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         WARN_ON(!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         WARN_ON(atomic_read(&sk->sk_wmem_alloc));
 354         WARN_ON(!sk_unhashed(sk));
 355         WARN_ON(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_long_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         if (other->sk_shutdown & RCV_SHUTDOWN)
1065                 goto out_unlock;
1066
1067         if (unix_recvq_full(other)) {
1068                 err = -EAGAIN;
1069                 if (!timeo)
1070                         goto out_unlock;
1071
1072                 timeo = unix_wait_for_peer(other, timeo);
1073
1074                 err = sock_intr_errno(timeo);
1075                 if (signal_pending(current))
1076                         goto out;
1077                 sock_put(other);
1078                 goto restart;
1079         }
1080
1081         /* Latch our state.
1082
1083            It is tricky place. We need to grab write lock and cannot
1084            drop lock on peer. It is dangerous because deadlock is
1085            possible. Connect to self case and simultaneous
1086            attempt to connect are eliminated by checking socket
1087            state. other is TCP_LISTEN, if sk is TCP_LISTEN we
1088            check this before attempt to grab lock.
1089
1090            Well, and we have to recheck the state after socket locked.
1091          */
1092         st = sk->sk_state;
1093
1094         switch (st) {
1095         case TCP_CLOSE:
1096                 /* This is ok... continue with connect */
1097                 break;
1098         case TCP_ESTABLISHED:
1099                 /* Socket is already connected */
1100                 err = -EISCONN;
1101                 goto out_unlock;
1102         default:
1103                 err = -EINVAL;
1104                 goto out_unlock;
1105         }
1106
1107         unix_state_lock_nested(sk);
1108
1109         if (sk->sk_state != st) {
1110                 unix_state_unlock(sk);
1111                 unix_state_unlock(other);
1112                 sock_put(other);
1113                 goto restart;
1114         }
1115
1116         err = security_unix_stream_connect(sock, other->sk_socket, newsk);
1117         if (err) {
1118                 unix_state_unlock(sk);
1119                 goto out_unlock;
1120         }
1121
1122         /* The way is open! Fastly set all the necessary fields... */
1123
1124         sock_hold(sk);
1125         unix_peer(newsk)        = sk;
1126         newsk->sk_state         = TCP_ESTABLISHED;
1127         newsk->sk_type          = sk->sk_type;
1128         newsk->sk_peercred.pid  = task_tgid_vnr(current);
1129         newsk->sk_peercred.uid  = current->euid;
1130         newsk->sk_peercred.gid  = current->egid;
1131         newu = unix_sk(newsk);
1132         newsk->sk_sleep         = &newu->peer_wait;
1133         otheru = unix_sk(other);
1134
1135         /* copy address information from listening to new sock*/
1136         if (otheru->addr) {
1137                 atomic_inc(&otheru->addr->refcnt);
1138                 newu->addr = otheru->addr;
1139         }
1140         if (otheru->dentry) {
1141                 newu->dentry    = dget(otheru->dentry);
1142                 newu->mnt       = mntget(otheru->mnt);
1143         }
1144
1145         /* Set credentials */
1146         sk->sk_peercred = other->sk_peercred;
1147
1148         sock->state     = SS_CONNECTED;
1149         sk->sk_state    = TCP_ESTABLISHED;
1150         sock_hold(newsk);
1151
1152         smp_mb__after_atomic_inc();     /* sock_hold() does an atomic_inc() */
1153         unix_peer(sk)   = newsk;
1154
1155         unix_state_unlock(sk);
1156
1157         /* take ten and and send info to listening sock */
1158         spin_lock(&other->sk_receive_queue.lock);
1159         __skb_queue_tail(&other->sk_receive_queue, skb);
1160         spin_unlock(&other->sk_receive_queue.lock);
1161         unix_state_unlock(other);
1162         other->sk_data_ready(other, 0);
1163         sock_put(other);
1164         return 0;
1165
1166 out_unlock:
1167         if (other)
1168                 unix_state_unlock(other);
1169
1170 out:
1171         if (skb)
1172                 kfree_skb(skb);
1173         if (newsk)
1174                 unix_release_sock(newsk, 0);
1175         if (other)
1176                 sock_put(other);
1177         return err;
1178 }
1179
1180 static int unix_socketpair(struct socket *socka, struct socket *sockb)
1181 {
1182         struct sock *ska=socka->sk, *skb = sockb->sk;
1183
1184         /* Join our sockets back to back */
1185         sock_hold(ska);
1186         sock_hold(skb);
1187         unix_peer(ska)=skb;
1188         unix_peer(skb)=ska;
1189         ska->sk_peercred.pid = skb->sk_peercred.pid = task_tgid_vnr(current);
1190         ska->sk_peercred.uid = skb->sk_peercred.uid = current->euid;
1191         ska->sk_peercred.gid = skb->sk_peercred.gid = current->egid;
1192
1193         if (ska->sk_type != SOCK_DGRAM) {
1194                 ska->sk_state = TCP_ESTABLISHED;
1195                 skb->sk_state = TCP_ESTABLISHED;
1196                 socka->state  = SS_CONNECTED;
1197                 sockb->state  = SS_CONNECTED;
1198         }
1199         return 0;
1200 }
1201
1202 static int unix_accept(struct socket *sock, struct socket *newsock, int flags)
1203 {
1204         struct sock *sk = sock->sk;
1205         struct sock *tsk;
1206         struct sk_buff *skb;
1207         int err;
1208
1209         err = -EOPNOTSUPP;
1210         if (sock->type!=SOCK_STREAM && sock->type!=SOCK_SEQPACKET)
1211                 goto out;
1212
1213         err = -EINVAL;
1214         if (sk->sk_state != TCP_LISTEN)
1215                 goto out;
1216
1217         /* If socket state is TCP_LISTEN it cannot change (for now...),
1218          * so that no locks are necessary.
1219          */
1220
1221         skb = skb_recv_datagram(sk, 0, flags&O_NONBLOCK, &err);
1222         if (!skb) {
1223                 /* This means receive shutdown. */
1224                 if (err == 0)
1225                         err = -EINVAL;
1226                 goto out;
1227         }
1228
1229         tsk = skb->sk;
1230         skb_free_datagram(sk, skb);
1231         wake_up_interruptible(&unix_sk(sk)->peer_wait);
1232
1233         /* attach accepted sock to socket */
1234         unix_state_lock(tsk);
1235         newsock->state = SS_CONNECTED;
1236         sock_graft(tsk, newsock);
1237         unix_state_unlock(tsk);
1238         return 0;
1239
1240 out:
1241         return err;
1242 }
1243
1244
1245 static int unix_getname(struct socket *sock, struct sockaddr *uaddr, int *uaddr_len, int peer)
1246 {
1247         struct sock *sk = sock->sk;
1248         struct unix_sock *u;
1249         struct sockaddr_un *sunaddr=(struct sockaddr_un *)uaddr;
1250         int err = 0;
1251
1252         if (peer) {
1253                 sk = unix_peer_get(sk);
1254
1255                 err = -ENOTCONN;
1256                 if (!sk)
1257                         goto out;
1258                 err = 0;
1259         } else {
1260                 sock_hold(sk);
1261         }
1262
1263         u = unix_sk(sk);
1264         unix_state_lock(sk);
1265         if (!u->addr) {
1266                 sunaddr->sun_family = AF_UNIX;
1267                 sunaddr->sun_path[0] = 0;
1268                 *uaddr_len = sizeof(short);
1269         } else {
1270                 struct unix_address *addr = u->addr;
1271
1272                 *uaddr_len = addr->len;
1273                 memcpy(sunaddr, addr->name, *uaddr_len);
1274         }
1275         unix_state_unlock(sk);
1276         sock_put(sk);
1277 out:
1278         return err;
1279 }
1280
1281 static void unix_detach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1282 {
1283         int i;
1284
1285         scm->fp = UNIXCB(skb).fp;
1286         skb->destructor = sock_wfree;
1287         UNIXCB(skb).fp = NULL;
1288
1289         for (i=scm->fp->count-1; i>=0; i--)
1290                 unix_notinflight(scm->fp->fp[i]);
1291 }
1292
1293 static void unix_destruct_fds(struct sk_buff *skb)
1294 {
1295         struct scm_cookie scm;
1296         memset(&scm, 0, sizeof(scm));
1297         unix_detach_fds(&scm, skb);
1298
1299         /* Alas, it calls VFS */
1300         /* So fscking what? fput() had been SMP-safe since the last Summer */
1301         scm_destroy(&scm);
1302         sock_wfree(skb);
1303 }
1304
1305 static int unix_attach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1306 {
1307         int i;
1308
1309         /*
1310          * Need to duplicate file references for the sake of garbage
1311          * collection.  Otherwise a socket in the fps might become a
1312          * candidate for GC while the skb is not yet queued.
1313          */
1314         UNIXCB(skb).fp = scm_fp_dup(scm->fp);
1315         if (!UNIXCB(skb).fp)
1316                 return -ENOMEM;
1317
1318         for (i=scm->fp->count-1; i>=0; i--)
1319                 unix_inflight(scm->fp->fp[i]);
1320         skb->destructor = unix_destruct_fds;
1321         return 0;
1322 }
1323
1324 /*
1325  *      Send AF_UNIX data.
1326  */
1327
1328 static int unix_dgram_sendmsg(struct kiocb *kiocb, struct socket *sock,
1329                               struct msghdr *msg, size_t len)
1330 {
1331         struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1332         struct sock *sk = sock->sk;
1333         struct net *net = sock_net(sk);
1334         struct unix_sock *u = unix_sk(sk);
1335         struct sockaddr_un *sunaddr=msg->msg_name;
1336         struct sock *other = NULL;
1337         int namelen = 0; /* fake GCC */
1338         int err;
1339         unsigned hash;
1340         struct sk_buff *skb;
1341         long timeo;
1342         struct scm_cookie tmp_scm;
1343
1344         if (NULL == siocb->scm)
1345                 siocb->scm = &tmp_scm;
1346         wait_for_unix_gc();
1347         err = scm_send(sock, msg, siocb->scm);
1348         if (err < 0)
1349                 return err;
1350
1351         err = -EOPNOTSUPP;
1352         if (msg->msg_flags&MSG_OOB)
1353                 goto out;
1354
1355         if (msg->msg_namelen) {
1356                 err = unix_mkname(sunaddr, msg->msg_namelen, &hash);
1357                 if (err < 0)
1358                         goto out;
1359                 namelen = err;
1360         } else {
1361                 sunaddr = NULL;
1362                 err = -ENOTCONN;
1363                 other = unix_peer_get(sk);
1364                 if (!other)
1365                         goto out;
1366         }
1367
1368         if (test_bit(SOCK_PASSCRED, &sock->flags)
1369                 && !u->addr && (err = unix_autobind(sock)) != 0)
1370                 goto out;
1371
1372         err = -EMSGSIZE;
1373         if (len > sk->sk_sndbuf - 32)
1374                 goto out;
1375
1376         skb = sock_alloc_send_skb(sk, len, msg->msg_flags&MSG_DONTWAIT, &err);
1377         if (skb==NULL)
1378                 goto out;
1379
1380         memcpy(UNIXCREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1381         if (siocb->scm->fp) {
1382                 err = unix_attach_fds(siocb->scm, skb);
1383                 if (err)
1384                         goto out_free;
1385         }
1386         unix_get_secdata(siocb->scm, skb);
1387
1388         skb_reset_transport_header(skb);
1389         err = memcpy_fromiovec(skb_put(skb,len), msg->msg_iov, len);
1390         if (err)
1391                 goto out_free;
1392
1393         timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1394
1395 restart:
1396         if (!other) {
1397                 err = -ECONNRESET;
1398                 if (sunaddr == NULL)
1399                         goto out_free;
1400
1401                 other = unix_find_other(net, sunaddr, namelen, sk->sk_type,
1402                                         hash, &err);
1403                 if (other==NULL)
1404                         goto out_free;
1405         }
1406
1407         unix_state_lock(other);
1408         err = -EPERM;
1409         if (!unix_may_send(sk, other))
1410                 goto out_unlock;
1411
1412         if (sock_flag(other, SOCK_DEAD)) {
1413                 /*
1414                  *      Check with 1003.1g - what should
1415                  *      datagram error
1416                  */
1417                 unix_state_unlock(other);
1418                 sock_put(other);
1419
1420                 err = 0;
1421                 unix_state_lock(sk);
1422                 if (unix_peer(sk) == other) {
1423                         unix_peer(sk)=NULL;
1424                         unix_state_unlock(sk);
1425
1426                         unix_dgram_disconnected(sk, other);
1427                         sock_put(other);
1428                         err = -ECONNREFUSED;
1429                 } else {
1430                         unix_state_unlock(sk);
1431                 }
1432
1433                 other = NULL;
1434                 if (err)
1435                         goto out_free;
1436                 goto restart;
1437         }
1438
1439         err = -EPIPE;
1440         if (other->sk_shutdown & RCV_SHUTDOWN)
1441                 goto out_unlock;
1442
1443         if (sk->sk_type != SOCK_SEQPACKET) {
1444                 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1445                 if (err)
1446                         goto out_unlock;
1447         }
1448
1449         if (unix_peer(other) != sk && unix_recvq_full(other)) {
1450                 if (!timeo) {
1451                         err = -EAGAIN;
1452                         goto out_unlock;
1453                 }
1454
1455                 timeo = unix_wait_for_peer(other, timeo);
1456
1457                 err = sock_intr_errno(timeo);
1458                 if (signal_pending(current))
1459                         goto out_free;
1460
1461                 goto restart;
1462         }
1463
1464         skb_queue_tail(&other->sk_receive_queue, skb);
1465         unix_state_unlock(other);
1466         other->sk_data_ready(other, len);
1467         sock_put(other);
1468         scm_destroy(siocb->scm);
1469         return len;
1470
1471 out_unlock:
1472         unix_state_unlock(other);
1473 out_free:
1474         kfree_skb(skb);
1475 out:
1476         if (other)
1477                 sock_put(other);
1478         scm_destroy(siocb->scm);
1479         return err;
1480 }
1481
1482
1483 static int unix_stream_sendmsg(struct kiocb *kiocb, struct socket *sock,
1484                                struct msghdr *msg, size_t len)
1485 {
1486         struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1487         struct sock *sk = sock->sk;
1488         struct sock *other = NULL;
1489         struct sockaddr_un *sunaddr=msg->msg_name;
1490         int err,size;
1491         struct sk_buff *skb;
1492         int sent=0;
1493         struct scm_cookie tmp_scm;
1494
1495         if (NULL == siocb->scm)
1496                 siocb->scm = &tmp_scm;
1497         wait_for_unix_gc();
1498         err = scm_send(sock, msg, siocb->scm);
1499         if (err < 0)
1500                 return err;
1501
1502         err = -EOPNOTSUPP;
1503         if (msg->msg_flags&MSG_OOB)
1504                 goto out_err;
1505
1506         if (msg->msg_namelen) {
1507                 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
1508                 goto out_err;
1509         } else {
1510                 sunaddr = NULL;
1511                 err = -ENOTCONN;
1512                 other = unix_peer(sk);
1513                 if (!other)
1514                         goto out_err;
1515         }
1516
1517         if (sk->sk_shutdown & SEND_SHUTDOWN)
1518                 goto pipe_err;
1519
1520         while(sent < len)
1521         {
1522                 /*
1523                  *      Optimisation for the fact that under 0.01% of X
1524                  *      messages typically need breaking up.
1525                  */
1526
1527                 size = len-sent;
1528
1529                 /* Keep two messages in the pipe so it schedules better */
1530                 if (size > ((sk->sk_sndbuf >> 1) - 64))
1531                         size = (sk->sk_sndbuf >> 1) - 64;
1532
1533                 if (size > SKB_MAX_ALLOC)
1534                         size = SKB_MAX_ALLOC;
1535
1536                 /*
1537                  *      Grab a buffer
1538                  */
1539
1540                 skb=sock_alloc_send_skb(sk,size,msg->msg_flags&MSG_DONTWAIT, &err);
1541
1542                 if (skb==NULL)
1543                         goto out_err;
1544
1545                 /*
1546                  *      If you pass two values to the sock_alloc_send_skb
1547                  *      it tries to grab the large buffer with GFP_NOFS
1548                  *      (which can fail easily), and if it fails grab the
1549                  *      fallback size buffer which is under a page and will
1550                  *      succeed. [Alan]
1551                  */
1552                 size = min_t(int, size, skb_tailroom(skb));
1553
1554                 memcpy(UNIXCREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1555                 if (siocb->scm->fp) {
1556                         err = unix_attach_fds(siocb->scm, skb);
1557                         if (err) {
1558                                 kfree_skb(skb);
1559                                 goto out_err;
1560                         }
1561                 }
1562
1563                 if ((err = memcpy_fromiovec(skb_put(skb,size), msg->msg_iov, size)) != 0) {
1564                         kfree_skb(skb);
1565                         goto out_err;
1566                 }
1567
1568                 unix_state_lock(other);
1569
1570                 if (sock_flag(other, SOCK_DEAD) ||
1571                     (other->sk_shutdown & RCV_SHUTDOWN))
1572                         goto pipe_err_free;
1573
1574                 skb_queue_tail(&other->sk_receive_queue, skb);
1575                 unix_state_unlock(other);
1576                 other->sk_data_ready(other, size);
1577                 sent+=size;
1578         }
1579
1580         scm_destroy(siocb->scm);
1581         siocb->scm = NULL;
1582
1583         return sent;
1584
1585 pipe_err_free:
1586         unix_state_unlock(other);
1587         kfree_skb(skb);
1588 pipe_err:
1589         if (sent==0 && !(msg->msg_flags&MSG_NOSIGNAL))
1590                 send_sig(SIGPIPE,current,0);
1591         err = -EPIPE;
1592 out_err:
1593         scm_destroy(siocb->scm);
1594         siocb->scm = NULL;
1595         return sent ? : err;
1596 }
1597
1598 static int unix_seqpacket_sendmsg(struct kiocb *kiocb, struct socket *sock,
1599                                   struct msghdr *msg, size_t len)
1600 {
1601         int err;
1602         struct sock *sk = sock->sk;
1603
1604         err = sock_error(sk);
1605         if (err)
1606                 return err;
1607
1608         if (sk->sk_state != TCP_ESTABLISHED)
1609                 return -ENOTCONN;
1610
1611         if (msg->msg_namelen)
1612                 msg->msg_namelen = 0;
1613
1614         return unix_dgram_sendmsg(kiocb, sock, msg, len);
1615 }
1616
1617 static void unix_copy_addr(struct msghdr *msg, struct sock *sk)
1618 {
1619         struct unix_sock *u = unix_sk(sk);
1620
1621         msg->msg_namelen = 0;
1622         if (u->addr) {
1623                 msg->msg_namelen = u->addr->len;
1624                 memcpy(msg->msg_name, u->addr->name, u->addr->len);
1625         }
1626 }
1627
1628 static int unix_dgram_recvmsg(struct kiocb *iocb, struct socket *sock,
1629                               struct msghdr *msg, size_t size,
1630                               int flags)
1631 {
1632         struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1633         struct scm_cookie tmp_scm;
1634         struct sock *sk = sock->sk;
1635         struct unix_sock *u = unix_sk(sk);
1636         int noblock = flags & MSG_DONTWAIT;
1637         struct sk_buff *skb;
1638         int err;
1639
1640         err = -EOPNOTSUPP;
1641         if (flags&MSG_OOB)
1642                 goto out;
1643
1644         msg->msg_namelen = 0;
1645
1646         mutex_lock(&u->readlock);
1647
1648         skb = skb_recv_datagram(sk, flags, noblock, &err);
1649         if (!skb) {
1650                 unix_state_lock(sk);
1651                 /* Signal EOF on disconnected non-blocking SEQPACKET socket. */
1652                 if (sk->sk_type == SOCK_SEQPACKET && err == -EAGAIN &&
1653                     (sk->sk_shutdown & RCV_SHUTDOWN))
1654                         err = 0;
1655                 unix_state_unlock(sk);
1656                 goto out_unlock;
1657         }
1658
1659         wake_up_interruptible_sync(&u->peer_wait);
1660
1661         if (msg->msg_name)
1662                 unix_copy_addr(msg, skb->sk);
1663
1664         if (size > skb->len)
1665                 size = skb->len;
1666         else if (size < skb->len)
1667                 msg->msg_flags |= MSG_TRUNC;
1668
1669         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, size);
1670         if (err)
1671                 goto out_free;
1672
1673         if (!siocb->scm) {
1674                 siocb->scm = &tmp_scm;
1675                 memset(&tmp_scm, 0, sizeof(tmp_scm));
1676         }
1677         siocb->scm->creds = *UNIXCREDS(skb);
1678         unix_set_secdata(siocb->scm, skb);
1679
1680         if (!(flags & MSG_PEEK))
1681         {
1682                 if (UNIXCB(skb).fp)
1683                         unix_detach_fds(siocb->scm, skb);
1684         }
1685         else
1686         {
1687                 /* It is questionable: on PEEK we could:
1688                    - do not return fds - good, but too simple 8)
1689                    - return fds, and do not return them on read (old strategy,
1690                      apparently wrong)
1691                    - clone fds (I chose it for now, it is the most universal
1692                      solution)
1693
1694                    POSIX 1003.1g does not actually define this clearly
1695                    at all. POSIX 1003.1g doesn't define a lot of things
1696                    clearly however!
1697
1698                 */
1699                 if (UNIXCB(skb).fp)
1700                         siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1701         }
1702         err = size;
1703
1704         scm_recv(sock, msg, siocb->scm, flags);
1705
1706 out_free:
1707         skb_free_datagram(sk,skb);
1708 out_unlock:
1709         mutex_unlock(&u->readlock);
1710 out:
1711         return err;
1712 }
1713
1714 /*
1715  *      Sleep until data has arrive. But check for races..
1716  */
1717
1718 static long unix_stream_data_wait(struct sock * sk, long timeo)
1719 {
1720         DEFINE_WAIT(wait);
1721
1722         unix_state_lock(sk);
1723
1724         for (;;) {
1725                 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1726
1727                 if (!skb_queue_empty(&sk->sk_receive_queue) ||
1728                     sk->sk_err ||
1729                     (sk->sk_shutdown & RCV_SHUTDOWN) ||
1730                     signal_pending(current) ||
1731                     !timeo)
1732                         break;
1733
1734                 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1735                 unix_state_unlock(sk);
1736                 timeo = schedule_timeout(timeo);
1737                 unix_state_lock(sk);
1738                 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1739         }
1740
1741         finish_wait(sk->sk_sleep, &wait);
1742         unix_state_unlock(sk);
1743         return timeo;
1744 }
1745
1746
1747
1748 static int unix_stream_recvmsg(struct kiocb *iocb, struct socket *sock,
1749                                struct msghdr *msg, size_t size,
1750                                int flags)
1751 {
1752         struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1753         struct scm_cookie tmp_scm;
1754         struct sock *sk = sock->sk;
1755         struct unix_sock *u = unix_sk(sk);
1756         struct sockaddr_un *sunaddr=msg->msg_name;
1757         int copied = 0;
1758         int check_creds = 0;
1759         int target;
1760         int err = 0;
1761         long timeo;
1762
1763         err = -EINVAL;
1764         if (sk->sk_state != TCP_ESTABLISHED)
1765                 goto out;
1766
1767         err = -EOPNOTSUPP;
1768         if (flags&MSG_OOB)
1769                 goto out;
1770
1771         target = sock_rcvlowat(sk, flags&MSG_WAITALL, size);
1772         timeo = sock_rcvtimeo(sk, flags&MSG_DONTWAIT);
1773
1774         msg->msg_namelen = 0;
1775
1776         /* Lock the socket to prevent queue disordering
1777          * while sleeps in memcpy_tomsg
1778          */
1779
1780         if (!siocb->scm) {
1781                 siocb->scm = &tmp_scm;
1782                 memset(&tmp_scm, 0, sizeof(tmp_scm));
1783         }
1784
1785         mutex_lock(&u->readlock);
1786
1787         do
1788         {
1789                 int chunk;
1790                 struct sk_buff *skb;
1791
1792                 unix_state_lock(sk);
1793                 skb = skb_dequeue(&sk->sk_receive_queue);
1794                 if (skb==NULL)
1795                 {
1796                         if (copied >= target)
1797                                 goto unlock;
1798
1799                         /*
1800                          *      POSIX 1003.1g mandates this order.
1801                          */
1802
1803                         if ((err = sock_error(sk)) != 0)
1804                                 goto unlock;
1805                         if (sk->sk_shutdown & RCV_SHUTDOWN)
1806                                 goto unlock;
1807
1808                         unix_state_unlock(sk);
1809                         err = -EAGAIN;
1810                         if (!timeo)
1811                                 break;
1812                         mutex_unlock(&u->readlock);
1813
1814                         timeo = unix_stream_data_wait(sk, timeo);
1815
1816                         if (signal_pending(current)) {
1817                                 err = sock_intr_errno(timeo);
1818                                 goto out;
1819                         }
1820                         mutex_lock(&u->readlock);
1821                         continue;
1822  unlock:
1823                         unix_state_unlock(sk);
1824                         break;
1825                 }
1826                 unix_state_unlock(sk);
1827
1828                 if (check_creds) {
1829                         /* Never glue messages from different writers */
1830                         if (memcmp(UNIXCREDS(skb), &siocb->scm->creds, sizeof(siocb->scm->creds)) != 0) {
1831                                 skb_queue_head(&sk->sk_receive_queue, skb);
1832                                 break;
1833                         }
1834                 } else {
1835                         /* Copy credentials */
1836                         siocb->scm->creds = *UNIXCREDS(skb);
1837                         check_creds = 1;
1838                 }
1839
1840                 /* Copy address just once */
1841                 if (sunaddr)
1842                 {
1843                         unix_copy_addr(msg, skb->sk);
1844                         sunaddr = NULL;
1845                 }
1846
1847                 chunk = min_t(unsigned int, skb->len, size);
1848                 if (memcpy_toiovec(msg->msg_iov, skb->data, chunk)) {
1849                         skb_queue_head(&sk->sk_receive_queue, skb);
1850                         if (copied == 0)
1851                                 copied = -EFAULT;
1852                         break;
1853                 }
1854                 copied += chunk;
1855                 size -= chunk;
1856
1857                 /* Mark read part of skb as used */
1858                 if (!(flags & MSG_PEEK))
1859                 {
1860                         skb_pull(skb, chunk);
1861
1862                         if (UNIXCB(skb).fp)
1863                                 unix_detach_fds(siocb->scm, skb);
1864
1865                         /* put the skb back if we didn't use it up.. */
1866                         if (skb->len)
1867                         {
1868                                 skb_queue_head(&sk->sk_receive_queue, skb);
1869                                 break;
1870                         }
1871
1872                         kfree_skb(skb);
1873
1874                         if (siocb->scm->fp)
1875                                 break;
1876                 }
1877                 else
1878                 {
1879                         /* It is questionable, see note in unix_dgram_recvmsg.
1880                          */
1881                         if (UNIXCB(skb).fp)
1882                                 siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1883
1884                         /* put message back and return */
1885                         skb_queue_head(&sk->sk_receive_queue, skb);
1886                         break;
1887                 }
1888         } while (size);
1889
1890         mutex_unlock(&u->readlock);
1891         scm_recv(sock, msg, siocb->scm, flags);
1892 out:
1893         return copied ? : err;
1894 }
1895
1896 static int unix_shutdown(struct socket *sock, int mode)
1897 {
1898         struct sock *sk = sock->sk;
1899         struct sock *other;
1900
1901         mode = (mode+1)&(RCV_SHUTDOWN|SEND_SHUTDOWN);
1902
1903         if (mode) {
1904                 unix_state_lock(sk);
1905                 sk->sk_shutdown |= mode;
1906                 other=unix_peer(sk);
1907                 if (other)
1908                         sock_hold(other);
1909                 unix_state_unlock(sk);
1910                 sk->sk_state_change(sk);
1911
1912                 if (other &&
1913                         (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET)) {
1914
1915                         int peer_mode = 0;
1916
1917                         if (mode&RCV_SHUTDOWN)
1918                                 peer_mode |= SEND_SHUTDOWN;
1919                         if (mode&SEND_SHUTDOWN)
1920                                 peer_mode |= RCV_SHUTDOWN;
1921                         unix_state_lock(other);
1922                         other->sk_shutdown |= peer_mode;
1923                         unix_state_unlock(other);
1924                         other->sk_state_change(other);
1925                         read_lock(&other->sk_callback_lock);
1926                         if (peer_mode == SHUTDOWN_MASK)
1927                                 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_HUP);
1928                         else if (peer_mode & RCV_SHUTDOWN)
1929                                 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_IN);
1930                         read_unlock(&other->sk_callback_lock);
1931                 }
1932                 if (other)
1933                         sock_put(other);
1934         }
1935         return 0;
1936 }
1937
1938 static int unix_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1939 {
1940         struct sock *sk = sock->sk;
1941         long amount=0;
1942         int err;
1943
1944         switch(cmd)
1945         {
1946                 case SIOCOUTQ:
1947                         amount = atomic_read(&sk->sk_wmem_alloc);
1948                         err = put_user(amount, (int __user *)arg);
1949                         break;
1950                 case SIOCINQ:
1951                 {
1952                         struct sk_buff *skb;
1953
1954                         if (sk->sk_state == TCP_LISTEN) {
1955                                 err = -EINVAL;
1956                                 break;
1957                         }
1958
1959                         spin_lock(&sk->sk_receive_queue.lock);
1960                         if (sk->sk_type == SOCK_STREAM ||
1961                             sk->sk_type == SOCK_SEQPACKET) {
1962                                 skb_queue_walk(&sk->sk_receive_queue, skb)
1963                                         amount += skb->len;
1964                         } else {
1965                                 skb = skb_peek(&sk->sk_receive_queue);
1966                                 if (skb)
1967                                         amount=skb->len;
1968                         }
1969                         spin_unlock(&sk->sk_receive_queue.lock);
1970                         err = put_user(amount, (int __user *)arg);
1971                         break;
1972                 }
1973
1974                 default:
1975                         err = -ENOIOCTLCMD;
1976                         break;
1977         }
1978         return err;
1979 }
1980
1981 static unsigned int unix_poll(struct file * file, struct socket *sock, poll_table *wait)
1982 {
1983         struct sock *sk = sock->sk;
1984         unsigned int mask;
1985
1986         poll_wait(file, sk->sk_sleep, wait);
1987         mask = 0;
1988
1989         /* exceptional events? */
1990         if (sk->sk_err)
1991                 mask |= POLLERR;
1992         if (sk->sk_shutdown == SHUTDOWN_MASK)
1993                 mask |= POLLHUP;
1994         if (sk->sk_shutdown & RCV_SHUTDOWN)
1995                 mask |= POLLRDHUP;
1996
1997         /* readable? */
1998         if (!skb_queue_empty(&sk->sk_receive_queue) ||
1999             (sk->sk_shutdown & RCV_SHUTDOWN))
2000                 mask |= POLLIN | POLLRDNORM;
2001
2002         /* Connection-based need to check for termination and startup */
2003         if ((sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) && sk->sk_state == TCP_CLOSE)
2004                 mask |= POLLHUP;
2005
2006         /*
2007          * we set writable also when the other side has shut down the
2008          * connection. This prevents stuck sockets.
2009          */
2010         if (unix_writable(sk))
2011                 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2012
2013         return mask;
2014 }
2015
2016 static unsigned int unix_dgram_poll(struct file *file, struct socket *sock,
2017                                     poll_table *wait)
2018 {
2019         struct sock *sk = sock->sk, *other;
2020         unsigned int mask, writable;
2021
2022         poll_wait(file, sk->sk_sleep, wait);
2023         mask = 0;
2024
2025         /* exceptional events? */
2026         if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue))
2027                 mask |= POLLERR;
2028         if (sk->sk_shutdown & RCV_SHUTDOWN)
2029                 mask |= POLLRDHUP;
2030         if (sk->sk_shutdown == SHUTDOWN_MASK)
2031                 mask |= POLLHUP;
2032
2033         /* readable? */
2034         if (!skb_queue_empty(&sk->sk_receive_queue) ||
2035             (sk->sk_shutdown & RCV_SHUTDOWN))
2036                 mask |= POLLIN | POLLRDNORM;
2037
2038         /* Connection-based need to check for termination and startup */
2039         if (sk->sk_type == SOCK_SEQPACKET) {
2040                 if (sk->sk_state == TCP_CLOSE)
2041                         mask |= POLLHUP;
2042                 /* connection hasn't started yet? */
2043                 if (sk->sk_state == TCP_SYN_SENT)
2044                         return mask;
2045         }
2046
2047         /* writable? */
2048         writable = unix_writable(sk);
2049         if (writable) {
2050                 other = unix_peer_get(sk);
2051                 if (other) {
2052                         if (unix_peer(other) != sk) {
2053                                 poll_wait(file, &unix_sk(other)->peer_wait,
2054                                           wait);
2055                                 if (unix_recvq_full(other))
2056                                         writable = 0;
2057                         }
2058
2059                         sock_put(other);
2060                 }
2061         }
2062
2063         if (writable)
2064                 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2065         else
2066                 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
2067
2068         return mask;
2069 }
2070
2071 #ifdef CONFIG_PROC_FS
2072 static struct sock *first_unix_socket(int *i)
2073 {
2074         for (*i = 0; *i <= UNIX_HASH_SIZE; (*i)++) {
2075                 if (!hlist_empty(&unix_socket_table[*i]))
2076                         return __sk_head(&unix_socket_table[*i]);
2077         }
2078         return NULL;
2079 }
2080
2081 static struct sock *next_unix_socket(int *i, struct sock *s)
2082 {
2083         struct sock *next = sk_next(s);
2084         /* More in this chain? */
2085         if (next)
2086                 return next;
2087         /* Look for next non-empty chain. */
2088         for ((*i)++; *i <= UNIX_HASH_SIZE; (*i)++) {
2089                 if (!hlist_empty(&unix_socket_table[*i]))
2090                         return __sk_head(&unix_socket_table[*i]);
2091         }
2092         return NULL;
2093 }
2094
2095 struct unix_iter_state {
2096         struct seq_net_private p;
2097         int i;
2098 };
2099 static struct sock *unix_seq_idx(struct seq_file *seq, loff_t pos)
2100 {
2101         struct unix_iter_state *iter = seq->private;
2102         loff_t off = 0;
2103         struct sock *s;
2104
2105         for (s = first_unix_socket(&iter->i); s; s = next_unix_socket(&iter->i, s)) {
2106                 if (sock_net(s) != seq_file_net(seq))
2107                         continue;
2108                 if (off == pos)
2109                         return s;
2110                 ++off;
2111         }
2112         return NULL;
2113 }
2114
2115
2116 static void *unix_seq_start(struct seq_file *seq, loff_t *pos)
2117         __acquires(unix_table_lock)
2118 {
2119         spin_lock(&unix_table_lock);
2120         return *pos ? unix_seq_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2121 }
2122
2123 static void *unix_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2124 {
2125         struct unix_iter_state *iter = seq->private;
2126         struct sock *sk = v;
2127         ++*pos;
2128
2129         if (v == SEQ_START_TOKEN)
2130                 sk = first_unix_socket(&iter->i);
2131         else
2132                 sk = next_unix_socket(&iter->i, sk);
2133         while (sk && (sock_net(sk) != seq_file_net(seq)))
2134                 sk = next_unix_socket(&iter->i, sk);
2135         return sk;
2136 }
2137
2138 static void unix_seq_stop(struct seq_file *seq, void *v)
2139         __releases(unix_table_lock)
2140 {
2141         spin_unlock(&unix_table_lock);
2142 }
2143
2144 static int unix_seq_show(struct seq_file *seq, void *v)
2145 {
2146
2147         if (v == SEQ_START_TOKEN)
2148                 seq_puts(seq, "Num       RefCount Protocol Flags    Type St "
2149                          "Inode Path\n");
2150         else {
2151                 struct sock *s = v;
2152                 struct unix_sock *u = unix_sk(s);
2153                 unix_state_lock(s);
2154
2155                 seq_printf(seq, "%p: %08X %08X %08X %04X %02X %5lu",
2156                         s,
2157                         atomic_read(&s->sk_refcnt),
2158                         0,
2159                         s->sk_state == TCP_LISTEN ? __SO_ACCEPTCON : 0,
2160                         s->sk_type,
2161                         s->sk_socket ?
2162                         (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTED : SS_UNCONNECTED) :
2163                         (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTING : SS_DISCONNECTING),
2164                         sock_i_ino(s));
2165
2166                 if (u->addr) {
2167                         int i, len;
2168                         seq_putc(seq, ' ');
2169
2170                         i = 0;
2171                         len = u->addr->len - sizeof(short);
2172                         if (!UNIX_ABSTRACT(s))
2173                                 len--;
2174                         else {
2175                                 seq_putc(seq, '@');
2176                                 i++;
2177                         }
2178                         for ( ; i < len; i++)
2179                                 seq_putc(seq, u->addr->name->sun_path[i]);
2180                 }
2181                 unix_state_unlock(s);
2182                 seq_putc(seq, '\n');
2183         }
2184
2185         return 0;
2186 }
2187
2188 static const struct seq_operations unix_seq_ops = {
2189         .start  = unix_seq_start,
2190         .next   = unix_seq_next,
2191         .stop   = unix_seq_stop,
2192         .show   = unix_seq_show,
2193 };
2194
2195
2196 static int unix_seq_open(struct inode *inode, struct file *file)
2197 {
2198         return seq_open_net(inode, file, &unix_seq_ops,
2199                             sizeof(struct unix_iter_state));
2200 }
2201
2202 static const struct file_operations unix_seq_fops = {
2203         .owner          = THIS_MODULE,
2204         .open           = unix_seq_open,
2205         .read           = seq_read,
2206         .llseek         = seq_lseek,
2207         .release        = seq_release_net,
2208 };
2209
2210 #endif
2211
2212 static struct net_proto_family unix_family_ops = {
2213         .family = PF_UNIX,
2214         .create = unix_create,
2215         .owner  = THIS_MODULE,
2216 };
2217
2218
2219 static int unix_net_init(struct net *net)
2220 {
2221         int error = -ENOMEM;
2222
2223         net->unx.sysctl_max_dgram_qlen = 10;
2224         if (unix_sysctl_register(net))
2225                 goto out;
2226
2227 #ifdef CONFIG_PROC_FS
2228         if (!proc_net_fops_create(net, "unix", 0, &unix_seq_fops)) {
2229                 unix_sysctl_unregister(net);
2230                 goto out;
2231         }
2232 #endif
2233         error = 0;
2234 out:
2235         return error;
2236 }
2237
2238 static void unix_net_exit(struct net *net)
2239 {
2240         unix_sysctl_unregister(net);
2241         proc_net_remove(net, "unix");
2242 }
2243
2244 static struct pernet_operations unix_net_ops = {
2245         .init = unix_net_init,
2246         .exit = unix_net_exit,
2247 };
2248
2249 static int __init af_unix_init(void)
2250 {
2251         int rc = -1;
2252         struct sk_buff *dummy_skb;
2253
2254         BUILD_BUG_ON(sizeof(struct unix_skb_parms) > sizeof(dummy_skb->cb));
2255
2256         rc = proto_register(&unix_proto, 1);
2257         if (rc != 0) {
2258                 printk(KERN_CRIT "%s: Cannot create unix_sock SLAB cache!\n",
2259                        __func__);
2260                 goto out;
2261         }
2262
2263         sock_register(&unix_family_ops);
2264         register_pernet_subsys(&unix_net_ops);
2265 out:
2266         return rc;
2267 }
2268
2269 static void __exit af_unix_exit(void)
2270 {
2271         sock_unregister(PF_UNIX);
2272         proto_unregister(&unix_proto);
2273         unregister_pernet_subsys(&unix_net_ops);
2274 }
2275
2276 /* Earlier than device_initcall() so that other drivers invoking
2277    request_module() don't end up in a loop when modprobe tries
2278    to use a UNIX socket. But later than subsys_initcall() because
2279    we depend on stuff initialised there */
2280 fs_initcall(af_unix_init);
2281 module_exit(af_unix_exit);
2282
2283 MODULE_LICENSE("GPL");
2284 MODULE_ALIAS_NETPROTO(PF_UNIX);