]>
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> | |
106 | #include <net/sock.h> | |
c752f073 | 107 | #include <net/tcp_states.h> |
1da177e4 LT |
108 | #include <net/af_unix.h> |
109 | #include <linux/proc_fs.h> | |
110 | #include <linux/seq_file.h> | |
111 | #include <net/scm.h> | |
112 | #include <linux/init.h> | |
113 | #include <linux/poll.h> | |
1da177e4 LT |
114 | #include <linux/rtnetlink.h> |
115 | #include <linux/mount.h> | |
116 | #include <net/checksum.h> | |
117 | #include <linux/security.h> | |
118 | ||
18adaf06 | 119 | int sysctl_unix_max_dgram_qlen __read_mostly = 10; |
1da177e4 | 120 | |
13111698 AB |
121 | static struct hlist_head unix_socket_table[UNIX_HASH_SIZE + 1]; |
122 | static DEFINE_SPINLOCK(unix_table_lock); | |
1da177e4 LT |
123 | static atomic_t unix_nr_socks = ATOMIC_INIT(0); |
124 | ||
125 | #define unix_sockets_unbound (&unix_socket_table[UNIX_HASH_SIZE]) | |
126 | ||
127 | #define UNIX_ABSTRACT(sk) (unix_sk(sk)->addr->hash != UNIX_HASH_SIZE) | |
128 | ||
13111698 AB |
129 | static struct sock *first_unix_socket(int *i) |
130 | { | |
131 | for (*i = 0; *i <= UNIX_HASH_SIZE; (*i)++) { | |
132 | if (!hlist_empty(&unix_socket_table[*i])) | |
133 | return __sk_head(&unix_socket_table[*i]); | |
134 | } | |
135 | return NULL; | |
136 | } | |
137 | ||
138 | static struct sock *next_unix_socket(int *i, struct sock *s) | |
139 | { | |
140 | struct sock *next = sk_next(s); | |
141 | /* More in this chain? */ | |
142 | if (next) | |
143 | return next; | |
144 | /* Look for next non-empty chain. */ | |
145 | for ((*i)++; *i <= UNIX_HASH_SIZE; (*i)++) { | |
146 | if (!hlist_empty(&unix_socket_table[*i])) | |
147 | return __sk_head(&unix_socket_table[*i]); | |
148 | } | |
149 | return NULL; | |
150 | } | |
151 | ||
152 | #define forall_unix_sockets(i, s) \ | |
153 | for (s = first_unix_socket(&(i)); s; s = next_unix_socket(&(i),(s))) | |
154 | ||
877ce7c1 | 155 | #ifdef CONFIG_SECURITY_NETWORK |
dc49c1f9 | 156 | static void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb) |
877ce7c1 | 157 | { |
dc49c1f9 | 158 | memcpy(UNIXSID(skb), &scm->secid, sizeof(u32)); |
877ce7c1 CZ |
159 | } |
160 | ||
161 | static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb) | |
162 | { | |
dc49c1f9 | 163 | scm->secid = *UNIXSID(skb); |
877ce7c1 CZ |
164 | } |
165 | #else | |
dc49c1f9 | 166 | static inline void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb) |
877ce7c1 CZ |
167 | { } |
168 | ||
169 | static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb) | |
170 | { } | |
171 | #endif /* CONFIG_SECURITY_NETWORK */ | |
172 | ||
1da177e4 LT |
173 | /* |
174 | * SMP locking strategy: | |
fbe9cc4a | 175 | * hash table is protected with spinlock unix_table_lock |
1da177e4 LT |
176 | * each socket state is protected by separate rwlock. |
177 | */ | |
178 | ||
44bb9363 | 179 | static inline unsigned unix_hash_fold(__wsum n) |
1da177e4 | 180 | { |
44bb9363 | 181 | unsigned hash = (__force unsigned)n; |
1da177e4 LT |
182 | hash ^= hash>>16; |
183 | hash ^= hash>>8; | |
184 | return hash&(UNIX_HASH_SIZE-1); | |
185 | } | |
186 | ||
187 | #define unix_peer(sk) (unix_sk(sk)->peer) | |
188 | ||
189 | static inline int unix_our_peer(struct sock *sk, struct sock *osk) | |
190 | { | |
191 | return unix_peer(osk) == sk; | |
192 | } | |
193 | ||
194 | static inline int unix_may_send(struct sock *sk, struct sock *osk) | |
195 | { | |
196 | return (unix_peer(osk) == NULL || unix_our_peer(sk, osk)); | |
197 | } | |
198 | ||
199 | static struct sock *unix_peer_get(struct sock *s) | |
200 | { | |
201 | struct sock *peer; | |
202 | ||
1c92b4e5 | 203 | unix_state_lock(s); |
1da177e4 LT |
204 | peer = unix_peer(s); |
205 | if (peer) | |
206 | sock_hold(peer); | |
1c92b4e5 | 207 | unix_state_unlock(s); |
1da177e4 LT |
208 | return peer; |
209 | } | |
210 | ||
211 | static inline void unix_release_addr(struct unix_address *addr) | |
212 | { | |
213 | if (atomic_dec_and_test(&addr->refcnt)) | |
214 | kfree(addr); | |
215 | } | |
216 | ||
217 | /* | |
218 | * Check unix socket name: | |
219 | * - should be not zero length. | |
220 | * - if started by not zero, should be NULL terminated (FS object) | |
221 | * - if started by zero, it is abstract name. | |
222 | */ | |
ac7bfa62 | 223 | |
1da177e4 LT |
224 | static int unix_mkname(struct sockaddr_un * sunaddr, int len, unsigned *hashp) |
225 | { | |
226 | if (len <= sizeof(short) || len > sizeof(*sunaddr)) | |
227 | return -EINVAL; | |
228 | if (!sunaddr || sunaddr->sun_family != AF_UNIX) | |
229 | return -EINVAL; | |
230 | if (sunaddr->sun_path[0]) { | |
231 | /* | |
232 | * This may look like an off by one error but it is a bit more | |
233 | * subtle. 108 is the longest valid AF_UNIX path for a binding. | |
234 | * sun_path[108] doesnt as such exist. However in kernel space | |
235 | * we are guaranteed that it is a valid memory location in our | |
236 | * kernel address buffer. | |
237 | */ | |
238 | ((char *)sunaddr)[len]=0; | |
239 | len = strlen(sunaddr->sun_path)+1+sizeof(short); | |
240 | return len; | |
241 | } | |
242 | ||
243 | *hashp = unix_hash_fold(csum_partial((char*)sunaddr, len, 0)); | |
244 | return len; | |
245 | } | |
246 | ||
247 | static void __unix_remove_socket(struct sock *sk) | |
248 | { | |
249 | sk_del_node_init(sk); | |
250 | } | |
251 | ||
252 | static void __unix_insert_socket(struct hlist_head *list, struct sock *sk) | |
253 | { | |
254 | BUG_TRAP(sk_unhashed(sk)); | |
255 | sk_add_node(sk, list); | |
256 | } | |
257 | ||
258 | static inline void unix_remove_socket(struct sock *sk) | |
259 | { | |
fbe9cc4a | 260 | spin_lock(&unix_table_lock); |
1da177e4 | 261 | __unix_remove_socket(sk); |
fbe9cc4a | 262 | spin_unlock(&unix_table_lock); |
1da177e4 LT |
263 | } |
264 | ||
265 | static inline void unix_insert_socket(struct hlist_head *list, struct sock *sk) | |
266 | { | |
fbe9cc4a | 267 | spin_lock(&unix_table_lock); |
1da177e4 | 268 | __unix_insert_socket(list, sk); |
fbe9cc4a | 269 | spin_unlock(&unix_table_lock); |
1da177e4 LT |
270 | } |
271 | ||
272 | static struct sock *__unix_find_socket_byname(struct sockaddr_un *sunname, | |
273 | int len, int type, unsigned hash) | |
274 | { | |
275 | struct sock *s; | |
276 | struct hlist_node *node; | |
277 | ||
278 | sk_for_each(s, node, &unix_socket_table[hash ^ type]) { | |
279 | struct unix_sock *u = unix_sk(s); | |
280 | ||
281 | if (u->addr->len == len && | |
282 | !memcmp(u->addr->name, sunname, len)) | |
283 | goto found; | |
284 | } | |
285 | s = NULL; | |
286 | found: | |
287 | return s; | |
288 | } | |
289 | ||
290 | static inline struct sock *unix_find_socket_byname(struct sockaddr_un *sunname, | |
291 | int len, int type, | |
292 | unsigned hash) | |
293 | { | |
294 | struct sock *s; | |
295 | ||
fbe9cc4a | 296 | spin_lock(&unix_table_lock); |
1da177e4 LT |
297 | s = __unix_find_socket_byname(sunname, len, type, hash); |
298 | if (s) | |
299 | sock_hold(s); | |
fbe9cc4a | 300 | spin_unlock(&unix_table_lock); |
1da177e4 LT |
301 | return s; |
302 | } | |
303 | ||
304 | static struct sock *unix_find_socket_byinode(struct inode *i) | |
305 | { | |
306 | struct sock *s; | |
307 | struct hlist_node *node; | |
308 | ||
fbe9cc4a | 309 | spin_lock(&unix_table_lock); |
1da177e4 LT |
310 | sk_for_each(s, node, |
311 | &unix_socket_table[i->i_ino & (UNIX_HASH_SIZE - 1)]) { | |
312 | struct dentry *dentry = unix_sk(s)->dentry; | |
313 | ||
314 | if(dentry && dentry->d_inode == i) | |
315 | { | |
316 | sock_hold(s); | |
317 | goto found; | |
318 | } | |
319 | } | |
320 | s = NULL; | |
321 | found: | |
fbe9cc4a | 322 | spin_unlock(&unix_table_lock); |
1da177e4 LT |
323 | return s; |
324 | } | |
325 | ||
326 | static inline int unix_writable(struct sock *sk) | |
327 | { | |
328 | return (atomic_read(&sk->sk_wmem_alloc) << 2) <= sk->sk_sndbuf; | |
329 | } | |
330 | ||
331 | static void unix_write_space(struct sock *sk) | |
332 | { | |
333 | read_lock(&sk->sk_callback_lock); | |
334 | if (unix_writable(sk)) { | |
335 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) | |
336 | wake_up_interruptible(sk->sk_sleep); | |
337 | sk_wake_async(sk, 2, POLL_OUT); | |
338 | } | |
339 | read_unlock(&sk->sk_callback_lock); | |
340 | } | |
341 | ||
342 | /* When dgram socket disconnects (or changes its peer), we clear its receive | |
343 | * queue of packets arrived from previous peer. First, it allows to do | |
344 | * flow control based only on wmem_alloc; second, sk connected to peer | |
345 | * may receive messages only from that peer. */ | |
346 | static void unix_dgram_disconnected(struct sock *sk, struct sock *other) | |
347 | { | |
b03efcfb | 348 | if (!skb_queue_empty(&sk->sk_receive_queue)) { |
1da177e4 LT |
349 | skb_queue_purge(&sk->sk_receive_queue); |
350 | wake_up_interruptible_all(&unix_sk(sk)->peer_wait); | |
351 | ||
352 | /* If one link of bidirectional dgram pipe is disconnected, | |
353 | * we signal error. Messages are lost. Do not make this, | |
354 | * when peer was not connected to us. | |
355 | */ | |
356 | if (!sock_flag(other, SOCK_DEAD) && unix_peer(other) == sk) { | |
357 | other->sk_err = ECONNRESET; | |
358 | other->sk_error_report(other); | |
359 | } | |
360 | } | |
361 | } | |
362 | ||
363 | static void unix_sock_destructor(struct sock *sk) | |
364 | { | |
365 | struct unix_sock *u = unix_sk(sk); | |
366 | ||
367 | skb_queue_purge(&sk->sk_receive_queue); | |
368 | ||
369 | BUG_TRAP(!atomic_read(&sk->sk_wmem_alloc)); | |
370 | BUG_TRAP(sk_unhashed(sk)); | |
371 | BUG_TRAP(!sk->sk_socket); | |
372 | if (!sock_flag(sk, SOCK_DEAD)) { | |
373 | printk("Attempt to release alive unix socket: %p\n", sk); | |
374 | return; | |
375 | } | |
376 | ||
377 | if (u->addr) | |
378 | unix_release_addr(u->addr); | |
379 | ||
380 | atomic_dec(&unix_nr_socks); | |
381 | #ifdef UNIX_REFCNT_DEBUG | |
382 | printk(KERN_DEBUG "UNIX %p is destroyed, %d are still alive.\n", sk, atomic_read(&unix_nr_socks)); | |
383 | #endif | |
384 | } | |
385 | ||
386 | static int unix_release_sock (struct sock *sk, int embrion) | |
387 | { | |
388 | struct unix_sock *u = unix_sk(sk); | |
389 | struct dentry *dentry; | |
390 | struct vfsmount *mnt; | |
391 | struct sock *skpair; | |
392 | struct sk_buff *skb; | |
393 | int state; | |
394 | ||
395 | unix_remove_socket(sk); | |
396 | ||
397 | /* Clear state */ | |
1c92b4e5 | 398 | unix_state_lock(sk); |
1da177e4 LT |
399 | sock_orphan(sk); |
400 | sk->sk_shutdown = SHUTDOWN_MASK; | |
401 | dentry = u->dentry; | |
402 | u->dentry = NULL; | |
403 | mnt = u->mnt; | |
404 | u->mnt = NULL; | |
405 | state = sk->sk_state; | |
406 | sk->sk_state = TCP_CLOSE; | |
1c92b4e5 | 407 | unix_state_unlock(sk); |
1da177e4 LT |
408 | |
409 | wake_up_interruptible_all(&u->peer_wait); | |
410 | ||
411 | skpair=unix_peer(sk); | |
412 | ||
413 | if (skpair!=NULL) { | |
414 | if (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) { | |
1c92b4e5 | 415 | unix_state_lock(skpair); |
1da177e4 LT |
416 | /* No more writes */ |
417 | skpair->sk_shutdown = SHUTDOWN_MASK; | |
418 | if (!skb_queue_empty(&sk->sk_receive_queue) || embrion) | |
419 | skpair->sk_err = ECONNRESET; | |
1c92b4e5 | 420 | unix_state_unlock(skpair); |
1da177e4 LT |
421 | skpair->sk_state_change(skpair); |
422 | read_lock(&skpair->sk_callback_lock); | |
423 | sk_wake_async(skpair,1,POLL_HUP); | |
424 | read_unlock(&skpair->sk_callback_lock); | |
425 | } | |
426 | sock_put(skpair); /* It may now die */ | |
427 | unix_peer(sk) = NULL; | |
428 | } | |
429 | ||
430 | /* Try to flush out this socket. Throw out buffers at least */ | |
431 | ||
432 | while ((skb = skb_dequeue(&sk->sk_receive_queue)) != NULL) { | |
433 | if (state==TCP_LISTEN) | |
434 | unix_release_sock(skb->sk, 1); | |
435 | /* passed fds are erased in the kfree_skb hook */ | |
436 | kfree_skb(skb); | |
437 | } | |
438 | ||
439 | if (dentry) { | |
440 | dput(dentry); | |
441 | mntput(mnt); | |
442 | } | |
443 | ||
444 | sock_put(sk); | |
445 | ||
446 | /* ---- Socket is dead now and most probably destroyed ---- */ | |
447 | ||
448 | /* | |
449 | * Fixme: BSD difference: In BSD all sockets connected to use get | |
450 | * ECONNRESET and we die on the spot. In Linux we behave | |
451 | * like files and pipes do and wait for the last | |
452 | * dereference. | |
453 | * | |
454 | * Can't we simply set sock->err? | |
455 | * | |
456 | * What the above comment does talk about? --ANK(980817) | |
457 | */ | |
458 | ||
459 | if (atomic_read(&unix_tot_inflight)) | |
ac7bfa62 | 460 | unix_gc(); /* Garbage collect fds */ |
1da177e4 LT |
461 | |
462 | return 0; | |
463 | } | |
464 | ||
465 | static int unix_listen(struct socket *sock, int backlog) | |
466 | { | |
467 | int err; | |
468 | struct sock *sk = sock->sk; | |
469 | struct unix_sock *u = unix_sk(sk); | |
470 | ||
471 | err = -EOPNOTSUPP; | |
472 | if (sock->type!=SOCK_STREAM && sock->type!=SOCK_SEQPACKET) | |
473 | goto out; /* Only stream/seqpacket sockets accept */ | |
474 | err = -EINVAL; | |
475 | if (!u->addr) | |
476 | goto out; /* No listens on an unbound socket */ | |
1c92b4e5 | 477 | unix_state_lock(sk); |
1da177e4 LT |
478 | if (sk->sk_state != TCP_CLOSE && sk->sk_state != TCP_LISTEN) |
479 | goto out_unlock; | |
480 | if (backlog > sk->sk_max_ack_backlog) | |
481 | wake_up_interruptible_all(&u->peer_wait); | |
482 | sk->sk_max_ack_backlog = backlog; | |
483 | sk->sk_state = TCP_LISTEN; | |
484 | /* set credentials so connect can copy them */ | |
485 | sk->sk_peercred.pid = current->tgid; | |
486 | sk->sk_peercred.uid = current->euid; | |
487 | sk->sk_peercred.gid = current->egid; | |
488 | err = 0; | |
489 | ||
490 | out_unlock: | |
1c92b4e5 | 491 | unix_state_unlock(sk); |
1da177e4 LT |
492 | out: |
493 | return err; | |
494 | } | |
495 | ||
496 | static int unix_release(struct socket *); | |
497 | static int unix_bind(struct socket *, struct sockaddr *, int); | |
498 | static int unix_stream_connect(struct socket *, struct sockaddr *, | |
499 | int addr_len, int flags); | |
500 | static int unix_socketpair(struct socket *, struct socket *); | |
501 | static int unix_accept(struct socket *, struct socket *, int); | |
502 | static int unix_getname(struct socket *, struct sockaddr *, int *, int); | |
503 | static unsigned int unix_poll(struct file *, struct socket *, poll_table *); | |
504 | static int unix_ioctl(struct socket *, unsigned int, unsigned long); | |
505 | static int unix_shutdown(struct socket *, int); | |
506 | static int unix_stream_sendmsg(struct kiocb *, struct socket *, | |
507 | struct msghdr *, size_t); | |
508 | static int unix_stream_recvmsg(struct kiocb *, struct socket *, | |
509 | struct msghdr *, size_t, int); | |
510 | static int unix_dgram_sendmsg(struct kiocb *, struct socket *, | |
511 | struct msghdr *, size_t); | |
512 | static int unix_dgram_recvmsg(struct kiocb *, struct socket *, | |
513 | struct msghdr *, size_t, int); | |
514 | static int unix_dgram_connect(struct socket *, struct sockaddr *, | |
515 | int, int); | |
516 | static int unix_seqpacket_sendmsg(struct kiocb *, struct socket *, | |
517 | struct msghdr *, size_t); | |
518 | ||
90ddc4f0 | 519 | static const struct proto_ops unix_stream_ops = { |
1da177e4 LT |
520 | .family = PF_UNIX, |
521 | .owner = THIS_MODULE, | |
522 | .release = unix_release, | |
523 | .bind = unix_bind, | |
524 | .connect = unix_stream_connect, | |
525 | .socketpair = unix_socketpair, | |
526 | .accept = unix_accept, | |
527 | .getname = unix_getname, | |
528 | .poll = unix_poll, | |
529 | .ioctl = unix_ioctl, | |
530 | .listen = unix_listen, | |
531 | .shutdown = unix_shutdown, | |
532 | .setsockopt = sock_no_setsockopt, | |
533 | .getsockopt = sock_no_getsockopt, | |
534 | .sendmsg = unix_stream_sendmsg, | |
535 | .recvmsg = unix_stream_recvmsg, | |
536 | .mmap = sock_no_mmap, | |
537 | .sendpage = sock_no_sendpage, | |
538 | }; | |
539 | ||
90ddc4f0 | 540 | static const struct proto_ops unix_dgram_ops = { |
1da177e4 LT |
541 | .family = PF_UNIX, |
542 | .owner = THIS_MODULE, | |
543 | .release = unix_release, | |
544 | .bind = unix_bind, | |
545 | .connect = unix_dgram_connect, | |
546 | .socketpair = unix_socketpair, | |
547 | .accept = sock_no_accept, | |
548 | .getname = unix_getname, | |
549 | .poll = datagram_poll, | |
550 | .ioctl = unix_ioctl, | |
551 | .listen = sock_no_listen, | |
552 | .shutdown = unix_shutdown, | |
553 | .setsockopt = sock_no_setsockopt, | |
554 | .getsockopt = sock_no_getsockopt, | |
555 | .sendmsg = unix_dgram_sendmsg, | |
556 | .recvmsg = unix_dgram_recvmsg, | |
557 | .mmap = sock_no_mmap, | |
558 | .sendpage = sock_no_sendpage, | |
559 | }; | |
560 | ||
90ddc4f0 | 561 | static const struct proto_ops unix_seqpacket_ops = { |
1da177e4 LT |
562 | .family = PF_UNIX, |
563 | .owner = THIS_MODULE, | |
564 | .release = unix_release, | |
565 | .bind = unix_bind, | |
566 | .connect = unix_stream_connect, | |
567 | .socketpair = unix_socketpair, | |
568 | .accept = unix_accept, | |
569 | .getname = unix_getname, | |
570 | .poll = datagram_poll, | |
571 | .ioctl = unix_ioctl, | |
572 | .listen = unix_listen, | |
573 | .shutdown = unix_shutdown, | |
574 | .setsockopt = sock_no_setsockopt, | |
575 | .getsockopt = sock_no_getsockopt, | |
576 | .sendmsg = unix_seqpacket_sendmsg, | |
577 | .recvmsg = unix_dgram_recvmsg, | |
578 | .mmap = sock_no_mmap, | |
579 | .sendpage = sock_no_sendpage, | |
580 | }; | |
581 | ||
582 | static struct proto unix_proto = { | |
583 | .name = "UNIX", | |
584 | .owner = THIS_MODULE, | |
585 | .obj_size = sizeof(struct unix_sock), | |
586 | }; | |
587 | ||
a09785a2 IM |
588 | /* |
589 | * AF_UNIX sockets do not interact with hardware, hence they | |
590 | * dont trigger interrupts - so it's safe for them to have | |
591 | * bh-unsafe locking for their sk_receive_queue.lock. Split off | |
592 | * this special lock-class by reinitializing the spinlock key: | |
593 | */ | |
594 | static struct lock_class_key af_unix_sk_receive_queue_lock_key; | |
595 | ||
1da177e4 LT |
596 | static struct sock * unix_create1(struct socket *sock) |
597 | { | |
598 | struct sock *sk = NULL; | |
599 | struct unix_sock *u; | |
600 | ||
529bf6be | 601 | if (atomic_read(&unix_nr_socks) >= 2*get_max_files()) |
1da177e4 LT |
602 | goto out; |
603 | ||
604 | sk = sk_alloc(PF_UNIX, GFP_KERNEL, &unix_proto, 1); | |
605 | if (!sk) | |
606 | goto out; | |
607 | ||
608 | atomic_inc(&unix_nr_socks); | |
609 | ||
610 | sock_init_data(sock,sk); | |
a09785a2 IM |
611 | lockdep_set_class(&sk->sk_receive_queue.lock, |
612 | &af_unix_sk_receive_queue_lock_key); | |
1da177e4 LT |
613 | |
614 | sk->sk_write_space = unix_write_space; | |
615 | sk->sk_max_ack_backlog = sysctl_unix_max_dgram_qlen; | |
616 | sk->sk_destruct = unix_sock_destructor; | |
617 | u = unix_sk(sk); | |
618 | u->dentry = NULL; | |
619 | u->mnt = NULL; | |
fd19f329 | 620 | spin_lock_init(&u->lock); |
1fd05ba5 MS |
621 | atomic_set(&u->inflight, 0); |
622 | INIT_LIST_HEAD(&u->link); | |
57b47a53 | 623 | mutex_init(&u->readlock); /* single task reading lock */ |
1da177e4 LT |
624 | init_waitqueue_head(&u->peer_wait); |
625 | unix_insert_socket(unix_sockets_unbound, sk); | |
626 | out: | |
627 | return sk; | |
628 | } | |
629 | ||
630 | static int unix_create(struct socket *sock, int protocol) | |
631 | { | |
632 | if (protocol && protocol != PF_UNIX) | |
633 | return -EPROTONOSUPPORT; | |
634 | ||
635 | sock->state = SS_UNCONNECTED; | |
636 | ||
637 | switch (sock->type) { | |
638 | case SOCK_STREAM: | |
639 | sock->ops = &unix_stream_ops; | |
640 | break; | |
641 | /* | |
642 | * Believe it or not BSD has AF_UNIX, SOCK_RAW though | |
643 | * nothing uses it. | |
644 | */ | |
645 | case SOCK_RAW: | |
646 | sock->type=SOCK_DGRAM; | |
647 | case SOCK_DGRAM: | |
648 | sock->ops = &unix_dgram_ops; | |
649 | break; | |
650 | case SOCK_SEQPACKET: | |
651 | sock->ops = &unix_seqpacket_ops; | |
652 | break; | |
653 | default: | |
654 | return -ESOCKTNOSUPPORT; | |
655 | } | |
656 | ||
657 | return unix_create1(sock) ? 0 : -ENOMEM; | |
658 | } | |
659 | ||
660 | static int unix_release(struct socket *sock) | |
661 | { | |
662 | struct sock *sk = sock->sk; | |
663 | ||
664 | if (!sk) | |
665 | return 0; | |
666 | ||
667 | sock->sk = NULL; | |
668 | ||
669 | return unix_release_sock (sk, 0); | |
670 | } | |
671 | ||
672 | static int unix_autobind(struct socket *sock) | |
673 | { | |
674 | struct sock *sk = sock->sk; | |
675 | struct unix_sock *u = unix_sk(sk); | |
676 | static u32 ordernum = 1; | |
677 | struct unix_address * addr; | |
678 | int err; | |
679 | ||
57b47a53 | 680 | mutex_lock(&u->readlock); |
1da177e4 LT |
681 | |
682 | err = 0; | |
683 | if (u->addr) | |
684 | goto out; | |
685 | ||
686 | err = -ENOMEM; | |
0da974f4 | 687 | addr = kzalloc(sizeof(*addr) + sizeof(short) + 16, GFP_KERNEL); |
1da177e4 LT |
688 | if (!addr) |
689 | goto out; | |
690 | ||
1da177e4 LT |
691 | addr->name->sun_family = AF_UNIX; |
692 | atomic_set(&addr->refcnt, 1); | |
693 | ||
694 | retry: | |
695 | addr->len = sprintf(addr->name->sun_path+1, "%05x", ordernum) + 1 + sizeof(short); | |
696 | addr->hash = unix_hash_fold(csum_partial((void*)addr->name, addr->len, 0)); | |
697 | ||
fbe9cc4a | 698 | spin_lock(&unix_table_lock); |
1da177e4 LT |
699 | ordernum = (ordernum+1)&0xFFFFF; |
700 | ||
701 | if (__unix_find_socket_byname(addr->name, addr->len, sock->type, | |
702 | addr->hash)) { | |
fbe9cc4a | 703 | spin_unlock(&unix_table_lock); |
1da177e4 LT |
704 | /* Sanity yield. It is unusual case, but yet... */ |
705 | if (!(ordernum&0xFF)) | |
706 | yield(); | |
707 | goto retry; | |
708 | } | |
709 | addr->hash ^= sk->sk_type; | |
710 | ||
711 | __unix_remove_socket(sk); | |
712 | u->addr = addr; | |
713 | __unix_insert_socket(&unix_socket_table[addr->hash], sk); | |
fbe9cc4a | 714 | spin_unlock(&unix_table_lock); |
1da177e4 LT |
715 | err = 0; |
716 | ||
57b47a53 | 717 | out: mutex_unlock(&u->readlock); |
1da177e4 LT |
718 | return err; |
719 | } | |
720 | ||
721 | static struct sock *unix_find_other(struct sockaddr_un *sunname, int len, | |
722 | int type, unsigned hash, int *error) | |
723 | { | |
724 | struct sock *u; | |
725 | struct nameidata nd; | |
726 | int err = 0; | |
ac7bfa62 | 727 | |
1da177e4 LT |
728 | if (sunname->sun_path[0]) { |
729 | err = path_lookup(sunname->sun_path, LOOKUP_FOLLOW, &nd); | |
730 | if (err) | |
731 | goto fail; | |
e4543edd | 732 | err = vfs_permission(&nd, MAY_WRITE); |
1da177e4 LT |
733 | if (err) |
734 | goto put_fail; | |
735 | ||
736 | err = -ECONNREFUSED; | |
737 | if (!S_ISSOCK(nd.dentry->d_inode->i_mode)) | |
738 | goto put_fail; | |
739 | u=unix_find_socket_byinode(nd.dentry->d_inode); | |
740 | if (!u) | |
741 | goto put_fail; | |
742 | ||
743 | if (u->sk_type == type) | |
744 | touch_atime(nd.mnt, nd.dentry); | |
745 | ||
746 | path_release(&nd); | |
747 | ||
748 | err=-EPROTOTYPE; | |
749 | if (u->sk_type != type) { | |
750 | sock_put(u); | |
751 | goto fail; | |
752 | } | |
753 | } else { | |
754 | err = -ECONNREFUSED; | |
755 | u=unix_find_socket_byname(sunname, len, type, hash); | |
756 | if (u) { | |
757 | struct dentry *dentry; | |
758 | dentry = unix_sk(u)->dentry; | |
759 | if (dentry) | |
760 | touch_atime(unix_sk(u)->mnt, dentry); | |
761 | } else | |
762 | goto fail; | |
763 | } | |
764 | return u; | |
765 | ||
766 | put_fail: | |
767 | path_release(&nd); | |
768 | fail: | |
769 | *error=err; | |
770 | return NULL; | |
771 | } | |
772 | ||
773 | ||
774 | static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len) | |
775 | { | |
776 | struct sock *sk = sock->sk; | |
777 | struct unix_sock *u = unix_sk(sk); | |
778 | struct sockaddr_un *sunaddr=(struct sockaddr_un *)uaddr; | |
779 | struct dentry * dentry = NULL; | |
780 | struct nameidata nd; | |
781 | int err; | |
782 | unsigned hash; | |
783 | struct unix_address *addr; | |
784 | struct hlist_head *list; | |
785 | ||
786 | err = -EINVAL; | |
787 | if (sunaddr->sun_family != AF_UNIX) | |
788 | goto out; | |
789 | ||
790 | if (addr_len==sizeof(short)) { | |
791 | err = unix_autobind(sock); | |
792 | goto out; | |
793 | } | |
794 | ||
795 | err = unix_mkname(sunaddr, addr_len, &hash); | |
796 | if (err < 0) | |
797 | goto out; | |
798 | addr_len = err; | |
799 | ||
57b47a53 | 800 | mutex_lock(&u->readlock); |
1da177e4 LT |
801 | |
802 | err = -EINVAL; | |
803 | if (u->addr) | |
804 | goto out_up; | |
805 | ||
806 | err = -ENOMEM; | |
807 | addr = kmalloc(sizeof(*addr)+addr_len, GFP_KERNEL); | |
808 | if (!addr) | |
809 | goto out_up; | |
810 | ||
811 | memcpy(addr->name, sunaddr, addr_len); | |
812 | addr->len = addr_len; | |
813 | addr->hash = hash ^ sk->sk_type; | |
814 | atomic_set(&addr->refcnt, 1); | |
815 | ||
816 | if (sunaddr->sun_path[0]) { | |
817 | unsigned int mode; | |
818 | err = 0; | |
819 | /* | |
820 | * Get the parent directory, calculate the hash for last | |
821 | * component. | |
822 | */ | |
823 | err = path_lookup(sunaddr->sun_path, LOOKUP_PARENT, &nd); | |
824 | if (err) | |
825 | goto out_mknod_parent; | |
f81a0bff CH |
826 | |
827 | dentry = lookup_create(&nd, 0); | |
1da177e4 LT |
828 | err = PTR_ERR(dentry); |
829 | if (IS_ERR(dentry)) | |
830 | goto out_mknod_unlock; | |
f81a0bff | 831 | |
1da177e4 LT |
832 | /* |
833 | * All right, let's create it. | |
834 | */ | |
835 | mode = S_IFSOCK | | |
836 | (SOCK_INODE(sock)->i_mode & ~current->fs->umask); | |
837 | err = vfs_mknod(nd.dentry->d_inode, dentry, mode, 0); | |
838 | if (err) | |
839 | goto out_mknod_dput; | |
1b1dcc1b | 840 | mutex_unlock(&nd.dentry->d_inode->i_mutex); |
1da177e4 LT |
841 | dput(nd.dentry); |
842 | nd.dentry = dentry; | |
843 | ||
844 | addr->hash = UNIX_HASH_SIZE; | |
845 | } | |
846 | ||
fbe9cc4a | 847 | spin_lock(&unix_table_lock); |
1da177e4 LT |
848 | |
849 | if (!sunaddr->sun_path[0]) { | |
850 | err = -EADDRINUSE; | |
851 | if (__unix_find_socket_byname(sunaddr, addr_len, | |
852 | sk->sk_type, hash)) { | |
853 | unix_release_addr(addr); | |
854 | goto out_unlock; | |
855 | } | |
856 | ||
857 | list = &unix_socket_table[addr->hash]; | |
858 | } else { | |
859 | list = &unix_socket_table[dentry->d_inode->i_ino & (UNIX_HASH_SIZE-1)]; | |
860 | u->dentry = nd.dentry; | |
861 | u->mnt = nd.mnt; | |
862 | } | |
863 | ||
864 | err = 0; | |
865 | __unix_remove_socket(sk); | |
866 | u->addr = addr; | |
867 | __unix_insert_socket(list, sk); | |
868 | ||
869 | out_unlock: | |
fbe9cc4a | 870 | spin_unlock(&unix_table_lock); |
1da177e4 | 871 | out_up: |
57b47a53 | 872 | mutex_unlock(&u->readlock); |
1da177e4 LT |
873 | out: |
874 | return err; | |
875 | ||
876 | out_mknod_dput: | |
877 | dput(dentry); | |
878 | out_mknod_unlock: | |
1b1dcc1b | 879 | mutex_unlock(&nd.dentry->d_inode->i_mutex); |
1da177e4 LT |
880 | path_release(&nd); |
881 | out_mknod_parent: | |
882 | if (err==-EEXIST) | |
883 | err=-EADDRINUSE; | |
884 | unix_release_addr(addr); | |
885 | goto out_up; | |
886 | } | |
887 | ||
278a3de5 DM |
888 | static void unix_state_double_lock(struct sock *sk1, struct sock *sk2) |
889 | { | |
890 | if (unlikely(sk1 == sk2) || !sk2) { | |
891 | unix_state_lock(sk1); | |
892 | return; | |
893 | } | |
894 | if (sk1 < sk2) { | |
895 | unix_state_lock(sk1); | |
896 | unix_state_lock_nested(sk2); | |
897 | } else { | |
898 | unix_state_lock(sk2); | |
899 | unix_state_lock_nested(sk1); | |
900 | } | |
901 | } | |
902 | ||
903 | static void unix_state_double_unlock(struct sock *sk1, struct sock *sk2) | |
904 | { | |
905 | if (unlikely(sk1 == sk2) || !sk2) { | |
906 | unix_state_unlock(sk1); | |
907 | return; | |
908 | } | |
909 | unix_state_unlock(sk1); | |
910 | unix_state_unlock(sk2); | |
911 | } | |
912 | ||
1da177e4 LT |
913 | static int unix_dgram_connect(struct socket *sock, struct sockaddr *addr, |
914 | int alen, int flags) | |
915 | { | |
916 | struct sock *sk = sock->sk; | |
917 | struct sockaddr_un *sunaddr=(struct sockaddr_un*)addr; | |
918 | struct sock *other; | |
919 | unsigned hash; | |
920 | int err; | |
921 | ||
922 | if (addr->sa_family != AF_UNSPEC) { | |
923 | err = unix_mkname(sunaddr, alen, &hash); | |
924 | if (err < 0) | |
925 | goto out; | |
926 | alen = err; | |
927 | ||
928 | if (test_bit(SOCK_PASSCRED, &sock->flags) && | |
929 | !unix_sk(sk)->addr && (err = unix_autobind(sock)) != 0) | |
930 | goto out; | |
931 | ||
278a3de5 | 932 | restart: |
1da177e4 LT |
933 | other=unix_find_other(sunaddr, alen, sock->type, hash, &err); |
934 | if (!other) | |
935 | goto out; | |
936 | ||
278a3de5 DM |
937 | unix_state_double_lock(sk, other); |
938 | ||
939 | /* Apparently VFS overslept socket death. Retry. */ | |
940 | if (sock_flag(other, SOCK_DEAD)) { | |
941 | unix_state_double_unlock(sk, other); | |
942 | sock_put(other); | |
943 | goto restart; | |
944 | } | |
1da177e4 LT |
945 | |
946 | err = -EPERM; | |
947 | if (!unix_may_send(sk, other)) | |
948 | goto out_unlock; | |
949 | ||
950 | err = security_unix_may_send(sk->sk_socket, other->sk_socket); | |
951 | if (err) | |
952 | goto out_unlock; | |
953 | ||
954 | } else { | |
955 | /* | |
956 | * 1003.1g breaking connected state with AF_UNSPEC | |
957 | */ | |
958 | other = NULL; | |
278a3de5 | 959 | unix_state_double_lock(sk, other); |
1da177e4 LT |
960 | } |
961 | ||
962 | /* | |
963 | * If it was connected, reconnect. | |
964 | */ | |
965 | if (unix_peer(sk)) { | |
966 | struct sock *old_peer = unix_peer(sk); | |
967 | unix_peer(sk)=other; | |
278a3de5 | 968 | unix_state_double_unlock(sk, other); |
1da177e4 LT |
969 | |
970 | if (other != old_peer) | |
971 | unix_dgram_disconnected(sk, old_peer); | |
972 | sock_put(old_peer); | |
973 | } else { | |
974 | unix_peer(sk)=other; | |
278a3de5 | 975 | unix_state_double_unlock(sk, other); |
1da177e4 | 976 | } |
ac7bfa62 | 977 | return 0; |
1da177e4 LT |
978 | |
979 | out_unlock: | |
278a3de5 | 980 | unix_state_double_unlock(sk, other); |
1da177e4 LT |
981 | sock_put(other); |
982 | out: | |
983 | return err; | |
984 | } | |
985 | ||
986 | static long unix_wait_for_peer(struct sock *other, long timeo) | |
987 | { | |
988 | struct unix_sock *u = unix_sk(other); | |
989 | int sched; | |
990 | DEFINE_WAIT(wait); | |
991 | ||
992 | prepare_to_wait_exclusive(&u->peer_wait, &wait, TASK_INTERRUPTIBLE); | |
993 | ||
994 | sched = !sock_flag(other, SOCK_DEAD) && | |
995 | !(other->sk_shutdown & RCV_SHUTDOWN) && | |
64a14651 | 996 | (skb_queue_len(&other->sk_receive_queue) > |
1da177e4 LT |
997 | other->sk_max_ack_backlog); |
998 | ||
1c92b4e5 | 999 | unix_state_unlock(other); |
1da177e4 LT |
1000 | |
1001 | if (sched) | |
1002 | timeo = schedule_timeout(timeo); | |
1003 | ||
1004 | finish_wait(&u->peer_wait, &wait); | |
1005 | return timeo; | |
1006 | } | |
1007 | ||
1008 | static int unix_stream_connect(struct socket *sock, struct sockaddr *uaddr, | |
1009 | int addr_len, int flags) | |
1010 | { | |
1011 | struct sockaddr_un *sunaddr=(struct sockaddr_un *)uaddr; | |
1012 | struct sock *sk = sock->sk; | |
1013 | struct unix_sock *u = unix_sk(sk), *newu, *otheru; | |
1014 | struct sock *newsk = NULL; | |
1015 | struct sock *other = NULL; | |
1016 | struct sk_buff *skb = NULL; | |
1017 | unsigned hash; | |
1018 | int st; | |
1019 | int err; | |
1020 | long timeo; | |
1021 | ||
1022 | err = unix_mkname(sunaddr, addr_len, &hash); | |
1023 | if (err < 0) | |
1024 | goto out; | |
1025 | addr_len = err; | |
1026 | ||
1027 | if (test_bit(SOCK_PASSCRED, &sock->flags) | |
1028 | && !u->addr && (err = unix_autobind(sock)) != 0) | |
1029 | goto out; | |
1030 | ||
1031 | timeo = sock_sndtimeo(sk, flags & O_NONBLOCK); | |
1032 | ||
1033 | /* First of all allocate resources. | |
1034 | If we will make it after state is locked, | |
1035 | we will have to recheck all again in any case. | |
1036 | */ | |
1037 | ||
1038 | err = -ENOMEM; | |
1039 | ||
1040 | /* create new sock for complete connection */ | |
1041 | newsk = unix_create1(NULL); | |
1042 | if (newsk == NULL) | |
1043 | goto out; | |
1044 | ||
1045 | /* Allocate skb for sending to listening sock */ | |
1046 | skb = sock_wmalloc(newsk, 1, 0, GFP_KERNEL); | |
1047 | if (skb == NULL) | |
1048 | goto out; | |
1049 | ||
1050 | restart: | |
1051 | /* Find listening sock. */ | |
1052 | other = unix_find_other(sunaddr, addr_len, sk->sk_type, hash, &err); | |
1053 | if (!other) | |
1054 | goto out; | |
1055 | ||
1056 | /* Latch state of peer */ | |
1c92b4e5 | 1057 | unix_state_lock(other); |
1da177e4 LT |
1058 | |
1059 | /* Apparently VFS overslept socket death. Retry. */ | |
1060 | if (sock_flag(other, SOCK_DEAD)) { | |
1c92b4e5 | 1061 | unix_state_unlock(other); |
1da177e4 LT |
1062 | sock_put(other); |
1063 | goto restart; | |
1064 | } | |
1065 | ||
1066 | err = -ECONNREFUSED; | |
1067 | if (other->sk_state != TCP_LISTEN) | |
1068 | goto out_unlock; | |
1069 | ||
64a14651 | 1070 | if (skb_queue_len(&other->sk_receive_queue) > |
1da177e4 LT |
1071 | other->sk_max_ack_backlog) { |
1072 | err = -EAGAIN; | |
1073 | if (!timeo) | |
1074 | goto out_unlock; | |
1075 | ||
1076 | timeo = unix_wait_for_peer(other, timeo); | |
1077 | ||
1078 | err = sock_intr_errno(timeo); | |
1079 | if (signal_pending(current)) | |
1080 | goto out; | |
1081 | sock_put(other); | |
1082 | goto restart; | |
ac7bfa62 | 1083 | } |
1da177e4 LT |
1084 | |
1085 | /* Latch our state. | |
1086 | ||
1087 | It is tricky place. We need to grab write lock and cannot | |
1088 | drop lock on peer. It is dangerous because deadlock is | |
1089 | possible. Connect to self case and simultaneous | |
1090 | attempt to connect are eliminated by checking socket | |
1091 | state. other is TCP_LISTEN, if sk is TCP_LISTEN we | |
1092 | check this before attempt to grab lock. | |
1093 | ||
1094 | Well, and we have to recheck the state after socket locked. | |
1095 | */ | |
1096 | st = sk->sk_state; | |
1097 | ||
1098 | switch (st) { | |
1099 | case TCP_CLOSE: | |
1100 | /* This is ok... continue with connect */ | |
1101 | break; | |
1102 | case TCP_ESTABLISHED: | |
1103 | /* Socket is already connected */ | |
1104 | err = -EISCONN; | |
1105 | goto out_unlock; | |
1106 | default: | |
1107 | err = -EINVAL; | |
1108 | goto out_unlock; | |
1109 | } | |
1110 | ||
1c92b4e5 | 1111 | unix_state_lock_nested(sk); |
1da177e4 LT |
1112 | |
1113 | if (sk->sk_state != st) { | |
1c92b4e5 DM |
1114 | unix_state_unlock(sk); |
1115 | unix_state_unlock(other); | |
1da177e4 LT |
1116 | sock_put(other); |
1117 | goto restart; | |
1118 | } | |
1119 | ||
1120 | err = security_unix_stream_connect(sock, other->sk_socket, newsk); | |
1121 | if (err) { | |
1c92b4e5 | 1122 | unix_state_unlock(sk); |
1da177e4 LT |
1123 | goto out_unlock; |
1124 | } | |
1125 | ||
1126 | /* The way is open! Fastly set all the necessary fields... */ | |
1127 | ||
1128 | sock_hold(sk); | |
1129 | unix_peer(newsk) = sk; | |
1130 | newsk->sk_state = TCP_ESTABLISHED; | |
1131 | newsk->sk_type = sk->sk_type; | |
1132 | newsk->sk_peercred.pid = current->tgid; | |
1133 | newsk->sk_peercred.uid = current->euid; | |
1134 | newsk->sk_peercred.gid = current->egid; | |
1135 | newu = unix_sk(newsk); | |
1136 | newsk->sk_sleep = &newu->peer_wait; | |
1137 | otheru = unix_sk(other); | |
1138 | ||
1139 | /* copy address information from listening to new sock*/ | |
1140 | if (otheru->addr) { | |
1141 | atomic_inc(&otheru->addr->refcnt); | |
1142 | newu->addr = otheru->addr; | |
1143 | } | |
1144 | if (otheru->dentry) { | |
1145 | newu->dentry = dget(otheru->dentry); | |
1146 | newu->mnt = mntget(otheru->mnt); | |
1147 | } | |
1148 | ||
1149 | /* Set credentials */ | |
1150 | sk->sk_peercred = other->sk_peercred; | |
1151 | ||
1da177e4 LT |
1152 | sock->state = SS_CONNECTED; |
1153 | sk->sk_state = TCP_ESTABLISHED; | |
830a1e5c BL |
1154 | sock_hold(newsk); |
1155 | ||
1156 | smp_mb__after_atomic_inc(); /* sock_hold() does an atomic_inc() */ | |
1157 | unix_peer(sk) = newsk; | |
1da177e4 | 1158 | |
1c92b4e5 | 1159 | unix_state_unlock(sk); |
1da177e4 LT |
1160 | |
1161 | /* take ten and and send info to listening sock */ | |
1162 | spin_lock(&other->sk_receive_queue.lock); | |
1163 | __skb_queue_tail(&other->sk_receive_queue, skb); | |
1da177e4 | 1164 | spin_unlock(&other->sk_receive_queue.lock); |
1c92b4e5 | 1165 | unix_state_unlock(other); |
1da177e4 LT |
1166 | other->sk_data_ready(other, 0); |
1167 | sock_put(other); | |
1168 | return 0; | |
1169 | ||
1170 | out_unlock: | |
1171 | if (other) | |
1c92b4e5 | 1172 | unix_state_unlock(other); |
1da177e4 LT |
1173 | |
1174 | out: | |
1175 | if (skb) | |
1176 | kfree_skb(skb); | |
1177 | if (newsk) | |
1178 | unix_release_sock(newsk, 0); | |
1179 | if (other) | |
1180 | sock_put(other); | |
1181 | return err; | |
1182 | } | |
1183 | ||
1184 | static int unix_socketpair(struct socket *socka, struct socket *sockb) | |
1185 | { | |
1186 | struct sock *ska=socka->sk, *skb = sockb->sk; | |
1187 | ||
1188 | /* Join our sockets back to back */ | |
1189 | sock_hold(ska); | |
1190 | sock_hold(skb); | |
1191 | unix_peer(ska)=skb; | |
1192 | unix_peer(skb)=ska; | |
1193 | ska->sk_peercred.pid = skb->sk_peercred.pid = current->tgid; | |
1194 | ska->sk_peercred.uid = skb->sk_peercred.uid = current->euid; | |
1195 | ska->sk_peercred.gid = skb->sk_peercred.gid = current->egid; | |
1196 | ||
1197 | if (ska->sk_type != SOCK_DGRAM) { | |
1198 | ska->sk_state = TCP_ESTABLISHED; | |
1199 | skb->sk_state = TCP_ESTABLISHED; | |
1200 | socka->state = SS_CONNECTED; | |
1201 | sockb->state = SS_CONNECTED; | |
1202 | } | |
1203 | return 0; | |
1204 | } | |
1205 | ||
1206 | static int unix_accept(struct socket *sock, struct socket *newsock, int flags) | |
1207 | { | |
1208 | struct sock *sk = sock->sk; | |
1209 | struct sock *tsk; | |
1210 | struct sk_buff *skb; | |
1211 | int err; | |
1212 | ||
1213 | err = -EOPNOTSUPP; | |
1214 | if (sock->type!=SOCK_STREAM && sock->type!=SOCK_SEQPACKET) | |
1215 | goto out; | |
1216 | ||
1217 | err = -EINVAL; | |
1218 | if (sk->sk_state != TCP_LISTEN) | |
1219 | goto out; | |
1220 | ||
1221 | /* If socket state is TCP_LISTEN it cannot change (for now...), | |
1222 | * so that no locks are necessary. | |
1223 | */ | |
1224 | ||
1225 | skb = skb_recv_datagram(sk, 0, flags&O_NONBLOCK, &err); | |
1226 | if (!skb) { | |
1227 | /* This means receive shutdown. */ | |
1228 | if (err == 0) | |
1229 | err = -EINVAL; | |
1230 | goto out; | |
1231 | } | |
1232 | ||
1233 | tsk = skb->sk; | |
1234 | skb_free_datagram(sk, skb); | |
1235 | wake_up_interruptible(&unix_sk(sk)->peer_wait); | |
1236 | ||
1237 | /* attach accepted sock to socket */ | |
1c92b4e5 | 1238 | unix_state_lock(tsk); |
1da177e4 LT |
1239 | newsock->state = SS_CONNECTED; |
1240 | sock_graft(tsk, newsock); | |
1c92b4e5 | 1241 | unix_state_unlock(tsk); |
1da177e4 LT |
1242 | return 0; |
1243 | ||
1244 | out: | |
1245 | return err; | |
1246 | } | |
1247 | ||
1248 | ||
1249 | static int unix_getname(struct socket *sock, struct sockaddr *uaddr, int *uaddr_len, int peer) | |
1250 | { | |
1251 | struct sock *sk = sock->sk; | |
1252 | struct unix_sock *u; | |
1253 | struct sockaddr_un *sunaddr=(struct sockaddr_un *)uaddr; | |
1254 | int err = 0; | |
1255 | ||
1256 | if (peer) { | |
1257 | sk = unix_peer_get(sk); | |
1258 | ||
1259 | err = -ENOTCONN; | |
1260 | if (!sk) | |
1261 | goto out; | |
1262 | err = 0; | |
1263 | } else { | |
1264 | sock_hold(sk); | |
1265 | } | |
1266 | ||
1267 | u = unix_sk(sk); | |
1c92b4e5 | 1268 | unix_state_lock(sk); |
1da177e4 LT |
1269 | if (!u->addr) { |
1270 | sunaddr->sun_family = AF_UNIX; | |
1271 | sunaddr->sun_path[0] = 0; | |
1272 | *uaddr_len = sizeof(short); | |
1273 | } else { | |
1274 | struct unix_address *addr = u->addr; | |
1275 | ||
1276 | *uaddr_len = addr->len; | |
1277 | memcpy(sunaddr, addr->name, *uaddr_len); | |
1278 | } | |
1c92b4e5 | 1279 | unix_state_unlock(sk); |
1da177e4 LT |
1280 | sock_put(sk); |
1281 | out: | |
1282 | return err; | |
1283 | } | |
1284 | ||
1285 | static void unix_detach_fds(struct scm_cookie *scm, struct sk_buff *skb) | |
1286 | { | |
1287 | int i; | |
1288 | ||
1289 | scm->fp = UNIXCB(skb).fp; | |
1290 | skb->destructor = sock_wfree; | |
1291 | UNIXCB(skb).fp = NULL; | |
1292 | ||
1293 | for (i=scm->fp->count-1; i>=0; i--) | |
1294 | unix_notinflight(scm->fp->fp[i]); | |
1295 | } | |
1296 | ||
1297 | static void unix_destruct_fds(struct sk_buff *skb) | |
1298 | { | |
1299 | struct scm_cookie scm; | |
1300 | memset(&scm, 0, sizeof(scm)); | |
1301 | unix_detach_fds(&scm, skb); | |
1302 | ||
1303 | /* Alas, it calls VFS */ | |
1304 | /* So fscking what? fput() had been SMP-safe since the last Summer */ | |
1305 | scm_destroy(&scm); | |
1306 | sock_wfree(skb); | |
1307 | } | |
1308 | ||
1309 | static void unix_attach_fds(struct scm_cookie *scm, struct sk_buff *skb) | |
1310 | { | |
1311 | int i; | |
1312 | for (i=scm->fp->count-1; i>=0; i--) | |
1313 | unix_inflight(scm->fp->fp[i]); | |
1314 | UNIXCB(skb).fp = scm->fp; | |
1315 | skb->destructor = unix_destruct_fds; | |
1316 | scm->fp = NULL; | |
1317 | } | |
1318 | ||
1319 | /* | |
1320 | * Send AF_UNIX data. | |
1321 | */ | |
1322 | ||
1323 | static int unix_dgram_sendmsg(struct kiocb *kiocb, struct socket *sock, | |
1324 | struct msghdr *msg, size_t len) | |
1325 | { | |
1326 | struct sock_iocb *siocb = kiocb_to_siocb(kiocb); | |
1327 | struct sock *sk = sock->sk; | |
1328 | struct unix_sock *u = unix_sk(sk); | |
1329 | struct sockaddr_un *sunaddr=msg->msg_name; | |
1330 | struct sock *other = NULL; | |
1331 | int namelen = 0; /* fake GCC */ | |
1332 | int err; | |
1333 | unsigned hash; | |
1334 | struct sk_buff *skb; | |
1335 | long timeo; | |
1336 | struct scm_cookie tmp_scm; | |
1337 | ||
1338 | if (NULL == siocb->scm) | |
1339 | siocb->scm = &tmp_scm; | |
1340 | err = scm_send(sock, msg, siocb->scm); | |
1341 | if (err < 0) | |
1342 | return err; | |
1343 | ||
1344 | err = -EOPNOTSUPP; | |
1345 | if (msg->msg_flags&MSG_OOB) | |
1346 | goto out; | |
1347 | ||
1348 | if (msg->msg_namelen) { | |
1349 | err = unix_mkname(sunaddr, msg->msg_namelen, &hash); | |
1350 | if (err < 0) | |
1351 | goto out; | |
1352 | namelen = err; | |
1353 | } else { | |
1354 | sunaddr = NULL; | |
1355 | err = -ENOTCONN; | |
1356 | other = unix_peer_get(sk); | |
1357 | if (!other) | |
1358 | goto out; | |
1359 | } | |
1360 | ||
1361 | if (test_bit(SOCK_PASSCRED, &sock->flags) | |
1362 | && !u->addr && (err = unix_autobind(sock)) != 0) | |
1363 | goto out; | |
1364 | ||
1365 | err = -EMSGSIZE; | |
1366 | if (len > sk->sk_sndbuf - 32) | |
1367 | goto out; | |
1368 | ||
1369 | skb = sock_alloc_send_skb(sk, len, msg->msg_flags&MSG_DONTWAIT, &err); | |
1370 | if (skb==NULL) | |
1371 | goto out; | |
1372 | ||
1373 | memcpy(UNIXCREDS(skb), &siocb->scm->creds, sizeof(struct ucred)); | |
1374 | if (siocb->scm->fp) | |
1375 | unix_attach_fds(siocb->scm, skb); | |
dc49c1f9 | 1376 | unix_get_secdata(siocb->scm, skb); |
877ce7c1 | 1377 | |
badff6d0 | 1378 | skb_reset_transport_header(skb); |
1da177e4 LT |
1379 | err = memcpy_fromiovec(skb_put(skb,len), msg->msg_iov, len); |
1380 | if (err) | |
1381 | goto out_free; | |
1382 | ||
1383 | timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT); | |
1384 | ||
1385 | restart: | |
1386 | if (!other) { | |
1387 | err = -ECONNRESET; | |
1388 | if (sunaddr == NULL) | |
1389 | goto out_free; | |
1390 | ||
1391 | other = unix_find_other(sunaddr, namelen, sk->sk_type, | |
1392 | hash, &err); | |
1393 | if (other==NULL) | |
1394 | goto out_free; | |
1395 | } | |
1396 | ||
1c92b4e5 | 1397 | unix_state_lock(other); |
1da177e4 LT |
1398 | err = -EPERM; |
1399 | if (!unix_may_send(sk, other)) | |
1400 | goto out_unlock; | |
1401 | ||
1402 | if (sock_flag(other, SOCK_DEAD)) { | |
1403 | /* | |
1404 | * Check with 1003.1g - what should | |
1405 | * datagram error | |
1406 | */ | |
1c92b4e5 | 1407 | unix_state_unlock(other); |
1da177e4 LT |
1408 | sock_put(other); |
1409 | ||
1410 | err = 0; | |
1c92b4e5 | 1411 | unix_state_lock(sk); |
1da177e4 LT |
1412 | if (unix_peer(sk) == other) { |
1413 | unix_peer(sk)=NULL; | |
1c92b4e5 | 1414 | unix_state_unlock(sk); |
1da177e4 LT |
1415 | |
1416 | unix_dgram_disconnected(sk, other); | |
1417 | sock_put(other); | |
1418 | err = -ECONNREFUSED; | |
1419 | } else { | |
1c92b4e5 | 1420 | unix_state_unlock(sk); |
1da177e4 LT |
1421 | } |
1422 | ||
1423 | other = NULL; | |
1424 | if (err) | |
1425 | goto out_free; | |
1426 | goto restart; | |
1427 | } | |
1428 | ||
1429 | err = -EPIPE; | |
1430 | if (other->sk_shutdown & RCV_SHUTDOWN) | |
1431 | goto out_unlock; | |
1432 | ||
1433 | if (sk->sk_type != SOCK_SEQPACKET) { | |
1434 | err = security_unix_may_send(sk->sk_socket, other->sk_socket); | |
1435 | if (err) | |
1436 | goto out_unlock; | |
1437 | } | |
1438 | ||
1439 | if (unix_peer(other) != sk && | |
64a14651 | 1440 | (skb_queue_len(&other->sk_receive_queue) > |
1da177e4 LT |
1441 | other->sk_max_ack_backlog)) { |
1442 | if (!timeo) { | |
1443 | err = -EAGAIN; | |
1444 | goto out_unlock; | |
1445 | } | |
1446 | ||
1447 | timeo = unix_wait_for_peer(other, timeo); | |
1448 | ||
1449 | err = sock_intr_errno(timeo); | |
1450 | if (signal_pending(current)) | |
1451 | goto out_free; | |
1452 | ||
1453 | goto restart; | |
1454 | } | |
1455 | ||
1456 | skb_queue_tail(&other->sk_receive_queue, skb); | |
1c92b4e5 | 1457 | unix_state_unlock(other); |
1da177e4 LT |
1458 | other->sk_data_ready(other, len); |
1459 | sock_put(other); | |
1460 | scm_destroy(siocb->scm); | |
1461 | return len; | |
1462 | ||
1463 | out_unlock: | |
1c92b4e5 | 1464 | unix_state_unlock(other); |
1da177e4 LT |
1465 | out_free: |
1466 | kfree_skb(skb); | |
1467 | out: | |
1468 | if (other) | |
1469 | sock_put(other); | |
1470 | scm_destroy(siocb->scm); | |
1471 | return err; | |
1472 | } | |
1473 | ||
ac7bfa62 | 1474 | |
1da177e4 LT |
1475 | static int unix_stream_sendmsg(struct kiocb *kiocb, struct socket *sock, |
1476 | struct msghdr *msg, size_t len) | |
1477 | { | |
1478 | struct sock_iocb *siocb = kiocb_to_siocb(kiocb); | |
1479 | struct sock *sk = sock->sk; | |
1480 | struct sock *other = NULL; | |
1481 | struct sockaddr_un *sunaddr=msg->msg_name; | |
1482 | int err,size; | |
1483 | struct sk_buff *skb; | |
1484 | int sent=0; | |
1485 | struct scm_cookie tmp_scm; | |
1486 | ||
1487 | if (NULL == siocb->scm) | |
1488 | siocb->scm = &tmp_scm; | |
1489 | err = scm_send(sock, msg, siocb->scm); | |
1490 | if (err < 0) | |
1491 | return err; | |
1492 | ||
1493 | err = -EOPNOTSUPP; | |
1494 | if (msg->msg_flags&MSG_OOB) | |
1495 | goto out_err; | |
1496 | ||
1497 | if (msg->msg_namelen) { | |
1498 | err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP; | |
1499 | goto out_err; | |
1500 | } else { | |
1501 | sunaddr = NULL; | |
1502 | err = -ENOTCONN; | |
830a1e5c | 1503 | other = unix_peer(sk); |
1da177e4 LT |
1504 | if (!other) |
1505 | goto out_err; | |
1506 | } | |
1507 | ||
1508 | if (sk->sk_shutdown & SEND_SHUTDOWN) | |
1509 | goto pipe_err; | |
1510 | ||
1511 | while(sent < len) | |
1512 | { | |
1513 | /* | |
e9df7d7f BL |
1514 | * Optimisation for the fact that under 0.01% of X |
1515 | * messages typically need breaking up. | |
1da177e4 LT |
1516 | */ |
1517 | ||
e9df7d7f | 1518 | size = len-sent; |
1da177e4 LT |
1519 | |
1520 | /* Keep two messages in the pipe so it schedules better */ | |
e9df7d7f BL |
1521 | if (size > ((sk->sk_sndbuf >> 1) - 64)) |
1522 | size = (sk->sk_sndbuf >> 1) - 64; | |
1da177e4 LT |
1523 | |
1524 | if (size > SKB_MAX_ALLOC) | |
1525 | size = SKB_MAX_ALLOC; | |
ac7bfa62 | 1526 | |
1da177e4 LT |
1527 | /* |
1528 | * Grab a buffer | |
1529 | */ | |
ac7bfa62 | 1530 | |
1da177e4 LT |
1531 | skb=sock_alloc_send_skb(sk,size,msg->msg_flags&MSG_DONTWAIT, &err); |
1532 | ||
1533 | if (skb==NULL) | |
1534 | goto out_err; | |
1535 | ||
1536 | /* | |
1537 | * If you pass two values to the sock_alloc_send_skb | |
1538 | * it tries to grab the large buffer with GFP_NOFS | |
1539 | * (which can fail easily), and if it fails grab the | |
1540 | * fallback size buffer which is under a page and will | |
1541 | * succeed. [Alan] | |
1542 | */ | |
1543 | size = min_t(int, size, skb_tailroom(skb)); | |
1544 | ||
1545 | memcpy(UNIXCREDS(skb), &siocb->scm->creds, sizeof(struct ucred)); | |
1546 | if (siocb->scm->fp) | |
1547 | unix_attach_fds(siocb->scm, skb); | |
1548 | ||
1549 | if ((err = memcpy_fromiovec(skb_put(skb,size), msg->msg_iov, size)) != 0) { | |
1550 | kfree_skb(skb); | |
1551 | goto out_err; | |
1552 | } | |
1553 | ||
1c92b4e5 | 1554 | unix_state_lock(other); |
1da177e4 LT |
1555 | |
1556 | if (sock_flag(other, SOCK_DEAD) || | |
1557 | (other->sk_shutdown & RCV_SHUTDOWN)) | |
1558 | goto pipe_err_free; | |
1559 | ||
1560 | skb_queue_tail(&other->sk_receive_queue, skb); | |
1c92b4e5 | 1561 | unix_state_unlock(other); |
1da177e4 LT |
1562 | other->sk_data_ready(other, size); |
1563 | sent+=size; | |
1564 | } | |
1da177e4 LT |
1565 | |
1566 | scm_destroy(siocb->scm); | |
1567 | siocb->scm = NULL; | |
1568 | ||
1569 | return sent; | |
1570 | ||
1571 | pipe_err_free: | |
1c92b4e5 | 1572 | unix_state_unlock(other); |
1da177e4 LT |
1573 | kfree_skb(skb); |
1574 | pipe_err: | |
1575 | if (sent==0 && !(msg->msg_flags&MSG_NOSIGNAL)) | |
1576 | send_sig(SIGPIPE,current,0); | |
1577 | err = -EPIPE; | |
1578 | out_err: | |
1da177e4 LT |
1579 | scm_destroy(siocb->scm); |
1580 | siocb->scm = NULL; | |
1581 | return sent ? : err; | |
1582 | } | |
1583 | ||
1584 | static int unix_seqpacket_sendmsg(struct kiocb *kiocb, struct socket *sock, | |
1585 | struct msghdr *msg, size_t len) | |
1586 | { | |
1587 | int err; | |
1588 | struct sock *sk = sock->sk; | |
ac7bfa62 | 1589 | |
1da177e4 LT |
1590 | err = sock_error(sk); |
1591 | if (err) | |
1592 | return err; | |
1593 | ||
1594 | if (sk->sk_state != TCP_ESTABLISHED) | |
1595 | return -ENOTCONN; | |
1596 | ||
1597 | if (msg->msg_namelen) | |
1598 | msg->msg_namelen = 0; | |
1599 | ||
1600 | return unix_dgram_sendmsg(kiocb, sock, msg, len); | |
1601 | } | |
ac7bfa62 | 1602 | |
1da177e4 LT |
1603 | static void unix_copy_addr(struct msghdr *msg, struct sock *sk) |
1604 | { | |
1605 | struct unix_sock *u = unix_sk(sk); | |
1606 | ||
1607 | msg->msg_namelen = 0; | |
1608 | if (u->addr) { | |
1609 | msg->msg_namelen = u->addr->len; | |
1610 | memcpy(msg->msg_name, u->addr->name, u->addr->len); | |
1611 | } | |
1612 | } | |
1613 | ||
1614 | static int unix_dgram_recvmsg(struct kiocb *iocb, struct socket *sock, | |
1615 | struct msghdr *msg, size_t size, | |
1616 | int flags) | |
1617 | { | |
1618 | struct sock_iocb *siocb = kiocb_to_siocb(iocb); | |
1619 | struct scm_cookie tmp_scm; | |
1620 | struct sock *sk = sock->sk; | |
1621 | struct unix_sock *u = unix_sk(sk); | |
1622 | int noblock = flags & MSG_DONTWAIT; | |
1623 | struct sk_buff *skb; | |
1624 | int err; | |
1625 | ||
1626 | err = -EOPNOTSUPP; | |
1627 | if (flags&MSG_OOB) | |
1628 | goto out; | |
1629 | ||
1630 | msg->msg_namelen = 0; | |
1631 | ||
57b47a53 | 1632 | mutex_lock(&u->readlock); |
1da177e4 LT |
1633 | |
1634 | skb = skb_recv_datagram(sk, flags, noblock, &err); | |
1635 | if (!skb) | |
1636 | goto out_unlock; | |
1637 | ||
1638 | wake_up_interruptible(&u->peer_wait); | |
1639 | ||
1640 | if (msg->msg_name) | |
1641 | unix_copy_addr(msg, skb->sk); | |
1642 | ||
1643 | if (size > skb->len) | |
1644 | size = skb->len; | |
1645 | else if (size < skb->len) | |
1646 | msg->msg_flags |= MSG_TRUNC; | |
1647 | ||
1648 | err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, size); | |
1649 | if (err) | |
1650 | goto out_free; | |
1651 | ||
1652 | if (!siocb->scm) { | |
1653 | siocb->scm = &tmp_scm; | |
1654 | memset(&tmp_scm, 0, sizeof(tmp_scm)); | |
1655 | } | |
1656 | siocb->scm->creds = *UNIXCREDS(skb); | |
877ce7c1 | 1657 | unix_set_secdata(siocb->scm, skb); |
1da177e4 LT |
1658 | |
1659 | if (!(flags & MSG_PEEK)) | |
1660 | { | |
1661 | if (UNIXCB(skb).fp) | |
1662 | unix_detach_fds(siocb->scm, skb); | |
1663 | } | |
ac7bfa62 | 1664 | else |
1da177e4 LT |
1665 | { |
1666 | /* It is questionable: on PEEK we could: | |
1667 | - do not return fds - good, but too simple 8) | |
1668 | - return fds, and do not return them on read (old strategy, | |
1669 | apparently wrong) | |
1670 | - clone fds (I chose it for now, it is the most universal | |
1671 | solution) | |
ac7bfa62 YH |
1672 | |
1673 | POSIX 1003.1g does not actually define this clearly | |
1674 | at all. POSIX 1003.1g doesn't define a lot of things | |
1675 | clearly however! | |
1676 | ||
1da177e4 LT |
1677 | */ |
1678 | if (UNIXCB(skb).fp) | |
1679 | siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp); | |
1680 | } | |
1681 | err = size; | |
1682 | ||
1683 | scm_recv(sock, msg, siocb->scm, flags); | |
1684 | ||
1685 | out_free: | |
1686 | skb_free_datagram(sk,skb); | |
1687 | out_unlock: | |
57b47a53 | 1688 | mutex_unlock(&u->readlock); |
1da177e4 LT |
1689 | out: |
1690 | return err; | |
1691 | } | |
1692 | ||
1693 | /* | |
1694 | * Sleep until data has arrive. But check for races.. | |
1695 | */ | |
ac7bfa62 | 1696 | |
1da177e4 LT |
1697 | static long unix_stream_data_wait(struct sock * sk, long timeo) |
1698 | { | |
1699 | DEFINE_WAIT(wait); | |
1700 | ||
1c92b4e5 | 1701 | unix_state_lock(sk); |
1da177e4 LT |
1702 | |
1703 | for (;;) { | |
1704 | prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE); | |
1705 | ||
b03efcfb | 1706 | if (!skb_queue_empty(&sk->sk_receive_queue) || |
1da177e4 LT |
1707 | sk->sk_err || |
1708 | (sk->sk_shutdown & RCV_SHUTDOWN) || | |
1709 | signal_pending(current) || | |
1710 | !timeo) | |
1711 | break; | |
1712 | ||
1713 | set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags); | |
1c92b4e5 | 1714 | unix_state_unlock(sk); |
1da177e4 | 1715 | timeo = schedule_timeout(timeo); |
1c92b4e5 | 1716 | unix_state_lock(sk); |
1da177e4 LT |
1717 | clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags); |
1718 | } | |
1719 | ||
1720 | finish_wait(sk->sk_sleep, &wait); | |
1c92b4e5 | 1721 | unix_state_unlock(sk); |
1da177e4 LT |
1722 | return timeo; |
1723 | } | |
1724 | ||
1725 | ||
1726 | ||
1727 | static int unix_stream_recvmsg(struct kiocb *iocb, struct socket *sock, | |
1728 | struct msghdr *msg, size_t size, | |
1729 | int flags) | |
1730 | { | |
1731 | struct sock_iocb *siocb = kiocb_to_siocb(iocb); | |
1732 | struct scm_cookie tmp_scm; | |
1733 | struct sock *sk = sock->sk; | |
1734 | struct unix_sock *u = unix_sk(sk); | |
1735 | struct sockaddr_un *sunaddr=msg->msg_name; | |
1736 | int copied = 0; | |
1737 | int check_creds = 0; | |
1738 | int target; | |
1739 | int err = 0; | |
1740 | long timeo; | |
1741 | ||
1742 | err = -EINVAL; | |
1743 | if (sk->sk_state != TCP_ESTABLISHED) | |
1744 | goto out; | |
1745 | ||
1746 | err = -EOPNOTSUPP; | |
1747 | if (flags&MSG_OOB) | |
1748 | goto out; | |
1749 | ||
1750 | target = sock_rcvlowat(sk, flags&MSG_WAITALL, size); | |
1751 | timeo = sock_rcvtimeo(sk, flags&MSG_DONTWAIT); | |
1752 | ||
1753 | msg->msg_namelen = 0; | |
1754 | ||
1755 | /* Lock the socket to prevent queue disordering | |
1756 | * while sleeps in memcpy_tomsg | |
1757 | */ | |
1758 | ||
1759 | if (!siocb->scm) { | |
1760 | siocb->scm = &tmp_scm; | |
1761 | memset(&tmp_scm, 0, sizeof(tmp_scm)); | |
1762 | } | |
1763 | ||
57b47a53 | 1764 | mutex_lock(&u->readlock); |
1da177e4 LT |
1765 | |
1766 | do | |
1767 | { | |
1768 | int chunk; | |
1769 | struct sk_buff *skb; | |
1770 | ||
3c0d2f37 | 1771 | unix_state_lock(sk); |
1da177e4 LT |
1772 | skb = skb_dequeue(&sk->sk_receive_queue); |
1773 | if (skb==NULL) | |
1774 | { | |
1775 | if (copied >= target) | |
3c0d2f37 | 1776 | goto unlock; |
1da177e4 LT |
1777 | |
1778 | /* | |
1779 | * POSIX 1003.1g mandates this order. | |
1780 | */ | |
ac7bfa62 | 1781 | |
1da177e4 | 1782 | if ((err = sock_error(sk)) != 0) |
3c0d2f37 | 1783 | goto unlock; |
1da177e4 | 1784 | if (sk->sk_shutdown & RCV_SHUTDOWN) |
3c0d2f37 MS |
1785 | goto unlock; |
1786 | ||
1787 | unix_state_unlock(sk); | |
1da177e4 LT |
1788 | err = -EAGAIN; |
1789 | if (!timeo) | |
1790 | break; | |
57b47a53 | 1791 | mutex_unlock(&u->readlock); |
1da177e4 LT |
1792 | |
1793 | timeo = unix_stream_data_wait(sk, timeo); | |
1794 | ||
1795 | if (signal_pending(current)) { | |
1796 | err = sock_intr_errno(timeo); | |
1797 | goto out; | |
1798 | } | |
57b47a53 | 1799 | mutex_lock(&u->readlock); |
1da177e4 | 1800 | continue; |
3c0d2f37 MS |
1801 | unlock: |
1802 | unix_state_unlock(sk); | |
1803 | break; | |
1da177e4 | 1804 | } |
3c0d2f37 | 1805 | unix_state_unlock(sk); |
1da177e4 LT |
1806 | |
1807 | if (check_creds) { | |
1808 | /* Never glue messages from different writers */ | |
1809 | if (memcmp(UNIXCREDS(skb), &siocb->scm->creds, sizeof(siocb->scm->creds)) != 0) { | |
1810 | skb_queue_head(&sk->sk_receive_queue, skb); | |
1811 | break; | |
1812 | } | |
1813 | } else { | |
1814 | /* Copy credentials */ | |
1815 | siocb->scm->creds = *UNIXCREDS(skb); | |
1816 | check_creds = 1; | |
1817 | } | |
1818 | ||
1819 | /* Copy address just once */ | |
1820 | if (sunaddr) | |
1821 | { | |
1822 | unix_copy_addr(msg, skb->sk); | |
1823 | sunaddr = NULL; | |
1824 | } | |
1825 | ||
1826 | chunk = min_t(unsigned int, skb->len, size); | |
1827 | if (memcpy_toiovec(msg->msg_iov, skb->data, chunk)) { | |
1828 | skb_queue_head(&sk->sk_receive_queue, skb); | |
1829 | if (copied == 0) | |
1830 | copied = -EFAULT; | |
1831 | break; | |
1832 | } | |
1833 | copied += chunk; | |
1834 | size -= chunk; | |
1835 | ||
1836 | /* Mark read part of skb as used */ | |
1837 | if (!(flags & MSG_PEEK)) | |
1838 | { | |
1839 | skb_pull(skb, chunk); | |
1840 | ||
1841 | if (UNIXCB(skb).fp) | |
1842 | unix_detach_fds(siocb->scm, skb); | |
1843 | ||
1844 | /* put the skb back if we didn't use it up.. */ | |
1845 | if (skb->len) | |
1846 | { | |
1847 | skb_queue_head(&sk->sk_receive_queue, skb); | |
1848 | break; | |
1849 | } | |
1850 | ||
1851 | kfree_skb(skb); | |
1852 | ||
1853 | if (siocb->scm->fp) | |
1854 | break; | |
1855 | } | |
1856 | else | |
1857 | { | |
1858 | /* It is questionable, see note in unix_dgram_recvmsg. | |
1859 | */ | |
1860 | if (UNIXCB(skb).fp) | |
1861 | siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp); | |
1862 | ||
1863 | /* put message back and return */ | |
1864 | skb_queue_head(&sk->sk_receive_queue, skb); | |
1865 | break; | |
1866 | } | |
1867 | } while (size); | |
1868 | ||
57b47a53 | 1869 | mutex_unlock(&u->readlock); |
1da177e4 LT |
1870 | scm_recv(sock, msg, siocb->scm, flags); |
1871 | out: | |
1872 | return copied ? : err; | |
1873 | } | |
1874 | ||
1875 | static int unix_shutdown(struct socket *sock, int mode) | |
1876 | { | |
1877 | struct sock *sk = sock->sk; | |
1878 | struct sock *other; | |
1879 | ||
1880 | mode = (mode+1)&(RCV_SHUTDOWN|SEND_SHUTDOWN); | |
1881 | ||
1882 | if (mode) { | |
1c92b4e5 | 1883 | unix_state_lock(sk); |
1da177e4 LT |
1884 | sk->sk_shutdown |= mode; |
1885 | other=unix_peer(sk); | |
1886 | if (other) | |
1887 | sock_hold(other); | |
1c92b4e5 | 1888 | unix_state_unlock(sk); |
1da177e4 LT |
1889 | sk->sk_state_change(sk); |
1890 | ||
1891 | if (other && | |
1892 | (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET)) { | |
1893 | ||
1894 | int peer_mode = 0; | |
1895 | ||
1896 | if (mode&RCV_SHUTDOWN) | |
1897 | peer_mode |= SEND_SHUTDOWN; | |
1898 | if (mode&SEND_SHUTDOWN) | |
1899 | peer_mode |= RCV_SHUTDOWN; | |
1c92b4e5 | 1900 | unix_state_lock(other); |
1da177e4 | 1901 | other->sk_shutdown |= peer_mode; |
1c92b4e5 | 1902 | unix_state_unlock(other); |
1da177e4 LT |
1903 | other->sk_state_change(other); |
1904 | read_lock(&other->sk_callback_lock); | |
1905 | if (peer_mode == SHUTDOWN_MASK) | |
1906 | sk_wake_async(other,1,POLL_HUP); | |
1907 | else if (peer_mode & RCV_SHUTDOWN) | |
1908 | sk_wake_async(other,1,POLL_IN); | |
1909 | read_unlock(&other->sk_callback_lock); | |
1910 | } | |
1911 | if (other) | |
1912 | sock_put(other); | |
1913 | } | |
1914 | return 0; | |
1915 | } | |
1916 | ||
1917 | static int unix_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) | |
1918 | { | |
1919 | struct sock *sk = sock->sk; | |
1920 | long amount=0; | |
1921 | int err; | |
1922 | ||
1923 | switch(cmd) | |
1924 | { | |
1925 | case SIOCOUTQ: | |
1926 | amount = atomic_read(&sk->sk_wmem_alloc); | |
1927 | err = put_user(amount, (int __user *)arg); | |
1928 | break; | |
1929 | case SIOCINQ: | |
1930 | { | |
1931 | struct sk_buff *skb; | |
1932 | ||
1933 | if (sk->sk_state == TCP_LISTEN) { | |
1934 | err = -EINVAL; | |
1935 | break; | |
1936 | } | |
1937 | ||
1938 | spin_lock(&sk->sk_receive_queue.lock); | |
1939 | if (sk->sk_type == SOCK_STREAM || | |
1940 | sk->sk_type == SOCK_SEQPACKET) { | |
1941 | skb_queue_walk(&sk->sk_receive_queue, skb) | |
1942 | amount += skb->len; | |
1943 | } else { | |
1944 | skb = skb_peek(&sk->sk_receive_queue); | |
1945 | if (skb) | |
1946 | amount=skb->len; | |
1947 | } | |
1948 | spin_unlock(&sk->sk_receive_queue.lock); | |
1949 | err = put_user(amount, (int __user *)arg); | |
1950 | break; | |
1951 | } | |
1952 | ||
1953 | default: | |
b5e5fa5e | 1954 | err = -ENOIOCTLCMD; |
1da177e4 LT |
1955 | break; |
1956 | } | |
1957 | return err; | |
1958 | } | |
1959 | ||
1960 | static unsigned int unix_poll(struct file * file, struct socket *sock, poll_table *wait) | |
1961 | { | |
1962 | struct sock *sk = sock->sk; | |
1963 | unsigned int mask; | |
1964 | ||
1965 | poll_wait(file, sk->sk_sleep, wait); | |
1966 | mask = 0; | |
1967 | ||
1968 | /* exceptional events? */ | |
1969 | if (sk->sk_err) | |
1970 | mask |= POLLERR; | |
1971 | if (sk->sk_shutdown == SHUTDOWN_MASK) | |
1972 | mask |= POLLHUP; | |
f348d70a DL |
1973 | if (sk->sk_shutdown & RCV_SHUTDOWN) |
1974 | mask |= POLLRDHUP; | |
1da177e4 LT |
1975 | |
1976 | /* readable? */ | |
1977 | if (!skb_queue_empty(&sk->sk_receive_queue) || | |
1978 | (sk->sk_shutdown & RCV_SHUTDOWN)) | |
1979 | mask |= POLLIN | POLLRDNORM; | |
1980 | ||
1981 | /* Connection-based need to check for termination and startup */ | |
1982 | if ((sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) && sk->sk_state == TCP_CLOSE) | |
1983 | mask |= POLLHUP; | |
1984 | ||
1985 | /* | |
1986 | * we set writable also when the other side has shut down the | |
1987 | * connection. This prevents stuck sockets. | |
1988 | */ | |
1989 | if (unix_writable(sk)) | |
1990 | mask |= POLLOUT | POLLWRNORM | POLLWRBAND; | |
1991 | ||
1992 | return mask; | |
1993 | } | |
1994 | ||
1995 | ||
1996 | #ifdef CONFIG_PROC_FS | |
1997 | static struct sock *unix_seq_idx(int *iter, loff_t pos) | |
1998 | { | |
1999 | loff_t off = 0; | |
2000 | struct sock *s; | |
2001 | ||
2002 | for (s = first_unix_socket(iter); s; s = next_unix_socket(iter, s)) { | |
ac7bfa62 | 2003 | if (off == pos) |
1da177e4 LT |
2004 | return s; |
2005 | ++off; | |
2006 | } | |
2007 | return NULL; | |
2008 | } | |
2009 | ||
2010 | ||
2011 | static void *unix_seq_start(struct seq_file *seq, loff_t *pos) | |
2012 | { | |
fbe9cc4a | 2013 | spin_lock(&unix_table_lock); |
1da177e4 LT |
2014 | return *pos ? unix_seq_idx(seq->private, *pos - 1) : ((void *) 1); |
2015 | } | |
2016 | ||
2017 | static void *unix_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
2018 | { | |
2019 | ++*pos; | |
2020 | ||
ac7bfa62 | 2021 | if (v == (void *)1) |
1da177e4 LT |
2022 | return first_unix_socket(seq->private); |
2023 | return next_unix_socket(seq->private, v); | |
2024 | } | |
2025 | ||
2026 | static void unix_seq_stop(struct seq_file *seq, void *v) | |
2027 | { | |
fbe9cc4a | 2028 | spin_unlock(&unix_table_lock); |
1da177e4 LT |
2029 | } |
2030 | ||
2031 | static int unix_seq_show(struct seq_file *seq, void *v) | |
2032 | { | |
ac7bfa62 | 2033 | |
1da177e4 LT |
2034 | if (v == (void *)1) |
2035 | seq_puts(seq, "Num RefCount Protocol Flags Type St " | |
2036 | "Inode Path\n"); | |
2037 | else { | |
2038 | struct sock *s = v; | |
2039 | struct unix_sock *u = unix_sk(s); | |
1c92b4e5 | 2040 | unix_state_lock(s); |
1da177e4 LT |
2041 | |
2042 | seq_printf(seq, "%p: %08X %08X %08X %04X %02X %5lu", | |
2043 | s, | |
2044 | atomic_read(&s->sk_refcnt), | |
2045 | 0, | |
2046 | s->sk_state == TCP_LISTEN ? __SO_ACCEPTCON : 0, | |
2047 | s->sk_type, | |
2048 | s->sk_socket ? | |
2049 | (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTED : SS_UNCONNECTED) : | |
2050 | (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTING : SS_DISCONNECTING), | |
2051 | sock_i_ino(s)); | |
2052 | ||
2053 | if (u->addr) { | |
2054 | int i, len; | |
2055 | seq_putc(seq, ' '); | |
2056 | ||
2057 | i = 0; | |
2058 | len = u->addr->len - sizeof(short); | |
2059 | if (!UNIX_ABSTRACT(s)) | |
2060 | len--; | |
2061 | else { | |
2062 | seq_putc(seq, '@'); | |
2063 | i++; | |
2064 | } | |
2065 | for ( ; i < len; i++) | |
2066 | seq_putc(seq, u->addr->name->sun_path[i]); | |
2067 | } | |
1c92b4e5 | 2068 | unix_state_unlock(s); |
1da177e4 LT |
2069 | seq_putc(seq, '\n'); |
2070 | } | |
2071 | ||
2072 | return 0; | |
2073 | } | |
2074 | ||
56b3d975 | 2075 | static const struct seq_operations unix_seq_ops = { |
1da177e4 LT |
2076 | .start = unix_seq_start, |
2077 | .next = unix_seq_next, | |
2078 | .stop = unix_seq_stop, | |
2079 | .show = unix_seq_show, | |
2080 | }; | |
2081 | ||
2082 | ||
2083 | static int unix_seq_open(struct inode *inode, struct file *file) | |
2084 | { | |
2085 | struct seq_file *seq; | |
2086 | int rc = -ENOMEM; | |
2087 | int *iter = kmalloc(sizeof(int), GFP_KERNEL); | |
2088 | ||
2089 | if (!iter) | |
2090 | goto out; | |
2091 | ||
2092 | rc = seq_open(file, &unix_seq_ops); | |
2093 | if (rc) | |
2094 | goto out_kfree; | |
2095 | ||
2096 | seq = file->private_data; | |
2097 | seq->private = iter; | |
2098 | *iter = 0; | |
2099 | out: | |
2100 | return rc; | |
2101 | out_kfree: | |
2102 | kfree(iter); | |
2103 | goto out; | |
2104 | } | |
2105 | ||
da7071d7 | 2106 | static const struct file_operations unix_seq_fops = { |
1da177e4 LT |
2107 | .owner = THIS_MODULE, |
2108 | .open = unix_seq_open, | |
2109 | .read = seq_read, | |
2110 | .llseek = seq_lseek, | |
2111 | .release = seq_release_private, | |
2112 | }; | |
2113 | ||
2114 | #endif | |
2115 | ||
2116 | static struct net_proto_family unix_family_ops = { | |
2117 | .family = PF_UNIX, | |
2118 | .create = unix_create, | |
2119 | .owner = THIS_MODULE, | |
2120 | }; | |
2121 | ||
1da177e4 LT |
2122 | static int __init af_unix_init(void) |
2123 | { | |
2124 | int rc = -1; | |
2125 | struct sk_buff *dummy_skb; | |
2126 | ||
ef047f5e | 2127 | BUILD_BUG_ON(sizeof(struct unix_skb_parms) > sizeof(dummy_skb->cb)); |
1da177e4 LT |
2128 | |
2129 | rc = proto_register(&unix_proto, 1); | |
ac7bfa62 YH |
2130 | if (rc != 0) { |
2131 | printk(KERN_CRIT "%s: Cannot create unix_sock SLAB cache!\n", | |
1da177e4 LT |
2132 | __FUNCTION__); |
2133 | goto out; | |
2134 | } | |
2135 | ||
2136 | sock_register(&unix_family_ops); | |
2137 | #ifdef CONFIG_PROC_FS | |
2138 | proc_net_fops_create("unix", 0, &unix_seq_fops); | |
2139 | #endif | |
2140 | unix_sysctl_register(); | |
2141 | out: | |
2142 | return rc; | |
2143 | } | |
2144 | ||
2145 | static void __exit af_unix_exit(void) | |
2146 | { | |
2147 | sock_unregister(PF_UNIX); | |
2148 | unix_sysctl_unregister(); | |
2149 | proc_net_remove("unix"); | |
2150 | proto_unregister(&unix_proto); | |
2151 | } | |
2152 | ||
2153 | module_init(af_unix_init); | |
2154 | module_exit(af_unix_exit); | |
2155 | ||
2156 | MODULE_LICENSE("GPL"); | |
2157 | MODULE_ALIAS_NETPROTO(PF_UNIX); |