Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * NET4: Implementation of BSD Unix domain sockets. | |
3 | * | |
4 | * Authors: Alan Cox, <alan.cox@linux.org> | |
5 | * | |
6 | * This program is free software; you can redistribute it and/or | |
7 | * modify it under the terms of the GNU General Public License | |
8 | * as published by the Free Software Foundation; either version | |
9 | * 2 of the License, or (at your option) any later version. | |
10 | * | |
11 | * Version: $Id: af_unix.c,v 1.133 2002/02/08 03:57:19 davem Exp $ | |
12 | * | |
13 | * Fixes: | |
14 | * Linus Torvalds : Assorted bug cures. | |
15 | * Niibe Yutaka : async I/O support. | |
16 | * Carsten Paeth : PF_UNIX check, address fixes. | |
17 | * Alan Cox : Limit size of allocated blocks. | |
18 | * Alan Cox : Fixed the stupid socketpair bug. | |
19 | * Alan Cox : BSD compatibility fine tuning. | |
20 | * Alan Cox : Fixed a bug in connect when interrupted. | |
21 | * Alan Cox : Sorted out a proper draft version of | |
22 | * file descriptor passing hacked up from | |
23 | * Mike Shaver's work. | |
24 | * Marty Leisner : Fixes to fd passing | |
25 | * Nick Nevin : recvmsg bugfix. | |
26 | * Alan Cox : Started proper garbage collector | |
27 | * Heiko EiBfeldt : Missing verify_area check | |
28 | * Alan Cox : Started POSIXisms | |
29 | * Andreas Schwab : Replace inode by dentry for proper | |
30 | * reference counting | |
31 | * Kirk Petersen : Made this a module | |
32 | * Christoph Rohland : Elegant non-blocking accept/connect algorithm. | |
33 | * Lots of bug fixes. | |
34 | * Alexey Kuznetosv : Repaired (I hope) bugs introduces | |
35 | * by above two patches. | |
36 | * Andrea Arcangeli : If possible we block in connect(2) | |
37 | * if the max backlog of the listen socket | |
38 | * is been reached. This won't break | |
39 | * old apps and it will avoid huge amount | |
40 | * of socks hashed (this for unix_gc() | |
41 | * performances reasons). | |
42 | * Security fix that limits the max | |
43 | * number of socks to 2*max_files and | |
44 | * the number of skb queueable in the | |
45 | * dgram receiver. | |
46 | * Artur Skawina : Hash function optimizations | |
47 | * Alexey Kuznetsov : Full scale SMP. Lot of bugs are introduced 8) | |
48 | * Malcolm Beattie : Set peercred for socketpair | |
49 | * Michal Ostrowski : Module initialization cleanup. | |
50 | * Arnaldo C. Melo : Remove MOD_{INC,DEC}_USE_COUNT, | |
51 | * the core infrastructure is doing that | |
52 | * for all net proto families now (2.5.69+) | |
53 | * | |
54 | * | |
55 | * Known differences from reference BSD that was tested: | |
56 | * | |
57 | * [TO FIX] | |
58 | * ECONNREFUSED is not returned from one end of a connected() socket to the | |
59 | * other the moment one end closes. | |
60 | * fstat() doesn't return st_dev=0, and give the blksize as high water mark | |
61 | * and a fake inode identifier (nor the BSD first socket fstat twice bug). | |
62 | * [NOT TO FIX] | |
63 | * accept() returns a path name even if the connecting socket has closed | |
64 | * in the meantime (BSD loses the path and gives up). | |
65 | * accept() returns 0 length path for an unbound connector. BSD returns 16 | |
66 | * and a null first byte in the path (but not for gethost/peername - BSD bug ??) | |
67 | * socketpair(...SOCK_RAW..) doesn't panic the kernel. | |
68 | * BSD af_unix apparently has connect forgetting to block properly. | |
69 | * (need to check this with the POSIX spec in detail) | |
70 | * | |
71 | * Differences from 2.0.0-11-... (ANK) | |
72 | * Bug fixes and improvements. | |
73 | * - client shutdown killed server socket. | |
74 | * - removed all useless cli/sti pairs. | |
75 | * | |
76 | * Semantic changes/extensions. | |
77 | * - generic control message passing. | |
78 | * - SCM_CREDENTIALS control message. | |
79 | * - "Abstract" (not FS based) socket bindings. | |
80 | * Abstract names are sequences of bytes (not zero terminated) | |
81 | * started by 0, so that this name space does not intersect | |
82 | * with BSD names. | |
83 | */ | |
84 | ||
85 | #include <linux/module.h> | |
1da177e4 | 86 | #include <linux/kernel.h> |
1da177e4 LT |
87 | #include <linux/signal.h> |
88 | #include <linux/sched.h> | |
89 | #include <linux/errno.h> | |
90 | #include <linux/string.h> | |
91 | #include <linux/stat.h> | |
92 | #include <linux/dcache.h> | |
93 | #include <linux/namei.h> | |
94 | #include <linux/socket.h> | |
95 | #include <linux/un.h> | |
96 | #include <linux/fcntl.h> | |
97 | #include <linux/termios.h> | |
98 | #include <linux/sockios.h> | |
99 | #include <linux/net.h> | |
100 | #include <linux/in.h> | |
101 | #include <linux/fs.h> | |
102 | #include <linux/slab.h> | |
103 | #include <asm/uaccess.h> | |
104 | #include <linux/skbuff.h> | |
105 | #include <linux/netdevice.h> | |
457c4cbc | 106 | #include <net/net_namespace.h> |
1da177e4 | 107 | #include <net/sock.h> |
c752f073 | 108 | #include <net/tcp_states.h> |
1da177e4 LT |
109 | #include <net/af_unix.h> |
110 | #include <linux/proc_fs.h> | |
111 | #include <linux/seq_file.h> | |
112 | #include <net/scm.h> | |
113 | #include <linux/init.h> | |
114 | #include <linux/poll.h> | |
1da177e4 LT |
115 | #include <linux/rtnetlink.h> |
116 | #include <linux/mount.h> | |
117 | #include <net/checksum.h> | |
118 | #include <linux/security.h> | |
119 | ||
18adaf06 | 120 | int sysctl_unix_max_dgram_qlen __read_mostly = 10; |
1da177e4 | 121 | |
13111698 AB |
122 | static struct hlist_head unix_socket_table[UNIX_HASH_SIZE + 1]; |
123 | static DEFINE_SPINLOCK(unix_table_lock); | |
1da177e4 LT |
124 | static atomic_t unix_nr_socks = ATOMIC_INIT(0); |
125 | ||
126 | #define unix_sockets_unbound (&unix_socket_table[UNIX_HASH_SIZE]) | |
127 | ||
128 | #define UNIX_ABSTRACT(sk) (unix_sk(sk)->addr->hash != UNIX_HASH_SIZE) | |
129 | ||
13111698 AB |
130 | static struct sock *first_unix_socket(int *i) |
131 | { | |
132 | for (*i = 0; *i <= UNIX_HASH_SIZE; (*i)++) { | |
133 | if (!hlist_empty(&unix_socket_table[*i])) | |
134 | return __sk_head(&unix_socket_table[*i]); | |
135 | } | |
136 | return NULL; | |
137 | } | |
138 | ||
139 | static struct sock *next_unix_socket(int *i, struct sock *s) | |
140 | { | |
141 | struct sock *next = sk_next(s); | |
142 | /* More in this chain? */ | |
143 | if (next) | |
144 | return next; | |
145 | /* Look for next non-empty chain. */ | |
146 | for ((*i)++; *i <= UNIX_HASH_SIZE; (*i)++) { | |
147 | if (!hlist_empty(&unix_socket_table[*i])) | |
148 | return __sk_head(&unix_socket_table[*i]); | |
149 | } | |
150 | return NULL; | |
151 | } | |
152 | ||
153 | #define forall_unix_sockets(i, s) \ | |
154 | for (s = first_unix_socket(&(i)); s; s = next_unix_socket(&(i),(s))) | |
155 | ||
877ce7c1 | 156 | #ifdef CONFIG_SECURITY_NETWORK |
dc49c1f9 | 157 | static void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb) |
877ce7c1 | 158 | { |
dc49c1f9 | 159 | memcpy(UNIXSID(skb), &scm->secid, sizeof(u32)); |
877ce7c1 CZ |
160 | } |
161 | ||
162 | static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb) | |
163 | { | |
dc49c1f9 | 164 | scm->secid = *UNIXSID(skb); |
877ce7c1 CZ |
165 | } |
166 | #else | |
dc49c1f9 | 167 | static inline void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb) |
877ce7c1 CZ |
168 | { } |
169 | ||
170 | static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb) | |
171 | { } | |
172 | #endif /* CONFIG_SECURITY_NETWORK */ | |
173 | ||
1da177e4 LT |
174 | /* |
175 | * SMP locking strategy: | |
fbe9cc4a | 176 | * hash table is protected with spinlock unix_table_lock |
1da177e4 LT |
177 | * each socket state is protected by separate rwlock. |
178 | */ | |
179 | ||
44bb9363 | 180 | static inline unsigned unix_hash_fold(__wsum n) |
1da177e4 | 181 | { |
44bb9363 | 182 | unsigned hash = (__force unsigned)n; |
1da177e4 LT |
183 | hash ^= hash>>16; |
184 | hash ^= hash>>8; | |
185 | return hash&(UNIX_HASH_SIZE-1); | |
186 | } | |
187 | ||
188 | #define unix_peer(sk) (unix_sk(sk)->peer) | |
189 | ||
190 | static inline int unix_our_peer(struct sock *sk, struct sock *osk) | |
191 | { | |
192 | return unix_peer(osk) == sk; | |
193 | } | |
194 | ||
195 | static inline int unix_may_send(struct sock *sk, struct sock *osk) | |
196 | { | |
197 | return (unix_peer(osk) == NULL || unix_our_peer(sk, osk)); | |
198 | } | |
199 | ||
200 | static struct sock *unix_peer_get(struct sock *s) | |
201 | { | |
202 | struct sock *peer; | |
203 | ||
1c92b4e5 | 204 | unix_state_lock(s); |
1da177e4 LT |
205 | peer = unix_peer(s); |
206 | if (peer) | |
207 | sock_hold(peer); | |
1c92b4e5 | 208 | unix_state_unlock(s); |
1da177e4 LT |
209 | return peer; |
210 | } | |
211 | ||
212 | static 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 | */ | |
ac7bfa62 | 224 | |
1da177e4 LT |
225 | static int unix_mkname(struct sockaddr_un * sunaddr, int len, unsigned *hashp) |
226 | { | |
227 | if (len <= sizeof(short) || len > sizeof(*sunaddr)) | |
228 | return -EINVAL; | |
229 | if (!sunaddr || sunaddr->sun_family != AF_UNIX) | |
230 | return -EINVAL; | |
231 | if (sunaddr->sun_path[0]) { | |
232 | /* | |
233 | * This may look like an off by one error but it is a bit more | |
234 | * subtle. 108 is the longest valid AF_UNIX path for a binding. | |
235 | * sun_path[108] doesnt as such exist. However in kernel space | |
236 | * we are guaranteed that it is a valid memory location in our | |
237 | * kernel address buffer. | |
238 | */ | |
239 | ((char *)sunaddr)[len]=0; | |
240 | len = strlen(sunaddr->sun_path)+1+sizeof(short); | |
241 | return len; | |
242 | } | |
243 | ||
244 | *hashp = unix_hash_fold(csum_partial((char*)sunaddr, len, 0)); | |
245 | return len; | |
246 | } | |
247 | ||
248 | static void __unix_remove_socket(struct sock *sk) | |
249 | { | |
250 | sk_del_node_init(sk); | |
251 | } | |
252 | ||
253 | static void __unix_insert_socket(struct hlist_head *list, struct sock *sk) | |
254 | { | |
255 | BUG_TRAP(sk_unhashed(sk)); | |
256 | sk_add_node(sk, list); | |
257 | } | |
258 | ||
259 | static inline void unix_remove_socket(struct sock *sk) | |
260 | { | |
fbe9cc4a | 261 | spin_lock(&unix_table_lock); |
1da177e4 | 262 | __unix_remove_socket(sk); |
fbe9cc4a | 263 | spin_unlock(&unix_table_lock); |
1da177e4 LT |
264 | } |
265 | ||
266 | static inline void unix_insert_socket(struct hlist_head *list, struct sock *sk) | |
267 | { | |
fbe9cc4a | 268 | spin_lock(&unix_table_lock); |
1da177e4 | 269 | __unix_insert_socket(list, sk); |
fbe9cc4a | 270 | spin_unlock(&unix_table_lock); |
1da177e4 LT |
271 | } |
272 | ||
273 | static struct sock *__unix_find_socket_byname(struct sockaddr_un *sunname, | |
274 | int len, int type, unsigned hash) | |
275 | { | |
276 | struct sock *s; | |
277 | struct hlist_node *node; | |
278 | ||
279 | sk_for_each(s, node, &unix_socket_table[hash ^ type]) { | |
280 | struct unix_sock *u = unix_sk(s); | |
281 | ||
282 | if (u->addr->len == len && | |
283 | !memcmp(u->addr->name, sunname, len)) | |
284 | goto found; | |
285 | } | |
286 | s = NULL; | |
287 | found: | |
288 | return s; | |
289 | } | |
290 | ||
291 | static inline struct sock *unix_find_socket_byname(struct sockaddr_un *sunname, | |
292 | int len, int type, | |
293 | unsigned hash) | |
294 | { | |
295 | struct sock *s; | |
296 | ||
fbe9cc4a | 297 | spin_lock(&unix_table_lock); |
1da177e4 LT |
298 | s = __unix_find_socket_byname(sunname, len, type, hash); |
299 | if (s) | |
300 | sock_hold(s); | |
fbe9cc4a | 301 | spin_unlock(&unix_table_lock); |
1da177e4 LT |
302 | return s; |
303 | } | |
304 | ||
305 | static struct sock *unix_find_socket_byinode(struct inode *i) | |
306 | { | |
307 | struct sock *s; | |
308 | struct hlist_node *node; | |
309 | ||
fbe9cc4a | 310 | spin_lock(&unix_table_lock); |
1da177e4 LT |
311 | sk_for_each(s, node, |
312 | &unix_socket_table[i->i_ino & (UNIX_HASH_SIZE - 1)]) { | |
313 | struct dentry *dentry = unix_sk(s)->dentry; | |
314 | ||
315 | if(dentry && dentry->d_inode == i) | |
316 | { | |
317 | sock_hold(s); | |
318 | goto found; | |
319 | } | |
320 | } | |
321 | s = NULL; | |
322 | found: | |
fbe9cc4a | 323 | spin_unlock(&unix_table_lock); |
1da177e4 LT |
324 | return s; |
325 | } | |
326 | ||
327 | static inline int unix_writable(struct sock *sk) | |
328 | { | |
329 | return (atomic_read(&sk->sk_wmem_alloc) << 2) <= sk->sk_sndbuf; | |
330 | } | |
331 | ||
332 | static void unix_write_space(struct sock *sk) | |
333 | { | |
334 | read_lock(&sk->sk_callback_lock); | |
335 | if (unix_writable(sk)) { | |
336 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) | |
71e20f18 | 337 | wake_up_interruptible_sync(sk->sk_sleep); |
1da177e4 LT |
338 | sk_wake_async(sk, 2, POLL_OUT); |
339 | } | |
340 | read_unlock(&sk->sk_callback_lock); | |
341 | } | |
342 | ||
343 | /* When dgram socket disconnects (or changes its peer), we clear its receive | |
344 | * queue of packets arrived from previous peer. First, it allows to do | |
345 | * flow control based only on wmem_alloc; second, sk connected to peer | |
346 | * may receive messages only from that peer. */ | |
347 | static void unix_dgram_disconnected(struct sock *sk, struct sock *other) | |
348 | { | |
b03efcfb | 349 | if (!skb_queue_empty(&sk->sk_receive_queue)) { |
1da177e4 LT |
350 | skb_queue_purge(&sk->sk_receive_queue); |
351 | wake_up_interruptible_all(&unix_sk(sk)->peer_wait); | |
352 | ||
353 | /* If one link of bidirectional dgram pipe is disconnected, | |
354 | * we signal error. Messages are lost. Do not make this, | |
355 | * when peer was not connected to us. | |
356 | */ | |
357 | if (!sock_flag(other, SOCK_DEAD) && unix_peer(other) == sk) { | |
358 | other->sk_err = ECONNRESET; | |
359 | other->sk_error_report(other); | |
360 | } | |
361 | } | |
362 | } | |
363 | ||
364 | static void unix_sock_destructor(struct sock *sk) | |
365 | { | |
366 | struct unix_sock *u = unix_sk(sk); | |
367 | ||
368 | skb_queue_purge(&sk->sk_receive_queue); | |
369 | ||
370 | BUG_TRAP(!atomic_read(&sk->sk_wmem_alloc)); | |
371 | BUG_TRAP(sk_unhashed(sk)); | |
372 | BUG_TRAP(!sk->sk_socket); | |
373 | if (!sock_flag(sk, SOCK_DEAD)) { | |
374 | printk("Attempt to release alive unix socket: %p\n", sk); | |
375 | return; | |
376 | } | |
377 | ||
378 | if (u->addr) | |
379 | unix_release_addr(u->addr); | |
380 | ||
381 | atomic_dec(&unix_nr_socks); | |
382 | #ifdef UNIX_REFCNT_DEBUG | |
383 | printk(KERN_DEBUG "UNIX %p is destroyed, %d are still alive.\n", sk, atomic_read(&unix_nr_socks)); | |
384 | #endif | |
385 | } | |
386 | ||
387 | static int unix_release_sock (struct sock *sk, int embrion) | |
388 | { | |
389 | struct unix_sock *u = unix_sk(sk); | |
390 | struct dentry *dentry; | |
391 | struct vfsmount *mnt; | |
392 | struct sock *skpair; | |
393 | struct sk_buff *skb; | |
394 | int state; | |
395 | ||
396 | unix_remove_socket(sk); | |
397 | ||
398 | /* Clear state */ | |
1c92b4e5 | 399 | unix_state_lock(sk); |
1da177e4 LT |
400 | sock_orphan(sk); |
401 | sk->sk_shutdown = SHUTDOWN_MASK; | |
402 | dentry = u->dentry; | |
403 | u->dentry = NULL; | |
404 | mnt = u->mnt; | |
405 | u->mnt = NULL; | |
406 | state = sk->sk_state; | |
407 | sk->sk_state = TCP_CLOSE; | |
1c92b4e5 | 408 | unix_state_unlock(sk); |
1da177e4 LT |
409 | |
410 | wake_up_interruptible_all(&u->peer_wait); | |
411 | ||
412 | skpair=unix_peer(sk); | |
413 | ||
414 | if (skpair!=NULL) { | |
415 | if (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) { | |
1c92b4e5 | 416 | unix_state_lock(skpair); |
1da177e4 LT |
417 | /* No more writes */ |
418 | skpair->sk_shutdown = SHUTDOWN_MASK; | |
419 | if (!skb_queue_empty(&sk->sk_receive_queue) || embrion) | |
420 | skpair->sk_err = ECONNRESET; | |
1c92b4e5 | 421 | unix_state_unlock(skpair); |
1da177e4 LT |
422 | skpair->sk_state_change(skpair); |
423 | read_lock(&skpair->sk_callback_lock); | |
424 | sk_wake_async(skpair,1,POLL_HUP); | |
425 | read_unlock(&skpair->sk_callback_lock); | |
426 | } | |
427 | sock_put(skpair); /* It may now die */ | |
428 | unix_peer(sk) = NULL; | |
429 | } | |
430 | ||
431 | /* Try to flush out this socket. Throw out buffers at least */ | |
432 | ||
433 | while ((skb = skb_dequeue(&sk->sk_receive_queue)) != NULL) { | |
434 | if (state==TCP_LISTEN) | |
435 | unix_release_sock(skb->sk, 1); | |
436 | /* passed fds are erased in the kfree_skb hook */ | |
437 | kfree_skb(skb); | |
438 | } | |
439 | ||
440 | if (dentry) { | |
441 | dput(dentry); | |
442 | mntput(mnt); | |
443 | } | |
444 | ||
445 | sock_put(sk); | |
446 | ||
447 | /* ---- Socket is dead now and most probably destroyed ---- */ | |
448 | ||
449 | /* | |
450 | * Fixme: BSD difference: In BSD all sockets connected to use get | |
451 | * ECONNRESET and we die on the spot. In Linux we behave | |
452 | * like files and pipes do and wait for the last | |
453 | * dereference. | |
454 | * | |
455 | * Can't we simply set sock->err? | |
456 | * | |
457 | * What the above comment does talk about? --ANK(980817) | |
458 | */ | |
459 | ||
460 | if (atomic_read(&unix_tot_inflight)) | |
ac7bfa62 | 461 | unix_gc(); /* Garbage collect fds */ |
1da177e4 LT |
462 | |
463 | return 0; | |
464 | } | |
465 | ||
466 | static int unix_listen(struct socket *sock, int backlog) | |
467 | { | |
468 | int err; | |
469 | struct sock *sk = sock->sk; | |
470 | struct unix_sock *u = unix_sk(sk); | |
471 | ||
472 | err = -EOPNOTSUPP; | |
473 | if (sock->type!=SOCK_STREAM && sock->type!=SOCK_SEQPACKET) | |
474 | goto out; /* Only stream/seqpacket sockets accept */ | |
475 | err = -EINVAL; | |
476 | if (!u->addr) | |
477 | goto out; /* No listens on an unbound socket */ | |
1c92b4e5 | 478 | unix_state_lock(sk); |
1da177e4 LT |
479 | if (sk->sk_state != TCP_CLOSE && sk->sk_state != TCP_LISTEN) |
480 | goto out_unlock; | |
481 | if (backlog > sk->sk_max_ack_backlog) | |
482 | wake_up_interruptible_all(&u->peer_wait); | |
483 | sk->sk_max_ack_backlog = backlog; | |
484 | sk->sk_state = TCP_LISTEN; | |
485 | /* set credentials so connect can copy them */ | |
b488893a | 486 | sk->sk_peercred.pid = task_tgid_vnr(current); |
1da177e4 LT |
487 | sk->sk_peercred.uid = current->euid; |
488 | sk->sk_peercred.gid = current->egid; | |
489 | err = 0; | |
490 | ||
491 | out_unlock: | |
1c92b4e5 | 492 | unix_state_unlock(sk); |
1da177e4 LT |
493 | out: |
494 | return err; | |
495 | } | |
496 | ||
497 | static int unix_release(struct socket *); | |
498 | static int unix_bind(struct socket *, struct sockaddr *, int); | |
499 | static int unix_stream_connect(struct socket *, struct sockaddr *, | |
500 | int addr_len, int flags); | |
501 | static int unix_socketpair(struct socket *, struct socket *); | |
502 | static int unix_accept(struct socket *, struct socket *, int); | |
503 | static int unix_getname(struct socket *, struct sockaddr *, int *, int); | |
504 | static unsigned int unix_poll(struct file *, struct socket *, poll_table *); | |
505 | static int unix_ioctl(struct socket *, unsigned int, unsigned long); | |
506 | static int unix_shutdown(struct socket *, int); | |
507 | static int unix_stream_sendmsg(struct kiocb *, struct socket *, | |
508 | struct msghdr *, size_t); | |
509 | static int unix_stream_recvmsg(struct kiocb *, struct socket *, | |
510 | struct msghdr *, size_t, int); | |
511 | static int unix_dgram_sendmsg(struct kiocb *, struct socket *, | |
512 | struct msghdr *, size_t); | |
513 | static int unix_dgram_recvmsg(struct kiocb *, struct socket *, | |
514 | struct msghdr *, size_t, int); | |
515 | static int unix_dgram_connect(struct socket *, struct sockaddr *, | |
516 | int, int); | |
517 | static int unix_seqpacket_sendmsg(struct kiocb *, struct socket *, | |
518 | struct msghdr *, size_t); | |
519 | ||
90ddc4f0 | 520 | static const struct proto_ops unix_stream_ops = { |
1da177e4 LT |
521 | .family = PF_UNIX, |
522 | .owner = THIS_MODULE, | |
523 | .release = unix_release, | |
524 | .bind = unix_bind, | |
525 | .connect = unix_stream_connect, | |
526 | .socketpair = unix_socketpair, | |
527 | .accept = unix_accept, | |
528 | .getname = unix_getname, | |
529 | .poll = unix_poll, | |
530 | .ioctl = unix_ioctl, | |
531 | .listen = unix_listen, | |
532 | .shutdown = unix_shutdown, | |
533 | .setsockopt = sock_no_setsockopt, | |
534 | .getsockopt = sock_no_getsockopt, | |
535 | .sendmsg = unix_stream_sendmsg, | |
536 | .recvmsg = unix_stream_recvmsg, | |
537 | .mmap = sock_no_mmap, | |
538 | .sendpage = sock_no_sendpage, | |
539 | }; | |
540 | ||
90ddc4f0 | 541 | static const struct proto_ops unix_dgram_ops = { |
1da177e4 LT |
542 | .family = PF_UNIX, |
543 | .owner = THIS_MODULE, | |
544 | .release = unix_release, | |
545 | .bind = unix_bind, | |
546 | .connect = unix_dgram_connect, | |
547 | .socketpair = unix_socketpair, | |
548 | .accept = sock_no_accept, | |
549 | .getname = unix_getname, | |
550 | .poll = datagram_poll, | |
551 | .ioctl = unix_ioctl, | |
552 | .listen = sock_no_listen, | |
553 | .shutdown = unix_shutdown, | |
554 | .setsockopt = sock_no_setsockopt, | |
555 | .getsockopt = sock_no_getsockopt, | |
556 | .sendmsg = unix_dgram_sendmsg, | |
557 | .recvmsg = unix_dgram_recvmsg, | |
558 | .mmap = sock_no_mmap, | |
559 | .sendpage = sock_no_sendpage, | |
560 | }; | |
561 | ||
90ddc4f0 | 562 | static const struct proto_ops unix_seqpacket_ops = { |
1da177e4 LT |
563 | .family = PF_UNIX, |
564 | .owner = THIS_MODULE, | |
565 | .release = unix_release, | |
566 | .bind = unix_bind, | |
567 | .connect = unix_stream_connect, | |
568 | .socketpair = unix_socketpair, | |
569 | .accept = unix_accept, | |
570 | .getname = unix_getname, | |
571 | .poll = datagram_poll, | |
572 | .ioctl = unix_ioctl, | |
573 | .listen = unix_listen, | |
574 | .shutdown = unix_shutdown, | |
575 | .setsockopt = sock_no_setsockopt, | |
576 | .getsockopt = sock_no_getsockopt, | |
577 | .sendmsg = unix_seqpacket_sendmsg, | |
578 | .recvmsg = unix_dgram_recvmsg, | |
579 | .mmap = sock_no_mmap, | |
580 | .sendpage = sock_no_sendpage, | |
581 | }; | |
582 | ||
583 | static struct proto unix_proto = { | |
584 | .name = "UNIX", | |
585 | .owner = THIS_MODULE, | |
586 | .obj_size = sizeof(struct unix_sock), | |
587 | }; | |
588 | ||
a09785a2 IM |
589 | /* |
590 | * AF_UNIX sockets do not interact with hardware, hence they | |
591 | * dont trigger interrupts - so it's safe for them to have | |
592 | * bh-unsafe locking for their sk_receive_queue.lock. Split off | |
593 | * this special lock-class by reinitializing the spinlock key: | |
594 | */ | |
595 | static struct lock_class_key af_unix_sk_receive_queue_lock_key; | |
596 | ||
1b8d7ae4 | 597 | static struct sock * unix_create1(struct net *net, struct socket *sock) |
1da177e4 LT |
598 | { |
599 | struct sock *sk = NULL; | |
600 | struct unix_sock *u; | |
601 | ||
529bf6be | 602 | if (atomic_read(&unix_nr_socks) >= 2*get_max_files()) |
1da177e4 LT |
603 | goto out; |
604 | ||
1b8d7ae4 | 605 | sk = sk_alloc(net, PF_UNIX, GFP_KERNEL, &unix_proto, 1); |
1da177e4 LT |
606 | if (!sk) |
607 | goto out; | |
608 | ||
609 | atomic_inc(&unix_nr_socks); | |
610 | ||
611 | sock_init_data(sock,sk); | |
a09785a2 IM |
612 | lockdep_set_class(&sk->sk_receive_queue.lock, |
613 | &af_unix_sk_receive_queue_lock_key); | |
1da177e4 LT |
614 | |
615 | sk->sk_write_space = unix_write_space; | |
616 | sk->sk_max_ack_backlog = sysctl_unix_max_dgram_qlen; | |
617 | sk->sk_destruct = unix_sock_destructor; | |
618 | u = unix_sk(sk); | |
619 | u->dentry = NULL; | |
620 | u->mnt = NULL; | |
fd19f329 | 621 | spin_lock_init(&u->lock); |
1fd05ba5 MS |
622 | atomic_set(&u->inflight, 0); |
623 | INIT_LIST_HEAD(&u->link); | |
57b47a53 | 624 | mutex_init(&u->readlock); /* single task reading lock */ |
1da177e4 LT |
625 | init_waitqueue_head(&u->peer_wait); |
626 | unix_insert_socket(unix_sockets_unbound, sk); | |
627 | out: | |
628 | return sk; | |
629 | } | |
630 | ||
1b8d7ae4 | 631 | static int unix_create(struct net *net, struct socket *sock, int protocol) |
1da177e4 | 632 | { |
1b8d7ae4 EB |
633 | if (net != &init_net) |
634 | return -EAFNOSUPPORT; | |
635 | ||
1da177e4 LT |
636 | if (protocol && protocol != PF_UNIX) |
637 | return -EPROTONOSUPPORT; | |
638 | ||
639 | sock->state = SS_UNCONNECTED; | |
640 | ||
641 | switch (sock->type) { | |
642 | case SOCK_STREAM: | |
643 | sock->ops = &unix_stream_ops; | |
644 | break; | |
645 | /* | |
646 | * Believe it or not BSD has AF_UNIX, SOCK_RAW though | |
647 | * nothing uses it. | |
648 | */ | |
649 | case SOCK_RAW: | |
650 | sock->type=SOCK_DGRAM; | |
651 | case SOCK_DGRAM: | |
652 | sock->ops = &unix_dgram_ops; | |
653 | break; | |
654 | case SOCK_SEQPACKET: | |
655 | sock->ops = &unix_seqpacket_ops; | |
656 | break; | |
657 | default: | |
658 | return -ESOCKTNOSUPPORT; | |
659 | } | |
660 | ||
1b8d7ae4 | 661 | return unix_create1(net, sock) ? 0 : -ENOMEM; |
1da177e4 LT |
662 | } |
663 | ||
664 | static int unix_release(struct socket *sock) | |
665 | { | |
666 | struct sock *sk = sock->sk; | |
667 | ||
668 | if (!sk) | |
669 | return 0; | |
670 | ||
671 | sock->sk = NULL; | |
672 | ||
673 | return unix_release_sock (sk, 0); | |
674 | } | |
675 | ||
676 | static int unix_autobind(struct socket *sock) | |
677 | { | |
678 | struct sock *sk = sock->sk; | |
679 | struct unix_sock *u = unix_sk(sk); | |
680 | static u32 ordernum = 1; | |
681 | struct unix_address * addr; | |
682 | int err; | |
683 | ||
57b47a53 | 684 | mutex_lock(&u->readlock); |
1da177e4 LT |
685 | |
686 | err = 0; | |
687 | if (u->addr) | |
688 | goto out; | |
689 | ||
690 | err = -ENOMEM; | |
0da974f4 | 691 | addr = kzalloc(sizeof(*addr) + sizeof(short) + 16, GFP_KERNEL); |
1da177e4 LT |
692 | if (!addr) |
693 | goto out; | |
694 | ||
1da177e4 LT |
695 | addr->name->sun_family = AF_UNIX; |
696 | atomic_set(&addr->refcnt, 1); | |
697 | ||
698 | retry: | |
699 | addr->len = sprintf(addr->name->sun_path+1, "%05x", ordernum) + 1 + sizeof(short); | |
700 | addr->hash = unix_hash_fold(csum_partial((void*)addr->name, addr->len, 0)); | |
701 | ||
fbe9cc4a | 702 | spin_lock(&unix_table_lock); |
1da177e4 LT |
703 | ordernum = (ordernum+1)&0xFFFFF; |
704 | ||
705 | if (__unix_find_socket_byname(addr->name, addr->len, sock->type, | |
706 | addr->hash)) { | |
fbe9cc4a | 707 | spin_unlock(&unix_table_lock); |
1da177e4 LT |
708 | /* Sanity yield. It is unusual case, but yet... */ |
709 | if (!(ordernum&0xFF)) | |
710 | yield(); | |
711 | goto retry; | |
712 | } | |
713 | addr->hash ^= sk->sk_type; | |
714 | ||
715 | __unix_remove_socket(sk); | |
716 | u->addr = addr; | |
717 | __unix_insert_socket(&unix_socket_table[addr->hash], sk); | |
fbe9cc4a | 718 | spin_unlock(&unix_table_lock); |
1da177e4 LT |
719 | err = 0; |
720 | ||
57b47a53 | 721 | out: mutex_unlock(&u->readlock); |
1da177e4 LT |
722 | return err; |
723 | } | |
724 | ||
725 | static struct sock *unix_find_other(struct sockaddr_un *sunname, int len, | |
726 | int type, unsigned hash, int *error) | |
727 | { | |
728 | struct sock *u; | |
729 | struct nameidata nd; | |
730 | int err = 0; | |
ac7bfa62 | 731 | |
1da177e4 LT |
732 | if (sunname->sun_path[0]) { |
733 | err = path_lookup(sunname->sun_path, LOOKUP_FOLLOW, &nd); | |
734 | if (err) | |
735 | goto fail; | |
e4543edd | 736 | err = vfs_permission(&nd, MAY_WRITE); |
1da177e4 LT |
737 | if (err) |
738 | goto put_fail; | |
739 | ||
740 | err = -ECONNREFUSED; | |
741 | if (!S_ISSOCK(nd.dentry->d_inode->i_mode)) | |
742 | goto put_fail; | |
743 | u=unix_find_socket_byinode(nd.dentry->d_inode); | |
744 | if (!u) | |
745 | goto put_fail; | |
746 | ||
747 | if (u->sk_type == type) | |
748 | touch_atime(nd.mnt, nd.dentry); | |
749 | ||
750 | path_release(&nd); | |
751 | ||
752 | err=-EPROTOTYPE; | |
753 | if (u->sk_type != type) { | |
754 | sock_put(u); | |
755 | goto fail; | |
756 | } | |
757 | } else { | |
758 | err = -ECONNREFUSED; | |
759 | u=unix_find_socket_byname(sunname, len, type, hash); | |
760 | if (u) { | |
761 | struct dentry *dentry; | |
762 | dentry = unix_sk(u)->dentry; | |
763 | if (dentry) | |
764 | touch_atime(unix_sk(u)->mnt, dentry); | |
765 | } else | |
766 | goto fail; | |
767 | } | |
768 | return u; | |
769 | ||
770 | put_fail: | |
771 | path_release(&nd); | |
772 | fail: | |
773 | *error=err; | |
774 | return NULL; | |
775 | } | |
776 | ||
777 | ||
778 | static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len) | |
779 | { | |
780 | struct sock *sk = sock->sk; | |
781 | struct unix_sock *u = unix_sk(sk); | |
782 | struct sockaddr_un *sunaddr=(struct sockaddr_un *)uaddr; | |
783 | struct dentry * dentry = NULL; | |
784 | struct nameidata nd; | |
785 | int err; | |
786 | unsigned hash; | |
787 | struct unix_address *addr; | |
788 | struct hlist_head *list; | |
789 | ||
790 | err = -EINVAL; | |
791 | if (sunaddr->sun_family != AF_UNIX) | |
792 | goto out; | |
793 | ||
794 | if (addr_len==sizeof(short)) { | |
795 | err = unix_autobind(sock); | |
796 | goto out; | |
797 | } | |
798 | ||
799 | err = unix_mkname(sunaddr, addr_len, &hash); | |
800 | if (err < 0) | |
801 | goto out; | |
802 | addr_len = err; | |
803 | ||
57b47a53 | 804 | mutex_lock(&u->readlock); |
1da177e4 LT |
805 | |
806 | err = -EINVAL; | |
807 | if (u->addr) | |
808 | goto out_up; | |
809 | ||
810 | err = -ENOMEM; | |
811 | addr = kmalloc(sizeof(*addr)+addr_len, GFP_KERNEL); | |
812 | if (!addr) | |
813 | goto out_up; | |
814 | ||
815 | memcpy(addr->name, sunaddr, addr_len); | |
816 | addr->len = addr_len; | |
817 | addr->hash = hash ^ sk->sk_type; | |
818 | atomic_set(&addr->refcnt, 1); | |
819 | ||
820 | if (sunaddr->sun_path[0]) { | |
821 | unsigned int mode; | |
822 | err = 0; | |
823 | /* | |
824 | * Get the parent directory, calculate the hash for last | |
825 | * component. | |
826 | */ | |
827 | err = path_lookup(sunaddr->sun_path, LOOKUP_PARENT, &nd); | |
828 | if (err) | |
829 | goto out_mknod_parent; | |
f81a0bff CH |
830 | |
831 | dentry = lookup_create(&nd, 0); | |
1da177e4 LT |
832 | err = PTR_ERR(dentry); |
833 | if (IS_ERR(dentry)) | |
834 | goto out_mknod_unlock; | |
f81a0bff | 835 | |
1da177e4 LT |
836 | /* |
837 | * All right, let's create it. | |
838 | */ | |
839 | mode = S_IFSOCK | | |
840 | (SOCK_INODE(sock)->i_mode & ~current->fs->umask); | |
841 | err = vfs_mknod(nd.dentry->d_inode, dentry, mode, 0); | |
842 | if (err) | |
843 | goto out_mknod_dput; | |
1b1dcc1b | 844 | mutex_unlock(&nd.dentry->d_inode->i_mutex); |
1da177e4 LT |
845 | dput(nd.dentry); |
846 | nd.dentry = dentry; | |
847 | ||
848 | addr->hash = UNIX_HASH_SIZE; | |
849 | } | |
850 | ||
fbe9cc4a | 851 | spin_lock(&unix_table_lock); |
1da177e4 LT |
852 | |
853 | if (!sunaddr->sun_path[0]) { | |
854 | err = -EADDRINUSE; | |
855 | if (__unix_find_socket_byname(sunaddr, addr_len, | |
856 | sk->sk_type, hash)) { | |
857 | unix_release_addr(addr); | |
858 | goto out_unlock; | |
859 | } | |
860 | ||
861 | list = &unix_socket_table[addr->hash]; | |
862 | } else { | |
863 | list = &unix_socket_table[dentry->d_inode->i_ino & (UNIX_HASH_SIZE-1)]; | |
864 | u->dentry = nd.dentry; | |
865 | u->mnt = nd.mnt; | |
866 | } | |
867 | ||
868 | err = 0; | |
869 | __unix_remove_socket(sk); | |
870 | u->addr = addr; | |
871 | __unix_insert_socket(list, sk); | |
872 | ||
873 | out_unlock: | |
fbe9cc4a | 874 | spin_unlock(&unix_table_lock); |
1da177e4 | 875 | out_up: |
57b47a53 | 876 | mutex_unlock(&u->readlock); |
1da177e4 LT |
877 | out: |
878 | return err; | |
879 | ||
880 | out_mknod_dput: | |
881 | dput(dentry); | |
882 | out_mknod_unlock: | |
1b1dcc1b | 883 | mutex_unlock(&nd.dentry->d_inode->i_mutex); |
1da177e4 LT |
884 | path_release(&nd); |
885 | out_mknod_parent: | |
886 | if (err==-EEXIST) | |
887 | err=-EADDRINUSE; | |
888 | unix_release_addr(addr); | |
889 | goto out_up; | |
890 | } | |
891 | ||
278a3de5 DM |
892 | static void unix_state_double_lock(struct sock *sk1, struct sock *sk2) |
893 | { | |
894 | if (unlikely(sk1 == sk2) || !sk2) { | |
895 | unix_state_lock(sk1); | |
896 | return; | |
897 | } | |
898 | if (sk1 < sk2) { | |
899 | unix_state_lock(sk1); | |
900 | unix_state_lock_nested(sk2); | |
901 | } else { | |
902 | unix_state_lock(sk2); | |
903 | unix_state_lock_nested(sk1); | |
904 | } | |
905 | } | |
906 | ||
907 | static void unix_state_double_unlock(struct sock *sk1, struct sock *sk2) | |
908 | { | |
909 | if (unlikely(sk1 == sk2) || !sk2) { | |
910 | unix_state_unlock(sk1); | |
911 | return; | |
912 | } | |
913 | unix_state_unlock(sk1); | |
914 | unix_state_unlock(sk2); | |
915 | } | |
916 | ||
1da177e4 LT |
917 | static int unix_dgram_connect(struct socket *sock, struct sockaddr *addr, |
918 | int alen, int flags) | |
919 | { | |
920 | struct sock *sk = sock->sk; | |
921 | struct sockaddr_un *sunaddr=(struct sockaddr_un*)addr; | |
922 | struct sock *other; | |
923 | unsigned hash; | |
924 | int err; | |
925 | ||
926 | if (addr->sa_family != AF_UNSPEC) { | |
927 | err = unix_mkname(sunaddr, alen, &hash); | |
928 | if (err < 0) | |
929 | goto out; | |
930 | alen = err; | |
931 | ||
932 | if (test_bit(SOCK_PASSCRED, &sock->flags) && | |
933 | !unix_sk(sk)->addr && (err = unix_autobind(sock)) != 0) | |
934 | goto out; | |
935 | ||
278a3de5 | 936 | restart: |
1da177e4 LT |
937 | other=unix_find_other(sunaddr, alen, sock->type, hash, &err); |
938 | if (!other) | |
939 | goto out; | |
940 | ||
278a3de5 DM |
941 | unix_state_double_lock(sk, other); |
942 | ||
943 | /* Apparently VFS overslept socket death. Retry. */ | |
944 | if (sock_flag(other, SOCK_DEAD)) { | |
945 | unix_state_double_unlock(sk, other); | |
946 | sock_put(other); | |
947 | goto restart; | |
948 | } | |
1da177e4 LT |
949 | |
950 | err = -EPERM; | |
951 | if (!unix_may_send(sk, other)) | |
952 | goto out_unlock; | |
953 | ||
954 | err = security_unix_may_send(sk->sk_socket, other->sk_socket); | |
955 | if (err) | |
956 | goto out_unlock; | |
957 | ||
958 | } else { | |
959 | /* | |
960 | * 1003.1g breaking connected state with AF_UNSPEC | |
961 | */ | |
962 | other = NULL; | |
278a3de5 | 963 | unix_state_double_lock(sk, other); |
1da177e4 LT |
964 | } |
965 | ||
966 | /* | |
967 | * If it was connected, reconnect. | |
968 | */ | |
969 | if (unix_peer(sk)) { | |
970 | struct sock *old_peer = unix_peer(sk); | |
971 | unix_peer(sk)=other; | |
278a3de5 | 972 | unix_state_double_unlock(sk, other); |
1da177e4 LT |
973 | |
974 | if (other != old_peer) | |
975 | unix_dgram_disconnected(sk, old_peer); | |
976 | sock_put(old_peer); | |
977 | } else { | |
978 | unix_peer(sk)=other; | |
278a3de5 | 979 | unix_state_double_unlock(sk, other); |
1da177e4 | 980 | } |
ac7bfa62 | 981 | return 0; |
1da177e4 LT |
982 | |
983 | out_unlock: | |
278a3de5 | 984 | unix_state_double_unlock(sk, other); |
1da177e4 LT |
985 | sock_put(other); |
986 | out: | |
987 | return err; | |
988 | } | |
989 | ||
990 | static long unix_wait_for_peer(struct sock *other, long timeo) | |
991 | { | |
992 | struct unix_sock *u = unix_sk(other); | |
993 | int sched; | |
994 | DEFINE_WAIT(wait); | |
995 | ||
996 | prepare_to_wait_exclusive(&u->peer_wait, &wait, TASK_INTERRUPTIBLE); | |
997 | ||
998 | sched = !sock_flag(other, SOCK_DEAD) && | |
999 | !(other->sk_shutdown & RCV_SHUTDOWN) && | |
64a14651 | 1000 | (skb_queue_len(&other->sk_receive_queue) > |
1da177e4 LT |
1001 | other->sk_max_ack_backlog); |
1002 | ||
1c92b4e5 | 1003 | unix_state_unlock(other); |
1da177e4 LT |
1004 | |
1005 | if (sched) | |
1006 | timeo = schedule_timeout(timeo); | |
1007 | ||
1008 | finish_wait(&u->peer_wait, &wait); | |
1009 | return timeo; | |
1010 | } | |
1011 | ||
1012 | static int unix_stream_connect(struct socket *sock, struct sockaddr *uaddr, | |
1013 | int addr_len, int flags) | |
1014 | { | |
1015 | struct sockaddr_un *sunaddr=(struct sockaddr_un *)uaddr; | |
1016 | struct sock *sk = sock->sk; | |
1017 | struct unix_sock *u = unix_sk(sk), *newu, *otheru; | |
1018 | struct sock *newsk = NULL; | |
1019 | struct sock *other = NULL; | |
1020 | struct sk_buff *skb = NULL; | |
1021 | unsigned hash; | |
1022 | int st; | |
1023 | int err; | |
1024 | long timeo; | |
1025 | ||
1026 | err = unix_mkname(sunaddr, addr_len, &hash); | |
1027 | if (err < 0) | |
1028 | goto out; | |
1029 | addr_len = err; | |
1030 | ||
1031 | if (test_bit(SOCK_PASSCRED, &sock->flags) | |
1032 | && !u->addr && (err = unix_autobind(sock)) != 0) | |
1033 | goto out; | |
1034 | ||
1035 | timeo = sock_sndtimeo(sk, flags & O_NONBLOCK); | |
1036 | ||
1037 | /* First of all allocate resources. | |
1038 | If we will make it after state is locked, | |
1039 | we will have to recheck all again in any case. | |
1040 | */ | |
1041 | ||
1042 | err = -ENOMEM; | |
1043 | ||
1044 | /* create new sock for complete connection */ | |
1b8d7ae4 | 1045 | newsk = unix_create1(sk->sk_net, NULL); |
1da177e4 LT |
1046 | if (newsk == NULL) |
1047 | goto out; | |
1048 | ||
1049 | /* Allocate skb for sending to listening sock */ | |
1050 | skb = sock_wmalloc(newsk, 1, 0, GFP_KERNEL); | |
1051 | if (skb == NULL) | |
1052 | goto out; | |
1053 | ||
1054 | restart: | |
1055 | /* Find listening sock. */ | |
1056 | other = unix_find_other(sunaddr, addr_len, sk->sk_type, hash, &err); | |
1057 | if (!other) | |
1058 | goto out; | |
1059 | ||
1060 | /* Latch state of peer */ | |
1c92b4e5 | 1061 | unix_state_lock(other); |
1da177e4 LT |
1062 | |
1063 | /* Apparently VFS overslept socket death. Retry. */ | |
1064 | if (sock_flag(other, SOCK_DEAD)) { | |
1c92b4e5 | 1065 | unix_state_unlock(other); |
1da177e4 LT |
1066 | sock_put(other); |
1067 | goto restart; | |
1068 | } | |
1069 | ||
1070 | err = -ECONNREFUSED; | |
1071 | if (other->sk_state != TCP_LISTEN) | |
1072 | goto out_unlock; | |
1073 | ||
64a14651 | 1074 | if (skb_queue_len(&other->sk_receive_queue) > |
1da177e4 LT |
1075 | other->sk_max_ack_backlog) { |
1076 | err = -EAGAIN; | |
1077 | if (!timeo) | |
1078 | goto out_unlock; | |
1079 | ||
1080 | timeo = unix_wait_for_peer(other, timeo); | |
1081 | ||
1082 | err = sock_intr_errno(timeo); | |
1083 | if (signal_pending(current)) | |
1084 | goto out; | |
1085 | sock_put(other); | |
1086 | goto restart; | |
ac7bfa62 | 1087 | } |
1da177e4 LT |
1088 | |
1089 | /* Latch our state. | |
1090 | ||
1091 | It is tricky place. We need to grab write lock and cannot | |
1092 | drop lock on peer. It is dangerous because deadlock is | |
1093 | possible. Connect to self case and simultaneous | |
1094 | attempt to connect are eliminated by checking socket | |
1095 | state. other is TCP_LISTEN, if sk is TCP_LISTEN we | |
1096 | check this before attempt to grab lock. | |
1097 | ||
1098 | Well, and we have to recheck the state after socket locked. | |
1099 | */ | |
1100 | st = sk->sk_state; | |
1101 | ||
1102 | switch (st) { | |
1103 | case TCP_CLOSE: | |
1104 | /* This is ok... continue with connect */ | |
1105 | break; | |
1106 | case TCP_ESTABLISHED: | |
1107 | /* Socket is already connected */ | |
1108 | err = -EISCONN; | |
1109 | goto out_unlock; | |
1110 | default: | |
1111 | err = -EINVAL; | |
1112 | goto out_unlock; | |
1113 | } | |
1114 | ||
1c92b4e5 | 1115 | unix_state_lock_nested(sk); |
1da177e4 LT |
1116 | |
1117 | if (sk->sk_state != st) { | |
1c92b4e5 DM |
1118 | unix_state_unlock(sk); |
1119 | unix_state_unlock(other); | |
1da177e4 LT |
1120 | sock_put(other); |
1121 | goto restart; | |
1122 | } | |
1123 | ||
1124 | err = security_unix_stream_connect(sock, other->sk_socket, newsk); | |
1125 | if (err) { | |
1c92b4e5 | 1126 | unix_state_unlock(sk); |
1da177e4 LT |
1127 | goto out_unlock; |
1128 | } | |
1129 | ||
1130 | /* The way is open! Fastly set all the necessary fields... */ | |
1131 | ||
1132 | sock_hold(sk); | |
1133 | unix_peer(newsk) = sk; | |
1134 | newsk->sk_state = TCP_ESTABLISHED; | |
1135 | newsk->sk_type = sk->sk_type; | |
b488893a | 1136 | newsk->sk_peercred.pid = task_tgid_vnr(current); |
1da177e4 LT |
1137 | newsk->sk_peercred.uid = current->euid; |
1138 | newsk->sk_peercred.gid = current->egid; | |
1139 | newu = unix_sk(newsk); | |
1140 | newsk->sk_sleep = &newu->peer_wait; | |
1141 | otheru = unix_sk(other); | |
1142 | ||
1143 | /* copy address information from listening to new sock*/ | |
1144 | if (otheru->addr) { | |
1145 | atomic_inc(&otheru->addr->refcnt); | |
1146 | newu->addr = otheru->addr; | |
1147 | } | |
1148 | if (otheru->dentry) { | |
1149 | newu->dentry = dget(otheru->dentry); | |
1150 | newu->mnt = mntget(otheru->mnt); | |
1151 | } | |
1152 | ||
1153 | /* Set credentials */ | |
1154 | sk->sk_peercred = other->sk_peercred; | |
1155 | ||
1da177e4 LT |
1156 | sock->state = SS_CONNECTED; |
1157 | sk->sk_state = TCP_ESTABLISHED; | |
830a1e5c BL |
1158 | sock_hold(newsk); |
1159 | ||
1160 | smp_mb__after_atomic_inc(); /* sock_hold() does an atomic_inc() */ | |
1161 | unix_peer(sk) = newsk; | |
1da177e4 | 1162 | |
1c92b4e5 | 1163 | unix_state_unlock(sk); |
1da177e4 LT |
1164 | |
1165 | /* take ten and and send info to listening sock */ | |
1166 | spin_lock(&other->sk_receive_queue.lock); | |
1167 | __skb_queue_tail(&other->sk_receive_queue, skb); | |
1da177e4 | 1168 | spin_unlock(&other->sk_receive_queue.lock); |
1c92b4e5 | 1169 | unix_state_unlock(other); |
1da177e4 LT |
1170 | other->sk_data_ready(other, 0); |
1171 | sock_put(other); | |
1172 | return 0; | |
1173 | ||
1174 | out_unlock: | |
1175 | if (other) | |
1c92b4e5 | 1176 | unix_state_unlock(other); |
1da177e4 LT |
1177 | |
1178 | out: | |
1179 | if (skb) | |
1180 | kfree_skb(skb); | |
1181 | if (newsk) | |
1182 | unix_release_sock(newsk, 0); | |
1183 | if (other) | |
1184 | sock_put(other); | |
1185 | return err; | |
1186 | } | |
1187 | ||
1188 | static int unix_socketpair(struct socket *socka, struct socket *sockb) | |
1189 | { | |
1190 | struct sock *ska=socka->sk, *skb = sockb->sk; | |
1191 | ||
1192 | /* Join our sockets back to back */ | |
1193 | sock_hold(ska); | |
1194 | sock_hold(skb); | |
1195 | unix_peer(ska)=skb; | |
1196 | unix_peer(skb)=ska; | |
b488893a | 1197 | ska->sk_peercred.pid = skb->sk_peercred.pid = task_tgid_vnr(current); |
1da177e4 LT |
1198 | ska->sk_peercred.uid = skb->sk_peercred.uid = current->euid; |
1199 | ska->sk_peercred.gid = skb->sk_peercred.gid = current->egid; | |
1200 | ||
1201 | if (ska->sk_type != SOCK_DGRAM) { | |
1202 | ska->sk_state = TCP_ESTABLISHED; | |
1203 | skb->sk_state = TCP_ESTABLISHED; | |
1204 | socka->state = SS_CONNECTED; | |
1205 | sockb->state = SS_CONNECTED; | |
1206 | } | |
1207 | return 0; | |
1208 | } | |
1209 | ||
1210 | static int unix_accept(struct socket *sock, struct socket *newsock, int flags) | |
1211 | { | |
1212 | struct sock *sk = sock->sk; | |
1213 | struct sock *tsk; | |
1214 | struct sk_buff *skb; | |
1215 | int err; | |
1216 | ||
1217 | err = -EOPNOTSUPP; | |
1218 | if (sock->type!=SOCK_STREAM && sock->type!=SOCK_SEQPACKET) | |
1219 | goto out; | |
1220 | ||
1221 | err = -EINVAL; | |
1222 | if (sk->sk_state != TCP_LISTEN) | |
1223 | goto out; | |
1224 | ||
1225 | /* If socket state is TCP_LISTEN it cannot change (for now...), | |
1226 | * so that no locks are necessary. | |
1227 | */ | |
1228 | ||
1229 | skb = skb_recv_datagram(sk, 0, flags&O_NONBLOCK, &err); | |
1230 | if (!skb) { | |
1231 | /* This means receive shutdown. */ | |
1232 | if (err == 0) | |
1233 | err = -EINVAL; | |
1234 | goto out; | |
1235 | } | |
1236 | ||
1237 | tsk = skb->sk; | |
1238 | skb_free_datagram(sk, skb); | |
1239 | wake_up_interruptible(&unix_sk(sk)->peer_wait); | |
1240 | ||
1241 | /* attach accepted sock to socket */ | |
1c92b4e5 | 1242 | unix_state_lock(tsk); |
1da177e4 LT |
1243 | newsock->state = SS_CONNECTED; |
1244 | sock_graft(tsk, newsock); | |
1c92b4e5 | 1245 | unix_state_unlock(tsk); |
1da177e4 LT |
1246 | return 0; |
1247 | ||
1248 | out: | |
1249 | return err; | |
1250 | } | |
1251 | ||
1252 | ||
1253 | static int unix_getname(struct socket *sock, struct sockaddr *uaddr, int *uaddr_len, int peer) | |
1254 | { | |
1255 | struct sock *sk = sock->sk; | |
1256 | struct unix_sock *u; | |
1257 | struct sockaddr_un *sunaddr=(struct sockaddr_un *)uaddr; | |
1258 | int err = 0; | |
1259 | ||
1260 | if (peer) { | |
1261 | sk = unix_peer_get(sk); | |
1262 | ||
1263 | err = -ENOTCONN; | |
1264 | if (!sk) | |
1265 | goto out; | |
1266 | err = 0; | |
1267 | } else { | |
1268 | sock_hold(sk); | |
1269 | } | |
1270 | ||
1271 | u = unix_sk(sk); | |
1c92b4e5 | 1272 | unix_state_lock(sk); |
1da177e4 LT |
1273 | if (!u->addr) { |
1274 | sunaddr->sun_family = AF_UNIX; | |
1275 | sunaddr->sun_path[0] = 0; | |
1276 | *uaddr_len = sizeof(short); | |
1277 | } else { | |
1278 | struct unix_address *addr = u->addr; | |
1279 | ||
1280 | *uaddr_len = addr->len; | |
1281 | memcpy(sunaddr, addr->name, *uaddr_len); | |
1282 | } | |
1c92b4e5 | 1283 | unix_state_unlock(sk); |
1da177e4 LT |
1284 | sock_put(sk); |
1285 | out: | |
1286 | return err; | |
1287 | } | |
1288 | ||
1289 | static void unix_detach_fds(struct scm_cookie *scm, struct sk_buff *skb) | |
1290 | { | |
1291 | int i; | |
1292 | ||
1293 | scm->fp = UNIXCB(skb).fp; | |
1294 | skb->destructor = sock_wfree; | |
1295 | UNIXCB(skb).fp = NULL; | |
1296 | ||
1297 | for (i=scm->fp->count-1; i>=0; i--) | |
1298 | unix_notinflight(scm->fp->fp[i]); | |
1299 | } | |
1300 | ||
1301 | static void unix_destruct_fds(struct sk_buff *skb) | |
1302 | { | |
1303 | struct scm_cookie scm; | |
1304 | memset(&scm, 0, sizeof(scm)); | |
1305 | unix_detach_fds(&scm, skb); | |
1306 | ||
1307 | /* Alas, it calls VFS */ | |
1308 | /* So fscking what? fput() had been SMP-safe since the last Summer */ | |
1309 | scm_destroy(&scm); | |
1310 | sock_wfree(skb); | |
1311 | } | |
1312 | ||
1313 | static void unix_attach_fds(struct scm_cookie *scm, struct sk_buff *skb) | |
1314 | { | |
1315 | int i; | |
1316 | for (i=scm->fp->count-1; i>=0; i--) | |
1317 | unix_inflight(scm->fp->fp[i]); | |
1318 | UNIXCB(skb).fp = scm->fp; | |
1319 | skb->destructor = unix_destruct_fds; | |
1320 | scm->fp = NULL; | |
1321 | } | |
1322 | ||
1323 | /* | |
1324 | * Send AF_UNIX data. | |
1325 | */ | |
1326 | ||
1327 | static int unix_dgram_sendmsg(struct kiocb *kiocb, struct socket *sock, | |
1328 | struct msghdr *msg, size_t len) | |
1329 | { | |
1330 | struct sock_iocb *siocb = kiocb_to_siocb(kiocb); | |
1331 | struct sock *sk = sock->sk; | |
1332 | struct unix_sock *u = unix_sk(sk); | |
1333 | struct sockaddr_un *sunaddr=msg->msg_name; | |
1334 | struct sock *other = NULL; | |
1335 | int namelen = 0; /* fake GCC */ | |
1336 | int err; | |
1337 | unsigned hash; | |
1338 | struct sk_buff *skb; | |
1339 | long timeo; | |
1340 | struct scm_cookie tmp_scm; | |
1341 | ||
1342 | if (NULL == siocb->scm) | |
1343 | siocb->scm = &tmp_scm; | |
1344 | err = scm_send(sock, msg, siocb->scm); | |
1345 | if (err < 0) | |
1346 | return err; | |
1347 | ||
1348 | err = -EOPNOTSUPP; | |
1349 | if (msg->msg_flags&MSG_OOB) | |
1350 | goto out; | |
1351 | ||
1352 | if (msg->msg_namelen) { | |
1353 | err = unix_mkname(sunaddr, msg->msg_namelen, &hash); | |
1354 | if (err < 0) | |
1355 | goto out; | |
1356 | namelen = err; | |
1357 | } else { | |
1358 | sunaddr = NULL; | |
1359 | err = -ENOTCONN; | |
1360 | other = unix_peer_get(sk); | |
1361 | if (!other) | |
1362 | goto out; | |
1363 | } | |
1364 | ||
1365 | if (test_bit(SOCK_PASSCRED, &sock->flags) | |
1366 | && !u->addr && (err = unix_autobind(sock)) != 0) | |
1367 | goto out; | |
1368 | ||
1369 | err = -EMSGSIZE; | |
1370 | if (len > sk->sk_sndbuf - 32) | |
1371 | goto out; | |
1372 | ||
1373 | skb = sock_alloc_send_skb(sk, len, msg->msg_flags&MSG_DONTWAIT, &err); | |
1374 | if (skb==NULL) | |
1375 | goto out; | |
1376 | ||
1377 | memcpy(UNIXCREDS(skb), &siocb->scm->creds, sizeof(struct ucred)); | |
1378 | if (siocb->scm->fp) | |
1379 | unix_attach_fds(siocb->scm, skb); | |
dc49c1f9 | 1380 | unix_get_secdata(siocb->scm, skb); |
877ce7c1 | 1381 | |
badff6d0 | 1382 | skb_reset_transport_header(skb); |
1da177e4 LT |
1383 | err = memcpy_fromiovec(skb_put(skb,len), msg->msg_iov, len); |
1384 | if (err) | |
1385 | goto out_free; | |
1386 | ||
1387 | timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT); | |
1388 | ||
1389 | restart: | |
1390 | if (!other) { | |
1391 | err = -ECONNRESET; | |
1392 | if (sunaddr == NULL) | |
1393 | goto out_free; | |
1394 | ||
1395 | other = unix_find_other(sunaddr, namelen, sk->sk_type, | |
1396 | hash, &err); | |
1397 | if (other==NULL) | |
1398 | goto out_free; | |
1399 | } | |
1400 | ||
1c92b4e5 | 1401 | unix_state_lock(other); |
1da177e4 LT |
1402 | err = -EPERM; |
1403 | if (!unix_may_send(sk, other)) | |
1404 | goto out_unlock; | |
1405 | ||
1406 | if (sock_flag(other, SOCK_DEAD)) { | |
1407 | /* | |
1408 | * Check with 1003.1g - what should | |
1409 | * datagram error | |
1410 | */ | |
1c92b4e5 | 1411 | unix_state_unlock(other); |
1da177e4 LT |
1412 | sock_put(other); |
1413 | ||
1414 | err = 0; | |
1c92b4e5 | 1415 | unix_state_lock(sk); |
1da177e4 LT |
1416 | if (unix_peer(sk) == other) { |
1417 | unix_peer(sk)=NULL; | |
1c92b4e5 | 1418 | unix_state_unlock(sk); |
1da177e4 LT |
1419 | |
1420 | unix_dgram_disconnected(sk, other); | |
1421 | sock_put(other); | |
1422 | err = -ECONNREFUSED; | |
1423 | } else { | |
1c92b4e5 | 1424 | unix_state_unlock(sk); |
1da177e4 LT |
1425 | } |
1426 | ||
1427 | other = NULL; | |
1428 | if (err) | |
1429 | goto out_free; | |
1430 | goto restart; | |
1431 | } | |
1432 | ||
1433 | err = -EPIPE; | |
1434 | if (other->sk_shutdown & RCV_SHUTDOWN) | |
1435 | goto out_unlock; | |
1436 | ||
1437 | if (sk->sk_type != SOCK_SEQPACKET) { | |
1438 | err = security_unix_may_send(sk->sk_socket, other->sk_socket); | |
1439 | if (err) | |
1440 | goto out_unlock; | |
1441 | } | |
1442 | ||
1443 | if (unix_peer(other) != sk && | |
64a14651 | 1444 | (skb_queue_len(&other->sk_receive_queue) > |
1da177e4 LT |
1445 | other->sk_max_ack_backlog)) { |
1446 | if (!timeo) { | |
1447 | err = -EAGAIN; | |
1448 | goto out_unlock; | |
1449 | } | |
1450 | ||
1451 | timeo = unix_wait_for_peer(other, timeo); | |
1452 | ||
1453 | err = sock_intr_errno(timeo); | |
1454 | if (signal_pending(current)) | |
1455 | goto out_free; | |
1456 | ||
1457 | goto restart; | |
1458 | } | |
1459 | ||
1460 | skb_queue_tail(&other->sk_receive_queue, skb); | |
1c92b4e5 | 1461 | unix_state_unlock(other); |
1da177e4 LT |
1462 | other->sk_data_ready(other, len); |
1463 | sock_put(other); | |
1464 | scm_destroy(siocb->scm); | |
1465 | return len; | |
1466 | ||
1467 | out_unlock: | |
1c92b4e5 | 1468 | unix_state_unlock(other); |
1da177e4 LT |
1469 | out_free: |
1470 | kfree_skb(skb); | |
1471 | out: | |
1472 | if (other) | |
1473 | sock_put(other); | |
1474 | scm_destroy(siocb->scm); | |
1475 | return err; | |
1476 | } | |
1477 | ||
ac7bfa62 | 1478 | |
1da177e4 LT |
1479 | static int unix_stream_sendmsg(struct kiocb *kiocb, struct socket *sock, |
1480 | struct msghdr *msg, size_t len) | |
1481 | { | |
1482 | struct sock_iocb *siocb = kiocb_to_siocb(kiocb); | |
1483 | struct sock *sk = sock->sk; | |
1484 | struct sock *other = NULL; | |
1485 | struct sockaddr_un *sunaddr=msg->msg_name; | |
1486 | int err,size; | |
1487 | struct sk_buff *skb; | |
1488 | int sent=0; | |
1489 | struct scm_cookie tmp_scm; | |
1490 | ||
1491 | if (NULL == siocb->scm) | |
1492 | siocb->scm = &tmp_scm; | |
1493 | err = scm_send(sock, msg, siocb->scm); | |
1494 | if (err < 0) | |
1495 | return err; | |
1496 | ||
1497 | err = -EOPNOTSUPP; | |
1498 | if (msg->msg_flags&MSG_OOB) | |
1499 | goto out_err; | |
1500 | ||
1501 | if (msg->msg_namelen) { | |
1502 | err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP; | |
1503 | goto out_err; | |
1504 | } else { | |
1505 | sunaddr = NULL; | |
1506 | err = -ENOTCONN; | |
830a1e5c | 1507 | other = unix_peer(sk); |
1da177e4 LT |
1508 | if (!other) |
1509 | goto out_err; | |
1510 | } | |
1511 | ||
1512 | if (sk->sk_shutdown & SEND_SHUTDOWN) | |
1513 | goto pipe_err; | |
1514 | ||
1515 | while(sent < len) | |
1516 | { | |
1517 | /* | |
e9df7d7f BL |
1518 | * Optimisation for the fact that under 0.01% of X |
1519 | * messages typically need breaking up. | |
1da177e4 LT |
1520 | */ |
1521 | ||
e9df7d7f | 1522 | size = len-sent; |
1da177e4 LT |
1523 | |
1524 | /* Keep two messages in the pipe so it schedules better */ | |
e9df7d7f BL |
1525 | if (size > ((sk->sk_sndbuf >> 1) - 64)) |
1526 | size = (sk->sk_sndbuf >> 1) - 64; | |
1da177e4 LT |
1527 | |
1528 | if (size > SKB_MAX_ALLOC) | |
1529 | size = SKB_MAX_ALLOC; | |
ac7bfa62 | 1530 | |
1da177e4 LT |
1531 | /* |
1532 | * Grab a buffer | |
1533 | */ | |
ac7bfa62 | 1534 | |
1da177e4 LT |
1535 | skb=sock_alloc_send_skb(sk,size,msg->msg_flags&MSG_DONTWAIT, &err); |
1536 | ||
1537 | if (skb==NULL) | |
1538 | goto out_err; | |
1539 | ||
1540 | /* | |
1541 | * If you pass two values to the sock_alloc_send_skb | |
1542 | * it tries to grab the large buffer with GFP_NOFS | |
1543 | * (which can fail easily), and if it fails grab the | |
1544 | * fallback size buffer which is under a page and will | |
1545 | * succeed. [Alan] | |
1546 | */ | |
1547 | size = min_t(int, size, skb_tailroom(skb)); | |
1548 | ||
1549 | memcpy(UNIXCREDS(skb), &siocb->scm->creds, sizeof(struct ucred)); | |
1550 | if (siocb->scm->fp) | |
1551 | unix_attach_fds(siocb->scm, skb); | |
1552 | ||
1553 | if ((err = memcpy_fromiovec(skb_put(skb,size), msg->msg_iov, size)) != 0) { | |
1554 | kfree_skb(skb); | |
1555 | goto out_err; | |
1556 | } | |
1557 | ||
1c92b4e5 | 1558 | unix_state_lock(other); |
1da177e4 LT |
1559 | |
1560 | if (sock_flag(other, SOCK_DEAD) || | |
1561 | (other->sk_shutdown & RCV_SHUTDOWN)) | |
1562 | goto pipe_err_free; | |
1563 | ||
1564 | skb_queue_tail(&other->sk_receive_queue, skb); | |
1c92b4e5 | 1565 | unix_state_unlock(other); |
1da177e4 LT |
1566 | other->sk_data_ready(other, size); |
1567 | sent+=size; | |
1568 | } | |
1da177e4 LT |
1569 | |
1570 | scm_destroy(siocb->scm); | |
1571 | siocb->scm = NULL; | |
1572 | ||
1573 | return sent; | |
1574 | ||
1575 | pipe_err_free: | |
1c92b4e5 | 1576 | unix_state_unlock(other); |
1da177e4 LT |
1577 | kfree_skb(skb); |
1578 | pipe_err: | |
1579 | if (sent==0 && !(msg->msg_flags&MSG_NOSIGNAL)) | |
1580 | send_sig(SIGPIPE,current,0); | |
1581 | err = -EPIPE; | |
1582 | out_err: | |
1da177e4 LT |
1583 | scm_destroy(siocb->scm); |
1584 | siocb->scm = NULL; | |
1585 | return sent ? : err; | |
1586 | } | |
1587 | ||
1588 | static int unix_seqpacket_sendmsg(struct kiocb *kiocb, struct socket *sock, | |
1589 | struct msghdr *msg, size_t len) | |
1590 | { | |
1591 | int err; | |
1592 | struct sock *sk = sock->sk; | |
ac7bfa62 | 1593 | |
1da177e4 LT |
1594 | err = sock_error(sk); |
1595 | if (err) | |
1596 | return err; | |
1597 | ||
1598 | if (sk->sk_state != TCP_ESTABLISHED) | |
1599 | return -ENOTCONN; | |
1600 | ||
1601 | if (msg->msg_namelen) | |
1602 | msg->msg_namelen = 0; | |
1603 | ||
1604 | return unix_dgram_sendmsg(kiocb, sock, msg, len); | |
1605 | } | |
ac7bfa62 | 1606 | |
1da177e4 LT |
1607 | static void unix_copy_addr(struct msghdr *msg, struct sock *sk) |
1608 | { | |
1609 | struct unix_sock *u = unix_sk(sk); | |
1610 | ||
1611 | msg->msg_namelen = 0; | |
1612 | if (u->addr) { | |
1613 | msg->msg_namelen = u->addr->len; | |
1614 | memcpy(msg->msg_name, u->addr->name, u->addr->len); | |
1615 | } | |
1616 | } | |
1617 | ||
1618 | static int unix_dgram_recvmsg(struct kiocb *iocb, struct socket *sock, | |
1619 | struct msghdr *msg, size_t size, | |
1620 | int flags) | |
1621 | { | |
1622 | struct sock_iocb *siocb = kiocb_to_siocb(iocb); | |
1623 | struct scm_cookie tmp_scm; | |
1624 | struct sock *sk = sock->sk; | |
1625 | struct unix_sock *u = unix_sk(sk); | |
1626 | int noblock = flags & MSG_DONTWAIT; | |
1627 | struct sk_buff *skb; | |
1628 | int err; | |
1629 | ||
1630 | err = -EOPNOTSUPP; | |
1631 | if (flags&MSG_OOB) | |
1632 | goto out; | |
1633 | ||
1634 | msg->msg_namelen = 0; | |
1635 | ||
57b47a53 | 1636 | mutex_lock(&u->readlock); |
1da177e4 LT |
1637 | |
1638 | skb = skb_recv_datagram(sk, flags, noblock, &err); | |
1639 | if (!skb) | |
1640 | goto out_unlock; | |
1641 | ||
71e20f18 | 1642 | wake_up_interruptible_sync(&u->peer_wait); |
1da177e4 LT |
1643 | |
1644 | if (msg->msg_name) | |
1645 | unix_copy_addr(msg, skb->sk); | |
1646 | ||
1647 | if (size > skb->len) | |
1648 | size = skb->len; | |
1649 | else if (size < skb->len) | |
1650 | msg->msg_flags |= MSG_TRUNC; | |
1651 | ||
1652 | err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, size); | |
1653 | if (err) | |
1654 | goto out_free; | |
1655 | ||
1656 | if (!siocb->scm) { | |
1657 | siocb->scm = &tmp_scm; | |
1658 | memset(&tmp_scm, 0, sizeof(tmp_scm)); | |
1659 | } | |
1660 | siocb->scm->creds = *UNIXCREDS(skb); | |
877ce7c1 | 1661 | unix_set_secdata(siocb->scm, skb); |
1da177e4 LT |
1662 | |
1663 | if (!(flags & MSG_PEEK)) | |
1664 | { | |
1665 | if (UNIXCB(skb).fp) | |
1666 | unix_detach_fds(siocb->scm, skb); | |
1667 | } | |
ac7bfa62 | 1668 | else |
1da177e4 LT |
1669 | { |
1670 | /* It is questionable: on PEEK we could: | |
1671 | - do not return fds - good, but too simple 8) | |
1672 | - return fds, and do not return them on read (old strategy, | |
1673 | apparently wrong) | |
1674 | - clone fds (I chose it for now, it is the most universal | |
1675 | solution) | |
ac7bfa62 YH |
1676 | |
1677 | POSIX 1003.1g does not actually define this clearly | |
1678 | at all. POSIX 1003.1g doesn't define a lot of things | |
1679 | clearly however! | |
1680 | ||
1da177e4 LT |
1681 | */ |
1682 | if (UNIXCB(skb).fp) | |
1683 | siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp); | |
1684 | } | |
1685 | err = size; | |
1686 | ||
1687 | scm_recv(sock, msg, siocb->scm, flags); | |
1688 | ||
1689 | out_free: | |
1690 | skb_free_datagram(sk,skb); | |
1691 | out_unlock: | |
57b47a53 | 1692 | mutex_unlock(&u->readlock); |
1da177e4 LT |
1693 | out: |
1694 | return err; | |
1695 | } | |
1696 | ||
1697 | /* | |
1698 | * Sleep until data has arrive. But check for races.. | |
1699 | */ | |
ac7bfa62 | 1700 | |
1da177e4 LT |
1701 | static long unix_stream_data_wait(struct sock * sk, long timeo) |
1702 | { | |
1703 | DEFINE_WAIT(wait); | |
1704 | ||
1c92b4e5 | 1705 | unix_state_lock(sk); |
1da177e4 LT |
1706 | |
1707 | for (;;) { | |
1708 | prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE); | |
1709 | ||
b03efcfb | 1710 | if (!skb_queue_empty(&sk->sk_receive_queue) || |
1da177e4 LT |
1711 | sk->sk_err || |
1712 | (sk->sk_shutdown & RCV_SHUTDOWN) || | |
1713 | signal_pending(current) || | |
1714 | !timeo) | |
1715 | break; | |
1716 | ||
1717 | set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags); | |
1c92b4e5 | 1718 | unix_state_unlock(sk); |
1da177e4 | 1719 | timeo = schedule_timeout(timeo); |
1c92b4e5 | 1720 | unix_state_lock(sk); |
1da177e4 LT |
1721 | clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags); |
1722 | } | |
1723 | ||
1724 | finish_wait(sk->sk_sleep, &wait); | |
1c92b4e5 | 1725 | unix_state_unlock(sk); |
1da177e4 LT |
1726 | return timeo; |
1727 | } | |
1728 | ||
1729 | ||
1730 | ||
1731 | static int unix_stream_recvmsg(struct kiocb *iocb, struct socket *sock, | |
1732 | struct msghdr *msg, size_t size, | |
1733 | int flags) | |
1734 | { | |
1735 | struct sock_iocb *siocb = kiocb_to_siocb(iocb); | |
1736 | struct scm_cookie tmp_scm; | |
1737 | struct sock *sk = sock->sk; | |
1738 | struct unix_sock *u = unix_sk(sk); | |
1739 | struct sockaddr_un *sunaddr=msg->msg_name; | |
1740 | int copied = 0; | |
1741 | int check_creds = 0; | |
1742 | int target; | |
1743 | int err = 0; | |
1744 | long timeo; | |
1745 | ||
1746 | err = -EINVAL; | |
1747 | if (sk->sk_state != TCP_ESTABLISHED) | |
1748 | goto out; | |
1749 | ||
1750 | err = -EOPNOTSUPP; | |
1751 | if (flags&MSG_OOB) | |
1752 | goto out; | |
1753 | ||
1754 | target = sock_rcvlowat(sk, flags&MSG_WAITALL, size); | |
1755 | timeo = sock_rcvtimeo(sk, flags&MSG_DONTWAIT); | |
1756 | ||
1757 | msg->msg_namelen = 0; | |
1758 | ||
1759 | /* Lock the socket to prevent queue disordering | |
1760 | * while sleeps in memcpy_tomsg | |
1761 | */ | |
1762 | ||
1763 | if (!siocb->scm) { | |
1764 | siocb->scm = &tmp_scm; | |
1765 | memset(&tmp_scm, 0, sizeof(tmp_scm)); | |
1766 | } | |
1767 | ||
57b47a53 | 1768 | mutex_lock(&u->readlock); |
1da177e4 LT |
1769 | |
1770 | do | |
1771 | { | |
1772 | int chunk; | |
1773 | struct sk_buff *skb; | |
1774 | ||
3c0d2f37 | 1775 | unix_state_lock(sk); |
1da177e4 LT |
1776 | skb = skb_dequeue(&sk->sk_receive_queue); |
1777 | if (skb==NULL) | |
1778 | { | |
1779 | if (copied >= target) | |
3c0d2f37 | 1780 | goto unlock; |
1da177e4 LT |
1781 | |
1782 | /* | |
1783 | * POSIX 1003.1g mandates this order. | |
1784 | */ | |
ac7bfa62 | 1785 | |
1da177e4 | 1786 | if ((err = sock_error(sk)) != 0) |
3c0d2f37 | 1787 | goto unlock; |
1da177e4 | 1788 | if (sk->sk_shutdown & RCV_SHUTDOWN) |
3c0d2f37 MS |
1789 | goto unlock; |
1790 | ||
1791 | unix_state_unlock(sk); | |
1da177e4 LT |
1792 | err = -EAGAIN; |
1793 | if (!timeo) | |
1794 | break; | |
57b47a53 | 1795 | mutex_unlock(&u->readlock); |
1da177e4 LT |
1796 | |
1797 | timeo = unix_stream_data_wait(sk, timeo); | |
1798 | ||
1799 | if (signal_pending(current)) { | |
1800 | err = sock_intr_errno(timeo); | |
1801 | goto out; | |
1802 | } | |
57b47a53 | 1803 | mutex_lock(&u->readlock); |
1da177e4 | 1804 | continue; |
3c0d2f37 MS |
1805 | unlock: |
1806 | unix_state_unlock(sk); | |
1807 | break; | |
1da177e4 | 1808 | } |
3c0d2f37 | 1809 | unix_state_unlock(sk); |
1da177e4 LT |
1810 | |
1811 | if (check_creds) { | |
1812 | /* Never glue messages from different writers */ | |
1813 | if (memcmp(UNIXCREDS(skb), &siocb->scm->creds, sizeof(siocb->scm->creds)) != 0) { | |
1814 | skb_queue_head(&sk->sk_receive_queue, skb); | |
1815 | break; | |
1816 | } | |
1817 | } else { | |
1818 | /* Copy credentials */ | |
1819 | siocb->scm->creds = *UNIXCREDS(skb); | |
1820 | check_creds = 1; | |
1821 | } | |
1822 | ||
1823 | /* Copy address just once */ | |
1824 | if (sunaddr) | |
1825 | { | |
1826 | unix_copy_addr(msg, skb->sk); | |
1827 | sunaddr = NULL; | |
1828 | } | |
1829 | ||
1830 | chunk = min_t(unsigned int, skb->len, size); | |
1831 | if (memcpy_toiovec(msg->msg_iov, skb->data, chunk)) { | |
1832 | skb_queue_head(&sk->sk_receive_queue, skb); | |
1833 | if (copied == 0) | |
1834 | copied = -EFAULT; | |
1835 | break; | |
1836 | } | |
1837 | copied += chunk; | |
1838 | size -= chunk; | |
1839 | ||
1840 | /* Mark read part of skb as used */ | |
1841 | if (!(flags & MSG_PEEK)) | |
1842 | { | |
1843 | skb_pull(skb, chunk); | |
1844 | ||
1845 | if (UNIXCB(skb).fp) | |
1846 | unix_detach_fds(siocb->scm, skb); | |
1847 | ||
1848 | /* put the skb back if we didn't use it up.. */ | |
1849 | if (skb->len) | |
1850 | { | |
1851 | skb_queue_head(&sk->sk_receive_queue, skb); | |
1852 | break; | |
1853 | } | |
1854 | ||
1855 | kfree_skb(skb); | |
1856 | ||
1857 | if (siocb->scm->fp) | |
1858 | break; | |
1859 | } | |
1860 | else | |
1861 | { | |
1862 | /* It is questionable, see note in unix_dgram_recvmsg. | |
1863 | */ | |
1864 | if (UNIXCB(skb).fp) | |
1865 | siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp); | |
1866 | ||
1867 | /* put message back and return */ | |
1868 | skb_queue_head(&sk->sk_receive_queue, skb); | |
1869 | break; | |
1870 | } | |
1871 | } while (size); | |
1872 | ||
57b47a53 | 1873 | mutex_unlock(&u->readlock); |
1da177e4 LT |
1874 | scm_recv(sock, msg, siocb->scm, flags); |
1875 | out: | |
1876 | return copied ? : err; | |
1877 | } | |
1878 | ||
1879 | static int unix_shutdown(struct socket *sock, int mode) | |
1880 | { | |
1881 | struct sock *sk = sock->sk; | |
1882 | struct sock *other; | |
1883 | ||
1884 | mode = (mode+1)&(RCV_SHUTDOWN|SEND_SHUTDOWN); | |
1885 | ||
1886 | if (mode) { | |
1c92b4e5 | 1887 | unix_state_lock(sk); |
1da177e4 LT |
1888 | sk->sk_shutdown |= mode; |
1889 | other=unix_peer(sk); | |
1890 | if (other) | |
1891 | sock_hold(other); | |
1c92b4e5 | 1892 | unix_state_unlock(sk); |
1da177e4 LT |
1893 | sk->sk_state_change(sk); |
1894 | ||
1895 | if (other && | |
1896 | (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET)) { | |
1897 | ||
1898 | int peer_mode = 0; | |
1899 | ||
1900 | if (mode&RCV_SHUTDOWN) | |
1901 | peer_mode |= SEND_SHUTDOWN; | |
1902 | if (mode&SEND_SHUTDOWN) | |
1903 | peer_mode |= RCV_SHUTDOWN; | |
1c92b4e5 | 1904 | unix_state_lock(other); |
1da177e4 | 1905 | other->sk_shutdown |= peer_mode; |
1c92b4e5 | 1906 | unix_state_unlock(other); |
1da177e4 LT |
1907 | other->sk_state_change(other); |
1908 | read_lock(&other->sk_callback_lock); | |
1909 | if (peer_mode == SHUTDOWN_MASK) | |
1910 | sk_wake_async(other,1,POLL_HUP); | |
1911 | else if (peer_mode & RCV_SHUTDOWN) | |
1912 | sk_wake_async(other,1,POLL_IN); | |
1913 | read_unlock(&other->sk_callback_lock); | |
1914 | } | |
1915 | if (other) | |
1916 | sock_put(other); | |
1917 | } | |
1918 | return 0; | |
1919 | } | |
1920 | ||
1921 | static int unix_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) | |
1922 | { | |
1923 | struct sock *sk = sock->sk; | |
1924 | long amount=0; | |
1925 | int err; | |
1926 | ||
1927 | switch(cmd) | |
1928 | { | |
1929 | case SIOCOUTQ: | |
1930 | amount = atomic_read(&sk->sk_wmem_alloc); | |
1931 | err = put_user(amount, (int __user *)arg); | |
1932 | break; | |
1933 | case SIOCINQ: | |
1934 | { | |
1935 | struct sk_buff *skb; | |
1936 | ||
1937 | if (sk->sk_state == TCP_LISTEN) { | |
1938 | err = -EINVAL; | |
1939 | break; | |
1940 | } | |
1941 | ||
1942 | spin_lock(&sk->sk_receive_queue.lock); | |
1943 | if (sk->sk_type == SOCK_STREAM || | |
1944 | sk->sk_type == SOCK_SEQPACKET) { | |
1945 | skb_queue_walk(&sk->sk_receive_queue, skb) | |
1946 | amount += skb->len; | |
1947 | } else { | |
1948 | skb = skb_peek(&sk->sk_receive_queue); | |
1949 | if (skb) | |
1950 | amount=skb->len; | |
1951 | } | |
1952 | spin_unlock(&sk->sk_receive_queue.lock); | |
1953 | err = put_user(amount, (int __user *)arg); | |
1954 | break; | |
1955 | } | |
1956 | ||
1957 | default: | |
b5e5fa5e | 1958 | err = -ENOIOCTLCMD; |
1da177e4 LT |
1959 | break; |
1960 | } | |
1961 | return err; | |
1962 | } | |
1963 | ||
1964 | static unsigned int unix_poll(struct file * file, struct socket *sock, poll_table *wait) | |
1965 | { | |
1966 | struct sock *sk = sock->sk; | |
1967 | unsigned int mask; | |
1968 | ||
1969 | poll_wait(file, sk->sk_sleep, wait); | |
1970 | mask = 0; | |
1971 | ||
1972 | /* exceptional events? */ | |
1973 | if (sk->sk_err) | |
1974 | mask |= POLLERR; | |
1975 | if (sk->sk_shutdown == SHUTDOWN_MASK) | |
1976 | mask |= POLLHUP; | |
f348d70a DL |
1977 | if (sk->sk_shutdown & RCV_SHUTDOWN) |
1978 | mask |= POLLRDHUP; | |
1da177e4 LT |
1979 | |
1980 | /* readable? */ | |
1981 | if (!skb_queue_empty(&sk->sk_receive_queue) || | |
1982 | (sk->sk_shutdown & RCV_SHUTDOWN)) | |
1983 | mask |= POLLIN | POLLRDNORM; | |
1984 | ||
1985 | /* Connection-based need to check for termination and startup */ | |
1986 | if ((sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) && sk->sk_state == TCP_CLOSE) | |
1987 | mask |= POLLHUP; | |
1988 | ||
1989 | /* | |
1990 | * we set writable also when the other side has shut down the | |
1991 | * connection. This prevents stuck sockets. | |
1992 | */ | |
1993 | if (unix_writable(sk)) | |
1994 | mask |= POLLOUT | POLLWRNORM | POLLWRBAND; | |
1995 | ||
1996 | return mask; | |
1997 | } | |
1998 | ||
1999 | ||
2000 | #ifdef CONFIG_PROC_FS | |
2001 | static struct sock *unix_seq_idx(int *iter, loff_t pos) | |
2002 | { | |
2003 | loff_t off = 0; | |
2004 | struct sock *s; | |
2005 | ||
2006 | for (s = first_unix_socket(iter); s; s = next_unix_socket(iter, s)) { | |
ac7bfa62 | 2007 | if (off == pos) |
1da177e4 LT |
2008 | return s; |
2009 | ++off; | |
2010 | } | |
2011 | return NULL; | |
2012 | } | |
2013 | ||
2014 | ||
2015 | static void *unix_seq_start(struct seq_file *seq, loff_t *pos) | |
2016 | { | |
fbe9cc4a | 2017 | spin_lock(&unix_table_lock); |
1da177e4 LT |
2018 | return *pos ? unix_seq_idx(seq->private, *pos - 1) : ((void *) 1); |
2019 | } | |
2020 | ||
2021 | static void *unix_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
2022 | { | |
2023 | ++*pos; | |
2024 | ||
ac7bfa62 | 2025 | if (v == (void *)1) |
1da177e4 LT |
2026 | return first_unix_socket(seq->private); |
2027 | return next_unix_socket(seq->private, v); | |
2028 | } | |
2029 | ||
2030 | static void unix_seq_stop(struct seq_file *seq, void *v) | |
2031 | { | |
fbe9cc4a | 2032 | spin_unlock(&unix_table_lock); |
1da177e4 LT |
2033 | } |
2034 | ||
2035 | static int unix_seq_show(struct seq_file *seq, void *v) | |
2036 | { | |
ac7bfa62 | 2037 | |
1da177e4 LT |
2038 | if (v == (void *)1) |
2039 | seq_puts(seq, "Num RefCount Protocol Flags Type St " | |
2040 | "Inode Path\n"); | |
2041 | else { | |
2042 | struct sock *s = v; | |
2043 | struct unix_sock *u = unix_sk(s); | |
1c92b4e5 | 2044 | unix_state_lock(s); |
1da177e4 LT |
2045 | |
2046 | seq_printf(seq, "%p: %08X %08X %08X %04X %02X %5lu", | |
2047 | s, | |
2048 | atomic_read(&s->sk_refcnt), | |
2049 | 0, | |
2050 | s->sk_state == TCP_LISTEN ? __SO_ACCEPTCON : 0, | |
2051 | s->sk_type, | |
2052 | s->sk_socket ? | |
2053 | (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTED : SS_UNCONNECTED) : | |
2054 | (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTING : SS_DISCONNECTING), | |
2055 | sock_i_ino(s)); | |
2056 | ||
2057 | if (u->addr) { | |
2058 | int i, len; | |
2059 | seq_putc(seq, ' '); | |
2060 | ||
2061 | i = 0; | |
2062 | len = u->addr->len - sizeof(short); | |
2063 | if (!UNIX_ABSTRACT(s)) | |
2064 | len--; | |
2065 | else { | |
2066 | seq_putc(seq, '@'); | |
2067 | i++; | |
2068 | } | |
2069 | for ( ; i < len; i++) | |
2070 | seq_putc(seq, u->addr->name->sun_path[i]); | |
2071 | } | |
1c92b4e5 | 2072 | unix_state_unlock(s); |
1da177e4 LT |
2073 | seq_putc(seq, '\n'); |
2074 | } | |
2075 | ||
2076 | return 0; | |
2077 | } | |
2078 | ||
56b3d975 | 2079 | static const struct seq_operations unix_seq_ops = { |
1da177e4 LT |
2080 | .start = unix_seq_start, |
2081 | .next = unix_seq_next, | |
2082 | .stop = unix_seq_stop, | |
2083 | .show = unix_seq_show, | |
2084 | }; | |
2085 | ||
2086 | ||
2087 | static int unix_seq_open(struct inode *inode, struct file *file) | |
2088 | { | |
cf7732e4 | 2089 | return seq_open_private(file, &unix_seq_ops, sizeof(int)); |
1da177e4 LT |
2090 | } |
2091 | ||
da7071d7 | 2092 | static const struct file_operations unix_seq_fops = { |
1da177e4 LT |
2093 | .owner = THIS_MODULE, |
2094 | .open = unix_seq_open, | |
2095 | .read = seq_read, | |
2096 | .llseek = seq_lseek, | |
2097 | .release = seq_release_private, | |
2098 | }; | |
2099 | ||
2100 | #endif | |
2101 | ||
2102 | static struct net_proto_family unix_family_ops = { | |
2103 | .family = PF_UNIX, | |
2104 | .create = unix_create, | |
2105 | .owner = THIS_MODULE, | |
2106 | }; | |
2107 | ||
1da177e4 LT |
2108 | static int __init af_unix_init(void) |
2109 | { | |
2110 | int rc = -1; | |
2111 | struct sk_buff *dummy_skb; | |
2112 | ||
ef047f5e | 2113 | BUILD_BUG_ON(sizeof(struct unix_skb_parms) > sizeof(dummy_skb->cb)); |
1da177e4 LT |
2114 | |
2115 | rc = proto_register(&unix_proto, 1); | |
ac7bfa62 YH |
2116 | if (rc != 0) { |
2117 | printk(KERN_CRIT "%s: Cannot create unix_sock SLAB cache!\n", | |
1da177e4 LT |
2118 | __FUNCTION__); |
2119 | goto out; | |
2120 | } | |
2121 | ||
2122 | sock_register(&unix_family_ops); | |
2123 | #ifdef CONFIG_PROC_FS | |
457c4cbc | 2124 | proc_net_fops_create(&init_net, "unix", 0, &unix_seq_fops); |
1da177e4 LT |
2125 | #endif |
2126 | unix_sysctl_register(); | |
2127 | out: | |
2128 | return rc; | |
2129 | } | |
2130 | ||
2131 | static void __exit af_unix_exit(void) | |
2132 | { | |
2133 | sock_unregister(PF_UNIX); | |
2134 | unix_sysctl_unregister(); | |
457c4cbc | 2135 | proc_net_remove(&init_net, "unix"); |
1da177e4 LT |
2136 | proto_unregister(&unix_proto); |
2137 | } | |
2138 | ||
2139 | module_init(af_unix_init); | |
2140 | module_exit(af_unix_exit); | |
2141 | ||
2142 | MODULE_LICENSE("GPL"); | |
2143 | MODULE_ALIAS_NETPROTO(PF_UNIX); |