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

   1 /*
   2  * NET4:        Implementation of BSD Unix domain sockets.
   3  *
   4  * Authors:     Alan Cox, <alan@lxorguk.ukuu.org.uk>
   5  *
   6  *              This program is free software; you can redistribute it and/or
   7  *              modify it under the terms of the GNU General Public License
   8  *              as published by the Free Software Foundation; either version
   9  *              2 of the License, or (at your option) any later version.
  10  *
  11  * Fixes:
  12  *              Linus Torvalds  :       Assorted bug cures.
  13  *              Niibe Yutaka    :       async I/O support.
  14  *              Carsten Paeth   :       PF_UNIX check, address fixes.
  15  *              Alan Cox        :       Limit size of allocated blocks.
  16  *              Alan Cox        :       Fixed the stupid socketpair bug.
  17  *              Alan Cox        :       BSD compatibility fine tuning.
  18  *              Alan Cox        :       Fixed a bug in connect when interrupted.
  19  *              Alan Cox        :       Sorted out a proper draft version of
  20  *                                      file descriptor passing hacked up from
  21  *                                      Mike Shaver's work.
  22  *              Marty Leisner   :       Fixes to fd passing
  23  *              Nick Nevin      :       recvmsg bugfix.
  24  *              Alan Cox        :       Started proper garbage collector
  25  *              Heiko EiBfeldt  :       Missing verify_area check
  26  *              Alan Cox        :       Started POSIXisms
  27  *              Andreas Schwab  :       Replace inode by dentry for proper
  28  *                                      reference counting
  29  *              Kirk Petersen   :       Made this a module
  30  *          Christoph Rohland   :       Elegant non-blocking accept/connect algorithm.
  31  *                                      Lots of bug fixes.
  32  *           Alexey Kuznetosv   :       Repaired (I hope) bugs introduces
  33  *                                      by above two patches.
  34  *           Andrea Arcangeli   :       If possible we block in connect(2)
  35  *                                      if the max backlog of the listen socket
  36  *                                      is been reached. This won't break
  37  *                                      old apps and it will avoid huge amount
  38  *                                      of socks hashed (this for unix_gc()
  39  *                                      performances reasons).
  40  *                                      Security fix that limits the max
  41  *                                      number of socks to 2*max_files and
  42  *                                      the number of skb queueable in the
  43  *                                      dgram receiver.
  44  *              Artur Skawina   :       Hash function optimizations
  45  *           Alexey Kuznetsov   :       Full scale SMP. Lot of bugs are introduced 8)
  46  *            Malcolm Beattie   :       Set peercred for socketpair
  47  *           Michal Ostrowski   :       Module initialization cleanup.
  48  *           Arnaldo C. Melo    :       Remove MOD_{INC,DEC}_USE_COUNT,
  49  *                                      the core infrastructure is doing that
  50  *                                      for all net proto families now (2.5.69+)
  51  *
  52  *
  53  * Known differences from reference BSD that was tested:
  54  *
  55  *      [TO FIX]
  56  *      ECONNREFUSED is not returned from one end of a connected() socket to the
  57  *              other the moment one end closes.
  58  *      fstat() doesn't return st_dev=0, and give the blksize as high water mark
  59  *              and a fake inode identifier (nor the BSD first socket fstat twice bug).
  60  *      [NOT TO FIX]
  61  *      accept() returns a path name even if the connecting socket has closed
  62  *              in the meantime (BSD loses the path and gives up).
  63  *      accept() returns 0 length path for an unbound connector. BSD returns 16
  64  *              and a null first byte in the path (but not for gethost/peername - BSD bug ??)
  65  *      socketpair(...SOCK_RAW..) doesn't panic the kernel.
  66  *      BSD af_unix apparently has connect forgetting to block properly.
  67  *              (need to check this with the POSIX spec in detail)
  68  *
  69  * Differences from 2.0.0-11-... (ANK)
  70  *      Bug fixes and improvements.
  71  *              - client shutdown killed server socket.
  72  *              - removed all useless cli/sti pairs.
  73  *
  74  *      Semantic changes/extensions.
  75  *              - generic control message passing.
  76  *              - SCM_CREDENTIALS control message.
  77  *              - "Abstract" (not FS based) socket bindings.
  78  *                Abstract names are sequences of bytes (not zero terminated)
  79  *                started by 0, so that this name space does not intersect
  80  *                with BSD names.
  81  */
  82
  83 #include <linux/module.h>
  84 #include <linux/kernel.h>
  85 #include <linux/signal.h>
  86 #include <linux/sched.h>
  87 #include <linux/errno.h>
  88 #include <linux/string.h>
  89 #include <linux/stat.h>
  90 #include <linux/dcache.h>
  91 #include <linux/namei.h>
  92 #include <linux/socket.h>
  93 #include <linux/un.h>
  94 #include <linux/fcntl.h>
  95 #include <linux/termios.h>
  96 #include <linux/sockios.h>
  97 #include <linux/net.h>
  98 #include <linux/in.h>
  99 #include <linux/fs.h>
 100 #include <linux/slab.h>
 101 #include <asm/uaccess.h>
 102 #include <linux/skbuff.h>
 103 #include <linux/netdevice.h>
 104 #include <net/net_namespace.h>
 105 #include <net/sock.h>
 106 #include <net/tcp_states.h>
 107 #include <net/af_unix.h>
 108 #include <linux/proc_fs.h>
 109 #include <linux/seq_file.h>
 110 #include <net/scm.h>
 111 #include <linux/init.h>
 112 #include <linux/poll.h>
 113 #include <linux/rtnetlink.h>
 114 #include <linux/mount.h>
 115 #include <net/checksum.h>
 116 #include <linux/security.h>
 117
 118 static struct hlist_head unix_socket_table[UNIX_HASH_SIZE + 1];
 119 static DEFINE_SPINLOCK(unix_table_lock);
 120 static atomic_t unix_nr_socks = ATOMIC_INIT(0);
 121
 122 #define unix_sockets_unbound    (&unix_socket_table[UNIX_HASH_SIZE])
 123
 124 #define UNIX_ABSTRACT(sk)       (unix_sk(sk)->addr->hash != UNIX_HASH_SIZE)
 125
 126 #ifdef CONFIG_SECURITY_NETWORK
 127 static void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb)
 128 {
 129         memcpy(UNIXSID(skb), &scm->secid, sizeof(u32));
 130 }
 131
 132 static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb)
 133 {
 134         scm->secid = *UNIXSID(skb);
 135 }
 136 #else
 137 static inline void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb)
 138 { }
 139
 140 static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb)
 141 { }
 142 #endif /* CONFIG_SECURITY_NETWORK */
 143
 144 /*
 145  *  SMP locking strategy:
 146  *    hash table is protected with spinlock unix_table_lock
 147  *    each socket state is protected by separate rwlock.
 148  */
 149
 150 static inline unsigned unix_hash_fold(__wsum n)
 151 {
 152         unsigned hash = (__force unsigned)n;
 153         hash ^= hash>>16;
 154         hash ^= hash>>8;
 155         return hash&(UNIX_HASH_SIZE-1);
 156 }
 157
 158 #define unix_peer(sk) (unix_sk(sk)->peer)
 159
 160 static inline int unix_our_peer(struct sock *sk, struct sock *osk)
 161 {
 162         return unix_peer(osk) == sk;
 163 }
 164
 165 static inline int unix_may_send(struct sock *sk, struct sock *osk)
 166 {
 167         return unix_peer(osk) == NULL || unix_our_peer(sk, osk);
 168 }
 169
 170 static inline int unix_recvq_full(struct sock const *sk)
 171 {
 172         return skb_queue_len(&sk->sk_receive_queue) > sk->sk_max_ack_backlog;
 173 }
 174
 175 static struct sock *unix_peer_get(struct sock *s)
 176 {
 177         struct sock *peer;
 178
 179         unix_state_lock(s);
 180         peer = unix_peer(s);
 181         if (peer)
 182                 sock_hold(peer);
 183         unix_state_unlock(s);
 184         return peer;
 185 }
 186
 187 static inline void unix_release_addr(struct unix_address *addr)
 188 {
 189         if (atomic_dec_and_test(&addr->refcnt))
 190                 kfree(addr);
 191 }
 192
 193 /*
 194  *      Check unix socket name:
 195  *              - should be not zero length.
 196  *              - if started by not zero, should be NULL terminated (FS object)
 197  *              - if started by zero, it is abstract name.
 198  */
 199
 200 static int unix_mkname(struct sockaddr_un *sunaddr, int len, unsigned *hashp)
 201 {
 202         if (len <= sizeof(short) || len > sizeof(*sunaddr))
 203                 return -EINVAL;
 204         if (!sunaddr || sunaddr->sun_family != AF_UNIX)
 205                 return -EINVAL;
 206         if (sunaddr->sun_path[0]) {
 207                 /*
 208                  * This may look like an off by one error but it is a bit more
 209                  * subtle. 108 is the longest valid AF_UNIX path for a binding.
 210                  * sun_path[108] doesnt as such exist.  However in kernel space
 211                  * we are guaranteed that it is a valid memory location in our
 212                  * kernel address buffer.
 213                  */
 214                 ((char *)sunaddr)[len] = 0;
 215                 len = strlen(sunaddr->sun_path)+1+sizeof(short);
 216                 return len;
 217         }
 218
 219         *hashp = unix_hash_fold(csum_partial(sunaddr, len, 0));
 220         return len;
 221 }
 222
 223 static void __unix_remove_socket(struct sock *sk)
 224 {
 225         sk_del_node_init(sk);
 226 }
 227
 228 static void __unix_insert_socket(struct hlist_head *list, struct sock *sk)
 229 {
 230         WARN_ON(!sk_unhashed(sk));
 231         sk_add_node(sk, list);
 232 }
 233
 234 static inline void unix_remove_socket(struct sock *sk)
 235 {
 236         spin_lock(&unix_table_lock);
 237         __unix_remove_socket(sk);
 238         spin_unlock(&unix_table_lock);
 239 }
 240
 241 static inline void unix_insert_socket(struct hlist_head *list, struct sock *sk)
 242 {
 243         spin_lock(&unix_table_lock);
 244         __unix_insert_socket(list, sk);
 245         spin_unlock(&unix_table_lock);
 246 }
 247
 248 static struct sock *__unix_find_socket_byname(struct net *net,
 249                                               struct sockaddr_un *sunname,
 250                                               int len, int type, unsigned hash)
 251 {
 252         struct sock *s;
 253         struct hlist_node *node;
 254
 255         sk_for_each(s, node, &unix_socket_table[hash ^ type]) {
 256                 struct unix_sock *u = unix_sk(s);
 257
 258                 if (!net_eq(sock_net(s), net))
 259                         continue;
 260
 261                 if (u->addr->len == len &&
 262                     !memcmp(u->addr->name, sunname, len))
 263                         goto found;
 264         }
 265         s = NULL;
 266 found:
 267         return s;
 268 }
 269
 270 static inline struct sock *unix_find_socket_byname(struct net *net,
 271                                                    struct sockaddr_un *sunname,
 272                                                    int len, int type,
 273                                                    unsigned hash)
 274 {
 275         struct sock *s;
 276
 277         spin_lock(&unix_table_lock);
 278         s = __unix_find_socket_byname(net, sunname, len, type, hash);
 279         if (s)
 280                 sock_hold(s);
 281         spin_unlock(&unix_table_lock);
 282         return s;
 283 }
 284
 285 static struct sock *unix_find_socket_byinode(struct net *net, struct inode *i)
 286 {
 287         struct sock *s;
 288         struct hlist_node *node;
 289
 290         spin_lock(&unix_table_lock);
 291         sk_for_each(s, node,
 292                     &unix_socket_table[i->i_ino & (UNIX_HASH_SIZE - 1)]) {
 293                 struct dentry *dentry = unix_sk(s)->dentry;
 294
 295                 if (!net_eq(sock_net(s), net))
 296                         continue;
 297
 298                 if (dentry && dentry->d_inode == i) {
 299                         sock_hold(s);
 300                         goto found;
 301                 }
 302         }
 303         s = NULL;
 304 found:
 305         spin_unlock(&unix_table_lock);
 306         return s;
 307 }
 308
 309 static inline int unix_writable(struct sock *sk)
 310 {
 311         return (atomic_read(&sk->sk_wmem_alloc) << 2) <= sk->sk_sndbuf;
 312 }
 313
 314 static void unix_write_space(struct sock *sk)
 315 {
 316         read_lock(&sk->sk_callback_lock);
 317         if (unix_writable(sk)) {
 318                 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
 319                         wake_up_interruptible_sync(sk->sk_sleep);
 320                 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
 321         }
 322         read_unlock(&sk->sk_callback_lock);
 323 }
 324
 325 /* When dgram socket disconnects (or changes its peer), we clear its receive
 326  * queue of packets arrived from previous peer. First, it allows to do
 327  * flow control based only on wmem_alloc; second, sk connected to peer
 328  * may receive messages only from that peer. */
 329 static void unix_dgram_disconnected(struct sock *sk, struct sock *other)
 330 {
 331         if (!skb_queue_empty(&sk->sk_receive_queue)) {
 332                 skb_queue_purge(&sk->sk_receive_queue);
 333                 wake_up_interruptible_all(&unix_sk(sk)->peer_wait);
 334
 335                 /* If one link of bidirectional dgram pipe is disconnected,
 336                  * we signal error. Messages are lost. Do not make this,
 337                  * when peer was not connected to us.
 338                  */
 339                 if (!sock_flag(other, SOCK_DEAD) && unix_peer(other) == sk) {
 340                         other->sk_err = ECONNRESET;
 341                         other->sk_error_report(other);
 342                 }
 343         }
 344 }
 345
 346 static void unix_sock_destructor(struct sock *sk)
 347 {
 348         struct unix_sock *u = unix_sk(sk);
 349
 350         skb_queue_purge(&sk->sk_receive_queue);
 351
 352         WARN_ON(atomic_read(&sk->sk_wmem_alloc));
 353         WARN_ON(!sk_unhashed(sk));
 354         WARN_ON(sk->sk_socket);
 355         if (!sock_flag(sk, SOCK_DEAD)) {
 356                 printk(KERN_INFO "Attempt to release alive unix socket: %p\n", sk);
 357                 return;
 358         }
 359
 360         if (u->addr)
 361                 unix_release_addr(u->addr);
 362
 363         atomic_dec(&unix_nr_socks);
 364         sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
 365 #ifdef UNIX_REFCNT_DEBUG
 366         printk(KERN_DEBUG "UNIX %p is destroyed, %d are still alive.\n", sk,
 367                 atomic_read(&unix_nr_socks));
 368 #endif
 369 }
 370
 371 static int unix_release_sock(struct sock *sk, int embrion)
 372 {
 373         struct unix_sock *u = unix_sk(sk);
 374         struct dentry *dentry;
 375         struct vfsmount *mnt;
 376         struct sock *skpair;
 377         struct sk_buff *skb;
 378         int state;
 379
 380         unix_remove_socket(sk);
 381
 382         /* Clear state */
 383         unix_state_lock(sk);
 384         sock_orphan(sk);
 385         sk->sk_shutdown = SHUTDOWN_MASK;
 386         dentry       = u->dentry;
 387         u->dentry    = NULL;
 388         mnt          = u->mnt;
 389         u->mnt       = NULL;
 390         state = sk->sk_state;
 391         sk->sk_state = TCP_CLOSE;
 392         unix_state_unlock(sk);
 393
 394         wake_up_interruptible_all(&u->peer_wait);
 395
 396         skpair = unix_peer(sk);
 397
 398         if (skpair != NULL) {
 399                 if (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) {
 400                         unix_state_lock(skpair);
 401                         /* No more writes */
 402                         skpair->sk_shutdown = SHUTDOWN_MASK;
 403                         if (!skb_queue_empty(&sk->sk_receive_queue) || embrion)
 404                                 skpair->sk_err = ECONNRESET;
 405                         unix_state_unlock(skpair);
 406                         skpair->sk_state_change(skpair);
 407                         read_lock(&skpair->sk_callback_lock);
 408                         sk_wake_async(skpair, SOCK_WAKE_WAITD, POLL_HUP);
 409                         read_unlock(&skpair->sk_callback_lock);
 410                 }
 411                 sock_put(skpair); /* It may now die */
 412                 unix_peer(sk) = NULL;
 413         }
 414
 415         /* Try to flush out this socket. Throw out buffers at least */
 416
 417         while ((skb = skb_dequeue(&sk->sk_receive_queue)) != NULL) {
 418                 if (state == TCP_LISTEN)
 419                         unix_release_sock(skb->sk, 1);
 420                 /* passed fds are erased in the kfree_skb hook        */
 421                 kfree_skb(skb);
 422         }
 423
 424         if (dentry) {
 425                 dput(dentry);
 426                 mntput(mnt);
 427         }
 428
 429         sock_put(sk);
 430
 431         /* ---- Socket is dead now and most probably destroyed ---- */
 432
 433         /*
 434          * Fixme: BSD difference: In BSD all sockets connected to use get
 435          *        ECONNRESET and we die on the spot. In Linux we behave
 436          *        like files and pipes do and wait for the last
 437          *        dereference.
 438          *
 439          * Can't we simply set sock->err?
 440          *
 441          *        What the above comment does talk about? --ANK(980817)
 442          */
 443
 444         if (unix_tot_inflight)
 445                 unix_gc();              /* Garbage collect fds */
 446
 447         return 0;
 448 }
 449
 450 static int unix_listen(struct socket *sock, int backlog)
 451 {
 452         int err;
 453         struct sock *sk = sock->sk;
 454         struct unix_sock *u = unix_sk(sk);
 455
 456         err = -EOPNOTSUPP;
 457         if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
 458                 goto out;       /* Only stream/seqpacket sockets accept */
 459         err = -EINVAL;
 460         if (!u->addr)
 461                 goto out;       /* No listens on an unbound socket */
 462         unix_state_lock(sk);
 463         if (sk->sk_state != TCP_CLOSE && sk->sk_state != TCP_LISTEN)
 464                 goto out_unlock;
 465         if (backlog > sk->sk_max_ack_backlog)
 466                 wake_up_interruptible_all(&u->peer_wait);
 467         sk->sk_max_ack_backlog  = backlog;
 468         sk->sk_state            = TCP_LISTEN;
 469         /* set credentials so connect can copy them */
 470         sk->sk_peercred.pid     = task_tgid_vnr(current);
 471         sk->sk_peercred.uid     = current->euid;
 472         sk->sk_peercred.gid     = current->egid;
 473         err = 0;
 474
 475 out_unlock:
 476         unix_state_unlock(sk);
 477 out:
 478         return err;
 479 }
 480
 481 static int unix_release(struct socket *);
 482 static int unix_bind(struct socket *, struct sockaddr *, int);
 483 static int unix_stream_connect(struct socket *, struct sockaddr *,
 484                                int addr_len, int flags);
 485 static int unix_socketpair(struct socket *, struct socket *);
 486 static int unix_accept(struct socket *, struct socket *, int);
 487 static int unix_getname(struct socket *, struct sockaddr *, int *, int);
 488 static unsigned int unix_poll(struct file *, struct socket *, poll_table *);
 489 static unsigned int unix_dgram_poll(struct file *, struct socket *,
 490                                     poll_table *);
 491 static int unix_ioctl(struct socket *, unsigned int, unsigned long);
 492 static int unix_shutdown(struct socket *, int);
 493 static int unix_stream_sendmsg(struct kiocb *, struct socket *,
 494                                struct msghdr *, size_t);
 495 static int unix_stream_recvmsg(struct kiocb *, struct socket *,
 496                                struct msghdr *, size_t, int);
 497 static int unix_dgram_sendmsg(struct kiocb *, struct socket *,
 498                               struct msghdr *, size_t);
 499 static int unix_dgram_recvmsg(struct kiocb *, struct socket *,
 500                               struct msghdr *, size_t, int);
 501 static int unix_dgram_connect(struct socket *, struct sockaddr *,
 502                               int, int);
 503 static int unix_seqpacket_sendmsg(struct kiocb *, struct socket *,
 504                                   struct msghdr *, size_t);
 505
 506 static const struct proto_ops unix_stream_ops = {
 507         .family =       PF_UNIX,
 508         .owner =        THIS_MODULE,
 509         .release =      unix_release,
 510         .bind =         unix_bind,
 511         .connect =      unix_stream_connect,
 512         .socketpair =   unix_socketpair,
 513         .accept =       unix_accept,
 514         .getname =      unix_getname,
 515         .poll =         unix_poll,
 516         .ioctl =        unix_ioctl,
 517         .listen =       unix_listen,
 518         .shutdown =     unix_shutdown,
 519         .setsockopt =   sock_no_setsockopt,
 520         .getsockopt =   sock_no_getsockopt,
 521         .sendmsg =      unix_stream_sendmsg,
 522         .recvmsg =      unix_stream_recvmsg,
 523         .mmap =         sock_no_mmap,
 524         .sendpage =     sock_no_sendpage,
 525 };
 526
 527 static const struct proto_ops unix_dgram_ops = {
 528         .family =       PF_UNIX,
 529         .owner =        THIS_MODULE,
 530         .release =      unix_release,
 531         .bind =         unix_bind,
 532         .connect =      unix_dgram_connect,
 533         .socketpair =   unix_socketpair,
 534         .accept =       sock_no_accept,
 535         .getname =      unix_getname,
 536         .poll =         unix_dgram_poll,
 537         .ioctl =        unix_ioctl,
 538         .listen =       sock_no_listen,
 539         .shutdown =     unix_shutdown,
 540         .setsockopt =   sock_no_setsockopt,
 541         .getsockopt =   sock_no_getsockopt,
 542         .sendmsg =      unix_dgram_sendmsg,
 543         .recvmsg =      unix_dgram_recvmsg,
 544         .mmap =         sock_no_mmap,
 545         .sendpage =     sock_no_sendpage,
 546 };
 547
 548 static const struct proto_ops unix_seqpacket_ops = {
 549         .family =       PF_UNIX,
 550         .owner =        THIS_MODULE,
 551         .release =      unix_release,
 552         .bind =         unix_bind,
 553         .connect =      unix_stream_connect,
 554         .socketpair =   unix_socketpair,
 555         .accept =       unix_accept,
 556         .getname =      unix_getname,
 557         .poll =         unix_dgram_poll,
 558         .ioctl =        unix_ioctl,
 559         .listen =       unix_listen,
 560         .shutdown =     unix_shutdown,
 561         .setsockopt =   sock_no_setsockopt,
 562         .getsockopt =   sock_no_getsockopt,
 563         .sendmsg =      unix_seqpacket_sendmsg,
 564         .recvmsg =      unix_dgram_recvmsg,
 565         .mmap =         sock_no_mmap,
 566         .sendpage =     sock_no_sendpage,
 567 };
 568
 569 static struct proto unix_proto = {
 570         .name                   = "UNIX",
 571         .owner                  = THIS_MODULE,
 572         .sockets_allocated      = &unix_nr_socks,
 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                 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
 618
 619         return sk;
 620 }
 621
 622 static int unix_create(struct net *net, struct socket *sock, int protocol)
 623 {
 624         if (protocol && protocol != PF_UNIX)
 625                 return -EPROTONOSUPPORT;
 626
 627         sock->state = SS_UNCONNECTED;
 628
 629         switch (sock->type) {
 630         case SOCK_STREAM:
 631                 sock->ops = &unix_stream_ops;
 632                 break;
 633                 /*
 634                  *      Believe it or not BSD has AF_UNIX, SOCK_RAW though
 635                  *      nothing uses it.
 636                  */
 637         case SOCK_RAW:
 638                 sock->type = SOCK_DGRAM;
 639         case SOCK_DGRAM:
 640                 sock->ops = &unix_dgram_ops;
 641                 break;
 642         case SOCK_SEQPACKET:
 643                 sock->ops = &unix_seqpacket_ops;
 644                 break;
 645         default:
 646                 return -ESOCKTNOSUPPORT;
 647         }
 648
 649         return unix_create1(net, sock) ? 0 : -ENOMEM;
 650 }
 651
 652 static int unix_release(struct socket *sock)
 653 {
 654         struct sock *sk = sock->sk;
 655
 656         if (!sk)
 657                 return 0;
 658
 659         sock->sk = NULL;
 660
 661         return unix_release_sock(sk, 0);
 662 }
 663
 664 static int unix_autobind(struct socket *sock)
 665 {
 666         struct sock *sk = sock->sk;
 667         struct net *net = sock_net(sk);
 668         struct unix_sock *u = unix_sk(sk);
 669         static u32 ordernum = 1;
 670         struct unix_address *addr;
 671         int err;
 672
 673         mutex_lock(&u->readlock);
 674
 675         err = 0;
 676         if (u->addr)
 677                 goto out;
 678
 679         err = -ENOMEM;
 680         addr = kzalloc(sizeof(*addr) + sizeof(short) + 16, GFP_KERNEL);
 681         if (!addr)
 682                 goto out;
 683
 684         addr->name->sun_family = AF_UNIX;
 685         atomic_set(&addr->refcnt, 1);
 686
 687 retry:
 688         addr->len = sprintf(addr->name->sun_path+1, "%05x", ordernum) + 1 + sizeof(short);
 689         addr->hash = unix_hash_fold(csum_partial(addr->name, addr->len, 0));
 690
 691         spin_lock(&unix_table_lock);
 692         ordernum = (ordernum+1)&0xFFFFF;
 693
 694         if (__unix_find_socket_byname(net, addr->name, addr->len, sock->type,
 695                                       addr->hash)) {
 696                 spin_unlock(&unix_table_lock);
 697                 /* Sanity yield. It is unusual case, but yet... */
 698                 if (!(ordernum&0xFF))
 699                         yield();
 700                 goto retry;
 701         }
 702         addr->hash ^= sk->sk_type;
 703
 704         __unix_remove_socket(sk);
 705         u->addr = addr;
 706         __unix_insert_socket(&unix_socket_table[addr->hash], sk);
 707         spin_unlock(&unix_table_lock);
 708         err = 0;
 709
 710 out:    mutex_unlock(&u->readlock);
 711         return err;
 712 }
 713
 714 static struct sock *unix_find_other(struct net *net,
 715                                     struct sockaddr_un *sunname, int len,
 716                                     int type, unsigned hash, int *error)
 717 {
 718         struct sock *u;
 719         struct path path;
 720         int err = 0;
 721
 722         if (sunname->sun_path[0]) {
 723                 struct inode *inode;
 724                 err = kern_path(sunname->sun_path, LOOKUP_FOLLOW, &path);
 725                 if (err)
 726                         goto fail;
 727                 inode = path.dentry->d_inode;
 728                 err = inode_permission(inode, MAY_WRITE);
 729                 if (err)
 730                         goto put_fail;
 731
 732                 err = -ECONNREFUSED;
 733                 if (!S_ISSOCK(inode->i_mode))
 734                         goto put_fail;
 735                 u = unix_find_socket_byinode(net, inode);
 736                 if (!u)
 737                         goto put_fail;
 738
 739                 if (u->sk_type == type)
 740                         touch_atime(path.mnt, path.dentry);
 741
 742                 path_put(&path);
 743
 744                 err = -EPROTOTYPE;
 745                 if (u->sk_type != type) {
 746                         sock_put(u);
 747                         goto fail;
 748                 }
 749         } else {
 750                 err = -ECONNREFUSED;
 751                 u = unix_find_socket_byname(net, sunname, len, type, hash);
 752                 if (u) {
 753                         struct dentry *dentry;
 754                         dentry = unix_sk(u)->dentry;
 755                         if (dentry)
 756                                 touch_atime(unix_sk(u)->mnt, dentry);
 757                 } else
 758                         goto fail;
 759         }
 760         return u;
 761
 762 put_fail:
 763         path_put(&path);
 764 fail:
 765         *error = err;
 766         return NULL;
 767 }
 768
 769
 770 static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
 771 {
 772         struct sock *sk = sock->sk;
 773         struct net *net = sock_net(sk);
 774         struct unix_sock *u = unix_sk(sk);
 775         struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
 776         struct dentry *dentry = NULL;
 777         struct nameidata nd;
 778         int err;
 779         unsigned hash;
 780         struct unix_address *addr;
 781         struct hlist_head *list;
 782
 783         err = -EINVAL;
 784         if (sunaddr->sun_family != AF_UNIX)
 785                 goto out;
 786
 787         if (addr_len == sizeof(short)) {
 788                 err = unix_autobind(sock);
 789                 goto out;
 790         }
 791
 792         err = unix_mkname(sunaddr, addr_len, &hash);
 793         if (err < 0)
 794                 goto out;
 795         addr_len = err;
 796
 797         mutex_lock(&u->readlock);
 798
 799         err = -EINVAL;
 800         if (u->addr)
 801                 goto out_up;
 802
 803         err = -ENOMEM;
 804         addr = kmalloc(sizeof(*addr)+addr_len, GFP_KERNEL);
 805         if (!addr)
 806                 goto out_up;
 807
 808         memcpy(addr->name, sunaddr, addr_len);
 809         addr->len = addr_len;
 810         addr->hash = hash ^ sk->sk_type;
 811         atomic_set(&addr->refcnt, 1);
 812
 813         if (sunaddr->sun_path[0]) {
 814                 unsigned int mode;
 815                 err = 0;
 816                 /*
 817                  * Get the parent directory, calculate the hash for last
 818                  * component.
 819                  */
 820                 err = path_lookup(sunaddr->sun_path, LOOKUP_PARENT, &nd);
 821                 if (err)
 822                         goto out_mknod_parent;
 823
 824                 dentry = lookup_create(&nd, 0);
 825                 err = PTR_ERR(dentry);
 826                 if (IS_ERR(dentry))
 827                         goto out_mknod_unlock;
 828
 829                 /*
 830                  * All right, let's create it.
 831                  */
 832                 mode = S_IFSOCK |
 833                        (SOCK_INODE(sock)->i_mode & ~current->fs->umask);
 834                 err = mnt_want_write(nd.path.mnt);
 835                 if (err)
 836                         goto out_mknod_dput;
 837                 err = vfs_mknod(nd.path.dentry->d_inode, dentry, mode, 0);
 838                 mnt_drop_write(nd.path.mnt);
 839                 if (err)
 840                         goto out_mknod_dput;
 841                 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
 842                 dput(nd.path.dentry);
 843                 nd.path.dentry = dentry;
 844
 845                 addr->hash = UNIX_HASH_SIZE;
 846         }
 847
 848         spin_lock(&unix_table_lock);
 849
 850         if (!sunaddr->sun_path[0]) {
 851                 err = -EADDRINUSE;
 852                 if (__unix_find_socket_byname(net, sunaddr, addr_len,
 853                                               sk->sk_type, hash)) {
 854                         unix_release_addr(addr);
 855                         goto out_unlock;
 856                 }
 857
 858                 list = &unix_socket_table[addr->hash];
 859         } else {
 860                 list = &unix_socket_table[dentry->d_inode->i_ino & (UNIX_HASH_SIZE-1)];
 861                 u->dentry = nd.path.dentry;
 862                 u->mnt    = nd.path.mnt;
 863         }
 864
 865         err = 0;
 866         __unix_remove_socket(sk);
 867         u->addr = addr;
 868         __unix_insert_socket(list, sk);
 869
 870 out_unlock:
 871         spin_unlock(&unix_table_lock);
 872 out_up:
 873         mutex_unlock(&u->readlock);
 874 out:
 875         return err;
 876
 877 out_mknod_dput:
 878         dput(dentry);
 879 out_mknod_unlock:
 880         mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
 881         path_put(&nd.path);
 882 out_mknod_parent:
 883         if (err == -EEXIST)
 884                 err = -EADDRINUSE;
 885         unix_release_addr(addr);
 886         goto out_up;
 887 }
 888
 889 static void unix_state_double_lock(struct sock *sk1, struct sock *sk2)
 890 {
 891         if (unlikely(sk1 == sk2) || !sk2) {
 892                 unix_state_lock(sk1);
 893                 return;
 894         }
 895         if (sk1 < sk2) {
 896                 unix_state_lock(sk1);
 897                 unix_state_lock_nested(sk2);
 898         } else {
 899                 unix_state_lock(sk2);
 900                 unix_state_lock_nested(sk1);
 901         }
 902 }
 903
 904 static void unix_state_double_unlock(struct sock *sk1, struct sock *sk2)
 905 {
 906         if (unlikely(sk1 == sk2) || !sk2) {
 907                 unix_state_unlock(sk1);
 908                 return;
 909         }
 910         unix_state_unlock(sk1);
 911         unix_state_unlock(sk2);
 912 }
 913
 914 static int unix_dgram_connect(struct socket *sock, struct sockaddr *addr,
 915                               int alen, int flags)
 916 {
 917         struct sock *sk = sock->sk;
 918         struct net *net = sock_net(sk);
 919         struct sockaddr_un *sunaddr = (struct sockaddr_un *)addr;
 920         struct sock *other;
 921         unsigned hash;
 922         int err;
 923
 924         if (addr->sa_family != AF_UNSPEC) {
 925                 err = unix_mkname(sunaddr, alen, &hash);
 926                 if (err < 0)
 927                         goto out;
 928                 alen = err;
 929
 930                 if (test_bit(SOCK_PASSCRED, &sock->flags) &&
 931                     !unix_sk(sk)->addr && (err = unix_autobind(sock)) != 0)
 932                         goto out;
 933
 934 restart:
 935                 other = unix_find_other(net, sunaddr, alen, sock->type, hash, &err);
 936                 if (!other)
 937                         goto out;
 938
 939                 unix_state_double_lock(sk, other);
 940
 941                 /* Apparently VFS overslept socket death. Retry. */
 942                 if (sock_flag(other, SOCK_DEAD)) {
 943                         unix_state_double_unlock(sk, other);
 944                         sock_put(other);
 945                         goto restart;
 946                 }
 947
 948                 err = -EPERM;
 949                 if (!unix_may_send(sk, other))
 950                         goto out_unlock;
 951
 952                 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
 953                 if (err)
 954                         goto out_unlock;
 955
 956         } else {
 957                 /*
 958                  *      1003.1g breaking connected state with AF_UNSPEC
 959                  */
 960                 other = NULL;
 961                 unix_state_double_lock(sk, other);
 962         }
 963
 964         /*
 965          * If it was connected, reconnect.
 966          */
 967         if (unix_peer(sk)) {
 968                 struct sock *old_peer = unix_peer(sk);
 969                 unix_peer(sk) = other;
 970                 unix_state_double_unlock(sk, other);
 971
 972                 if (other != old_peer)
 973                         unix_dgram_disconnected(sk, old_peer);
 974                 sock_put(old_peer);
 975         } else {
 976                 unix_peer(sk) = other;
 977                 unix_state_double_unlock(sk, other);
 978         }
 979         return 0;
 980
 981 out_unlock:
 982         unix_state_double_unlock(sk, other);
 983         sock_put(other);
 984 out:
 985         return err;
 986 }
 987
 988 static long unix_wait_for_peer(struct sock *other, long timeo)
 989 {
 990         struct unix_sock *u = unix_sk(other);
 991         int sched;
 992         DEFINE_WAIT(wait);
 993
 994         prepare_to_wait_exclusive(&u->peer_wait, &wait, TASK_INTERRUPTIBLE);
 995
 996         sched = !sock_flag(other, SOCK_DEAD) &&
 997                 !(other->sk_shutdown & RCV_SHUTDOWN) &&
 998                 unix_recvq_full(other);
 999
1000         unix_state_unlock(other);
1001
1002         if (sched)
1003                 timeo = schedule_timeout(timeo);
1004
1005         finish_wait(&u->peer_wait, &wait);
1006         return timeo;
1007 }
1008
1009 static int unix_stream_connect(struct socket *sock, struct sockaddr *uaddr,
1010                                int addr_len, int flags)
1011 {
1012         struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
1013         struct sock *sk = sock->sk;
1014         struct net *net = sock_net(sk);
1015         struct unix_sock *u = unix_sk(sk), *newu, *otheru;
1016         struct sock *newsk = NULL;
1017         struct sock *other = NULL;
1018         struct sk_buff *skb = NULL;
1019         unsigned hash;
1020         int st;
1021         int err;
1022         long timeo;
1023
1024         err = unix_mkname(sunaddr, addr_len, &hash);
1025         if (err < 0)
1026                 goto out;
1027         addr_len = err;
1028
1029         if (test_bit(SOCK_PASSCRED, &sock->flags)
1030                 && !u->addr && (err = unix_autobind(sock)) != 0)
1031                 goto out;
1032
1033         timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
1034
1035         /* First of all allocate resources.
1036            If we will make it after state is locked,
1037            we will have to recheck all again in any case.
1038          */
1039
1040         err = -ENOMEM;
1041
1042         /* create new sock for complete connection */
1043         newsk = unix_create1(sock_net(sk), NULL);
1044         if (newsk == NULL)
1045                 goto out;
1046
1047         /* Allocate skb for sending to listening sock */
1048         skb = sock_wmalloc(newsk, 1, 0, GFP_KERNEL);
1049         if (skb == NULL)
1050                 goto out;
1051
1052 restart:
1053         /*  Find listening sock. */
1054         other = unix_find_other(net, sunaddr, addr_len, sk->sk_type, hash, &err);
1055         if (!other)
1056                 goto out;
1057
1058         /* Latch state of peer */
1059         unix_state_lock(other);
1060
1061         /* Apparently VFS overslept socket death. Retry. */
1062         if (sock_flag(other, SOCK_DEAD)) {
1063                 unix_state_unlock(other);
1064                 sock_put(other);
1065                 goto restart;
1066         }
1067
1068         err = -ECONNREFUSED;
1069         if (other->sk_state != TCP_LISTEN)
1070                 goto out_unlock;
1071
1072         if (unix_recvq_full(other)) {
1073                 err = -EAGAIN;
1074                 if (!timeo)
1075                         goto out_unlock;
1076
1077                 timeo = unix_wait_for_peer(other, timeo);
1078
1079                 err = sock_intr_errno(timeo);
1080                 if (signal_pending(current))
1081                         goto out;
1082                 sock_put(other);
1083                 goto restart;
1084         }
1085
1086         /* Latch our state.
1087
1088            It is tricky place. We need to grab write lock and cannot
1089            drop lock on peer. It is dangerous because deadlock is
1090            possible. Connect to self case and simultaneous
1091            attempt to connect are eliminated by checking socket
1092            state. other is TCP_LISTEN, if sk is TCP_LISTEN we
1093            check this before attempt to grab lock.
1094
1095            Well, and we have to recheck the state after socket locked.
1096          */
1097         st = sk->sk_state;
1098
1099         switch (st) {
1100         case TCP_CLOSE:
1101                 /* This is ok... continue with connect */
1102                 break;
1103         case TCP_ESTABLISHED:
1104                 /* Socket is already connected */
1105                 err = -EISCONN;
1106                 goto out_unlock;
1107         default:
1108                 err = -EINVAL;
1109                 goto out_unlock;
1110         }
1111
1112         unix_state_lock_nested(sk);
1113
1114         if (sk->sk_state != st) {
1115                 unix_state_unlock(sk);
1116                 unix_state_unlock(other);
1117                 sock_put(other);
1118                 goto restart;
1119         }
1120
1121         err = security_unix_stream_connect(sock, other->sk_socket, newsk);
1122         if (err) {
1123                 unix_state_unlock(sk);
1124                 goto out_unlock;
1125         }
1126
1127         /* The way is open! Fastly set all the necessary fields... */
1128
1129         sock_hold(sk);
1130         unix_peer(newsk)        = sk;
1131         newsk->sk_state         = TCP_ESTABLISHED;
1132         newsk->sk_type          = sk->sk_type;
1133         newsk->sk_peercred.pid  = task_tgid_vnr(current);
1134         newsk->sk_peercred.uid  = current->euid;
1135         newsk->sk_peercred.gid  = current->egid;
1136         newu = unix_sk(newsk);
1137         newsk->sk_sleep         = &newu->peer_wait;
1138         otheru = unix_sk(other);
1139
1140         /* copy address information from listening to new sock*/
1141         if (otheru->addr) {
1142                 atomic_inc(&otheru->addr->refcnt);
1143                 newu->addr = otheru->addr;
1144         }
1145         if (otheru->dentry) {
1146                 newu->dentry    = dget(otheru->dentry);
1147                 newu->mnt       = mntget(otheru->mnt);
1148         }
1149
1150         /* Set credentials */
1151         sk->sk_peercred = other->sk_peercred;
1152
1153         sock->state     = SS_CONNECTED;
1154         sk->sk_state    = TCP_ESTABLISHED;
1155         sock_hold(newsk);
1156
1157         smp_mb__after_atomic_inc();     /* sock_hold() does an atomic_inc() */
1158         unix_peer(sk)   = newsk;
1159
1160         unix_state_unlock(sk);
1161
1162         /* take ten and and send info to listening sock */
1163         spin_lock(&other->sk_receive_queue.lock);
1164         __skb_queue_tail(&other->sk_receive_queue, skb);
1165         spin_unlock(&other->sk_receive_queue.lock);
1166         unix_state_unlock(other);
1167         other->sk_data_ready(other, 0);
1168         sock_put(other);
1169         return 0;
1170
1171 out_unlock:
1172         if (other)
1173                 unix_state_unlock(other);
1174
1175 out:
1176         if (skb)
1177                 kfree_skb(skb);
1178         if (newsk)
1179                 unix_release_sock(newsk, 0);
1180         if (other)
1181                 sock_put(other);
1182         return err;
1183 }
1184
1185 static int unix_socketpair(struct socket *socka, struct socket *sockb)
1186 {
1187         struct sock *ska = socka->sk, *skb = sockb->sk;
1188
1189         /* Join our sockets back to back */
1190         sock_hold(ska);
1191         sock_hold(skb);
1192         unix_peer(ska) = skb;
1193         unix_peer(skb) = ska;
1194         ska->sk_peercred.pid = skb->sk_peercred.pid = task_tgid_vnr(current);
1195         ska->sk_peercred.uid = skb->sk_peercred.uid = current->euid;
1196         ska->sk_peercred.gid = skb->sk_peercred.gid = current->egid;
1197
1198         if (ska->sk_type != SOCK_DGRAM) {
1199                 ska->sk_state = TCP_ESTABLISHED;
1200                 skb->sk_state = TCP_ESTABLISHED;
1201                 socka->state  = SS_CONNECTED;
1202                 sockb->state  = SS_CONNECTED;
1203         }
1204         return 0;
1205 }
1206
1207 static int unix_accept(struct socket *sock, struct socket *newsock, int flags)
1208 {
1209         struct sock *sk = sock->sk;
1210         struct sock *tsk;
1211         struct sk_buff *skb;
1212         int err;
1213
1214         err = -EOPNOTSUPP;
1215         if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
1216                 goto out;
1217
1218         err = -EINVAL;
1219         if (sk->sk_state != TCP_LISTEN)
1220                 goto out;
1221
1222         /* If socket state is TCP_LISTEN it cannot change (for now...),
1223          * so that no locks are necessary.
1224          */
1225
1226         skb = skb_recv_datagram(sk, 0, flags&O_NONBLOCK, &err);
1227         if (!skb) {
1228                 /* This means receive shutdown. */
1229                 if (err == 0)
1230                         err = -EINVAL;
1231                 goto out;
1232         }
1233
1234         tsk = skb->sk;
1235         skb_free_datagram(sk, skb);
1236         wake_up_interruptible(&unix_sk(sk)->peer_wait);
1237
1238         /* attach accepted sock to socket */
1239         unix_state_lock(tsk);
1240         newsock->state = SS_CONNECTED;
1241         sock_graft(tsk, newsock);
1242         unix_state_unlock(tsk);
1243         return 0;
1244
1245 out:
1246         return err;
1247 }
1248
1249
1250 static int unix_getname(struct socket *sock, struct sockaddr *uaddr, int *uaddr_len, int peer)
1251 {
1252         struct sock *sk = sock->sk;
1253         struct unix_sock *u;
1254         struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
1255         int err = 0;
1256
1257         if (peer) {
1258                 sk = unix_peer_get(sk);
1259
1260                 err = -ENOTCONN;
1261                 if (!sk)
1262                         goto out;
1263                 err = 0;
1264         } else {
1265                 sock_hold(sk);
1266         }
1267
1268         u = unix_sk(sk);
1269         unix_state_lock(sk);
1270         if (!u->addr) {
1271                 sunaddr->sun_family = AF_UNIX;
1272                 sunaddr->sun_path[0] = 0;
1273                 *uaddr_len = sizeof(short);
1274         } else {
1275                 struct unix_address *addr = u->addr;
1276
1277                 *uaddr_len = addr->len;
1278                 memcpy(sunaddr, addr->name, *uaddr_len);
1279         }
1280         unix_state_unlock(sk);
1281         sock_put(sk);
1282 out:
1283         return err;
1284 }
1285
1286 static void unix_detach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1287 {
1288         int i;
1289
1290         scm->fp = UNIXCB(skb).fp;
1291         skb->destructor = sock_wfree;
1292         UNIXCB(skb).fp = NULL;
1293
1294         for (i = scm->fp->count-1; i >= 0; i--)
1295                 unix_notinflight(scm->fp->fp[i]);
1296 }
1297
1298 static void unix_destruct_fds(struct sk_buff *skb)
1299 {
1300         struct scm_cookie scm;
1301         memset(&scm, 0, sizeof(scm));
1302         unix_detach_fds(&scm, skb);
1303
1304         /* Alas, it calls VFS */
1305         /* So fscking what? fput() had been SMP-safe since the last Summer */
1306         scm_destroy(&scm);
1307         sock_wfree(skb);
1308 }
1309
1310 static int unix_attach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1311 {
1312         int i;
1313
1314         /*
1315          * Need to duplicate file references for the sake of garbage
1316          * collection.  Otherwise a socket in the fps might become a
1317          * candidate for GC while the skb is not yet queued.
1318          */
1319         UNIXCB(skb).fp = scm_fp_dup(scm->fp);
1320         if (!UNIXCB(skb).fp)
1321                 return -ENOMEM;
1322
1323         for (i = scm->fp->count-1; i >= 0; i--)
1324                 unix_inflight(scm->fp->fp[i]);
1325         skb->destructor = unix_destruct_fds;
1326         return 0;
1327 }
1328
1329 /*
1330  *      Send AF_UNIX data.
1331  */
1332
1333 static int unix_dgram_sendmsg(struct kiocb *kiocb, struct socket *sock,
1334                               struct msghdr *msg, size_t len)
1335 {
1336         struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1337         struct sock *sk = sock->sk;
1338         struct net *net = sock_net(sk);
1339         struct unix_sock *u = unix_sk(sk);
1340         struct sockaddr_un *sunaddr = msg->msg_name;
1341         struct sock *other = NULL;
1342         int namelen = 0; /* fake GCC */
1343         int err;
1344         unsigned hash;
1345         struct sk_buff *skb;
1346         long timeo;
1347         struct scm_cookie tmp_scm;
1348
1349         if (NULL == siocb->scm)
1350                 siocb->scm = &tmp_scm;
1351         err = scm_send(sock, msg, siocb->scm);
1352         if (err < 0)
1353                 return err;
1354
1355         err = -EOPNOTSUPP;
1356         if (msg->msg_flags&MSG_OOB)
1357                 goto out;
1358
1359         if (msg->msg_namelen) {
1360                 err = unix_mkname(sunaddr, msg->msg_namelen, &hash);
1361                 if (err < 0)
1362                         goto out;
1363                 namelen = err;
1364         } else {
1365                 sunaddr = NULL;
1366                 err = -ENOTCONN;
1367                 other = unix_peer_get(sk);
1368                 if (!other)
1369                         goto out;
1370         }
1371
1372         if (test_bit(SOCK_PASSCRED, &sock->flags)
1373                 && !u->addr && (err = unix_autobind(sock)) != 0)
1374                 goto out;
1375
1376         err = -EMSGSIZE;
1377         if (len > sk->sk_sndbuf - 32)
1378                 goto out;
1379
1380         skb = sock_alloc_send_skb(sk, len, msg->msg_flags&MSG_DONTWAIT, &err);
1381         if (skb == NULL)
1382                 goto out;
1383
1384         memcpy(UNIXCREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1385         if (siocb->scm->fp) {
1386                 err = unix_attach_fds(siocb->scm, skb);
1387                 if (err)
1388                         goto out_free;
1389         }
1390         unix_get_secdata(siocb->scm, skb);
1391
1392         skb_reset_transport_header(skb);
1393         err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
1394         if (err)
1395                 goto out_free;
1396
1397         timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1398
1399 restart:
1400         if (!other) {
1401                 err = -ECONNRESET;
1402                 if (sunaddr == NULL)
1403                         goto out_free;
1404
1405                 other = unix_find_other(net, sunaddr, namelen, sk->sk_type,
1406                                         hash, &err);
1407                 if (other == NULL)
1408                         goto out_free;
1409         }
1410
1411         unix_state_lock(other);
1412         err = -EPERM;
1413         if (!unix_may_send(sk, other))
1414                 goto out_unlock;
1415
1416         if (sock_flag(other, SOCK_DEAD)) {
1417                 /*
1418                  *      Check with 1003.1g - what should
1419                  *      datagram error
1420                  */
1421                 unix_state_unlock(other);
1422                 sock_put(other);
1423
1424                 err = 0;
1425                 unix_state_lock(sk);
1426                 if (unix_peer(sk) == other) {
1427                         unix_peer(sk) = NULL;
1428                         unix_state_unlock(sk);
1429
1430                         unix_dgram_disconnected(sk, other);
1431                         sock_put(other);
1432                         err = -ECONNREFUSED;
1433                 } else {
1434                         unix_state_unlock(sk);
1435                 }
1436
1437                 other = NULL;
1438                 if (err)
1439                         goto out_free;
1440                 goto restart;
1441         }
1442
1443         err = -EPIPE;
1444         if (other->sk_shutdown & RCV_SHUTDOWN)
1445                 goto out_unlock;
1446
1447         if (sk->sk_type != SOCK_SEQPACKET) {
1448                 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1449                 if (err)
1450                         goto out_unlock;
1451         }
1452
1453         if (unix_peer(other) != sk && unix_recvq_full(other)) {
1454                 if (!timeo) {
1455                         err = -EAGAIN;
1456                         goto out_unlock;
1457                 }
1458
1459                 timeo = unix_wait_for_peer(other, timeo);
1460
1461                 err = sock_intr_errno(timeo);
1462                 if (signal_pending(current))
1463                         goto out_free;
1464
1465                 goto restart;
1466         }
1467
1468         skb_queue_tail(&other->sk_receive_queue, skb);
1469         unix_state_unlock(other);
1470         other->sk_data_ready(other, len);
1471         sock_put(other);
1472         scm_destroy(siocb->scm);
1473         return len;
1474
1475 out_unlock:
1476         unix_state_unlock(other);
1477 out_free:
1478         kfree_skb(skb);
1479 out:
1480         if (other)
1481                 sock_put(other);
1482         scm_destroy(siocb->scm);
1483         return err;
1484 }
1485
1486
1487 static int unix_stream_sendmsg(struct kiocb *kiocb, struct socket *sock,
1488                                struct msghdr *msg, size_t len)
1489 {
1490         struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1491         struct sock *sk = sock->sk;
1492         struct sock *other = NULL;
1493         struct sockaddr_un *sunaddr = msg->msg_name;
1494         int err, size;
1495         struct sk_buff *skb;
1496         int sent = 0;
1497         struct scm_cookie tmp_scm;
1498
1499         if (NULL == siocb->scm)
1500                 siocb->scm = &tmp_scm;
1501         err = scm_send(sock, msg, siocb->scm);
1502         if (err < 0)
1503                 return err;
1504
1505         err = -EOPNOTSUPP;
1506         if (msg->msg_flags&MSG_OOB)
1507                 goto out_err;
1508
1509         if (msg->msg_namelen) {
1510                 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
1511                 goto out_err;
1512         } else {
1513                 sunaddr = NULL;
1514                 err = -ENOTCONN;
1515                 other = unix_peer(sk);
1516                 if (!other)
1517                         goto out_err;
1518         }
1519
1520         if (sk->sk_shutdown & SEND_SHUTDOWN)
1521                 goto pipe_err;
1522
1523         while (sent < len) {
1524                 /*
1525                  *      Optimisation for the fact that under 0.01% of X
1526                  *      messages typically need breaking up.
1527                  */
1528
1529                 size = len-sent;
1530
1531                 /* Keep two messages in the pipe so it schedules better */
1532                 if (size > ((sk->sk_sndbuf >> 1) - 64))
1533                         size = (sk->sk_sndbuf >> 1) - 64;
1534
1535                 if (size > SKB_MAX_ALLOC)
1536                         size = SKB_MAX_ALLOC;
1537
1538                 /*
1539                  *      Grab a buffer
1540                  */
1541
1542                 skb = sock_alloc_send_skb(sk, size, msg->msg_flags&MSG_DONTWAIT,
1543                                           &err);
1544
1545                 if (skb == NULL)
1546                         goto out_err;
1547
1548                 /*
1549                  *      If you pass two values to the sock_alloc_send_skb
1550                  *      it tries to grab the large buffer with GFP_NOFS
1551                  *      (which can fail easily), and if it fails grab the
1552                  *      fallback size buffer which is under a page and will
1553                  *      succeed. [Alan]
1554                  */
1555                 size = min_t(int, size, skb_tailroom(skb));
1556
1557                 memcpy(UNIXCREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1558                 if (siocb->scm->fp) {
1559                         err = unix_attach_fds(siocb->scm, skb);
1560                         if (err) {
1561                                 kfree_skb(skb);
1562                                 goto out_err;
1563                         }
1564                 }
1565
1566                 err = memcpy_fromiovec(skb_put(skb, size), msg->msg_iov, size);
1567                 if (err) {
1568                         kfree_skb(skb);
1569                         goto out_err;
1570                 }
1571
1572                 unix_state_lock(other);
1573
1574                 if (sock_flag(other, SOCK_DEAD) ||
1575                     (other->sk_shutdown & RCV_SHUTDOWN))
1576                         goto pipe_err_free;
1577
1578                 skb_queue_tail(&other->sk_receive_queue, skb);
1579                 unix_state_unlock(other);
1580                 other->sk_data_ready(other, size);
1581                 sent += size;
1582         }
1583
1584         scm_destroy(siocb->scm);
1585         siocb->scm = NULL;
1586
1587         return sent;
1588
1589 pipe_err_free:
1590         unix_state_unlock(other);
1591         kfree_skb(skb);
1592 pipe_err:
1593         if (sent == 0 && !(msg->msg_flags&MSG_NOSIGNAL))
1594                 send_sig(SIGPIPE, current, 0);
1595         err = -EPIPE;
1596 out_err:
1597         scm_destroy(siocb->scm);
1598         siocb->scm = NULL;
1599         return sent ? : err;
1600 }
1601
1602 static int unix_seqpacket_sendmsg(struct kiocb *kiocb, struct socket *sock,
1603                                   struct msghdr *msg, size_t len)
1604 {
1605         int err;
1606         struct sock *sk = sock->sk;
1607
1608         err = sock_error(sk);
1609         if (err)
1610                 return err;
1611
1612         if (sk->sk_state != TCP_ESTABLISHED)
1613                 return -ENOTCONN;
1614
1615         if (msg->msg_namelen)
1616                 msg->msg_namelen = 0;
1617
1618         return unix_dgram_sendmsg(kiocb, sock, msg, len);
1619 }
1620
1621 static void unix_copy_addr(struct msghdr *msg, struct sock *sk)
1622 {
1623         struct unix_sock *u = unix_sk(sk);
1624
1625         msg->msg_namelen = 0;
1626         if (u->addr) {
1627                 msg->msg_namelen = u->addr->len;
1628                 memcpy(msg->msg_name, u->addr->name, u->addr->len);
1629         }
1630 }
1631
1632 static int unix_dgram_recvmsg(struct kiocb *iocb, struct socket *sock,
1633                               struct msghdr *msg, size_t size,
1634                               int flags)
1635 {
1636         struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1637         struct scm_cookie tmp_scm;
1638         struct sock *sk = sock->sk;
1639         struct unix_sock *u = unix_sk(sk);
1640         int noblock = flags & MSG_DONTWAIT;
1641         struct sk_buff *skb;
1642         int err;
1643
1644         err = -EOPNOTSUPP;
1645         if (flags&MSG_OOB)
1646                 goto out;
1647
1648         msg->msg_namelen = 0;
1649
1650         mutex_lock(&u->readlock);
1651
1652         skb = skb_recv_datagram(sk, flags, noblock, &err);
1653         if (!skb) {
1654                 unix_state_lock(sk);
1655                 /* Signal EOF on disconnected non-blocking SEQPACKET socket. */
1656                 if (sk->sk_type == SOCK_SEQPACKET && err == -EAGAIN &&
1657                     (sk->sk_shutdown & RCV_SHUTDOWN))
1658                         err = 0;
1659                 unix_state_unlock(sk);
1660                 goto out_unlock;
1661         }
1662
1663         wake_up_interruptible_sync(&u->peer_wait);
1664
1665         if (msg->msg_name)
1666                 unix_copy_addr(msg, skb->sk);
1667
1668         if (size > skb->len)
1669                 size = skb->len;
1670         else if (size < skb->len)
1671                 msg->msg_flags |= MSG_TRUNC;
1672
1673         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, size);
1674         if (err)
1675                 goto out_free;
1676
1677         if (!siocb->scm) {
1678                 siocb->scm = &tmp_scm;
1679                 memset(&tmp_scm, 0, sizeof(tmp_scm));
1680         }
1681         siocb->scm->creds = *UNIXCREDS(skb);
1682         unix_set_secdata(siocb->scm, skb);
1683
1684         if (!(flags & MSG_PEEK)) {
1685                 if (UNIXCB(skb).fp)
1686                         unix_detach_fds(siocb->scm, skb);
1687         } else {
1688                 /* It is questionable: on PEEK we could:
1689                    - do not return fds - good, but too simple 8)
1690                    - return fds, and do not return them on read (old strategy,
1691                      apparently wrong)
1692                    - clone fds (I chose it for now, it is the most universal
1693                      solution)
1694
1695                    POSIX 1003.1g does not actually define this clearly
1696                    at all. POSIX 1003.1g doesn't define a lot of things
1697                    clearly however!
1698
1699                 */
1700                 if (UNIXCB(skb).fp)
1701                         siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1702         }
1703         err = size;
1704
1705         scm_recv(sock, msg, siocb->scm, flags);
1706
1707 out_free:
1708         skb_free_datagram(sk, skb);
1709 out_unlock:
1710         mutex_unlock(&u->readlock);
1711 out:
1712         return err;
1713 }
1714
1715 /*
1716  *      Sleep until data has arrive. But check for races..
1717  */
1718
1719 static long unix_stream_data_wait(struct sock *sk, long timeo)
1720 {
1721         DEFINE_WAIT(wait);
1722
1723         unix_state_lock(sk);
1724
1725         for (;;) {
1726                 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1727
1728                 if (!skb_queue_empty(&sk->sk_receive_queue) ||
1729                     sk->sk_err ||
1730                     (sk->sk_shutdown & RCV_SHUTDOWN) ||
1731                     signal_pending(current) ||
1732                     !timeo)
1733                         break;
1734
1735                 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1736                 unix_state_unlock(sk);
1737                 timeo = schedule_timeout(timeo);
1738                 unix_state_lock(sk);
1739                 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1740         }
1741
1742         finish_wait(sk->sk_sleep, &wait);
1743         unix_state_unlock(sk);
1744         return timeo;
1745 }
1746
1747
1748
1749 static int unix_stream_recvmsg(struct kiocb *iocb, struct socket *sock,
1750                                struct msghdr *msg, size_t size,
1751                                int flags)
1752 {
1753         struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1754         struct scm_cookie tmp_scm;
1755         struct sock *sk = sock->sk;
1756         struct unix_sock *u = unix_sk(sk);
1757         struct sockaddr_un *sunaddr = msg->msg_name;
1758         int copied = 0;
1759         int check_creds = 0;
1760         int target;
1761         int err = 0;
1762         long timeo;
1763
1764         err = -EINVAL;
1765         if (sk->sk_state != TCP_ESTABLISHED)
1766                 goto out;
1767
1768         err = -EOPNOTSUPP;
1769         if (flags&MSG_OOB)
1770                 goto out;
1771
1772         target = sock_rcvlowat(sk, flags&MSG_WAITALL, size);
1773         timeo = sock_rcvtimeo(sk, flags&MSG_DONTWAIT);
1774
1775         msg->msg_namelen = 0;
1776
1777         /* Lock the socket to prevent queue disordering
1778          * while sleeps in memcpy_tomsg
1779          */
1780
1781         if (!siocb->scm) {
1782                 siocb->scm = &tmp_scm;
1783                 memset(&tmp_scm, 0, sizeof(tmp_scm));
1784         }
1785
1786         mutex_lock(&u->readlock);
1787
1788         do {
1789                 int chunk;
1790                 struct sk_buff *skb;
1791
1792                 unix_state_lock(sk);
1793                 skb = skb_dequeue(&sk->sk_receive_queue);
1794                 if (skb == NULL) {
1795                         if (copied >= target)
1796                                 goto unlock;
1797
1798                         /*
1799                          *      POSIX 1003.1g mandates this order.
1800                          */
1801
1802                         err = sock_error(sk);
1803                         if (err)
1804                                 goto unlock;
1805                         if (sk->sk_shutdown & RCV_SHUTDOWN)
1806                                 goto unlock;
1807
1808                         unix_state_unlock(sk);
1809                         err = -EAGAIN;
1810                         if (!timeo)
1811                                 break;
1812                         mutex_unlock(&u->readlock);
1813
1814                         timeo = unix_stream_data_wait(sk, timeo);
1815
1816                         if (signal_pending(current)) {
1817                                 err = sock_intr_errno(timeo);
1818                                 goto out;
1819                         }
1820                         mutex_lock(&u->readlock);
1821                         continue;
1822  unlock:
1823                         unix_state_unlock(sk);
1824                         break;
1825                 }
1826                 unix_state_unlock(sk);
1827
1828                 if (check_creds) {
1829                         /* Never glue messages from different writers */
1830                         if (memcmp(UNIXCREDS(skb), &siocb->scm->creds,
1831                                    sizeof(siocb->scm->creds)) != 0) {
1832                                 skb_queue_head(&sk->sk_receive_queue, skb);
1833                                 break;
1834                         }
1835                 } else {
1836                         /* Copy credentials */
1837                         siocb->scm->creds = *UNIXCREDS(skb);
1838                         check_creds = 1;
1839                 }
1840
1841                 /* Copy address just once */
1842                 if (sunaddr) {
1843                         unix_copy_addr(msg, skb->sk);
1844                         sunaddr = NULL;
1845                 }
1846
1847                 chunk = min_t(unsigned int, skb->len, size);
1848                 if (memcpy_toiovec(msg->msg_iov, skb->data, chunk)) {
1849                         skb_queue_head(&sk->sk_receive_queue, skb);
1850                         if (copied == 0)
1851                                 copied = -EFAULT;
1852                         break;
1853                 }
1854                 copied += chunk;
1855                 size -= chunk;
1856
1857                 /* Mark read part of skb as used */
1858                 if (!(flags & MSG_PEEK)) {
1859                         skb_pull(skb, chunk);
1860
1861                         if (UNIXCB(skb).fp)
1862                                 unix_detach_fds(siocb->scm, skb);
1863
1864                         /* put the skb back if we didn't use it up.. */
1865                         if (skb->len) {
1866                                 skb_queue_head(&sk->sk_receive_queue, skb);
1867                                 break;
1868                         }
1869
1870                         kfree_skb(skb);
1871
1872                         if (siocb->scm->fp)
1873                                 break;
1874                 } else {
1875                         /* It is questionable, see note in unix_dgram_recvmsg.
1876                          */
1877                         if (UNIXCB(skb).fp)
1878                                 siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1879
1880                         /* put message back and return */
1881                         skb_queue_head(&sk->sk_receive_queue, skb);
1882                         break;
1883                 }
1884         } while (size);
1885
1886         mutex_unlock(&u->readlock);
1887         scm_recv(sock, msg, siocb->scm, flags);
1888 out:
1889         return copied ? : err;
1890 }
1891
1892 static int unix_shutdown(struct socket *sock, int mode)
1893 {
1894         struct sock *sk = sock->sk;
1895         struct sock *other;
1896
1897         mode = (mode+1)&(RCV_SHUTDOWN|SEND_SHUTDOWN);
1898
1899         if (mode) {
1900                 unix_state_lock(sk);
1901                 sk->sk_shutdown |= mode;
1902                 other = unix_peer(sk);
1903                 if (other)
1904                         sock_hold(other);
1905                 unix_state_unlock(sk);
1906                 sk->sk_state_change(sk);
1907
1908                 if (other &&
1909                         (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET)) {
1910
1911                         int peer_mode = 0;
1912
1913                         if (mode&RCV_SHUTDOWN)
1914                                 peer_mode |= SEND_SHUTDOWN;
1915                         if (mode&SEND_SHUTDOWN)
1916                                 peer_mode |= RCV_SHUTDOWN;
1917                         unix_state_lock(other);
1918                         other->sk_shutdown |= peer_mode;
1919                         unix_state_unlock(other);
1920                         other->sk_state_change(other);
1921                         read_lock(&other->sk_callback_lock);
1922                         if (peer_mode == SHUTDOWN_MASK)
1923                                 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_HUP);
1924                         else if (peer_mode & RCV_SHUTDOWN)
1925                                 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_IN);
1926                         read_unlock(&other->sk_callback_lock);
1927                 }
1928                 if (other)
1929                         sock_put(other);
1930         }
1931         return 0;
1932 }
1933
1934 static int unix_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1935 {
1936         struct sock *sk = sock->sk;
1937         long amount = 0;
1938         int err;
1939
1940         switch (cmd) {
1941         case SIOCOUTQ:
1942                 amount = atomic_read(&sk->sk_wmem_alloc);
1943                 err = put_user(amount, (int __user *)arg);
1944                 break;
1945         case SIOCINQ:
1946                 {
1947                         struct sk_buff *skb;
1948
1949                         if (sk->sk_state == TCP_LISTEN) {
1950                                 err = -EINVAL;
1951                                 break;
1952                         }
1953
1954                         spin_lock(&sk->sk_receive_queue.lock);
1955                         if (sk->sk_type == SOCK_STREAM ||
1956                             sk->sk_type == SOCK_SEQPACKET) {
1957                                 skb_queue_walk(&sk->sk_receive_queue, skb)
1958                                         amount += skb->len;
1959                         } else {
1960                                 skb = skb_peek(&sk->sk_receive_queue);
1961                                 if (skb)
1962                                         amount = skb->len;
1963                         }
1964                         spin_unlock(&sk->sk_receive_queue.lock);
1965                         err = put_user(amount, (int __user *)arg);
1966                         break;
1967                 }
1968
1969         default:
1970                 err = -ENOIOCTLCMD;
1971                 break;
1972         }
1973         return err;
1974 }
1975
1976 static unsigned int unix_poll(struct file *file, struct socket *sock, poll_table *wait)
1977 {
1978         struct sock *sk = sock->sk;
1979         unsigned int mask;
1980
1981         poll_wait(file, sk->sk_sleep, wait);
1982         mask = 0;
1983
1984         /* exceptional events? */
1985         if (sk->sk_err)
1986                 mask |= POLLERR;
1987         if (sk->sk_shutdown == SHUTDOWN_MASK)
1988                 mask |= POLLHUP;
1989         if (sk->sk_shutdown & RCV_SHUTDOWN)
1990                 mask |= POLLRDHUP;
1991
1992         /* readable? */
1993         if (!skb_queue_empty(&sk->sk_receive_queue) ||
1994             (sk->sk_shutdown & RCV_SHUTDOWN))
1995                 mask |= POLLIN | POLLRDNORM;
1996
1997         /* Connection-based need to check for termination and startup */
1998         if ((sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) &&
1999             sk->sk_state == TCP_CLOSE)
2000                 mask |= POLLHUP;
2001
2002         /*
2003          * we set writable also when the other side has shut down the
2004          * connection. This prevents stuck sockets.
2005          */
2006         if (unix_writable(sk))
2007                 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2008
2009         return mask;
2010 }
2011
2012 static unsigned int unix_dgram_poll(struct file *file, struct socket *sock,
2013                                     poll_table *wait)
2014 {
2015         struct sock *sk = sock->sk, *other;
2016         unsigned int mask, writable;
2017
2018         poll_wait(file, sk->sk_sleep, wait);
2019         mask = 0;
2020
2021         /* exceptional events? */
2022         if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue))
2023                 mask |= POLLERR;
2024         if (sk->sk_shutdown & RCV_SHUTDOWN)
2025                 mask |= POLLRDHUP;
2026         if (sk->sk_shutdown == SHUTDOWN_MASK)
2027                 mask |= POLLHUP;
2028
2029         /* readable? */
2030         if (!skb_queue_empty(&sk->sk_receive_queue) ||
2031             (sk->sk_shutdown & RCV_SHUTDOWN))
2032                 mask |= POLLIN | POLLRDNORM;
2033
2034         /* Connection-based need to check for termination and startup */
2035         if (sk->sk_type == SOCK_SEQPACKET) {
2036                 if (sk->sk_state == TCP_CLOSE)
2037                         mask |= POLLHUP;
2038                 /* connection hasn't started yet? */
2039                 if (sk->sk_state == TCP_SYN_SENT)
2040                         return mask;
2041         }
2042
2043         /* writable? */
2044         writable = unix_writable(sk);
2045         if (writable) {
2046                 other = unix_peer_get(sk);
2047                 if (other) {
2048                         if (unix_peer(other) != sk) {
2049                                 poll_wait(file, &unix_sk(other)->peer_wait,
2050                                           wait);
2051                                 if (unix_recvq_full(other))
2052                                         writable = 0;
2053                         }
2054
2055                         sock_put(other);
2056                 }
2057         }
2058
2059         if (writable)
2060                 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2061         else
2062                 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
2063
2064         return mask;
2065 }
2066
2067 #ifdef CONFIG_PROC_FS
2068 static struct sock *first_unix_socket(int *i)
2069 {
2070         for (*i = 0; *i <= UNIX_HASH_SIZE; (*i)++) {
2071                 if (!hlist_empty(&unix_socket_table[*i]))
2072                         return __sk_head(&unix_socket_table[*i]);
2073         }
2074         return NULL;
2075 }
2076
2077 static struct sock *next_unix_socket(int *i, struct sock *s)
2078 {
2079         struct sock *next = sk_next(s);
2080         /* More in this chain? */
2081         if (next)
2082                 return next;
2083         /* Look for next non-empty chain. */
2084         for ((*i)++; *i <= UNIX_HASH_SIZE; (*i)++) {
2085                 if (!hlist_empty(&unix_socket_table[*i]))
2086                         return __sk_head(&unix_socket_table[*i]);
2087         }
2088         return NULL;
2089 }
2090
2091 struct unix_iter_state {
2092         struct seq_net_private p;
2093         int i;
2094 };
2095
2096 static struct sock *unix_seq_idx(struct seq_file *seq, loff_t pos)
2097 {
2098         struct unix_iter_state *iter = seq->private;
2099         loff_t off = 0;
2100         struct sock *s;
2101
2102         for (s = first_unix_socket(&iter->i); s; s = next_unix_socket(&iter->i, s)) {
2103                 if (sock_net(s) != seq_file_net(seq))
2104                         continue;
2105                 if (off == pos)
2106                         return s;
2107                 ++off;
2108         }
2109         return NULL;
2110 }
2111
2112 static void *unix_seq_start(struct seq_file *seq, loff_t *pos)
2113         __acquires(unix_table_lock)
2114 {
2115         spin_lock(&unix_table_lock);
2116         return *pos ? unix_seq_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2117 }
2118
2119 static void *unix_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2120 {
2121         struct unix_iter_state *iter = seq->private;
2122         struct sock *sk = v;
2123         ++*pos;
2124
2125         if (v == SEQ_START_TOKEN)
2126                 sk = first_unix_socket(&iter->i);
2127         else
2128                 sk = next_unix_socket(&iter->i, sk);
2129         while (sk && (sock_net(sk) != seq_file_net(seq)))
2130                 sk = next_unix_socket(&iter->i, sk);
2131         return sk;
2132 }
2133
2134 static void unix_seq_stop(struct seq_file *seq, void *v)
2135         __releases(unix_table_lock)
2136 {
2137         spin_unlock(&unix_table_lock);
2138 }
2139
2140 static int unix_seq_show(struct seq_file *seq, void *v)
2141 {
2142
2143         if (v == SEQ_START_TOKEN)
2144                 seq_puts(seq, "Num       RefCount Protocol Flags    Type St "
2145                          "Inode Path\n");
2146         else {
2147                 struct sock *s = v;
2148                 struct unix_sock *u = unix_sk(s);
2149                 unix_state_lock(s);
2150
2151                 seq_printf(seq, "%p: %08X %08X %08X %04X %02X %5lu",
2152                         s,
2153                         atomic_read(&s->sk_refcnt),
2154                         0,
2155                         s->sk_state == TCP_LISTEN ? __SO_ACCEPTCON : 0,
2156                         s->sk_type,
2157                         s->sk_socket ?
2158                         (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTED : SS_UNCONNECTED) :
2159                         (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTING : SS_DISCONNECTING),
2160                         sock_i_ino(s));
2161
2162                 if (u->addr) {
2163                         int i, len;
2164                         seq_putc(seq, ' ');
2165
2166                         i = 0;
2167                         len = u->addr->len - sizeof(short);
2168                         if (!UNIX_ABSTRACT(s))
2169                                 len--;
2170                         else {
2171                                 seq_putc(seq, '@');
2172                                 i++;
2173                         }
2174                         for ( ; i < len; i++)
2175                                 seq_putc(seq, u->addr->name->sun_path[i]);
2176                 }
2177                 unix_state_unlock(s);
2178                 seq_putc(seq, '\n');
2179         }
2180
2181         return 0;
2182 }
2183
2184 static const struct seq_operations unix_seq_ops = {
2185         .start  = unix_seq_start,
2186         .next   = unix_seq_next,
2187         .stop   = unix_seq_stop,
2188         .show   = unix_seq_show,
2189 };
2190
2191 static int unix_seq_open(struct inode *inode, struct file *file)
2192 {
2193         return seq_open_net(inode, file, &unix_seq_ops,
2194                             sizeof(struct unix_iter_state));
2195 }
2196
2197 static const struct file_operations unix_seq_fops = {
2198         .owner          = THIS_MODULE,
2199         .open           = unix_seq_open,
2200         .read           = seq_read,
2201         .llseek         = seq_lseek,
2202         .release        = seq_release_net,
2203 };
2204
2205 #endif
2206
2207 static struct net_proto_family unix_family_ops = {
2208         .family = PF_UNIX,
2209         .create = unix_create,
2210         .owner  = THIS_MODULE,
2211 };
2212
2213
2214 static int unix_net_init(struct net *net)
2215 {
2216         int error = -ENOMEM;
2217
2218         net->unx.sysctl_max_dgram_qlen = 10;
2219         if (unix_sysctl_register(net))
2220                 goto out;
2221
2222 #ifdef CONFIG_PROC_FS
2223         if (!proc_net_fops_create(net, "unix", 0, &unix_seq_fops)) {
2224                 unix_sysctl_unregister(net);
2225                 goto out;
2226         }
2227 #endif
2228         error = 0;
2229 out:
2230         return error;
2231 }
2232
2233 static void unix_net_exit(struct net *net)
2234 {
2235         unix_sysctl_unregister(net);
2236         proc_net_remove(net, "unix");
2237 }
2238
2239 static struct pernet_operations unix_net_ops = {
2240         .init = unix_net_init,
2241         .exit = unix_net_exit,
2242 };
2243
2244 static int __init af_unix_init(void)
2245 {
2246         int rc = -1;
2247         struct sk_buff *dummy_skb;
2248
2249         BUILD_BUG_ON(sizeof(struct unix_skb_parms) > sizeof(dummy_skb->cb));
2250
2251         rc = proto_register(&unix_proto, 1);
2252         if (rc != 0) {
2253                 printk(KERN_CRIT "%s: Cannot create unix_sock SLAB cache!\n",
2254                        __func__);
2255                 goto out;
2256         }
2257
2258         sock_register(&unix_family_ops);
2259         register_pernet_subsys(&unix_net_ops);
2260 out:
2261         return rc;
2262 }
2263
2264 static void __exit af_unix_exit(void)
2265 {
2266         sock_unregister(PF_UNIX);
2267         proto_unregister(&unix_proto);
2268         unregister_pernet_subsys(&unix_net_ops);
2269 }
2270
2271 /* Earlier than device_initcall() so that other drivers invoking
2272    request_module() don't end up in a loop when modprobe tries
2273    to use a UNIX socket. But later than subsys_initcall() because
2274    we depend on stuff initialised there */
2275 fs_initcall(af_unix_init);
2276 module_exit(af_unix_exit);
2277
2278 MODULE_LICENSE("GPL");
2279 MODULE_ALIAS_NETPROTO(PF_UNIX);