git.oblomov.eu Git - linux-2.6/blob - net/unix/af_unix.c

   1 /*
   2  * NET4:        Implementation of BSD Unix domain sockets.
   3  *
   4  * Authors:     Alan Cox, <alan@lxorguk.ukuu.org.uk>
   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         current_euid_egid(&sk->sk_peercred.uid, &sk->sk_peercred.gid);
 471         err = 0;
 472
 473 out_unlock:
 474         unix_state_unlock(sk);
 475 out:
 476         return err;
 477 }
 478
 479 static int unix_release(struct socket *);
 480 static int unix_bind(struct socket *, struct sockaddr *, int);
 481 static int unix_stream_connect(struct socket *, struct sockaddr *,
 482                                int addr_len, int flags);
 483 static int unix_socketpair(struct socket *, struct socket *);
 484 static int unix_accept(struct socket *, struct socket *, int);
 485 static int unix_getname(struct socket *, struct sockaddr *, int *, int);
 486 static unsigned int unix_poll(struct file *, struct socket *, poll_table *);
 487 static unsigned int unix_dgram_poll(struct file *, struct socket *,
 488                                     poll_table *);
 489 static int unix_ioctl(struct socket *, unsigned int, unsigned long);
 490 static int unix_shutdown(struct socket *, int);
 491 static int unix_stream_sendmsg(struct kiocb *, struct socket *,
 492                                struct msghdr *, size_t);
 493 static int unix_stream_recvmsg(struct kiocb *, struct socket *,
 494                                struct msghdr *, size_t, int);
 495 static int unix_dgram_sendmsg(struct kiocb *, struct socket *,
 496                               struct msghdr *, size_t);
 497 static int unix_dgram_recvmsg(struct kiocb *, struct socket *,
 498                               struct msghdr *, size_t, int);
 499 static int unix_dgram_connect(struct socket *, struct sockaddr *,
 500                               int, int);
 501 static int unix_seqpacket_sendmsg(struct kiocb *, struct socket *,
 502                                   struct msghdr *, size_t);
 503
 504 static const struct proto_ops unix_stream_ops = {
 505         .family =       PF_UNIX,
 506         .owner =        THIS_MODULE,
 507         .release =      unix_release,
 508         .bind =         unix_bind,
 509         .connect =      unix_stream_connect,
 510         .socketpair =   unix_socketpair,
 511         .accept =       unix_accept,
 512         .getname =      unix_getname,
 513         .poll =         unix_poll,
 514         .ioctl =        unix_ioctl,
 515         .listen =       unix_listen,
 516         .shutdown =     unix_shutdown,
 517         .setsockopt =   sock_no_setsockopt,
 518         .getsockopt =   sock_no_getsockopt,
 519         .sendmsg =      unix_stream_sendmsg,
 520         .recvmsg =      unix_stream_recvmsg,
 521         .mmap =         sock_no_mmap,
 522         .sendpage =     sock_no_sendpage,
 523 };
 524
 525 static const struct proto_ops unix_dgram_ops = {
 526         .family =       PF_UNIX,
 527         .owner =        THIS_MODULE,
 528         .release =      unix_release,
 529         .bind =         unix_bind,
 530         .connect =      unix_dgram_connect,
 531         .socketpair =   unix_socketpair,
 532         .accept =       sock_no_accept,
 533         .getname =      unix_getname,
 534         .poll =         unix_dgram_poll,
 535         .ioctl =        unix_ioctl,
 536         .listen =       sock_no_listen,
 537         .shutdown =     unix_shutdown,
 538         .setsockopt =   sock_no_setsockopt,
 539         .getsockopt =   sock_no_getsockopt,
 540         .sendmsg =      unix_dgram_sendmsg,
 541         .recvmsg =      unix_dgram_recvmsg,
 542         .mmap =         sock_no_mmap,
 543         .sendpage =     sock_no_sendpage,
 544 };
 545
 546 static const struct proto_ops unix_seqpacket_ops = {
 547         .family =       PF_UNIX,
 548         .owner =        THIS_MODULE,
 549         .release =      unix_release,
 550         .bind =         unix_bind,
 551         .connect =      unix_stream_connect,
 552         .socketpair =   unix_socketpair,
 553         .accept =       unix_accept,
 554         .getname =      unix_getname,
 555         .poll =         unix_dgram_poll,
 556         .ioctl =        unix_ioctl,
 557         .listen =       unix_listen,
 558         .shutdown =     unix_shutdown,
 559         .setsockopt =   sock_no_setsockopt,
 560         .getsockopt =   sock_no_getsockopt,
 561         .sendmsg =      unix_seqpacket_sendmsg,
 562         .recvmsg =      unix_dgram_recvmsg,
 563         .mmap =         sock_no_mmap,
 564         .sendpage =     sock_no_sendpage,
 565 };
 566
 567 static struct proto unix_proto = {
 568         .name     = "UNIX",
 569         .owner    = THIS_MODULE,
 570         .obj_size = sizeof(struct unix_sock),
 571 };
 572
 573 /*
 574  * AF_UNIX sockets do not interact with hardware, hence they
 575  * dont trigger interrupts - so it's safe for them to have
 576  * bh-unsafe locking for their sk_receive_queue.lock. Split off
 577  * this special lock-class by reinitializing the spinlock key:
 578  */
 579 static struct lock_class_key af_unix_sk_receive_queue_lock_key;
 580
 581 static struct sock * unix_create1(struct net *net, struct socket *sock)
 582 {
 583         struct sock *sk = NULL;
 584         struct unix_sock *u;
 585
 586         atomic_inc(&unix_nr_socks);
 587         if (atomic_read(&unix_nr_socks) > 2 * get_max_files())
 588                 goto out;
 589
 590         sk = sk_alloc(net, PF_UNIX, GFP_KERNEL, &unix_proto);
 591         if (!sk)
 592                 goto out;
 593
 594         sock_init_data(sock,sk);
 595         lockdep_set_class(&sk->sk_receive_queue.lock,
 596                                 &af_unix_sk_receive_queue_lock_key);
 597
 598         sk->sk_write_space      = unix_write_space;
 599         sk->sk_max_ack_backlog  = net->unx.sysctl_max_dgram_qlen;
 600         sk->sk_destruct         = unix_sock_destructor;
 601         u         = unix_sk(sk);
 602         u->dentry = NULL;
 603         u->mnt    = NULL;
 604         spin_lock_init(&u->lock);
 605         atomic_long_set(&u->inflight, 0);
 606         INIT_LIST_HEAD(&u->link);
 607         mutex_init(&u->readlock); /* single task reading lock */
 608         init_waitqueue_head(&u->peer_wait);
 609         unix_insert_socket(unix_sockets_unbound, sk);
 610 out:
 611         if (sk == NULL)
 612                 atomic_dec(&unix_nr_socks);
 613         return sk;
 614 }
 615
 616 static int unix_create(struct net *net, struct socket *sock, int protocol)
 617 {
 618         if (protocol && protocol != PF_UNIX)
 619                 return -EPROTONOSUPPORT;
 620
 621         sock->state = SS_UNCONNECTED;
 622
 623         switch (sock->type) {
 624         case SOCK_STREAM:
 625                 sock->ops = &unix_stream_ops;
 626                 break;
 627                 /*
 628                  *      Believe it or not BSD has AF_UNIX, SOCK_RAW though
 629                  *      nothing uses it.
 630                  */
 631         case SOCK_RAW:
 632                 sock->type=SOCK_DGRAM;
 633         case SOCK_DGRAM:
 634                 sock->ops = &unix_dgram_ops;
 635                 break;
 636         case SOCK_SEQPACKET:
 637                 sock->ops = &unix_seqpacket_ops;
 638                 break;
 639         default:
 640                 return -ESOCKTNOSUPPORT;
 641         }
 642
 643         return unix_create1(net, sock) ? 0 : -ENOMEM;
 644 }
 645
 646 static int unix_release(struct socket *sock)
 647 {
 648         struct sock *sk = sock->sk;
 649
 650         if (!sk)
 651                 return 0;
 652
 653         sock->sk = NULL;
 654
 655         return unix_release_sock (sk, 0);
 656 }
 657
 658 static int unix_autobind(struct socket *sock)
 659 {
 660         struct sock *sk = sock->sk;
 661         struct net *net = sock_net(sk);
 662         struct unix_sock *u = unix_sk(sk);
 663         static u32 ordernum = 1;
 664         struct unix_address * addr;
 665         int err;
 666
 667         mutex_lock(&u->readlock);
 668
 669         err = 0;
 670         if (u->addr)
 671                 goto out;
 672
 673         err = -ENOMEM;
 674         addr = kzalloc(sizeof(*addr) + sizeof(short) + 16, GFP_KERNEL);
 675         if (!addr)
 676                 goto out;
 677
 678         addr->name->sun_family = AF_UNIX;
 679         atomic_set(&addr->refcnt, 1);
 680
 681 retry:
 682         addr->len = sprintf(addr->name->sun_path+1, "%05x", ordernum) + 1 + sizeof(short);
 683         addr->hash = unix_hash_fold(csum_partial((void*)addr->name, addr->len, 0));
 684
 685         spin_lock(&unix_table_lock);
 686         ordernum = (ordernum+1)&0xFFFFF;
 687
 688         if (__unix_find_socket_byname(net, addr->name, addr->len, sock->type,
 689                                       addr->hash)) {
 690                 spin_unlock(&unix_table_lock);
 691                 /* Sanity yield. It is unusual case, but yet... */
 692                 if (!(ordernum&0xFF))
 693                         yield();
 694                 goto retry;
 695         }
 696         addr->hash ^= sk->sk_type;
 697
 698         __unix_remove_socket(sk);
 699         u->addr = addr;
 700         __unix_insert_socket(&unix_socket_table[addr->hash], sk);
 701         spin_unlock(&unix_table_lock);
 702         err = 0;
 703
 704 out:    mutex_unlock(&u->readlock);
 705         return err;
 706 }
 707
 708 static struct sock *unix_find_other(struct net *net,
 709                                     struct sockaddr_un *sunname, int len,
 710                                     int type, unsigned hash, int *error)
 711 {
 712         struct sock *u;
 713         struct path path;
 714         int err = 0;
 715
 716         if (sunname->sun_path[0]) {
 717                 struct inode *inode;
 718                 err = kern_path(sunname->sun_path, LOOKUP_FOLLOW, &path);
 719                 if (err)
 720                         goto fail;
 721                 inode = path.dentry->d_inode;
 722                 err = inode_permission(inode, MAY_WRITE);
 723                 if (err)
 724                         goto put_fail;
 725
 726                 err = -ECONNREFUSED;
 727                 if (!S_ISSOCK(inode->i_mode))
 728                         goto put_fail;
 729                 u = unix_find_socket_byinode(net, inode);
 730                 if (!u)
 731                         goto put_fail;
 732
 733                 if (u->sk_type == type)
 734                         touch_atime(path.mnt, path.dentry);
 735
 736                 path_put(&path);
 737
 738                 err=-EPROTOTYPE;
 739                 if (u->sk_type != type) {
 740                         sock_put(u);
 741                         goto fail;
 742                 }
 743         } else {
 744                 err = -ECONNREFUSED;
 745                 u=unix_find_socket_byname(net, sunname, len, type, hash);
 746                 if (u) {
 747                         struct dentry *dentry;
 748                         dentry = unix_sk(u)->dentry;
 749                         if (dentry)
 750                                 touch_atime(unix_sk(u)->mnt, dentry);
 751                 } else
 752                         goto fail;
 753         }
 754         return u;
 755
 756 put_fail:
 757         path_put(&path);
 758 fail:
 759         *error=err;
 760         return NULL;
 761 }
 762
 763
 764 static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
 765 {
 766         struct sock *sk = sock->sk;
 767         struct net *net = sock_net(sk);
 768         struct unix_sock *u = unix_sk(sk);
 769         struct sockaddr_un *sunaddr=(struct sockaddr_un *)uaddr;
 770         struct dentry * dentry = NULL;
 771         struct nameidata nd;
 772         int err;
 773         unsigned hash;
 774         struct unix_address *addr;
 775         struct hlist_head *list;
 776
 777         err = -EINVAL;
 778         if (sunaddr->sun_family != AF_UNIX)
 779                 goto out;
 780
 781         if (addr_len==sizeof(short)) {
 782                 err = unix_autobind(sock);
 783                 goto out;
 784         }
 785
 786         err = unix_mkname(sunaddr, addr_len, &hash);
 787         if (err < 0)
 788                 goto out;
 789         addr_len = err;
 790
 791         mutex_lock(&u->readlock);
 792
 793         err = -EINVAL;
 794         if (u->addr)
 795                 goto out_up;
 796
 797         err = -ENOMEM;
 798         addr = kmalloc(sizeof(*addr)+addr_len, GFP_KERNEL);
 799         if (!addr)
 800                 goto out_up;
 801
 802         memcpy(addr->name, sunaddr, addr_len);
 803         addr->len = addr_len;
 804         addr->hash = hash ^ sk->sk_type;
 805         atomic_set(&addr->refcnt, 1);
 806
 807         if (sunaddr->sun_path[0]) {
 808                 unsigned int mode;
 809                 err = 0;
 810                 /*
 811                  * Get the parent directory, calculate the hash for last
 812                  * component.
 813                  */
 814                 err = path_lookup(sunaddr->sun_path, LOOKUP_PARENT, &nd);
 815                 if (err)
 816                         goto out_mknod_parent;
 817
 818                 dentry = lookup_create(&nd, 0);
 819                 err = PTR_ERR(dentry);
 820                 if (IS_ERR(dentry))
 821                         goto out_mknod_unlock;
 822
 823                 /*
 824                  * All right, let's create it.
 825                  */
 826                 mode = S_IFSOCK |
 827                        (SOCK_INODE(sock)->i_mode & ~current->fs->umask);
 828                 err = mnt_want_write(nd.path.mnt);
 829                 if (err)
 830                         goto out_mknod_dput;
 831                 err = vfs_mknod(nd.path.dentry->d_inode, dentry, mode, 0);
 832                 mnt_drop_write(nd.path.mnt);
 833                 if (err)
 834                         goto out_mknod_dput;
 835                 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
 836                 dput(nd.path.dentry);
 837                 nd.path.dentry = dentry;
 838
 839                 addr->hash = UNIX_HASH_SIZE;
 840         }
 841
 842         spin_lock(&unix_table_lock);
 843
 844         if (!sunaddr->sun_path[0]) {
 845                 err = -EADDRINUSE;
 846                 if (__unix_find_socket_byname(net, sunaddr, addr_len,
 847                                               sk->sk_type, hash)) {
 848                         unix_release_addr(addr);
 849                         goto out_unlock;
 850                 }
 851
 852                 list = &unix_socket_table[addr->hash];
 853         } else {
 854                 list = &unix_socket_table[dentry->d_inode->i_ino & (UNIX_HASH_SIZE-1)];
 855                 u->dentry = nd.path.dentry;
 856                 u->mnt    = nd.path.mnt;
 857         }
 858
 859         err = 0;
 860         __unix_remove_socket(sk);
 861         u->addr = addr;
 862         __unix_insert_socket(list, sk);
 863
 864 out_unlock:
 865         spin_unlock(&unix_table_lock);
 866 out_up:
 867         mutex_unlock(&u->readlock);
 868 out:
 869         return err;
 870
 871 out_mknod_dput:
 872         dput(dentry);
 873 out_mknod_unlock:
 874         mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
 875         path_put(&nd.path);
 876 out_mknod_parent:
 877         if (err==-EEXIST)
 878                 err=-EADDRINUSE;
 879         unix_release_addr(addr);
 880         goto out_up;
 881 }
 882
 883 static void unix_state_double_lock(struct sock *sk1, struct sock *sk2)
 884 {
 885         if (unlikely(sk1 == sk2) || !sk2) {
 886                 unix_state_lock(sk1);
 887                 return;
 888         }
 889         if (sk1 < sk2) {
 890                 unix_state_lock(sk1);
 891                 unix_state_lock_nested(sk2);
 892         } else {
 893                 unix_state_lock(sk2);
 894                 unix_state_lock_nested(sk1);
 895         }
 896 }
 897
 898 static void unix_state_double_unlock(struct sock *sk1, struct sock *sk2)
 899 {
 900         if (unlikely(sk1 == sk2) || !sk2) {
 901                 unix_state_unlock(sk1);
 902                 return;
 903         }
 904         unix_state_unlock(sk1);
 905         unix_state_unlock(sk2);
 906 }
 907
 908 static int unix_dgram_connect(struct socket *sock, struct sockaddr *addr,
 909                               int alen, int flags)
 910 {
 911         struct sock *sk = sock->sk;
 912         struct net *net = sock_net(sk);
 913         struct sockaddr_un *sunaddr=(struct sockaddr_un*)addr;
 914         struct sock *other;
 915         unsigned hash;
 916         int err;
 917
 918         if (addr->sa_family != AF_UNSPEC) {
 919                 err = unix_mkname(sunaddr, alen, &hash);
 920                 if (err < 0)
 921                         goto out;
 922                 alen = err;
 923
 924                 if (test_bit(SOCK_PASSCRED, &sock->flags) &&
 925                     !unix_sk(sk)->addr && (err = unix_autobind(sock)) != 0)
 926                         goto out;
 927
 928 restart:
 929                 other=unix_find_other(net, sunaddr, alen, sock->type, hash, &err);
 930                 if (!other)
 931                         goto out;
 932
 933                 unix_state_double_lock(sk, other);
 934
 935                 /* Apparently VFS overslept socket death. Retry. */
 936                 if (sock_flag(other, SOCK_DEAD)) {
 937                         unix_state_double_unlock(sk, other);
 938                         sock_put(other);
 939                         goto restart;
 940                 }
 941
 942                 err = -EPERM;
 943                 if (!unix_may_send(sk, other))
 944                         goto out_unlock;
 945
 946                 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
 947                 if (err)
 948                         goto out_unlock;
 949
 950         } else {
 951                 /*
 952                  *      1003.1g breaking connected state with AF_UNSPEC
 953                  */
 954                 other = NULL;
 955                 unix_state_double_lock(sk, other);
 956         }
 957
 958         /*
 959          * If it was connected, reconnect.
 960          */
 961         if (unix_peer(sk)) {
 962                 struct sock *old_peer = unix_peer(sk);
 963                 unix_peer(sk)=other;
 964                 unix_state_double_unlock(sk, other);
 965
 966                 if (other != old_peer)
 967                         unix_dgram_disconnected(sk, old_peer);
 968                 sock_put(old_peer);
 969         } else {
 970                 unix_peer(sk)=other;
 971                 unix_state_double_unlock(sk, other);
 972         }
 973         return 0;
 974
 975 out_unlock:
 976         unix_state_double_unlock(sk, other);
 977         sock_put(other);
 978 out:
 979         return err;
 980 }
 981
 982 static long unix_wait_for_peer(struct sock *other, long timeo)
 983 {
 984         struct unix_sock *u = unix_sk(other);
 985         int sched;
 986         DEFINE_WAIT(wait);
 987
 988         prepare_to_wait_exclusive(&u->peer_wait, &wait, TASK_INTERRUPTIBLE);
 989
 990         sched = !sock_flag(other, SOCK_DEAD) &&
 991                 !(other->sk_shutdown & RCV_SHUTDOWN) &&
 992                 unix_recvq_full(other);
 993
 994         unix_state_unlock(other);
 995
 996         if (sched)
 997                 timeo = schedule_timeout(timeo);
 998
 999         finish_wait(&u->peer_wait, &wait);
1000         return timeo;
1001 }
1002
1003 static int unix_stream_connect(struct socket *sock, struct sockaddr *uaddr,
1004                                int addr_len, int flags)
1005 {
1006         struct sockaddr_un *sunaddr=(struct sockaddr_un *)uaddr;
1007         struct sock *sk = sock->sk;
1008         struct net *net = sock_net(sk);
1009         struct unix_sock *u = unix_sk(sk), *newu, *otheru;
1010         struct sock *newsk = NULL;
1011         struct sock *other = NULL;
1012         struct sk_buff *skb = NULL;
1013         unsigned hash;
1014         int st;
1015         int err;
1016         long timeo;
1017
1018         err = unix_mkname(sunaddr, addr_len, &hash);
1019         if (err < 0)
1020                 goto out;
1021         addr_len = err;
1022
1023         if (test_bit(SOCK_PASSCRED, &sock->flags)
1024                 && !u->addr && (err = unix_autobind(sock)) != 0)
1025                 goto out;
1026
1027         timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
1028
1029         /* First of all allocate resources.
1030            If we will make it after state is locked,
1031            we will have to recheck all again in any case.
1032          */
1033
1034         err = -ENOMEM;
1035
1036         /* create new sock for complete connection */
1037         newsk = unix_create1(sock_net(sk), NULL);
1038         if (newsk == NULL)
1039                 goto out;
1040
1041         /* Allocate skb for sending to listening sock */
1042         skb = sock_wmalloc(newsk, 1, 0, GFP_KERNEL);
1043         if (skb == NULL)
1044                 goto out;
1045
1046 restart:
1047         /*  Find listening sock. */
1048         other = unix_find_other(net, sunaddr, addr_len, sk->sk_type, hash, &err);
1049         if (!other)
1050                 goto out;
1051
1052         /* Latch state of peer */
1053         unix_state_lock(other);
1054
1055         /* Apparently VFS overslept socket death. Retry. */
1056         if (sock_flag(other, SOCK_DEAD)) {
1057                 unix_state_unlock(other);
1058                 sock_put(other);
1059                 goto restart;
1060         }
1061
1062         err = -ECONNREFUSED;
1063         if (other->sk_state != TCP_LISTEN)
1064                 goto out_unlock;
1065
1066         if (unix_recvq_full(other)) {
1067                 err = -EAGAIN;
1068                 if (!timeo)
1069                         goto out_unlock;
1070
1071                 timeo = unix_wait_for_peer(other, timeo);
1072
1073                 err = sock_intr_errno(timeo);
1074                 if (signal_pending(current))
1075                         goto out;
1076                 sock_put(other);
1077                 goto restart;
1078         }
1079
1080         /* Latch our state.
1081
1082            It is tricky place. We need to grab write lock and cannot
1083            drop lock on peer. It is dangerous because deadlock is
1084            possible. Connect to self case and simultaneous
1085            attempt to connect are eliminated by checking socket
1086            state. other is TCP_LISTEN, if sk is TCP_LISTEN we
1087            check this before attempt to grab lock.
1088
1089            Well, and we have to recheck the state after socket locked.
1090          */
1091         st = sk->sk_state;
1092
1093         switch (st) {
1094         case TCP_CLOSE:
1095                 /* This is ok... continue with connect */
1096                 break;
1097         case TCP_ESTABLISHED:
1098                 /* Socket is already connected */
1099                 err = -EISCONN;
1100                 goto out_unlock;
1101         default:
1102                 err = -EINVAL;
1103                 goto out_unlock;
1104         }
1105
1106         unix_state_lock_nested(sk);
1107
1108         if (sk->sk_state != st) {
1109                 unix_state_unlock(sk);
1110                 unix_state_unlock(other);
1111                 sock_put(other);
1112                 goto restart;
1113         }
1114
1115         err = security_unix_stream_connect(sock, other->sk_socket, newsk);
1116         if (err) {
1117                 unix_state_unlock(sk);
1118                 goto out_unlock;
1119         }
1120
1121         /* The way is open! Fastly set all the necessary fields... */
1122
1123         sock_hold(sk);
1124         unix_peer(newsk)        = sk;
1125         newsk->sk_state         = TCP_ESTABLISHED;
1126         newsk->sk_type          = sk->sk_type;
1127         newsk->sk_peercred.pid  = task_tgid_vnr(current);
1128         current_euid_egid(&newsk->sk_peercred.uid, &newsk->sk_peercred.gid);
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         current_euid_egid(&skb->sk_peercred.uid, &skb->sk_peercred.gid);
1189         ska->sk_peercred.uid = skb->sk_peercred.uid;
1190         ska->sk_peercred.gid = skb->sk_peercred.gid;
1191
1192         if (ska->sk_type != SOCK_DGRAM) {
1193                 ska->sk_state = TCP_ESTABLISHED;
1194                 skb->sk_state = TCP_ESTABLISHED;
1195                 socka->state  = SS_CONNECTED;
1196                 sockb->state  = SS_CONNECTED;
1197         }
1198         return 0;
1199 }
1200
1201 static int unix_accept(struct socket *sock, struct socket *newsock, int flags)
1202 {
1203         struct sock *sk = sock->sk;
1204         struct sock *tsk;
1205         struct sk_buff *skb;
1206         int err;
1207
1208         err = -EOPNOTSUPP;
1209         if (sock->type!=SOCK_STREAM && sock->type!=SOCK_SEQPACKET)
1210                 goto out;
1211
1212         err = -EINVAL;
1213         if (sk->sk_state != TCP_LISTEN)
1214                 goto out;
1215
1216         /* If socket state is TCP_LISTEN it cannot change (for now...),
1217          * so that no locks are necessary.
1218          */
1219
1220         skb = skb_recv_datagram(sk, 0, flags&O_NONBLOCK, &err);
1221         if (!skb) {
1222                 /* This means receive shutdown. */
1223                 if (err == 0)
1224                         err = -EINVAL;
1225                 goto out;
1226         }
1227
1228         tsk = skb->sk;
1229         skb_free_datagram(sk, skb);
1230         wake_up_interruptible(&unix_sk(sk)->peer_wait);
1231
1232         /* attach accepted sock to socket */
1233         unix_state_lock(tsk);
1234         newsock->state = SS_CONNECTED;
1235         sock_graft(tsk, newsock);
1236         unix_state_unlock(tsk);
1237         return 0;
1238
1239 out:
1240         return err;
1241 }
1242
1243
1244 static int unix_getname(struct socket *sock, struct sockaddr *uaddr, int *uaddr_len, int peer)
1245 {
1246         struct sock *sk = sock->sk;
1247         struct unix_sock *u;
1248         struct sockaddr_un *sunaddr=(struct sockaddr_un *)uaddr;
1249         int err = 0;
1250
1251         if (peer) {
1252                 sk = unix_peer_get(sk);
1253
1254                 err = -ENOTCONN;
1255                 if (!sk)
1256                         goto out;
1257                 err = 0;
1258         } else {
1259                 sock_hold(sk);
1260         }
1261
1262         u = unix_sk(sk);
1263         unix_state_lock(sk);
1264         if (!u->addr) {
1265                 sunaddr->sun_family = AF_UNIX;
1266                 sunaddr->sun_path[0] = 0;
1267                 *uaddr_len = sizeof(short);
1268         } else {
1269                 struct unix_address *addr = u->addr;
1270
1271                 *uaddr_len = addr->len;
1272                 memcpy(sunaddr, addr->name, *uaddr_len);
1273         }
1274         unix_state_unlock(sk);
1275         sock_put(sk);
1276 out:
1277         return err;
1278 }
1279
1280 static void unix_detach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1281 {
1282         int i;
1283
1284         scm->fp = UNIXCB(skb).fp;
1285         skb->destructor = sock_wfree;
1286         UNIXCB(skb).fp = NULL;
1287
1288         for (i=scm->fp->count-1; i>=0; i--)
1289                 unix_notinflight(scm->fp->fp[i]);
1290 }
1291
1292 static void unix_destruct_fds(struct sk_buff *skb)
1293 {
1294         struct scm_cookie scm;
1295         memset(&scm, 0, sizeof(scm));
1296         unix_detach_fds(&scm, skb);
1297
1298         /* Alas, it calls VFS */
1299         /* So fscking what? fput() had been SMP-safe since the last Summer */
1300         scm_destroy(&scm);
1301         sock_wfree(skb);
1302 }
1303
1304 static int unix_attach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1305 {
1306         int i;
1307
1308         /*
1309          * Need to duplicate file references for the sake of garbage
1310          * collection.  Otherwise a socket in the fps might become a
1311          * candidate for GC while the skb is not yet queued.
1312          */
1313         UNIXCB(skb).fp = scm_fp_dup(scm->fp);
1314         if (!UNIXCB(skb).fp)
1315                 return -ENOMEM;
1316
1317         for (i=scm->fp->count-1; i>=0; i--)
1318                 unix_inflight(scm->fp->fp[i]);
1319         skb->destructor = unix_destruct_fds;
1320         return 0;
1321 }
1322
1323 /*
1324  *      Send AF_UNIX data.
1325  */
1326
1327 static int unix_dgram_sendmsg(struct kiocb *kiocb, struct socket *sock,
1328                               struct msghdr *msg, size_t len)
1329 {
1330         struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1331         struct sock *sk = sock->sk;
1332         struct net *net = sock_net(sk);
1333         struct unix_sock *u = unix_sk(sk);
1334         struct sockaddr_un *sunaddr=msg->msg_name;
1335         struct sock *other = NULL;
1336         int namelen = 0; /* fake GCC */
1337         int err;
1338         unsigned hash;
1339         struct sk_buff *skb;
1340         long timeo;
1341         struct scm_cookie tmp_scm;
1342
1343         if (NULL == siocb->scm)
1344                 siocb->scm = &tmp_scm;
1345         wait_for_unix_gc();
1346         err = scm_send(sock, msg, siocb->scm);
1347         if (err < 0)
1348                 return err;
1349
1350         err = -EOPNOTSUPP;
1351         if (msg->msg_flags&MSG_OOB)
1352                 goto out;
1353
1354         if (msg->msg_namelen) {
1355                 err = unix_mkname(sunaddr, msg->msg_namelen, &hash);
1356                 if (err < 0)
1357                         goto out;
1358                 namelen = err;
1359         } else {
1360                 sunaddr = NULL;
1361                 err = -ENOTCONN;
1362                 other = unix_peer_get(sk);
1363                 if (!other)
1364                         goto out;
1365         }
1366
1367         if (test_bit(SOCK_PASSCRED, &sock->flags)
1368                 && !u->addr && (err = unix_autobind(sock)) != 0)
1369                 goto out;
1370
1371         err = -EMSGSIZE;
1372         if (len > sk->sk_sndbuf - 32)
1373                 goto out;
1374
1375         skb = sock_alloc_send_skb(sk, len, msg->msg_flags&MSG_DONTWAIT, &err);
1376         if (skb==NULL)
1377                 goto out;
1378
1379         memcpy(UNIXCREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1380         if (siocb->scm->fp) {
1381                 err = unix_attach_fds(siocb->scm, skb);
1382                 if (err)
1383                         goto out_free;
1384         }
1385         unix_get_secdata(siocb->scm, skb);
1386
1387         skb_reset_transport_header(skb);
1388         err = memcpy_fromiovec(skb_put(skb,len), msg->msg_iov, len);
1389         if (err)
1390                 goto out_free;
1391
1392         timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1393
1394 restart:
1395         if (!other) {
1396                 err = -ECONNRESET;
1397                 if (sunaddr == NULL)
1398                         goto out_free;
1399
1400                 other = unix_find_other(net, sunaddr, namelen, sk->sk_type,
1401                                         hash, &err);
1402                 if (other==NULL)
1403                         goto out_free;
1404         }
1405
1406         unix_state_lock(other);
1407         err = -EPERM;
1408         if (!unix_may_send(sk, other))
1409                 goto out_unlock;
1410
1411         if (sock_flag(other, SOCK_DEAD)) {
1412                 /*
1413                  *      Check with 1003.1g - what should
1414                  *      datagram error
1415                  */
1416                 unix_state_unlock(other);
1417                 sock_put(other);
1418
1419                 err = 0;
1420                 unix_state_lock(sk);
1421                 if (unix_peer(sk) == other) {
1422                         unix_peer(sk)=NULL;
1423                         unix_state_unlock(sk);
1424
1425                         unix_dgram_disconnected(sk, other);
1426                         sock_put(other);
1427                         err = -ECONNREFUSED;
1428                 } else {
1429                         unix_state_unlock(sk);
1430                 }
1431
1432                 other = NULL;
1433                 if (err)
1434                         goto out_free;
1435                 goto restart;
1436         }
1437
1438         err = -EPIPE;
1439         if (other->sk_shutdown & RCV_SHUTDOWN)
1440                 goto out_unlock;
1441
1442         if (sk->sk_type != SOCK_SEQPACKET) {
1443                 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1444                 if (err)
1445                         goto out_unlock;
1446         }
1447
1448         if (unix_peer(other) != sk && unix_recvq_full(other)) {
1449                 if (!timeo) {
1450                         err = -EAGAIN;
1451                         goto out_unlock;
1452                 }
1453
1454                 timeo = unix_wait_for_peer(other, timeo);
1455
1456                 err = sock_intr_errno(timeo);
1457                 if (signal_pending(current))
1458                         goto out_free;
1459
1460                 goto restart;
1461         }
1462
1463         skb_queue_tail(&other->sk_receive_queue, skb);
1464         unix_state_unlock(other);
1465         other->sk_data_ready(other, len);
1466         sock_put(other);
1467         scm_destroy(siocb->scm);
1468         return len;
1469
1470 out_unlock:
1471         unix_state_unlock(other);
1472 out_free:
1473         kfree_skb(skb);
1474 out:
1475         if (other)
1476                 sock_put(other);
1477         scm_destroy(siocb->scm);
1478         return err;
1479 }
1480
1481
1482 static int unix_stream_sendmsg(struct kiocb *kiocb, struct socket *sock,
1483                                struct msghdr *msg, size_t len)
1484 {
1485         struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1486         struct sock *sk = sock->sk;
1487         struct sock *other = NULL;
1488         struct sockaddr_un *sunaddr=msg->msg_name;
1489         int err,size;
1490         struct sk_buff *skb;
1491         int sent=0;
1492         struct scm_cookie tmp_scm;
1493
1494         if (NULL == siocb->scm)
1495                 siocb->scm = &tmp_scm;
1496         wait_for_unix_gc();
1497         err = scm_send(sock, msg, siocb->scm);
1498         if (err < 0)
1499                 return err;
1500
1501         err = -EOPNOTSUPP;
1502         if (msg->msg_flags&MSG_OOB)
1503                 goto out_err;
1504
1505         if (msg->msg_namelen) {
1506                 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
1507                 goto out_err;
1508         } else {
1509                 sunaddr = NULL;
1510                 err = -ENOTCONN;
1511                 other = unix_peer(sk);
1512                 if (!other)
1513                         goto out_err;
1514         }
1515
1516         if (sk->sk_shutdown & SEND_SHUTDOWN)
1517                 goto pipe_err;
1518
1519         while(sent < len)
1520         {
1521                 /*
1522                  *      Optimisation for the fact that under 0.01% of X
1523                  *      messages typically need breaking up.
1524                  */
1525
1526                 size = len-sent;
1527
1528                 /* Keep two messages in the pipe so it schedules better */
1529                 if (size > ((sk->sk_sndbuf >> 1) - 64))
1530                         size = (sk->sk_sndbuf >> 1) - 64;
1531
1532                 if (size > SKB_MAX_ALLOC)
1533                         size = SKB_MAX_ALLOC;
1534
1535                 /*
1536                  *      Grab a buffer
1537                  */
1538
1539                 skb=sock_alloc_send_skb(sk,size,msg->msg_flags&MSG_DONTWAIT, &err);
1540
1541                 if (skb==NULL)
1542                         goto out_err;
1543
1544                 /*
1545                  *      If you pass two values to the sock_alloc_send_skb
1546                  *      it tries to grab the large buffer with GFP_NOFS
1547                  *      (which can fail easily), and if it fails grab the
1548                  *      fallback size buffer which is under a page and will
1549                  *      succeed. [Alan]
1550                  */
1551                 size = min_t(int, size, skb_tailroom(skb));
1552
1553                 memcpy(UNIXCREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1554                 if (siocb->scm->fp) {
1555                         err = unix_attach_fds(siocb->scm, skb);
1556                         if (err) {
1557                                 kfree_skb(skb);
1558                                 goto out_err;
1559                         }
1560                 }
1561
1562                 if ((err = memcpy_fromiovec(skb_put(skb,size), msg->msg_iov, size)) != 0) {
1563                         kfree_skb(skb);
1564                         goto out_err;
1565                 }
1566
1567                 unix_state_lock(other);
1568
1569                 if (sock_flag(other, SOCK_DEAD) ||
1570                     (other->sk_shutdown & RCV_SHUTDOWN))
1571                         goto pipe_err_free;
1572
1573                 skb_queue_tail(&other->sk_receive_queue, skb);
1574                 unix_state_unlock(other);
1575                 other->sk_data_ready(other, size);
1576                 sent+=size;
1577         }
1578
1579         scm_destroy(siocb->scm);
1580         siocb->scm = NULL;
1581
1582         return sent;
1583
1584 pipe_err_free:
1585         unix_state_unlock(other);
1586         kfree_skb(skb);
1587 pipe_err:
1588         if (sent==0 && !(msg->msg_flags&MSG_NOSIGNAL))
1589                 send_sig(SIGPIPE,current,0);
1590         err = -EPIPE;
1591 out_err:
1592         scm_destroy(siocb->scm);
1593         siocb->scm = NULL;
1594         return sent ? : err;
1595 }
1596
1597 static int unix_seqpacket_sendmsg(struct kiocb *kiocb, struct socket *sock,
1598                                   struct msghdr *msg, size_t len)
1599 {
1600         int err;
1601         struct sock *sk = sock->sk;
1602
1603         err = sock_error(sk);
1604         if (err)
1605                 return err;
1606
1607         if (sk->sk_state != TCP_ESTABLISHED)
1608                 return -ENOTCONN;
1609
1610         if (msg->msg_namelen)
1611                 msg->msg_namelen = 0;
1612
1613         return unix_dgram_sendmsg(kiocb, sock, msg, len);
1614 }
1615
1616 static void unix_copy_addr(struct msghdr *msg, struct sock *sk)
1617 {
1618         struct unix_sock *u = unix_sk(sk);
1619
1620         msg->msg_namelen = 0;
1621         if (u->addr) {
1622                 msg->msg_namelen = u->addr->len;
1623                 memcpy(msg->msg_name, u->addr->name, u->addr->len);
1624         }
1625 }
1626
1627 static int unix_dgram_recvmsg(struct kiocb *iocb, struct socket *sock,
1628                               struct msghdr *msg, size_t size,
1629                               int flags)
1630 {
1631         struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1632         struct scm_cookie tmp_scm;
1633         struct sock *sk = sock->sk;
1634         struct unix_sock *u = unix_sk(sk);
1635         int noblock = flags & MSG_DONTWAIT;
1636         struct sk_buff *skb;
1637         int err;
1638
1639         err = -EOPNOTSUPP;
1640         if (flags&MSG_OOB)
1641                 goto out;
1642
1643         msg->msg_namelen = 0;
1644
1645         mutex_lock(&u->readlock);
1646
1647         skb = skb_recv_datagram(sk, flags, noblock, &err);
1648         if (!skb) {
1649                 unix_state_lock(sk);
1650                 /* Signal EOF on disconnected non-blocking SEQPACKET socket. */
1651                 if (sk->sk_type == SOCK_SEQPACKET && err == -EAGAIN &&
1652                     (sk->sk_shutdown & RCV_SHUTDOWN))
1653                         err = 0;
1654                 unix_state_unlock(sk);
1655                 goto out_unlock;
1656         }
1657
1658         wake_up_interruptible_sync(&u->peer_wait);
1659
1660         if (msg->msg_name)
1661                 unix_copy_addr(msg, skb->sk);
1662
1663         if (size > skb->len)
1664                 size = skb->len;
1665         else if (size < skb->len)
1666                 msg->msg_flags |= MSG_TRUNC;
1667
1668         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, size);
1669         if (err)
1670                 goto out_free;
1671
1672         if (!siocb->scm) {
1673                 siocb->scm = &tmp_scm;
1674                 memset(&tmp_scm, 0, sizeof(tmp_scm));
1675         }
1676         siocb->scm->creds = *UNIXCREDS(skb);
1677         unix_set_secdata(siocb->scm, skb);
1678
1679         if (!(flags & MSG_PEEK))
1680         {
1681                 if (UNIXCB(skb).fp)
1682                         unix_detach_fds(siocb->scm, skb);
1683         }
1684         else
1685         {
1686                 /* It is questionable: on PEEK we could:
1687                    - do not return fds - good, but too simple 8)
1688                    - return fds, and do not return them on read (old strategy,
1689                      apparently wrong)
1690                    - clone fds (I chose it for now, it is the most universal
1691                      solution)
1692
1693                    POSIX 1003.1g does not actually define this clearly
1694                    at all. POSIX 1003.1g doesn't define a lot of things
1695                    clearly however!
1696
1697                 */
1698                 if (UNIXCB(skb).fp)
1699                         siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1700         }
1701         err = size;
1702
1703         scm_recv(sock, msg, siocb->scm, flags);
1704
1705 out_free:
1706         skb_free_datagram(sk,skb);
1707 out_unlock:
1708         mutex_unlock(&u->readlock);
1709 out:
1710         return err;
1711 }
1712
1713 /*
1714  *      Sleep until data has arrive. But check for races..
1715  */
1716
1717 static long unix_stream_data_wait(struct sock * sk, long timeo)
1718 {
1719         DEFINE_WAIT(wait);
1720
1721         unix_state_lock(sk);
1722
1723         for (;;) {
1724                 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1725
1726                 if (!skb_queue_empty(&sk->sk_receive_queue) ||
1727                     sk->sk_err ||
1728                     (sk->sk_shutdown & RCV_SHUTDOWN) ||
1729                     signal_pending(current) ||
1730                     !timeo)
1731                         break;
1732
1733                 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1734                 unix_state_unlock(sk);
1735                 timeo = schedule_timeout(timeo);
1736                 unix_state_lock(sk);
1737                 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1738         }
1739
1740         finish_wait(sk->sk_sleep, &wait);
1741         unix_state_unlock(sk);
1742         return timeo;
1743 }
1744
1745
1746
1747 static int unix_stream_recvmsg(struct kiocb *iocb, struct socket *sock,
1748                                struct msghdr *msg, size_t size,
1749                                int flags)
1750 {
1751         struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1752         struct scm_cookie tmp_scm;
1753         struct sock *sk = sock->sk;
1754         struct unix_sock *u = unix_sk(sk);
1755         struct sockaddr_un *sunaddr=msg->msg_name;
1756         int copied = 0;
1757         int check_creds = 0;
1758         int target;
1759         int err = 0;
1760         long timeo;
1761
1762         err = -EINVAL;
1763         if (sk->sk_state != TCP_ESTABLISHED)
1764                 goto out;
1765
1766         err = -EOPNOTSUPP;
1767         if (flags&MSG_OOB)
1768                 goto out;
1769
1770         target = sock_rcvlowat(sk, flags&MSG_WAITALL, size);
1771         timeo = sock_rcvtimeo(sk, flags&MSG_DONTWAIT);
1772
1773         msg->msg_namelen = 0;
1774
1775         /* Lock the socket to prevent queue disordering
1776          * while sleeps in memcpy_tomsg
1777          */
1778
1779         if (!siocb->scm) {
1780                 siocb->scm = &tmp_scm;
1781                 memset(&tmp_scm, 0, sizeof(tmp_scm));
1782         }
1783
1784         mutex_lock(&u->readlock);
1785
1786         do
1787         {
1788                 int chunk;
1789                 struct sk_buff *skb;
1790
1791                 unix_state_lock(sk);
1792                 skb = skb_dequeue(&sk->sk_receive_queue);
1793                 if (skb==NULL)
1794                 {
1795                         if (copied >= target)
1796                                 goto unlock;
1797
1798                         /*
1799                          *      POSIX 1003.1g mandates this order.
1800                          */
1801
1802                         if ((err = sock_error(sk)) != 0)
1803                                 goto unlock;
1804                         if (sk->sk_shutdown & RCV_SHUTDOWN)
1805                                 goto unlock;
1806
1807                         unix_state_unlock(sk);
1808                         err = -EAGAIN;
1809                         if (!timeo)
1810                                 break;
1811                         mutex_unlock(&u->readlock);
1812
1813                         timeo = unix_stream_data_wait(sk, timeo);
1814
1815                         if (signal_pending(current)) {
1816                                 err = sock_intr_errno(timeo);
1817                                 goto out;
1818                         }
1819                         mutex_lock(&u->readlock);
1820                         continue;
1821  unlock:
1822                         unix_state_unlock(sk);
1823                         break;
1824                 }
1825                 unix_state_unlock(sk);
1826
1827                 if (check_creds) {
1828                         /* Never glue messages from different writers */
1829                         if (memcmp(UNIXCREDS(skb), &siocb->scm->creds, sizeof(siocb->scm->creds)) != 0) {
1830                                 skb_queue_head(&sk->sk_receive_queue, skb);
1831                                 break;
1832                         }
1833                 } else {
1834                         /* Copy credentials */
1835                         siocb->scm->creds = *UNIXCREDS(skb);
1836                         check_creds = 1;
1837                 }
1838
1839                 /* Copy address just once */
1840                 if (sunaddr)
1841                 {
1842                         unix_copy_addr(msg, skb->sk);
1843                         sunaddr = NULL;
1844                 }
1845
1846                 chunk = min_t(unsigned int, skb->len, size);
1847                 if (memcpy_toiovec(msg->msg_iov, skb->data, chunk)) {
1848                         skb_queue_head(&sk->sk_receive_queue, skb);
1849                         if (copied == 0)
1850                                 copied = -EFAULT;
1851                         break;
1852                 }
1853                 copied += chunk;
1854                 size -= chunk;
1855
1856                 /* Mark read part of skb as used */
1857                 if (!(flags & MSG_PEEK))
1858                 {
1859                         skb_pull(skb, chunk);
1860
1861                         if (UNIXCB(skb).fp)
1862                                 unix_detach_fds(siocb->scm, skb);
1863
1864                         /* put the skb back if we didn't use it up.. */
1865                         if (skb->len)
1866                         {
1867                                 skb_queue_head(&sk->sk_receive_queue, skb);
1868                                 break;
1869                         }
1870
1871                         kfree_skb(skb);
1872
1873                         if (siocb->scm->fp)
1874                                 break;
1875                 }
1876                 else
1877                 {
1878                         /* It is questionable, see note in unix_dgram_recvmsg.
1879                          */
1880                         if (UNIXCB(skb).fp)
1881                                 siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1882
1883                         /* put message back and return */
1884                         skb_queue_head(&sk->sk_receive_queue, skb);
1885                         break;
1886                 }
1887         } while (size);
1888
1889         mutex_unlock(&u->readlock);
1890         scm_recv(sock, msg, siocb->scm, flags);
1891 out:
1892         return copied ? : err;
1893 }
1894
1895 static int unix_shutdown(struct socket *sock, int mode)
1896 {
1897         struct sock *sk = sock->sk;
1898         struct sock *other;
1899
1900         mode = (mode+1)&(RCV_SHUTDOWN|SEND_SHUTDOWN);
1901
1902         if (mode) {
1903                 unix_state_lock(sk);
1904                 sk->sk_shutdown |= mode;
1905                 other=unix_peer(sk);
1906                 if (other)
1907                         sock_hold(other);
1908                 unix_state_unlock(sk);
1909                 sk->sk_state_change(sk);
1910
1911                 if (other &&
1912                         (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET)) {
1913
1914                         int peer_mode = 0;
1915
1916                         if (mode&RCV_SHUTDOWN)
1917                                 peer_mode |= SEND_SHUTDOWN;
1918                         if (mode&SEND_SHUTDOWN)
1919                                 peer_mode |= RCV_SHUTDOWN;
1920                         unix_state_lock(other);
1921                         other->sk_shutdown |= peer_mode;
1922                         unix_state_unlock(other);
1923                         other->sk_state_change(other);
1924                         read_lock(&other->sk_callback_lock);
1925                         if (peer_mode == SHUTDOWN_MASK)
1926                                 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_HUP);
1927                         else if (peer_mode & RCV_SHUTDOWN)
1928                                 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_IN);
1929                         read_unlock(&other->sk_callback_lock);
1930                 }
1931                 if (other)
1932                         sock_put(other);
1933         }
1934         return 0;
1935 }
1936
1937 static int unix_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1938 {
1939         struct sock *sk = sock->sk;
1940         long amount=0;
1941         int err;
1942
1943         switch(cmd)
1944         {
1945                 case SIOCOUTQ:
1946                         amount = atomic_read(&sk->sk_wmem_alloc);
1947                         err = put_user(amount, (int __user *)arg);
1948                         break;
1949                 case SIOCINQ:
1950                 {
1951                         struct sk_buff *skb;
1952
1953                         if (sk->sk_state == TCP_LISTEN) {
1954                                 err = -EINVAL;
1955                                 break;
1956                         }
1957
1958                         spin_lock(&sk->sk_receive_queue.lock);
1959                         if (sk->sk_type == SOCK_STREAM ||
1960                             sk->sk_type == SOCK_SEQPACKET) {
1961                                 skb_queue_walk(&sk->sk_receive_queue, skb)
1962                                         amount += skb->len;
1963                         } else {
1964                                 skb = skb_peek(&sk->sk_receive_queue);
1965                                 if (skb)
1966                                         amount=skb->len;
1967                         }
1968                         spin_unlock(&sk->sk_receive_queue.lock);
1969                         err = put_user(amount, (int __user *)arg);
1970                         break;
1971                 }
1972
1973                 default:
1974                         err = -ENOIOCTLCMD;
1975                         break;
1976         }
1977         return err;
1978 }
1979
1980 static unsigned int unix_poll(struct file * file, struct socket *sock, poll_table *wait)
1981 {
1982         struct sock *sk = sock->sk;
1983         unsigned int mask;
1984
1985         poll_wait(file, sk->sk_sleep, wait);
1986         mask = 0;
1987
1988         /* exceptional events? */
1989         if (sk->sk_err)
1990                 mask |= POLLERR;
1991         if (sk->sk_shutdown == SHUTDOWN_MASK)
1992                 mask |= POLLHUP;
1993         if (sk->sk_shutdown & RCV_SHUTDOWN)
1994                 mask |= POLLRDHUP;
1995
1996         /* readable? */
1997         if (!skb_queue_empty(&sk->sk_receive_queue) ||
1998             (sk->sk_shutdown & RCV_SHUTDOWN))
1999                 mask |= POLLIN | POLLRDNORM;
2000
2001         /* Connection-based need to check for termination and startup */
2002         if ((sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) && sk->sk_state == TCP_CLOSE)
2003                 mask |= POLLHUP;
2004
2005         /*
2006          * we set writable also when the other side has shut down the
2007          * connection. This prevents stuck sockets.
2008          */
2009         if (unix_writable(sk))
2010                 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2011
2012         return mask;
2013 }
2014
2015 static unsigned int unix_dgram_poll(struct file *file, struct socket *sock,
2016                                     poll_table *wait)
2017 {
2018         struct sock *sk = sock->sk, *other;
2019         unsigned int mask, writable;
2020
2021         poll_wait(file, sk->sk_sleep, wait);
2022         mask = 0;
2023
2024         /* exceptional events? */
2025         if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue))
2026                 mask |= POLLERR;
2027         if (sk->sk_shutdown & RCV_SHUTDOWN)
2028                 mask |= POLLRDHUP;
2029         if (sk->sk_shutdown == SHUTDOWN_MASK)
2030                 mask |= POLLHUP;
2031
2032         /* readable? */
2033         if (!skb_queue_empty(&sk->sk_receive_queue) ||
2034             (sk->sk_shutdown & RCV_SHUTDOWN))
2035                 mask |= POLLIN | POLLRDNORM;
2036
2037         /* Connection-based need to check for termination and startup */
2038         if (sk->sk_type == SOCK_SEQPACKET) {
2039                 if (sk->sk_state == TCP_CLOSE)
2040                         mask |= POLLHUP;
2041                 /* connection hasn't started yet? */
2042                 if (sk->sk_state == TCP_SYN_SENT)
2043                         return mask;
2044         }
2045
2046         /* writable? */
2047         writable = unix_writable(sk);
2048         if (writable) {
2049                 other = unix_peer_get(sk);
2050                 if (other) {
2051                         if (unix_peer(other) != sk) {
2052                                 poll_wait(file, &unix_sk(other)->peer_wait,
2053                                           wait);
2054                                 if (unix_recvq_full(other))
2055                                         writable = 0;
2056                         }
2057
2058                         sock_put(other);
2059                 }
2060         }
2061
2062         if (writable)
2063                 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2064         else
2065                 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
2066
2067         return mask;
2068 }
2069
2070 #ifdef CONFIG_PROC_FS
2071 static struct sock *first_unix_socket(int *i)
2072 {
2073         for (*i = 0; *i <= UNIX_HASH_SIZE; (*i)++) {
2074                 if (!hlist_empty(&unix_socket_table[*i]))
2075                         return __sk_head(&unix_socket_table[*i]);
2076         }
2077         return NULL;
2078 }
2079
2080 static struct sock *next_unix_socket(int *i, struct sock *s)
2081 {
2082         struct sock *next = sk_next(s);
2083         /* More in this chain? */
2084         if (next)
2085                 return next;
2086         /* Look for next non-empty chain. */
2087         for ((*i)++; *i <= UNIX_HASH_SIZE; (*i)++) {
2088                 if (!hlist_empty(&unix_socket_table[*i]))
2089                         return __sk_head(&unix_socket_table[*i]);
2090         }
2091         return NULL;
2092 }
2093
2094 struct unix_iter_state {
2095         struct seq_net_private p;
2096         int i;
2097 };
2098 static struct sock *unix_seq_idx(struct seq_file *seq, loff_t pos)
2099 {
2100         struct unix_iter_state *iter = seq->private;
2101         loff_t off = 0;
2102         struct sock *s;
2103
2104         for (s = first_unix_socket(&iter->i); s; s = next_unix_socket(&iter->i, s)) {
2105                 if (sock_net(s) != seq_file_net(seq))
2106                         continue;
2107                 if (off == pos)
2108                         return s;
2109                 ++off;
2110         }
2111         return NULL;
2112 }
2113
2114
2115 static void *unix_seq_start(struct seq_file *seq, loff_t *pos)
2116         __acquires(unix_table_lock)
2117 {
2118         spin_lock(&unix_table_lock);
2119         return *pos ? unix_seq_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2120 }
2121
2122 static void *unix_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2123 {
2124         struct unix_iter_state *iter = seq->private;
2125         struct sock *sk = v;
2126         ++*pos;
2127
2128         if (v == SEQ_START_TOKEN)
2129                 sk = first_unix_socket(&iter->i);
2130         else
2131                 sk = next_unix_socket(&iter->i, sk);
2132         while (sk && (sock_net(sk) != seq_file_net(seq)))
2133                 sk = next_unix_socket(&iter->i, sk);
2134         return sk;
2135 }
2136
2137 static void unix_seq_stop(struct seq_file *seq, void *v)
2138         __releases(unix_table_lock)
2139 {
2140         spin_unlock(&unix_table_lock);
2141 }
2142
2143 static int unix_seq_show(struct seq_file *seq, void *v)
2144 {
2145
2146         if (v == SEQ_START_TOKEN)
2147                 seq_puts(seq, "Num       RefCount Protocol Flags    Type St "
2148                          "Inode Path\n");
2149         else {
2150                 struct sock *s = v;
2151                 struct unix_sock *u = unix_sk(s);
2152                 unix_state_lock(s);
2153
2154                 seq_printf(seq, "%p: %08X %08X %08X %04X %02X %5lu",
2155                         s,
2156                         atomic_read(&s->sk_refcnt),
2157                         0,
2158                         s->sk_state == TCP_LISTEN ? __SO_ACCEPTCON : 0,
2159                         s->sk_type,
2160                         s->sk_socket ?
2161                         (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTED : SS_UNCONNECTED) :
2162                         (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTING : SS_DISCONNECTING),
2163                         sock_i_ino(s));
2164
2165                 if (u->addr) {
2166                         int i, len;
2167                         seq_putc(seq, ' ');
2168
2169                         i = 0;
2170                         len = u->addr->len - sizeof(short);
2171                         if (!UNIX_ABSTRACT(s))
2172                                 len--;
2173                         else {
2174                                 seq_putc(seq, '@');
2175                                 i++;
2176                         }
2177                         for ( ; i < len; i++)
2178                                 seq_putc(seq, u->addr->name->sun_path[i]);
2179                 }
2180                 unix_state_unlock(s);
2181                 seq_putc(seq, '\n');
2182         }
2183
2184         return 0;
2185 }
2186
2187 static const struct seq_operations unix_seq_ops = {
2188         .start  = unix_seq_start,
2189         .next   = unix_seq_next,
2190         .stop   = unix_seq_stop,
2191         .show   = unix_seq_show,
2192 };
2193
2194
2195 static int unix_seq_open(struct inode *inode, struct file *file)
2196 {
2197         return seq_open_net(inode, file, &unix_seq_ops,
2198                             sizeof(struct unix_iter_state));
2199 }
2200
2201 static const struct file_operations unix_seq_fops = {
2202         .owner          = THIS_MODULE,
2203         .open           = unix_seq_open,
2204         .read           = seq_read,
2205         .llseek         = seq_lseek,
2206         .release        = seq_release_net,
2207 };
2208
2209 #endif
2210
2211 static struct net_proto_family unix_family_ops = {
2212         .family = PF_UNIX,
2213         .create = unix_create,
2214         .owner  = THIS_MODULE,
2215 };
2216
2217
2218 static int unix_net_init(struct net *net)
2219 {
2220         int error = -ENOMEM;
2221
2222         net->unx.sysctl_max_dgram_qlen = 10;
2223         if (unix_sysctl_register(net))
2224                 goto out;
2225
2226 #ifdef CONFIG_PROC_FS
2227         if (!proc_net_fops_create(net, "unix", 0, &unix_seq_fops)) {
2228                 unix_sysctl_unregister(net);
2229                 goto out;
2230         }
2231 #endif
2232         error = 0;
2233 out:
2234         return error;
2235 }
2236
2237 static void unix_net_exit(struct net *net)
2238 {
2239         unix_sysctl_unregister(net);
2240         proc_net_remove(net, "unix");
2241 }
2242
2243 static struct pernet_operations unix_net_ops = {
2244         .init = unix_net_init,
2245         .exit = unix_net_exit,
2246 };
2247
2248 static int __init af_unix_init(void)
2249 {
2250         int rc = -1;
2251         struct sk_buff *dummy_skb;
2252
2253         BUILD_BUG_ON(sizeof(struct unix_skb_parms) > sizeof(dummy_skb->cb));
2254
2255         rc = proto_register(&unix_proto, 1);
2256         if (rc != 0) {
2257                 printk(KERN_CRIT "%s: Cannot create unix_sock SLAB cache!\n",
2258                        __func__);
2259                 goto out;
2260         }
2261
2262         sock_register(&unix_family_ops);
2263         register_pernet_subsys(&unix_net_ops);
2264 out:
2265         return rc;
2266 }
2267
2268 static void __exit af_unix_exit(void)
2269 {
2270         sock_unregister(PF_UNIX);
2271         proto_unregister(&unix_proto);
2272         unregister_pernet_subsys(&unix_net_ops);
2273 }
2274
2275 /* Earlier than device_initcall() so that other drivers invoking
2276    request_module() don't end up in a loop when modprobe tries
2277    to use a UNIX socket. But later than subsys_initcall() because
2278    we depend on stuff initialised there */
2279 fs_initcall(af_unix_init);
2280 module_exit(af_unix_exit);
2281
2282 MODULE_LICENSE("GPL");
2283 MODULE_ALIAS_NETPROTO(PF_UNIX);