Linux kernel & device driver programming

Cross-Referenced Linux and Device Driver Code

[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ]
Version: [ 2.6.11.8 ] [ 2.6.25 ] [ 2.6.25.8 ] [ 2.6.31.13 ] Architecture: [ i386 ]
  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>
 86 #include <linux/config.h>
 87 #include <linux/kernel.h>
 88 #include <linux/major.h>
 89 #include <linux/signal.h>
 90 #include <linux/sched.h>
 91 #include <linux/errno.h>
 92 #include <linux/string.h>
 93 #include <linux/stat.h>
 94 #include <linux/dcache.h>
 95 #include <linux/namei.h>
 96 #include <linux/socket.h>
 97 #include <linux/un.h>
 98 #include <linux/fcntl.h>
 99 #include <linux/termios.h>
100 #include <linux/sockios.h>
101 #include <linux/net.h>
102 #include <linux/in.h>
103 #include <linux/fs.h>
104 #include <linux/slab.h>
105 #include <asm/uaccess.h>
106 #include <linux/skbuff.h>
107 #include <linux/netdevice.h>
108 #include <net/sock.h>
109 #include <linux/tcp.h>
110 #include <net/af_unix.h>
111 #include <linux/proc_fs.h>
112 #include <linux/seq_file.h>
113 #include <net/scm.h>
114 #include <linux/init.h>
115 #include <linux/poll.h>
116 #include <linux/smp_lock.h>
117 #include <linux/rtnetlink.h>
118 #include <linux/mount.h>
119 #include <net/checksum.h>
120 #include <linux/security.h>
121 
122 int sysctl_unix_max_dgram_qlen = 10;
123 
124 static kmem_cache_t *unix_sk_cachep;
125 
126 struct hlist_head unix_socket_table[UNIX_HASH_SIZE + 1];
127 DEFINE_RWLOCK(unix_table_lock);
128 static atomic_t unix_nr_socks = ATOMIC_INIT(0);
129 
130 #define unix_sockets_unbound    (&unix_socket_table[UNIX_HASH_SIZE])
131 
132 #define UNIX_ABSTRACT(sk)       (unix_sk(sk)->addr->hash != UNIX_HASH_SIZE)
133 
134 /*
135  *  SMP locking strategy:
136  *    hash table is protected with rwlock unix_table_lock
137  *    each socket state is protected by separate rwlock.
138  */
139 
140 static inline unsigned unix_hash_fold(unsigned hash)
141 {
142         hash ^= hash>>16;
143         hash ^= hash>>8;
144         return hash&(UNIX_HASH_SIZE-1);
145 }
146 
147 #define unix_peer(sk) (unix_sk(sk)->peer)
148 
149 static inline int unix_our_peer(struct sock *sk, struct sock *osk)
150 {
151         return unix_peer(osk) == sk;
152 }
153 
154 static inline int unix_may_send(struct sock *sk, struct sock *osk)
155 {
156         return (unix_peer(osk) == NULL || unix_our_peer(sk, osk));
157 }
158 
159 static struct sock *unix_peer_get(struct sock *s)
160 {
161         struct sock *peer;
162 
163         unix_state_rlock(s);
164         peer = unix_peer(s);
165         if (peer)
166                 sock_hold(peer);
167         unix_state_runlock(s);
168         return peer;
169 }
170 
171 static inline void unix_release_addr(struct unix_address *addr)
172 {
173         if (atomic_dec_and_test(&addr->refcnt))
174                 kfree(addr);
175 }
176 
177 /*
178  *      Check unix socket name:
179  *              - should be not zero length.
180  *              - if started by not zero, should be NULL terminated (FS object)
181  *              - if started by zero, it is abstract name.
182  */
183  
184 static int unix_mkname(struct sockaddr_un * sunaddr, int len, unsigned *hashp)
185 {
186         if (len <= sizeof(short) || len > sizeof(*sunaddr))
187                 return -EINVAL;
188         if (!sunaddr || sunaddr->sun_family != AF_UNIX)
189                 return -EINVAL;
190         if (sunaddr->sun_path[0]) {
191                 ((char *)sunaddr)[len]=0;
192                 len = strlen(sunaddr->sun_path)+1+sizeof(short);
193                 return len;
194         }
195 
196         *hashp = unix_hash_fold(csum_partial((char*)sunaddr, len, 0));
197         return len;
198 }
199 
200 static void __unix_remove_socket(struct sock *sk)
201 {
202         sk_del_node_init(sk);
203 }
204 
205 static void __unix_insert_socket(struct hlist_head *list, struct sock *sk)
206 {
207         BUG_TRAP(sk_unhashed(sk));
208         sk_add_node(sk, list);
209 }
210 
211 static inline void unix_remove_socket(struct sock *sk)
212 {
213         write_lock(&unix_table_lock);
214         __unix_remove_socket(sk);
215         write_unlock(&unix_table_lock);
216 }
217 
218 static inline void unix_insert_socket(struct hlist_head *list, struct sock *sk)
219 {
220         write_lock(&unix_table_lock);
221         __unix_insert_socket(list, sk);
222         write_unlock(&unix_table_lock);
223 }
224 
225 static struct sock *__unix_find_socket_byname(struct sockaddr_un *sunname,
226                                               int len, int type, unsigned hash)
227 {
228         struct sock *s;
229         struct hlist_node *node;
230 
231         sk_for_each(s, node, &unix_socket_table[hash ^ type]) {
232                 struct unix_sock *u = unix_sk(s);
233 
234                 if (u->addr->len == len &&
235                     !memcmp(u->addr->name, sunname, len))
236                         goto found;
237         }
238         s = NULL;
239 found:
240         return s;
241 }
242 
243 static inline struct sock *unix_find_socket_byname(struct sockaddr_un *sunname,
244                                                    int len, int type,
245                                                    unsigned hash)
246 {
247         struct sock *s;
248 
249         read_lock(&unix_table_lock);
250         s = __unix_find_socket_byname(sunname, len, type, hash);
251         if (s)
252                 sock_hold(s);
253         read_unlock(&unix_table_lock);
254         return s;
255 }
256 
257 static struct sock *unix_find_socket_byinode(struct inode *i)
258 {
259         struct sock *s;
260         struct hlist_node *node;
261 
262         read_lock(&unix_table_lock);
263         sk_for_each(s, node,
264                     &unix_socket_table[i->i_ino & (UNIX_HASH_SIZE - 1)]) {
265                 struct dentry *dentry = unix_sk(s)->dentry;
266 
267                 if(dentry && dentry->d_inode == i)
268                 {
269                         sock_hold(s);
270                         goto found;
271                 }
272         }
273         s = NULL;
274 found:
275         read_unlock(&unix_table_lock);
276         return s;
277 }
278 
279 static inline int unix_writable(struct sock *sk)
280 {
281         return (atomic_read(&sk->sk_wmem_alloc) << 2) <= sk->sk_sndbuf;
282 }
283 
284 static void unix_write_space(struct sock *sk)
285 {
286         read_lock(&sk->sk_callback_lock);
287         if (unix_writable(sk)) {
288                 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
289                         wake_up_interruptible(sk->sk_sleep);
290                 sk_wake_async(sk, 2, POLL_OUT);
291         }
292         read_unlock(&sk->sk_callback_lock);
293 }
294 
295 /* When dgram socket disconnects (or changes its peer), we clear its receive
296  * queue of packets arrived from previous peer. First, it allows to do
297  * flow control based only on wmem_alloc; second, sk connected to peer
298  * may receive messages only from that peer. */
299 static void unix_dgram_disconnected(struct sock *sk, struct sock *other)
300 {
301         if (skb_queue_len(&sk->sk_receive_queue)) {
302                 skb_queue_purge(&sk->sk_receive_queue);
303                 wake_up_interruptible_all(&unix_sk(sk)->peer_wait);
304 
305                 /* If one link of bidirectional dgram pipe is disconnected,
306                  * we signal error. Messages are lost. Do not make this,
307                  * when peer was not connected to us.
308                  */
309                 if (!sock_flag(other, SOCK_DEAD) && unix_peer(other) == sk) {
310                         other->sk_err = ECONNRESET;
311                         other->sk_error_report(other);
312                 }
313         }
314 }
315 
316 static void unix_sock_destructor(struct sock *sk)
317 {
318         struct unix_sock *u = unix_sk(sk);
319 
320         skb_queue_purge(&sk->sk_receive_queue);
321 
322         BUG_TRAP(!atomic_read(&sk->sk_wmem_alloc));
323         BUG_TRAP(sk_unhashed(sk));
324         BUG_TRAP(!sk->sk_socket);
325         if (!sock_flag(sk, SOCK_DEAD)) {
326                 printk("Attempt to release alive unix socket: %p\n", sk);
327                 return;
328         }
329 
330         if (u->addr)
331                 unix_release_addr(u->addr);
332 
333         atomic_dec(&unix_nr_socks);
334 #ifdef UNIX_REFCNT_DEBUG
335         printk(KERN_DEBUG "UNIX %p is destroyed, %d are still alive.\n", sk, atomic_read(&unix_nr_socks));
336 #endif
337 }
338 
339 static int unix_release_sock (struct sock *sk, int embrion)
340 {
341         struct unix_sock *u = unix_sk(sk);
342         struct dentry *dentry;
343         struct vfsmount *mnt;
344         struct sock *skpair;
345         struct sk_buff *skb;
346         int state;
347 
348         unix_remove_socket(sk);
349 
350         /* Clear state */
351         unix_state_wlock(sk);
352         sock_orphan(sk);
353         sk->sk_shutdown = SHUTDOWN_MASK;
354         dentry       = u->dentry;
355         u->dentry    = NULL;
356         mnt          = u->mnt;
357         u->mnt       = NULL;
358         state = sk->sk_state;
359         sk->sk_state = TCP_CLOSE;
360         unix_state_wunlock(sk);
361 
362         wake_up_interruptible_all(&u->peer_wait);
363 
364         skpair=unix_peer(sk);
365 
366         if (skpair!=NULL) {
367                 if (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) {
368                         unix_state_wlock(skpair);
369                         /* No more writes */
370                         skpair->sk_shutdown = SHUTDOWN_MASK;
371                         if (!skb_queue_empty(&sk->sk_receive_queue) || embrion)
372                                 skpair->sk_err = ECONNRESET;
373                         unix_state_wunlock(skpair);
374                         skpair->sk_state_change(skpair);
375                         read_lock(&skpair->sk_callback_lock);
376                         sk_wake_async(skpair,1,POLL_HUP);
377                         read_unlock(&skpair->sk_callback_lock);
378                 }
379                 sock_put(skpair); /* It may now die */
380                 unix_peer(sk) = NULL;
381         }
382 
383         /* Try to flush out this socket. Throw out buffers at least */
384 
385         while ((skb = skb_dequeue(&sk->sk_receive_queue)) != NULL) {
386                 if (state==TCP_LISTEN)
387                         unix_release_sock(skb->sk, 1);
388                 /* passed fds are erased in the kfree_skb hook        */
389                 kfree_skb(skb);
390         }
391 
392         if (dentry) {
393                 dput(dentry);
394                 mntput(mnt);
395         }
396 
397         sock_put(sk);
398 
399         /* ---- Socket is dead now and most probably destroyed ---- */
400 
401         /*
402          * Fixme: BSD difference: In BSD all sockets connected to use get
403          *        ECONNRESET and we die on the spot. In Linux we behave
404          *        like files and pipes do and wait for the last
405          *        dereference.
406          *
407          * Can't we simply set sock->err?
408          *
409          *        What the above comment does talk about? --ANK(980817)
410          */
411 
412         if (atomic_read(&unix_tot_inflight))
413                 unix_gc();              /* Garbage collect fds */       
414 
415         return 0;
416 }
417 
418 static int unix_listen(struct socket *sock, int backlog)
419 {
420         int err;
421         struct sock *sk = sock->sk;
422         struct unix_sock *u = unix_sk(sk);
423 
424         err = -EOPNOTSUPP;
425         if (sock->type!=SOCK_STREAM && sock->type!=SOCK_SEQPACKET)
426                 goto out;                       /* Only stream/seqpacket sockets accept */
427         err = -EINVAL;
428         if (!u->addr)
429                 goto out;                       /* No listens on an unbound socket */
430         unix_state_wlock(sk);
431         if (sk->sk_state != TCP_CLOSE && sk->sk_state != TCP_LISTEN)
432                 goto out_unlock;
433         if (backlog > sk->sk_max_ack_backlog)
434                 wake_up_interruptible_all(&u->peer_wait);
435         sk->sk_max_ack_backlog  = backlog;
436         sk->sk_state            = TCP_LISTEN;
437         /* set credentials so connect can copy them */
438         sk->sk_peercred.pid     = current->tgid;
439         sk->sk_peercred.uid     = current->euid;
440         sk->sk_peercred.gid     = current->egid;
441         err = 0;
442 
443 out_unlock:
444         unix_state_wunlock(sk);
445 out:
446         return err;
447 }
448 
449 static int unix_release(struct socket *);
450 static int unix_bind(struct socket *, struct sockaddr *, int);
451 static int unix_stream_connect(struct socket *, struct sockaddr *,
452                                int addr_len, int flags);
453 static int unix_socketpair(struct socket *, struct socket *);
454 static int unix_accept(struct socket *, struct socket *, int);
455 static int unix_getname(struct socket *, struct sockaddr *, int *, int);
456 static unsigned int unix_poll(struct file *, struct socket *, poll_table *);
457 static int unix_ioctl(struct socket *, unsigned int, unsigned long);
458 static int unix_shutdown(struct socket *, int);
459 static int unix_stream_sendmsg(struct kiocb *, struct socket *,
460                                struct msghdr *, size_t);
461 static int unix_stream_recvmsg(struct kiocb *, struct socket *,
462                                struct msghdr *, size_t, int);
463 static int unix_dgram_sendmsg(struct kiocb *, struct socket *,
464                               struct msghdr *, size_t);
465 static int unix_dgram_recvmsg(struct kiocb *, struct socket *,
466                               struct msghdr *, size_t, int);
467 static int unix_dgram_connect(struct socket *, struct sockaddr *,
468                               int, int);
469 static int unix_seqpacket_sendmsg(struct kiocb *, struct socket *,
470                                   struct msghdr *, size_t);
471 
472 static struct proto_ops unix_stream_ops = {
473         .family =       PF_UNIX,
474         .owner =        THIS_MODULE,
475         .release =      unix_release,
476         .bind =         unix_bind,
477         .connect =      unix_stream_connect,
478         .socketpair =   unix_socketpair,
479         .accept =       unix_accept,
480         .getname =      unix_getname,
481         .poll =         unix_poll,
482         .ioctl =        unix_ioctl,
483         .listen =       unix_listen,
484         .shutdown =     unix_shutdown,
485         .setsockopt =   sock_no_setsockopt,
486         .getsockopt =   sock_no_getsockopt,
487         .sendmsg =      unix_stream_sendmsg,
488         .recvmsg =      unix_stream_recvmsg,
489         .mmap =         sock_no_mmap,
490         .sendpage =     sock_no_sendpage,
491 };
492 
493 static struct proto_ops unix_dgram_ops = {
494         .family =       PF_UNIX,
495         .owner =        THIS_MODULE,
496         .release =      unix_release,
497         .bind =         unix_bind,
498         .connect =      unix_dgram_connect,
499         .socketpair =   unix_socketpair,
500         .accept =       sock_no_accept,
501         .getname =      unix_getname,
502         .poll =         datagram_poll,
503         .ioctl =        unix_ioctl,
504         .listen =       sock_no_listen,
505         .shutdown =     unix_shutdown,
506         .setsockopt =   sock_no_setsockopt,
507         .getsockopt =   sock_no_getsockopt,
508         .sendmsg =      unix_dgram_sendmsg,
509         .recvmsg =      unix_dgram_recvmsg,
510         .mmap =         sock_no_mmap,
511         .sendpage =     sock_no_sendpage,
512 };
513 
514 static struct proto_ops unix_seqpacket_ops = {
515         .family =       PF_UNIX,
516         .owner =        THIS_MODULE,
517         .release =      unix_release,
518         .bind =         unix_bind,
519         .connect =      unix_stream_connect,
520         .socketpair =   unix_socketpair,
521         .accept =       unix_accept,
522         .getname =      unix_getname,
523         .poll =         datagram_poll,
524         .ioctl =        unix_ioctl,
525         .listen =       unix_listen,
526         .shutdown =     unix_shutdown,
527         .setsockopt =   sock_no_setsockopt,
528         .getsockopt =   sock_no_getsockopt,
529         .sendmsg =      unix_seqpacket_sendmsg,
530         .recvmsg =      unix_dgram_recvmsg,
531         .mmap =         sock_no_mmap,
532         .sendpage =     sock_no_sendpage,
533 };
534 
535 static struct sock * unix_create1(struct socket *sock)
536 {
537         struct sock *sk = NULL;
538         struct unix_sock *u;
539 
540         if (atomic_read(&unix_nr_socks) >= 2*files_stat.max_files)
541                 goto out;
542 
543         sk = sk_alloc(PF_UNIX, GFP_KERNEL, sizeof(struct unix_sock),
544                       unix_sk_cachep);
545         if (!sk)
546                 goto out;
547 
548         atomic_inc(&unix_nr_socks);
549 
550         sock_init_data(sock,sk);
551         sk_set_owner(sk, THIS_MODULE);
552 
553         sk->sk_write_space      = unix_write_space;
554         sk->sk_max_ack_backlog  = sysctl_unix_max_dgram_qlen;
555         sk->sk_destruct         = unix_sock_destructor;
556         u         = unix_sk(sk);
557         u->dentry = NULL;
558         u->mnt    = NULL;
559         rwlock_init(&u->lock);
560         atomic_set(&u->inflight, sock ? 0 : -1);
561         init_MUTEX(&u->readsem); /* single task reading lock */
562         init_waitqueue_head(&u->peer_wait);
563         unix_insert_socket(unix_sockets_unbound, sk);
564 out:
565         return sk;
566 }
567 
568 static int unix_create(struct socket *sock, int protocol)
569 {
570         if (protocol && protocol != PF_UNIX)
571                 return -EPROTONOSUPPORT;
572 
573         sock->state = SS_UNCONNECTED;
574 
575         switch (sock->type) {
576         case SOCK_STREAM:
577                 sock->ops = &unix_stream_ops;
578                 break;
579                 /*
580                  *      Believe it or not BSD has AF_UNIX, SOCK_RAW though
581                  *      nothing uses it.
582                  */
583         case SOCK_RAW:
584                 sock->type=SOCK_DGRAM;
585         case SOCK_DGRAM:
586                 sock->ops = &unix_dgram_ops;
587                 break;
588         case SOCK_SEQPACKET:
589                 sock->ops = &unix_seqpacket_ops;
590                 break;
591         default:
592                 return -ESOCKTNOSUPPORT;
593         }
594 
595         return unix_create1(sock) ? 0 : -ENOMEM;
596 }
597 
598 static int unix_release(struct socket *sock)
599 {
600         struct sock *sk = sock->sk;
601 
602         if (!sk)
603                 return 0;
604 
605         sock->sk = NULL;
606 
607         return unix_release_sock (sk, 0);
608 }
609 
610 static int unix_autobind(struct socket *sock)
611 {
612         struct sock *sk = sock->sk;
613         struct unix_sock *u = unix_sk(sk);
614         static u32 ordernum = 1;
615         struct unix_address * addr;
616         int err;
617 
618         down(&u->readsem);
619 
620         err = 0;
621         if (u->addr)
622                 goto out;
623 
624         err = -ENOMEM;
625         addr = kmalloc(sizeof(*addr) + sizeof(short) + 16, GFP_KERNEL);
626         if (!addr)
627                 goto out;
628 
629         memset(addr, 0, sizeof(*addr) + sizeof(short) + 16);
630         addr->name->sun_family = AF_UNIX;
631         atomic_set(&addr->refcnt, 1);
632 
633 retry:
634         addr->len = sprintf(addr->name->sun_path+1, "%05x", ordernum) + 1 + sizeof(short);
635         addr->hash = unix_hash_fold(csum_partial((void*)addr->name, addr->len, 0));
636 
637         write_lock(&unix_table_lock);
638         ordernum = (ordernum+1)&0xFFFFF;
639 
640         if (__unix_find_socket_byname(addr->name, addr->len, sock->type,
641                                       addr->hash)) {
642                 write_unlock(&unix_table_lock);
643                 /* Sanity yield. It is unusual case, but yet... */
644                 if (!(ordernum&0xFF))
645                         yield();
646                 goto retry;
647         }
648         addr->hash ^= sk->sk_type;
649 
650         __unix_remove_socket(sk);
651         u->addr = addr;
652         __unix_insert_socket(&unix_socket_table[addr->hash], sk);
653         write_unlock(&unix_table_lock);
654         err = 0;
655 
656 out:    up(&u->readsem);
657         return err;
658 }
659 
660 static struct sock *unix_find_other(struct sockaddr_un *sunname, int len,
661                                     int type, unsigned hash, int *error)
662 {
663         struct sock *u;
664         struct nameidata nd;
665         int err = 0;
666         
667         if (sunname->sun_path[0]) {
668                 err = path_lookup(sunname->sun_path, LOOKUP_FOLLOW, &nd);
669                 if (err)
670                         goto fail;
671                 err = permission(nd.dentry->d_inode,MAY_WRITE, &nd);
672                 if (err)
673                         goto put_fail;
674 
675                 err = -ECONNREFUSED;
676                 if (!S_ISSOCK(nd.dentry->d_inode->i_mode))
677                         goto put_fail;
678                 u=unix_find_socket_byinode(nd.dentry->d_inode);
679                 if (!u)
680                         goto put_fail;
681 
682                 if (u->sk_type == type)
683                         touch_atime(nd.mnt, nd.dentry);
684 
685                 path_release(&nd);
686 
687                 err=-EPROTOTYPE;
688                 if (u->sk_type != type) {
689                         sock_put(u);
690                         goto fail;
691                 }
692         } else {
693                 err = -ECONNREFUSED;
694                 u=unix_find_socket_byname(sunname, len, type, hash);
695                 if (u) {
696                         struct dentry *dentry;
697                         dentry = unix_sk(u)->dentry;
698                         if (dentry)
699                                 touch_atime(unix_sk(u)->mnt, dentry);
700                 } else
701                         goto fail;
702         }
703         return u;
704 
705 put_fail:
706         path_release(&nd);
707 fail:
708         *error=err;
709         return NULL;
710 }
711 
712 
713 static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
714 {
715         struct sock *sk = sock->sk;
716         struct unix_sock *u = unix_sk(sk);
717         struct sockaddr_un *sunaddr=(struct sockaddr_un *)uaddr;
718         struct dentry * dentry = NULL;
719         struct nameidata nd;
720         int err;
721         unsigned hash;
722         struct unix_address *addr;
723         struct hlist_head *list;
724 
725         err = -EINVAL;
726         if (sunaddr->sun_family != AF_UNIX)
727                 goto out;
728 
729         if (addr_len==sizeof(short)) {
730                 err = unix_autobind(sock);
731                 goto out;
732         }
733 
734         err = unix_mkname(sunaddr, addr_len, &hash);
735         if (err < 0)
736                 goto out;
737         addr_len = err;
738 
739         down(&u->readsem);
740 
741         err = -EINVAL;
742         if (u->addr)
743                 goto out_up;
744 
745         err = -ENOMEM;
746         addr = kmalloc(sizeof(*addr)+addr_len, GFP_KERNEL);
747         if (!addr)
748                 goto out_up;
749 
750         memcpy(addr->name, sunaddr, addr_len);
751         addr->len = addr_len;
752         addr->hash = hash ^ sk->sk_type;
753         atomic_set(&addr->refcnt, 1);
754 
755         if (sunaddr->sun_path[0]) {
756                 unsigned int mode;
757                 err = 0;
758                 /*
759                  * Get the parent directory, calculate the hash for last
760                  * component.
761                  */
762                 err = path_lookup(sunaddr->sun_path, LOOKUP_PARENT, &nd);
763                 if (err)
764                         goto out_mknod_parent;
765                 /*
766                  * Yucky last component or no last component at all?
767                  * (foo/., foo/.., /////)
768                  */
769                 err = -EEXIST;
770                 if (nd.last_type != LAST_NORM)
771                         goto out_mknod;
772                 /*
773                  * Lock the directory.
774                  */
775                 down(&nd.dentry->d_inode->i_sem);
776                 /*
777                  * Do the final lookup.
778                  */
779                 dentry = lookup_hash(&nd.last, nd.dentry);
780                 err = PTR_ERR(dentry);
781                 if (IS_ERR(dentry))
782                         goto out_mknod_unlock;
783                 err = -ENOENT;
784                 /*
785                  * Special case - lookup gave negative, but... we had foo/bar/
786                  * From the vfs_mknod() POV we just have a negative dentry -
787                  * all is fine. Let's be bastards - you had / on the end, you've
788                  * been asking for (non-existent) directory. -ENOENT for you.
789                  */
790                 if (nd.last.name[nd.last.len] && !dentry->d_inode)
791                         goto out_mknod_dput;
792                 /*
793                  * All right, let's create it.
794                  */
795                 mode = S_IFSOCK |
796                        (SOCK_INODE(sock)->i_mode & ~current->fs->umask);
797                 err = vfs_mknod(nd.dentry->d_inode, dentry, mode, 0);
798                 if (err)
799                         goto out_mknod_dput;
800                 up(&nd.dentry->d_inode->i_sem);
801                 dput(nd.dentry);
802                 nd.dentry = dentry;
803 
804                 addr->hash = UNIX_HASH_SIZE;
805         }
806 
807         write_lock(&unix_table_lock);
808 
809         if (!sunaddr->sun_path[0]) {
810                 err = -EADDRINUSE;
811                 if (__unix_find_socket_byname(sunaddr, addr_len,
812                                               sk->sk_type, hash)) {
813                         unix_release_addr(addr);
814                         goto out_unlock;
815                 }
816 
817                 list = &unix_socket_table[addr->hash];
818         } else {
819                 list = &unix_socket_table[dentry->d_inode->i_ino & (UNIX_HASH_SIZE-1)];
820                 u->dentry = nd.dentry;
821                 u->mnt    = nd.mnt;
822         }
823 
824         err = 0;
825         __unix_remove_socket(sk);
826         u->addr = addr;
827         __unix_insert_socket(list, sk);
828 
829 out_unlock:
830         write_unlock(&unix_table_lock);
831 out_up:
832         up(&u->readsem);
833 out:
834         return err;
835 
836 out_mknod_dput:
837         dput(dentry);
838 out_mknod_unlock:
839         up(&nd.dentry->d_inode->i_sem);
840 out_mknod:
841         path_release(&nd);
842 out_mknod_parent:
843         if (err==-EEXIST)
844                 err=-EADDRINUSE;
845         unix_release_addr(addr);
846         goto out_up;
847 }
848 
849 static int unix_dgram_connect(struct socket *sock, struct sockaddr *addr,
850                               int alen, int flags)
851 {
852         struct sock *sk = sock->sk;
853         struct sockaddr_un *sunaddr=(struct sockaddr_un*)addr;
854         struct sock *other;
855         unsigned hash;
856         int err;
857 
858         if (addr->sa_family != AF_UNSPEC) {
859                 err = unix_mkname(sunaddr, alen, &hash);
860                 if (err < 0)
861                         goto out;
862                 alen = err;
863 
864                 if (sock->passcred && !unix_sk(sk)->addr &&
865                     (err = unix_autobind(sock)) != 0)
866                         goto out;
867 
868                 other=unix_find_other(sunaddr, alen, sock->type, hash, &err);
869                 if (!other)
870                         goto out;
871 
872                 unix_state_wlock(sk);
873 
874                 err = -EPERM;
875                 if (!unix_may_send(sk, other))
876                         goto out_unlock;
877 
878                 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
879                 if (err)
880                         goto out_unlock;
881 
882         } else {
883                 /*
884                  *      1003.1g breaking connected state with AF_UNSPEC
885                  */
886                 other = NULL;
887                 unix_state_wlock(sk);
888         }
889 
890         /*
891          * If it was connected, reconnect.
892          */
893         if (unix_peer(sk)) {
894                 struct sock *old_peer = unix_peer(sk);
895                 unix_peer(sk)=other;
896                 unix_state_wunlock(sk);
897 
898                 if (other != old_peer)
899                         unix_dgram_disconnected(sk, old_peer);
900                 sock_put(old_peer);
901         } else {
902                 unix_peer(sk)=other;
903                 unix_state_wunlock(sk);
904         }
905         return 0;
906 
907 out_unlock:
908         unix_state_wunlock(sk);
909         sock_put(other);
910 out:
911         return err;
912 }
913 
914 static long unix_wait_for_peer(struct sock *other, long timeo)
915 {
916         struct unix_sock *u = unix_sk(other);
917         int sched;
918         DEFINE_WAIT(wait);
919 
920         prepare_to_wait_exclusive(&u->peer_wait, &wait, TASK_INTERRUPTIBLE);
921 
922         sched = !sock_flag(other, SOCK_DEAD) &&
923                 !(other->sk_shutdown & RCV_SHUTDOWN) &&
924                 (skb_queue_len(&other->sk_receive_queue) >
925                  other->sk_max_ack_backlog);
926 
927         unix_state_runlock(other);
928 
929         if (sched)
930                 timeo = schedule_timeout(timeo);
931 
932         finish_wait(&u->peer_wait, &wait);
933         return timeo;
934 }
935 
936 static int unix_stream_connect(struct socket *sock, struct sockaddr *uaddr,
937                                int addr_len, int flags)
938 {
939         struct sockaddr_un *sunaddr=(struct sockaddr_un *)uaddr;
940         struct sock *sk = sock->sk;
941         struct unix_sock *u = unix_sk(sk), *newu, *otheru;
942         struct sock *newsk = NULL;
943         struct sock *other = NULL;
944         struct sk_buff *skb = NULL;
945         unsigned hash;
946         int st;
947         int err;
948         long timeo;
949 
950         err = unix_mkname(sunaddr, addr_len, &hash);
951         if (err < 0)
952                 goto out;
953         addr_len = err;
954 
955         if (sock->passcred && !u->addr && (err = unix_autobind(sock)) != 0)
956                 goto out;
957 
958         timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
959 
960         /* First of all allocate resources.
961            If we will make it after state is locked,
962            we will have to recheck all again in any case.
963          */
964 
965         err = -ENOMEM;
966 
967         /* create new sock for complete connection */
968         newsk = unix_create1(NULL);
969         if (newsk == NULL)
970                 goto out;
971 
972         /* Allocate skb for sending to listening sock */
973         skb = sock_wmalloc(newsk, 1, 0, GFP_KERNEL);
974         if (skb == NULL)
975                 goto out;
976 
977 restart:
978         /*  Find listening sock. */
979         other = unix_find_other(sunaddr, addr_len, sk->sk_type, hash, &err);
980         if (!other)
981                 goto out;
982 
983         /* Latch state of peer */
984         unix_state_rlock(other);
985 
986         /* Apparently VFS overslept socket death. Retry. */
987         if (sock_flag(other, SOCK_DEAD)) {
988                 unix_state_runlock(other);
989                 sock_put(other);
990                 goto restart;
991         }
992 
993         err = -ECONNREFUSED;
994         if (other->sk_state != TCP_LISTEN)
995                 goto out_unlock;
996 
997         if (skb_queue_len(&other->sk_receive_queue) >
998             other->sk_max_ack_backlog) {
999                 err = -EAGAIN;
1000                 if (!timeo)
1001                         goto out_unlock;
1002 
1003                 timeo = unix_wait_for_peer(other, timeo);
1004 
1005                 err = sock_intr_errno(timeo);
1006                 if (signal_pending(current))
1007                         goto out;
1008                 sock_put(other);
1009                 goto restart;
1010         }
1011 
1012         /* Latch our state.
1013 
1014            It is tricky place. We need to grab write lock and cannot
1015            drop lock on peer. It is dangerous because deadlock is
1016            possible. Connect to self case and simultaneous
1017            attempt to connect are eliminated by checking socket
1018            state. other is TCP_LISTEN, if sk is TCP_LISTEN we
1019            check this before attempt to grab lock.
1020 
1021            Well, and we have to recheck the state after socket locked.
1022          */
1023         st = sk->sk_state;
1024 
1025         switch (st) {
1026         case TCP_CLOSE:
1027                 /* This is ok... continue with connect */
1028                 break;
1029         case TCP_ESTABLISHED:
1030                 /* Socket is already connected */
1031                 err = -EISCONN;
1032                 goto out_unlock;
1033         default:
1034                 err = -EINVAL;
1035                 goto out_unlock;
1036         }
1037 
1038         unix_state_wlock(sk);
1039 
1040         if (sk->sk_state != st) {
1041                 unix_state_wunlock(sk);
1042                 unix_state_runlock(other);
1043                 sock_put(other);
1044                 goto restart;
1045         }
1046 
1047         err = security_unix_stream_connect(sock, other->sk_socket, newsk);
1048         if (err) {
1049                 unix_state_wunlock(sk);
1050                 goto out_unlock;
1051         }
1052 
1053         /* The way is open! Fastly set all the necessary fields... */
1054 
1055         sock_hold(sk);
1056         unix_peer(newsk)        = sk;
1057         newsk->sk_state         = TCP_ESTABLISHED;
1058         newsk->sk_type          = sk->sk_type;
1059         newsk->sk_peercred.pid  = current->tgid;
1060         newsk->sk_peercred.uid  = current->euid;
1061         newsk->sk_peercred.gid  = current->egid;
1062         newu = unix_sk(newsk);
1063         newsk->sk_sleep         = &newu->peer_wait;
1064         otheru = unix_sk(other);
1065 
1066         /* copy address information from listening to new sock*/
1067         if (otheru->addr) {
1068                 atomic_inc(&otheru->addr->refcnt);
1069                 newu->addr = otheru->addr;
1070         }
1071         if (otheru->dentry) {
1072                 newu->dentry    = dget(otheru->dentry);
1073                 newu->mnt       = mntget(otheru->mnt);
1074         }
1075 
1076         /* Set credentials */
1077         sk->sk_peercred = other->sk_peercred;
1078 
1079         sock_hold(newsk);
1080         unix_peer(sk)   = newsk;
1081         sock->state     = SS_CONNECTED;
1082         sk->sk_state    = TCP_ESTABLISHED;
1083 
1084         unix_state_wunlock(sk);
1085 
1086         /* take ten and and send info to listening sock */
1087         spin_lock(&other->sk_receive_queue.lock);
1088         __skb_queue_tail(&other->sk_receive_queue, skb);
1089         /* Undo artificially decreased inflight after embrion
1090          * is installed to listening socket. */
1091         atomic_inc(&newu->inflight);
1092         spin_unlock(&other->sk_receive_queue.lock);
1093         unix_state_runlock(other);
1094         other->sk_data_ready(other, 0);
1095         sock_put(other);
1096         return 0;
1097 
1098 out_unlock:
1099         if (other)
1100                 unix_state_runlock(other);
1101 
1102 out:
1103         if (skb)
1104                 kfree_skb(skb);
1105         if (newsk)
1106                 unix_release_sock(newsk, 0);
1107         if (other)
1108                 sock_put(other);
1109         return err;
1110 }
1111 
1112 static int unix_socketpair(struct socket *socka, struct socket *sockb)
1113 {
1114         struct sock *ska=socka->sk, *skb = sockb->sk;
1115 
1116         /* Join our sockets back to back */
1117         sock_hold(ska);
1118         sock_hold(skb);
1119         unix_peer(ska)=skb;
1120         unix_peer(skb)=ska;
1121         ska->sk_peercred.pid = skb->sk_peercred.pid = current->tgid;
1122         ska->sk_peercred.uid = skb->sk_peercred.uid = current->euid;
1123         ska->sk_peercred.gid = skb->sk_peercred.gid = current->egid;
1124 
1125         if (ska->sk_type != SOCK_DGRAM) {
1126                 ska->sk_state = TCP_ESTABLISHED;
1127                 skb->sk_state = TCP_ESTABLISHED;
1128                 socka->state  = SS_CONNECTED;
1129                 sockb->state  = SS_CONNECTED;
1130         }
1131         return 0;
1132 }
1133 
1134 static int unix_accept(struct socket *sock, struct socket *newsock, int flags)
1135 {
1136         struct sock *sk = sock->sk;
1137         struct sock *tsk;
1138         struct sk_buff *skb;
1139         int err;
1140 
1141         err = -EOPNOTSUPP;
1142         if (sock->type!=SOCK_STREAM && sock->type!=SOCK_SEQPACKET)
1143                 goto out;
1144 
1145         err = -EINVAL;
1146         if (sk->sk_state != TCP_LISTEN)
1147                 goto out;
1148 
1149         /* If socket state is TCP_LISTEN it cannot change (for now...),
1150          * so that no locks are necessary.
1151          */
1152 
1153         skb = skb_recv_datagram(sk, 0, flags&O_NONBLOCK, &err);
1154         if (!skb) {
1155                 /* This means receive shutdown. */
1156                 if (err == 0)
1157                         err = -EINVAL;
1158                 goto out;
1159         }
1160 
1161         tsk = skb->sk;
1162         skb_free_datagram(sk, skb);
1163         wake_up_interruptible(&unix_sk(sk)->peer_wait);
1164 
1165         /* attach accepted sock to socket */
1166         unix_state_wlock(tsk);
1167         newsock->state = SS_CONNECTED;
1168         sock_graft(tsk, newsock);
1169         unix_state_wunlock(tsk);
1170         return 0;
1171 
1172 out:
1173         return err;
1174 }
1175 
1176 
1177 static int unix_getname(struct socket *sock, struct sockaddr *uaddr, int *uaddr_len, int peer)
1178 {
1179         struct sock *sk = sock->sk;
1180         struct unix_sock *u;
1181         struct sockaddr_un *sunaddr=(struct sockaddr_un *)uaddr;
1182         int err = 0;
1183 
1184         if (peer) {
1185                 sk = unix_peer_get(sk);
1186 
1187                 err = -ENOTCONN;
1188                 if (!sk)
1189                         goto out;
1190                 err = 0;
1191         } else {
1192                 sock_hold(sk);
1193         }
1194 
1195         u = unix_sk(sk);
1196         unix_state_rlock(sk);
1197         if (!u->addr) {
1198                 sunaddr->sun_family = AF_UNIX;
1199                 sunaddr->sun_path[0] = 0;
1200                 *uaddr_len = sizeof(short);
1201         } else {
1202                 struct unix_address *addr = u->addr;
1203 
1204                 *uaddr_len = addr->len;
1205                 memcpy(sunaddr, addr->name, *uaddr_len);
1206         }
1207         unix_state_runlock(sk);
1208         sock_put(sk);
1209 out:
1210         return err;
1211 }
1212 
1213 static void unix_detach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1214 {
1215         int i;
1216 
1217         scm->fp = UNIXCB(skb).fp;
1218         skb->destructor = sock_wfree;
1219         UNIXCB(skb).fp = NULL;
1220 
1221         for (i=scm->fp->count-1; i>=0; i--)
1222                 unix_notinflight(scm->fp->fp[i]);
1223 }
1224 
1225 static void unix_destruct_fds(struct sk_buff *skb)
1226 {
1227         struct scm_cookie scm;
1228         memset(&scm, 0, sizeof(scm));
1229         unix_detach_fds(&scm, skb);
1230 
1231         /* Alas, it calls VFS */
1232         /* So fscking what? fput() had been SMP-safe since the last Summer */
1233         scm_destroy(&scm);
1234         sock_wfree(skb);
1235 }
1236 
1237 static void unix_attach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1238 {
1239         int i;
1240         for (i=scm->fp->count-1; i>=0; i--)
1241                 unix_inflight(scm->fp->fp[i]);
1242         UNIXCB(skb).fp = scm->fp;
1243         skb->destructor = unix_destruct_fds;
1244         scm->fp = NULL;
1245 }
1246 
1247 /*
1248  *      Send AF_UNIX data.
1249  */
1250 
1251 static int unix_dgram_sendmsg(struct kiocb *kiocb, struct socket *sock,
1252                               struct msghdr *msg, size_t len)
1253 {
1254         struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1255         struct sock *sk = sock->sk;
1256         struct unix_sock *u = unix_sk(sk);
1257         struct sockaddr_un *sunaddr=msg->msg_name;
1258         struct sock *other = NULL;
1259         int namelen = 0; /* fake GCC */
1260         int err;
1261         unsigned hash;
1262         struct sk_buff *skb;
1263         long timeo;
1264         struct scm_cookie tmp_scm;
1265 
1266         if (NULL == siocb->scm)
1267                 siocb->scm = &tmp_scm;
1268         err = scm_send(sock, msg, siocb->scm);
1269         if (err < 0)
1270                 return err;
1271 
1272         err = -EOPNOTSUPP;
1273         if (msg->msg_flags&MSG_OOB)
1274                 goto out;
1275 
1276         if (msg->msg_namelen) {
1277                 err = unix_mkname(sunaddr, msg->msg_namelen, &hash);
1278                 if (err < 0)
1279                         goto out;
1280                 namelen = err;
1281         } else {
1282                 sunaddr = NULL;
1283                 err = -ENOTCONN;
1284                 other = unix_peer_get(sk);
1285                 if (!other)
1286                         goto out;
1287         }
1288 
1289         if (sock->passcred && !u->addr && (err = unix_autobind(sock)) != 0)
1290                 goto out;
1291 
1292         err = -EMSGSIZE;
1293         if (len > sk->sk_sndbuf - 32)
1294                 goto out;
1295 
1296         skb = sock_alloc_send_skb(sk, len, msg->msg_flags&MSG_DONTWAIT, &err);
1297         if (skb==NULL)
1298                 goto out;
1299 
1300         memcpy(UNIXCREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1301         if (siocb->scm->fp)
1302                 unix_attach_fds(siocb->scm, skb);
1303 
1304         skb->h.raw = skb->data;
1305         err = memcpy_fromiovec(skb_put(skb,len), msg->msg_iov, len);
1306         if (err)
1307                 goto out_free;
1308 
1309         timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1310 
1311 restart:
1312         if (!other) {
1313                 err = -ECONNRESET;
1314                 if (sunaddr == NULL)
1315                         goto out_free;
1316 
1317                 other = unix_find_other(sunaddr, namelen, sk->sk_type,
1318                                         hash, &err);
1319                 if (other==NULL)
1320                         goto out_free;
1321         }
1322 
1323         unix_state_rlock(other);
1324         err = -EPERM;
1325         if (!unix_may_send(sk, other))
1326                 goto out_unlock;
1327 
1328         if (sock_flag(other, SOCK_DEAD)) {
1329                 /*
1330                  *      Check with 1003.1g - what should
1331                  *      datagram error
1332                  */
1333                 unix_state_runlock(other);
1334                 sock_put(other);
1335 
1336                 err = 0;
1337                 unix_state_wlock(sk);
1338                 if (unix_peer(sk) == other) {
1339                         unix_peer(sk)=NULL;
1340                         unix_state_wunlock(sk);
1341 
1342                         unix_dgram_disconnected(sk, other);
1343                         sock_put(other);
1344                         err = -ECONNREFUSED;
1345                 } else {
1346                         unix_state_wunlock(sk);
1347                 }
1348 
1349                 other = NULL;
1350                 if (err)
1351                         goto out_free;
1352                 goto restart;
1353         }
1354 
1355         err = -EPIPE;
1356         if (other->sk_shutdown & RCV_SHUTDOWN)
1357                 goto out_unlock;
1358 
1359         if (sk->sk_type != SOCK_SEQPACKET) {
1360                 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1361                 if (err)
1362                         goto out_unlock;
1363         }
1364 
1365         if (unix_peer(other) != sk &&
1366             (skb_queue_len(&other->sk_receive_queue) >
1367              other->sk_max_ack_backlog)) {
1368                 if (!timeo) {
1369                         err = -EAGAIN;
1370                         goto out_unlock;
1371                 }
1372 
1373                 timeo = unix_wait_for_peer(other, timeo);
1374 
1375                 err = sock_intr_errno(timeo);
1376                 if (signal_pending(current))
1377                         goto out_free;
1378 
1379                 goto restart;
1380         }
1381 
1382         skb_queue_tail(&other->sk_receive_queue, skb);
1383         unix_state_runlock(other);
1384         other->sk_data_ready(other, len);
1385         sock_put(other);
1386         scm_destroy(siocb->scm);
1387         return len;
1388 
1389 out_unlock:
1390         unix_state_runlock(other);
1391 out_free:
1392         kfree_skb(skb);
1393 out:
1394         if (other)
1395                 sock_put(other);
1396         scm_destroy(siocb->scm);
1397         return err;
1398 }
1399 
1400                 
1401 static int unix_stream_sendmsg(struct kiocb *kiocb, struct socket *sock,
1402                                struct msghdr *msg, size_t len)
1403 {
1404         struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1405         struct sock *sk = sock->sk;
1406         struct sock *other = NULL;
1407         struct sockaddr_un *sunaddr=msg->msg_name;
1408         int err,size;
1409         struct sk_buff *skb;
1410         int sent=0;
1411         struct scm_cookie tmp_scm;
1412 
1413         if (NULL == siocb->scm)
1414                 siocb->scm = &tmp_scm;
1415         err = scm_send(sock, msg, siocb->scm);
1416         if (err < 0)
1417                 return err;
1418 
1419         err = -EOPNOTSUPP;
1420         if (msg->msg_flags&MSG_OOB)
1421                 goto out_err;
1422 
1423         if (msg->msg_namelen) {
1424                 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
1425                 goto out_err;
1426         } else {
1427                 sunaddr = NULL;
1428                 err = -ENOTCONN;
1429                 other = unix_peer_get(sk);
1430                 if (!other)
1431                         goto out_err;
1432         }
1433 
1434         if (sk->sk_shutdown & SEND_SHUTDOWN)
1435                 goto pipe_err;
1436 
1437         while(sent < len)
1438         {
1439                 /*
1440                  *      Optimisation for the fact that under 0.01% of X messages typically
1441                  *      need breaking up.
1442                  */
1443 
1444                 size=len-sent;
1445 
1446                 /* Keep two messages in the pipe so it schedules better */
1447                 if (size > sk->sk_sndbuf / 2 - 64)
1448                         size = sk->sk_sndbuf / 2 - 64;
1449 
1450                 if (size > SKB_MAX_ALLOC)
1451                         size = SKB_MAX_ALLOC;
1452                         
1453                 /*
1454                  *      Grab a buffer
1455                  */
1456                  
1457                 skb=sock_alloc_send_skb(sk,size,msg->msg_flags&MSG_DONTWAIT, &err);
1458 
1459                 if (skb==NULL)
1460                         goto out_err;
1461 
1462                 /*
1463                  *      If you pass two values to the sock_alloc_send_skb
1464                  *      it tries to grab the large buffer with GFP_NOFS
1465                  *      (which can fail easily), and if it fails grab the
1466                  *      fallback size buffer which is under a page and will
1467                  *      succeed. [Alan]
1468                  */
1469                 size = min_t(int, size, skb_tailroom(skb));
1470 
1471                 memcpy(UNIXCREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1472                 if (siocb->scm->fp)
1473                         unix_attach_fds(siocb->scm, skb);
1474 
1475                 if ((err = memcpy_fromiovec(skb_put(skb,size), msg->msg_iov, size)) != 0) {
1476                         kfree_skb(skb);
1477                         goto out_err;
1478                 }
1479 
1480                 unix_state_rlock(other);
1481 
1482                 if (sock_flag(other, SOCK_DEAD) ||
1483                     (other->sk_shutdown & RCV_SHUTDOWN))
1484                         goto pipe_err_free;
1485 
1486                 skb_queue_tail(&other->sk_receive_queue, skb);
1487                 unix_state_runlock(other);
1488                 other->sk_data_ready(other, size);
1489                 sent+=size;
1490         }
1491         sock_put(other);
1492 
1493         scm_destroy(siocb->scm);
1494         siocb->scm = NULL;
1495 
1496         return sent;
1497 
1498 pipe_err_free:
1499         unix_state_runlock(other);
1500         kfree_skb(skb);
1501 pipe_err:
1502         if (sent==0 && !(msg->msg_flags&MSG_NOSIGNAL))
1503                 send_sig(SIGPIPE,current,0);
1504         err = -EPIPE;
1505 out_err:
1506         if (other)
1507                 sock_put(other);
1508         scm_destroy(siocb->scm);
1509         siocb->scm = NULL;
1510         return sent ? : err;
1511 }
1512 
1513 static int unix_seqpacket_sendmsg(struct kiocb *kiocb, struct socket *sock,
1514                                   struct msghdr *msg, size_t len)
1515 {
1516         int err;
1517         struct sock *sk = sock->sk;
1518         
1519         err = sock_error(sk);
1520         if (err)
1521                 return err;
1522 
1523         if (sk->sk_state != TCP_ESTABLISHED)
1524                 return -ENOTCONN;
1525 
1526         if (msg->msg_namelen)
1527                 msg->msg_namelen = 0;
1528 
1529         return unix_dgram_sendmsg(kiocb, sock, msg, len);
1530 }
1531                                                                                             
1532 static void unix_copy_addr(struct msghdr *msg, struct sock *sk)
1533 {
1534         struct unix_sock *u = unix_sk(sk);
1535 
1536         msg->msg_namelen = 0;
1537         if (u->addr) {
1538                 msg->msg_namelen = u->addr->len;
1539                 memcpy(msg->msg_name, u->addr->name, u->addr->len);
1540         }
1541 }
1542 
1543 static int unix_dgram_recvmsg(struct kiocb *iocb, struct socket *sock,
1544                               struct msghdr *msg, size_t size,
1545                               int flags)
1546 {
1547         struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1548         struct scm_cookie tmp_scm;
1549         struct sock *sk = sock->sk;
1550         struct unix_sock *u = unix_sk(sk);
1551         int noblock = flags & MSG_DONTWAIT;
1552         struct sk_buff *skb;
1553         int err;
1554 
1555         err = -EOPNOTSUPP;
1556         if (flags&MSG_OOB)
1557                 goto out;
1558 
1559         msg->msg_namelen = 0;
1560 
1561         down(&u->readsem);
1562 
1563         skb = skb_recv_datagram(sk, flags, noblock, &err);
1564         if (!skb)
1565                 goto out_unlock;
1566 
1567         wake_up_interruptible(&u->peer_wait);
1568 
1569         if (msg->msg_name)
1570                 unix_copy_addr(msg, skb->sk);
1571 
1572         if (size > skb->len)
1573                 size = skb->len;
1574         else if (size < skb->len)
1575                 msg->msg_flags |= MSG_TRUNC;
1576 
1577         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, size);
1578         if (err)
1579                 goto out_free;
1580 
1581         if (!siocb->scm) {
1582                 siocb->scm = &tmp_scm;
1583                 memset(&tmp_scm, 0, sizeof(tmp_scm));
1584         }
1585         siocb->scm->creds = *UNIXCREDS(skb);
1586 
1587         if (!(flags & MSG_PEEK))
1588         {
1589                 if (UNIXCB(skb).fp)
1590                         unix_detach_fds(siocb->scm, skb);
1591         }
1592         else 
1593         {
1594                 /* It is questionable: on PEEK we could:
1595                    - do not return fds - good, but too simple 8)
1596                    - return fds, and do not return them on read (old strategy,
1597                      apparently wrong)
1598                    - clone fds (I chose it for now, it is the most universal
1599                      solution)
1600                 
1601                    POSIX 1003.1g does not actually define this clearly
1602                    at all. POSIX 1003.1g doesn't define a lot of things
1603                    clearly however!                  
1604                    
1605                 */
1606                 if (UNIXCB(skb).fp)
1607                         siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1608         }
1609         err = size;
1610 
1611         scm_recv(sock, msg, siocb->scm, flags);
1612 
1613 out_free:
1614         skb_free_datagram(sk,skb);
1615 out_unlock:
1616         up(&u->readsem);
1617 out:
1618         return err;
1619 }
1620 
1621 /*
1622  *      Sleep until data has arrive. But check for races..
1623  */
1624  
1625 static long unix_stream_data_wait(struct sock * sk, long timeo)
1626 {
1627         DEFINE_WAIT(wait);
1628 
1629         unix_state_rlock(sk);
1630 
1631         for (;;) {
1632                 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1633 
1634                 if (skb_queue_len(&sk->sk_receive_queue) ||
1635                     sk->sk_err ||
1636                     (sk->sk_shutdown & RCV_SHUTDOWN) ||
1637                     signal_pending(current) ||
1638                     !timeo)
1639                         break;
1640 
1641                 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1642                 unix_state_runlock(sk);
1643                 timeo = schedule_timeout(timeo);
1644                 unix_state_rlock(sk);
1645                 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1646         }
1647 
1648         finish_wait(sk->sk_sleep, &wait);
1649         unix_state_runlock(sk);
1650         return timeo;
1651 }
1652 
1653 
1654 
1655 static int unix_stream_recvmsg(struct kiocb *iocb, struct socket *sock,
1656                                struct msghdr *msg, size_t size,
1657                                int flags)
1658 {
1659         struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1660         struct scm_cookie tmp_scm;
1661         struct sock *sk = sock->sk;
1662         struct unix_sock *u = unix_sk(sk);
1663         struct sockaddr_un *sunaddr=msg->msg_name;
1664         int copied = 0;
1665         int check_creds = 0;
1666         int target;
1667         int err = 0;
1668         long timeo;
1669 
1670         err = -EINVAL;
1671         if (sk->sk_state != TCP_ESTABLISHED)
1672                 goto out;
1673 
1674         err = -EOPNOTSUPP;
1675         if (flags&MSG_OOB)
1676                 goto out;
1677 
1678         target = sock_rcvlowat(sk, flags&MSG_WAITALL, size);
1679         timeo = sock_rcvtimeo(sk, flags&MSG_DONTWAIT);
1680 
1681         msg->msg_namelen = 0;
1682 
1683         /* Lock the socket to prevent queue disordering
1684          * while sleeps in memcpy_tomsg
1685          */
1686 
1687         if (!siocb->scm) {
1688                 siocb->scm = &tmp_scm;
1689                 memset(&tmp_scm, 0, sizeof(tmp_scm));
1690         }
1691 
1692         down(&u->readsem);
1693 
1694         do
1695         {
1696                 int chunk;
1697                 struct sk_buff *skb;
1698 
1699                 skb = skb_dequeue(&sk->sk_receive_queue);
1700                 if (skb==NULL)
1701                 {
1702                         if (copied >= target)
1703                                 break;
1704 
1705                         /*
1706                          *      POSIX 1003.1g mandates this order.
1707                          */
1708                          
1709                         if ((err = sock_error(sk)) != 0)
1710                                 break;
1711                         if (sk->sk_shutdown & RCV_SHUTDOWN)
1712                                 break;
1713                         err = -EAGAIN;
1714                         if (!timeo)
1715                                 break;
1716                         up(&u->readsem);
1717 
1718                         timeo = unix_stream_data_wait(sk, timeo);
1719 
1720                         if (signal_pending(current)) {
1721                                 err = sock_intr_errno(timeo);
1722                                 goto out;
1723                         }
1724                         down(&u->readsem);
1725                         continue;
1726                 }
1727 
1728                 if (check_creds) {
1729                         /* Never glue messages from different writers */
1730                         if (memcmp(UNIXCREDS(skb), &siocb->scm->creds, sizeof(siocb->scm->creds)) != 0) {
1731                                 skb_queue_head(&sk->sk_receive_queue, skb);
1732                                 break;
1733                         }
1734                 } else {
1735                         /* Copy credentials */
1736                         siocb->scm->creds = *UNIXCREDS(skb);
1737                         check_creds = 1;
1738                 }
1739 
1740                 /* Copy address just once */
1741                 if (sunaddr)
1742                 {
1743                         unix_copy_addr(msg, skb->sk);
1744                         sunaddr = NULL;
1745                 }
1746 
1747                 chunk = min_t(unsigned int, skb->len, size);
1748                 if (memcpy_toiovec(msg->msg_iov, skb->data, chunk)) {
1749                         skb_queue_head(&sk->sk_receive_queue, skb);
1750                         if (copied == 0)
1751                                 copied = -EFAULT;
1752                         break;
1753                 }
1754                 copied += chunk;
1755                 size -= chunk;
1756 
1757                 /* Mark read part of skb as used */
1758                 if (!(flags & MSG_PEEK))
1759                 {
1760                         skb_pull(skb, chunk);
1761 
1762                         if (UNIXCB(skb).fp)
1763                                 unix_detach_fds(siocb->scm, skb);
1764 
1765                         /* put the skb back if we didn't use it up.. */
1766                         if (skb->len)
1767                         {
1768                                 skb_queue_head(&sk->sk_receive_queue, skb);
1769                                 break;
1770                         }
1771 
1772                         kfree_skb(skb);
1773 
1774                         if (siocb->scm->fp)
1775                                 break;
1776                 }
1777                 else
1778                 {
1779                         /* It is questionable, see note in unix_dgram_recvmsg.
1780                          */
1781                         if (UNIXCB(skb).fp)
1782                                 siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1783 
1784                         /* put message back and return */
1785                         skb_queue_head(&sk->sk_receive_queue, skb);
1786                         break;
1787                 }
1788         } while (size);
1789 
1790         up(&u->readsem);
1791         scm_recv(sock, msg, siocb->scm, flags);
1792 out:
1793         return copied ? : err;
1794 }
1795 
1796 static int unix_shutdown(struct socket *sock, int mode)
1797 {
1798         struct sock *sk = sock->sk;
1799         struct sock *other;
1800 
1801         mode = (mode+1)&(RCV_SHUTDOWN|SEND_SHUTDOWN);
1802 
1803         if (mode) {
1804                 unix_state_wlock(sk);
1805                 sk->sk_shutdown |= mode;
1806                 other=unix_peer(sk);
1807                 if (other)
1808                         sock_hold(other);
1809                 unix_state_wunlock(sk);
1810                 sk->sk_state_change(sk);
1811 
1812                 if (other &&
1813                         (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET)) {
1814 
1815                         int peer_mode = 0;
1816 
1817                         if (mode&RCV_SHUTDOWN)
1818                                 peer_mode |= SEND_SHUTDOWN;
1819                         if (mode&SEND_SHUTDOWN)
1820                                 peer_mode |= RCV_SHUTDOWN;
1821                         unix_state_wlock(other);
1822                         other->sk_shutdown |= peer_mode;
1823                         unix_state_wunlock(other);
1824                         other->sk_state_change(other);
1825                         read_lock(&other->sk_callback_lock);
1826                         if (peer_mode == SHUTDOWN_MASK)
1827                                 sk_wake_async(other,1,POLL_HUP);
1828                         else if (peer_mode & RCV_SHUTDOWN)
1829                                 sk_wake_async(other,1,POLL_IN);
1830                         read_unlock(&other->sk_callback_lock);
1831                 }
1832                 if (other)
1833                         sock_put(other);
1834         }
1835         return 0;
1836 }
1837 
1838 static int unix_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1839 {
1840         struct sock *sk = sock->sk;
1841         long amount=0;
1842         int err;
1843 
1844         switch(cmd)
1845         {
1846                 case SIOCOUTQ:
1847                         amount = atomic_read(&sk->sk_wmem_alloc);
1848                         err = put_user(amount, (int __user *)arg);
1849                         break;
1850                 case SIOCINQ:
1851                 {
1852                         struct sk_buff *skb;
1853 
1854                         if (sk->sk_state == TCP_LISTEN) {
1855                                 err = -EINVAL;
1856                                 break;
1857                         }
1858 
1859                         spin_lock(&sk->sk_receive_queue.lock);
1860                         if (sk->sk_type == SOCK_STREAM ||
1861                             sk->sk_type == SOCK_SEQPACKET) {
1862                                 skb_queue_walk(&sk->sk_receive_queue, skb)
1863                                         amount += skb->len;
1864                         } else {
1865                                 skb = skb_peek(&sk->sk_receive_queue);
1866                                 if (skb)
1867                                         amount=skb->len;
1868                         }
1869                         spin_unlock(&sk->sk_receive_queue.lock);
1870                         err = put_user(amount, (int __user *)arg);
1871                         break;
1872                 }
1873 
1874                 default:
1875                         err = dev_ioctl(cmd, (void __user *)arg);
1876                         break;
1877         }
1878         return err;
1879 }
1880 
1881 static unsigned int unix_poll(struct file * file, struct socket *sock, poll_table *wait)
1882 {
1883         struct sock *sk = sock->sk;
1884         unsigned int mask;
1885 
1886         poll_wait(file, sk->sk_sleep, wait);
1887         mask = 0;
1888 
1889         /* exceptional events? */
1890         if (sk->sk_err)
1891                 mask |= POLLERR;
1892         if (sk->sk_shutdown == SHUTDOWN_MASK)
1893                 mask |= POLLHUP;
1894 
1895         /* readable? */
1896         if (!skb_queue_empty(&sk->sk_receive_queue) ||
1897             (sk->sk_shutdown & RCV_SHUTDOWN))
1898                 mask |= POLLIN | POLLRDNORM;
1899 
1900         /* Connection-based need to check for termination and startup */
1901         if ((sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) && sk->sk_state == TCP_CLOSE)
1902                 mask |= POLLHUP;
1903 
1904         /*
1905          * we set writable also when the other side has shut down the
1906          * connection. This prevents stuck sockets.
1907          */
1908         if (unix_writable(sk))
1909                 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
1910 
1911         return mask;
1912 }
1913 
1914 
1915 #ifdef CONFIG_PROC_FS
1916 static struct sock *unix_seq_idx(int *iter, loff_t pos)
1917 {
1918         loff_t off = 0;
1919         struct sock *s;
1920 
1921         for (s = first_unix_socket(iter); s; s = next_unix_socket(iter, s)) {
1922                 if (off == pos) 
1923                         return s;
1924                 ++off;
1925         }
1926         return NULL;
1927 }
1928 
1929 
1930 static void *unix_seq_start(struct seq_file *seq, loff_t *pos)
1931 {
1932         read_lock(&unix_table_lock);
1933         return *pos ? unix_seq_idx(seq->private, *pos - 1) : ((void *) 1);
1934 }
1935 
1936 static void *unix_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1937 {
1938         ++*pos;
1939 
1940         if (v == (void *)1) 
1941                 return first_unix_socket(seq->private);
1942         return next_unix_socket(seq->private, v);
1943 }
1944 
1945 static void unix_seq_stop(struct seq_file *seq, void *v)
1946 {
1947         read_unlock(&unix_table_lock);
1948 }
1949 
1950 static int unix_seq_show(struct seq_file *seq, void *v)
1951 {
1952         
1953         if (v == (void *)1)
1954                 seq_puts(seq, "Num       RefCount Protocol Flags    Type St "
1955                          "Inode Path\n");
1956         else {
1957                 struct sock *s = v;
1958                 struct unix_sock *u = unix_sk(s);
1959                 unix_state_rlock(s);
1960 
1961                 seq_printf(seq, "%p: %08X %08X %08X %04X %02X %5lu",
1962                         s,
1963                         atomic_read(&s->sk_refcnt),
1964                         0,
1965                         s->sk_state == TCP_LISTEN ? __SO_ACCEPTCON : 0,
1966                         s->sk_type,
1967                         s->sk_socket ?
1968                         (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTED : SS_UNCONNECTED) :
1969                         (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTING : SS_DISCONNECTING),
1970                         sock_i_ino(s));
1971 
1972                 if (u->addr) {
1973                         int i, len;
1974                         seq_putc(seq, ' ');
1975 
1976                         i = 0;
1977                         len = u->addr->len - sizeof(short);
1978                         if (!UNIX_ABSTRACT(s))
1979                                 len--;
1980                         else {
1981                                 seq_putc(seq, '@');
1982                                 i++;
1983                         }
1984                         for ( ; i < len; i++)
1985                                 seq_putc(seq, u->addr->name->sun_path[i]);
1986                 }
1987                 unix_state_runlock(s);
1988                 seq_putc(seq, '\n');
1989         }
1990 
1991         return 0;
1992 }
1993 
1994 static struct seq_operations unix_seq_ops = {
1995         .start  = unix_seq_start,
1996         .next   = unix_seq_next,
1997         .stop   = unix_seq_stop,
1998         .show   = unix_seq_show,
1999 };
2000 
2001 
2002 static int unix_seq_open(struct inode *inode, struct file *file)
2003 {
2004         struct seq_file *seq;
2005         int rc = -ENOMEM;
2006         int *iter = kmalloc(sizeof(int), GFP_KERNEL);
2007 
2008         if (!iter)
2009                 goto out;
2010 
2011         rc = seq_open(file, &unix_seq_ops);
2012         if (rc)
2013                 goto out_kfree;
2014 
2015         seq          = file->private_data;
2016         seq->private = iter;
2017         *iter = 0;
2018 out:
2019         return rc;
2020 out_kfree:
2021         kfree(iter);
2022         goto out;
2023 }
2024 
2025 static struct file_operations unix_seq_fops = {
2026         .owner          = THIS_MODULE,
2027         .open           = unix_seq_open,
2028         .read           = seq_read,
2029         .llseek         = seq_lseek,
2030         .release        = seq_release_private,
2031 };
2032 
2033 #endif
2034 
2035 static struct net_proto_family unix_family_ops = {
2036         .family = PF_UNIX,
2037         .create = unix_create,
2038         .owner  = THIS_MODULE,
2039 };
2040 
2041 #ifdef CONFIG_SYSCTL
2042 extern void unix_sysctl_register(void);
2043 extern void unix_sysctl_unregister(void);
2044 #else
2045 static inline void unix_sysctl_register(void) {}
2046 static inline void unix_sysctl_unregister(void) {}
2047 #endif
2048 
2049 static int __init af_unix_init(void)
2050 {
2051         struct sk_buff *dummy_skb;
2052 
2053         if (sizeof(struct unix_skb_parms) > sizeof(dummy_skb->cb)) {
2054                 printk(KERN_CRIT "%s: panic\n", __FUNCTION__);
2055                 return -1;
2056         }
2057         /* allocate our sock slab cache */
2058         unix_sk_cachep = kmem_cache_create("unix_sock",
2059                                            sizeof(struct unix_sock), 0,
2060                                            SLAB_HWCACHE_ALIGN, NULL, NULL);
2061         if (!unix_sk_cachep)
2062                 printk(KERN_CRIT
2063                         "af_unix_init: Cannot create unix_sock SLAB cache!\n");
2064 
2065         sock_register(&unix_family_ops);
2066 #ifdef CONFIG_PROC_FS
2067         proc_net_fops_create("unix", 0, &unix_seq_fops);
2068 #endif
2069         unix_sysctl_register();
2070         return 0;
2071 }
2072 
2073 static void __exit af_unix_exit(void)
2074 {
2075         sock_unregister(PF_UNIX);
2076         unix_sysctl_unregister();
2077         proc_net_remove("unix");
2078         kmem_cache_destroy(unix_sk_cachep);
2079 }
2080 
2081 module_init(af_unix_init);
2082 module_exit(af_unix_exit);
2083 
2084 MODULE_LICENSE("GPL");
2085 MODULE_ALIAS_NETPROTO(PF_UNIX);
2086 
  This page was automatically generated by the LXR engine.