net/unix/af_unix.c

   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  * Fixes:
  12  *              Linus Torvalds  :       Assorted bug cures.
  13  *              Niibe Yutaka    :       async I/O support.
  14  *              Carsten Paeth   :       PF_UNIX check, address fixes.
  15  *              Alan Cox        :       Limit size of allocated blocks.
  16  *              Alan Cox        :       Fixed the stupid socketpair bug.
  17  *              Alan Cox        :       BSD compatibility fine tuning.
  18  *              Alan Cox        :       Fixed a bug in connect when interrupted.
  19  *              Alan Cox        :       Sorted out a proper draft version of
  20  *                                      file descriptor passing hacked up from
  21  *                                      Mike Shaver's work.
  22  *              Marty Leisner   :       Fixes to fd passing
  23  *              Nick Nevin      :       recvmsg bugfix.
  24  *              Alan Cox        :       Started proper garbage collector
  25  *              Heiko EiBfeldt  :       Missing verify_area check
  26  *              Alan Cox        :       Started POSIXisms
  27  *              Andreas Schwab  :       Replace inode by dentry for proper
  28  *                                      reference counting
  29  *              Kirk Petersen   :       Made this a module
  30  *          Christoph Rohland   :       Elegant non-blocking accept/connect algorithm.
  31  *                                      Lots of bug fixes.
  32  *           Alexey Kuznetosv   :       Repaired (I hope) bugs introduces
  33  *                                      by above two patches.
  34  *           Andrea Arcangeli   :       If possible we block in connect(2)
  35  *                                      if the max backlog of the listen socket
  36  *                                      is been reached. This won't break
  37  *                                      old apps and it will avoid huge amount
  38  *                                      of socks hashed (this for unix_gc()
  39  *                                      performances reasons).
  40  *                                      Security fix that limits the max
  41  *                                      number of socks to 2*max_files and
  42  *                                      the number of skb queueable in the
  43  *                                      dgram receiver.
  44  *              Artur Skawina   :       Hash function optimizations
  45  *           Alexey Kuznetsov   :       Full scale SMP. Lot of bugs are introduced 8)
  46  *            Malcolm Beattie   :       Set peercred for socketpair
  47  *           Michal Ostrowski   :       Module initialization cleanup.
  48  *           Arnaldo C. Melo    :       Remove MOD_{INC,DEC}_USE_COUNT,
  49  *                                      the core infrastructure is doing that
  50  *                                      for all net proto families now (2.5.69+)
  51  *
  52  *
  53  * Known differences from reference BSD that was tested:
  54  *
  55  *      [TO FIX]
  56  *      ECONNREFUSED is not returned from one end of a connected() socket to the
  57  *              other the moment one end closes.
  58  *      fstat() doesn't return st_dev=0, and give the blksize as high water mark
  59  *              and a fake inode identifier (nor the BSD first socket fstat twice bug).
  60  *      [NOT TO FIX]
  61  *      accept() returns a path name even if the connecting socket has closed
  62  *              in the meantime (BSD loses the path and gives up).
  63  *      accept() returns 0 length path for an unbound connector. BSD returns 16
  64  *              and a null first byte in the path (but not for gethost/peername - BSD bug ??)
  65  *      socketpair(...SOCK_RAW..) doesn't panic the kernel.
  66  *      BSD af_unix apparently has connect forgetting to block properly.
  67  *              (need to check this with the POSIX spec in detail)
  68  *
  69  * Differences from 2.0.0-11-... (ANK)
  70  *      Bug fixes and improvements.
  71  *              - client shutdown killed server socket.
  72  *              - removed all useless cli/sti pairs.
  73  *
  74  *      Semantic changes/extensions.
  75  *              - generic control message passing.
  76  *              - SCM_CREDENTIALS control message.
  77  *              - "Abstract" (not FS based) socket bindings.
  78  *                Abstract names are sequences of bytes (not zero terminated)
  79  *                started by 0, so that this name space does not intersect
  80  *                with BSD names.
  81  */
  82
  83 #include <linux/module.h>
  84 #include <linux/kernel.h>
  85 #include <linux/signal.h>
  86 #include <linux/sched.h>
  87 #include <linux/errno.h>
  88 #include <linux/string.h>
  89 #include <linux/stat.h>
  90 #include <linux/dcache.h>
  91 #include <linux/namei.h>
  92 #include <linux/socket.h>
  93 #include <linux/un.h>
  94 #include <linux/fcntl.h>
  95 #include <linux/termios.h>
  96 #include <linux/sockios.h>
  97 #include <linux/net.h>
  98 #include <linux/in.h>
  99 #include <linux/fs.h>
 100 #include <linux/slab.h>
 101 #include <asm/uaccess.h>
 102 #include <linux/skbuff.h>
 103 #include <linux/netdevice.h>
 104 #include <net/net_namespace.h>
 105 #include <net/sock.h>
 106 #include <net/tcp_states.h>
 107 #include <net/af_unix.h>
 108 #include <linux/proc_fs.h>
 109 #include <linux/seq_file.h>
 110 #include <net/scm.h>
 111 #include <linux/init.h>
 112 #include <linux/poll.h>
 113 #include <linux/rtnetlink.h>
 114 #include <linux/mount.h>
 115 #include <net/checksum.h>
 116 #include <linux/security.h>
 117
 118 static struct hlist_head unix_socket_table[UNIX_HASH_SIZE + 1];
 119 static DEFINE_SPINLOCK(unix_table_lock);
 120 static atomic_t unix_nr_socks = ATOMIC_INIT(0);
 121
 122 #define unix_sockets_unbound    (&unix_socket_table[UNIX_HASH_SIZE])
 123
 124 #define UNIX_ABSTRACT(sk)       (unix_sk(sk)->addr->hash != UNIX_HASH_SIZE)
 125
 126 #ifdef CONFIG_SECURITY_NETWORK
 127 static void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb)
 128 {
 129         memcpy(UNIXSID(skb), &scm->secid, sizeof(u32));
 130 }
 131
 132 static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb)
 133 {
 134         scm->secid = *UNIXSID(skb);
 135 }
 136 #else
 137 static inline void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb)
 138 { }
 139
 140 static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb)
 141 { }
 142 #endif /* CONFIG_SECURITY_NETWORK */
 143
 144 /*
 145  *  SMP locking strategy:
 146  *    hash table is protected with spinlock unix_table_lock
 147  *    each socket state is protected by separate rwlock.
 148  */
 149
 150 static inline unsigned unix_hash_fold(__wsum n)
 151 {
 152         unsigned hash = (__force unsigned)n;
 153         hash ^= hash>>16;
 154         hash ^= hash>>8;
 155         return hash&(UNIX_HASH_SIZE-1);
 156 }
 157
 158 #define unix_peer(sk) (unix_sk(sk)->peer)
 159
 160 static inline int unix_our_peer(struct sock *sk, struct sock *osk)
 161 {
 162         return unix_peer(osk) == sk;
 163 }
 164
 165 static inline int unix_may_send(struct sock *sk, struct sock *osk)
 166 {
 167         return (unix_peer(osk) == NULL || unix_our_peer(sk, osk));
 168 }
 169
 170 static inline int unix_recvq_full(struct sock const *sk)
 171 {
 172         return skb_queue_len(&sk->sk_receive_queue) > sk->sk_max_ack_backlog;
 173 }
 174
 175 static struct sock *unix_peer_get(struct sock *s)
 176 {
 177         struct sock *peer;
 178
 179         unix_state_lock(s);
 180         peer = unix_peer(s);
 181         if (peer)
 182                 sock_hold(peer);
 183         unix_state_unlock(s);
 184         return peer;
 185 }
 186
 187 static inline void unix_release_addr(struct unix_address *addr)
 188 {
 189         if (atomic_dec_and_test(&addr->refcnt))
 190                 kfree(addr);
 191 }
 192
 193 /*
 194  *      Check unix socket name:
 195  *              - should be not zero length.
 196  *              - if started by not zero, should be NULL terminated (FS object)
 197  *              - if started by zero, it is abstract name.
 198  */
 199
 200 static int unix_mkname(struct sockaddr_un * sunaddr, int len, unsigned *hashp)
 201 {
 202         if (len <= sizeof(short) || len > sizeof(*sunaddr))
 203                 return -EINVAL;
 204         if (!sunaddr || sunaddr->sun_family != AF_UNIX)
 205                 return -EINVAL;
 206         if (sunaddr->sun_path[0]) {
 207                 /*
 208                  * This may look like an off by one error but it is a bit more
 209                  * subtle. 108 is the longest valid AF_UNIX path for a binding.
 210                  * sun_path[108] doesnt as such exist.  However in kernel space
 211                  * we are guaranteed that it is a valid memory location in our
 212                  * kernel address buffer.
 213                  */
 214                 ((char *)sunaddr)[len]=0;
 215                 len = strlen(sunaddr->sun_path)+1+sizeof(short);
 216                 return len;
 217         }
 218
 219         *hashp = unix_hash_fold(csum_partial((char*)sunaddr, len, 0));
 220         return len;
 221 }
 222
 223 static void __unix_remove_socket(struct sock *sk)
 224 {
 225         sk_del_node_init(sk);
 226 }
 227
 228 static void __unix_insert_socket(struct hlist_head *list, struct sock *sk)
 229 {
 230         BUG_TRAP(sk_unhashed(sk));
 231         sk_add_node(sk, list);
 232 }
 233
 234 static inline void unix_remove_socket(struct sock *sk)
 235 {
 236         spin_lock(&unix_table_lock);
 237         __unix_remove_socket(sk);
 238         spin_unlock(&unix_table_lock);
 239 }
 240
 241 static inline void unix_insert_socket(struct hlist_head *list, struct sock *sk)
 242 {
 243         spin_lock(&unix_table_lock);
 244         __unix_insert_socket(list, sk);
 245         spin_unlock(&unix_table_lock);
 246 }
 247
 248 static struct sock *__unix_find_socket_byname(struct net *net,
 249                                               struct sockaddr_un *sunname,
 250                                               int len, int type, unsigned hash)
 251 {
 252         struct sock *s;
 253         struct hlist_node *node;
 254
 255         sk_for_each(s, node, &unix_socket_table[hash ^ type]) {
 256                 struct unix_sock *u = unix_sk(s);
 257
 258                 if (!net_eq(sock_net(s), net))
 259                         continue;
 260
 261                 if (u->addr->len == len &&
 262                     !memcmp(u->addr->name, sunname, len))
 263                         goto found;
 264         }
 265         s = NULL;
 266 found:
 267         return s;
 268 }
 269
 270 static inline struct sock *unix_find_socket_byname(struct net *net,
 271                                                    struct sockaddr_un *sunname,
 272                                                    int len, int type,
 273                                                    unsigned hash)
 274 {
 275         struct sock *s;
 276
 277         spin_lock(&unix_table_lock);
 278         s = __unix_find_socket_byname(net, sunname, len, type, hash);
 279         if (s)
 280                 sock_hold(s);
 281         spin_unlock(&unix_table_lock);
 282         return s;
 283 }
 284
 285 static struct sock *unix_find_socket_byinode(struct net *net, struct inode *i)
 286 {
 287         struct sock *s;
 288         struct hlist_node *node;
 289
 290         spin_lock(&unix_table_lock);
 291         sk_for_each(s, node,
 292                     &unix_socket_table[i->i_ino & (UNIX_HASH_SIZE - 1)]) {
 293                 struct dentry *dentry = unix_sk(s)->dentry;
 294
 295                 if (!net_eq(sock_net(s), net))
 296                         continue;
 297
 298                 if(dentry && dentry->d_inode == i)
 299                 {
 300                         sock_hold(s);
 301                         goto found;
 302                 }
 303         }
 304         s = NULL;
 305 found:
 306         spin_unlock(&unix_table_lock);
 307         return s;
 308 }
 309
 310 static inline int unix_writable(struct sock *sk)
 311 {
 312         return (atomic_read(&sk->sk_wmem_alloc) << 2) <= sk->sk_sndbuf;
 313 }
 314
 315 static void unix_write_space(struct sock *sk)
 316 {
 317         read_lock(&sk->sk_callback_lock);
 318         if (unix_writable(sk)) {
 319                 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
 320                         wake_up_interruptible_sync(sk->sk_sleep);
 321                 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
 322         }
 323         read_unlock(&sk->sk_callback_lock);
 324 }
 325
 326 /* When dgram socket disconnects (or changes its peer), we clear its receive
 327  * queue of packets arrived from previous peer. First, it allows to do
 328  * flow control based only on wmem_alloc; second, sk connected to peer
 329  * may receive messages only from that peer. */
 330 static void unix_dgram_disconnected(struct sock *sk, struct sock *other)
 331 {
 332         if (!skb_queue_empty(&sk->sk_receive_queue)) {
 333                 skb_queue_purge(&sk->sk_receive_queue);
 334                 wake_up_interruptible_all(&unix_sk(sk)->peer_wait);
 335
 336                 /* If one link of bidirectional dgram pipe is disconnected,
 337                  * we signal error. Messages are lost. Do not make this,
 338                  * when peer was not connected to us.
 339                  */
 340                 if (!sock_flag(other, SOCK_DEAD) && unix_peer(other) == sk) {
 341                         other->sk_err = ECONNRESET;
 342                         other->sk_error_report(other);
 343                 }
 344         }
 345 }
 346
 347 static void unix_sock_destructor(struct sock *sk)
 348 {
 349         struct unix_sock *u = unix_sk(sk);
 350
 351         skb_queue_purge(&sk->sk_receive_queue);
 352
 353         BUG_TRAP(!atomic_read(&sk->sk_wmem_alloc));
 354         BUG_TRAP(sk_unhashed(sk));
 355         BUG_TRAP(!sk->sk_socket);
 356         if (!sock_flag(sk, SOCK_DEAD)) {
 357                 printk("Attempt to release alive unix socket: %p\n", sk);
 358                 return;
 359         }
 360
 361         if (u->addr)
 362                 unix_release_addr(u->addr);
 363
 364         atomic_dec(&unix_nr_socks);
 365 #ifdef UNIX_REFCNT_DEBUG
 366         printk(KERN_DEBUG "UNIX %p is destroyed, %d are still alive.\n", sk, atomic_read(&unix_nr_socks));
 367 #endif
 368 }
 369
 370 static int unix_release_sock (struct sock *sk, int embrion)
 371 {
 372         struct unix_sock *u = unix_sk(sk);
 373         struct dentry *dentry;
 374         struct vfsmount *mnt;
 375         struct sock *skpair;
 376         struct sk_buff *skb;
 377         int state;
 378
 379         unix_remove_socket(sk);
 380
 381         /* Clear state */
 382         unix_state_lock(sk);
 383         sock_orphan(sk);
 384         sk->sk_shutdown = SHUTDOWN_MASK;
 385         dentry       = u->dentry;
 386         u->dentry    = NULL;
 387         mnt          = u->mnt;
 388         u->mnt       = NULL;
 389         state = sk->sk_state;
 390         sk->sk_state = TCP_CLOSE;
 391         unix_state_unlock(sk);
 392
 393         wake_up_interruptible_all(&u->peer_wait);
 394
 395         skpair=unix_peer(sk);
 396
 397         if (skpair!=NULL) {
 398                 if (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) {
 399                         unix_state_lock(skpair);
 400                         /* No more writes */
 401                         skpair->sk_shutdown = SHUTDOWN_MASK;
 402                         if (!skb_queue_empty(&sk->sk_receive_queue) || embrion)
 403                                 skpair->sk_err = ECONNRESET;
 404                         unix_state_unlock(skpair);
 405                         skpair->sk_state_change(skpair);
 406                         read_lock(&skpair->sk_callback_lock);
 407                         sk_wake_async(skpair, SOCK_WAKE_WAITD, POLL_HUP);
 408                         read_unlock(&skpair->sk_callback_lock);
 409                 }
 410                 sock_put(skpair); /* It may now die */
 411                 unix_peer(sk) = NULL;
 412         }
 413
 414         /* Try to flush out this socket. Throw out buffers at least */
 415
 416         while ((skb = skb_dequeue(&sk->sk_receive_queue)) != NULL) {
 417                 if (state==TCP_LISTEN)
 418                         unix_release_sock(skb->sk, 1);
 419                 /* passed fds are erased in the kfree_skb hook        */
 420                 kfree_skb(skb);
 421         }
 422
 423         if (dentry) {
 424                 dput(dentry);
 425                 mntput(mnt);
 426         }
 427
 428         sock_put(sk);
 429
 430         /* ---- Socket is dead now and most probably destroyed ---- */
 431
 432         /*
 433          * Fixme: BSD difference: In BSD all sockets connected to use get
 434          *        ECONNRESET and we die on the spot. In Linux we behave
 435          *        like files and pipes do and wait for the last
 436          *        dereference.
 437          *
 438          * Can't we simply set sock->err?
 439          *
 440          *        What the above comment does talk about? --ANK(980817)
 441          */
 442
 443         if (unix_tot_inflight)
 444                 unix_gc();              /* Garbage collect fds */
 445
 446         return 0;
 447 }
 448
 449 static int unix_listen(struct socket *sock, int backlog)
 450 {
 451         int err;
 452         struct sock *sk = sock->sk;
 453         struct unix_sock *u = unix_sk(sk);
 454
 455         err = -EOPNOTSUPP;
 456         if (sock->type!=SOCK_STREAM && sock->type!=SOCK_SEQPACKET)
 457                 goto out;                       /* Only stream/seqpacket sockets accept */
 458         err = -EINVAL;
 459         if (!u->addr)
 460                 goto out;                       /* No listens on an unbound socket */
 461         unix_state_lock(sk);
 462         if (sk->sk_state != TCP_CLOSE && sk->sk_state != TCP_LISTEN)
 463                 goto out_unlock;
 464         if (backlog > sk->sk_max_ack_backlog)
 465                 wake_up_interruptible_all(&u->peer_wait);
 466         sk->sk_max_ack_backlog  = backlog;
 467         sk->sk_state            = TCP_LISTEN;
 468         /* set credentials so connect can copy them */
 469         sk->sk_peercred.pid     = task_tgid_vnr(current);
 470         sk->sk_peercred.uid     = current->euid;
 471         sk->sk_peercred.gid     = current->egid;
 472         err = 0;
 473
 474 out_unlock:
 475         unix_state_unlock(sk);
 476 out:
 477         return err;
 478 }
 479
 480 static int unix_release(struct socket *);
 481 static int unix_bind(struct socket *, struct sockaddr *, int);
 482 static int unix_stream_connect(struct socket *, struct sockaddr *,
 483                                int addr_len, int flags);
 484 static int unix_socketpair(struct socket *, struct socket *);
 485 static int unix_accept(struct socket *, struct socket *, int);
 486 static int unix_getname(struct socket *, struct sockaddr *, int *, int);
 487 static unsigned int unix_poll(struct file *, struct socket *, poll_table *);
 488 static unsigned int unix_dgram_poll(struct file *, struct socket *,
 489                                     poll_table *);
 490 static int unix_ioctl(struct socket *, unsigned int, unsigned long);
 491 static int unix_shutdown(struct socket *, int);
 492 static int unix_stream_sendmsg(struct kiocb *, struct socket *,
 493                                struct msghdr *, size_t);
 494 static int unix_stream_recvmsg(struct kiocb *, struct socket *,
 495                                struct msghdr *, size_t, int);
 496 static int unix_dgram_sendmsg(struct kiocb *, struct socket *,
 497                               struct msghdr *, size_t);
 498 static int unix_dgram_recvmsg(struct kiocb *, struct socket *,
 499                               struct msghdr *, size_t, int);
 500 static int unix_dgram_connect(struct socket *, struct sockaddr *,
 501                               int, int);
 502 static int unix_seqpacket_sendmsg(struct kiocb *, struct socket *,
 503                                   struct msghdr *, size_t);
 504
 505 static const struct proto_ops unix_stream_ops = {
 506         .family =       PF_UNIX,
 507         .owner =        THIS_MODULE,
 508         .release =      unix_release,
 509         .bind =         unix_bind,
 510         .connect =      unix_stream_connect,
 511         .socketpair =   unix_socketpair,
 512         .accept =       unix_accept,
 513         .getname =      unix_getname,
 514         .poll =         unix_poll,
 515         .ioctl =        unix_ioctl,
 516         .listen =       unix_listen,
 517         .shutdown =     unix_shutdown,
 518         .setsockopt =   sock_no_setsockopt,
 519         .getsockopt =   sock_no_getsockopt,
 520         .sendmsg =      unix_stream_sendmsg,
 521         .recvmsg =      unix_stream_recvmsg,
 522         .mmap =         sock_no_mmap,
 523         .sendpage =     sock_no_sendpage,
 524 };
 525
 526 static const struct proto_ops unix_dgram_ops = {
 527         .family =       PF_UNIX,
 528         .owner =        THIS_MODULE,
 529         .release =      unix_release,
 530         .bind =         unix_bind,
 531         .connect =      unix_dgram_connect,
 532         .socketpair =   unix_socketpair,
 533         .accept =       sock_no_accept,
 534         .getname =      unix_getname,
 535         .poll =         unix_dgram_poll,
 536         .ioctl =        unix_ioctl,
 537         .listen =       sock_no_listen,
 538         .shutdown =     unix_shutdown,
 539         .setsockopt =   sock_no_setsockopt,
 540         .getsockopt =   sock_no_getsockopt,
 541         .sendmsg =      unix_dgram_sendmsg,
 542         .recvmsg =      unix_dgram_recvmsg,
 543         .mmap =         sock_no_mmap,
 544         .sendpage =     sock_no_sendpage,
 545 };
 546
 547 static const struct proto_ops unix_seqpacket_ops = {
 548         .family =       PF_UNIX,
 549         .owner =        THIS_MODULE,
 550         .release =      unix_release,
 551         .bind =         unix_bind,
 552         .connect =      unix_stream_connect,
 553         .socketpair =   unix_socketpair,
 554         .accept =       unix_accept,
 555         .getname =      unix_getname,
 556         .poll =         unix_dgram_poll,
 557         .ioctl =        unix_ioctl,
 558         .listen =       unix_listen,
 559         .shutdown =     unix_shutdown,
 560         .setsockopt =   sock_no_setsockopt,
 561         .getsockopt =   sock_no_getsockopt,
 562         .sendmsg =      unix_seqpacket_sendmsg,
 563         .recvmsg =      unix_dgram_recvmsg,
 564         .mmap =         sock_no_mmap,
 565         .sendpage =     sock_no_sendpage,
 566 };
 567
 568 static struct proto unix_proto = {
 569         .name     = "UNIX",
 570         .owner    = THIS_MODULE,
 571         .obj_size = sizeof(struct unix_sock),
 572 };
 573
 574 /*
 575  * AF_UNIX sockets do not interact with hardware, hence they
 576  * dont trigger interrupts - so it's safe for them to have
 577  * bh-unsafe locking for their sk_receive_queue.lock. Split off
 578  * this special lock-class by reinitializing the spinlock key:
 579  */
 580 static struct lock_class_key af_unix_sk_receive_queue_lock_key;
 581
 582 static struct sock * unix_create1(struct net *net, struct socket *sock)
 583 {
 584         struct sock *sk = NULL;
 585         struct unix_sock *u;
 586
 587         atomic_inc(&unix_nr_socks);
 588         if (atomic_read(&unix_nr_socks) > 2 * get_max_files())
 589                 goto out;
 590
 591         sk = sk_alloc(net, PF_UNIX, GFP_KERNEL, &unix_proto);
 592         if (!sk)
 593                 goto out;
 594
 595         sock_init_data(sock,sk);
 596         lockdep_set_class(&sk->sk_receive_queue.lock,
 597                                 &af_unix_sk_receive_queue_lock_key);
 598
 599         sk->sk_write_space      = unix_write_space;
 600         sk->sk_max_ack_backlog  = net->unx.sysctl_max_dgram_qlen;
 601         sk->sk_destruct         = unix_sock_destructor;
 602         u         = unix_sk(sk);
 603         u->dentry = NULL;
 604         u->mnt    = NULL;
 605         spin_lock_init(&u->lock);
 606         atomic_set(&u->inflight, 0);
 607         INIT_LIST_HEAD(&u->link);
 608         mutex_init(&u->readlock); /* single task reading lock */
 609         init_waitqueue_head(&u->peer_wait);
 610         unix_insert_socket(unix_sockets_unbound, sk);
 611 out:
 612         if (sk == NULL)
 613                 atomic_dec(&unix_nr_socks);
 614         return sk;
 615 }
 616
 617 static int unix_create(struct net *net, struct socket *sock, int protocol)
 618 {
 619         if (protocol && protocol != PF_UNIX)
 620                 return -EPROTONOSUPPORT;
 621
 622         sock->state = SS_UNCONNECTED;
 623
 624         switch (sock->type) {
 625         case SOCK_STREAM:
 626                 sock->ops = &unix_stream_ops;
 627                 break;
 628                 /*
 629                  *      Believe it or not BSD has AF_UNIX, SOCK_RAW though
 630                  *      nothing uses it.
 631                  */
 632         case SOCK_RAW:
 633                 sock->type=SOCK_DGRAM;
 634         case SOCK_DGRAM:
 635                 sock->ops = &unix_dgram_ops;
 636                 break;
 637         case SOCK_SEQPACKET:
 638                 sock->ops = &unix_seqpacket_ops;
 639                 break;
 640         default:
 641                 return -ESOCKTNOSUPPORT;
 642         }
 643
 644         return unix_create1(net, sock) ? 0 : -ENOMEM;
 645 }
 646
 647 static int unix_release(struct socket *sock)
 648 {
 649         struct sock *sk = sock->sk;
 650
 651         if (!sk)
 652                 return 0;
 653
 654         sock->sk = NULL;
 655
 656         return unix_release_sock (sk, 0);
 657 }
 658
 659 static int unix_autobind(struct socket *sock)
 660 {
 661         struct sock *sk = sock->sk;
 662         struct net *net = sock_net(sk);
 663         struct unix_sock *u = unix_sk(sk);
 664         static u32 ordernum = 1;
 665         struct unix_address * addr;
 666         int err;
 667
 668         mutex_lock(&u->readlock);
 669
 670         err = 0;
 671         if (u->addr)
 672                 goto out;
 673
 674         err = -ENOMEM;
 675         addr = kzalloc(sizeof(*addr) + sizeof(short) + 16, GFP_KERNEL);
 676         if (!addr)
 677                 goto out;
 678
 679         addr->name->sun_family = AF_UNIX;
 680         atomic_set(&addr->refcnt, 1);
 681
 682 retry:
 683         addr->len = sprintf(addr->name->sun_path+1, "%05x", ordernum) + 1 + sizeof(short);
 684         addr->hash = unix_hash_fold(csum_partial((void*)addr->name, addr->len, 0));
 685
 686         spin_lock(&unix_table_lock);
 687         ordernum = (ordernum+1)&0xFFFFF;
 688
 689         if (__unix_find_socket_byname(net, addr->name, addr->len, sock->type,
 690                                       addr->hash)) {
 691                 spin_unlock(&unix_table_lock);
 692                 /* Sanity yield. It is unusual case, but yet... */
 693                 if (!(ordernum&0xFF))
 694                         yield();
 695                 goto retry;
 696         }
 697         addr->hash ^= sk->sk_type;
 698
 699         __unix_remove_socket(sk);
 700         u->addr = addr;
 701         __unix_insert_socket(&unix_socket_table[addr->hash], sk);
 702         spin_unlock(&unix_table_lock);
 703         err = 0;
 704
 705 out:    mutex_unlock(&u->readlock);
 706         return err;
 707 }
 708
 709 static struct sock *unix_find_other(struct net *net,
 710                                     struct sockaddr_un *sunname, int len,
 711                                     int type, unsigned hash, int *error)
 712 {
 713         struct sock *u;
 714         struct nameidata nd;
 715         int err = 0;
 716
 717         if (sunname->sun_path[0]) {
 718                 err = path_lookup(sunname->sun_path, LOOKUP_FOLLOW, &nd);
 719                 if (err)
 720                         goto fail;
 721                 err = vfs_permission(&nd, MAY_WRITE);
 722                 if (err)
 723                         goto put_fail;
 724
 725                 err = -ECONNREFUSED;
 726                 if (!S_ISSOCK(nd.path.dentry->d_inode->i_mode))
 727                         goto put_fail;
 728                 u = unix_find_socket_byinode(net, nd.path.dentry->d_inode);
 729                 if (!u)
 730                         goto put_fail;
 731
 732                 if (u->sk_type == type)
 733                         touch_atime(nd.path.mnt, nd.path.dentry);
 734
 735                 path_put(&nd.path);
 736
 737                 err=-EPROTOTYPE;
 738                 if (u->sk_type != type) {
 739                         sock_put(u);
 740                         goto fail;
 741                 }
 742         } else {
 743                 err = -ECONNREFUSED;
 744                 u=unix_find_socket_byname(net, sunname, len, type, hash);
 745                 if (u) {
 746                         struct dentry *dentry;
 747                         dentry = unix_sk(u)->dentry;
 748                         if (dentry)
 749                                 touch_atime(unix_sk(u)->mnt, dentry);
 750                 } else
 751                         goto fail;
 752         }
 753         return u;
 754
 755 put_fail:
 756         path_put(&nd.path);
 757 fail:
 758         *error=err;
 759         return NULL;
 760 }
 761
 762
 763 static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
 764 {
 765         struct sock *sk = sock->sk;
 766         struct net *net = sock_net(sk);
 767         struct unix_sock *u = unix_sk(sk);
 768         struct sockaddr_un *sunaddr=(struct sockaddr_un *)uaddr;
 769         struct dentry * dentry = NULL;
 770         struct nameidata nd;
 771         int err;
 772         unsigned hash;
 773         struct unix_address *addr;
 774         struct hlist_head *list;
 775
 776         err = -EINVAL;
 777         if (sunaddr->sun_family != AF_UNIX)
 778                 goto out;
 779
 780         if (addr_len==sizeof(short)) {
 781                 err = unix_autobind(sock);
 782                 goto out;
 783         }
 784
 785         err = unix_mkname(sunaddr, addr_len, &hash);
 786         if (err < 0)
 787                 goto out;
 788         addr_len = err;
 789
 790         mutex_lock(&u->readlock);
 791
 792         err = -EINVAL;
 793         if (u->addr)
 794                 goto out_up;
 795
 796         err = -ENOMEM;
 797         addr = kmalloc(sizeof(*addr)+addr_len, GFP_KERNEL);
 798         if (!addr)
 799                 goto out_up;
 800
 801         memcpy(addr->name, sunaddr, addr_len);
 802         addr->len = addr_len;
 803         addr->hash = hash ^ sk->sk_type;
 804         atomic_set(&addr->refcnt, 1);
 805
 806         if (sunaddr->sun_path[0]) {
 807                 unsigned int mode;
 808                 err = 0;
 809                 /*
 810                  * Get the parent directory, calculate the hash for last
 811                  * component.
 812                  */
 813                 err = path_lookup(sunaddr->sun_path, LOOKUP_PARENT, &nd);
 814                 if (err)
 815                         goto out_mknod_parent;
 816
 817                 dentry = lookup_create(&nd, 0);
 818                 err = PTR_ERR(dentry);
 819                 if (IS_ERR(dentry))
 820                         goto out_mknod_unlock;
 821
 822                 /*
 823                  * All right, let's create it.
 824                  */
 825                 mode = S_IFSOCK |
 826                        (SOCK_INODE(sock)->i_mode & ~current->fs->umask);
 827                 err = mnt_want_write(nd.path.mnt);
 828                 if (err)
 829                         goto out_mknod_dput;
 830                 err = vfs_mknod(nd.path.dentry->d_inode, dentry, mode, 0);
 831                 mnt_drop_write(nd.path.mnt);
 832                 if (err)
 833                         goto out_mknod_dput;
 834                 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
 835                 dput(nd.path.dentry);
 836                 nd.path.dentry = dentry;
 837
 838                 addr->hash = UNIX_HASH_SIZE;
 839         }
 840
 841         spin_lock(&unix_table_lock);
 842
 843         if (!sunaddr->sun_path[0]) {
 844                 err = -EADDRINUSE;
 845                 if (__unix_find_socket_byname(net, sunaddr, addr_len,
 846                                               sk->sk_type, hash)) {
 847                         unix_release_addr(addr);
 848                         goto out_unlock;
 849                 }
 850
 851                 list = &unix_socket_table[addr->hash];
 852         } else {
 853                 list = &unix_socket_table[dentry->d_inode->i_ino & (UNIX_HASH_SIZE-1)];
 854                 u->dentry = nd.path.dentry;
 855                 u->mnt    = nd.path.mnt;
 856         }
 857
 858         err = 0;
 859         __unix_remove_socket(sk);
 860         u->addr = addr;
 861         __unix_insert_socket(list, sk);
 862
 863 out_unlock:
 864         spin_unlock(&unix_table_lock);
 865 out_up:
 866         mutex_unlock(&u->readlock);
 867 out:
 868         return err;
 869
 870 out_mknod_dput:
 871         dput(dentry);
 872 out_mknod_unlock:
 873         mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
 874         path_put(&nd.path);
 875 out_mknod_parent:
 876         if (err==-EEXIST)
 877                 err=-EADDRINUSE;
 878         unix_release_addr(addr);
 879         goto out_up;
 880 }
 881
 882 static void unix_state_double_lock(struct sock *sk1, struct sock *sk2)
 883 {
 884         if (unlikely(sk1 == sk2) || !sk2) {
 885                 unix_state_lock(sk1);
 886                 return;
 887         }
 888         if (sk1 < sk2) {
 889                 unix_state_lock(sk1);
 890                 unix_state_lock_nested(sk2);
 891         } else {
 892                 unix_state_lock(sk2);
 893                 unix_state_lock_nested(sk1);
 894         }
 895 }
 896
 897 static void unix_state_double_unlock(struct sock *sk1, struct sock *sk2)
 898 {
 899         if (unlikely(sk1 == sk2) || !sk2) {
 900                 unix_state_unlock(sk1);
 901                 return;
 902         }
 903         unix_state_unlock(sk1);
 904         unix_state_unlock(sk2);
 905 }
 906
 907 static int unix_dgram_connect(struct socket *sock, struct sockaddr *addr,
 908                               int alen, int flags)
 909 {
 910         struct sock *sk = sock->sk;
 911         struct net *net = sock_net(sk);
 912         struct sockaddr_un *sunaddr=(struct sockaddr_un*)addr;
 913         struct sock *other;
 914         unsigned hash;
 915         int err;
 916
 917         if (addr->sa_family != AF_UNSPEC) {
 918                 err = unix_mkname(sunaddr, alen, &hash);
 919                 if (err < 0)
 920                         goto out;
 921                 alen = err;
 922
 923                 if (test_bit(SOCK_PASSCRED, &sock->flags) &&
 924                     !unix_sk(sk)->addr && (err = unix_autobind(sock)) != 0)
 925                         goto out;
 926
 927 restart:
 928                 other=unix_find_other(net, sunaddr, alen, sock->type, hash, &err);
 929                 if (!other)
 930                         goto out;
 931
 932                 unix_state_double_lock(sk, other);
 933
 934                 /* Apparently VFS overslept socket death. Retry. */
 935                 if (sock_flag(other, SOCK_DEAD)) {
 936                         unix_state_double_unlock(sk, other);
 937                         sock_put(other);
 938                         goto restart;
 939                 }
 940
 941                 err = -EPERM;
 942                 if (!unix_may_send(sk, other))
 943                         goto out_unlock;
 944
 945                 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
 946                 if (err)
 947                         goto out_unlock;
 948
 949         } else {
 950                 /*
 951                  *      1003.1g breaking connected state with AF_UNSPEC
 952                  */
 953                 other = NULL;
 954                 unix_state_double_lock(sk, other);
 955         }
 956
 957         /*
 958          * If it was connected, reconnect.
 959          */
 960         if (unix_peer(sk)) {
 961                 struct sock *old_peer = unix_peer(sk);
 962                 unix_peer(sk)=other;
 963                 unix_state_double_unlock(sk, other);
 964
 965                 if (other != old_peer)
 966                         unix_dgram_disconnected(sk, old_peer);
 967                 sock_put(old_peer);
 968         } else {
 969                 unix_peer(sk)=other;
 970                 unix_state_double_unlock(sk, other);
 971         }
 972         return 0;
 973
 974 out_unlock:
 975         unix_state_double_unlock(sk, other);
 976         sock_put(other);
 977 out:
 978         return err;
 979 }
 980
 981 static long unix_wait_for_peer(struct sock *other, long timeo)
 982 {
 983         struct unix_sock *u = unix_sk(other);
 984         int sched;
 985         DEFINE_WAIT(wait);
 986
 987         prepare_to_wait_exclusive(&u->peer_wait, &wait, TASK_INTERRUPTIBLE);
 988
 989         sched = !sock_flag(other, SOCK_DEAD) &&
 990                 !(other->sk_shutdown & RCV_SHUTDOWN) &&
 991                 unix_recvq_full(other);
 992
 993         unix_state_unlock(other);
 994
 995         if (sched)
 996                 timeo = schedule_timeout(timeo);
 997
 998         finish_wait(&u->peer_wait, &wait);
 999         return timeo;
1000 }
1001
1002 static int unix_stream_connect(struct socket *sock, struct sockaddr *uaddr,
1003                                int addr_len, int flags)
1004 {
1005         struct sockaddr_un *sunaddr=(struct sockaddr_un *)uaddr;
1006         struct sock *sk = sock->sk;
1007         struct net *net = sock_net(sk);
1008         struct unix_sock *u = unix_sk(sk), *newu, *otheru;
1009         struct sock *newsk = NULL;
1010         struct sock *other = NULL;
1011         struct sk_buff *skb = NULL;
1012         unsigned hash;
1013         int st;
1014         int err;
1015         long timeo;
1016
1017         err = unix_mkname(sunaddr, addr_len, &hash);
1018         if (err < 0)
1019                 goto out;
1020         addr_len = err;
1021
1022         if (test_bit(SOCK_PASSCRED, &sock->flags)
1023                 && !u->addr && (err = unix_autobind(sock)) != 0)
1024                 goto out;
1025
1026         timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
1027
1028         /* First of all allocate resources.
1029            If we will make it after state is locked,
1030            we will have to recheck all again in any case.
1031          */
1032
1033         err = -ENOMEM;
1034
1035         /* create new sock for complete connection */
1036         newsk = unix_create1(sock_net(sk), NULL);
1037         if (newsk == NULL)
1038                 goto out;
1039
1040         /* Allocate skb for sending to listening sock */
1041         skb = sock_wmalloc(newsk, 1, 0, GFP_KERNEL);
1042         if (skb == NULL)
1043                 goto out;
1044
1045 restart:
1046         /*  Find listening sock. */
1047         other = unix_find_other(net, sunaddr, addr_len, sk->sk_type, hash, &err);
1048         if (!other)
1049                 goto out;
1050
1051         /* Latch state of peer */
1052         unix_state_lock(other);
1053
1054         /* Apparently VFS overslept socket death. Retry. */
1055         if (sock_flag(other, SOCK_DEAD)) {
1056                 unix_state_unlock(other);
1057                 sock_put(other);
1058                 goto restart;
1059         }
1060
1061         err = -ECONNREFUSED;
1062         if (other->sk_state != TCP_LISTEN)
1063                 goto out_unlock;
1064
1065         if (unix_recvq_full(other)) {
1066                 err = -EAGAIN;
1067                 if (!timeo)
1068                         goto out_unlock;
1069
1070                 timeo = unix_wait_for_peer(other, timeo);
1071
1072                 err = sock_intr_errno(timeo);
1073                 if (signal_pending(current))
1074                         goto out;
1075                 sock_put(other);
1076                 goto restart;
1077         }
1078
1079         /* Latch our state.
1080
1081            It is tricky place. We need to grab write lock and cannot
1082            drop lock on peer. It is dangerous because deadlock is
1083            possible. Connect to self case and simultaneous
1084            attempt to connect are eliminated by checking socket
1085            state. other is TCP_LISTEN, if sk is TCP_LISTEN we
1086            check this before attempt to grab lock.
1087
1088            Well, and we have to recheck the state after socket locked.
1089          */
1090         st = sk->sk_state;
1091
1092         switch (st) {
1093         case TCP_CLOSE:
1094                 /* This is ok... continue with connect */
1095                 break;
1096         case TCP_ESTABLISHED:
1097                 /* Socket is already connected */
1098                 err = -EISCONN;
1099                 goto out_unlock;
1100         default:
1101                 err = -EINVAL;
1102                 goto out_unlock;
1103         }
1104
1105         unix_state_lock_nested(sk);
1106
1107         if (sk->sk_state != st) {
1108                 unix_state_unlock(sk);
1109                 unix_state_unlock(other);
1110                 sock_put(other);
1111                 goto restart;
1112         }
1113
1114         err = security_unix_stream_connect(sock, other->sk_socket, newsk);
1115         if (err) {
1116                 unix_state_unlock(sk);
1117                 goto out_unlock;
1118         }
1119
1120         /* The way is open! Fastly set all the necessary fields... */
1121
1122         sock_hold(sk);
1123         unix_peer(newsk)        = sk;
1124         newsk->sk_state         = TCP_ESTABLISHED;
1125         newsk->sk_type          = sk->sk_type;
1126         newsk->sk_peercred.pid  = task_tgid_vnr(current);
1127         newsk->sk_peercred.uid  = current->euid;
1128         newsk->sk_peercred.gid  = current->egid;
1129         newu = unix_sk(newsk);
1130         newsk->sk_sleep         = &newu->peer_wait;
1131         otheru = unix_sk(other);
1132
1133         /* copy address information from listening to new sock*/
1134         if (otheru->addr) {
1135                 atomic_inc(&otheru->addr->refcnt);
1136                 newu->addr = otheru->addr;
1137         }
1138         if (otheru->dentry) {
1139                 newu->dentry    = dget(otheru->dentry);
1140                 newu->mnt       = mntget(otheru->mnt);
1141         }
1142
1143         /* Set credentials */
1144         sk->sk_peercred = other->sk_peercred;
1145
1146         sock->state     = SS_CONNECTED;
1147         sk->sk_state    = TCP_ESTABLISHED;
1148         sock_hold(newsk);
1149
1150         smp_mb__after_atomic_inc();     /* sock_hold() does an atomic_inc() */
1151         unix_peer(sk)   = newsk;
1152
1153         unix_state_unlock(sk);
1154
1155         /* take ten and and send info to listening sock */
1156         spin_lock(&other->sk_receive_queue.lock);
1157         __skb_queue_tail(&other->sk_receive_queue, skb);
1158         spin_unlock(&other->sk_receive_queue.lock);
1159         unix_state_unlock(other);
1160         other->sk_data_ready(other, 0);
1161         sock_put(other);
1162         return 0;
1163
1164 out_unlock:
1165         if (other)
1166                 unix_state_unlock(other);
1167
1168 out:
1169         if (skb)
1170                 kfree_skb(skb);
1171         if (newsk)
1172                 unix_release_sock(newsk, 0);
1173         if (other)
1174                 sock_put(other);
1175         return err;
1176 }
1177
1178 static int unix_socketpair(struct socket *socka, struct socket *sockb)
1179 {
1180         struct sock *ska=socka->sk, *skb = sockb->sk;
1181
1182         /* Join our sockets back to back */
1183         sock_hold(ska);
1184         sock_hold(skb);
1185         unix_peer(ska)=skb;
1186         unix_peer(skb)=ska;
1187         ska->sk_peercred.pid = skb->sk_peercred.pid = task_tgid_vnr(current);
1188         ska->sk_peercred.uid = skb->sk_peercred.uid = current->euid;
1189         ska->sk_peercred.gid = skb->sk_peercred.gid = current->egid;
1190
1191         if (ska->sk_type != SOCK_DGRAM) {
1192                 ska->sk_state = TCP_ESTABLISHED;
1193                 skb->sk_state = TCP_ESTABLISHED;
1194                 socka->state  = SS_CONNECTED;
1195                 sockb->state  = SS_CONNECTED;
1196         }
1197         return 0;
1198 }
1199
1200 static int unix_accept(struct socket *sock, struct socket *newsock, int flags)
1201 {
1202         struct sock *sk = sock->sk;
1203         struct sock *tsk;
1204         struct sk_buff *skb;
1205         int err;
1206
1207         err = -EOPNOTSUPP;
1208         if (sock->type!=SOCK_STREAM && sock->type!=SOCK_SEQPACKET)
1209                 goto out;
1210
1211         err = -EINVAL;
1212         if (sk->sk_state != TCP_LISTEN)
1213                 goto out;
1214
1215         /* If socket state is TCP_LISTEN it cannot change (for now...),
1216          * so that no locks are necessary.
1217          */
1218
1219         skb = skb_recv_datagram(sk, 0, flags&O_NONBLOCK, &err);
1220         if (!skb) {
1221                 /* This means receive shutdown. */
1222                 if (err == 0)
1223                         err = -EINVAL;
1224                 goto out;
1225         }
1226
1227         tsk = skb->sk;
1228         skb_free_datagram(sk, skb);
1229         wake_up_interruptible(&unix_sk(sk)->peer_wait);
1230
1231         /* attach accepted sock to socket */
1232         unix_state_lock(tsk);
1233         newsock->state = SS_CONNECTED;
1234         sock_graft(tsk, newsock);
1235         unix_state_unlock(tsk);
1236         return 0;
1237
1238 out:
1239         return err;
1240 }
1241
1242
1243 static int unix_getname(struct socket *sock, struct sockaddr *uaddr, int *uaddr_len, int peer)
1244 {
1245         struct sock *sk = sock->sk;
1246         struct unix_sock *u;
1247         struct sockaddr_un *sunaddr=(struct sockaddr_un *)uaddr;
1248         int err = 0;
1249
1250         if (peer) {
1251                 sk = unix_peer_get(sk);
1252
1253                 err = -ENOTCONN;
1254                 if (!sk)
1255                         goto out;
1256                 err = 0;
1257         } else {
1258                 sock_hold(sk);
1259         }
1260
1261         u = unix_sk(sk);
1262         unix_state_lock(sk);
1263         if (!u->addr) {
1264                 sunaddr->sun_family = AF_UNIX;
1265                 sunaddr->sun_path[0] = 0;
1266                 *uaddr_len = sizeof(short);
1267         } else {
1268                 struct unix_address *addr = u->addr;
1269
1270                 *uaddr_len = addr->len;
1271                 memcpy(sunaddr, addr->name, *uaddr_len);
1272         }
1273         unix_state_unlock(sk);
1274         sock_put(sk);
1275 out:
1276         return err;
1277 }
1278
1279 static void unix_detach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1280 {
1281         int i;
1282
1283         scm->fp = UNIXCB(skb).fp;
1284         skb->destructor = sock_wfree;
1285         UNIXCB(skb).fp = NULL;
1286
1287         for (i=scm->fp->count-1; i>=0; i--)
1288                 unix_notinflight(scm->fp->fp[i]);
1289 }
1290
1291 static void unix_destruct_fds(struct sk_buff *skb)
1292 {
1293         struct scm_cookie scm;
1294         memset(&scm, 0, sizeof(scm));
1295         unix_detach_fds(&scm, skb);
1296
1297         /* Alas, it calls VFS */
1298         /* So fscking what? fput() had been SMP-safe since the last Summer */
1299         scm_destroy(&scm);
1300         sock_wfree(skb);
1301 }
1302
1303 static void unix_attach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1304 {
1305         int i;
1306         for (i=scm->fp->count-1; i>=0; i--)
1307                 unix_inflight(scm->fp->fp[i]);
1308         UNIXCB(skb).fp = scm->fp;
1309         skb->destructor = unix_destruct_fds;
1310         scm->fp = NULL;
1311 }
1312
1313 /*
1314  *      Send AF_UNIX data.
1315  */
1316
1317 static int unix_dgram_sendmsg(struct kiocb *kiocb, struct socket *sock,
1318                               struct msghdr *msg, size_t len)
1319 {
1320         struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1321         struct sock *sk = sock->sk;
1322         struct net *net = sock_net(sk);
1323         struct unix_sock *u = unix_sk(sk);
1324         struct sockaddr_un *sunaddr=msg->msg_name;
1325         struct sock *other = NULL;
1326         int namelen = 0; /* fake GCC */
1327         int err;
1328         unsigned hash;
1329         struct sk_buff *skb;
1330         long timeo;
1331         struct scm_cookie tmp_scm;
1332
1333         if (NULL == siocb->scm)
1334                 siocb->scm = &tmp_scm;
1335         err = scm_send(sock, msg, siocb->scm);
1336         if (err < 0)
1337                 return err;
1338
1339         err = -EOPNOTSUPP;
1340         if (msg->msg_flags&MSG_OOB)
1341                 goto out;
1342
1343         if (msg->msg_namelen) {
1344                 err = unix_mkname(sunaddr, msg->msg_namelen, &hash);
1345                 if (err < 0)
1346                         goto out;
1347                 namelen = err;
1348         } else {
1349                 sunaddr = NULL;
1350                 err = -ENOTCONN;
1351                 other = unix_peer_get(sk);
1352                 if (!other)
1353                         goto out;
1354         }
1355
1356         if (test_bit(SOCK_PASSCRED, &sock->flags)
1357                 && !u->addr && (err = unix_autobind(sock)) != 0)
1358                 goto out;
1359
1360         err = -EMSGSIZE;
1361         if (len > sk->sk_sndbuf - 32)
1362                 goto out;
1363
1364         skb = sock_alloc_send_skb(sk, len, msg->msg_flags&MSG_DONTWAIT, &err);
1365         if (skb==NULL)
1366                 goto out;
1367
1368         memcpy(UNIXCREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1369         if (siocb->scm->fp)
1370                 unix_attach_fds(siocb->scm, skb);
1371         unix_get_secdata(siocb->scm, skb);
1372
1373         skb_reset_transport_header(skb);
1374         err = memcpy_fromiovec(skb_put(skb,len), msg->msg_iov, len);
1375         if (err)
1376                 goto out_free;
1377
1378         timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1379
1380 restart:
1381         if (!other) {
1382                 err = -ECONNRESET;
1383                 if (sunaddr == NULL)
1384                         goto out_free;
1385
1386                 other = unix_find_other(net, sunaddr, namelen, sk->sk_type,
1387                                         hash, &err);
1388                 if (other==NULL)
1389                         goto out_free;
1390         }
1391
1392         unix_state_lock(other);
1393         err = -EPERM;
1394         if (!unix_may_send(sk, other))
1395                 goto out_unlock;
1396
1397         if (sock_flag(other, SOCK_DEAD)) {
1398                 /*
1399                  *      Check with 1003.1g - what should
1400                  *      datagram error
1401                  */
1402                 unix_state_unlock(other);
1403                 sock_put(other);
1404
1405                 err = 0;
1406                 unix_state_lock(sk);
1407                 if (unix_peer(sk) == other) {
1408                         unix_peer(sk)=NULL;
1409                         unix_state_unlock(sk);
1410
1411                         unix_dgram_disconnected(sk, other);
1412                         sock_put(other);
1413                         err = -ECONNREFUSED;
1414                 } else {
1415                         unix_state_unlock(sk);
1416                 }
1417
1418                 other = NULL;
1419                 if (err)
1420                         goto out_free;
1421                 goto restart;
1422         }
1423
1424         err = -EPIPE;
1425         if (other->sk_shutdown & RCV_SHUTDOWN)
1426                 goto out_unlock;
1427
1428         if (sk->sk_type != SOCK_SEQPACKET) {
1429                 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1430                 if (err)
1431                         goto out_unlock;
1432         }
1433
1434         if (unix_peer(other) != sk && unix_recvq_full(other)) {
1435                 if (!timeo) {
1436                         err = -EAGAIN;
1437                         goto out_unlock;
1438                 }
1439
1440                 timeo = unix_wait_for_peer(other, timeo);
1441
1442                 err = sock_intr_errno(timeo);
1443                 if (signal_pending(current))
1444                         goto out_free;
1445
1446                 goto restart;
1447         }
1448
1449         skb_queue_tail(&other->sk_receive_queue, skb);
1450         unix_state_unlock(other);
1451         other->sk_data_ready(other, len);
1452         sock_put(other);
1453         scm_destroy(siocb->scm);
1454         return len;
1455
1456 out_unlock:
1457         unix_state_unlock(other);
1458 out_free:
1459         kfree_skb(skb);
1460 out:
1461         if (other)
1462                 sock_put(other);
1463         scm_destroy(siocb->scm);
1464         return err;
1465 }
1466
1467
1468 static int unix_stream_sendmsg(struct kiocb *kiocb, struct socket *sock,
1469                                struct msghdr *msg, size_t len)
1470 {
1471         struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1472         struct sock *sk = sock->sk;
1473         struct sock *other = NULL;
1474         struct sockaddr_un *sunaddr=msg->msg_name;
1475         int err,size;
1476         struct sk_buff *skb;
1477         int sent=0;
1478         struct scm_cookie tmp_scm;
1479
1480         if (NULL == siocb->scm)
1481                 siocb->scm = &tmp_scm;
1482         err = scm_send(sock, msg, siocb->scm);
1483         if (err < 0)
1484                 return err;
1485
1486         err = -EOPNOTSUPP;
1487         if (msg->msg_flags&MSG_OOB)
1488                 goto out_err;
1489
1490         if (msg->msg_namelen) {
1491                 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
1492                 goto out_err;
1493         } else {
1494                 sunaddr = NULL;
1495                 err = -ENOTCONN;
1496                 other = unix_peer(sk);
1497                 if (!other)
1498                         goto out_err;
1499         }
1500
1501         if (sk->sk_shutdown & SEND_SHUTDOWN)
1502                 goto pipe_err;
1503
1504         while(sent < len)
1505         {
1506                 /*
1507                  *      Optimisation for the fact that under 0.01% of X
1508                  *      messages typically need breaking up.
1509                  */
1510
1511                 size = len-sent;
1512
1513                 /* Keep two messages in the pipe so it schedules better */
1514                 if (size > ((sk->sk_sndbuf >> 1) - 64))
1515                         size = (sk->sk_sndbuf >> 1) - 64;
1516
1517                 if (size > SKB_MAX_ALLOC)
1518                         size = SKB_MAX_ALLOC;
1519
1520                 /*
1521                  *      Grab a buffer
1522                  */
1523
1524                 skb=sock_alloc_send_skb(sk,size,msg->msg_flags&MSG_DONTWAIT, &err);
1525
1526                 if (skb==NULL)
1527                         goto out_err;
1528
1529                 /*
1530                  *      If you pass two values to the sock_alloc_send_skb
1531                  *      it tries to grab the large buffer with GFP_NOFS
1532                  *      (which can fail easily), and if it fails grab the
1533                  *      fallback size buffer which is under a page and will
1534                  *      succeed. [Alan]
1535                  */
1536                 size = min_t(int, size, skb_tailroom(skb));
1537
1538                 memcpy(UNIXCREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1539                 if (siocb->scm->fp)
1540                         unix_attach_fds(siocb->scm, skb);
1541
1542                 if ((err = memcpy_fromiovec(skb_put(skb,size), msg->msg_iov, size)) != 0) {
1543                         kfree_skb(skb);
1544                         goto out_err;
1545                 }
1546
1547                 unix_state_lock(other);
1548
1549                 if (sock_flag(other, SOCK_DEAD) ||
1550                     (other->sk_shutdown & RCV_SHUTDOWN))
1551                         goto pipe_err_free;
1552
1553                 skb_queue_tail(&other->sk_receive_queue, skb);
1554                 unix_state_unlock(other);
1555                 other->sk_data_ready(other, size);
1556                 sent+=size;
1557         }
1558
1559         scm_destroy(siocb->scm);
1560         siocb->scm = NULL;
1561
1562         return sent;
1563
1564 pipe_err_free:
1565         unix_state_unlock(other);
1566         kfree_skb(skb);
1567 pipe_err:
1568         if (sent==0 && !(msg->msg_flags&MSG_NOSIGNAL))
1569                 send_sig(SIGPIPE,current,0);
1570         err = -EPIPE;
1571 out_err:
1572         scm_destroy(siocb->scm);
1573         siocb->scm = NULL;
1574         return sent ? : err;
1575 }
1576
1577 static int unix_seqpacket_sendmsg(struct kiocb *kiocb, struct socket *sock,
1578                                   struct msghdr *msg, size_t len)
1579 {
1580         int err;
1581         struct sock *sk = sock->sk;
1582
1583         err = sock_error(sk);
1584         if (err)
1585                 return err;
1586
1587         if (sk->sk_state != TCP_ESTABLISHED)
1588                 return -ENOTCONN;
1589
1590         if (msg->msg_namelen)
1591                 msg->msg_namelen = 0;
1592
1593         return unix_dgram_sendmsg(kiocb, sock, msg, len);
1594 }
1595
1596 static void unix_copy_addr(struct msghdr *msg, struct sock *sk)
1597 {
1598         struct unix_sock *u = unix_sk(sk);
1599
1600         msg->msg_namelen = 0;
1601         if (u->addr) {
1602                 msg->msg_namelen = u->addr->len;
1603                 memcpy(msg->msg_name, u->addr->name, u->addr->len);
1604         }
1605 }
1606
1607 static int unix_dgram_recvmsg(struct kiocb *iocb, struct socket *sock,
1608                               struct msghdr *msg, size_t size,
1609                               int flags)
1610 {
1611         struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1612         struct scm_cookie tmp_scm;
1613         struct sock *sk = sock->sk;
1614         struct unix_sock *u = unix_sk(sk);
1615         int noblock = flags & MSG_DONTWAIT;
1616         struct sk_buff *skb;
1617         int err;
1618
1619         err = -EOPNOTSUPP;
1620         if (flags&MSG_OOB)
1621                 goto out;
1622
1623         msg->msg_namelen = 0;
1624
1625         mutex_lock(&u->readlock);
1626
1627         skb = skb_recv_datagram(sk, flags, noblock, &err);
1628         if (!skb) {
1629                 unix_state_lock(sk);
1630                 /* Signal EOF on disconnected non-blocking SEQPACKET socket. */
1631                 if (sk->sk_type == SOCK_SEQPACKET && err == -EAGAIN &&
1632                     (sk->sk_shutdown & RCV_SHUTDOWN))
1633                         err = 0;
1634                 unix_state_unlock(sk);
1635                 goto out_unlock;
1636         }
1637
1638         wake_up_interruptible_sync(&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);
1657         unix_set_secdata(siocb->scm, skb);
1658
1659         if (!(flags & MSG_PEEK))
1660         {
1661                 if (UNIXCB(skb).fp)
1662                         unix_detach_fds(siocb->scm, skb);
1663         }
1664         else
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)
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
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:
1688         mutex_unlock(&u->readlock);
1689 out:
1690         return err;
1691 }
1692
1693 /*
1694  *      Sleep until data has arrive. But check for races..
1695  */
1696
1697 static long unix_stream_data_wait(struct sock * sk, long timeo)
1698 {
1699         DEFINE_WAIT(wait);
1700
1701         unix_state_lock(sk);
1702
1703         for (;;) {
1704                 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1705
1706                 if (!skb_queue_empty(&sk->sk_receive_queue) ||
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);
1714                 unix_state_unlock(sk);
1715                 timeo = schedule_timeout(timeo);
1716                 unix_state_lock(sk);
1717                 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1718         }
1719
1720         finish_wait(sk->sk_sleep, &wait);
1721         unix_state_unlock(sk);
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
1764         mutex_lock(&u->readlock);
1765
1766         do
1767         {
1768                 int chunk;
1769                 struct sk_buff *skb;
1770
1771                 unix_state_lock(sk);
1772                 skb = skb_dequeue(&sk->sk_receive_queue);
1773                 if (skb==NULL)
1774                 {
1775                         if (copied >= target)
1776                                 goto unlock;
1777
1778                         /*
1779                          *      POSIX 1003.1g mandates this order.
1780                          */
1781
1782                         if ((err = sock_error(sk)) != 0)
1783                                 goto unlock;
1784                         if (sk->sk_shutdown & RCV_SHUTDOWN)
1785                                 goto unlock;
1786
1787                         unix_state_unlock(sk);
1788                         err = -EAGAIN;
1789                         if (!timeo)
1790                                 break;
1791                         mutex_unlock(&u->readlock);
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                         }
1799                         mutex_lock(&u->readlock);
1800                         continue;
1801  unlock:
1802                         unix_state_unlock(sk);
1803                         break;
1804                 }
1805                 unix_state_unlock(sk);
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
1869         mutex_unlock(&u->readlock);
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) {
1883                 unix_state_lock(sk);
1884                 sk->sk_shutdown |= mode;
1885                 other=unix_peer(sk);
1886                 if (other)
1887                         sock_hold(other);
1888                 unix_state_unlock(sk);
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;
1900                         unix_state_lock(other);
1901                         other->sk_shutdown |= peer_mode;
1902                         unix_state_unlock(other);
1903                         other->sk_state_change(other);
1904                         read_lock(&other->sk_callback_lock);
1905                         if (peer_mode == SHUTDOWN_MASK)
1906                                 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_HUP);
1907                         else if (peer_mode & RCV_SHUTDOWN)
1908                                 sk_wake_async(other, SOCK_WAKE_WAITD, 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:
1954                         err = -ENOIOCTLCMD;
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;
1973         if (sk->sk_shutdown & RCV_SHUTDOWN)
1974                 mask |= POLLRDHUP;
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 static unsigned int unix_dgram_poll(struct file *file, struct socket *sock,
1996                                     poll_table *wait)
1997 {
1998         struct sock *sk = sock->sk, *other;
1999         unsigned int mask, writable;
2000
2001         poll_wait(file, sk->sk_sleep, wait);
2002         mask = 0;
2003
2004         /* exceptional events? */
2005         if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue))
2006                 mask |= POLLERR;
2007         if (sk->sk_shutdown & RCV_SHUTDOWN)
2008                 mask |= POLLRDHUP;
2009         if (sk->sk_shutdown == SHUTDOWN_MASK)
2010                 mask |= POLLHUP;
2011
2012         /* readable? */
2013         if (!skb_queue_empty(&sk->sk_receive_queue) ||
2014             (sk->sk_shutdown & RCV_SHUTDOWN))
2015                 mask |= POLLIN | POLLRDNORM;
2016
2017         /* Connection-based need to check for termination and startup */
2018         if (sk->sk_type == SOCK_SEQPACKET) {
2019                 if (sk->sk_state == TCP_CLOSE)
2020                         mask |= POLLHUP;
2021                 /* connection hasn't started yet? */
2022                 if (sk->sk_state == TCP_SYN_SENT)
2023                         return mask;
2024         }
2025
2026         /* writable? */
2027         writable = unix_writable(sk);
2028         if (writable) {
2029                 other = unix_peer_get(sk);
2030                 if (other) {
2031                         if (unix_peer(other) != sk) {
2032                                 poll_wait(file, &unix_sk(other)->peer_wait,
2033                                           wait);
2034                                 if (unix_recvq_full(other))
2035                                         writable = 0;
2036                         }
2037
2038                         sock_put(other);
2039                 }
2040         }
2041
2042         if (writable)
2043                 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2044         else
2045                 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
2046
2047         return mask;
2048 }
2049
2050 #ifdef CONFIG_PROC_FS
2051 static struct sock *first_unix_socket(int *i)
2052 {
2053         for (*i = 0; *i <= UNIX_HASH_SIZE; (*i)++) {
2054                 if (!hlist_empty(&unix_socket_table[*i]))
2055                         return __sk_head(&unix_socket_table[*i]);
2056         }
2057         return NULL;
2058 }
2059
2060 static struct sock *next_unix_socket(int *i, struct sock *s)
2061 {
2062         struct sock *next = sk_next(s);
2063         /* More in this chain? */
2064         if (next)
2065                 return next;
2066         /* Look for next non-empty chain. */
2067         for ((*i)++; *i <= UNIX_HASH_SIZE; (*i)++) {
2068                 if (!hlist_empty(&unix_socket_table[*i]))
2069                         return __sk_head(&unix_socket_table[*i]);
2070         }
2071         return NULL;
2072 }
2073
2074 struct unix_iter_state {
2075         struct seq_net_private p;
2076         int i;
2077 };
2078 static struct sock *unix_seq_idx(struct seq_file *seq, loff_t pos)
2079 {
2080         struct unix_iter_state *iter = seq->private;
2081         loff_t off = 0;
2082         struct sock *s;
2083
2084         for (s = first_unix_socket(&iter->i); s; s = next_unix_socket(&iter->i, s)) {
2085                 if (sock_net(s) != seq_file_net(seq))
2086                         continue;
2087                 if (off == pos)
2088                         return s;
2089                 ++off;
2090         }
2091         return NULL;
2092 }
2093
2094
2095 static void *unix_seq_start(struct seq_file *seq, loff_t *pos)
2096         __acquires(unix_table_lock)
2097 {
2098         spin_lock(&unix_table_lock);
2099         return *pos ? unix_seq_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2100 }
2101
2102 static void *unix_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2103 {
2104         struct unix_iter_state *iter = seq->private;
2105         struct sock *sk = v;
2106         ++*pos;
2107
2108         if (v == SEQ_START_TOKEN)
2109                 sk = first_unix_socket(&iter->i);
2110         else
2111                 sk = next_unix_socket(&iter->i, sk);
2112         while (sk && (sock_net(sk) != seq_file_net(seq)))
2113                 sk = next_unix_socket(&iter->i, sk);
2114         return sk;
2115 }
2116
2117 static void unix_seq_stop(struct seq_file *seq, void *v)
2118         __releases(unix_table_lock)
2119 {
2120         spin_unlock(&unix_table_lock);
2121 }
2122
2123 static int unix_seq_show(struct seq_file *seq, void *v)
2124 {
2125
2126         if (v == SEQ_START_TOKEN)
2127                 seq_puts(seq, "Num       RefCount Protocol Flags    Type St "
2128                          "Inode Path\n");
2129         else {
2130                 struct sock *s = v;
2131                 struct unix_sock *u = unix_sk(s);
2132                 unix_state_lock(s);
2133
2134                 seq_printf(seq, "%p: %08X %08X %08X %04X %02X %5lu",
2135                         s,
2136                         atomic_read(&s->sk_refcnt),
2137                         0,
2138                         s->sk_state == TCP_LISTEN ? __SO_ACCEPTCON : 0,
2139                         s->sk_type,
2140                         s->sk_socket ?
2141                         (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTED : SS_UNCONNECTED) :
2142                         (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTING : SS_DISCONNECTING),
2143                         sock_i_ino(s));
2144
2145                 if (u->addr) {
2146                         int i, len;
2147                         seq_putc(seq, ' ');
2148
2149                         i = 0;
2150                         len = u->addr->len - sizeof(short);
2151                         if (!UNIX_ABSTRACT(s))
2152                                 len--;
2153                         else {
2154                                 seq_putc(seq, '@');
2155                                 i++;
2156                         }
2157                         for ( ; i < len; i++)
2158                                 seq_putc(seq, u->addr->name->sun_path[i]);
2159                 }
2160                 unix_state_unlock(s);
2161                 seq_putc(seq, '\n');
2162         }
2163
2164         return 0;
2165 }
2166
2167 static const struct seq_operations unix_seq_ops = {
2168         .start  = unix_seq_start,
2169         .next   = unix_seq_next,
2170         .stop   = unix_seq_stop,
2171         .show   = unix_seq_show,
2172 };
2173
2174
2175 static int unix_seq_open(struct inode *inode, struct file *file)
2176 {
2177         return seq_open_net(inode, file, &unix_seq_ops,
2178                             sizeof(struct unix_iter_state));
2179 }
2180
2181 static const struct file_operations unix_seq_fops = {
2182         .owner          = THIS_MODULE,
2183         .open           = unix_seq_open,
2184         .read           = seq_read,
2185         .llseek         = seq_lseek,
2186         .release        = seq_release_net,
2187 };
2188
2189 #endif
2190
2191 static struct net_proto_family unix_family_ops = {
2192         .family = PF_UNIX,
2193         .create = unix_create,
2194         .owner  = THIS_MODULE,
2195 };
2196
2197
2198 static int unix_net_init(struct net *net)
2199 {
2200         int error = -ENOMEM;
2201
2202         net->unx.sysctl_max_dgram_qlen = 10;
2203         if (unix_sysctl_register(net))
2204                 goto out;
2205
2206 #ifdef CONFIG_PROC_FS
2207         if (!proc_net_fops_create(net, "unix", 0, &unix_seq_fops)) {
2208                 unix_sysctl_unregister(net);
2209                 goto out;
2210         }
2211 #endif
2212         error = 0;
2213 out:
2214         return 0;
2215 }
2216
2217 static void unix_net_exit(struct net *net)
2218 {
2219         unix_sysctl_unregister(net);
2220         proc_net_remove(net, "unix");
2221 }
2222
2223 static struct pernet_operations unix_net_ops = {
2224         .init = unix_net_init,
2225         .exit = unix_net_exit,
2226 };
2227
2228 static int __init af_unix_init(void)
2229 {
2230         int rc = -1;
2231         struct sk_buff *dummy_skb;
2232
2233         BUILD_BUG_ON(sizeof(struct unix_skb_parms) > sizeof(dummy_skb->cb));
2234
2235         rc = proto_register(&unix_proto, 1);
2236         if (rc != 0) {
2237                 printk(KERN_CRIT "%s: Cannot create unix_sock SLAB cache!\n",
2238                        __func__);
2239                 goto out;
2240         }
2241
2242         sock_register(&unix_family_ops);
2243         register_pernet_subsys(&unix_net_ops);
2244 out:
2245         return rc;
2246 }
2247
2248 static void __exit af_unix_exit(void)
2249 {
2250         sock_unregister(PF_UNIX);
2251         proto_unregister(&unix_proto);
2252         unregister_pernet_subsys(&unix_net_ops);
2253 }
2254
2255 /* Earlier than device_initcall() so that other drivers invoking
2256    request_module() don't end up in a loop when modprobe tries
2257    to use a UNIX socket. But later than subsys_initcall() because
2258    we depend on stuff initialised there */
2259 fs_initcall(af_unix_init);
2260 module_exit(af_unix_exit);
2261
2262 MODULE_LICENSE("GPL");
2263 MODULE_ALIAS_NETPROTO(PF_UNIX);