive_queue);
Hannes Frederic Sowa2b514572015-05-21 17:00:01 +02002314 last_len = last ? last->len : 0;
Pavel Emelyanovfc0d7532012-02-21 07:32:06 +00002315again:
Eric Dumazet6eba6a32008-11-16 22:58:44 -08002316 if (skb == NULL) {
Eric Dumazet25888e32010-11-25 04:11:39 +00002317 unix_sk(sk)->recursion_level = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002318 if (copied >= target)
Miklos Szeredi3c0d2f32007-06-05 13:10:29 -07002319 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002320
2321 /*
2322 * POSIX 1003.1g mandates this order.
2323 */
YOSHIFUJI Hideakiac7bfa62007-02-09 23:25:23 +09002324
Eric Dumazet6eba6a32008-11-16 22:58:44 -08002325 err = sock_error(sk);
2326 if (err)
Miklos Szeredi3c0d2f32007-06-05 13:10:29 -07002327 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002328 if (sk->sk_shutdown & RCV_SHUTDOWN)
Miklos Szeredi3c0d2f32007-06-05 13:10:29 -07002329 goto unlock;
2330
2331 unix_state_unlock(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002332 err = -EAGAIN;
2333 if (!timeo)
2334 break;
Ingo Molnar57b47a52006-03-20 22:35:41 -08002335 mutex_unlock(&u->readlock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002336
Hannes Frederic Sowa2b514572015-05-21 17:00:01 +02002337 timeo = unix_stream_data_wait(sk, timeo, last,
2338 last_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002339
Rainer Weikusat3822b5c2015-12-16 20:09:25 +00002340 if (signal_pending(current)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002341 err = sock_intr_errno(timeo);
2342 goto out;
2343 }
Rainer Weikusatb3ca9b02011-02-28 04:50:55 +00002344
Rainer Weikusat3822b5c2015-12-16 20:09:25 +00002345 mutex_lock(&u->readlock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002346 continue;
Hannes Frederic Sowa2b514572015-05-21 17:00:01 +02002347unlock:
Miklos Szeredi3c0d2f32007-06-05 13:10:29 -07002348 unix_state_unlock(sk);
2349 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002350 }
Pavel Emelyanovfc0d7532012-02-21 07:32:06 +00002351
Eric Dumazete370a722013-08-08 14:37:32 -07002352 while (skip >= unix_skb_len(skb)) {
2353 skip -= unix_skb_len(skb);
Benjamin Poirier79f632c2013-04-29 11:42:14 +00002354 last = skb;
Hannes Frederic Sowa2b514572015-05-21 17:00:01 +02002355 last_len = skb->len;
Pavel Emelyanovfc0d7532012-02-21 07:32:06 +00002356 skb = skb_peek_next(skb, &sk->sk_receive_queue);
Benjamin Poirier79f632c2013-04-29 11:42:14 +00002357 if (!skb)
2358 goto again;
Pavel Emelyanovfc0d7532012-02-21 07:32:06 +00002359 }
2360
Miklos Szeredi3c0d2f32007-06-05 13:10:29 -07002361 unix_state_unlock(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002362
2363 if (check_creds) {
2364 /* Never glue messages from different writers */
Hannes Frederic Sowa9490f882015-11-26 12:08:18 +01002365 if (!unix_skb_scm_eq(skb, &scm))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002366 break;
Eric W. Biederman0e82e7f6d2013-04-03 16:14:47 +00002367 } else if (test_bit(SOCK_PASSCRED, &sock->flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002368 /* Copy credentials */
Christoph Hellwig7cc05662015-01-28 18:04:53 +01002369 scm_set_cred(&scm, UNIXCB(skb).pid, UNIXCB(skb).uid, UNIXCB(skb).gid);
Stephen Smalley37a9a8d2015-06-10 08:44:59 -04002370 unix_set_secdata(&scm, skb);
Hannes Frederic Sowa2b514572015-05-21 17:00:01 +02002371 check_creds = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002372 }
2373
2374 /* Copy address just once */
Hannes Frederic Sowa2b514572015-05-21 17:00:01 +02002375 if (state->msg && state->msg->msg_name) {
2376 DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr,
2377 state->msg->msg_name);
2378 unix_copy_addr(state->msg, skb->sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002379 sunaddr = NULL;
2380 }
2381
Eric Dumazete370a722013-08-08 14:37:32 -07002382 chunk = min_t(unsigned int, unix_skb_len(skb) - skip, size);
Hannes Frederic Sowa73ed5d22015-11-10 16:23:15 +01002383 skb_get(skb);
Hannes Frederic Sowa2b514572015-05-21 17:00:01 +02002384 chunk = state->recv_actor(skb, skip, chunk, state);
Hannes Frederic Sowa73ed5d22015-11-10 16:23:15 +01002385 drop_skb = !unix_skb_len(skb);
2386 /* skb is only safe to use if !drop_skb */
2387 consume_skb(skb);
Hannes Frederic Sowa2b514572015-05-21 17:00:01 +02002388 if (chunk < 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002389 if (copied == 0)
2390 copied = -EFAULT;
2391 break;
2392 }
2393 copied += chunk;
2394 size -= chunk;
2395
Hannes Frederic Sowa73ed5d22015-11-10 16:23:15 +01002396 if (drop_skb) {
2397 /* the skb was touched by a concurrent reader;
2398 * we should not expect anything from this skb
2399 * anymore and assume it invalid - we can be
2400 * sure it was dropped from the socket queue
2401 *
2402 * let's report a short read
2403 */
2404 err = 0;
2405 break;
2406 }
2407
Linus Torvalds1da177e2005-04-16 15:20:36 -07002408 /* Mark read part of skb as used */
Eric Dumazet6eba6a32008-11-16 22:58:44 -08002409 if (!(flags & MSG_PEEK)) {
Eric Dumazete370a722013-08-08 14:37:32 -07002410 UNIXCB(skb).consumed += chunk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002411
Pavel Emelyanovfc0d7532012-02-21 07:32:06 +00002412 sk_peek_offset_bwd(sk, chunk);
2413
Linus Torvalds1da177e2005-04-16 15:20:36 -07002414 if (UNIXCB(skb).fp)
Christoph Hellwig7cc05662015-01-28 18:04:53 +01002415 unix_detach_fds(&scm, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002416
Eric Dumazete370a722013-08-08 14:37:32 -07002417 if (unix_skb_len(skb))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002418 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002419
Eric Dumazet6f01fd62012-01-28 16:11:03 +00002420 skb_unlink(skb, &sk->sk_receive_queue);
Neil Horman70d4bf62010-07-20 06:45:56 +00002421 consume_skb(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002422
Christoph Hellwig7cc05662015-01-28 18:04:53 +01002423 if (scm.fp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002424 break;
Eric Dumazet6eba6a32008-11-16 22:58:44 -08002425 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002426 /* It is questionable, see note in unix_dgram_recvmsg.
2427 */
2428 if (UNIXCB(skb).fp)
Christoph Hellwig7cc05662015-01-28 18:04:53 +01002429 scm.fp = scm_fp_dup(UNIXCB(skb).fp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002430
Andrey Vagine9193d62015-10-02 00:05:36 +03002431 sk_peek_offset_fwd(sk, chunk);
Pavel Emelyanovfc0d7532012-02-21 07:32:06 +00002432
Aaron Conole9f389e32015-09-26 18:50:43 -04002433 if (UNIXCB(skb).fp)
2434 break;
2435
Andrey Vagine9193d62015-10-02 00:05:36 +03002436 skip = 0;
Aaron Conole9f389e32015-09-26 18:50:43 -04002437 last = skb;
2438 last_len = skb->len;
2439 unix_state_lock(sk);
2440 skb = skb_peek_next(skb, &sk->sk_receive_queue);
2441 if (skb)
2442 goto again;
2443 unix_state_unlock(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002444 break;
2445 }
2446 } while (size);
2447
Ingo Molnar57b47a52006-03-20 22:35:41 -08002448 mutex_unlock(&u->readlock);
Hannes Frederic Sowa2b514572015-05-21 17:00:01 +02002449 if (state->msg)
2450 scm_recv(sock, state->msg, &scm, flags);
2451 else
2452 scm_destroy(&scm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002453out:
2454 return copied ? : err;
2455}
2456
Hannes Frederic Sowa2b514572015-05-21 17:00:01 +02002457static int unix_stream_read_actor(struct sk_buff *skb,
2458 int skip, int chunk,
2459 struct unix_stream_read_state *state)
2460{
2461 int ret;
2462
2463 ret = skb_copy_datagram_msg(skb, UNIXCB(skb).consumed + skip,
2464 state->msg, chunk);
2465 return ret ?: chunk;
2466}
2467
2468static int unix_stream_recvmsg(struct socket *sock, struct msghdr *msg,
2469 size_t size, int flags)
2470{
2471 struct unix_stream_read_state state = {
2472 .recv_actor = unix_stream_read_actor,
2473 .socket = sock,
2474 .msg = msg,
2475 .size = size,
2476 .flags = flags
2477 };
2478
2479 return unix_stream_read_generic(&state);
2480}
2481
2482static ssize_t skb_unix_socket_splice(struct sock *sk,
2483 struct pipe_inode_info *pipe,
2484 struct splice_pipe_desc *spd)
2485{
2486 int ret;
2487 struct unix_sock *u = unix_sk(sk);
2488
2489 mutex_unlock(&u->readlock);
2490 ret = splice_to_pipe(pipe, spd);
2491 mutex_lock(&u->readlock);
2492
2493 return ret;
2494}
2495
2496static int unix_stream_splice_actor(struct sk_buff *skb,
2497 int skip, int chunk,
2498 struct unix_stream_read_state *state)
2499{
2500 return skb_splice_bits(skb, state->socket->sk,
2501 UNIXCB(skb).consumed + skip,
2502 state->pipe, chunk, state->splice_flags,
2503 skb_unix_socket_splice);
2504}
2505
2506static ssize_t unix_stream_splice_read(struct socket *sock, loff_t *ppos,
2507 struct pipe_inode_info *pipe,
2508 size_t size, unsigned int flags)
2509{
2510 struct unix_stream_read_state state = {
2511 .recv_actor = unix_stream_splice_actor,
2512 .socket = sock,
2513 .pipe = pipe,
2514 .size = size,
2515 .splice_flags = flags,
2516 };
2517
2518 if (unlikely(*ppos))
2519 return -ESPIPE;
2520
2521 if (sock->file->f_flags & O_NONBLOCK ||
2522 flags & SPLICE_F_NONBLOCK)
2523 state.flags = MSG_DONTWAIT;
2524
2525 return unix_stream_read_generic(&state);
2526}
2527
Linus Torvalds1da177e2005-04-16 15:20:36 -07002528static int unix_shutdown(struct socket *sock, int mode)
2529{
2530 struct sock *sk = sock->sk;
2531 struct sock *other;
2532
Xi Wangfc61b922012-08-26 16:47:13 +00002533 if (mode < SHUT_RD || mode > SHUT_RDWR)
2534 return -EINVAL;
2535 /* This maps:
2536 * SHUT_RD (0) -> RCV_SHUTDOWN (1)
2537 * SHUT_WR (1) -> SEND_SHUTDOWN (2)
2538 * SHUT_RDWR (2) -> SHUTDOWN_MASK (3)
2539 */
2540 ++mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002541
Alban Crequy7180a032011-01-19 04:56:36 +00002542 unix_state_lock(sk);
2543 sk->sk_shutdown |= mode;
2544 other = unix_peer(sk);
2545 if (other)
2546 sock_hold(other);
2547 unix_state_unlock(sk);
2548 sk->sk_state_change(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002549
Alban Crequy7180a032011-01-19 04:56:36 +00002550 if (other &&
2551 (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002552
Alban Crequy7180a032011-01-19 04:56:36 +00002553 int peer_mode = 0;
2554
2555 if (mode&RCV_SHUTDOWN)
2556 peer_mode |= SEND_SHUTDOWN;
2557 if (mode&SEND_SHUTDOWN)
2558 peer_mode |= RCV_SHUTDOWN;
2559 unix_state_lock(other);
2560 other->sk_shutdown |= peer_mode;
2561 unix_state_unlock(other);
2562 other->sk_state_change(other);
2563 if (peer_mode == SHUTDOWN_MASK)
2564 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_HUP);
2565 else if (peer_mode & RCV_SHUTDOWN)
2566 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_IN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002567 }
Alban Crequy7180a032011-01-19 04:56:36 +00002568 if (other)
2569 sock_put(other);
2570
Linus Torvalds1da177e2005-04-16 15:20:36 -07002571 return 0;
2572}
2573
Pavel Emelyanov885ee742011-12-30 00:54:11 +00002574long unix_inq_len(struct sock *sk)
2575{
2576 struct sk_buff *skb;
2577 long amount = 0;
2578
2579 if (sk->sk_state == TCP_LISTEN)
2580 return -EINVAL;
2581
2582 spin_lock(&sk->sk_receive_queue.lock);
2583 if (sk->sk_type == SOCK_STREAM ||
2584 sk->sk_type == SOCK_SEQPACKET) {
2585 skb_queue_walk(&sk->sk_receive_queue, skb)
Eric Dumazete370a722013-08-08 14:37:32 -07002586 amount += unix_skb_len(skb);
Pavel Emelyanov885ee742011-12-30 00:54:11 +00002587 } else {
2588 skb = skb_peek(&sk->sk_receive_queue);
2589 if (skb)
2590 amount = skb->len;
2591 }
2592 spin_unlock(&sk->sk_receive_queue.lock);
2593
2594 return amount;
2595}
2596EXPORT_SYMBOL_GPL(unix_inq_len);
2597
2598long unix_outq_len(struct sock *sk)
2599{
2600 return sk_wmem_alloc_get(sk);
2601}
2602EXPORT_SYMBOL_GPL(unix_outq_len);
2603
Linus Torvalds1da177e2005-04-16 15:20:36 -07002604static int unix_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2605{
2606 struct sock *sk = sock->sk;
Jianjun Konge27dfce2008-11-01 21:38:31 -07002607 long amount = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002608 int err;
2609
Eric Dumazet6eba6a32008-11-16 22:58:44 -08002610 switch (cmd) {
2611 case SIOCOUTQ:
Pavel Emelyanov885ee742011-12-30 00:54:11 +00002612 amount = unix_outq_len(sk);
Eric Dumazet6eba6a32008-11-16 22:58:44 -08002613 err = put_user(amount, (int __user *)arg);
2614 break;
2615 case SIOCINQ:
Pavel Emelyanov885ee742011-12-30 00:54:11 +00002616 amount = unix_inq_len(sk);
2617 if (amount < 0)
2618 err = amount;
2619 else
Linus Torvalds1da177e2005-04-16 15:20:36 -07002620 err = put_user(amount, (int __user *)arg);
Pavel Emelyanov885ee742011-12-30 00:54:11 +00002621 break;
Eric Dumazet6eba6a32008-11-16 22:58:44 -08002622 default:
2623 err = -ENOIOCTLCMD;
2624 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002625 }
2626 return err;
2627}
2628
Eric Dumazet6eba6a32008-11-16 22:58:44 -08002629static unsigned int unix_poll(struct file *file, struct socket *sock, poll_table *wait)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002630{
2631 struct sock *sk = sock->sk;
2632 unsigned int mask;
2633
Eric Dumazetaa395142010-04-20 13:03:51 +00002634 sock_poll_wait(file, sk_sleep(sk), wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002635 mask = 0;
2636
2637 /* exceptional events? */
2638 if (sk->sk_err)
2639 mask |= POLLERR;
2640 if (sk->sk_shutdown == SHUTDOWN_MASK)
2641 mask |= POLLHUP;
Davide Libenzif348d702006-03-25 03:07:39 -08002642 if (sk->sk_shutdown & RCV_SHUTDOWN)
Eric Dumazetdb409802010-09-06 11:13:50 +00002643 mask |= POLLRDHUP | POLLIN | POLLRDNORM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002644
2645 /* readable? */
Eric Dumazetdb409802010-09-06 11:13:50 +00002646 if (!skb_queue_empty(&sk->sk_receive_queue))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002647 mask |= POLLIN | POLLRDNORM;
2648
2649 /* Connection-based need to check for termination and startup */
Eric Dumazet6eba6a32008-11-16 22:58:44 -08002650 if ((sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) &&
2651 sk->sk_state == TCP_CLOSE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002652 mask |= POLLHUP;
2653
2654 /*
2655 * we set writable also when the other side has shut down the
2656 * connection. This prevents stuck sockets.
2657 */
2658 if (unix_writable(sk))
2659 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2660
2661 return mask;
2662}
2663
Rainer Weikusatec0d2152008-06-27 19:34:18 -07002664static unsigned int unix_dgram_poll(struct file *file, struct socket *sock,
2665 poll_table *wait)
Rainer Weikusat3c734192008-06-17 22:28:05 -07002666{
Rainer Weikusatec0d2152008-06-27 19:34:18 -07002667 struct sock *sk = sock->sk, *other;
2668 unsigned int mask, writable;
Rainer Weikusat3c734192008-06-17 22:28:05 -07002669
Eric Dumazetaa395142010-04-20 13:03:51 +00002670 sock_poll_wait(file, sk_sleep(sk), wait);
Rainer Weikusat3c734192008-06-17 22:28:05 -07002671 mask = 0;
2672
2673 /* exceptional events? */
2674 if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue))
Keller, Jacob E7d4c04f2013-03-28 11:19:25 +00002675 mask |= POLLERR |
Jacob Keller8facd5f2013-04-02 13:55:40 -07002676 (sock_flag(sk, SOCK_SELECT_ERR_QUEUE) ? POLLPRI : 0);
Keller, Jacob E7d4c04f2013-03-28 11:19:25 +00002677
Rainer Weikusat3c734192008-06-17 22:28:05 -07002678 if (sk->sk_shutdown & RCV_SHUTDOWN)
Eric Dumazet5456f092010-10-31 05:36:23 +00002679 mask |= POLLRDHUP | POLLIN | POLLRDNORM;
Rainer Weikusat3c734192008-06-17 22:28:05 -07002680 if (sk->sk_shutdown == SHUTDOWN_MASK)
2681 mask |= POLLHUP;
2682
2683 /* readable? */
Eric Dumazet5456f092010-10-31 05:36:23 +00002684 if (!skb_queue_empty(&sk->sk_receive_queue))
Rainer Weikusat3c734192008-06-17 22:28:05 -07002685 mask |= POLLIN | POLLRDNORM;
2686
2687 /* Connection-based need to check for termination and startup */
2688 if (sk->sk_type == SOCK_SEQPACKET) {
2689 if (sk->sk_state == TCP_CLOSE)
2690 mask |= POLLHUP;
2691 /* connection hasn't started yet? */
2692 if (sk->sk_state == TCP_SYN_SENT)
2693 return mask;
2694 }
2695
Eric Dumazet973a34a2010-10-31 05:38:25 +00002696 /* No write status requested, avoid expensive OUT tests. */
Hans Verkuil626cf232012-03-23 15:02:27 -07002697 if (!(poll_requested_events(wait) & (POLLWRBAND|POLLWRNORM|POLLOUT)))
Eric Dumazet973a34a2010-10-31 05:38:25 +00002698 return mask;
2699
Rainer Weikusatec0d2152008-06-27 19:34:18 -07002700 writable = unix_writable(sk);
Rainer Weikusat7d267272015-11-20 22:07:23 +00002701 if (writable) {
2702 unix_state_lock(sk);
2703
2704 other = unix_peer(sk);
2705 if (other && unix_peer(other) != sk &&
2706 unix_recvq_full(other) &&
2707 unix_dgram_peer_wake_me(sk, other))
2708 writable = 0;
2709
2710 unix_state_unlock(sk);
Rainer Weikusatec0d2152008-06-27 19:34:18 -07002711 }
2712
2713 if (writable)
Rainer Weikusat3c734192008-06-17 22:28:05 -07002714 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2715 else
Eric Dumazet9cd3e072015-11-29 20:03:10 -08002716 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
Rainer Weikusat3c734192008-06-17 22:28:05 -07002717
Rainer Weikusat3c734192008-06-17 22:28:05 -07002718 return mask;
2719}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002720
2721#ifdef CONFIG_PROC_FS
Pavel Emelyanova53eb3f2007-11-23 20:30:01 +08002722
Eric Dumazet7123aaa2012-06-08 05:03:21 +00002723#define BUCKET_SPACE (BITS_PER_LONG - (UNIX_HASH_BITS + 1) - 1)
2724
2725#define get_bucket(x) ((x) >> BUCKET_SPACE)
2726#define get_offset(x) ((x) & ((1L << BUCKET_SPACE) - 1))
2727#define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
Pavel Emelyanova53eb3f2007-11-23 20:30:01 +08002728
Eric Dumazet7123aaa2012-06-08 05:03:21 +00002729static struct sock *unix_from_bucket(struct seq_file *seq, loff_t *pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002730{
Eric Dumazet7123aaa2012-06-08 05:03:21 +00002731 unsigned long offset = get_offset(*pos);
2732 unsigned long bucket = get_bucket(*pos);
2733 struct sock *sk;
2734 unsigned long count = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002735
Eric Dumazet7123aaa2012-06-08 05:03:21 +00002736 for (sk = sk_head(&unix_socket_table[bucket]); sk; sk = sk_next(sk)) {
2737 if (sock_net(sk) != seq_file_net(seq))
Denis V. Lunev097e66c2007-11-19 22:29:30 -08002738 continue;
Eric Dumazet7123aaa2012-06-08 05:03:21 +00002739 if (++count == offset)
2740 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002741 }
Eric Dumazet7123aaa2012-06-08 05:03:21 +00002742
2743 return sk;
2744}
2745
2746static struct sock *unix_next_socket(struct seq_file *seq,
2747 struct sock *sk,
2748 loff_t *pos)
2749{
2750 unsigned long bucket;
2751
2752 while (sk > (struct sock *)SEQ_START_TOKEN) {
2753 sk = sk_next(sk);
2754 if (!sk)
2755 goto next_bucket;
2756 if (sock_net(sk) == seq_file_net(seq))
2757 return sk;
2758 }
2759
2760 do {
2761 sk = unix_from_bucket(seq, pos);
2762 if (sk)
2763 return sk;
2764
2765next_bucket:
2766 bucket = get_bucket(*pos) + 1;
2767 *pos = set_bucket_offset(bucket, 1);
2768 } while (bucket < ARRAY_SIZE(unix_socket_table));
2769
Linus Torvalds1da177e2005-04-16 15:20:36 -07002770 return NULL;
2771}
2772
Linus Torvalds1da177e2005-04-16 15:20:36 -07002773static void *unix_seq_start(struct seq_file *seq, loff_t *pos)
Eric Dumazet9a429c42008-01-01 21:58:02 -08002774 __acquires(unix_table_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002775{
David S. Millerfbe9cc42005-12-13 23:26:29 -08002776 spin_lock(&unix_table_lock);
Eric Dumazet7123aaa2012-06-08 05:03:21 +00002777
2778 if (!*pos)
2779 return SEQ_START_TOKEN;
2780
2781 if (get_bucket(*pos) >= ARRAY_SIZE(unix_socket_table))
2782 return NULL;
2783
2784 return unix_next_socket(seq, NULL, pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002785}
2786
2787static void *unix_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2788{
2789 ++*pos;
Eric Dumazet7123aaa2012-06-08 05:03:21 +00002790 return unix_next_socket(seq, v, pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002791}
2792
2793static void unix_seq_stop(struct seq_file *seq, void *v)
Eric Dumazet9a429c42008-01-01 21:58:02 -08002794 __releases(unix_table_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002795{
David S. Millerfbe9cc42005-12-13 23:26:29 -08002796 spin_unlock(&unix_table_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002797}
2798
2799static int unix_seq_show(struct seq_file *seq, void *v)
2800{
YOSHIFUJI Hideakiac7bfa62007-02-09 23:25:23 +09002801
Joe Perchesb9f31242008-04-12 19:04:38 -07002802 if (v == SEQ_START_TOKEN)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002803 seq_puts(seq, "Num RefCount Protocol Flags Type St "
2804 "Inode Path\n");
2805 else {
2806 struct sock *s = v;
2807 struct unix_sock *u = unix_sk(s);
David S. Miller1c92b4e2007-05-31 13:24:26 -07002808 unix_state_lock(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002809
Dan Rosenberg71338aa2011-05-23 12:17:35 +00002810 seq_printf(seq, "%pK: %08X %08X %08X %04X %02X %5lu",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002811 s,
2812 atomic_read(&s->sk_refcnt),
2813 0,
2814 s->sk_state == TCP_LISTEN ? __SO_ACCEPTCON : 0,
2815 s->sk_type,
2816 s->sk_socket ?
2817 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTED : SS_UNCONNECTED) :
2818 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTING : SS_DISCONNECTING),
2819 sock_i_ino(s));
2820
2821 if (u->addr) {
2822 int i, len;
2823 seq_putc(seq, ' ');
2824
2825 i = 0;
2826 len = u->addr->len - sizeof(short);
2827 if (!UNIX_ABSTRACT(s))
2828 len--;
2829 else {
2830 seq_putc(seq, '@');
2831 i++;
2832 }
2833 for ( ; i < len; i++)
2834 seq_putc(seq, u->addr->name->sun_path[i]);
2835 }
David S. Miller1c92b4e2007-05-31 13:24:26 -07002836 unix_state_unlock(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002837 seq_putc(seq, '\n');
2838 }
2839
2840 return 0;
2841}
2842
Philippe De Muyter56b3d972007-07-10 23:07:31 -07002843static const struct seq_operations unix_seq_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002844 .start = unix_seq_start,
2845 .next = unix_seq_next,
2846 .stop = unix_seq_stop,
2847 .show = unix_seq_show,
2848};
2849
Linus Torvalds1da177e2005-04-16 15:20:36 -07002850static int unix_seq_open(struct inode *inode, struct file *file)
2851{
Denis V. Luneve372c412007-11-19 22:31:54 -08002852 return seq_open_net(inode, file, &unix_seq_ops,
Eric Dumazet8b51b062012-06-08 22:10:20 +00002853 sizeof(struct seq_net_private));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002854}
2855
Arjan van de Venda7071d2007-02-12 00:55:36 -08002856static const struct file_operations unix_seq_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002857 .owner = THIS_MODULE,
2858 .open = unix_seq_open,
2859 .read = seq_read,
2860 .llseek = seq_lseek,
Denis V. Luneve372c412007-11-19 22:31:54 -08002861 .release = seq_release_net,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002862};
2863
2864#endif
2865
Stephen Hemmingerec1b4cf2009-10-05 05:58:39 +00002866static const struct net_proto_family unix_family_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002867 .family = PF_UNIX,
2868 .create = unix_create,
2869 .owner = THIS_MODULE,
2870};
2871
Denis V. Lunev097e66c2007-11-19 22:29:30 -08002872
Alexey Dobriyan2c8c1e72010-01-17 03:35:32 +00002873static int __net_init unix_net_init(struct net *net)
Denis V. Lunev097e66c2007-11-19 22:29:30 -08002874{
2875 int error = -ENOMEM;
2876
Denis V. Luneva0a53c82007-12-11 04:19:17 -08002877 net->unx.sysctl_max_dgram_qlen = 10;
Pavel Emelyanov1597fbc2007-12-01 23:51:01 +11002878 if (unix_sysctl_register(net))
2879 goto out;
Pavel Emelyanovd392e492007-12-01 23:44:15 +11002880
Denis V. Lunev097e66c2007-11-19 22:29:30 -08002881#ifdef CONFIG_PROC_FS
Gao fengd4beaa62013-02-18 01:34:54 +00002882 if (!proc_create("unix", 0, net->proc_net, &unix_seq_fops)) {
Pavel Emelyanov1597fbc2007-12-01 23:51:01 +11002883 unix_sysctl_unregister(net);
Denis V. Lunev097e66c2007-11-19 22:29:30 -08002884 goto out;
Pavel Emelyanov1597fbc2007-12-01 23:51:01 +11002885 }
Denis V. Lunev097e66c2007-11-19 22:29:30 -08002886#endif
2887 error = 0;
2888out:
Jianjun Kong48dcc33e2008-11-01 21:37:27 -07002889 return error;
Denis V. Lunev097e66c2007-11-19 22:29:30 -08002890}
2891
Alexey Dobriyan2c8c1e72010-01-17 03:35:32 +00002892static void __net_exit unix_net_exit(struct net *net)
Denis V. Lunev097e66c2007-11-19 22:29:30 -08002893{
Pavel Emelyanov1597fbc2007-12-01 23:51:01 +11002894 unix_sysctl_unregister(net);
Gao fengece31ff2013-02-18 01:34:56 +00002895 remove_proc_entry("unix", net->proc_net);
Denis V. Lunev097e66c2007-11-19 22:29:30 -08002896}
2897
2898static struct pernet_operations unix_net_ops = {
2899 .init = unix_net_init,
2900 .exit = unix_net_exit,
2901};
2902
Linus Torvalds1da177e2005-04-16 15:20:36 -07002903static int __init af_unix_init(void)
2904{
2905 int rc = -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002906
YOSHIFUJI Hideaki / 吉藤英明b4fff5f2013-01-09 07:20:07 +00002907 BUILD_BUG_ON(sizeof(struct unix_skb_parms) > FIELD_SIZEOF(struct sk_buff, cb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002908
2909 rc = proto_register(&unix_proto, 1);
YOSHIFUJI Hideakiac7bfa62007-02-09 23:25:23 +09002910 if (rc != 0) {
wangweidong5cc208b2013-12-06 18:03:36 +08002911 pr_crit("%s: Cannot create unix_sock SLAB cache!\n", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002912 goto out;
2913 }
2914
2915 sock_register(&unix_family_ops);
Denis V. Lunev097e66c2007-11-19 22:29:30 -08002916 register_pernet_subsys(&unix_net_ops);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002917out:
2918 return rc;
2919}
2920
2921static void __exit af_unix_exit(void)
2922{
2923 sock_unregister(PF_UNIX);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002924 proto_unregister(&unix_proto);
Denis V. Lunev097e66c2007-11-19 22:29:30 -08002925 unregister_pernet_subsys(&unix_net_ops);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002926}
2927
David Woodhouse3d366962008-04-24 00:59:25 -07002928/* Earlier than device_initcall() so that other drivers invoking
2929 request_module() don't end up in a loop when modprobe tries
2930 to use a UNIX socket. But later than subsys_initcall() because
2931 we depend on stuff initialised there */
2932fs_initcall(af_unix_init);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002933module_exit(af_unix_exit);
2934
2935MODULE_LICENSE("GPL");
2936MODULE_ALIAS_NETPROTO(PF_UNIX);