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