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 spin lock.
 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         struct socket_wq *wq;
 317
 318         rcu_read_lock();
 319         if (unix_writable(sk)) {
 320                 wq = rcu_dereference(sk->sk_wq);
 321                 if (wq_has_sleeper(wq))
 322                         wake_up_interruptible_sync(&wq->wait);
 323                 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
 324         }
 325         rcu_read_unlock();
 326 }
 327
 328 /* When dgram socket disconnects (or changes its peer), we clear its receive
 329  * queue of packets arrived from previous peer. First, it allows to do
 330  * flow control based only on wmem_alloc; second, sk connected to peer
 331  * may receive messages only from that peer. */
 332 static void unix_dgram_disconnected(struct sock *sk, struct sock *other)
 333 {
 334         if (!skb_queue_empty(&sk->sk_receive_queue)) {
 335                 skb_queue_purge(&sk->sk_receive_queue);
 336                 wake_up_interruptible_all(&unix_sk(sk)->peer_wait);
 337
 338                 /* If one link of bidirectional dgram pipe is disconnected,
 339                  * we signal error. Messages are lost. Do not make this,
 340                  * when peer was not connected to us.
 341                  */
 342                 if (!sock_flag(other, SOCK_DEAD) && unix_peer(other) == sk) {
 343                         other->sk_err = ECONNRESET;
 344                         other->sk_error_report(other);
 345                 }
 346         }
 347 }
 348
 349 static void unix_sock_destructor(struct sock *sk)
 350 {
 351         struct unix_sock *u = unix_sk(sk);
 352
 353         skb_queue_purge(&sk->sk_receive_queue);
 354
 355         WARN_ON(atomic_read(&sk->sk_wmem_alloc));
 356         WARN_ON(!sk_unhashed(sk));
 357         WARN_ON(sk->sk_socket);
 358         if (!sock_flag(sk, SOCK_DEAD)) {
 359                 printk(KERN_INFO "Attempt to release alive unix socket: %p\n", sk);
 360                 return;
 361         }
 362
 363         if (u->addr)
 364                 unix_release_addr(u->addr);
 365
 366         atomic_dec(&unix_nr_socks);
 367         local_bh_disable();
 368         sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
 369         local_bh_enable();
 370 #ifdef UNIX_REFCNT_DEBUG
 371         printk(KERN_DEBUG "UNIX %p is destroyed, %d are still alive.\n", sk,
 372                 atomic_read(&unix_nr_socks));
 373 #endif
 374 }
 375
 376 static int unix_release_sock(struct sock *sk, int embrion)
 377 {
 378         struct unix_sock *u = unix_sk(sk);
 379         struct dentry *dentry;
 380         struct vfsmount *mnt;
 381         struct sock *skpair;
 382         struct sk_buff *skb;
 383         int state;
 384
 385         unix_remove_socket(sk);
 386
 387         /* Clear state */
 388         unix_state_lock(sk);
 389         sock_orphan(sk);
 390         sk->sk_shutdown = SHUTDOWN_MASK;
 391         dentry       = u->dentry;
 392         u->dentry    = NULL;
 393         mnt          = u->mnt;
 394         u->mnt       = NULL;
 395         state = sk->sk_state;
 396         sk->sk_state = TCP_CLOSE;
 397         unix_state_unlock(sk);
 398
 399         wake_up_interruptible_all(&u->peer_wait);
 400
 401         skpair = unix_peer(sk);
 402
 403         if (skpair != NULL) {
 404                 if (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) {
 405                         unix_state_lock(skpair);
 406                         /* No more writes */
 407                         skpair->sk_shutdown = SHUTDOWN_MASK;
 408                         if (!skb_queue_empty(&sk->sk_receive_queue) || embrion)
 409                                 skpair->sk_err = ECONNRESET;
 410                         unix_state_unlock(skpair);
 411                         skpair->sk_state_change(skpair);
 412                         sk_wake_async(skpair, SOCK_WAKE_WAITD, POLL_HUP);
 413                 }
 414                 sock_put(skpair); /* It may now die */
 415                 unix_peer(sk) = NULL;
 416         }
 417
 418         /* Try to flush out this socket. Throw out buffers at least */
 419
 420         while ((skb = skb_dequeue(&sk->sk_receive_queue)) != NULL) {
 421                 if (state == TCP_LISTEN)
 422                         unix_release_sock(skb->sk, 1);
 423                 /* passed fds are erased in the kfree_skb hook        */
 424                 kfree_skb(skb);
 425         }
 426
 427         if (dentry) {
 428                 dput(dentry);
 429                 mntput(mnt);
 430         }
 431
 432         sock_put(sk);
 433
 434         /* ---- Socket is dead now and most probably destroyed ---- */
 435
 436         /*
 437          * Fixme: BSD difference: In BSD all sockets connected to use get
 438          *        ECONNRESET and we die on the spot. In Linux we behave
 439          *        like files and pipes do and wait for the last
 440          *        dereference.
 441          *
 442          * Can't we simply set sock->err?
 443          *
 444          *        What the above comment does talk about? --ANK(980817)
 445          */
 446
 447         if (unix_tot_inflight)
 448                 unix_gc();              /* Garbage collect fds */
 449
 450         return 0;
 451 }
 452
 453 static int unix_listen(struct socket *sock, int backlog)
 454 {
 455         int err;
 456         struct sock *sk = sock->sk;
 457         struct unix_sock *u = unix_sk(sk);
 458
 459         err = -EOPNOTSUPP;
 460         if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
 461                 goto out;       /* Only stream/seqpacket sockets accept */
 462         err = -EINVAL;
 463         if (!u->addr)
 464                 goto out;       /* No listens on an unbound socket */
 465         unix_state_lock(sk);
 466         if (sk->sk_state != TCP_CLOSE && sk->sk_state != TCP_LISTEN)
 467                 goto out_unlock;
 468         if (backlog > sk->sk_max_ack_backlog)
 469                 wake_up_interruptible_all(&u->peer_wait);
 470         sk->sk_max_ack_backlog  = backlog;
 471         sk->sk_state            = TCP_LISTEN;
 472         /* set credentials so connect can copy them */
 473         sk->sk_peercred.pid     = task_tgid_vnr(current);
 474         current_euid_egid(&sk->sk_peercred.uid, &sk->sk_peercred.gid);
 475         err = 0;
 476
 477 out_unlock:
 478         unix_state_unlock(sk);
 479 out:
 480         return err;
 481 }
 482
 483 static int unix_release(struct socket *);
 484 static int unix_bind(struct socket *, struct sockaddr *, int);
 485 static int unix_stream_connect(struct socket *, struct sockaddr *,
 486                                int addr_len, int flags);
 487 static int unix_socketpair(struct socket *, struct socket *);
 488 static int unix_accept(struct socket *, struct socket *, int);
 489 static int unix_getname(struct socket *, struct sockaddr *, int *, int);
 490 static unsigned int unix_poll(struct file *, struct socket *, poll_table *);
 491 static unsigned int unix_dgram_poll(struct file *, struct socket *,
 492                                     poll_table *);
 493 static int unix_ioctl(struct socket *, unsigned int, unsigned long);
 494 static int unix_shutdown(struct socket *, int);
 495 static int unix_stream_sendmsg(struct kiocb *, struct socket *,
 496                                struct msghdr *, size_t);
 497 static int unix_stream_recvmsg(struct kiocb *, struct socket *,
 498                                struct msghdr *, size_t, int);
 499 static int unix_dgram_sendmsg(struct kiocb *, struct socket *,
 500                               struct msghdr *, size_t);
 501 static int unix_dgram_recvmsg(struct kiocb *, struct socket *,
 502                               struct msghdr *, size_t, int);
 503 static int unix_dgram_connect(struct socket *, struct sockaddr *,
 504                               int, int);
 505 static int unix_seqpacket_sendmsg(struct kiocb *, struct socket *,
 506                                   struct msghdr *, size_t);
 507
 508 static const struct proto_ops unix_stream_ops = {
 509         .family =       PF_UNIX,
 510         .owner =        THIS_MODULE,
 511         .release =      unix_release,
 512         .bind =         unix_bind,
 513         .connect =      unix_stream_connect,
 514         .socketpair =   unix_socketpair,
 515         .accept =       unix_accept,
 516         .getname =      unix_getname,
 517         .poll =         unix_poll,
 518         .ioctl =        unix_ioctl,
 519         .listen =       unix_listen,
 520         .shutdown =     unix_shutdown,
 521         .setsockopt =   sock_no_setsockopt,
 522         .getsockopt =   sock_no_getsockopt,
 523         .sendmsg =      unix_stream_sendmsg,
 524         .recvmsg =      unix_stream_recvmsg,
 525         .mmap =         sock_no_mmap,
 526         .sendpage =     sock_no_sendpage,
 527 };
 528
 529 static const struct proto_ops unix_dgram_ops = {
 530         .family =       PF_UNIX,
 531         .owner =        THIS_MODULE,
 532         .release =      unix_release,
 533         .bind =         unix_bind,
 534         .connect =      unix_dgram_connect,
 535         .socketpair =   unix_socketpair,
 536         .accept =       sock_no_accept,
 537         .getname =      unix_getname,
 538         .poll =         unix_dgram_poll,
 539         .ioctl =        unix_ioctl,
 540         .listen =       sock_no_listen,
 541         .shutdown =     unix_shutdown,
 542         .setsockopt =   sock_no_setsockopt,
 543         .getsockopt =   sock_no_getsockopt,
 544         .sendmsg =      unix_dgram_sendmsg,
 545         .recvmsg =      unix_dgram_recvmsg,
 546         .mmap =         sock_no_mmap,
 547         .sendpage =     sock_no_sendpage,
 548 };
 549
 550 static const struct proto_ops unix_seqpacket_ops = {
 551         .family =       PF_UNIX,
 552         .owner =        THIS_MODULE,
 553         .release =      unix_release,
 554         .bind =         unix_bind,
 555         .connect =      unix_stream_connect,
 556         .socketpair =   unix_socketpair,
 557         .accept =       unix_accept,
 558         .getname =      unix_getname,
 559         .poll =         unix_dgram_poll,
 560         .ioctl =        unix_ioctl,
 561         .listen =       unix_listen,
 562         .shutdown =     unix_shutdown,
 563         .setsockopt =   sock_no_setsockopt,
 564         .getsockopt =   sock_no_getsockopt,
 565         .sendmsg =      unix_seqpacket_sendmsg,
 566         .recvmsg =      unix_dgram_recvmsg,
 567         .mmap =         sock_no_mmap,
 568         .sendpage =     sock_no_sendpage,
 569 };
 570
 571 static struct proto unix_proto = {
 572         .name                   = "UNIX",
 573         .owner                  = THIS_MODULE,
 574         .obj_size               = sizeof(struct unix_sock),
 575 };
 576
 577 /*
 578  * AF_UNIX sockets do not interact with hardware, hence they
 579  * dont trigger interrupts - so it's safe for them to have
 580  * bh-unsafe locking for their sk_receive_queue.lock. Split off
 581  * this special lock-class by reinitializing the spinlock key:
 582  */
 583 static struct lock_class_key af_unix_sk_receive_queue_lock_key;
 584
 585 static struct sock *unix_create1(struct net *net, struct socket *sock)
 586 {
 587         struct sock *sk = NULL;
 588         struct unix_sock *u;
 589
 590         atomic_inc(&unix_nr_socks);
 591         if (atomic_read(&unix_nr_socks) > 2 * get_max_files())
 592                 goto out;
 593
 594         sk = sk_alloc(net, PF_UNIX, GFP_KERNEL, &unix_proto);
 595         if (!sk)
 596                 goto out;
 597
 598         sock_init_data(sock, sk);
 599         lockdep_set_class(&sk->sk_receive_queue.lock,
 600                                 &af_unix_sk_receive_queue_lock_key);
 601
 602         sk->sk_write_space      = unix_write_space;
 603         sk->sk_max_ack_backlog  = net->unx.sysctl_max_dgram_qlen;
 604         sk->sk_destruct         = unix_sock_destructor;
 605         u         = unix_sk(sk);
 606         u->dentry = NULL;
 607         u->mnt    = NULL;
 608         spin_lock_init(&u->lock);
 609         atomic_long_set(&u->inflight, 0);
 610         INIT_LIST_HEAD(&u->link);
 611         mutex_init(&u->readlock); /* single task reading lock */
 612         init_waitqueue_head(&u->peer_wait);
 613         unix_insert_socket(unix_sockets_unbound, sk);
 614 out:
 615         if (sk == NULL)
 616                 atomic_dec(&unix_nr_socks);
 617         else {
 618                 local_bh_disable();
 619                 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
 620                 local_bh_enable();
 621         }
 622         return sk;
 623 }
 624
 625 static int unix_create(struct net *net, struct socket *sock, int protocol,
 626                        int kern)
 627 {
 628         if (protocol && protocol != PF_UNIX)
 629                 return -EPROTONOSUPPORT;
 630
 631         sock->state = SS_UNCONNECTED;
 632
 633         switch (sock->type) {
 634         case SOCK_STREAM:
 635                 sock->ops = &unix_stream_ops;
 636                 break;
 637                 /*
 638                  *      Believe it or not BSD has AF_UNIX, SOCK_RAW though
 639                  *      nothing uses it.
 640                  */
 641         case SOCK_RAW:
 642                 sock->type = SOCK_DGRAM;
 643         case SOCK_DGRAM:
 644                 sock->ops = &unix_dgram_ops;
 645                 break;
 646         case SOCK_SEQPACKET:
 647                 sock->ops = &unix_seqpacket_ops;
 648                 break;
 649         default:
 650                 return -ESOCKTNOSUPPORT;
 651         }
 652
 653         return unix_create1(net, sock) ? 0 : -ENOMEM;
 654 }
 655
 656 static int unix_release(struct socket *sock)
 657 {
 658         struct sock *sk = sock->sk;
 659
 660         if (!sk)
 661                 return 0;
 662
 663         sock->sk = NULL;
 664
 665         return unix_release_sock(sk, 0);
 666 }
 667
 668 static int unix_autobind(struct socket *sock)
 669 {
 670         struct sock *sk = sock->sk;
 671         struct net *net = sock_net(sk);
 672         struct unix_sock *u = unix_sk(sk);
 673         static u32 ordernum = 1;
 674         struct unix_address *addr;
 675         int err;
 676
 677         mutex_lock(&u->readlock);
 678
 679         err = 0;
 680         if (u->addr)
 681                 goto out;
 682
 683         err = -ENOMEM;
 684         addr = kzalloc(sizeof(*addr) + sizeof(short) + 16, GFP_KERNEL);
 685         if (!addr)
 686                 goto out;
 687
 688         addr->name->sun_family = AF_UNIX;
 689         atomic_set(&addr->refcnt, 1);
 690
 691 retry:
 692         addr->len = sprintf(addr->name->sun_path+1, "%05x", ordernum) + 1 + sizeof(short);
 693         addr->hash = unix_hash_fold(csum_partial(addr->name, addr->len, 0));
 694
 695         spin_lock(&unix_table_lock);
 696         ordernum = (ordernum+1)&0xFFFFF;
 697
 698         if (__unix_find_socket_byname(net, addr->name, addr->len, sock->type,
 699                                       addr->hash)) {
 700                 spin_unlock(&unix_table_lock);
 701                 /* Sanity yield. It is unusual case, but yet... */
 702                 if (!(ordernum&0xFF))
 703                         yield();
 704                 goto retry;
 705         }
 706         addr->hash ^= sk->sk_type;
 707
 708         __unix_remove_socket(sk);
 709         u->addr = addr;
 710         __unix_insert_socket(&unix_socket_table[addr->hash], sk);
 711         spin_unlock(&unix_table_lock);
 712         err = 0;
 713
 714 out:    mutex_unlock(&u->readlock);
 715         return err;
 716 }
 717
 718 static struct sock *unix_find_other(struct net *net,
 719                                     struct sockaddr_un *sunname, int len,
 720                                     int type, unsigned hash, int *error)
 721 {
 722         struct sock *u;
 723         struct path path;
 724         int err = 0;
 725
 726         if (sunname->sun_path[0]) {
 727                 struct inode *inode;
 728                 err = kern_path(sunname->sun_path, LOOKUP_FOLLOW, &path);
 729                 if (err)
 730                         goto fail;
 731                 inode = path.dentry->d_inode;
 732                 err = inode_permission(inode, MAY_WRITE);
 733                 if (err)
 734                         goto put_fail;
 735
 736                 err = -ECONNREFUSED;
 737                 if (!S_ISSOCK(inode->i_mode))
 738                         goto put_fail;
 739                 u = unix_find_socket_byinode(net, inode);
 740                 if (!u)
 741                         goto put_fail;
 742
 743                 if (u->sk_type == type)
 744                         touch_atime(path.mnt, path.dentry);
 745
 746                 path_put(&path);
 747
 748                 err = -EPROTOTYPE;
 749                 if (u->sk_type != type) {
 750                         sock_put(u);
 751                         goto fail;
 752                 }
 753         } else {
 754                 err = -ECONNREFUSED;
 755                 u = unix_find_socket_byname(net, sunname, len, type, hash);
 756                 if (u) {
 757                         struct dentry *dentry;
 758                         dentry = unix_sk(u)->dentry;
 759                         if (dentry)
 760                                 touch_atime(unix_sk(u)->mnt, dentry);
 761                 } else
 762                         goto fail;
 763         }
 764         return u;
 765
 766 put_fail:
 767         path_put(&path);
 768 fail:
 769         *error = err;
 770         return NULL;
 771 }
 772
 773
 774 static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
 775 {
 776         struct sock *sk = sock->sk;
 777         struct net *net = sock_net(sk);
 778         struct unix_sock *u = unix_sk(sk);
 779         struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
 780         struct dentry *dentry = NULL;
 781         struct nameidata nd;
 782         int err;
 783         unsigned hash;
 784         struct unix_address *addr;
 785         struct hlist_head *list;
 786
 787         err = -EINVAL;
 788         if (sunaddr->sun_family != AF_UNIX)
 789                 goto out;
 790
 791         if (addr_len == sizeof(short)) {
 792                 err = unix_autobind(sock);
 793                 goto out;
 794         }
 795
 796         err = unix_mkname(sunaddr, addr_len, &hash);
 797         if (err < 0)
 798                 goto out;
 799         addr_len = err;
 800
 801         mutex_lock(&u->readlock);
 802
 803         err = -EINVAL;
 804         if (u->addr)
 805                 goto out_up;
 806
 807         err = -ENOMEM;
 808         addr = kmalloc(sizeof(*addr)+addr_len, GFP_KERNEL);
 809         if (!addr)
 810                 goto out_up;
 811
 812         memcpy(addr->name, sunaddr, addr_len);
 813         addr->len = addr_len;
 814         addr->hash = hash ^ sk->sk_type;
 815         atomic_set(&addr->refcnt, 1);
 816
 817         if (sunaddr->sun_path[0]) {
 818                 unsigned int mode;
 819                 err = 0;
 820                 /*
 821                  * Get the parent directory, calculate the hash for last
 822                  * component.
 823                  */
 824                 err = path_lookup(sunaddr->sun_path, LOOKUP_PARENT, &nd);
 825                 if (err)
 826                         goto out_mknod_parent;
 827
 828                 dentry = lookup_create(&nd, 0);
 829                 err = PTR_ERR(dentry);
 830                 if (IS_ERR(dentry))
 831                         goto out_mknod_unlock;
 832
 833                 /*
 834                  * All right, let's create it.
 835                  */
 836                 mode = S_IFSOCK |
 837                        (SOCK_INODE(sock)->i_mode & ~current_umask());
 838                 err = mnt_want_write(nd.path.mnt);
 839                 if (err)
 840                         goto out_mknod_dput;
 841                 err = security_path_mknod(&nd.path, dentry, mode, 0);
 842                 if (err)
 843                         goto out_mknod_drop_write;
 844                 err = vfs_mknod(nd.path.dentry->d_inode, dentry, mode, 0);
 845 out_mknod_drop_write:
 846                 mnt_drop_write(nd.path.mnt);
 847                 if (err)
 848                         goto out_mknod_dput;
 849                 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
 850                 dput(nd.path.dentry);
 851                 nd.path.dentry = dentry;
 852
 853                 addr->hash = UNIX_HASH_SIZE;
 854         }
 855
 856         spin_lock(&unix_table_lock);
 857
 858         if (!sunaddr->sun_path[0]) {
 859                 err = -EADDRINUSE;
 860                 if (__unix_find_socket_byname(net, sunaddr, addr_len,
 861                                               sk->sk_type, hash)) {
 862                         unix_release_addr(addr);
 863                         goto out_unlock;
 864                 }
 865
 866                 list = &unix_socket_table[addr->hash];
 867         } else {
 868                 list = &unix_socket_table[dentry->d_inode->i_ino & (UNIX_HASH_SIZE-1)];
 869                 u->dentry = nd.path.dentry;
 870                 u->mnt    = nd.path.mnt;
 871         }
 872
 873         err = 0;
 874         __unix_remove_socket(sk);
 875         u->addr = addr;
 876         __unix_insert_socket(list, sk);
 877
 878 out_unlock:
 879         spin_unlock(&unix_table_lock);
 880 out_up:
 881         mutex_unlock(&u->readlock);
 882 out:
 883         return err;
 884
 885 out_mknod_dput:
 886         dput(dentry);
 887 out_mknod_unlock:
 888         mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
 889         path_put(&nd.path);
 890 out_mknod_parent:
 891         if (err == -EEXIST)
 892                 err = -EADDRINUSE;
 893         unix_release_addr(addr);
 894         goto out_up;
 895 }
 896
 897 static void unix_state_double_lock(struct sock *sk1, struct sock *sk2)
 898 {
 899         if (unlikely(sk1 == sk2) || !sk2) {
 900                 unix_state_lock(sk1);
 901                 return;
 902         }
 903         if (sk1 < sk2) {
 904                 unix_state_lock(sk1);
 905                 unix_state_lock_nested(sk2);
 906         } else {
 907                 unix_state_lock(sk2);
 908                 unix_state_lock_nested(sk1);
 909         }
 910 }
 911
 912 static void unix_state_double_unlock(struct sock *sk1, struct sock *sk2)
 913 {
 914         if (unlikely(sk1 == sk2) || !sk2) {
 915                 unix_state_unlock(sk1);
 916                 return;
 917         }
 918         unix_state_unlock(sk1);
 919         unix_state_unlock(sk2);
 920 }
 921
 922 static int unix_dgram_connect(struct socket *sock, struct sockaddr *addr,
 923                               int alen, int flags)
 924 {
 925         struct sock *sk = sock->sk;
 926         struct net *net = sock_net(sk);
 927         struct sockaddr_un *sunaddr = (struct sockaddr_un *)addr;
 928         struct sock *other;
 929         unsigned hash;
 930         int err;
 931
 932         if (addr->sa_family != AF_UNSPEC) {
 933                 err = unix_mkname(sunaddr, alen, &hash);
 934                 if (err < 0)
 935                         goto out;
 936                 alen = err;
 937
 938                 if (test_bit(SOCK_PASSCRED, &sock->flags) &&
 939                     !unix_sk(sk)->addr && (err = unix_autobind(sock)) != 0)
 940                         goto out;
 941
 942 restart:
 943                 other = unix_find_other(net, sunaddr, alen, sock->type, hash, &err);
 944                 if (!other)
 945                         goto out;
 946
 947                 unix_state_double_lock(sk, other);
 948
 949                 /* Apparently VFS overslept socket death. Retry. */
 950                 if (sock_flag(other, SOCK_DEAD)) {
 951                         unix_state_double_unlock(sk, other);
 952                         sock_put(other);
 953                         goto restart;
 954                 }
 955
 956                 err = -EPERM;
 957                 if (!unix_may_send(sk, other))
 958                         goto out_unlock;
 959
 960                 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
 961                 if (err)
 962                         goto out_unlock;
 963
 964         } else {
 965                 /*
 966                  *      1003.1g breaking connected state with AF_UNSPEC
 967                  */
 968                 other = NULL;
 969                 unix_state_double_lock(sk, other);
 970         }
 971
 972         /*
 973          * If it was connected, reconnect.
 974          */
 975         if (unix_peer(sk)) {
 976                 struct sock *old_peer = unix_peer(sk);
 977                 unix_peer(sk) = other;
 978                 unix_state_double_unlock(sk, other);
 979
 980                 if (other != old_peer)
 981                         unix_dgram_disconnected(sk, old_peer);
 982                 sock_put(old_peer);
 983         } else {
 984                 unix_peer(sk) = other;
 985                 unix_state_double_unlock(sk, other);
 986         }
 987         return 0;
 988
 989 out_unlock:
 990         unix_state_double_unlock(sk, other);
 991         sock_put(other);
 992 out:
 993         return err;
 994 }
 995
 996 static long unix_wait_for_peer(struct sock *other, long timeo)
 997 {
 998         struct unix_sock *u = unix_sk(other);
 999         int sched;
1000         DEFINE_WAIT(wait);
1001
1002         prepare_to_wait_exclusive(&u->peer_wait, &wait, TASK_INTERRUPTIBLE);
1003
1004         sched = !sock_flag(other, SOCK_DEAD) &&
1005                 !(other->sk_shutdown & RCV_SHUTDOWN) &&
1006                 unix_recvq_full(other);
1007
1008         unix_state_unlock(other);
1009
1010         if (sched)
1011                 timeo = schedule_timeout(timeo);
1012
1013         finish_wait(&u->peer_wait, &wait);
1014         return timeo;
1015 }
1016
1017 static int unix_stream_connect(struct socket *sock, struct sockaddr *uaddr,
1018                                int addr_len, int flags)
1019 {
1020         struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
1021         struct sock *sk = sock->sk;
1022         struct net *net = sock_net(sk);
1023         struct unix_sock *u = unix_sk(sk), *newu, *otheru;
1024         struct sock *newsk = NULL;
1025         struct sock *other = NULL;
1026         struct sk_buff *skb = NULL;
1027         unsigned hash;
1028         int st;
1029         int err;
1030         long timeo;
1031
1032         err = unix_mkname(sunaddr, addr_len, &hash);
1033         if (err < 0)
1034                 goto out;
1035         addr_len = err;
1036
1037         if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr &&
1038             (err = unix_autobind(sock)) != 0)
1039                 goto out;
1040
1041         timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
1042
1043         /* First of all allocate resources.
1044            If we will make it after state is locked,
1045            we will have to recheck all again in any case.
1046          */
1047
1048         err = -ENOMEM;
1049
1050         /* create new sock for complete connection */
1051         newsk = unix_create1(sock_net(sk), NULL);
1052         if (newsk == NULL)
1053                 goto out;
1054
1055         /* Allocate skb for sending to listening sock */
1056         skb = sock_wmalloc(newsk, 1, 0, GFP_KERNEL);
1057         if (skb == NULL)
1058                 goto out;
1059
1060 restart:
1061         /*  Find listening sock. */
1062         other = unix_find_other(net, sunaddr, addr_len, sk->sk_type, hash, &err);
1063         if (!other)
1064                 goto out;
1065
1066         /* Latch state of peer */
1067         unix_state_lock(other);
1068
1069         /* Apparently VFS overslept socket death. Retry. */
1070         if (sock_flag(other, SOCK_DEAD)) {
1071                 unix_state_unlock(other);
1072                 sock_put(other);
1073                 goto restart;
1074         }
1075
1076         err = -ECONNREFUSED;
1077         if (other->sk_state != TCP_LISTEN)
1078                 goto out_unlock;
1079         if (other->sk_shutdown & RCV_SHUTDOWN)
1080                 goto out_unlock;
1081
1082         if (unix_recvq_full(other)) {
1083                 err = -EAGAIN;
1084                 if (!timeo)
1085                         goto out_unlock;
1086
1087                 timeo = unix_wait_for_peer(other, timeo);
1088
1089                 err = sock_intr_errno(timeo);
1090                 if (signal_pending(current))
1091                         goto out;
1092                 sock_put(other);
1093                 goto restart;
1094         }
1095
1096         /* Latch our state.
1097
1098            It is tricky place. We need to grab write lock and cannot
1099            drop lock on peer. It is dangerous because deadlock is
1100            possible. Connect to self case and simultaneous
1101            attempt to connect are eliminated by checking socket
1102            state. other is TCP_LISTEN, if sk is TCP_LISTEN we
1103            check this before attempt to grab lock.
1104
1105            Well, and we have to recheck the state after socket locked.
1106          */
1107         st = sk->sk_state;
1108
1109         switch (st) {
1110         case TCP_CLOSE:
1111                 /* This is ok... continue with connect */
1112                 break;
1113         case TCP_ESTABLISHED:
1114                 /* Socket is already connected */
1115                 err = -EISCONN;
1116                 goto out_unlock;
1117         default:
1118                 err = -EINVAL;
1119                 goto out_unlock;
1120         }
1121
1122         unix_state_lock_nested(sk);
1123
1124         if (sk->sk_state != st) {
1125                 unix_state_unlock(sk);
1126                 unix_state_unlock(other);
1127                 sock_put(other);
1128                 goto restart;
1129         }
1130
1131         err = security_unix_stream_connect(sock, other->sk_socket, newsk);
1132         if (err) {
1133                 unix_state_unlock(sk);
1134                 goto out_unlock;
1135         }
1136
1137         /* The way is open! Fastly set all the necessary fields... */
1138
1139         sock_hold(sk);
1140         unix_peer(newsk)        = sk;
1141         newsk->sk_state         = TCP_ESTABLISHED;
1142         newsk->sk_type          = sk->sk_type;
1143         newsk->sk_peercred.pid  = task_tgid_vnr(current);
1144         current_euid_egid(&newsk->sk_peercred.uid, &newsk->sk_peercred.gid);
1145         newu = unix_sk(newsk);
1146         newsk->sk_wq            = &newu->peer_wq;
1147         otheru = unix_sk(other);
1148
1149         /* copy address information from listening to new sock*/
1150         if (otheru->addr) {
1151                 atomic_inc(&otheru->addr->refcnt);
1152                 newu->addr = otheru->addr;
1153         }
1154         if (otheru->dentry) {
1155                 newu->dentry    = dget(otheru->dentry);
1156                 newu->mnt       = mntget(otheru->mnt);
1157         }
1158
1159         /* Set credentials */
1160         sk->sk_peercred = other->sk_peercred;
1161
1162         sock->state     = SS_CONNECTED;
1163         sk->sk_state    = TCP_ESTABLISHED;
1164         sock_hold(newsk);
1165
1166         smp_mb__after_atomic_inc();     /* sock_hold() does an atomic_inc() */
1167         unix_peer(sk)   = newsk;
1168
1169         unix_state_unlock(sk);
1170
1171         /* take ten and and send info to listening sock */
1172         spin_lock(&other->sk_receive_queue.lock);
1173         __skb_queue_tail(&other->sk_receive_queue, skb);
1174         spin_unlock(&other->sk_receive_queue.lock);
1175         unix_state_unlock(other);
1176         other->sk_data_ready(other, 0);
1177         sock_put(other);
1178         return 0;
1179
1180 out_unlock:
1181         if (other)
1182                 unix_state_unlock(other);
1183
1184 out:
1185         kfree_skb(skb);
1186         if (newsk)
1187                 unix_release_sock(newsk, 0);
1188         if (other)
1189                 sock_put(other);
1190         return err;
1191 }
1192
1193 static int unix_socketpair(struct socket *socka, struct socket *sockb)
1194 {
1195         struct sock *ska = socka->sk, *skb = sockb->sk;
1196
1197         /* Join our sockets back to back */
1198         sock_hold(ska);
1199         sock_hold(skb);
1200         unix_peer(ska) = skb;
1201         unix_peer(skb) = ska;
1202         ska->sk_peercred.pid = skb->sk_peercred.pid = task_tgid_vnr(current);
1203         current_euid_egid(&skb->sk_peercred.uid, &skb->sk_peercred.gid);
1204         ska->sk_peercred.uid = skb->sk_peercred.uid;
1205         ska->sk_peercred.gid = skb->sk_peercred.gid;
1206
1207         if (ska->sk_type != SOCK_DGRAM) {
1208                 ska->sk_state = TCP_ESTABLISHED;
1209                 skb->sk_state = TCP_ESTABLISHED;
1210                 socka->state  = SS_CONNECTED;
1211                 sockb->state  = SS_CONNECTED;
1212         }
1213         return 0;
1214 }
1215
1216 static int unix_accept(struct socket *sock, struct socket *newsock, int flags)
1217 {
1218         struct sock *sk = sock->sk;
1219         struct sock *tsk;
1220         struct sk_buff *skb;
1221         int err;
1222
1223         err = -EOPNOTSUPP;
1224         if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
1225                 goto out;
1226
1227         err = -EINVAL;
1228         if (sk->sk_state != TCP_LISTEN)
1229                 goto out;
1230
1231         /* If socket state is TCP_LISTEN it cannot change (for now...),
1232          * so that no locks are necessary.
1233          */
1234
1235         skb = skb_recv_datagram(sk, 0, flags&O_NONBLOCK, &err);
1236         if (!skb) {
1237                 /* This means receive shutdown. */
1238                 if (err == 0)
1239                         err = -EINVAL;
1240                 goto out;
1241         }
1242
1243         tsk = skb->sk;
1244         skb_free_datagram(sk, skb);
1245         wake_up_interruptible(&unix_sk(sk)->peer_wait);
1246
1247         /* attach accepted sock to socket */
1248         unix_state_lock(tsk);
1249         newsock->state = SS_CONNECTED;
1250         sock_graft(tsk, newsock);
1251         unix_state_unlock(tsk);
1252         return 0;
1253
1254 out:
1255         return err;
1256 }
1257
1258
1259 static int unix_getname(struct socket *sock, struct sockaddr *uaddr, int *uaddr_len, int peer)
1260 {
1261         struct sock *sk = sock->sk;
1262         struct unix_sock *u;
1263         DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr, uaddr);
1264         int err = 0;
1265
1266         if (peer) {
1267                 sk = unix_peer_get(sk);
1268
1269                 err = -ENOTCONN;
1270                 if (!sk)
1271                         goto out;
1272                 err = 0;
1273         } else {
1274                 sock_hold(sk);
1275         }
1276
1277         u = unix_sk(sk);
1278         unix_state_lock(sk);
1279         if (!u->addr) {
1280                 sunaddr->sun_family = AF_UNIX;
1281                 sunaddr->sun_path[0] = 0;
1282                 *uaddr_len = sizeof(short);
1283         } else {
1284                 struct unix_address *addr = u->addr;
1285
1286                 *uaddr_len = addr->len;
1287                 memcpy(sunaddr, addr->name, *uaddr_len);
1288         }
1289         unix_state_unlock(sk);
1290         sock_put(sk);
1291 out:
1292         return err;
1293 }
1294
1295 static void unix_detach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1296 {
1297         int i;
1298
1299         scm->fp = UNIXCB(skb).fp;
1300         skb->destructor = sock_wfree;
1301         UNIXCB(skb).fp = NULL;
1302
1303         for (i = scm->fp->count-1; i >= 0; i--)
1304                 unix_notinflight(scm->fp->fp[i]);
1305 }
1306
1307 static void unix_destruct_fds(struct sk_buff *skb)
1308 {
1309         struct scm_cookie scm;
1310         memset(&scm, 0, sizeof(scm));
1311         unix_detach_fds(&scm, skb);
1312
1313         /* Alas, it calls VFS */
1314         /* So fscking what? fput() had been SMP-safe since the last Summer */
1315         scm_destroy(&scm);
1316         sock_wfree(skb);
1317 }
1318
1319 static int unix_attach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1320 {
1321         int i;
1322
1323         /*
1324          * Need to duplicate file references for the sake of garbage
1325          * collection.  Otherwise a socket in the fps might become a
1326          * candidate for GC while the skb is not yet queued.
1327          */
1328         UNIXCB(skb).fp = scm_fp_dup(scm->fp);
1329         if (!UNIXCB(skb).fp)
1330                 return -ENOMEM;
1331
1332         for (i = scm->fp->count-1; i >= 0; i--)
1333                 unix_inflight(scm->fp->fp[i]);
1334         skb->destructor = unix_destruct_fds;
1335         return 0;
1336 }
1337
1338 /*
1339  *      Send AF_UNIX data.
1340  */
1341
1342 static int unix_dgram_sendmsg(struct kiocb *kiocb, struct socket *sock,
1343                               struct msghdr *msg, size_t len)
1344 {
1345         struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1346         struct sock *sk = sock->sk;
1347         struct net *net = sock_net(sk);
1348         struct unix_sock *u = unix_sk(sk);
1349         struct sockaddr_un *sunaddr = msg->msg_name;
1350         struct sock *other = NULL;
1351         int namelen = 0; /* fake GCC */
1352         int err;
1353         unsigned hash;
1354         struct sk_buff *skb;
1355         long timeo;
1356         struct scm_cookie tmp_scm;
1357
1358         if (NULL == siocb->scm)
1359                 siocb->scm = &tmp_scm;
1360         wait_for_unix_gc();
1361         err = scm_send(sock, msg, siocb->scm);
1362         if (err < 0)
1363                 return err;
1364
1365         err = -EOPNOTSUPP;
1366         if (msg->msg_flags&MSG_OOB)
1367                 goto out;
1368
1369         if (msg->msg_namelen) {
1370                 err = unix_mkname(sunaddr, msg->msg_namelen, &hash);
1371                 if (err < 0)
1372                         goto out;
1373                 namelen = err;
1374         } else {
1375                 sunaddr = NULL;
1376                 err = -ENOTCONN;
1377                 other = unix_peer_get(sk);
1378                 if (!other)
1379                         goto out;
1380         }
1381
1382         if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr
1383             && (err = unix_autobind(sock)) != 0)
1384                 goto out;
1385
1386         err = -EMSGSIZE;
1387         if (len > sk->sk_sndbuf - 32)
1388                 goto out;
1389
1390         skb = sock_alloc_send_skb(sk, len, msg->msg_flags&MSG_DONTWAIT, &err);
1391         if (skb == NULL)
1392                 goto out;
1393
1394         memcpy(UNIXCREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1395         if (siocb->scm->fp) {
1396                 err = unix_attach_fds(siocb->scm, skb);
1397                 if (err)
1398                         goto out_free;
1399         }
1400         unix_get_secdata(siocb->scm, skb);
1401
1402         skb_reset_transport_header(skb);
1403         err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
1404         if (err)
1405                 goto out_free;
1406
1407         timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1408
1409 restart:
1410         if (!other) {
1411                 err = -ECONNRESET;
1412                 if (sunaddr == NULL)
1413                         goto out_free;
1414
1415                 other = unix_find_other(net, sunaddr, namelen, sk->sk_type,
1416                                         hash, &err);
1417                 if (other == NULL)
1418                         goto out_free;
1419         }
1420
1421         unix_state_lock(other);
1422         err = -EPERM;
1423         if (!unix_may_send(sk, other))
1424                 goto out_unlock;
1425
1426         if (sock_flag(other, SOCK_DEAD)) {
1427                 /*
1428                  *      Check with 1003.1g - what should
1429                  *      datagram error
1430                  */
1431                 unix_state_unlock(other);
1432                 sock_put(other);
1433
1434                 err = 0;
1435                 unix_state_lock(sk);
1436                 if (unix_peer(sk) == other) {
1437                         unix_peer(sk) = NULL;
1438                         unix_state_unlock(sk);
1439
1440                         unix_dgram_disconnected(sk, other);
1441                         sock_put(other);
1442                         err = -ECONNREFUSED;
1443                 } else {
1444                         unix_state_unlock(sk);
1445                 }
1446
1447                 other = NULL;
1448                 if (err)
1449                         goto out_free;
1450                 goto restart;
1451         }
1452
1453         err = -EPIPE;
1454         if (other->sk_shutdown & RCV_SHUTDOWN)
1455                 goto out_unlock;
1456
1457         if (sk->sk_type != SOCK_SEQPACKET) {
1458                 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1459                 if (err)
1460                         goto out_unlock;
1461         }
1462
1463         if (unix_peer(other) != sk && unix_recvq_full(other)) {
1464                 if (!timeo) {
1465                         err = -EAGAIN;
1466                         goto out_unlock;
1467                 }
1468
1469                 timeo = unix_wait_for_peer(other, timeo);
1470
1471                 err = sock_intr_errno(timeo);
1472                 if (signal_pending(current))
1473                         goto out_free;
1474
1475                 goto restart;
1476         }
1477
1478         skb_queue_tail(&other->sk_receive_queue, skb);
1479         unix_state_unlock(other);
1480         other->sk_data_ready(other, len);
1481         sock_put(other);
1482         scm_destroy(siocb->scm);
1483         return len;
1484
1485 out_unlock:
1486         unix_state_unlock(other);
1487 out_free:
1488         kfree_skb(skb);
1489 out:
1490         if (other)
1491                 sock_put(other);
1492         scm_destroy(siocb->scm);
1493         return err;
1494 }
1495
1496
1497 static int unix_stream_sendmsg(struct kiocb *kiocb, struct socket *sock,
1498                                struct msghdr *msg, size_t len)
1499 {
1500         struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1501         struct sock *sk = sock->sk;
1502         struct sock *other = NULL;
1503         struct sockaddr_un *sunaddr = msg->msg_name;
1504         int err, size;
1505         struct sk_buff *skb;
1506         int sent = 0;
1507         struct scm_cookie tmp_scm;
1508         bool fds_sent = false;
1509
1510         if (NULL == siocb->scm)
1511                 siocb->scm = &tmp_scm;
1512         wait_for_unix_gc();
1513         err = scm_send(sock, msg, siocb->scm);
1514         if (err < 0)
1515                 return err;
1516
1517         err = -EOPNOTSUPP;
1518         if (msg->msg_flags&MSG_OOB)
1519                 goto out_err;
1520
1521         if (msg->msg_namelen) {
1522                 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
1523                 goto out_err;
1524         } else {
1525                 sunaddr = NULL;
1526                 err = -ENOTCONN;
1527                 other = unix_peer(sk);
1528                 if (!other)
1529                         goto out_err;
1530         }
1531
1532         if (sk->sk_shutdown & SEND_SHUTDOWN)
1533                 goto pipe_err;
1534
1535         while (sent < len) {
1536                 /*
1537                  *      Optimisation for the fact that under 0.01% of X
1538                  *      messages typically need breaking up.
1539                  */
1540
1541                 size = len-sent;
1542
1543                 /* Keep two messages in the pipe so it schedules better */
1544                 if (size > ((sk->sk_sndbuf >> 1) - 64))
1545                         size = (sk->sk_sndbuf >> 1) - 64;
1546
1547                 if (size > SKB_MAX_ALLOC)
1548                         size = SKB_MAX_ALLOC;
1549
1550                 /*
1551                  *      Grab a buffer
1552                  */
1553
1554                 skb = sock_alloc_send_skb(sk, size, msg->msg_flags&MSG_DONTWAIT,
1555                                           &err);
1556
1557                 if (skb == NULL)
1558                         goto out_err;
1559
1560                 /*
1561                  *      If you pass two values to the sock_alloc_send_skb
1562                  *      it tries to grab the large buffer with GFP_NOFS
1563                  *      (which can fail easily), and if it fails grab the
1564                  *      fallback size buffer which is under a page and will
1565                  *      succeed. [Alan]
1566                  */
1567                 size = min_t(int, size, skb_tailroom(skb));
1568
1569                 memcpy(UNIXCREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1570                 /* Only send the fds in the first buffer */
1571                 if (siocb->scm->fp && !fds_sent) {
1572                         err = unix_attach_fds(siocb->scm, skb);
1573                         if (err) {
1574                                 kfree_skb(skb);
1575                                 goto out_err;
1576                         }
1577                         fds_sent = true;
1578                 }
1579
1580                 err = memcpy_fromiovec(skb_put(skb, size), msg->msg_iov, size);
1581                 if (err) {
1582                         kfree_skb(skb);
1583                         goto out_err;
1584                 }
1585
1586                 unix_state_lock(other);
1587
1588                 if (sock_flag(other, SOCK_DEAD) ||
1589                     (other->sk_shutdown & RCV_SHUTDOWN))
1590                         goto pipe_err_free;
1591
1592                 skb_queue_tail(&other->sk_receive_queue, skb);
1593                 unix_state_unlock(other);
1594                 other->sk_data_ready(other, size);
1595                 sent += size;
1596         }
1597
1598         scm_destroy(siocb->scm);
1599         siocb->scm = NULL;
1600
1601         return sent;
1602
1603 pipe_err_free:
1604         unix_state_unlock(other);
1605         kfree_skb(skb);
1606 pipe_err:
1607         if (sent == 0 && !(msg->msg_flags&MSG_NOSIGNAL))
1608                 send_sig(SIGPIPE, current, 0);
1609         err = -EPIPE;
1610 out_err:
1611         scm_destroy(siocb->scm);
1612         siocb->scm = NULL;
1613         return sent ? : err;
1614 }
1615
1616 static int unix_seqpacket_sendmsg(struct kiocb *kiocb, struct socket *sock,
1617                                   struct msghdr *msg, size_t len)
1618 {
1619         int err;
1620         struct sock *sk = sock->sk;
1621
1622         err = sock_error(sk);
1623         if (err)
1624                 return err;
1625
1626         if (sk->sk_state != TCP_ESTABLISHED)
1627                 return -ENOTCONN;
1628
1629         if (msg->msg_namelen)
1630                 msg->msg_namelen = 0;
1631
1632         return unix_dgram_sendmsg(kiocb, sock, msg, len);
1633 }
1634
1635 static void unix_copy_addr(struct msghdr *msg, struct sock *sk)
1636 {
1637         struct unix_sock *u = unix_sk(sk);
1638
1639         msg->msg_namelen = 0;
1640         if (u->addr) {
1641                 msg->msg_namelen = u->addr->len;
1642                 memcpy(msg->msg_name, u->addr->name, u->addr->len);
1643         }
1644 }
1645
1646 static int unix_dgram_recvmsg(struct kiocb *iocb, struct socket *sock,
1647                               struct msghdr *msg, size_t size,
1648                               int flags)
1649 {
1650         struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1651         struct scm_cookie tmp_scm;
1652         struct sock *sk = sock->sk;
1653         struct unix_sock *u = unix_sk(sk);
1654         int noblock = flags & MSG_DONTWAIT;
1655         struct sk_buff *skb;
1656         int err;
1657
1658         err = -EOPNOTSUPP;
1659         if (flags&MSG_OOB)
1660                 goto out;
1661
1662         msg->msg_namelen = 0;
1663
1664         mutex_lock(&u->readlock);
1665
1666         skb = skb_recv_datagram(sk, flags, noblock, &err);
1667         if (!skb) {
1668                 unix_state_lock(sk);
1669                 /* Signal EOF on disconnected non-blocking SEQPACKET socket. */
1670                 if (sk->sk_type == SOCK_SEQPACKET && err == -EAGAIN &&
1671                     (sk->sk_shutdown & RCV_SHUTDOWN))
1672                         err = 0;
1673                 unix_state_unlock(sk);
1674                 goto out_unlock;
1675         }
1676
1677         wake_up_interruptible_sync(&u->peer_wait);
1678
1679         if (msg->msg_name)
1680                 unix_copy_addr(msg, skb->sk);
1681
1682         if (size > skb->len)
1683                 size = skb->len;
1684         else if (size < skb->len)
1685                 msg->msg_flags |= MSG_TRUNC;
1686
1687         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, size);
1688         if (err)
1689                 goto out_free;
1690
1691         if (!siocb->scm) {
1692                 siocb->scm = &tmp_scm;
1693                 memset(&tmp_scm, 0, sizeof(tmp_scm));
1694         }
1695         siocb->scm->creds = *UNIXCREDS(skb);
1696         unix_set_secdata(siocb->scm, skb);
1697
1698         if (!(flags & MSG_PEEK)) {
1699                 if (UNIXCB(skb).fp)
1700                         unix_detach_fds(siocb->scm, skb);
1701         } else {
1702                 /* It is questionable: on PEEK we could:
1703                    - do not return fds - good, but too simple 8)
1704                    - return fds, and do not return them on read (old strategy,
1705                      apparently wrong)
1706                    - clone fds (I chose it for now, it is the most universal
1707                      solution)
1708
1709                    POSIX 1003.1g does not actually define this clearly
1710                    at all. POSIX 1003.1g doesn't define a lot of things
1711                    clearly however!
1712
1713                 */
1714                 if (UNIXCB(skb).fp)
1715                         siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1716         }
1717         err = size;
1718
1719         scm_recv(sock, msg, siocb->scm, flags);
1720
1721 out_free:
1722         skb_free_datagram(sk, skb);
1723 out_unlock:
1724         mutex_unlock(&u->readlock);
1725 out:
1726         return err;
1727 }
1728
1729 /*
1730  *      Sleep until data has arrive. But check for races..
1731  */
1732
1733 static long unix_stream_data_wait(struct sock *sk, long timeo)
1734 {
1735         DEFINE_WAIT(wait);
1736
1737         unix_state_lock(sk);
1738
1739         for (;;) {
1740                 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1741
1742                 if (!skb_queue_empty(&sk->sk_receive_queue) ||
1743                     sk->sk_err ||
1744                     (sk->sk_shutdown & RCV_SHUTDOWN) ||
1745                     signal_pending(current) ||
1746                     !timeo)
1747                         break;
1748
1749                 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1750                 unix_state_unlock(sk);
1751                 timeo = schedule_timeout(timeo);
1752                 unix_state_lock(sk);
1753                 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1754         }
1755
1756         finish_wait(sk_sleep(sk), &wait);
1757         unix_state_unlock(sk);
1758         return timeo;
1759 }
1760
1761
1762
1763 static int unix_stream_recvmsg(struct kiocb *iocb, struct socket *sock,
1764                                struct msghdr *msg, size_t size,
1765                                int flags)
1766 {
1767         struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1768         struct scm_cookie tmp_scm;
1769         struct sock *sk = sock->sk;
1770         struct unix_sock *u = unix_sk(sk);
1771         struct sockaddr_un *sunaddr = msg->msg_name;
1772         int copied = 0;
1773         int check_creds = 0;
1774         int target;
1775         int err = 0;
1776         long timeo;
1777
1778         err = -EINVAL;
1779         if (sk->sk_state != TCP_ESTABLISHED)
1780                 goto out;
1781
1782         err = -EOPNOTSUPP;
1783         if (flags&MSG_OOB)
1784                 goto out;
1785
1786         target = sock_rcvlowat(sk, flags&MSG_WAITALL, size);
1787         timeo = sock_rcvtimeo(sk, flags&MSG_DONTWAIT);
1788
1789         msg->msg_namelen = 0;
1790
1791         /* Lock the socket to prevent queue disordering
1792          * while sleeps in memcpy_tomsg
1793          */
1794
1795         if (!siocb->scm) {
1796                 siocb->scm = &tmp_scm;
1797                 memset(&tmp_scm, 0, sizeof(tmp_scm));
1798         }
1799
1800         mutex_lock(&u->readlock);
1801
1802         do {
1803                 int chunk;
1804                 struct sk_buff *skb;
1805
1806                 unix_state_lock(sk);
1807                 skb = skb_dequeue(&sk->sk_receive_queue);
1808                 if (skb == NULL) {
1809                         if (copied >= target)
1810                                 goto unlock;
1811
1812                         /*
1813                          *      POSIX 1003.1g mandates this order.
1814                          */
1815
1816                         err = sock_error(sk);
1817                         if (err)
1818                                 goto unlock;
1819                         if (sk->sk_shutdown & RCV_SHUTDOWN)
1820                                 goto unlock;
1821
1822                         unix_state_unlock(sk);
1823                         err = -EAGAIN;
1824                         if (!timeo)
1825                                 break;
1826                         mutex_unlock(&u->readlock);
1827
1828                         timeo = unix_stream_data_wait(sk, timeo);
1829
1830                         if (signal_pending(current)) {
1831                                 err = sock_intr_errno(timeo);
1832                                 goto out;
1833                         }
1834                         mutex_lock(&u->readlock);
1835                         continue;
1836  unlock:
1837                         unix_state_unlock(sk);
1838                         break;
1839                 }
1840                 unix_state_unlock(sk);
1841
1842                 if (check_creds) {
1843                         /* Never glue messages from different writers */
1844                         if (memcmp(UNIXCREDS(skb), &siocb->scm->creds,
1845                                    sizeof(siocb->scm->creds)) != 0) {
1846                                 skb_queue_head(&sk->sk_receive_queue, skb);
1847                                 break;
1848                         }
1849                 } else {
1850                         /* Copy credentials */
1851                         siocb->scm->creds = *UNIXCREDS(skb);
1852                         check_creds = 1;
1853                 }
1854
1855                 /* Copy address just once */
1856                 if (sunaddr) {
1857                         unix_copy_addr(msg, skb->sk);
1858                         sunaddr = NULL;
1859                 }
1860
1861                 chunk = min_t(unsigned int, skb->len, size);
1862                 if (memcpy_toiovec(msg->msg_iov, skb->data, chunk)) {
1863                         skb_queue_head(&sk->sk_receive_queue, skb);
1864                         if (copied == 0)
1865                                 copied = -EFAULT;
1866                         break;
1867                 }
1868                 copied += chunk;
1869                 size -= chunk;
1870
1871                 /* Mark read part of skb as used */
1872                 if (!(flags & MSG_PEEK)) {
1873                         skb_pull(skb, chunk);
1874
1875                         if (UNIXCB(skb).fp)
1876                                 unix_detach_fds(siocb->scm, skb);
1877
1878                         /* put the skb back if we didn't use it up.. */
1879                         if (skb->len) {
1880                                 skb_queue_head(&sk->sk_receive_queue, skb);
1881                                 break;
1882                         }
1883
1884                         kfree_skb(skb);
1885
1886                         if (siocb->scm->fp)
1887                                 break;
1888                 } else {
1889                         /* It is questionable, see note in unix_dgram_recvmsg.
1890                          */
1891                         if (UNIXCB(skb).fp)
1892                                 siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1893
1894                         /* put message back and return */
1895                         skb_queue_head(&sk->sk_receive_queue, skb);
1896                         break;
1897                 }
1898         } while (size);
1899
1900         mutex_unlock(&u->readlock);
1901         scm_recv(sock, msg, siocb->scm, flags);
1902 out:
1903         return copied ? : err;
1904 }
1905
1906 static int unix_shutdown(struct socket *sock, int mode)
1907 {
1908         struct sock *sk = sock->sk;
1909         struct sock *other;
1910
1911         mode = (mode+1)&(RCV_SHUTDOWN|SEND_SHUTDOWN);
1912
1913         if (mode) {
1914                 unix_state_lock(sk);
1915                 sk->sk_shutdown |= mode;
1916                 other = unix_peer(sk);
1917                 if (other)
1918                         sock_hold(other);
1919                 unix_state_unlock(sk);
1920                 sk->sk_state_change(sk);
1921
1922                 if (other &&
1923                         (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET)) {
1924
1925                         int peer_mode = 0;
1926
1927                         if (mode&RCV_SHUTDOWN)
1928                                 peer_mode |= SEND_SHUTDOWN;
1929                         if (mode&SEND_SHUTDOWN)
1930                                 peer_mode |= RCV_SHUTDOWN;
1931                         unix_state_lock(other);
1932                         other->sk_shutdown |= peer_mode;
1933                         unix_state_unlock(other);
1934                         other->sk_state_change(other);
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                 }
1940                 if (other)
1941                         sock_put(other);
1942         }
1943         return 0;
1944 }
1945
1946 static int unix_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1947 {
1948         struct sock *sk = sock->sk;
1949         long amount = 0;
1950         int err;
1951
1952         switch (cmd) {
1953         case SIOCOUTQ:
1954                 amount = sk_wmem_alloc_get(sk);
1955                 err = put_user(amount, (int __user *)arg);
1956                 break;
1957         case SIOCINQ:
1958                 {
1959                         struct sk_buff *skb;
1960
1961                         if (sk->sk_state == TCP_LISTEN) {
1962                                 err = -EINVAL;
1963                                 break;
1964                         }
1965
1966                         spin_lock(&sk->sk_receive_queue.lock);
1967                         if (sk->sk_type == SOCK_STREAM ||
1968                             sk->sk_type == SOCK_SEQPACKET) {
1969                                 skb_queue_walk(&sk->sk_receive_queue, skb)
1970                                         amount += skb->len;
1971                         } else {
1972                                 skb = skb_peek(&sk->sk_receive_queue);
1973                                 if (skb)
1974                                         amount = skb->len;
1975                         }
1976                         spin_unlock(&sk->sk_receive_queue.lock);
1977                         err = put_user(amount, (int __user *)arg);
1978                         break;
1979                 }
1980
1981         default:
1982                 err = -ENOIOCTLCMD;
1983                 break;
1984         }
1985         return err;
1986 }
1987
1988 static unsigned int unix_poll(struct file *file, struct socket *sock, poll_table *wait)
1989 {
1990         struct sock *sk = sock->sk;
1991         unsigned int mask;
1992
1993         sock_poll_wait(file, sk_sleep(sk), wait);
1994         mask = 0;
1995
1996         /* exceptional events? */
1997         if (sk->sk_err)
1998                 mask |= POLLERR;
1999         if (sk->sk_shutdown == SHUTDOWN_MASK)
2000                 mask |= POLLHUP;
2001         if (sk->sk_shutdown & RCV_SHUTDOWN)
2002                 mask |= POLLRDHUP;
2003
2004         /* readable? */
2005         if (!skb_queue_empty(&sk->sk_receive_queue) ||
2006             (sk->sk_shutdown & RCV_SHUTDOWN))
2007                 mask |= POLLIN | POLLRDNORM;
2008
2009         /* Connection-based need to check for termination and startup */
2010         if ((sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) &&
2011             sk->sk_state == TCP_CLOSE)
2012                 mask |= POLLHUP;
2013
2014         /*
2015          * we set writable also when the other side has shut down the
2016          * connection. This prevents stuck sockets.
2017          */
2018         if (unix_writable(sk))
2019                 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2020
2021         return mask;
2022 }
2023
2024 static unsigned int unix_dgram_poll(struct file *file, struct socket *sock,
2025                                     poll_table *wait)
2026 {
2027         struct sock *sk = sock->sk, *other;
2028         unsigned int mask, writable;
2029
2030         sock_poll_wait(file, sk_sleep(sk), wait);
2031         mask = 0;
2032
2033         /* exceptional events? */
2034         if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue))
2035                 mask |= POLLERR;
2036         if (sk->sk_shutdown & RCV_SHUTDOWN)
2037                 mask |= POLLRDHUP;
2038         if (sk->sk_shutdown == SHUTDOWN_MASK)
2039                 mask |= POLLHUP;
2040
2041         /* readable? */
2042         if (!skb_queue_empty(&sk->sk_receive_queue) ||
2043             (sk->sk_shutdown & RCV_SHUTDOWN))
2044                 mask |= POLLIN | POLLRDNORM;
2045
2046         /* Connection-based need to check for termination and startup */
2047         if (sk->sk_type == SOCK_SEQPACKET) {
2048                 if (sk->sk_state == TCP_CLOSE)
2049                         mask |= POLLHUP;
2050                 /* connection hasn't started yet? */
2051                 if (sk->sk_state == TCP_SYN_SENT)
2052                         return mask;
2053         }
2054
2055         /* writable? */
2056         writable = unix_writable(sk);
2057         if (writable) {
2058                 other = unix_peer_get(sk);
2059                 if (other) {
2060                         if (unix_peer(other) != sk) {
2061                                 sock_poll_wait(file, &unix_sk(other)->peer_wait,
2062                                           wait);
2063                                 if (unix_recvq_full(other))
2064                                         writable = 0;
2065                         }
2066
2067                         sock_put(other);
2068                 }
2069         }
2070
2071         if (writable)
2072                 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2073         else
2074                 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
2075
2076         return mask;
2077 }
2078
2079 #ifdef CONFIG_PROC_FS
2080 static struct sock *first_unix_socket(int *i)
2081 {
2082         for (*i = 0; *i <= UNIX_HASH_SIZE; (*i)++) {
2083                 if (!hlist_empty(&unix_socket_table[*i]))
2084                         return __sk_head(&unix_socket_table[*i]);
2085         }
2086         return NULL;
2087 }
2088
2089 static struct sock *next_unix_socket(int *i, struct sock *s)
2090 {
2091         struct sock *next = sk_next(s);
2092         /* More in this chain? */
2093         if (next)
2094                 return next;
2095         /* Look for next non-empty chain. */
2096         for ((*i)++; *i <= UNIX_HASH_SIZE; (*i)++) {
2097                 if (!hlist_empty(&unix_socket_table[*i]))
2098                         return __sk_head(&unix_socket_table[*i]);
2099         }
2100         return NULL;
2101 }
2102
2103 struct unix_iter_state {
2104         struct seq_net_private p;
2105         int i;
2106 };
2107
2108 static struct sock *unix_seq_idx(struct seq_file *seq, loff_t pos)
2109 {
2110         struct unix_iter_state *iter = seq->private;
2111         loff_t off = 0;
2112         struct sock *s;
2113
2114         for (s = first_unix_socket(&iter->i); s; s = next_unix_socket(&iter->i, s)) {
2115                 if (sock_net(s) != seq_file_net(seq))
2116                         continue;
2117                 if (off == pos)
2118                         return s;
2119                 ++off;
2120         }
2121         return NULL;
2122 }
2123
2124 static void *unix_seq_start(struct seq_file *seq, loff_t *pos)
2125         __acquires(unix_table_lock)
2126 {
2127         spin_lock(&unix_table_lock);
2128         return *pos ? unix_seq_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2129 }
2130
2131 static void *unix_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2132 {
2133         struct unix_iter_state *iter = seq->private;
2134         struct sock *sk = v;
2135         ++*pos;
2136
2137         if (v == SEQ_START_TOKEN)
2138                 sk = first_unix_socket(&iter->i);
2139         else
2140                 sk = next_unix_socket(&iter->i, sk);
2141         while (sk && (sock_net(sk) != seq_file_net(seq)))
2142                 sk = next_unix_socket(&iter->i, sk);
2143         return sk;
2144 }
2145
2146 static void unix_seq_stop(struct seq_file *seq, void *v)
2147         __releases(unix_table_lock)
2148 {
2149         spin_unlock(&unix_table_lock);
2150 }
2151
2152 static int unix_seq_show(struct seq_file *seq, void *v)
2153 {
2154
2155         if (v == SEQ_START_TOKEN)
2156                 seq_puts(seq, "Num       RefCount Protocol Flags    Type St "
2157                          "Inode Path\n");
2158         else {
2159                 struct sock *s = v;
2160                 struct unix_sock *u = unix_sk(s);
2161                 unix_state_lock(s);
2162
2163                 seq_printf(seq, "%p: %08X %08X %08X %04X %02X %5lu",
2164                         s,
2165                         atomic_read(&s->sk_refcnt),
2166                         0,
2167                         s->sk_state == TCP_LISTEN ? __SO_ACCEPTCON : 0,
2168                         s->sk_type,
2169                         s->sk_socket ?
2170                         (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTED : SS_UNCONNECTED) :
2171                         (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTING : SS_DISCONNECTING),
2172                         sock_i_ino(s));
2173
2174                 if (u->addr) {
2175                         int i, len;
2176                         seq_putc(seq, ' ');
2177
2178                         i = 0;
2179                         len = u->addr->len - sizeof(short);
2180                         if (!UNIX_ABSTRACT(s))
2181                                 len--;
2182                         else {
2183                                 seq_putc(seq, '@');
2184                                 i++;
2185                         }
2186                         for ( ; i < len; i++)
2187                                 seq_putc(seq, u->addr->name->sun_path[i]);
2188                 }
2189                 unix_state_unlock(s);
2190                 seq_putc(seq, '\n');
2191         }
2192
2193         return 0;
2194 }
2195
2196 static const struct seq_operations unix_seq_ops = {
2197         .start  = unix_seq_start,
2198         .next   = unix_seq_next,
2199         .stop   = unix_seq_stop,
2200         .show   = unix_seq_show,
2201 };
2202
2203 static int unix_seq_open(struct inode *inode, struct file *file)
2204 {
2205         return seq_open_net(inode, file, &unix_seq_ops,
2206                             sizeof(struct unix_iter_state));
2207 }
2208
2209 static const struct file_operations unix_seq_fops = {
2210         .owner          = THIS_MODULE,
2211         .open           = unix_seq_open,
2212         .read           = seq_read,
2213         .llseek         = seq_lseek,
2214         .release        = seq_release_net,
2215 };
2216
2217 #endif
2218
2219 static const struct net_proto_family unix_family_ops = {
2220         .family = PF_UNIX,
2221         .create = unix_create,
2222         .owner  = THIS_MODULE,
2223 };
2224
2225
2226 static int __net_init unix_net_init(struct net *net)
2227 {
2228         int error = -ENOMEM;
2229
2230         net->unx.sysctl_max_dgram_qlen = 10;
2231         if (unix_sysctl_register(net))
2232                 goto out;
2233
2234 #ifdef CONFIG_PROC_FS
2235         if (!proc_net_fops_create(net, "unix", 0, &unix_seq_fops)) {
2236                 unix_sysctl_unregister(net);
2237                 goto out;
2238         }
2239 #endif
2240         error = 0;
2241 out:
2242         return error;
2243 }
2244
2245 static void __net_exit unix_net_exit(struct net *net)
2246 {
2247         unix_sysctl_unregister(net);
2248         proc_net_remove(net, "unix");
2249 }
2250
2251 static struct pernet_operations unix_net_ops = {
2252         .init = unix_net_init,
2253         .exit = unix_net_exit,
2254 };
2255
2256 static int __init af_unix_init(void)
2257 {
2258         int rc = -1;
2259         struct sk_buff *dummy_skb;
2260
2261         BUILD_BUG_ON(sizeof(struct unix_skb_parms) > sizeof(dummy_skb->cb));
2262
2263         rc = proto_register(&unix_proto, 1);
2264         if (rc != 0) {
2265                 printk(KERN_CRIT "%s: Cannot create unix_sock SLAB cache!\n",
2266                        __func__);
2267                 goto out;
2268         }
2269
2270         sock_register(&unix_family_ops);
2271         register_pernet_subsys(&unix_net_ops);
2272 out:
2273         return rc;
2274 }
2275
2276 static void __exit af_unix_exit(void)
2277 {
2278         sock_unregister(PF_UNIX);
2279         proto_unregister(&unix_proto);
2280         unregister_pernet_subsys(&unix_net_ops);
2281 }
2282
2283 /* Earlier than device_initcall() so that other drivers invoking
2284    request_module() don't end up in a loop when modprobe tries
2285    to use a UNIX socket. But later than subsys_initcall() because
2286    we depend on stuff initialised there */
2287 fs_initcall(af_unix_init);
2288 module_exit(af_unix_exit);
2289
2290 MODULE_LICENSE("GPL");
2291 MODULE_ALIAS_NETPROTO(PF_UNIX);