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(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                         sock_hold(s);
 300                         goto found;
 301                 }
 302         }
 303         s = NULL;
 304 found:
 305         spin_unlock(&unix_table_lock);
 306         return s;
 307 }
 308
 309 static inline int unix_writable(struct sock *sk)
 310 {
 311         return (atomic_read(&sk->sk_wmem_alloc) << 2) <= sk->sk_sndbuf;
 312 }
 313
 314 static void unix_write_space(struct sock *sk)
 315 {
 316         read_lock(&sk->sk_callback_lock);
 317         if (unix_writable(sk)) {
 318                 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
 319                         wake_up_interruptible_sync(sk->sk_sleep);
 320                 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
 321         }
 322         read_unlock(&sk->sk_callback_lock);
 323 }
 324
 325 /* When dgram socket disconnects (or changes its peer), we clear its receive
 326  * queue of packets arrived from previous peer. First, it allows to do
 327  * flow control based only on wmem_alloc; second, sk connected to peer
 328  * may receive messages only from that peer. */
 329 static void unix_dgram_disconnected(struct sock *sk, struct sock *other)
 330 {
 331         if (!skb_queue_empty(&sk->sk_receive_queue)) {
 332                 skb_queue_purge(&sk->sk_receive_queue);
 333                 wake_up_interruptible_all(&unix_sk(sk)->peer_wait);
 334
 335                 /* If one link of bidirectional dgram pipe is disconnected,
 336                  * we signal error. Messages are lost. Do not make this,
 337                  * when peer was not connected to us.
 338                  */
 339                 if (!sock_flag(other, SOCK_DEAD) && unix_peer(other) == sk) {
 340                         other->sk_err = ECONNRESET;
 341                         other->sk_error_report(other);
 342                 }
 343         }
 344 }
 345
 346 static void unix_sock_destructor(struct sock *sk)
 347 {
 348         struct unix_sock *u = unix_sk(sk);
 349
 350         skb_queue_purge(&sk->sk_receive_queue);
 351
 352         WARN_ON(atomic_read(&sk->sk_wmem_alloc));
 353         WARN_ON(!sk_unhashed(sk));
 354         WARN_ON(sk->sk_socket);
 355         if (!sock_flag(sk, SOCK_DEAD)) {
 356                 printk(KERN_INFO "Attempt to release alive unix socket: %p\n", sk);
 357                 return;
 358         }
 359
 360         if (u->addr)
 361                 unix_release_addr(u->addr);
 362
 363         atomic_dec(&unix_nr_socks);
 364         local_bh_disable();
 365         sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
 366         local_bh_enable();
 367 #ifdef UNIX_REFCNT_DEBUG
 368         printk(KERN_DEBUG "UNIX %p is destroyed, %d are still alive.\n", sk,
 369                 atomic_read(&unix_nr_socks));
 370 #endif
 371 }
 372
 373 static int unix_release_sock(struct sock *sk, int embrion)
 374 {
 375         struct unix_sock *u = unix_sk(sk);
 376         struct dentry *dentry;
 377         struct vfsmount *mnt;
 378         struct sock *skpair;
 379         struct sk_buff *skb;
 380         int state;
 381
 382         unix_remove_socket(sk);
 383
 384         /* Clear state */
 385         unix_state_lock(sk);
 386         sock_orphan(sk);
 387         sk->sk_shutdown = SHUTDOWN_MASK;
 388         dentry       = u->dentry;
 389         u->dentry    = NULL;
 390         mnt          = u->mnt;
 391         u->mnt       = NULL;
 392         state = sk->sk_state;
 393         sk->sk_state = TCP_CLOSE;
 394         unix_state_unlock(sk);
 395
 396         wake_up_interruptible_all(&u->peer_wait);
 397
 398         skpair = unix_peer(sk);
 399
 400         if (skpair != NULL) {
 401                 if (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) {
 402                         unix_state_lock(skpair);
 403                         /* No more writes */
 404                         skpair->sk_shutdown = SHUTDOWN_MASK;
 405                         if (!skb_queue_empty(&sk->sk_receive_queue) || embrion)
 406                                 skpair->sk_err = ECONNRESET;
 407                         unix_state_unlock(skpair);
 408                         skpair->sk_state_change(skpair);
 409                         read_lock(&skpair->sk_callback_lock);
 410                         sk_wake_async(skpair, SOCK_WAKE_WAITD, POLL_HUP);
 411                         read_unlock(&skpair->sk_callback_lock);
 412                 }
 413                 sock_put(skpair); /* It may now die */
 414                 unix_peer(sk) = NULL;
 415         }
 416
 417         /* Try to flush out this socket. Throw out buffers at least */
 418
 419         while ((skb = skb_dequeue(&sk->sk_receive_queue)) != NULL) {
 420                 if (state == TCP_LISTEN)
 421                         unix_release_sock(skb->sk, 1);
 422                 /* passed fds are erased in the kfree_skb hook        */
 423                 kfree_skb(skb);
 424         }
 425
 426         if (dentry) {
 427                 dput(dentry);
 428                 mntput(mnt);
 429         }
 430
 431         sock_put(sk);
 432
 433         /* ---- Socket is dead now and most probably destroyed ---- */
 434
 435         /*
 436          * Fixme: BSD difference: In BSD all sockets connected to use get
 437          *        ECONNRESET and we die on the spot. In Linux we behave
 438          *        like files and pipes do and wait for the last
 439          *        dereference.
 440          *
 441          * Can't we simply set sock->err?
 442          *
 443          *        What the above comment does talk about? --ANK(980817)
 444          */
 445
 446         if (unix_tot_inflight)
 447                 unix_gc();              /* Garbage collect fds */
 448
 449         return 0;
 450 }
 451
 452 static int unix_listen(struct socket *sock, int backlog)
 453 {
 454         int err;
 455         struct sock *sk = sock->sk;
 456         struct unix_sock *u = unix_sk(sk);
 457
 458         err = -EOPNOTSUPP;
 459         if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
 460                 goto out;       /* Only stream/seqpacket sockets accept */
 461         err = -EINVAL;
 462         if (!u->addr)
 463                 goto out;       /* No listens on an unbound socket */
 464         unix_state_lock(sk);
 465         if (sk->sk_state != TCP_CLOSE && sk->sk_state != TCP_LISTEN)
 466                 goto out_unlock;
 467         if (backlog > sk->sk_max_ack_backlog)
 468                 wake_up_interruptible_all(&u->peer_wait);
 469         sk->sk_max_ack_backlog  = backlog;
 470         sk->sk_state            = TCP_LISTEN;
 471         /* set credentials so connect can copy them */
 472         sk->sk_peercred.pid     = task_tgid_vnr(current);
 473         current_euid_egid(&sk->sk_peercred.uid, &sk->sk_peercred.gid);
 474         err = 0;
 475
 476 out_unlock:
 477         unix_state_unlock(sk);
 478 out:
 479         return err;
 480 }
 481
 482 static int unix_release(struct socket *);
 483 static int unix_bind(struct socket *, struct sockaddr *, int);
 484 static int unix_stream_connect(struct socket *, struct sockaddr *,
 485                                int addr_len, int flags);
 486 static int unix_socketpair(struct socket *, struct socket *);
 487 static int unix_accept(struct socket *, struct socket *, int);
 488 static int unix_getname(struct socket *, struct sockaddr *, int *, int);
 489 static unsigned int unix_poll(struct file *, struct socket *, poll_table *);
 490 static unsigned int unix_dgram_poll(struct file *, struct socket *,
 491                                     poll_table *);
 492 static int unix_ioctl(struct socket *, unsigned int, unsigned long);
 493 static int unix_shutdown(struct socket *, int);
 494 static int unix_stream_sendmsg(struct kiocb *, struct socket *,
 495                                struct msghdr *, size_t);
 496 static int unix_stream_recvmsg(struct kiocb *, struct socket *,
 497                                struct msghdr *, size_t, int);
 498 static int unix_dgram_sendmsg(struct kiocb *, struct socket *,
 499                               struct msghdr *, size_t);
 500 static int unix_dgram_recvmsg(struct kiocb *, struct socket *,
 501                               struct msghdr *, size_t, int);
 502 static int unix_dgram_connect(struct socket *, struct sockaddr *,
 503                               int, int);
 504 static int unix_seqpacket_sendmsg(struct kiocb *, struct socket *,
 505                                   struct msghdr *, size_t);
 506
 507 static const struct proto_ops unix_stream_ops = {
 508         .family =       PF_UNIX,
 509         .owner =        THIS_MODULE,
 510         .release =      unix_release,
 511         .bind =         unix_bind,
 512         .connect =      unix_stream_connect,
 513         .socketpair =   unix_socketpair,
 514         .accept =       unix_accept,
 515         .getname =      unix_getname,
 516         .poll =         unix_poll,
 517         .ioctl =        unix_ioctl,
 518         .listen =       unix_listen,
 519         .shutdown =     unix_shutdown,
 520         .setsockopt =   sock_no_setsockopt,
 521         .getsockopt =   sock_no_getsockopt,
 522         .sendmsg =      unix_stream_sendmsg,
 523         .recvmsg =      unix_stream_recvmsg,
 524         .mmap =         sock_no_mmap,
 525         .sendpage =     sock_no_sendpage,
 526 };
 527
 528 static const struct proto_ops unix_dgram_ops = {
 529         .family =       PF_UNIX,
 530         .owner =        THIS_MODULE,
 531         .release =      unix_release,
 532         .bind =         unix_bind,
 533         .connect =      unix_dgram_connect,
 534         .socketpair =   unix_socketpair,
 535         .accept =       sock_no_accept,
 536         .getname =      unix_getname,
 537         .poll =         unix_dgram_poll,
 538         .ioctl =        unix_ioctl,
 539         .listen =       sock_no_listen,
 540         .shutdown =     unix_shutdown,
 541         .setsockopt =   sock_no_setsockopt,
 542         .getsockopt =   sock_no_getsockopt,
 543         .sendmsg =      unix_dgram_sendmsg,
 544         .recvmsg =      unix_dgram_recvmsg,
 545         .mmap =         sock_no_mmap,
 546         .sendpage =     sock_no_sendpage,
 547 };
 548
 549 static const struct proto_ops unix_seqpacket_ops = {
 550         .family =       PF_UNIX,
 551         .owner =        THIS_MODULE,
 552         .release =      unix_release,
 553         .bind =         unix_bind,
 554         .connect =      unix_stream_connect,
 555         .socketpair =   unix_socketpair,
 556         .accept =       unix_accept,
 557         .getname =      unix_getname,
 558         .poll =         unix_dgram_poll,
 559         .ioctl =        unix_ioctl,
 560         .listen =       unix_listen,
 561         .shutdown =     unix_shutdown,
 562         .setsockopt =   sock_no_setsockopt,
 563         .getsockopt =   sock_no_getsockopt,
 564         .sendmsg =      unix_seqpacket_sendmsg,
 565         .recvmsg =      unix_dgram_recvmsg,
 566         .mmap =         sock_no_mmap,
 567         .sendpage =     sock_no_sendpage,
 568 };
 569
 570 static struct proto unix_proto = {
 571         .name                   = "UNIX",
 572         .owner                  = THIS_MODULE,
 573         .obj_size               = sizeof(struct unix_sock),
 574 };
 575
 576 /*
 577  * AF_UNIX sockets do not interact with hardware, hence they
 578  * dont trigger interrupts - so it's safe for them to have
 579  * bh-unsafe locking for their sk_receive_queue.lock. Split off
 580  * this special lock-class by reinitializing the spinlock key:
 581  */
 582 static struct lock_class_key af_unix_sk_receive_queue_lock_key;
 583
 584 static struct sock *unix_create1(struct net *net, struct socket *sock)
 585 {
 586         struct sock *sk = NULL;
 587         struct unix_sock *u;
 588
 589         atomic_inc(&unix_nr_socks);
 590         if (atomic_read(&unix_nr_socks) > 2 * get_max_files())
 591                 goto out;
 592
 593         sk = sk_alloc(net, PF_UNIX, GFP_KERNEL, &unix_proto);
 594         if (!sk)
 595                 goto out;
 596
 597         sock_init_data(sock, sk);
 598         lockdep_set_class(&sk->sk_receive_queue.lock,
 599                                 &af_unix_sk_receive_queue_lock_key);
 600
 601         sk->sk_write_space      = unix_write_space;
 602         sk->sk_max_ack_backlog  = net->unx.sysctl_max_dgram_qlen;
 603         sk->sk_destruct         = unix_sock_destructor;
 604         u         = unix_sk(sk);
 605         u->dentry = NULL;
 606         u->mnt    = NULL;
 607         spin_lock_init(&u->lock);
 608         atomic_long_set(&u->inflight, 0);
 609         INIT_LIST_HEAD(&u->link);
 610         mutex_init(&u->readlock); /* single task reading lock */
 611         init_waitqueue_head(&u->peer_wait);
 612         unix_insert_socket(unix_sockets_unbound, sk);
 613 out:
 614         if (sk == NULL)
 615                 atomic_dec(&unix_nr_socks);
 616         else {
 617                 local_bh_disable();
 618                 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
 619                 local_bh_enable();
 620         }
 621         return sk;
 622 }
 623
 624 static int unix_create(struct net *net, struct socket *sock, int protocol)
 625 {
 626         if (protocol && protocol != PF_UNIX)
 627                 return -EPROTONOSUPPORT;
 628
 629         sock->state = SS_UNCONNECTED;
 630
 631         switch (sock->type) {
 632         case SOCK_STREAM:
 633                 sock->ops = &unix_stream_ops;
 634                 break;
 635                 /*
 636                  *      Believe it or not BSD has AF_UNIX, SOCK_RAW though
 637                  *      nothing uses it.
 638                  */
 639         case SOCK_RAW:
 640                 sock->type = SOCK_DGRAM;
 641         case SOCK_DGRAM:
 642                 sock->ops = &unix_dgram_ops;
 643                 break;
 644         case SOCK_SEQPACKET:
 645                 sock->ops = &unix_seqpacket_ops;
 646                 break;
 647         default:
 648                 return -ESOCKTNOSUPPORT;
 649         }
 650
 651         return unix_create1(net, sock) ? 0 : -ENOMEM;
 652 }
 653
 654 static int unix_release(struct socket *sock)
 655 {
 656         struct sock *sk = sock->sk;
 657
 658         if (!sk)
 659                 return 0;
 660
 661         sock->sk = NULL;
 662
 663         return unix_release_sock(sk, 0);
 664 }
 665
 666 static int unix_autobind(struct socket *sock)
 667 {
 668         struct sock *sk = sock->sk;
 669         struct net *net = sock_net(sk);
 670         struct unix_sock *u = unix_sk(sk);
 671         static u32 ordernum = 1;
 672         struct unix_address *addr;
 673         int err;
 674
 675         mutex_lock(&u->readlock);
 676
 677         err = 0;
 678         if (u->addr)
 679                 goto out;
 680
 681         err = -ENOMEM;
 682         addr = kzalloc(sizeof(*addr) + sizeof(short) + 16, GFP_KERNEL);
 683         if (!addr)
 684                 goto out;
 685
 686         addr->name->sun_family = AF_UNIX;
 687         atomic_set(&addr->refcnt, 1);
 688
 689 retry:
 690         addr->len = sprintf(addr->name->sun_path+1, "%05x", ordernum) + 1 + sizeof(short);
 691         addr->hash = unix_hash_fold(csum_partial(addr->name, addr->len, 0));
 692
 693         spin_lock(&unix_table_lock);
 694         ordernum = (ordernum+1)&0xFFFFF;
 695
 696         if (__unix_find_socket_byname(net, addr->name, addr->len, sock->type,
 697                                       addr->hash)) {
 698                 spin_unlock(&unix_table_lock);
 699                 /* Sanity yield. It is unusual case, but yet... */
 700                 if (!(ordernum&0xFF))
 701                         yield();
 702                 goto retry;
 703         }
 704         addr->hash ^= sk->sk_type;
 705
 706         __unix_remove_socket(sk);
 707         u->addr = addr;
 708         __unix_insert_socket(&unix_socket_table[addr->hash], sk);
 709         spin_unlock(&unix_table_lock);
 710         err = 0;
 711
 712 out:    mutex_unlock(&u->readlock);
 713         return err;
 714 }
 715
 716 static struct sock *unix_find_other(struct net *net,
 717                                     struct sockaddr_un *sunname, int len,
 718                                     int type, unsigned hash, int *error)
 719 {
 720         struct sock *u;
 721         struct path path;
 722         int err = 0;
 723
 724         if (sunname->sun_path[0]) {
 725                 struct inode *inode;
 726                 err = kern_path(sunname->sun_path, LOOKUP_FOLLOW, &path);
 727                 if (err)
 728                         goto fail;
 729                 inode = path.dentry->d_inode;
 730                 err = inode_permission(inode, MAY_WRITE);
 731                 if (err)
 732                         goto put_fail;
 733
 734                 err = -ECONNREFUSED;
 735                 if (!S_ISSOCK(inode->i_mode))
 736                         goto put_fail;
 737                 u = unix_find_socket_byinode(net, inode);
 738                 if (!u)
 739                         goto put_fail;
 740
 741                 if (u->sk_type == type)
 742                         touch_atime(path.mnt, path.dentry);
 743
 744                 path_put(&path);
 745
 746                 err = -EPROTOTYPE;
 747                 if (u->sk_type != type) {
 748                         sock_put(u);
 749                         goto fail;
 750                 }
 751         } else {
 752                 err = -ECONNREFUSED;
 753                 u = unix_find_socket_byname(net, sunname, len, type, hash);
 754                 if (u) {
 755                         struct dentry *dentry;
 756                         dentry = unix_sk(u)->dentry;
 757                         if (dentry)
 758                                 touch_atime(unix_sk(u)->mnt, dentry);
 759                 } else
 760                         goto fail;
 761         }
 762         return u;
 763
 764 put_fail:
 765         path_put(&path);
 766 fail:
 767         *error = err;
 768         return NULL;
 769 }
 770
 771
 772 static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
 773 {
 774         struct sock *sk = sock->sk;
 775         struct net *net = sock_net(sk);
 776         struct unix_sock *u = unix_sk(sk);
 777         struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
 778         struct dentry *dentry = NULL;
 779         struct nameidata nd;
 780         int err;
 781         unsigned hash;
 782         struct unix_address *addr;
 783         struct hlist_head *list;
 784
 785         err = -EINVAL;
 786         if (sunaddr->sun_family != AF_UNIX)
 787                 goto out;
 788
 789         if (addr_len == sizeof(short)) {
 790                 err = unix_autobind(sock);
 791                 goto out;
 792         }
 793
 794         err = unix_mkname(sunaddr, addr_len, &hash);
 795         if (err < 0)
 796                 goto out;
 797         addr_len = err;
 798
 799         mutex_lock(&u->readlock);
 800
 801         err = -EINVAL;
 802         if (u->addr)
 803                 goto out_up;
 804
 805         err = -ENOMEM;
 806         addr = kmalloc(sizeof(*addr)+addr_len, GFP_KERNEL);
 807         if (!addr)
 808                 goto out_up;
 809
 810         memcpy(addr->name, sunaddr, addr_len);
 811         addr->len = addr_len;
 812         addr->hash = hash ^ sk->sk_type;
 813         atomic_set(&addr->refcnt, 1);
 814
 815         if (sunaddr->sun_path[0]) {
 816                 unsigned int mode;
 817                 err = 0;
 818                 /*
 819                  * Get the parent directory, calculate the hash for last
 820                  * component.
 821                  */
 822                 err = path_lookup(sunaddr->sun_path, LOOKUP_PARENT, &nd);
 823                 if (err)
 824                         goto out_mknod_parent;
 825
 826                 dentry = lookup_create(&nd, 0);
 827                 err = PTR_ERR(dentry);
 828                 if (IS_ERR(dentry))
 829                         goto out_mknod_unlock;
 830
 831                 /*
 832                  * All right, let's create it.
 833                  */
 834                 mode = S_IFSOCK |
 835                        (SOCK_INODE(sock)->i_mode & ~current->fs->umask);
 836                 err = mnt_want_write(nd.path.mnt);
 837                 if (err)
 838                         goto out_mknod_dput;
 839                 err = vfs_mknod(nd.path.dentry->d_inode, dentry, mode, 0);
 840                 mnt_drop_write(nd.path.mnt);
 841                 if (err)
 842                         goto out_mknod_dput;
 843                 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
 844                 dput(nd.path.dentry);
 845                 nd.path.dentry = dentry;
 846
 847                 addr->hash = UNIX_HASH_SIZE;
 848         }
 849
 850         spin_lock(&unix_table_lock);
 851
 852         if (!sunaddr->sun_path[0]) {
 853                 err = -EADDRINUSE;
 854                 if (__unix_find_socket_byname(net, sunaddr, addr_len,
 855                                               sk->sk_type, hash)) {
 856                         unix_release_addr(addr);
 857                         goto out_unlock;
 858                 }
 859
 860                 list = &unix_socket_table[addr->hash];
 861         } else {
 862                 list = &unix_socket_table[dentry->d_inode->i_ino & (UNIX_HASH_SIZE-1)];
 863                 u->dentry = nd.path.dentry;
 864                 u->mnt    = nd.path.mnt;
 865         }
 866
 867         err = 0;
 868         __unix_remove_socket(sk);
 869         u->addr = addr;
 870         __unix_insert_socket(list, sk);
 871
 872 out_unlock:
 873         spin_unlock(&unix_table_lock);
 874 out_up:
 875         mutex_unlock(&u->readlock);
 876 out:
 877         return err;
 878
 879 out_mknod_dput:
 880         dput(dentry);
 881 out_mknod_unlock:
 882         mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
 883         path_put(&nd.path);
 884 out_mknod_parent:
 885         if (err == -EEXIST)
 886                 err = -EADDRINUSE;
 887         unix_release_addr(addr);
 888         goto out_up;
 889 }
 890
 891 static void unix_state_double_lock(struct sock *sk1, struct sock *sk2)
 892 {
 893         if (unlikely(sk1 == sk2) || !sk2) {
 894                 unix_state_lock(sk1);
 895                 return;
 896         }
 897         if (sk1 < sk2) {
 898                 unix_state_lock(sk1);
 899                 unix_state_lock_nested(sk2);
 900         } else {
 901                 unix_state_lock(sk2);
 902                 unix_state_lock_nested(sk1);
 903         }
 904 }
 905
 906 static void unix_state_double_unlock(struct sock *sk1, struct sock *sk2)
 907 {
 908         if (unlikely(sk1 == sk2) || !sk2) {
 909                 unix_state_unlock(sk1);
 910                 return;
 911         }
 912         unix_state_unlock(sk1);
 913         unix_state_unlock(sk2);
 914 }
 915
 916 static int unix_dgram_connect(struct socket *sock, struct sockaddr *addr,
 917                               int alen, int flags)
 918 {
 919         struct sock *sk = sock->sk;
 920         struct net *net = sock_net(sk);
 921         struct sockaddr_un *sunaddr = (struct sockaddr_un *)addr;
 922         struct sock *other;
 923         unsigned hash;
 924         int err;
 925
 926         if (addr->sa_family != AF_UNSPEC) {
 927                 err = unix_mkname(sunaddr, alen, &hash);
 928                 if (err < 0)
 929                         goto out;
 930                 alen = err;
 931
 932                 if (test_bit(SOCK_PASSCRED, &sock->flags) &&
 933                     !unix_sk(sk)->addr && (err = unix_autobind(sock)) != 0)
 934                         goto out;
 935
 936 restart:
 937                 other = unix_find_other(net, sunaddr, alen, sock->type, hash, &err);
 938                 if (!other)
 939                         goto out;
 940
 941                 unix_state_double_lock(sk, other);
 942
 943                 /* Apparently VFS overslept socket death. Retry. */
 944                 if (sock_flag(other, SOCK_DEAD)) {
 945                         unix_state_double_unlock(sk, other);
 946                         sock_put(other);
 947                         goto restart;
 948                 }
 949
 950                 err = -EPERM;
 951                 if (!unix_may_send(sk, other))
 952                         goto out_unlock;
 953
 954                 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
 955                 if (err)
 956                         goto out_unlock;
 957
 958         } else {
 959                 /*
 960                  *      1003.1g breaking connected state with AF_UNSPEC
 961                  */
 962                 other = NULL;
 963                 unix_state_double_lock(sk, other);
 964         }
 965
 966         /*
 967          * If it was connected, reconnect.
 968          */
 969         if (unix_peer(sk)) {
 970                 struct sock *old_peer = unix_peer(sk);
 971                 unix_peer(sk) = other;
 972                 unix_state_double_unlock(sk, other);
 973
 974                 if (other != old_peer)
 975                         unix_dgram_disconnected(sk, old_peer);
 976                 sock_put(old_peer);
 977         } else {
 978                 unix_peer(sk) = other;
 979                 unix_state_double_unlock(sk, other);
 980         }
 981         return 0;
 982
 983 out_unlock:
 984         unix_state_double_unlock(sk, other);
 985         sock_put(other);
 986 out:
 987         return err;
 988 }
 989
 990 static long unix_wait_for_peer(struct sock *other, long timeo)
 991 {
 992         struct unix_sock *u = unix_sk(other);
 993         int sched;
 994         DEFINE_WAIT(wait);
 995
 996         prepare_to_wait_exclusive(&u->peer_wait, &wait, TASK_INTERRUPTIBLE);
 997
 998         sched = !sock_flag(other, SOCK_DEAD) &&
 999                 !(other->sk_shutdown & RCV_SHUTDOWN) &&
1000                 unix_recvq_full(other);
1001
1002         unix_state_unlock(other);
1003
1004         if (sched)
1005                 timeo = schedule_timeout(timeo);
1006
1007         finish_wait(&u->peer_wait, &wait);
1008         return timeo;
1009 }
1010
1011 static int unix_stream_connect(struct socket *sock, struct sockaddr *uaddr,
1012                                int addr_len, int flags)
1013 {
1014         struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
1015         struct sock *sk = sock->sk;
1016         struct net *net = sock_net(sk);
1017         struct unix_sock *u = unix_sk(sk), *newu, *otheru;
1018         struct sock *newsk = NULL;
1019         struct sock *other = NULL;
1020         struct sk_buff *skb = NULL;
1021         unsigned hash;
1022         int st;
1023         int err;
1024         long timeo;
1025
1026         err = unix_mkname(sunaddr, addr_len, &hash);
1027         if (err < 0)
1028                 goto out;
1029         addr_len = err;
1030
1031         if (test_bit(SOCK_PASSCRED, &sock->flags)
1032                 && !u->addr && (err = unix_autobind(sock)) != 0)
1033                 goto out;
1034
1035         timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
1036
1037         /* First of all allocate resources.
1038            If we will make it after state is locked,
1039            we will have to recheck all again in any case.
1040          */
1041
1042         err = -ENOMEM;
1043
1044         /* create new sock for complete connection */
1045         newsk = unix_create1(sock_net(sk), NULL);
1046         if (newsk == NULL)
1047                 goto out;
1048
1049         /* Allocate skb for sending to listening sock */
1050         skb = sock_wmalloc(newsk, 1, 0, GFP_KERNEL);
1051         if (skb == NULL)
1052                 goto out;
1053
1054 restart:
1055         /*  Find listening sock. */
1056         other = unix_find_other(net, sunaddr, addr_len, sk->sk_type, hash, &err);
1057         if (!other)
1058                 goto out;
1059
1060         /* Latch state of peer */
1061         unix_state_lock(other);
1062
1063         /* Apparently VFS overslept socket death. Retry. */
1064         if (sock_flag(other, SOCK_DEAD)) {
1065                 unix_state_unlock(other);
1066                 sock_put(other);
1067                 goto restart;
1068         }
1069
1070         err = -ECONNREFUSED;
1071         if (other->sk_state != TCP_LISTEN)
1072                 goto out_unlock;
1073
1074         if (unix_recvq_full(other)) {
1075                 err = -EAGAIN;
1076                 if (!timeo)
1077                         goto out_unlock;
1078
1079                 timeo = unix_wait_for_peer(other, timeo);
1080
1081                 err = sock_intr_errno(timeo);
1082                 if (signal_pending(current))
1083                         goto out;
1084                 sock_put(other);
1085                 goto restart;
1086         }
1087
1088         /* Latch our state.
1089
1090            It is tricky place. We need to grab write lock and cannot
1091            drop lock on peer. It is dangerous because deadlock is
1092            possible. Connect to self case and simultaneous
1093            attempt to connect are eliminated by checking socket
1094            state. other is TCP_LISTEN, if sk is TCP_LISTEN we
1095            check this before attempt to grab lock.
1096
1097            Well, and we have to recheck the state after socket locked.
1098          */
1099         st = sk->sk_state;
1100
1101         switch (st) {
1102         case TCP_CLOSE:
1103                 /* This is ok... continue with connect */
1104                 break;
1105         case TCP_ESTABLISHED:
1106                 /* Socket is already connected */
1107                 err = -EISCONN;
1108                 goto out_unlock;
1109         default:
1110                 err = -EINVAL;
1111                 goto out_unlock;
1112         }
1113
1114         unix_state_lock_nested(sk);
1115
1116         if (sk->sk_state != st) {
1117                 unix_state_unlock(sk);
1118                 unix_state_unlock(other);
1119                 sock_put(other);
1120                 goto restart;
1121         }
1122
1123         err = security_unix_stream_connect(sock, other->sk_socket, newsk);
1124         if (err) {
1125                 unix_state_unlock(sk);
1126                 goto out_unlock;
1127         }
1128
1129         /* The way is open! Fastly set all the necessary fields... */
1130
1131         sock_hold(sk);
1132         unix_peer(newsk)        = sk;
1133         newsk->sk_state         = TCP_ESTABLISHED;
1134         newsk->sk_type          = sk->sk_type;
1135         newsk->sk_peercred.pid  = task_tgid_vnr(current);
1136         current_euid_egid(&newsk->sk_peercred.uid, &newsk->sk_peercred.gid);
1137         newu = unix_sk(newsk);
1138         newsk->sk_sleep         = &newu->peer_wait;
1139         otheru = unix_sk(other);
1140
1141         /* copy address information from listening to new sock*/
1142         if (otheru->addr) {
1143                 atomic_inc(&otheru->addr->refcnt);
1144                 newu->addr = otheru->addr;
1145         }
1146         if (otheru->dentry) {
1147                 newu->dentry    = dget(otheru->dentry);
1148                 newu->mnt       = mntget(otheru->mnt);
1149         }
1150
1151         /* Set credentials */
1152         sk->sk_peercred = other->sk_peercred;
1153
1154         sock->state     = SS_CONNECTED;
1155         sk->sk_state    = TCP_ESTABLISHED;
1156         sock_hold(newsk);
1157
1158         smp_mb__after_atomic_inc();     /* sock_hold() does an atomic_inc() */
1159         unix_peer(sk)   = newsk;
1160
1161         unix_state_unlock(sk);
1162
1163         /* take ten and and send info to listening sock */
1164         spin_lock(&other->sk_receive_queue.lock);
1165         __skb_queue_tail(&other->sk_receive_queue, skb);
1166         spin_unlock(&other->sk_receive_queue.lock);
1167         unix_state_unlock(other);
1168         other->sk_data_ready(other, 0);
1169         sock_put(other);
1170         return 0;
1171
1172 out_unlock:
1173         if (other)
1174                 unix_state_unlock(other);
1175
1176 out:
1177         if (skb)
1178                 kfree_skb(skb);
1179         if (newsk)
1180                 unix_release_sock(newsk, 0);
1181         if (other)
1182                 sock_put(other);
1183         return err;
1184 }
1185
1186 static int unix_socketpair(struct socket *socka, struct socket *sockb)
1187 {
1188         struct sock *ska = socka->sk, *skb = sockb->sk;
1189
1190         /* Join our sockets back to back */
1191         sock_hold(ska);
1192         sock_hold(skb);
1193         unix_peer(ska) = skb;
1194         unix_peer(skb) = ska;
1195         ska->sk_peercred.pid = skb->sk_peercred.pid = task_tgid_vnr(current);
1196         current_euid_egid(&skb->sk_peercred.uid, &skb->sk_peercred.gid);
1197         ska->sk_peercred.uid = skb->sk_peercred.uid;
1198         ska->sk_peercred.gid = skb->sk_peercred.gid;
1199
1200         if (ska->sk_type != SOCK_DGRAM) {
1201                 ska->sk_state = TCP_ESTABLISHED;
1202                 skb->sk_state = TCP_ESTABLISHED;
1203                 socka->state  = SS_CONNECTED;
1204                 sockb->state  = SS_CONNECTED;
1205         }
1206         return 0;
1207 }
1208
1209 static int unix_accept(struct socket *sock, struct socket *newsock, int flags)
1210 {
1211         struct sock *sk = sock->sk;
1212         struct sock *tsk;
1213         struct sk_buff *skb;
1214         int err;
1215
1216         err = -EOPNOTSUPP;
1217         if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
1218                 goto out;
1219
1220         err = -EINVAL;
1221         if (sk->sk_state != TCP_LISTEN)
1222                 goto out;
1223
1224         /* If socket state is TCP_LISTEN it cannot change (for now...),
1225          * so that no locks are necessary.
1226          */
1227
1228         skb = skb_recv_datagram(sk, 0, flags&O_NONBLOCK, &err);
1229         if (!skb) {
1230                 /* This means receive shutdown. */
1231                 if (err == 0)
1232                         err = -EINVAL;
1233                 goto out;
1234         }
1235
1236         tsk = skb->sk;
1237         skb_free_datagram(sk, skb);
1238         wake_up_interruptible(&unix_sk(sk)->peer_wait);
1239
1240         /* attach accepted sock to socket */
1241         unix_state_lock(tsk);
1242         newsock->state = SS_CONNECTED;
1243         sock_graft(tsk, newsock);
1244         unix_state_unlock(tsk);
1245         return 0;
1246
1247 out:
1248         return err;
1249 }
1250
1251
1252 static int unix_getname(struct socket *sock, struct sockaddr *uaddr, int *uaddr_len, int peer)
1253 {
1254         struct sock *sk = sock->sk;
1255         struct unix_sock *u;
1256         struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
1257         int err = 0;
1258
1259         if (peer) {
1260                 sk = unix_peer_get(sk);
1261
1262                 err = -ENOTCONN;
1263                 if (!sk)
1264                         goto out;
1265                 err = 0;
1266         } else {
1267                 sock_hold(sk);
1268         }
1269
1270         u = unix_sk(sk);
1271         unix_state_lock(sk);
1272         if (!u->addr) {
1273                 sunaddr->sun_family = AF_UNIX;
1274                 sunaddr->sun_path[0] = 0;
1275                 *uaddr_len = sizeof(short);
1276         } else {
1277                 struct unix_address *addr = u->addr;
1278
1279                 *uaddr_len = addr->len;
1280                 memcpy(sunaddr, addr->name, *uaddr_len);
1281         }
1282         unix_state_unlock(sk);
1283         sock_put(sk);
1284 out:
1285         return err;
1286 }
1287
1288 static void unix_detach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1289 {
1290         int i;
1291
1292         scm->fp = UNIXCB(skb).fp;
1293         skb->destructor = sock_wfree;
1294         UNIXCB(skb).fp = NULL;
1295
1296         for (i = scm->fp->count-1; i >= 0; i--)
1297                 unix_notinflight(scm->fp->fp[i]);
1298 }
1299
1300 static void unix_destruct_fds(struct sk_buff *skb)
1301 {
1302         struct scm_cookie scm;
1303         memset(&scm, 0, sizeof(scm));
1304         unix_detach_fds(&scm, skb);
1305
1306         /* Alas, it calls VFS */
1307         /* So fscking what? fput() had been SMP-safe since the last Summer */
1308         scm_destroy(&scm);
1309         sock_wfree(skb);
1310 }
1311
1312 static int unix_attach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1313 {
1314         int i;
1315
1316         /*
1317          * Need to duplicate file references for the sake of garbage
1318          * collection.  Otherwise a socket in the fps might become a
1319          * candidate for GC while the skb is not yet queued.
1320          */
1321         UNIXCB(skb).fp = scm_fp_dup(scm->fp);
1322         if (!UNIXCB(skb).fp)
1323                 return -ENOMEM;
1324
1325         for (i = scm->fp->count-1; i >= 0; i--)
1326                 unix_inflight(scm->fp->fp[i]);
1327         skb->destructor = unix_destruct_fds;
1328         return 0;
1329 }
1330
1331 /*
1332  *      Send AF_UNIX data.
1333  */
1334
1335 static int unix_dgram_sendmsg(struct kiocb *kiocb, struct socket *sock,
1336                               struct msghdr *msg, size_t len)
1337 {
1338         struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1339         struct sock *sk = sock->sk;
1340         struct net *net = sock_net(sk);
1341         struct unix_sock *u = unix_sk(sk);
1342         struct sockaddr_un *sunaddr = msg->msg_name;
1343         struct sock *other = NULL;
1344         int namelen = 0; /* fake GCC */
1345         int err;
1346         unsigned hash;
1347         struct sk_buff *skb;
1348         long timeo;
1349         struct scm_cookie tmp_scm;
1350
1351         if (NULL == siocb->scm)
1352                 siocb->scm = &tmp_scm;
1353         wait_for_unix_gc();
1354         err = scm_send(sock, msg, siocb->scm);
1355         if (err < 0)
1356                 return err;
1357
1358         err = -EOPNOTSUPP;
1359         if (msg->msg_flags&MSG_OOB)
1360                 goto out;
1361
1362         if (msg->msg_namelen) {
1363                 err = unix_mkname(sunaddr, msg->msg_namelen, &hash);
1364                 if (err < 0)
1365                         goto out;
1366                 namelen = err;
1367         } else {
1368                 sunaddr = NULL;
1369                 err = -ENOTCONN;
1370                 other = unix_peer_get(sk);
1371                 if (!other)
1372                         goto out;
1373         }
1374
1375         if (test_bit(SOCK_PASSCRED, &sock->flags)
1376                 && !u->addr && (err = unix_autobind(sock)) != 0)
1377                 goto out;
1378
1379         err = -EMSGSIZE;
1380         if (len > sk->sk_sndbuf - 32)
1381                 goto out;
1382
1383         skb = sock_alloc_send_skb(sk, len, msg->msg_flags&MSG_DONTWAIT, &err);
1384         if (skb == NULL)
1385                 goto out;
1386
1387         memcpy(UNIXCREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1388         if (siocb->scm->fp) {
1389                 err = unix_attach_fds(siocb->scm, skb);
1390                 if (err)
1391                         goto out_free;
1392         }
1393         unix_get_secdata(siocb->scm, skb);
1394
1395         skb_reset_transport_header(skb);
1396         err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
1397         if (err)
1398                 goto out_free;
1399
1400         timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1401
1402 restart:
1403         if (!other) {
1404                 err = -ECONNRESET;
1405                 if (sunaddr == NULL)
1406                         goto out_free;
1407
1408                 other = unix_find_other(net, sunaddr, namelen, sk->sk_type,
1409                                         hash, &err);
1410                 if (other == NULL)
1411                         goto out_free;
1412         }
1413
1414         unix_state_lock(other);
1415         err = -EPERM;
1416         if (!unix_may_send(sk, other))
1417                 goto out_unlock;
1418
1419         if (sock_flag(other, SOCK_DEAD)) {
1420                 /*
1421                  *      Check with 1003.1g - what should
1422                  *      datagram error
1423                  */
1424                 unix_state_unlock(other);
1425                 sock_put(other);
1426
1427                 err = 0;
1428                 unix_state_lock(sk);
1429                 if (unix_peer(sk) == other) {
1430                         unix_peer(sk) = NULL;
1431                         unix_state_unlock(sk);
1432
1433                         unix_dgram_disconnected(sk, other);
1434                         sock_put(other);
1435                         err = -ECONNREFUSED;
1436                 } else {
1437                         unix_state_unlock(sk);
1438                 }
1439
1440                 other = NULL;
1441                 if (err)
1442                         goto out_free;
1443                 goto restart;
1444         }
1445
1446         err = -EPIPE;
1447         if (other->sk_shutdown & RCV_SHUTDOWN)
1448                 goto out_unlock;
1449
1450         if (sk->sk_type != SOCK_SEQPACKET) {
1451                 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1452                 if (err)
1453                         goto out_unlock;
1454         }
1455
1456         if (unix_peer(other) != sk && unix_recvq_full(other)) {
1457                 if (!timeo) {
1458                         err = -EAGAIN;
1459                         goto out_unlock;
1460                 }
1461
1462                 timeo = unix_wait_for_peer(other, timeo);
1463
1464                 err = sock_intr_errno(timeo);
1465                 if (signal_pending(current))
1466                         goto out_free;
1467
1468                 goto restart;
1469         }
1470
1471         skb_queue_tail(&other->sk_receive_queue, skb);
1472         unix_state_unlock(other);
1473         other->sk_data_ready(other, len);
1474         sock_put(other);
1475         scm_destroy(siocb->scm);
1476         return len;
1477
1478 out_unlock:
1479         unix_state_unlock(other);
1480 out_free:
1481         kfree_skb(skb);
1482 out:
1483         if (other)
1484                 sock_put(other);
1485         scm_destroy(siocb->scm);
1486         return err;
1487 }
1488
1489
1490 static int unix_stream_sendmsg(struct kiocb *kiocb, struct socket *sock,
1491                                struct msghdr *msg, size_t len)
1492 {
1493         struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1494         struct sock *sk = sock->sk;
1495         struct sock *other = NULL;
1496         struct sockaddr_un *sunaddr = msg->msg_name;
1497         int err, size;
1498         struct sk_buff *skb;
1499         int sent = 0;
1500         struct scm_cookie tmp_scm;
1501
1502         if (NULL == siocb->scm)
1503                 siocb->scm = &tmp_scm;
1504         wait_for_unix_gc();
1505         err = scm_send(sock, msg, siocb->scm);
1506         if (err < 0)
1507                 return err;
1508
1509         err = -EOPNOTSUPP;
1510         if (msg->msg_flags&MSG_OOB)
1511                 goto out_err;
1512
1513         if (msg->msg_namelen) {
1514                 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
1515                 goto out_err;
1516         } else {
1517                 sunaddr = NULL;
1518                 err = -ENOTCONN;
1519                 other = unix_peer(sk);
1520                 if (!other)
1521                         goto out_err;
1522         }
1523
1524         if (sk->sk_shutdown & SEND_SHUTDOWN)
1525                 goto pipe_err;
1526
1527         while (sent < len) {
1528                 /*
1529                  *      Optimisation for the fact that under 0.01% of X
1530                  *      messages typically need breaking up.
1531                  */
1532
1533                 size = len-sent;
1534
1535                 /* Keep two messages in the pipe so it schedules better */
1536                 if (size > ((sk->sk_sndbuf >> 1) - 64))
1537                         size = (sk->sk_sndbuf >> 1) - 64;
1538
1539                 if (size > SKB_MAX_ALLOC)
1540                         size = SKB_MAX_ALLOC;
1541
1542                 /*
1543                  *      Grab a buffer
1544                  */
1545
1546                 skb = sock_alloc_send_skb(sk, size, msg->msg_flags&MSG_DONTWAIT,
1547                                           &err);
1548
1549                 if (skb == NULL)
1550                         goto out_err;
1551
1552                 /*
1553                  *      If you pass two values to the sock_alloc_send_skb
1554                  *      it tries to grab the large buffer with GFP_NOFS
1555                  *      (which can fail easily), and if it fails grab the
1556                  *      fallback size buffer which is under a page and will
1557                  *      succeed. [Alan]
1558                  */
1559                 size = min_t(int, size, skb_tailroom(skb));
1560
1561                 memcpy(UNIXCREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1562                 if (siocb->scm->fp) {
1563                         err = unix_attach_fds(siocb->scm, skb);
1564                         if (err) {
1565                                 kfree_skb(skb);
1566                                 goto out_err;
1567                         }
1568                 }
1569
1570                 err = memcpy_fromiovec(skb_put(skb, size), msg->msg_iov, size);
1571                 if (err) {
1572                         kfree_skb(skb);
1573                         goto out_err;
1574                 }
1575
1576                 unix_state_lock(other);
1577
1578                 if (sock_flag(other, SOCK_DEAD) ||
1579                     (other->sk_shutdown & RCV_SHUTDOWN))
1580                         goto pipe_err_free;
1581
1582                 skb_queue_tail(&other->sk_receive_queue, skb);
1583                 unix_state_unlock(other);
1584                 other->sk_data_ready(other, size);
1585                 sent += size;
1586         }
1587
1588         scm_destroy(siocb->scm);
1589         siocb->scm = NULL;
1590
1591         return sent;
1592
1593 pipe_err_free:
1594         unix_state_unlock(other);
1595         kfree_skb(skb);
1596 pipe_err:
1597         if (sent == 0 && !(msg->msg_flags&MSG_NOSIGNAL))
1598                 send_sig(SIGPIPE, current, 0);
1599         err = -EPIPE;
1600 out_err:
1601         scm_destroy(siocb->scm);
1602         siocb->scm = NULL;
1603         return sent ? : err;
1604 }
1605
1606 static int unix_seqpacket_sendmsg(struct kiocb *kiocb, struct socket *sock,
1607                                   struct msghdr *msg, size_t len)
1608 {
1609         int err;
1610         struct sock *sk = sock->sk;
1611
1612         err = sock_error(sk);
1613         if (err)
1614                 return err;
1615
1616         if (sk->sk_state != TCP_ESTABLISHED)
1617                 return -ENOTCONN;
1618
1619         if (msg->msg_namelen)
1620                 msg->msg_namelen = 0;
1621
1622         return unix_dgram_sendmsg(kiocb, sock, msg, len);
1623 }
1624
1625 static void unix_copy_addr(struct msghdr *msg, struct sock *sk)
1626 {
1627         struct unix_sock *u = unix_sk(sk);
1628
1629         msg->msg_namelen = 0;
1630         if (u->addr) {
1631                 msg->msg_namelen = u->addr->len;
1632                 memcpy(msg->msg_name, u->addr->name, u->addr->len);
1633         }
1634 }
1635
1636 static int unix_dgram_recvmsg(struct kiocb *iocb, struct socket *sock,
1637                               struct msghdr *msg, size_t size,
1638                               int flags)
1639 {
1640         struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1641         struct scm_cookie tmp_scm;
1642         struct sock *sk = sock->sk;
1643         struct unix_sock *u = unix_sk(sk);
1644         int noblock = flags & MSG_DONTWAIT;
1645         struct sk_buff *skb;
1646         int err;
1647
1648         err = -EOPNOTSUPP;
1649         if (flags&MSG_OOB)
1650                 goto out;
1651
1652         msg->msg_namelen = 0;
1653
1654         mutex_lock(&u->readlock);
1655
1656         skb = skb_recv_datagram(sk, flags, noblock, &err);
1657         if (!skb) {
1658                 unix_state_lock(sk);
1659                 /* Signal EOF on disconnected non-blocking SEQPACKET socket. */
1660                 if (sk->sk_type == SOCK_SEQPACKET && err == -EAGAIN &&
1661                     (sk->sk_shutdown & RCV_SHUTDOWN))
1662                         err = 0;
1663                 unix_state_unlock(sk);
1664                 goto out_unlock;
1665         }
1666
1667         wake_up_interruptible_sync(&u->peer_wait);
1668
1669         if (msg->msg_name)
1670                 unix_copy_addr(msg, skb->sk);
1671
1672         if (size > skb->len)
1673                 size = skb->len;
1674         else if (size < skb->len)
1675                 msg->msg_flags |= MSG_TRUNC;
1676
1677         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, size);
1678         if (err)
1679                 goto out_free;
1680
1681         if (!siocb->scm) {
1682                 siocb->scm = &tmp_scm;
1683                 memset(&tmp_scm, 0, sizeof(tmp_scm));
1684         }
1685         siocb->scm->creds = *UNIXCREDS(skb);
1686         unix_set_secdata(siocb->scm, skb);
1687
1688         if (!(flags & MSG_PEEK)) {
1689                 if (UNIXCB(skb).fp)
1690                         unix_detach_fds(siocb->scm, skb);
1691         } else {
1692                 /* It is questionable: on PEEK we could:
1693                    - do not return fds - good, but too simple 8)
1694                    - return fds, and do not return them on read (old strategy,
1695                      apparently wrong)
1696                    - clone fds (I chose it for now, it is the most universal
1697                      solution)
1698
1699                    POSIX 1003.1g does not actually define this clearly
1700                    at all. POSIX 1003.1g doesn't define a lot of things
1701                    clearly however!
1702
1703                 */
1704                 if (UNIXCB(skb).fp)
1705                         siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1706         }
1707         err = size;
1708
1709         scm_recv(sock, msg, siocb->scm, flags);
1710
1711 out_free:
1712         skb_free_datagram(sk, skb);
1713 out_unlock:
1714         mutex_unlock(&u->readlock);
1715 out:
1716         return err;
1717 }
1718
1719 /*
1720  *      Sleep until data has arrive. But check for races..
1721  */
1722
1723 static long unix_stream_data_wait(struct sock *sk, long timeo)
1724 {
1725         DEFINE_WAIT(wait);
1726
1727         unix_state_lock(sk);
1728
1729         for (;;) {
1730                 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1731
1732                 if (!skb_queue_empty(&sk->sk_receive_queue) ||
1733                     sk->sk_err ||
1734                     (sk->sk_shutdown & RCV_SHUTDOWN) ||
1735                     signal_pending(current) ||
1736                     !timeo)
1737                         break;
1738
1739                 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1740                 unix_state_unlock(sk);
1741                 timeo = schedule_timeout(timeo);
1742                 unix_state_lock(sk);
1743                 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1744         }
1745
1746         finish_wait(sk->sk_sleep, &wait);
1747         unix_state_unlock(sk);
1748         return timeo;
1749 }
1750
1751
1752
1753 static int unix_stream_recvmsg(struct kiocb *iocb, struct socket *sock,
1754                                struct msghdr *msg, size_t size,
1755                                int flags)
1756 {
1757         struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1758         struct scm_cookie tmp_scm;
1759         struct sock *sk = sock->sk;
1760         struct unix_sock *u = unix_sk(sk);
1761         struct sockaddr_un *sunaddr = msg->msg_name;
1762         int copied = 0;
1763         int check_creds = 0;
1764         int target;
1765         int err = 0;
1766         long timeo;
1767
1768         err = -EINVAL;
1769         if (sk->sk_state != TCP_ESTABLISHED)
1770                 goto out;
1771
1772         err = -EOPNOTSUPP;
1773         if (flags&MSG_OOB)
1774                 goto out;
1775
1776         target = sock_rcvlowat(sk, flags&MSG_WAITALL, size);
1777         timeo = sock_rcvtimeo(sk, flags&MSG_DONTWAIT);
1778
1779         msg->msg_namelen = 0;
1780
1781         /* Lock the socket to prevent queue disordering
1782          * while sleeps in memcpy_tomsg
1783          */
1784
1785         if (!siocb->scm) {
1786                 siocb->scm = &tmp_scm;
1787                 memset(&tmp_scm, 0, sizeof(tmp_scm));
1788         }
1789
1790         mutex_lock(&u->readlock);
1791
1792         do {
1793                 int chunk;
1794                 struct sk_buff *skb;
1795
1796                 unix_state_lock(sk);
1797                 skb = skb_dequeue(&sk->sk_receive_queue);
1798                 if (skb == NULL) {
1799                         if (copied >= target)
1800                                 goto unlock;
1801
1802                         /*
1803                          *      POSIX 1003.1g mandates this order.
1804                          */
1805
1806                         err = sock_error(sk);
1807                         if (err)
1808                                 goto unlock;
1809                         if (sk->sk_shutdown & RCV_SHUTDOWN)
1810                                 goto unlock;
1811
1812                         unix_state_unlock(sk);
1813                         err = -EAGAIN;
1814                         if (!timeo)
1815                                 break;
1816                         mutex_unlock(&u->readlock);
1817
1818                         timeo = unix_stream_data_wait(sk, timeo);
1819
1820                         if (signal_pending(current)) {
1821                                 err = sock_intr_errno(timeo);
1822                                 goto out;
1823                         }
1824                         mutex_lock(&u->readlock);
1825                         continue;
1826  unlock:
1827                         unix_state_unlock(sk);
1828                         break;
1829                 }
1830                 unix_state_unlock(sk);
1831
1832                 if (check_creds) {
1833                         /* Never glue messages from different writers */
1834                         if (memcmp(UNIXCREDS(skb), &siocb->scm->creds,
1835                                    sizeof(siocb->scm->creds)) != 0) {
1836                                 skb_queue_head(&sk->sk_receive_queue, skb);
1837                                 break;
1838                         }
1839                 } else {
1840                         /* Copy credentials */
1841                         siocb->scm->creds = *UNIXCREDS(skb);
1842                         check_creds = 1;
1843                 }
1844
1845                 /* Copy address just once */
1846                 if (sunaddr) {
1847                         unix_copy_addr(msg, skb->sk);
1848                         sunaddr = NULL;
1849                 }
1850
1851                 chunk = min_t(unsigned int, skb->len, size);
1852                 if (memcpy_toiovec(msg->msg_iov, skb->data, chunk)) {
1853                         skb_queue_head(&sk->sk_receive_queue, skb);
1854                         if (copied == 0)
1855                                 copied = -EFAULT;
1856                         break;
1857                 }
1858                 copied += chunk;
1859                 size -= chunk;
1860
1861                 /* Mark read part of skb as used */
1862                 if (!(flags & MSG_PEEK)) {
1863                         skb_pull(skb, chunk);
1864
1865                         if (UNIXCB(skb).fp)
1866                                 unix_detach_fds(siocb->scm, skb);
1867
1868                         /* put the skb back if we didn't use it up.. */
1869                         if (skb->len) {
1870                                 skb_queue_head(&sk->sk_receive_queue, skb);
1871                                 break;
1872                         }
1873
1874                         kfree_skb(skb);
1875
1876                         if (siocb->scm->fp)
1877                                 break;
1878                 } else {
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         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) &&
2003             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
2100 static struct sock *unix_seq_idx(struct seq_file *seq, loff_t pos)
2101 {
2102         struct unix_iter_state *iter = seq->private;
2103         loff_t off = 0;
2104         struct sock *s;
2105
2106         for (s = first_unix_socket(&iter->i); s; s = next_unix_socket(&iter->i, s)) {
2107                 if (sock_net(s) != seq_file_net(seq))
2108                         continue;
2109                 if (off == pos)
2110                         return s;
2111                 ++off;
2112         }
2113         return NULL;
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 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);