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_has_sleeper(sk))
 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                        int kern)
 626 {
 627         if (protocol && protocol != PF_UNIX)
 628                 return -EPROTONOSUPPORT;
 629
 630         sock->state = SS_UNCONNECTED;
 631
 632         switch (sock->type) {
 633         case SOCK_STREAM:
 634                 sock->ops = &unix_stream_ops;
 635                 break;
 636                 /*
 637                  *      Believe it or not BSD has AF_UNIX, SOCK_RAW though
 638                  *      nothing uses it.
 639                  */
 640         case SOCK_RAW:
 641                 sock->type = SOCK_DGRAM;
 642         case SOCK_DGRAM:
 643                 sock->ops = &unix_dgram_ops;
 644                 break;
 645         case SOCK_SEQPACKET:
 646                 sock->ops = &unix_seqpacket_ops;
 647                 break;
 648         default:
 649                 return -ESOCKTNOSUPPORT;
 650         }
 651
 652         return unix_create1(net, sock) ? 0 : -ENOMEM;
 653 }
 654
 655 static int unix_release(struct socket *sock)
 656 {
 657         struct sock *sk = sock->sk;
 658
 659         if (!sk)
 660                 return 0;
 661
 662         sock->sk = NULL;
 663
 664         return unix_release_sock(sk, 0);
 665 }
 666
 667 static int unix_autobind(struct socket *sock)
 668 {
 669         struct sock *sk = sock->sk;
 670         struct net *net = sock_net(sk);
 671         struct unix_sock *u = unix_sk(sk);
 672         static u32 ordernum = 1;
 673         struct unix_address *addr;
 674         int err;
 675
 676         mutex_lock(&u->readlock);
 677
 678         err = 0;
 679         if (u->addr)
 680                 goto out;
 681
 682         err = -ENOMEM;
 683         addr = kzalloc(sizeof(*addr) + sizeof(short) + 16, GFP_KERNEL);
 684         if (!addr)
 685                 goto out;
 686
 687         addr->name->sun_family = AF_UNIX;
 688         atomic_set(&addr->refcnt, 1);
 689
 690 retry:
 691         addr->len = sprintf(addr->name->sun_path+1, "%05x", ordernum) + 1 + sizeof(short);
 692         addr->hash = unix_hash_fold(csum_partial(addr->name, addr->len, 0));
 693
 694         spin_lock(&unix_table_lock);
 695         ordernum = (ordernum+1)&0xFFFFF;
 696
 697         if (__unix_find_socket_byname(net, addr->name, addr->len, sock->type,
 698                                       addr->hash)) {
 699                 spin_unlock(&unix_table_lock);
 700                 /* Sanity yield. It is unusual case, but yet... */
 701                 if (!(ordernum&0xFF))
 702                         yield();
 703                 goto retry;
 704         }
 705         addr->hash ^= sk->sk_type;
 706
 707         __unix_remove_socket(sk);
 708         u->addr = addr;
 709         __unix_insert_socket(&unix_socket_table[addr->hash], sk);
 710         spin_unlock(&unix_table_lock);
 711         err = 0;
 712
 713 out:    mutex_unlock(&u->readlock);
 714         return err;
 715 }
 716
 717 static struct sock *unix_find_other(struct net *net,
 718                                     struct sockaddr_un *sunname, int len,
 719                                     int type, unsigned hash, int *error)
 720 {
 721         struct sock *u;
 722         struct path path;
 723         int err = 0;
 724
 725         if (sunname->sun_path[0]) {
 726                 struct inode *inode;
 727                 err = kern_path(sunname->sun_path, LOOKUP_FOLLOW, &path);
 728                 if (err)
 729                         goto fail;
 730                 inode = path.dentry->d_inode;
 731                 err = inode_permission(inode, MAY_WRITE);
 732                 if (err)
 733                         goto put_fail;
 734
 735                 err = -ECONNREFUSED;
 736                 if (!S_ISSOCK(inode->i_mode))
 737                         goto put_fail;
 738                 u = unix_find_socket_byinode(net, inode);
 739                 if (!u)
 740                         goto put_fail;
 741
 742                 if (u->sk_type == type)
 743                         touch_atime(path.mnt, path.dentry);
 744
 745                 path_put(&path);
 746
 747                 err = -EPROTOTYPE;
 748                 if (u->sk_type != type) {
 749                         sock_put(u);
 750                         goto fail;
 751                 }
 752         } else {
 753                 err = -ECONNREFUSED;
 754                 u = unix_find_socket_byname(net, sunname, len, type, hash);
 755                 if (u) {
 756                         struct dentry *dentry;
 757                         dentry = unix_sk(u)->dentry;
 758                         if (dentry)
 759                                 touch_atime(unix_sk(u)->mnt, dentry);
 760                 } else
 761                         goto fail;
 762         }
 763         return u;
 764
 765 put_fail:
 766         path_put(&path);
 767 fail:
 768         *error = err;
 769         return NULL;
 770 }
 771
 772
 773 static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
 774 {
 775         struct sock *sk = sock->sk;
 776         struct net *net = sock_net(sk);
 777         struct unix_sock *u = unix_sk(sk);
 778         struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
 779         struct dentry *dentry = NULL;
 780         struct nameidata nd;
 781         int err;
 782         unsigned hash;
 783         struct unix_address *addr;
 784         struct hlist_head *list;
 785
 786         err = -EINVAL;
 787         if (sunaddr->sun_family != AF_UNIX)
 788                 goto out;
 789
 790         if (addr_len == sizeof(short)) {
 791                 err = unix_autobind(sock);
 792                 goto out;
 793         }
 794
 795         err = unix_mkname(sunaddr, addr_len, &hash);
 796         if (err < 0)
 797                 goto out;
 798         addr_len = err;
 799
 800         mutex_lock(&u->readlock);
 801
 802         err = -EINVAL;
 803         if (u->addr)
 804                 goto out_up;
 805
 806         err = -ENOMEM;
 807         addr = kmalloc(sizeof(*addr)+addr_len, GFP_KERNEL);
 808         if (!addr)
 809                 goto out_up;
 810
 811         memcpy(addr->name, sunaddr, addr_len);
 812         addr->len = addr_len;
 813         addr->hash = hash ^ sk->sk_type;
 814         atomic_set(&addr->refcnt, 1);
 815
 816         if (sunaddr->sun_path[0]) {
 817                 unsigned int mode;
 818                 err = 0;
 819                 /*
 820                  * Get the parent directory, calculate the hash for last
 821                  * component.
 822                  */
 823                 err = path_lookup(sunaddr->sun_path, LOOKUP_PARENT, &nd);
 824                 if (err)
 825                         goto out_mknod_parent;
 826
 827                 dentry = lookup_create(&nd, 0);
 828                 err = PTR_ERR(dentry);
 829                 if (IS_ERR(dentry))
 830                         goto out_mknod_unlock;
 831
 832                 /*
 833                  * All right, let's create it.
 834                  */
 835                 mode = S_IFSOCK |
 836                        (SOCK_INODE(sock)->i_mode & ~current_umask());
 837                 err = mnt_want_write(nd.path.mnt);
 838                 if (err)
 839                         goto out_mknod_dput;
 840                 err = security_path_mknod(&nd.path, dentry, mode, 0);
 841                 if (err)
 842                         goto out_mknod_drop_write;
 843                 err = vfs_mknod(nd.path.dentry->d_inode, dentry, mode, 0);
 844 out_mknod_drop_write:
 845                 mnt_drop_write(nd.path.mnt);
 846                 if (err)
 847                         goto out_mknod_dput;
 848                 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
 849                 dput(nd.path.dentry);
 850                 nd.path.dentry = dentry;
 851
 852                 addr->hash = UNIX_HASH_SIZE;
 853         }
 854
 855         spin_lock(&unix_table_lock);
 856
 857         if (!sunaddr->sun_path[0]) {
 858                 err = -EADDRINUSE;
 859                 if (__unix_find_socket_byname(net, sunaddr, addr_len,
 860                                               sk->sk_type, hash)) {
 861                         unix_release_addr(addr);
 862                         goto out_unlock;
 863                 }
 864
 865                 list = &unix_socket_table[addr->hash];
 866         } else {
 867                 list = &unix_socket_table[dentry->d_inode->i_ino & (UNIX_HASH_SIZE-1)];
 868                 u->dentry = nd.path.dentry;
 869                 u->mnt    = nd.path.mnt;
 870         }
 871
 872         err = 0;
 873         __unix_remove_socket(sk);
 874         u->addr = addr;
 875         __unix_insert_socket(list, sk);
 876
 877 out_unlock:
 878         spin_unlock(&unix_table_lock);
 879 out_up:
 880         mutex_unlock(&u->readlock);
 881 out:
 882         return err;
 883
 884 out_mknod_dput:
 885         dput(dentry);
 886 out_mknod_unlock:
 887         mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
 888         path_put(&nd.path);
 889 out_mknod_parent:
 890         if (err == -EEXIST)
 891                 err = -EADDRINUSE;
 892         unix_release_addr(addr);
 893         goto out_up;
 894 }
 895
 896 static void unix_state_double_lock(struct sock *sk1, struct sock *sk2)
 897 {
 898         if (unlikely(sk1 == sk2) || !sk2) {
 899                 unix_state_lock(sk1);
 900                 return;
 901         }
 902         if (sk1 < sk2) {
 903                 unix_state_lock(sk1);
 904                 unix_state_lock_nested(sk2);
 905         } else {
 906                 unix_state_lock(sk2);
 907                 unix_state_lock_nested(sk1);
 908         }
 909 }
 910
 911 static void unix_state_double_unlock(struct sock *sk1, struct sock *sk2)
 912 {
 913         if (unlikely(sk1 == sk2) || !sk2) {
 914                 unix_state_unlock(sk1);
 915                 return;
 916         }
 917         unix_state_unlock(sk1);
 918         unix_state_unlock(sk2);
 919 }
 920
 921 static int unix_dgram_connect(struct socket *sock, struct sockaddr *addr,
 922                               int alen, int flags)
 923 {
 924         struct sock *sk = sock->sk;
 925         struct net *net = sock_net(sk);
 926         struct sockaddr_un *sunaddr = (struct sockaddr_un *)addr;
 927         struct sock *other;
 928         unsigned hash;
 929         int err;
 930
 931         if (addr->sa_family != AF_UNSPEC) {
 932                 err = unix_mkname(sunaddr, alen, &hash);
 933                 if (err < 0)
 934                         goto out;
 935                 alen = err;
 936
 937                 if (test_bit(SOCK_PASSCRED, &sock->flags) &&
 938                     !unix_sk(sk)->addr && (err = unix_autobind(sock)) != 0)
 939                         goto out;
 940
 941 restart:
 942                 other = unix_find_other(net, sunaddr, alen, sock->type, hash, &err);
 943                 if (!other)
 944                         goto out;
 945
 946                 unix_state_double_lock(sk, other);
 947
 948                 /* Apparently VFS overslept socket death. Retry. */
 949                 if (sock_flag(other, SOCK_DEAD)) {
 950                         unix_state_double_unlock(sk, other);
 951                         sock_put(other);
 952                         goto restart;
 953                 }
 954
 955                 err = -EPERM;
 956                 if (!unix_may_send(sk, other))
 957                         goto out_unlock;
 958
 959                 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
 960                 if (err)
 961                         goto out_unlock;
 962
 963         } else {
 964                 /*
 965                  *      1003.1g breaking connected state with AF_UNSPEC
 966                  */
 967                 other = NULL;
 968                 unix_state_double_lock(sk, other);
 969         }
 970
 971         /*
 972          * If it was connected, reconnect.
 973          */
 974         if (unix_peer(sk)) {
 975                 struct sock *old_peer = unix_peer(sk);
 976                 unix_peer(sk) = other;
 977                 unix_state_double_unlock(sk, other);
 978
 979                 if (other != old_peer)
 980                         unix_dgram_disconnected(sk, old_peer);
 981                 sock_put(old_peer);
 982         } else {
 983                 unix_peer(sk) = other;
 984                 unix_state_double_unlock(sk, other);
 985         }
 986         return 0;
 987
 988 out_unlock:
 989         unix_state_double_unlock(sk, other);
 990         sock_put(other);
 991 out:
 992         return err;
 993 }
 994
 995 static long unix_wait_for_peer(struct sock *other, long timeo)
 996 {
 997         struct unix_sock *u = unix_sk(other);
 998         int sched;
 999         DEFINE_WAIT(wait);
1000
1001         prepare_to_wait_exclusive(&u->peer_wait, &wait, TASK_INTERRUPTIBLE);
1002
1003         sched = !sock_flag(other, SOCK_DEAD) &&
1004                 !(other->sk_shutdown & RCV_SHUTDOWN) &&
1005                 unix_recvq_full(other);
1006
1007         unix_state_unlock(other);
1008
1009         if (sched)
1010                 timeo = schedule_timeout(timeo);
1011
1012         finish_wait(&u->peer_wait, &wait);
1013         return timeo;
1014 }
1015
1016 static int unix_stream_connect(struct socket *sock, struct sockaddr *uaddr,
1017                                int addr_len, int flags)
1018 {
1019         struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
1020         struct sock *sk = sock->sk;
1021         struct net *net = sock_net(sk);
1022         struct unix_sock *u = unix_sk(sk), *newu, *otheru;
1023         struct sock *newsk = NULL;
1024         struct sock *other = NULL;
1025         struct sk_buff *skb = NULL;
1026         unsigned hash;
1027         int st;
1028         int err;
1029         long timeo;
1030
1031         err = unix_mkname(sunaddr, addr_len, &hash);
1032         if (err < 0)
1033                 goto out;
1034         addr_len = err;
1035
1036         if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr &&
1037             (err = unix_autobind(sock)) != 0)
1038                 goto out;
1039
1040         timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
1041
1042         /* First of all allocate resources.
1043            If we will make it after state is locked,
1044            we will have to recheck all again in any case.
1045          */
1046
1047         err = -ENOMEM;
1048
1049         /* create new sock for complete connection */
1050         newsk = unix_create1(sock_net(sk), NULL);
1051         if (newsk == NULL)
1052                 goto out;
1053
1054         /* Allocate skb for sending to listening sock */
1055         skb = sock_wmalloc(newsk, 1, 0, GFP_KERNEL);
1056         if (skb == NULL)
1057                 goto out;
1058
1059 restart:
1060         /*  Find listening sock. */
1061         other = unix_find_other(net, sunaddr, addr_len, sk->sk_type, hash, &err);
1062         if (!other)
1063                 goto out;
1064
1065         /* Latch state of peer */
1066         unix_state_lock(other);
1067
1068         /* Apparently VFS overslept socket death. Retry. */
1069         if (sock_flag(other, SOCK_DEAD)) {
1070                 unix_state_unlock(other);
1071                 sock_put(other);
1072                 goto restart;
1073         }
1074
1075         err = -ECONNREFUSED;
1076         if (other->sk_state != TCP_LISTEN)
1077                 goto out_unlock;
1078         if (other->sk_shutdown & RCV_SHUTDOWN)
1079                 goto out_unlock;
1080
1081         if (unix_recvq_full(other)) {
1082                 err = -EAGAIN;
1083                 if (!timeo)
1084                         goto out_unlock;
1085
1086                 timeo = unix_wait_for_peer(other, timeo);
1087
1088                 err = sock_intr_errno(timeo);
1089                 if (signal_pending(current))
1090                         goto out;
1091                 sock_put(other);
1092                 goto restart;
1093         }
1094
1095         /* Latch our state.
1096
1097            It is tricky place. We need to grab write lock and cannot
1098            drop lock on peer. It is dangerous because deadlock is
1099            possible. Connect to self case and simultaneous
1100            attempt to connect are eliminated by checking socket
1101            state. other is TCP_LISTEN, if sk is TCP_LISTEN we
1102            check this before attempt to grab lock.
1103
1104            Well, and we have to recheck the state after socket locked.
1105          */
1106         st = sk->sk_state;
1107
1108         switch (st) {
1109         case TCP_CLOSE:
1110                 /* This is ok... continue with connect */
1111                 break;
1112         case TCP_ESTABLISHED:
1113                 /* Socket is already connected */
1114                 err = -EISCONN;
1115                 goto out_unlock;
1116         default:
1117                 err = -EINVAL;
1118                 goto out_unlock;
1119         }
1120
1121         unix_state_lock_nested(sk);
1122
1123         if (sk->sk_state != st) {
1124                 unix_state_unlock(sk);
1125                 unix_state_unlock(other);
1126                 sock_put(other);
1127                 goto restart;
1128         }
1129
1130         err = security_unix_stream_connect(sock, other->sk_socket, newsk);
1131         if (err) {
1132                 unix_state_unlock(sk);
1133                 goto out_unlock;
1134         }
1135
1136         /* The way is open! Fastly set all the necessary fields... */
1137
1138         sock_hold(sk);
1139         unix_peer(newsk)        = sk;
1140         newsk->sk_state         = TCP_ESTABLISHED;
1141         newsk->sk_type          = sk->sk_type;
1142         newsk->sk_peercred.pid  = task_tgid_vnr(current);
1143         current_euid_egid(&newsk->sk_peercred.uid, &newsk->sk_peercred.gid);
1144         newu = unix_sk(newsk);
1145         newsk->sk_sleep         = &newu->peer_wait;
1146         otheru = unix_sk(other);
1147
1148         /* copy address information from listening to new sock*/
1149         if (otheru->addr) {
1150                 atomic_inc(&otheru->addr->refcnt);
1151                 newu->addr = otheru->addr;
1152         }
1153         if (otheru->dentry) {
1154                 newu->dentry    = dget(otheru->dentry);
1155                 newu->mnt       = mntget(otheru->mnt);
1156         }
1157
1158         /* Set credentials */
1159         sk->sk_peercred = other->sk_peercred;
1160
1161         sock->state     = SS_CONNECTED;
1162         sk->sk_state    = TCP_ESTABLISHED;
1163         sock_hold(newsk);
1164
1165         smp_mb__after_atomic_inc();     /* sock_hold() does an atomic_inc() */
1166         unix_peer(sk)   = newsk;
1167
1168         unix_state_unlock(sk);
1169
1170         /* take ten and and send info to listening sock */
1171         spin_lock(&other->sk_receive_queue.lock);
1172         __skb_queue_tail(&other->sk_receive_queue, skb);
1173         spin_unlock(&other->sk_receive_queue.lock);
1174         unix_state_unlock(other);
1175         other->sk_data_ready(other, 0);
1176         sock_put(other);
1177         return 0;
1178
1179 out_unlock:
1180         if (other)
1181                 unix_state_unlock(other);
1182
1183 out:
1184         kfree_skb(skb);
1185         if (newsk)
1186                 unix_release_sock(newsk, 0);
1187         if (other)
1188                 sock_put(other);
1189         return err;
1190 }
1191
1192 static int unix_socketpair(struct socket *socka, struct socket *sockb)
1193 {
1194         struct sock *ska = socka->sk, *skb = sockb->sk;
1195
1196         /* Join our sockets back to back */
1197         sock_hold(ska);
1198         sock_hold(skb);
1199         unix_peer(ska) = skb;
1200         unix_peer(skb) = ska;
1201         ska->sk_peercred.pid = skb->sk_peercred.pid = task_tgid_vnr(current);
1202         current_euid_egid(&skb->sk_peercred.uid, &skb->sk_peercred.gid);
1203         ska->sk_peercred.uid = skb->sk_peercred.uid;
1204         ska->sk_peercred.gid = skb->sk_peercred.gid;
1205
1206         if (ska->sk_type != SOCK_DGRAM) {
1207                 ska->sk_state = TCP_ESTABLISHED;
1208                 skb->sk_state = TCP_ESTABLISHED;
1209                 socka->state  = SS_CONNECTED;
1210                 sockb->state  = SS_CONNECTED;
1211         }
1212         return 0;
1213 }
1214
1215 static int unix_accept(struct socket *sock, struct socket *newsock, int flags)
1216 {
1217         struct sock *sk = sock->sk;
1218         struct sock *tsk;
1219         struct sk_buff *skb;
1220         int err;
1221
1222         err = -EOPNOTSUPP;
1223         if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
1224                 goto out;
1225
1226         err = -EINVAL;
1227         if (sk->sk_state != TCP_LISTEN)
1228                 goto out;
1229
1230         /* If socket state is TCP_LISTEN it cannot change (for now...),
1231          * so that no locks are necessary.
1232          */
1233
1234         skb = skb_recv_datagram(sk, 0, flags&O_NONBLOCK, &err);
1235         if (!skb) {
1236                 /* This means receive shutdown. */
1237                 if (err == 0)
1238                         err = -EINVAL;
1239                 goto out;
1240         }
1241
1242         tsk = skb->sk;
1243         skb_free_datagram(sk, skb);
1244         wake_up_interruptible(&unix_sk(sk)->peer_wait);
1245
1246         /* attach accepted sock to socket */
1247         unix_state_lock(tsk);
1248         newsock->state = SS_CONNECTED;
1249         sock_graft(tsk, newsock);
1250         unix_state_unlock(tsk);
1251         return 0;
1252
1253 out:
1254         return err;
1255 }
1256
1257
1258 static int unix_getname(struct socket *sock, struct sockaddr *uaddr, int *uaddr_len, int peer)
1259 {
1260         struct sock *sk = sock->sk;
1261         struct unix_sock *u;
1262         DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr, uaddr);
1263         int err = 0;
1264
1265         if (peer) {
1266                 sk = unix_peer_get(sk);
1267
1268                 err = -ENOTCONN;
1269                 if (!sk)
1270                         goto out;
1271                 err = 0;
1272         } else {
1273                 sock_hold(sk);
1274         }
1275
1276         u = unix_sk(sk);
1277         unix_state_lock(sk);
1278         if (!u->addr) {
1279                 sunaddr->sun_family = AF_UNIX;
1280                 sunaddr->sun_path[0] = 0;
1281                 *uaddr_len = sizeof(short);
1282         } else {
1283                 struct unix_address *addr = u->addr;
1284
1285                 *uaddr_len = addr->len;
1286                 memcpy(sunaddr, addr->name, *uaddr_len);
1287         }
1288         unix_state_unlock(sk);
1289         sock_put(sk);
1290 out:
1291         return err;
1292 }
1293
1294 static void unix_detach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1295 {
1296         int i;
1297
1298         scm->fp = UNIXCB(skb).fp;
1299         skb->destructor = sock_wfree;
1300         UNIXCB(skb).fp = NULL;
1301
1302         for (i = scm->fp->count-1; i >= 0; i--)
1303                 unix_notinflight(scm->fp->fp[i]);
1304 }
1305
1306 static void unix_destruct_fds(struct sk_buff *skb)
1307 {
1308         struct scm_cookie scm;
1309         memset(&scm, 0, sizeof(scm));
1310         unix_detach_fds(&scm, skb);
1311
1312         /* Alas, it calls VFS */
1313         /* So fscking what? fput() had been SMP-safe since the last Summer */
1314         scm_destroy(&scm);
1315         sock_wfree(skb);
1316 }
1317
1318 static int unix_attach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1319 {
1320         int i;
1321
1322         /*
1323          * Need to duplicate file references for the sake of garbage
1324          * collection.  Otherwise a socket in the fps might become a
1325          * candidate for GC while the skb is not yet queued.
1326          */
1327         UNIXCB(skb).fp = scm_fp_dup(scm->fp);
1328         if (!UNIXCB(skb).fp)
1329                 return -ENOMEM;
1330
1331         for (i = scm->fp->count-1; i >= 0; i--)
1332                 unix_inflight(scm->fp->fp[i]);
1333         skb->destructor = unix_destruct_fds;
1334         return 0;
1335 }
1336
1337 /*
1338  *      Send AF_UNIX data.
1339  */
1340
1341 static int unix_dgram_sendmsg(struct kiocb *kiocb, struct socket *sock,
1342                               struct msghdr *msg, size_t len)
1343 {
1344         struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1345         struct sock *sk = sock->sk;
1346         struct net *net = sock_net(sk);
1347         struct unix_sock *u = unix_sk(sk);
1348         struct sockaddr_un *sunaddr = msg->msg_name;
1349         struct sock *other = NULL;
1350         int namelen = 0; /* fake GCC */
1351         int err;
1352         unsigned hash;
1353         struct sk_buff *skb;
1354         long timeo;
1355         struct scm_cookie tmp_scm;
1356
1357         if (NULL == siocb->scm)
1358                 siocb->scm = &tmp_scm;
1359         wait_for_unix_gc();
1360         err = scm_send(sock, msg, siocb->scm);
1361         if (err < 0)
1362                 return err;
1363
1364         err = -EOPNOTSUPP;
1365         if (msg->msg_flags&MSG_OOB)
1366                 goto out;
1367
1368         if (msg->msg_namelen) {
1369                 err = unix_mkname(sunaddr, msg->msg_namelen, &hash);
1370                 if (err < 0)
1371                         goto out;
1372                 namelen = err;
1373         } else {
1374                 sunaddr = NULL;
1375                 err = -ENOTCONN;
1376                 other = unix_peer_get(sk);
1377                 if (!other)
1378                         goto out;
1379         }
1380
1381         if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr
1382             && (err = unix_autobind(sock)) != 0)
1383                 goto out;
1384
1385         err = -EMSGSIZE;
1386         if (len > sk->sk_sndbuf - 32)
1387                 goto out;
1388
1389         skb = sock_alloc_send_skb(sk, len, msg->msg_flags&MSG_DONTWAIT, &err);
1390         if (skb == NULL)
1391                 goto out;
1392
1393         memcpy(UNIXCREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1394         if (siocb->scm->fp) {
1395                 err = unix_attach_fds(siocb->scm, skb);
1396                 if (err)
1397                         goto out_free;
1398         }
1399         unix_get_secdata(siocb->scm, skb);
1400
1401         skb_reset_transport_header(skb);
1402         err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
1403         if (err)
1404                 goto out_free;
1405
1406         timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1407
1408 restart:
1409         if (!other) {
1410                 err = -ECONNRESET;
1411                 if (sunaddr == NULL)
1412                         goto out_free;
1413
1414                 other = unix_find_other(net, sunaddr, namelen, sk->sk_type,
1415                                         hash, &err);
1416                 if (other == NULL)
1417                         goto out_free;
1418         }
1419
1420         unix_state_lock(other);
1421         err = -EPERM;
1422         if (!unix_may_send(sk, other))
1423                 goto out_unlock;
1424
1425         if (sock_flag(other, SOCK_DEAD)) {
1426                 /*
1427                  *      Check with 1003.1g - what should
1428                  *      datagram error
1429                  */
1430                 unix_state_unlock(other);
1431                 sock_put(other);
1432
1433                 err = 0;
1434                 unix_state_lock(sk);
1435                 if (unix_peer(sk) == other) {
1436                         unix_peer(sk) = NULL;
1437                         unix_state_unlock(sk);
1438
1439                         unix_dgram_disconnected(sk, other);
1440                         sock_put(other);
1441                         err = -ECONNREFUSED;
1442                 } else {
1443                         unix_state_unlock(sk);
1444                 }
1445
1446                 other = NULL;
1447                 if (err)
1448                         goto out_free;
1449                 goto restart;
1450         }
1451
1452         err = -EPIPE;
1453         if (other->sk_shutdown & RCV_SHUTDOWN)
1454                 goto out_unlock;
1455
1456         if (sk->sk_type != SOCK_SEQPACKET) {
1457                 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1458                 if (err)
1459                         goto out_unlock;
1460         }
1461
1462         if (unix_peer(other) != sk && unix_recvq_full(other)) {
1463                 if (!timeo) {
1464                         err = -EAGAIN;
1465                         goto out_unlock;
1466                 }
1467
1468                 timeo = unix_wait_for_peer(other, timeo);
1469
1470                 err = sock_intr_errno(timeo);
1471                 if (signal_pending(current))
1472                         goto out_free;
1473
1474                 goto restart;
1475         }
1476
1477         skb_queue_tail(&other->sk_receive_queue, skb);
1478         unix_state_unlock(other);
1479         other->sk_data_ready(other, len);
1480         sock_put(other);
1481         scm_destroy(siocb->scm);
1482         return len;
1483
1484 out_unlock:
1485         unix_state_unlock(other);
1486 out_free:
1487         kfree_skb(skb);
1488 out:
1489         if (other)
1490                 sock_put(other);
1491         scm_destroy(siocb->scm);
1492         return err;
1493 }
1494
1495
1496 static int unix_stream_sendmsg(struct kiocb *kiocb, struct socket *sock,
1497                                struct msghdr *msg, size_t len)
1498 {
1499         struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1500         struct sock *sk = sock->sk;
1501         struct sock *other = NULL;
1502         struct sockaddr_un *sunaddr = msg->msg_name;
1503         int err, size;
1504         struct sk_buff *skb;
1505         int sent = 0;
1506         struct scm_cookie tmp_scm;
1507         bool fds_sent = false;
1508
1509         if (NULL == siocb->scm)
1510                 siocb->scm = &tmp_scm;
1511         wait_for_unix_gc();
1512         err = scm_send(sock, msg, siocb->scm);
1513         if (err < 0)
1514                 return err;
1515
1516         err = -EOPNOTSUPP;
1517         if (msg->msg_flags&MSG_OOB)
1518                 goto out_err;
1519
1520         if (msg->msg_namelen) {
1521                 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
1522                 goto out_err;
1523         } else {
1524                 sunaddr = NULL;
1525                 err = -ENOTCONN;
1526                 other = unix_peer(sk);
1527                 if (!other)
1528                         goto out_err;
1529         }
1530
1531         if (sk->sk_shutdown & SEND_SHUTDOWN)
1532                 goto pipe_err;
1533
1534         while (sent < len) {
1535                 /*
1536                  *      Optimisation for the fact that under 0.01% of X
1537                  *      messages typically need breaking up.
1538                  */
1539
1540                 size = len-sent;
1541
1542                 /* Keep two messages in the pipe so it schedules better */
1543                 if (size > ((sk->sk_sndbuf >> 1) - 64))
1544                         size = (sk->sk_sndbuf >> 1) - 64;
1545
1546                 if (size > SKB_MAX_ALLOC)
1547                         size = SKB_MAX_ALLOC;
1548
1549                 /*
1550                  *      Grab a buffer
1551                  */
1552
1553                 skb = sock_alloc_send_skb(sk, size, msg->msg_flags&MSG_DONTWAIT,
1554                                           &err);
1555
1556                 if (skb == NULL)
1557                         goto out_err;
1558
1559                 /*
1560                  *      If you pass two values to the sock_alloc_send_skb
1561                  *      it tries to grab the large buffer with GFP_NOFS
1562                  *      (which can fail easily), and if it fails grab the
1563                  *      fallback size buffer which is under a page and will
1564                  *      succeed. [Alan]
1565                  */
1566                 size = min_t(int, size, skb_tailroom(skb));
1567
1568                 memcpy(UNIXCREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1569                 /* Only send the fds in the first buffer */
1570                 if (siocb->scm->fp && !fds_sent) {
1571                         err = unix_attach_fds(siocb->scm, skb);
1572                         if (err) {
1573                                 kfree_skb(skb);
1574                                 goto out_err;
1575                         }
1576                         fds_sent = true;
1577                 }
1578
1579                 err = memcpy_fromiovec(skb_put(skb, size), msg->msg_iov, size);
1580                 if (err) {
1581                         kfree_skb(skb);
1582                         goto out_err;
1583                 }
1584
1585                 unix_state_lock(other);
1586
1587                 if (sock_flag(other, SOCK_DEAD) ||
1588                     (other->sk_shutdown & RCV_SHUTDOWN))
1589                         goto pipe_err_free;
1590
1591                 skb_queue_tail(&other->sk_receive_queue, skb);
1592                 unix_state_unlock(other);
1593                 other->sk_data_ready(other, size);
1594                 sent += size;
1595         }
1596
1597         scm_destroy(siocb->scm);
1598         siocb->scm = NULL;
1599
1600         return sent;
1601
1602 pipe_err_free:
1603         unix_state_unlock(other);
1604         kfree_skb(skb);
1605 pipe_err:
1606         if (sent == 0 && !(msg->msg_flags&MSG_NOSIGNAL))
1607                 send_sig(SIGPIPE, current, 0);
1608         err = -EPIPE;
1609 out_err:
1610         scm_destroy(siocb->scm);
1611         siocb->scm = NULL;
1612         return sent ? : err;
1613 }
1614
1615 static int unix_seqpacket_sendmsg(struct kiocb *kiocb, struct socket *sock,
1616                                   struct msghdr *msg, size_t len)
1617 {
1618         int err;
1619         struct sock *sk = sock->sk;
1620
1621         err = sock_error(sk);
1622         if (err)
1623                 return err;
1624
1625         if (sk->sk_state != TCP_ESTABLISHED)
1626                 return -ENOTCONN;
1627
1628         if (msg->msg_namelen)
1629                 msg->msg_namelen = 0;
1630
1631         return unix_dgram_sendmsg(kiocb, sock, msg, len);
1632 }
1633
1634 static void unix_copy_addr(struct msghdr *msg, struct sock *sk)
1635 {
1636         struct unix_sock *u = unix_sk(sk);
1637
1638         msg->msg_namelen = 0;
1639         if (u->addr) {
1640                 msg->msg_namelen = u->addr->len;
1641                 memcpy(msg->msg_name, u->addr->name, u->addr->len);
1642         }
1643 }
1644
1645 static int unix_dgram_recvmsg(struct kiocb *iocb, struct socket *sock,
1646                               struct msghdr *msg, size_t size,
1647                               int flags)
1648 {
1649         struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1650         struct scm_cookie tmp_scm;
1651         struct sock *sk = sock->sk;
1652         struct unix_sock *u = unix_sk(sk);
1653         int noblock = flags & MSG_DONTWAIT;
1654         struct sk_buff *skb;
1655         int err;
1656
1657         err = -EOPNOTSUPP;
1658         if (flags&MSG_OOB)
1659                 goto out;
1660
1661         msg->msg_namelen = 0;
1662
1663         mutex_lock(&u->readlock);
1664
1665         skb = skb_recv_datagram(sk, flags, noblock, &err);
1666         if (!skb) {
1667                 unix_state_lock(sk);
1668                 /* Signal EOF on disconnected non-blocking SEQPACKET socket. */
1669                 if (sk->sk_type == SOCK_SEQPACKET && err == -EAGAIN &&
1670                     (sk->sk_shutdown & RCV_SHUTDOWN))
1671                         err = 0;
1672                 unix_state_unlock(sk);
1673                 goto out_unlock;
1674         }
1675
1676         wake_up_interruptible_sync(&u->peer_wait);
1677
1678         if (msg->msg_name)
1679                 unix_copy_addr(msg, skb->sk);
1680
1681         if (size > skb->len)
1682                 size = skb->len;
1683         else if (size < skb->len)
1684                 msg->msg_flags |= MSG_TRUNC;
1685
1686         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, size);
1687         if (err)
1688                 goto out_free;
1689
1690         if (!siocb->scm) {
1691                 siocb->scm = &tmp_scm;
1692                 memset(&tmp_scm, 0, sizeof(tmp_scm));
1693         }
1694         siocb->scm->creds = *UNIXCREDS(skb);
1695         unix_set_secdata(siocb->scm, skb);
1696
1697         if (!(flags & MSG_PEEK)) {
1698                 if (UNIXCB(skb).fp)
1699                         unix_detach_fds(siocb->scm, skb);
1700         } else {
1701                 /* It is questionable: on PEEK we could:
1702                    - do not return fds - good, but too simple 8)
1703                    - return fds, and do not return them on read (old strategy,
1704                      apparently wrong)
1705                    - clone fds (I chose it for now, it is the most universal
1706                      solution)
1707
1708                    POSIX 1003.1g does not actually define this clearly
1709                    at all. POSIX 1003.1g doesn't define a lot of things
1710                    clearly however!
1711
1712                 */
1713                 if (UNIXCB(skb).fp)
1714                         siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1715         }
1716         err = size;
1717
1718         scm_recv(sock, msg, siocb->scm, flags);
1719
1720 out_free:
1721         skb_free_datagram(sk, skb);
1722 out_unlock:
1723         mutex_unlock(&u->readlock);
1724 out:
1725         return err;
1726 }
1727
1728 /*
1729  *      Sleep until data has arrive. But check for races..
1730  */
1731
1732 static long unix_stream_data_wait(struct sock *sk, long timeo)
1733 {
1734         DEFINE_WAIT(wait);
1735
1736         unix_state_lock(sk);
1737
1738         for (;;) {
1739                 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1740
1741                 if (!skb_queue_empty(&sk->sk_receive_queue) ||
1742                     sk->sk_err ||
1743                     (sk->sk_shutdown & RCV_SHUTDOWN) ||
1744                     signal_pending(current) ||
1745                     !timeo)
1746                         break;
1747
1748                 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1749                 unix_state_unlock(sk);
1750                 timeo = schedule_timeout(timeo);
1751                 unix_state_lock(sk);
1752                 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1753         }
1754
1755         finish_wait(sk->sk_sleep, &wait);
1756         unix_state_unlock(sk);
1757         return timeo;
1758 }
1759
1760
1761
1762 static int unix_stream_recvmsg(struct kiocb *iocb, struct socket *sock,
1763                                struct msghdr *msg, size_t size,
1764                                int flags)
1765 {
1766         struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1767         struct scm_cookie tmp_scm;
1768         struct sock *sk = sock->sk;
1769         struct unix_sock *u = unix_sk(sk);
1770         struct sockaddr_un *sunaddr = msg->msg_name;
1771         int copied = 0;
1772         int check_creds = 0;
1773         int target;
1774         int err = 0;
1775         long timeo;
1776
1777         err = -EINVAL;
1778         if (sk->sk_state != TCP_ESTABLISHED)
1779                 goto out;
1780
1781         err = -EOPNOTSUPP;
1782         if (flags&MSG_OOB)
1783                 goto out;
1784
1785         target = sock_rcvlowat(sk, flags&MSG_WAITALL, size);
1786         timeo = sock_rcvtimeo(sk, flags&MSG_DONTWAIT);
1787
1788         msg->msg_namelen = 0;
1789
1790         /* Lock the socket to prevent queue disordering
1791          * while sleeps in memcpy_tomsg
1792          */
1793
1794         if (!siocb->scm) {
1795                 siocb->scm = &tmp_scm;
1796                 memset(&tmp_scm, 0, sizeof(tmp_scm));
1797         }
1798
1799         mutex_lock(&u->readlock);
1800
1801         do {
1802                 int chunk;
1803                 struct sk_buff *skb;
1804
1805                 unix_state_lock(sk);
1806                 skb = skb_dequeue(&sk->sk_receive_queue);
1807                 if (skb == NULL) {
1808                         if (copied >= target)
1809                                 goto unlock;
1810
1811                         /*
1812                          *      POSIX 1003.1g mandates this order.
1813                          */
1814
1815                         err = sock_error(sk);
1816                         if (err)
1817                                 goto unlock;
1818                         if (sk->sk_shutdown & RCV_SHUTDOWN)
1819                                 goto unlock;
1820
1821                         unix_state_unlock(sk);
1822                         err = -EAGAIN;
1823                         if (!timeo)
1824                                 break;
1825                         mutex_unlock(&u->readlock);
1826
1827                         timeo = unix_stream_data_wait(sk, timeo);
1828
1829                         if (signal_pending(current)) {
1830                                 err = sock_intr_errno(timeo);
1831                                 goto out;
1832                         }
1833                         mutex_lock(&u->readlock);
1834                         continue;
1835  unlock:
1836                         unix_state_unlock(sk);
1837                         break;
1838                 }
1839                 unix_state_unlock(sk);
1840
1841                 if (check_creds) {
1842                         /* Never glue messages from different writers */
1843                         if (memcmp(UNIXCREDS(skb), &siocb->scm->creds,
1844                                    sizeof(siocb->scm->creds)) != 0) {
1845                                 skb_queue_head(&sk->sk_receive_queue, skb);
1846                                 break;
1847                         }
1848                 } else {
1849                         /* Copy credentials */
1850                         siocb->scm->creds = *UNIXCREDS(skb);
1851                         check_creds = 1;
1852                 }
1853
1854                 /* Copy address just once */
1855                 if (sunaddr) {
1856                         unix_copy_addr(msg, skb->sk);
1857                         sunaddr = NULL;
1858                 }
1859
1860                 chunk = min_t(unsigned int, skb->len, size);
1861                 if (memcpy_toiovec(msg->msg_iov, skb->data, chunk)) {
1862                         skb_queue_head(&sk->sk_receive_queue, skb);
1863                         if (copied == 0)
1864                                 copied = -EFAULT;
1865                         break;
1866                 }
1867                 copied += chunk;
1868                 size -= chunk;
1869
1870                 /* Mark read part of skb as used */
1871                 if (!(flags & MSG_PEEK)) {
1872                         skb_pull(skb, chunk);
1873
1874                         if (UNIXCB(skb).fp)
1875                                 unix_detach_fds(siocb->scm, skb);
1876
1877                         /* put the skb back if we didn't use it up.. */
1878                         if (skb->len) {
1879                                 skb_queue_head(&sk->sk_receive_queue, skb);
1880                                 break;
1881                         }
1882
1883                         kfree_skb(skb);
1884
1885                         if (siocb->scm->fp)
1886                                 break;
1887                 } else {
1888                         /* It is questionable, see note in unix_dgram_recvmsg.
1889                          */
1890                         if (UNIXCB(skb).fp)
1891                                 siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1892
1893                         /* put message back and return */
1894                         skb_queue_head(&sk->sk_receive_queue, skb);
1895                         break;
1896                 }
1897         } while (size);
1898
1899         mutex_unlock(&u->readlock);
1900         scm_recv(sock, msg, siocb->scm, flags);
1901 out:
1902         return copied ? : err;
1903 }
1904
1905 static int unix_shutdown(struct socket *sock, int mode)
1906 {
1907         struct sock *sk = sock->sk;
1908         struct sock *other;
1909
1910         mode = (mode+1)&(RCV_SHUTDOWN|SEND_SHUTDOWN);
1911
1912         if (mode) {
1913                 unix_state_lock(sk);
1914                 sk->sk_shutdown |= mode;
1915                 other = unix_peer(sk);
1916                 if (other)
1917                         sock_hold(other);
1918                 unix_state_unlock(sk);
1919                 sk->sk_state_change(sk);
1920
1921                 if (other &&
1922                         (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET)) {
1923
1924                         int peer_mode = 0;
1925
1926                         if (mode&RCV_SHUTDOWN)
1927                                 peer_mode |= SEND_SHUTDOWN;
1928                         if (mode&SEND_SHUTDOWN)
1929                                 peer_mode |= RCV_SHUTDOWN;
1930                         unix_state_lock(other);
1931                         other->sk_shutdown |= peer_mode;
1932                         unix_state_unlock(other);
1933                         other->sk_state_change(other);
1934                         read_lock(&other->sk_callback_lock);
1935                         if (peer_mode == SHUTDOWN_MASK)
1936                                 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_HUP);
1937                         else if (peer_mode & RCV_SHUTDOWN)
1938                                 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_IN);
1939                         read_unlock(&other->sk_callback_lock);
1940                 }
1941                 if (other)
1942                         sock_put(other);
1943         }
1944         return 0;
1945 }
1946
1947 static int unix_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1948 {
1949         struct sock *sk = sock->sk;
1950         long amount = 0;
1951         int err;
1952
1953         switch (cmd) {
1954         case SIOCOUTQ:
1955                 amount = sk_wmem_alloc_get(sk);
1956                 err = put_user(amount, (int __user *)arg);
1957                 break;
1958         case SIOCINQ:
1959                 {
1960                         struct sk_buff *skb;
1961
1962                         if (sk->sk_state == TCP_LISTEN) {
1963                                 err = -EINVAL;
1964                                 break;
1965                         }
1966
1967                         spin_lock(&sk->sk_receive_queue.lock);
1968                         if (sk->sk_type == SOCK_STREAM ||
1969                             sk->sk_type == SOCK_SEQPACKET) {
1970                                 skb_queue_walk(&sk->sk_receive_queue, skb)
1971                                         amount += skb->len;
1972                         } else {
1973                                 skb = skb_peek(&sk->sk_receive_queue);
1974                                 if (skb)
1975                                         amount = skb->len;
1976                         }
1977                         spin_unlock(&sk->sk_receive_queue.lock);
1978                         err = put_user(amount, (int __user *)arg);
1979                         break;
1980                 }
1981
1982         default:
1983                 err = -ENOIOCTLCMD;
1984                 break;
1985         }
1986         return err;
1987 }
1988
1989 static unsigned int unix_poll(struct file *file, struct socket *sock, poll_table *wait)
1990 {
1991         struct sock *sk = sock->sk;
1992         unsigned int mask;
1993
1994         sock_poll_wait(file, sk->sk_sleep, wait);
1995         mask = 0;
1996
1997         /* exceptional events? */
1998         if (sk->sk_err)
1999                 mask |= POLLERR;
2000         if (sk->sk_shutdown == SHUTDOWN_MASK)
2001                 mask |= POLLHUP;
2002         if (sk->sk_shutdown & RCV_SHUTDOWN)
2003                 mask |= POLLRDHUP;
2004
2005         /* readable? */
2006         if (!skb_queue_empty(&sk->sk_receive_queue) ||
2007             (sk->sk_shutdown & RCV_SHUTDOWN))
2008                 mask |= POLLIN | POLLRDNORM;
2009
2010         /* Connection-based need to check for termination and startup */
2011         if ((sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) &&
2012             sk->sk_state == TCP_CLOSE)
2013                 mask |= POLLHUP;
2014
2015         /*
2016          * we set writable also when the other side has shut down the
2017          * connection. This prevents stuck sockets.
2018          */
2019         if (unix_writable(sk))
2020                 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2021
2022         return mask;
2023 }
2024
2025 static unsigned int unix_dgram_poll(struct file *file, struct socket *sock,
2026                                     poll_table *wait)
2027 {
2028         struct sock *sk = sock->sk, *other;
2029         unsigned int mask, writable;
2030
2031         sock_poll_wait(file, sk->sk_sleep, wait);
2032         mask = 0;
2033
2034         /* exceptional events? */
2035         if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue))
2036                 mask |= POLLERR;
2037         if (sk->sk_shutdown & RCV_SHUTDOWN)
2038                 mask |= POLLRDHUP;
2039         if (sk->sk_shutdown == SHUTDOWN_MASK)
2040                 mask |= POLLHUP;
2041
2042         /* readable? */
2043         if (!skb_queue_empty(&sk->sk_receive_queue) ||
2044             (sk->sk_shutdown & RCV_SHUTDOWN))
2045                 mask |= POLLIN | POLLRDNORM;
2046
2047         /* Connection-based need to check for termination and startup */
2048         if (sk->sk_type == SOCK_SEQPACKET) {
2049                 if (sk->sk_state == TCP_CLOSE)
2050                         mask |= POLLHUP;
2051                 /* connection hasn't started yet? */
2052                 if (sk->sk_state == TCP_SYN_SENT)
2053                         return mask;
2054         }
2055
2056         /* writable? */
2057         writable = unix_writable(sk);
2058         if (writable) {
2059                 other = unix_peer_get(sk);
2060                 if (other) {
2061                         if (unix_peer(other) != sk) {
2062                                 sock_poll_wait(file, &unix_sk(other)->peer_wait,
2063                                           wait);
2064                                 if (unix_recvq_full(other))
2065                                         writable = 0;
2066                         }
2067
2068                         sock_put(other);
2069                 }
2070         }
2071
2072         if (writable)
2073                 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2074         else
2075                 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
2076
2077         return mask;
2078 }
2079
2080 #ifdef CONFIG_PROC_FS
2081 static struct sock *first_unix_socket(int *i)
2082 {
2083         for (*i = 0; *i <= UNIX_HASH_SIZE; (*i)++) {
2084                 if (!hlist_empty(&unix_socket_table[*i]))
2085                         return __sk_head(&unix_socket_table[*i]);
2086         }
2087         return NULL;
2088 }
2089
2090 static struct sock *next_unix_socket(int *i, struct sock *s)
2091 {
2092         struct sock *next = sk_next(s);
2093         /* More in this chain? */
2094         if (next)
2095                 return next;
2096         /* Look for next non-empty chain. */
2097         for ((*i)++; *i <= UNIX_HASH_SIZE; (*i)++) {
2098                 if (!hlist_empty(&unix_socket_table[*i]))
2099                         return __sk_head(&unix_socket_table[*i]);
2100         }
2101         return NULL;
2102 }
2103
2104 struct unix_iter_state {
2105         struct seq_net_private p;
2106         int i;
2107 };
2108
2109 static struct sock *unix_seq_idx(struct seq_file *seq, loff_t pos)
2110 {
2111         struct unix_iter_state *iter = seq->private;
2112         loff_t off = 0;
2113         struct sock *s;
2114
2115         for (s = first_unix_socket(&iter->i); s; s = next_unix_socket(&iter->i, s)) {
2116                 if (sock_net(s) != seq_file_net(seq))
2117                         continue;
2118                 if (off == pos)
2119                         return s;
2120                 ++off;
2121         }
2122         return NULL;
2123 }
2124
2125 static void *unix_seq_start(struct seq_file *seq, loff_t *pos)
2126         __acquires(unix_table_lock)
2127 {
2128         spin_lock(&unix_table_lock);
2129         return *pos ? unix_seq_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2130 }
2131
2132 static void *unix_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2133 {
2134         struct unix_iter_state *iter = seq->private;
2135         struct sock *sk = v;
2136         ++*pos;
2137
2138         if (v == SEQ_START_TOKEN)
2139                 sk = first_unix_socket(&iter->i);
2140         else
2141                 sk = next_unix_socket(&iter->i, sk);
2142         while (sk && (sock_net(sk) != seq_file_net(seq)))
2143                 sk = next_unix_socket(&iter->i, sk);
2144         return sk;
2145 }
2146
2147 static void unix_seq_stop(struct seq_file *seq, void *v)
2148         __releases(unix_table_lock)
2149 {
2150         spin_unlock(&unix_table_lock);
2151 }
2152
2153 static int unix_seq_show(struct seq_file *seq, void *v)
2154 {
2155
2156         if (v == SEQ_START_TOKEN)
2157                 seq_puts(seq, "Num       RefCount Protocol Flags    Type St "
2158                          "Inode Path\n");
2159         else {
2160                 struct sock *s = v;
2161                 struct unix_sock *u = unix_sk(s);
2162                 unix_state_lock(s);
2163
2164                 seq_printf(seq, "%p: %08X %08X %08X %04X %02X %5lu",
2165                         s,
2166                         atomic_read(&s->sk_refcnt),
2167                         0,
2168                         s->sk_state == TCP_LISTEN ? __SO_ACCEPTCON : 0,
2169                         s->sk_type,
2170                         s->sk_socket ?
2171                         (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTED : SS_UNCONNECTED) :
2172                         (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTING : SS_DISCONNECTING),
2173                         sock_i_ino(s));
2174
2175                 if (u->addr) {
2176                         int i, len;
2177                         seq_putc(seq, ' ');
2178
2179                         i = 0;
2180                         len = u->addr->len - sizeof(short);
2181                         if (!UNIX_ABSTRACT(s))
2182                                 len--;
2183                         else {
2184                                 seq_putc(seq, '@');
2185                                 i++;
2186                         }
2187                         for ( ; i < len; i++)
2188                                 seq_putc(seq, u->addr->name->sun_path[i]);
2189                 }
2190                 unix_state_unlock(s);
2191                 seq_putc(seq, '\n');
2192         }
2193
2194         return 0;
2195 }
2196
2197 static const struct seq_operations unix_seq_ops = {
2198         .start  = unix_seq_start,
2199         .next   = unix_seq_next,
2200         .stop   = unix_seq_stop,
2201         .show   = unix_seq_show,
2202 };
2203
2204 static int unix_seq_open(struct inode *inode, struct file *file)
2205 {
2206         return seq_open_net(inode, file, &unix_seq_ops,
2207                             sizeof(struct unix_iter_state));
2208 }
2209
2210 static const struct file_operations unix_seq_fops = {
2211         .owner          = THIS_MODULE,
2212         .open           = unix_seq_open,
2213         .read           = seq_read,
2214         .llseek         = seq_lseek,
2215         .release        = seq_release_net,
2216 };
2217
2218 #endif
2219
2220 static const struct net_proto_family unix_family_ops = {
2221         .family = PF_UNIX,
2222         .create = unix_create,
2223         .owner  = THIS_MODULE,
2224 };
2225
2226
2227 static int unix_net_init(struct net *net)
2228 {
2229         int error = -ENOMEM;
2230
2231         net->unx.sysctl_max_dgram_qlen = 10;
2232         if (unix_sysctl_register(net))
2233                 goto out;
2234
2235 #ifdef CONFIG_PROC_FS
2236         if (!proc_net_fops_create(net, "unix", 0, &unix_seq_fops)) {
2237                 unix_sysctl_unregister(net);
2238                 goto out;
2239         }
2240 #endif
2241         error = 0;
2242 out:
2243         return error;
2244 }
2245
2246 static void unix_net_exit(struct net *net)
2247 {
2248         unix_sysctl_unregister(net);
2249         proc_net_remove(net, "unix");
2250 }
2251
2252 static struct pernet_operations unix_net_ops = {
2253         .init = unix_net_init,
2254         .exit = unix_net_exit,
2255 };
2256
2257 static int __init af_unix_init(void)
2258 {
2259         int rc = -1;
2260         struct sk_buff *dummy_skb;
2261
2262         BUILD_BUG_ON(sizeof(struct unix_skb_parms) > sizeof(dummy_skb->cb));
2263
2264         rc = proto_register(&unix_proto, 1);
2265         if (rc != 0) {
2266                 printk(KERN_CRIT "%s: Cannot create unix_sock SLAB cache!\n",
2267                        __func__);
2268                 goto out;
2269         }
2270
2271         sock_register(&unix_family_ops);
2272         register_pernet_subsys(&unix_net_ops);
2273 out:
2274         return rc;
2275 }
2276
2277 static void __exit af_unix_exit(void)
2278 {
2279         sock_unregister(PF_UNIX);
2280         proto_unregister(&unix_proto);
2281         unregister_pernet_subsys(&unix_net_ops);
2282 }
2283
2284 /* Earlier than device_initcall() so that other drivers invoking
2285    request_module() don't end up in a loop when modprobe tries
2286    to use a UNIX socket. But later than subsys_initcall() because
2287    we depend on stuff initialised there */
2288 fs_initcall(af_unix_init);
2289 module_exit(af_unix_exit);
2290
2291 MODULE_LICENSE("GPL");
2292 MODULE_ALIAS_NETPROTO(PF_UNIX);