net/unix/af_unix.c

   1 /*
   2  * NET4:        Implementation of BSD Unix domain sockets.
   3  *
   4  * Authors:     Alan Cox, <alan@lxorguk.ukuu.org.uk>
   5  *
   6  *              This program is free software; you can redistribute it and/or
   7  *              modify it under the terms of the GNU General Public License
   8  *              as published by the Free Software Foundation; either version
   9  *              2 of the License, or (at your option) any later version.
  10  *
  11  * Fixes:
  12  *              Linus Torvalds  :       Assorted bug cures.
  13  *              Niibe Yutaka    :       async I/O support.
  14  *              Carsten Paeth   :       PF_UNIX check, address fixes.
  15  *              Alan Cox        :       Limit size of allocated blocks.
  16  *              Alan Cox        :       Fixed the stupid socketpair bug.
  17  *              Alan Cox        :       BSD compatibility fine tuning.
  18  *              Alan Cox        :       Fixed a bug in connect when interrupted.
  19  *              Alan Cox        :       Sorted out a proper draft version of
  20  *                                      file descriptor passing hacked up from
  21  *                                      Mike Shaver's work.
  22  *              Marty Leisner   :       Fixes to fd passing
  23  *              Nick Nevin      :       recvmsg bugfix.
  24  *              Alan Cox        :       Started proper garbage collector
  25  *              Heiko EiBfeldt  :       Missing verify_area check
  26  *              Alan Cox        :       Started POSIXisms
  27  *              Andreas Schwab  :       Replace inode by dentry for proper
  28  *                                      reference counting
  29  *              Kirk Petersen   :       Made this a module
  30  *          Christoph Rohland   :       Elegant non-blocking accept/connect algorithm.
  31  *                                      Lots of bug fixes.
  32  *           Alexey Kuznetosv   :       Repaired (I hope) bugs introduces
  33  *                                      by above two patches.
  34  *           Andrea Arcangeli   :       If possible we block in connect(2)
  35  *                                      if the max backlog of the listen socket
  36  *                                      is been reached. This won't break
  37  *                                      old apps and it will avoid huge amount
  38  *                                      of socks hashed (this for unix_gc()
  39  *                                      performances reasons).
  40  *                                      Security fix that limits the max
  41  *                                      number of socks to 2*max_files and
  42  *                                      the number of skb queueable in the
  43  *                                      dgram receiver.
  44  *              Artur Skawina   :       Hash function optimizations
  45  *           Alexey Kuznetsov   :       Full scale SMP. Lot of bugs are introduced 8)
  46  *            Malcolm Beattie   :       Set peercred for socketpair
  47  *           Michal Ostrowski   :       Module initialization cleanup.
  48  *           Arnaldo C. Melo    :       Remove MOD_{INC,DEC}_USE_COUNT,
  49  *                                      the core infrastructure is doing that
  50  *                                      for all net proto families now (2.5.69+)
  51  *
  52  *
  53  * Known differences from reference BSD that was tested:
  54  *
  55  *      [TO FIX]
  56  *      ECONNREFUSED is not returned from one end of a connected() socket to the
  57  *              other the moment one end closes.
  58  *      fstat() doesn't return st_dev=0, and give the blksize as high water mark
  59  *              and a fake inode identifier (nor the BSD first socket fstat twice bug).
  60  *      [NOT TO FIX]
  61  *      accept() returns a path name even if the connecting socket has closed
  62  *              in the meantime (BSD loses the path and gives up).
  63  *      accept() returns 0 length path for an unbound connector. BSD returns 16
  64  *              and a null first byte in the path (but not for gethost/peername - BSD bug ??)
  65  *      socketpair(...SOCK_RAW..) doesn't panic the kernel.
  66  *      BSD af_unix apparently has connect forgetting to block properly.
  67  *              (need to check this with the POSIX spec in detail)
  68  *
  69  * Differences from 2.0.0-11-... (ANK)
  70  *      Bug fixes and improvements.
  71  *              - client shutdown killed server socket.
  72  *              - removed all useless cli/sti pairs.
  73  *
  74  *      Semantic changes/extensions.
  75  *              - generic control message passing.
  76  *              - SCM_CREDENTIALS control message.
  77  *              - "Abstract" (not FS based) socket bindings.
  78  *                Abstract names are sequences of bytes (not zero terminated)
  79  *                started by 0, so that this name space does not intersect
  80  *                with BSD names.
  81  */
  82
  83 #include <linux/module.h>
  84 #include <linux/kernel.h>
  85 #include <linux/signal.h>
  86 #include <linux/sched.h>
  87 #include <linux/errno.h>
  88 #include <linux/string.h>
  89 #include <linux/stat.h>
  90 #include <linux/dcache.h>
  91 #include <linux/namei.h>
  92 #include <linux/socket.h>
  93 #include <linux/un.h>
  94 #include <linux/fcntl.h>
  95 #include <linux/termios.h>
  96 #include <linux/sockios.h>
  97 #include <linux/net.h>
  98 #include <linux/in.h>
  99 #include <linux/fs.h>
 100 #include <linux/slab.h>
 101 #include <asm/uaccess.h>
 102 #include <linux/skbuff.h>
 103 #include <linux/netdevice.h>
 104 #include <net/net_namespace.h>
 105 #include <net/sock.h>
 106 #include <net/tcp_states.h>
 107 #include <net/af_unix.h>
 108 #include <linux/proc_fs.h>
 109 #include <linux/seq_file.h>
 110 #include <net/scm.h>
 111 #include <linux/init.h>
 112 #include <linux/poll.h>
 113 #include <linux/rtnetlink.h>
 114 #include <linux/mount.h>
 115 #include <net/checksum.h>
 116 #include <linux/security.h>
 117
 118 static struct hlist_head unix_socket_table[UNIX_HASH_SIZE + 1];
 119 static DEFINE_SPINLOCK(unix_table_lock);
 120 static atomic_t unix_nr_socks = ATOMIC_INIT(0);
 121
 122 #define unix_sockets_unbound    (&unix_socket_table[UNIX_HASH_SIZE])
 123
 124 #define UNIX_ABSTRACT(sk)       (unix_sk(sk)->addr->hash != UNIX_HASH_SIZE)
 125
 126 #ifdef CONFIG_SECURITY_NETWORK
 127 static void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb)
 128 {
 129         memcpy(UNIXSID(skb), &scm->secid, sizeof(u32));
 130 }
 131
 132 static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb)
 133 {
 134         scm->secid = *UNIXSID(skb);
 135 }
 136 #else
 137 static inline void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb)
 138 { }
 139
 140 static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb)
 141 { }
 142 #endif /* CONFIG_SECURITY_NETWORK */
 143
 144 /*
 145  *  SMP locking strategy:
 146  *    hash table is protected with spinlock unix_table_lock
 147  *    each socket state is protected by separate rwlock.
 148  */
 149
 150 static inline unsigned unix_hash_fold(__wsum n)
 151 {
 152         unsigned hash = (__force unsigned)n;
 153         hash ^= hash>>16;
 154         hash ^= hash>>8;
 155         return hash&(UNIX_HASH_SIZE-1);
 156 }
 157
 158 #define unix_peer(sk) (unix_sk(sk)->peer)
 159
 160 static inline int unix_our_peer(struct sock *sk, struct sock *osk)
 161 {
 162         return unix_peer(osk) == sk;
 163 }
 164
 165 static inline int unix_may_send(struct sock *sk, struct sock *osk)
 166 {
 167         return unix_peer(osk) == NULL || unix_our_peer(sk, osk);
 168 }
 169
 170 static inline int unix_recvq_full(struct sock const *sk)
 171 {
 172         return skb_queue_len(&sk->sk_receive_queue) > sk->sk_max_ack_backlog;
 173 }
 174
 175 static struct sock *unix_peer_get(struct sock *s)
 176 {
 177         struct sock *peer;
 178
 179         unix_state_lock(s);
 180         peer = unix_peer(s);
 181         if (peer)
 182                 sock_hold(peer);
 183         unix_state_unlock(s);
 184         return peer;
 185 }
 186
 187 static inline void unix_release_addr(struct unix_address *addr)
 188 {
 189         if (atomic_dec_and_test(&addr->refcnt))
 190                 kfree(addr);
 191 }
 192
 193 /*
 194  *      Check unix socket name:
 195  *              - should be not zero length.
 196  *              - if started by not zero, should be NULL terminated (FS object)
 197  *              - if started by zero, it is abstract name.
 198  */
 199
 200 static int unix_mkname(struct sockaddr_un *sunaddr, int len, unsigned *hashp)
 201 {
 202         if (len <= sizeof(short) || len > sizeof(*sunaddr))
 203                 return -EINVAL;
 204         if (!sunaddr || sunaddr->sun_family != AF_UNIX)
 205                 return -EINVAL;
 206         if (sunaddr->sun_path[0]) {
 207                 /*
 208                  * This may look like an off by one error but it is a bit more
 209                  * subtle. 108 is the longest valid AF_UNIX path for a binding.
 210                  * sun_path[108] doesnt as such exist.  However in kernel space
 211                  * we are guaranteed that it is a valid memory location in our
 212                  * kernel address buffer.
 213                  */
 214                 ((char *)sunaddr)[len] = 0;
 215                 len = strlen(sunaddr->sun_path)+1+sizeof(short);
 216                 return len;
 217         }
 218
 219         *hashp = unix_hash_fold(csum_partial(sunaddr, len, 0));
 220         return len;
 221 }
 222
 223 static void __unix_remove_socket(struct sock *sk)
 224 {
 225         sk_del_node_init(sk);
 226 }
 227
 228 static void __unix_insert_socket(struct hlist_head *list, struct sock *sk)
 229 {
 230         WARN_ON(!sk_unhashed(sk));
 231         sk_add_node(sk, list);
 232 }
 233
 234 static inline void unix_remove_socket(struct sock *sk)
 235 {
 236         spin_lock(&unix_table_lock);
 237         __unix_remove_socket(sk);
 238         spin_unlock(&unix_table_lock);
 239 }
 240
 241 static inline void unix_insert_socket(struct hlist_head *list, struct sock *sk)
 242 {
 243         spin_lock(&unix_table_lock);
 244         __unix_insert_socket(list, sk);
 245         spin_unlock(&unix_table_lock);
 246 }
 247
 248 static struct sock *__unix_find_socket_byname(struct net *net,
 249                                               struct sockaddr_un *sunname,
 250                                               int len, int type, unsigned hash)
 251 {
 252         struct sock *s;
 253         struct hlist_node *node;
 254
 255         sk_for_each(s, node, &unix_socket_table[hash ^ type]) {
 256                 struct unix_sock *u = unix_sk(s);
 257
 258                 if (!net_eq(sock_net(s), net))
 259                         continue;
 260
 261                 if (u->addr->len == len &&
 262                     !memcmp(u->addr->name, sunname, len))
 263                         goto found;
 264         }
 265         s = NULL;
 266 found:
 267         return s;
 268 }
 269
 270 static inline struct sock *unix_find_socket_byname(struct net *net,
 271                                                    struct sockaddr_un *sunname,
 272                                                    int len, int type,
 273                                                    unsigned hash)
 274 {
 275         struct sock *s;
 276
 277         spin_lock(&unix_table_lock);
 278         s = __unix_find_socket_byname(net, sunname, len, type, hash);
 279         if (s)
 280                 sock_hold(s);
 281         spin_unlock(&unix_table_lock);
 282         return s;
 283 }
 284
 285 static struct sock *unix_find_socket_byinode(struct net *net, struct inode *i)
 286 {
 287         struct sock *s;
 288         struct hlist_node *node;
 289
 290         spin_lock(&unix_table_lock);
 291         sk_for_each(s, node,
 292                     &unix_socket_table[i->i_ino & (UNIX_HASH_SIZE - 1)]) {
 293                 struct dentry *dentry = unix_sk(s)->dentry;
 294
 295                 if (!net_eq(sock_net(s), net))
 296                         continue;
 297
 298                 if (dentry && dentry->d_inode == i) {
 299                         sock_hold(s);
 300                         goto found;
 301                 }
 302         }
 303         s = NULL;
 304 found:
 305         spin_unlock(&unix_table_lock);
 306         return s;
 307 }
 308
 309 static inline int unix_writable(struct sock *sk)
 310 {
 311         return (atomic_read(&sk->sk_wmem_alloc) << 2) <= sk->sk_sndbuf;
 312 }
 313
 314 static void unix_write_space(struct sock *sk)
 315 {
 316         read_lock(&sk->sk_callback_lock);
 317         if (unix_writable(sk)) {
 318                 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
 319                         wake_up_interruptible_sync(sk->sk_sleep);
 320                 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
 321         }
 322         read_unlock(&sk->sk_callback_lock);
 323 }
 324
 325 /* When dgram socket disconnects (or changes its peer), we clear its receive
 326  * queue of packets arrived from previous peer. First, it allows to do
 327  * flow control based only on wmem_alloc; second, sk connected to peer
 328  * may receive messages only from that peer. */
 329 static void unix_dgram_disconnected(struct sock *sk, struct sock *other)
 330 {
 331         if (!skb_queue_empty(&sk->sk_receive_queue)) {
 332                 skb_queue_purge(&sk->sk_receive_queue);
 333                 wake_up_interruptible_all(&unix_sk(sk)->peer_wait);
 334
 335                 /* If one link of bidirectional dgram pipe is disconnected,
 336                  * we signal error. Messages are lost. Do not make this,
 337                  * when peer was not connected to us.
 338                  */
 339                 if (!sock_flag(other, SOCK_DEAD) && unix_peer(other) == sk) {
 340                         other->sk_err = ECONNRESET;
 341                         other->sk_error_report(other);
 342                 }
 343         }
 344 }
 345
 346 static void unix_sock_destructor(struct sock *sk)
 347 {
 348         struct unix_sock *u = unix_sk(sk);
 349
 350         skb_queue_purge(&sk->sk_receive_queue);
 351
 352         WARN_ON(atomic_read(&sk->sk_wmem_alloc));
 353         WARN_ON(!sk_unhashed(sk));
 354         WARN_ON(sk->sk_socket);
 355         if (!sock_flag(sk, SOCK_DEAD)) {
 356                 printk(KERN_INFO "Attempt to release alive unix socket: %p\n", sk);
 357                 return;
 358         }
 359
 360         if (u->addr)
 361                 unix_release_addr(u->addr);
 362
 363         atomic_dec(&unix_nr_socks);
 364         local_bh_disable();
 365         sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
 366         local_bh_enable();
 367 #ifdef UNIX_REFCNT_DEBUG
 368         printk(KERN_DEBUG "UNIX %p is destroyed, %d are still alive.\n", sk,
 369                 atomic_read(&unix_nr_socks));
 370 #endif
 371 }
 372
 373 static int unix_release_sock(struct sock *sk, int embrion)
 374 {
 375         struct unix_sock *u = unix_sk(sk);
 376         struct dentry *dentry;
 377         struct vfsmount *mnt;
 378         struct sock *skpair;
 379         struct sk_buff *skb;
 380         int state;
 381
 382         unix_remove_socket(sk);
 383
 384         /* Clear state */
 385         unix_state_lock(sk);
 386         sock_orphan(sk);
 387         sk->sk_shutdown = SHUTDOWN_MASK;
 388         dentry       = u->dentry;
 389         u->dentry    = NULL;
 390         mnt          = u->mnt;
 391         u->mnt       = NULL;
 392         state = sk->sk_state;
 393         sk->sk_state = TCP_CLOSE;
 394         unix_state_unlock(sk);
 395
 396         wake_up_interruptible_all(&u->peer_wait);
 397
 398         skpair = unix_peer(sk);
 399
 400         if (skpair != NULL) {
 401                 if (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) {
 402                         unix_state_lock(skpair);
 403                         /* No more writes */
 404                         skpair->sk_shutdown = SHUTDOWN_MASK;
 405                         if (!skb_queue_empty(&sk->sk_receive_queue) || embrion)
 406                                 skpair->sk_err = ECONNRESET;
 407                         unix_state_unlock(skpair);
 408                         skpair->sk_state_change(skpair);
 409                         read_lock(&skpair->sk_callback_lock);
 410                         sk_wake_async(skpair, SOCK_WAKE_WAITD, POLL_HUP);
 411                         read_unlock(&skpair->sk_callback_lock);
 412                 }
 413                 sock_put(skpair); /* It may now die */
 414                 unix_peer(sk) = NULL;
 415         }
 416
 417         /* Try to flush out this socket. Throw out buffers at least */
 418
 419         while ((skb = skb_dequeue(&sk->sk_receive_queue)) != NULL) {
 420                 if (state == TCP_LISTEN)
 421                         unix_release_sock(skb->sk, 1);
 422                 /* passed fds are erased in the kfree_skb hook        */
 423                 kfree_skb(skb);
 424         }
 425
 426         if (dentry) {
 427                 dput(dentry);
 428                 mntput(mnt);
 429         }
 430
 431         sock_put(sk);
 432
 433         /* ---- Socket is dead now and most probably destroyed ---- */
 434
 435         /*
 436          * Fixme: BSD difference: In BSD all sockets connected to use get
 437          *        ECONNRESET and we die on the spot. In Linux we behave
 438          *        like files and pipes do and wait for the last
 439          *        dereference.
 440          *
 441          * Can't we simply set sock->err?
 442          *
 443          *        What the above comment does talk about? --ANK(980817)
 444          */
 445
 446         if (unix_tot_inflight)
 447                 unix_gc();              /* Garbage collect fds */
 448
 449         return 0;
 450 }
 451
 452 static int unix_listen(struct socket *sock, int backlog)
 453 {
 454         int err;
 455         struct sock *sk = sock->sk;
 456         struct unix_sock *u = unix_sk(sk);
 457
 458         err = -EOPNOTSUPP;
 459         if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
 460                 goto out;       /* Only stream/seqpacket sockets accept */
 461         err = -EINVAL;
 462         if (!u->addr)
 463                 goto out;       /* No listens on an unbound socket */
 464         unix_state_lock(sk);
 465         if (sk->sk_state != TCP_CLOSE && sk->sk_state != TCP_LISTEN)
 466                 goto out_unlock;
 467         if (backlog > sk->sk_max_ack_backlog)
 468                 wake_up_interruptible_all(&u->peer_wait);
 469         sk->sk_max_ack_backlog  = backlog;
 470         sk->sk_state            = TCP_LISTEN;
 471         /* set credentials so connect can copy them */
 472         sk->sk_peercred.pid     = task_tgid_vnr(current);
 473         current_euid_egid(&sk->sk_peercred.uid, &sk->sk_peercred.gid);
 474         err = 0;
 475
 476 out_unlock:
 477         unix_state_unlock(sk);
 478 out:
 479         return err;
 480 }
 481
 482 static int unix_release(struct socket *);
 483 static int unix_bind(struct socket *, struct sockaddr *, int);
 484 static int unix_stream_connect(struct socket *, struct sockaddr *,
 485                                int addr_len, int flags);
 486 static int unix_socketpair(struct socket *, struct socket *);
 487 static int unix_accept(struct socket *, struct socket *, int);
 488 static int unix_getname(struct socket *, struct sockaddr *, int *, int);
 489 static unsigned int unix_poll(struct file *, struct socket *, poll_table *);
 490 static unsigned int unix_dgram_poll(struct file *, struct socket *,
 491                                     poll_table *);
 492 static int unix_ioctl(struct socket *, unsigned int, unsigned long);
 493 static int unix_shutdown(struct socket *, int);
 494 static int unix_stream_sendmsg(struct kiocb *, struct socket *,
 495                                struct msghdr *, size_t);
 496 static int unix_stream_recvmsg(struct kiocb *, struct socket *,
 497                                struct msghdr *, size_t, int);
 498 static int unix_dgram_sendmsg(struct kiocb *, struct socket *,
 499                               struct msghdr *, size_t);
 500 static int unix_dgram_recvmsg(struct kiocb *, struct socket *,
 501                               struct msghdr *, size_t, int);
 502 static int unix_dgram_connect(struct socket *, struct sockaddr *,
 503                               int, int);
 504 static int unix_seqpacket_sendmsg(struct kiocb *, struct socket *,
 505                                   struct msghdr *, size_t);
 506
 507 static const struct proto_ops unix_stream_ops = {
 508         .family =       PF_UNIX,
 509         .owner =        THIS_MODULE,
 510         .release =      unix_release,
 511         .bind =         unix_bind,
 512         .connect =      unix_stream_connect,
 513         .socketpair =   unix_socketpair,
 514         .accept =       unix_accept,
 515         .getname =      unix_getname,
 516         .poll =         unix_poll,
 517         .ioctl =        unix_ioctl,
 518         .listen =       unix_listen,
 519         .shutdown =     unix_shutdown,
 520         .setsockopt =   sock_no_setsockopt,
 521         .getsockopt =   sock_no_getsockopt,
 522         .sendmsg =      unix_stream_sendmsg,
 523         .recvmsg =      unix_stream_recvmsg,
 524         .mmap =         sock_no_mmap,
 525         .sendpage =     sock_no_sendpage,
 526 };
 527
 528 static const struct proto_ops unix_dgram_ops = {
 529         .family =       PF_UNIX,
 530         .owner =        THIS_MODULE,
 531         .release =      unix_release,
 532         .bind =         unix_bind,
 533         .connect =      unix_dgram_connect,
 534         .socketpair =   unix_socketpair,
 535         .accept =       sock_no_accept,
 536         .getname =      unix_getname,
 537         .poll =         unix_dgram_poll,
 538         .ioctl =        unix_ioctl,
 539         .listen =       sock_no_listen,
 540         .shutdown =     unix_shutdown,
 541         .setsockopt =   sock_no_setsockopt,
 542         .getsockopt =   sock_no_getsockopt,
 543         .sendmsg =      unix_dgram_sendmsg,
 544         .recvmsg =      unix_dgram_recvmsg,
 545         .mmap =         sock_no_mmap,
 546         .sendpage =     sock_no_sendpage,
 547 };
 548
 549 static const struct proto_ops unix_seqpacket_ops = {
 550         .family =       PF_UNIX,
 551         .owner =        THIS_MODULE,
 552         .release =      unix_release,
 553         .bind =         unix_bind,
 554         .connect =      unix_stream_connect,
 555         .socketpair =   unix_socketpair,
 556         .accept =       unix_accept,
 557         .getname =      unix_getname,
 558         .poll =         unix_dgram_poll,
 559         .ioctl =        unix_ioctl,
 560         .listen =       unix_listen,
 561         .shutdown =     unix_shutdown,
 562         .setsockopt =   sock_no_setsockopt,
 563         .getsockopt =   sock_no_getsockopt,
 564         .sendmsg =      unix_seqpacket_sendmsg,
 565         .recvmsg =      unix_dgram_recvmsg,
 566         .mmap =         sock_no_mmap,
 567         .sendpage =     sock_no_sendpage,
 568 };
 569
 570 static struct proto unix_proto = {
 571         .name                   = "UNIX",
 572         .owner                  = THIS_MODULE,
 573         .obj_size               = sizeof(struct unix_sock),
 574 };
 575
 576 /*
 577  * AF_UNIX sockets do not interact with hardware, hence they
 578  * dont trigger interrupts - so it's safe for them to have
 579  * bh-unsafe locking for their sk_receive_queue.lock. Split off
 580  * this special lock-class by reinitializing the spinlock key:
 581  */
 582 static struct lock_class_key af_unix_sk_receive_queue_lock_key;
 583
 584 static struct sock *unix_create1(struct net *net, struct socket *sock)
 585 {
 586         struct sock *sk = NULL;
 587         struct unix_sock *u;
 588
 589         atomic_inc(&unix_nr_socks);
 590         if (atomic_read(&unix_nr_socks) > 2 * get_max_files())
 591                 goto out;
 592
 593         sk = sk_alloc(net, PF_UNIX, GFP_KERNEL, &unix_proto);
 594         if (!sk)
 595                 goto out;
 596
 597         sock_init_data(sock, sk);
 598         lockdep_set_class(&sk->sk_receive_queue.lock,
 599                                 &af_unix_sk_receive_queue_lock_key);
 600
 601         sk->sk_write_space      = unix_write_space;
 602         sk->sk_max_ack_backlog  = net->unx.sysctl_max_dgram_qlen;
 603         sk->sk_destruct         = unix_sock_destructor;
 604         u         = unix_sk(sk);
 605         u->dentry = NULL;
 606         u->mnt    = NULL;
 607         spin_lock_init(&u->lock);
 608         atomic_long_set(&u->inflight, 0);
 609         INIT_LIST_HEAD(&u->link);
 610         mutex_init(&u->readlock); /* single task reading lock */
 611         init_waitqueue_head(&u->peer_wait);
 612         unix_insert_socket(unix_sockets_unbound, sk);
 613 out:
 614         if (sk == NULL)
 615                 atomic_dec(&unix_nr_socks);
 616         else {
 617                 local_bh_disable();
 618                 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
 619                 local_bh_enable();
 620         }
 621         return sk;
 622 }
 623
 624 static int unix_create(struct net *net, struct socket *sock, int protocol)
 625 {
 626         if (protocol && protocol != PF_UNIX)
 627                 return -EPROTONOSUPPORT;
 628
 629         sock->state = SS_UNCONNECTED;
 630
 631         switch (sock->type) {
 632         case SOCK_STREAM:
 633                 sock->ops = &unix_stream_ops;
 634                 break;
 635                 /*
 636                  *      Believe it or not BSD has AF_UNIX, SOCK_RAW though
 637                  *      nothing uses it.
 638                  */
 639         case SOCK_RAW:
 640                 sock->type = SOCK_DGRAM;
 641         case SOCK_DGRAM:
 642                 sock->ops = &unix_dgram_ops;
 643                 break;
 644         case SOCK_SEQPACKET:
 645                 sock->ops = &unix_seqpacket_ops;
 646                 break;
 647         default:
 648                 return -ESOCKTNOSUPPORT;
 649         }
 650
 651         return unix_create1(net, sock) ? 0 : -ENOMEM;
 652 }
 653
 654 static int unix_release(struct socket *sock)
 655 {
 656         struct sock *sk = sock->sk;
 657
 658         if (!sk)
 659                 return 0;
 660
 661         sock->sk = NULL;
 662
 663         return unix_release_sock(sk, 0);
 664 }
 665
 666 static int unix_autobind(struct socket *sock)
 667 {
 668         struct sock *sk = sock->sk;
 669         struct net *net = sock_net(sk);
 670         struct unix_sock *u = unix_sk(sk);
 671         static u32 ordernum = 1;
 672         struct unix_address *addr;
 673         int err;
 674
 675         mutex_lock(&u->readlock);
 676
 677         err = 0;
 678         if (u->addr)
 679                 goto out;
 680
 681         err = -ENOMEM;
 682         addr = kzalloc(sizeof(*addr) + sizeof(short) + 16, GFP_KERNEL);
 683         if (!addr)
 684                 goto out;
 685
 686         addr->name->sun_family = AF_UNIX;
 687         atomic_set(&addr->refcnt, 1);
 688
 689 retry:
 690         addr->len = sprintf(addr->name->sun_path+1, "%05x", ordernum) + 1 + sizeof(short);
 691         addr->hash = unix_hash_fold(csum_partial(addr->name, addr->len, 0));
 692
 693         spin_lock(&unix_table_lock);
 694         ordernum = (ordernum+1)&0xFFFFF;
 695
 696         if (__unix_find_socket_byname(net, addr->name, addr->len, sock->type,
 697                                       addr->hash)) {
 698                 spin_unlock(&unix_table_lock);
 699                 /* Sanity yield. It is unusual case, but yet... */
 700                 if (!(ordernum&0xFF))
 701                         yield();
 702                 goto retry;
 703         }
 704         addr->hash ^= sk->sk_type;
 705
 706         __unix_remove_socket(sk);
 707         u->addr = addr;
 708         __unix_insert_socket(&unix_socket_table[addr->hash], sk);
 709         spin_unlock(&unix_table_lock);
 710         err = 0;
 711
 712 out:    mutex_unlock(&u->readlock);
 713         return err;
 714 }
 715
 716 static struct sock *unix_find_other(struct net *net,
 717                                     struct sockaddr_un *sunname, int len,
 718                                     int type, unsigned hash, int *error)
 719 {
 720         struct sock *u;
 721         struct path path;
 722         int err = 0;
 723
 724         if (sunname->sun_path[0]) {
 725                 struct inode *inode;
 726                 err = kern_path(sunname->sun_path, LOOKUP_FOLLOW, &path);
 727                 if (err)
 728                         goto fail;
 729                 inode = path.dentry->d_inode;
 730                 err = inode_permission(inode, MAY_WRITE);
 731                 if (err)
 732                         goto put_fail;
 733
 734                 err = -ECONNREFUSED;
 735                 if (!S_ISSOCK(inode->i_mode))
 736                         goto put_fail;
 737                 u = unix_find_socket_byinode(net, inode);
 738                 if (!u)
 739                         goto put_fail;
 740
 741                 if (u->sk_type == type)
 742                         touch_atime(path.mnt, path.dentry);
 743
 744                 path_put(&path);
 745
 746                 err = -EPROTOTYPE;
 747                 if (u->sk_type != type) {
 748                         sock_put(u);
 749                         goto fail;
 750                 }
 751         } else {
 752                 err = -ECONNREFUSED;
 753                 u = unix_find_socket_byname(net, sunname, len, type, hash);
 754                 if (u) {
 755                         struct dentry *dentry;
 756                         dentry = unix_sk(u)->dentry;
 757                         if (dentry)
 758                                 touch_atime(unix_sk(u)->mnt, dentry);
 759                 } else
 760                         goto fail;
 761         }
 762         return u;
 763
 764 put_fail:
 765         path_put(&path);
 766 fail:
 767         *error = err;
 768         return NULL;
 769 }
 770
 771
 772 static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
 773 {
 774         struct sock *sk = sock->sk;
 775         struct net *net = sock_net(sk);
 776         struct unix_sock *u = unix_sk(sk);
 777         struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
 778         struct dentry *dentry = NULL;
 779         struct nameidata nd;
 780         int err;
 781         unsigned hash;
 782         struct unix_address *addr;
 783         struct hlist_head *list;
 784
 785         err = -EINVAL;
 786         if (sunaddr->sun_family != AF_UNIX)
 787                 goto out;
 788
 789         if (addr_len == sizeof(short)) {
 790                 err = unix_autobind(sock);
 791                 goto out;
 792         }
 793
 794         err = unix_mkname(sunaddr, addr_len, &hash);
 795         if (err < 0)
 796                 goto out;
 797         addr_len = err;
 798
 799         mutex_lock(&u->readlock);
 800
 801         err = -EINVAL;
 802         if (u->addr)
 803                 goto out_up;
 804
 805         err = -ENOMEM;
 806         addr = kmalloc(sizeof(*addr)+addr_len, GFP_KERNEL);
 807         if (!addr)
 808                 goto out_up;
 809
 810         memcpy(addr->name, sunaddr, addr_len);
 811         addr->len = addr_len;
 812         addr->hash = hash ^ sk->sk_type;
 813         atomic_set(&addr->refcnt, 1);
 814
 815         if (sunaddr->sun_path[0]) {
 816                 unsigned int mode;
 817                 err = 0;
 818                 /*
 819                  * Get the parent directory, calculate the hash for last
 820                  * component.
 821                  */
 822                 err = path_lookup(sunaddr->sun_path, LOOKUP_PARENT, &nd);
 823                 if (err)
 824                         goto out_mknod_parent;
 825
 826                 dentry = lookup_create(&nd, 0);
 827                 err = PTR_ERR(dentry);
 828                 if (IS_ERR(dentry))
 829                         goto out_mknod_unlock;
 830
 831                 /*
 832                  * All right, let's create it.
 833                  */
 834                 mode = S_IFSOCK |
 835                        (SOCK_INODE(sock)->i_mode & ~current->fs->umask);
 836                 err = mnt_want_write(nd.path.mnt);
 837                 if (err)
 838                         goto out_mknod_dput;
 839                 err = security_path_mknod(&nd.path, dentry, mode, 0);
 840                 if (err)
 841                         goto out_mknod_drop_write;
 842                 err = vfs_mknod(nd.path.dentry->d_inode, dentry, mode, 0);
 843 out_mknod_drop_write:
 844                 mnt_drop_write(nd.path.mnt);
 845                 if (err)
 846                         goto out_mknod_dput;
 847                 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
 848                 dput(nd.path.dentry);
 849                 nd.path.dentry = dentry;
 850
 851                 addr->hash = UNIX_HASH_SIZE;
 852         }
 853
 854         spin_lock(&unix_table_lock);
 855
 856         if (!sunaddr->sun_path[0]) {
 857                 err = -EADDRINUSE;
 858                 if (__unix_find_socket_byname(net, sunaddr, addr_len,
 859                                               sk->sk_type, hash)) {
 860                         unix_release_addr(addr);
 861                         goto out_unlock;
 862                 }
 863
 864                 list = &unix_socket_table[addr->hash];
 865         } else {
 866                 list = &unix_socket_table[dentry->d_inode->i_ino & (UNIX_HASH_SIZE-1)];
 867                 u->dentry = nd.path.dentry;
 868                 u->mnt    = nd.path.mnt;
 869         }
 870
 871         err = 0;
 872         __unix_remove_socket(sk);
 873         u->addr = addr;
 874         __unix_insert_socket(list, sk);
 875
 876 out_unlock:
 877         spin_unlock(&unix_table_lock);
 878 out_up:
 879         mutex_unlock(&u->readlock);
 880 out:
 881         return err;
 882
 883 out_mknod_dput:
 884         dput(dentry);
 885 out_mknod_unlock:
 886         mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
 887         path_put(&nd.path);
 888 out_mknod_parent:
 889         if (err == -EEXIST)
 890                 err = -EADDRINUSE;
 891         unix_release_addr(addr);
 892         goto out_up;
 893 }
 894
 895 static void unix_state_double_lock(struct sock *sk1, struct sock *sk2)
 896 {
 897         if (unlikely(sk1 == sk2) || !sk2) {
 898                 unix_state_lock(sk1);
 899                 return;
 900         }
 901         if (sk1 < sk2) {
 902                 unix_state_lock(sk1);
 903                 unix_state_lock_nested(sk2);
 904         } else {
 905                 unix_state_lock(sk2);
 906                 unix_state_lock_nested(sk1);
 907         }
 908 }
 909
 910 static void unix_state_double_unlock(struct sock *sk1, struct sock *sk2)
 911 {
 912         if (unlikely(sk1 == sk2) || !sk2) {
 913                 unix_state_unlock(sk1);
 914                 return;
 915         }
 916         unix_state_unlock(sk1);
 917         unix_state_unlock(sk2);
 918 }
 919
 920 static int unix_dgram_connect(struct socket *sock, struct sockaddr *addr,
 921                               int alen, int flags)
 922 {
 923         struct sock *sk = sock->sk;
 924         struct net *net = sock_net(sk);
 925         struct sockaddr_un *sunaddr = (struct sockaddr_un *)addr;
 926         struct sock *other;
 927         unsigned hash;
 928         int err;
 929
 930         if (addr->sa_family != AF_UNSPEC) {
 931                 err = unix_mkname(sunaddr, alen, &hash);
 932                 if (err < 0)
 933                         goto out;
 934                 alen = err;
 935
 936                 if (test_bit(SOCK_PASSCRED, &sock->flags) &&
 937                     !unix_sk(sk)->addr && (err = unix_autobind(sock)) != 0)
 938                         goto out;
 939
 940 restart:
 941                 other = unix_find_other(net, sunaddr, alen, sock->type, hash, &err);
 942                 if (!other)
 943                         goto out;
 944
 945                 unix_state_double_lock(sk, other);
 946
 947                 /* Apparently VFS overslept socket death. Retry. */
 948                 if (sock_flag(other, SOCK_DEAD)) {
 949                         unix_state_double_unlock(sk, other);
 950                         sock_put(other);
 951                         goto restart;
 952                 }
 953
 954                 err = -EPERM;
 955                 if (!unix_may_send(sk, other))
 956                         goto out_unlock;
 957
 958                 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
 959                 if (err)
 960                         goto out_unlock;
 961
 962         } else {
 963                 /*
 964                  *      1003.1g breaking connected state with AF_UNSPEC
 965                  */
 966                 other = NULL;
 967                 unix_state_double_lock(sk, other);
 968         }
 969
 970         /*
 971          * If it was connected, reconnect.
 972          */
 973         if (unix_peer(sk)) {
 974                 struct sock *old_peer = unix_peer(sk);
 975                 unix_peer(sk) = other;
 976                 unix_state_double_unlock(sk, other);
 977
 978                 if (other != old_peer)
 979                         unix_dgram_disconnected(sk, old_peer);
 980                 sock_put(old_peer);
 981         } else {
 982                 unix_peer(sk) = other;
 983                 unix_state_double_unlock(sk, other);
 984         }
 985         return 0;
 986
 987 out_unlock:
 988         unix_state_double_unlock(sk, other);
 989         sock_put(other);
 990 out:
 991         return err;
 992 }
 993
 994 static long unix_wait_for_peer(struct sock *other, long timeo)
 995 {
 996         struct unix_sock *u = unix_sk(other);
 997         int sched;
 998         DEFINE_WAIT(wait);
 999
1000         prepare_to_wait_exclusive(&u->peer_wait, &wait, TASK_INTERRUPTIBLE);
1001
1002         sched = !sock_flag(other, SOCK_DEAD) &&
1003                 !(other->sk_shutdown & RCV_SHUTDOWN) &&
1004                 unix_recvq_full(other);
1005
1006         unix_state_unlock(other);
1007
1008         if (sched)
1009                 timeo = schedule_timeout(timeo);
1010
1011         finish_wait(&u->peer_wait, &wait);
1012         return timeo;
1013 }
1014
1015 static int unix_stream_connect(struct socket *sock, struct sockaddr *uaddr,
1016                                int addr_len, int flags)
1017 {
1018         struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
1019         struct sock *sk = sock->sk;
1020         struct net *net = sock_net(sk);
1021         struct unix_sock *u = unix_sk(sk), *newu, *otheru;
1022         struct sock *newsk = NULL;
1023         struct sock *other = NULL;
1024         struct sk_buff *skb = NULL;
1025         unsigned hash;
1026         int st;
1027         int err;
1028         long timeo;
1029
1030         err = unix_mkname(sunaddr, addr_len, &hash);
1031         if (err < 0)
1032                 goto out;
1033         addr_len = err;
1034
1035         if (test_bit(SOCK_PASSCRED, &sock->flags)
1036                 && !u->addr && (err = unix_autobind(sock)) != 0)
1037                 goto out;
1038
1039         timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
1040
1041         /* First of all allocate resources.
1042            If we will make it after state is locked,
1043            we will have to recheck all again in any case.
1044          */
1045
1046         err = -ENOMEM;
1047
1048         /* create new sock for complete connection */
1049         newsk = unix_create1(sock_net(sk), NULL);
1050         if (newsk == NULL)
1051                 goto out;
1052
1053         /* Allocate skb for sending to listening sock */
1054         skb = sock_wmalloc(newsk, 1, 0, GFP_KERNEL);
1055         if (skb == NULL)
1056                 goto out;
1057
1058 restart:
1059         /*  Find listening sock. */
1060         other = unix_find_other(net, sunaddr, addr_len, sk->sk_type, hash, &err);
1061         if (!other)
1062                 goto out;
1063
1064         /* Latch state of peer */
1065         unix_state_lock(other);
1066
1067         /* Apparently VFS overslept socket death. Retry. */
1068         if (sock_flag(other, SOCK_DEAD)) {
1069                 unix_state_unlock(other);
1070                 sock_put(other);
1071                 goto restart;
1072         }
1073
1074         err = -ECONNREFUSED;
1075         if (other->sk_state != TCP_LISTEN)
1076                 goto out_unlock;
1077
1078         if (unix_recvq_full(other)) {
1079                 err = -EAGAIN;
1080                 if (!timeo)
1081                         goto out_unlock;
1082
1083                 timeo = unix_wait_for_peer(other, timeo);
1084
1085                 err = sock_intr_errno(timeo);
1086                 if (signal_pending(current))
1087                         goto out;
1088                 sock_put(other);
1089                 goto restart;
1090         }
1091
1092         /* Latch our state.
1093
1094            It is tricky place. We need to grab write lock and cannot
1095            drop lock on peer. It is dangerous because deadlock is
1096            possible. Connect to self case and simultaneous
1097            attempt to connect are eliminated by checking socket
1098            state. other is TCP_LISTEN, if sk is TCP_LISTEN we
1099            check this before attempt to grab lock.
1100
1101            Well, and we have to recheck the state after socket locked.
1102          */
1103         st = sk->sk_state;
1104
1105         switch (st) {
1106         case TCP_CLOSE:
1107                 /* This is ok... continue with connect */
1108                 break;
1109         case TCP_ESTABLISHED:
1110                 /* Socket is already connected */
1111                 err = -EISCONN;
1112                 goto out_unlock;
1113         default:
1114                 err = -EINVAL;
1115                 goto out_unlock;
1116         }
1117
1118         unix_state_lock_nested(sk);
1119
1120         if (sk->sk_state != st) {
1121                 unix_state_unlock(sk);
1122                 unix_state_unlock(other);
1123                 sock_put(other);
1124                 goto restart;
1125         }
1126
1127         err = security_unix_stream_connect(sock, other->sk_socket, newsk);
1128         if (err) {
1129                 unix_state_unlock(sk);
1130                 goto out_unlock;
1131         }
1132
1133         /* The way is open! Fastly set all the necessary fields... */
1134
1135         sock_hold(sk);
1136         unix_peer(newsk)        = sk;
1137         newsk->sk_state         = TCP_ESTABLISHED;
1138         newsk->sk_type          = sk->sk_type;
1139         newsk->sk_peercred.pid  = task_tgid_vnr(current);
1140         current_euid_egid(&newsk->sk_peercred.uid, &newsk->sk_peercred.gid);
1141         newu = unix_sk(newsk);
1142         newsk->sk_sleep         = &newu->peer_wait;
1143         otheru = unix_sk(other);
1144
1145         /* copy address information from listening to new sock*/
1146         if (otheru->addr) {
1147                 atomic_inc(&otheru->addr->refcnt);
1148                 newu->addr = otheru->addr;
1149         }
1150         if (otheru->dentry) {
1151                 newu->dentry    = dget(otheru->dentry);
1152                 newu->mnt       = mntget(otheru->mnt);
1153         }
1154
1155         /* Set credentials */
1156         sk->sk_peercred = other->sk_peercred;
1157
1158         sock->state     = SS_CONNECTED;
1159         sk->sk_state    = TCP_ESTABLISHED;
1160         sock_hold(newsk);
1161
1162         smp_mb__after_atomic_inc();     /* sock_hold() does an atomic_inc() */
1163         unix_peer(sk)   = newsk;
1164
1165         unix_state_unlock(sk);
1166
1167         /* take ten and and send info to listening sock */
1168         spin_lock(&other->sk_receive_queue.lock);
1169         __skb_queue_tail(&other->sk_receive_queue, skb);
1170         spin_unlock(&other->sk_receive_queue.lock);
1171         unix_state_unlock(other);
1172         other->sk_data_ready(other, 0);
1173         sock_put(other);
1174         return 0;
1175
1176 out_unlock:
1177         if (other)
1178                 unix_state_unlock(other);
1179
1180 out:
1181         if (skb)
1182                 kfree_skb(skb);
1183         if (newsk)
1184                 unix_release_sock(newsk, 0);
1185         if (other)
1186                 sock_put(other);
1187         return err;
1188 }
1189
1190 static int unix_socketpair(struct socket *socka, struct socket *sockb)
1191 {
1192         struct sock *ska = socka->sk, *skb = sockb->sk;
1193
1194         /* Join our sockets back to back */
1195         sock_hold(ska);
1196         sock_hold(skb);
1197         unix_peer(ska) = skb;
1198         unix_peer(skb) = ska;
1199         ska->sk_peercred.pid = skb->sk_peercred.pid = task_tgid_vnr(current);
1200         current_euid_egid(&skb->sk_peercred.uid, &skb->sk_peercred.gid);
1201         ska->sk_peercred.uid = skb->sk_peercred.uid;
1202         ska->sk_peercred.gid = skb->sk_peercred.gid;
1203
1204         if (ska->sk_type != SOCK_DGRAM) {
1205                 ska->sk_state = TCP_ESTABLISHED;
1206                 skb->sk_state = TCP_ESTABLISHED;
1207                 socka->state  = SS_CONNECTED;
1208                 sockb->state  = SS_CONNECTED;
1209         }
1210         return 0;
1211 }
1212
1213 static int unix_accept(struct socket *sock, struct socket *newsock, int flags)
1214 {
1215         struct sock *sk = sock->sk;
1216         struct sock *tsk;
1217         struct sk_buff *skb;
1218         int err;
1219
1220         err = -EOPNOTSUPP;
1221         if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
1222                 goto out;
1223
1224         err = -EINVAL;
1225         if (sk->sk_state != TCP_LISTEN)
1226                 goto out;
1227
1228         /* If socket state is TCP_LISTEN it cannot change (for now...),
1229          * so that no locks are necessary.
1230          */
1231
1232         skb = skb_recv_datagram(sk, 0, flags&O_NONBLOCK, &err);
1233         if (!skb) {
1234                 /* This means receive shutdown. */
1235                 if (err == 0)
1236                         err = -EINVAL;
1237                 goto out;
1238         }
1239
1240         tsk = skb->sk;
1241         skb_free_datagram(sk, skb);
1242         wake_up_interruptible(&unix_sk(sk)->peer_wait);
1243
1244         /* attach accepted sock to socket */
1245         unix_state_lock(tsk);
1246         newsock->state = SS_CONNECTED;
1247         sock_graft(tsk, newsock);
1248         unix_state_unlock(tsk);
1249         return 0;
1250
1251 out:
1252         return err;
1253 }
1254
1255
1256 static int unix_getname(struct socket *sock, struct sockaddr *uaddr, int *uaddr_len, int peer)
1257 {
1258         struct sock *sk = sock->sk;
1259         struct unix_sock *u;
1260         struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
1261         int err = 0;
1262
1263         if (peer) {
1264                 sk = unix_peer_get(sk);
1265
1266                 err = -ENOTCONN;
1267                 if (!sk)
1268                         goto out;
1269                 err = 0;
1270         } else {
1271                 sock_hold(sk);
1272         }
1273
1274         u = unix_sk(sk);
1275         unix_state_lock(sk);
1276         if (!u->addr) {
1277                 sunaddr->sun_family = AF_UNIX;
1278                 sunaddr->sun_path[0] = 0;
1279                 *uaddr_len = sizeof(short);
1280         } else {
1281                 struct unix_address *addr = u->addr;
1282
1283                 *uaddr_len = addr->len;
1284                 memcpy(sunaddr, addr->name, *uaddr_len);
1285         }
1286         unix_state_unlock(sk);
1287         sock_put(sk);
1288 out:
1289         return err;
1290 }
1291
1292 static void unix_detach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1293 {
1294         int i;
1295
1296         scm->fp = UNIXCB(skb).fp;
1297         skb->destructor = sock_wfree;
1298         UNIXCB(skb).fp = NULL;
1299
1300         for (i = scm->fp->count-1; i >= 0; i--)
1301                 unix_notinflight(scm->fp->fp[i]);
1302 }
1303
1304 static void unix_destruct_fds(struct sk_buff *skb)
1305 {
1306         struct scm_cookie scm;
1307         memset(&scm, 0, sizeof(scm));
1308         unix_detach_fds(&scm, skb);
1309
1310         /* Alas, it calls VFS */
1311         /* So fscking what? fput() had been SMP-safe since the last Summer */
1312         scm_destroy(&scm);
1313         sock_wfree(skb);
1314 }
1315
1316 static int unix_attach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1317 {
1318         int i;
1319
1320         /*
1321          * Need to duplicate file references for the sake of garbage
1322          * collection.  Otherwise a socket in the fps might become a
1323          * candidate for GC while the skb is not yet queued.
1324          */
1325         UNIXCB(skb).fp = scm_fp_dup(scm->fp);
1326         if (!UNIXCB(skb).fp)
1327                 return -ENOMEM;
1328
1329         for (i = scm->fp->count-1; i >= 0; i--)
1330                 unix_inflight(scm->fp->fp[i]);
1331         skb->destructor = unix_destruct_fds;
1332         return 0;
1333 }
1334
1335 /*
1336  *      Send AF_UNIX data.
1337  */
1338
1339 static int unix_dgram_sendmsg(struct kiocb *kiocb, struct socket *sock,
1340                               struct msghdr *msg, size_t len)
1341 {
1342         struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1343         struct sock *sk = sock->sk;
1344         struct net *net = sock_net(sk);
1345         struct unix_sock *u = unix_sk(sk);
1346         struct sockaddr_un *sunaddr = msg->msg_name;
1347         struct sock *other = NULL;
1348         int namelen = 0; /* fake GCC */
1349         int err;
1350         unsigned hash;
1351         struct sk_buff *skb;
1352         long timeo;
1353         struct scm_cookie tmp_scm;
1354
1355         if (NULL == siocb->scm)
1356                 siocb->scm = &tmp_scm;
1357         wait_for_unix_gc();
1358         err = scm_send(sock, msg, siocb->scm);
1359         if (err < 0)
1360                 return err;
1361
1362         err = -EOPNOTSUPP;
1363         if (msg->msg_flags&MSG_OOB)
1364                 goto out;
1365
1366         if (msg->msg_namelen) {
1367                 err = unix_mkname(sunaddr, msg->msg_namelen, &hash);
1368                 if (err < 0)
1369                         goto out;
1370                 namelen = err;
1371         } else {
1372                 sunaddr = NULL;
1373                 err = -ENOTCONN;
1374                 other = unix_peer_get(sk);
1375                 if (!other)
1376                         goto out;
1377         }
1378
1379         if (test_bit(SOCK_PASSCRED, &sock->flags)
1380                 && !u->addr && (err = unix_autobind(sock)) != 0)
1381                 goto out;
1382
1383         err = -EMSGSIZE;
1384         if (len > sk->sk_sndbuf - 32)
1385                 goto out;
1386
1387         skb = sock_alloc_send_skb(sk, len, msg->msg_flags&MSG_DONTWAIT, &err);
1388         if (skb == NULL)
1389                 goto out;
1390
1391         memcpy(UNIXCREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1392         if (siocb->scm->fp) {
1393                 err = unix_attach_fds(siocb->scm, skb);
1394                 if (err)
1395                         goto out_free;
1396         }
1397         unix_get_secdata(siocb->scm, skb);
1398
1399         skb_reset_transport_header(skb);
1400         err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
1401         if (err)
1402                 goto out_free;
1403
1404         timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1405
1406 restart:
1407         if (!other) {
1408                 err = -ECONNRESET;
1409                 if (sunaddr == NULL)
1410                         goto out_free;
1411
1412                 other = unix_find_other(net, sunaddr, namelen, sk->sk_type,
1413                                         hash, &err);
1414                 if (other == NULL)
1415                         goto out_free;
1416         }
1417
1418         unix_state_lock(other);
1419         err = -EPERM;
1420         if (!unix_may_send(sk, other))
1421                 goto out_unlock;
1422
1423         if (sock_flag(other, SOCK_DEAD)) {
1424                 /*
1425                  *      Check with 1003.1g - what should
1426                  *      datagram error
1427                  */
1428                 unix_state_unlock(other);
1429                 sock_put(other);
1430
1431                 err = 0;
1432                 unix_state_lock(sk);
1433                 if (unix_peer(sk) == other) {
1434                         unix_peer(sk) = NULL;
1435                         unix_state_unlock(sk);
1436
1437                         unix_dgram_disconnected(sk, other);
1438                         sock_put(other);
1439                         err = -ECONNREFUSED;
1440                 } else {
1441                         unix_state_unlock(sk);
1442                 }
1443
1444                 other = NULL;
1445                 if (err)
1446                         goto out_free;
1447                 goto restart;
1448         }
1449
1450         err = -EPIPE;
1451         if (other->sk_shutdown & RCV_SHUTDOWN)
1452                 goto out_unlock;
1453
1454         if (sk->sk_type != SOCK_SEQPACKET) {
1455                 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1456                 if (err)
1457                         goto out_unlock;
1458         }
1459
1460         if (unix_peer(other) != sk && unix_recvq_full(other)) {
1461                 if (!timeo) {
1462                         err = -EAGAIN;
1463                         goto out_unlock;
1464                 }
1465
1466                 timeo = unix_wait_for_peer(other, timeo);
1467
1468                 err = sock_intr_errno(timeo);
1469                 if (signal_pending(current))
1470                         goto out_free;
1471
1472                 goto restart;
1473         }
1474
1475         skb_queue_tail(&other->sk_receive_queue, skb);
1476         unix_state_unlock(other);
1477         other->sk_data_ready(other, len);
1478         sock_put(other);
1479         scm_destroy(siocb->scm);
1480         return len;
1481
1482 out_unlock:
1483         unix_state_unlock(other);
1484 out_free:
1485         kfree_skb(skb);
1486 out:
1487         if (other)
1488                 sock_put(other);
1489         scm_destroy(siocb->scm);
1490         return err;
1491 }
1492
1493
1494 static int unix_stream_sendmsg(struct kiocb *kiocb, struct socket *sock,
1495                                struct msghdr *msg, size_t len)
1496 {
1497         struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1498         struct sock *sk = sock->sk;
1499         struct sock *other = NULL;
1500         struct sockaddr_un *sunaddr = msg->msg_name;
1501         int err, size;
1502         struct sk_buff *skb;
1503         int sent = 0;
1504         struct scm_cookie tmp_scm;
1505
1506         if (NULL == siocb->scm)
1507                 siocb->scm = &tmp_scm;
1508         wait_for_unix_gc();
1509         err = scm_send(sock, msg, siocb->scm);
1510         if (err < 0)
1511                 return err;
1512
1513         err = -EOPNOTSUPP;
1514         if (msg->msg_flags&MSG_OOB)
1515                 goto out_err;
1516
1517         if (msg->msg_namelen) {
1518                 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
1519                 goto out_err;
1520         } else {
1521                 sunaddr = NULL;
1522                 err = -ENOTCONN;
1523                 other = unix_peer(sk);
1524                 if (!other)
1525                         goto out_err;
1526         }
1527
1528         if (sk->sk_shutdown & SEND_SHUTDOWN)
1529                 goto pipe_err;
1530
1531         while (sent < len) {
1532                 /*
1533                  *      Optimisation for the fact that under 0.01% of X
1534                  *      messages typically need breaking up.
1535                  */
1536
1537                 size = len-sent;
1538
1539                 /* Keep two messages in the pipe so it schedules better */
1540                 if (size > ((sk->sk_sndbuf >> 1) - 64))
1541                         size = (sk->sk_sndbuf >> 1) - 64;
1542
1543                 if (size > SKB_MAX_ALLOC)
1544                         size = SKB_MAX_ALLOC;
1545
1546                 /*
1547                  *      Grab a buffer
1548                  */
1549
1550                 skb = sock_alloc_send_skb(sk, size, msg->msg_flags&MSG_DONTWAIT,
1551                                           &err);
1552
1553                 if (skb == NULL)
1554                         goto out_err;
1555
1556                 /*
1557                  *      If you pass two values to the sock_alloc_send_skb
1558                  *      it tries to grab the large buffer with GFP_NOFS
1559                  *      (which can fail easily), and if it fails grab the
1560                  *      fallback size buffer which is under a page and will
1561                  *      succeed. [Alan]
1562                  */
1563                 size = min_t(int, size, skb_tailroom(skb));
1564
1565                 memcpy(UNIXCREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1566                 if (siocb->scm->fp) {
1567                         err = unix_attach_fds(siocb->scm, skb);
1568                         if (err) {
1569                                 kfree_skb(skb);
1570                                 goto out_err;
1571                         }
1572                 }
1573
1574                 err = memcpy_fromiovec(skb_put(skb, size), msg->msg_iov, size);
1575                 if (err) {
1576                         kfree_skb(skb);
1577                         goto out_err;
1578                 }
1579
1580                 unix_state_lock(other);
1581
1582                 if (sock_flag(other, SOCK_DEAD) ||
1583                     (other->sk_shutdown & RCV_SHUTDOWN))
1584                         goto pipe_err_free;
1585
1586                 skb_queue_tail(&other->sk_receive_queue, skb);
1587                 unix_state_unlock(other);
1588                 other->sk_data_ready(other, size);
1589                 sent += size;
1590         }
1591
1592         scm_destroy(siocb->scm);
1593         siocb->scm = NULL;
1594
1595         return sent;
1596
1597 pipe_err_free:
1598         unix_state_unlock(other);
1599         kfree_skb(skb);
1600 pipe_err:
1601         if (sent == 0 && !(msg->msg_flags&MSG_NOSIGNAL))
1602                 send_sig(SIGPIPE, current, 0);
1603         err = -EPIPE;
1604 out_err:
1605         scm_destroy(siocb->scm);
1606         siocb->scm = NULL;
1607         return sent ? : err;
1608 }
1609
1610 static int unix_seqpacket_sendmsg(struct kiocb *kiocb, struct socket *sock,
1611                                   struct msghdr *msg, size_t len)
1612 {
1613         int err;
1614         struct sock *sk = sock->sk;
1615
1616         err = sock_error(sk);
1617         if (err)
1618                 return err;
1619
1620         if (sk->sk_state != TCP_ESTABLISHED)
1621                 return -ENOTCONN;
1622
1623         if (msg->msg_namelen)
1624                 msg->msg_namelen = 0;
1625
1626         return unix_dgram_sendmsg(kiocb, sock, msg, len);
1627 }
1628
1629 static void unix_copy_addr(struct msghdr *msg, struct sock *sk)
1630 {
1631         struct unix_sock *u = unix_sk(sk);
1632
1633         msg->msg_namelen = 0;
1634         if (u->addr) {
1635                 msg->msg_namelen = u->addr->len;
1636                 memcpy(msg->msg_name, u->addr->name, u->addr->len);
1637         }
1638 }
1639
1640 static int unix_dgram_recvmsg(struct kiocb *iocb, struct socket *sock,
1641                               struct msghdr *msg, size_t size,
1642                               int flags)
1643 {
1644         struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1645         struct scm_cookie tmp_scm;
1646         struct sock *sk = sock->sk;
1647         struct unix_sock *u = unix_sk(sk);
1648         int noblock = flags & MSG_DONTWAIT;
1649         struct sk_buff *skb;
1650         int err;
1651
1652         err = -EOPNOTSUPP;
1653         if (flags&MSG_OOB)
1654                 goto out;
1655
1656         msg->msg_namelen = 0;
1657
1658         mutex_lock(&u->readlock);
1659
1660         skb = skb_recv_datagram(sk, flags, noblock, &err);
1661         if (!skb) {
1662                 unix_state_lock(sk);
1663                 /* Signal EOF on disconnected non-blocking SEQPACKET socket. */
1664                 if (sk->sk_type == SOCK_SEQPACKET && err == -EAGAIN &&
1665                     (sk->sk_shutdown & RCV_SHUTDOWN))
1666                         err = 0;
1667                 unix_state_unlock(sk);
1668                 goto out_unlock;
1669         }
1670
1671         wake_up_interruptible_sync(&u->peer_wait);
1672
1673         if (msg->msg_name)
1674                 unix_copy_addr(msg, skb->sk);
1675
1676         if (size > skb->len)
1677                 size = skb->len;
1678         else if (size < skb->len)
1679                 msg->msg_flags |= MSG_TRUNC;
1680
1681         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, size);
1682         if (err)
1683                 goto out_free;
1684
1685         if (!siocb->scm) {
1686                 siocb->scm = &tmp_scm;
1687                 memset(&tmp_scm, 0, sizeof(tmp_scm));
1688         }
1689         siocb->scm->creds = *UNIXCREDS(skb);
1690         unix_set_secdata(siocb->scm, skb);
1691
1692         if (!(flags & MSG_PEEK)) {
1693                 if (UNIXCB(skb).fp)
1694                         unix_detach_fds(siocb->scm, skb);
1695         } else {
1696                 /* It is questionable: on PEEK we could:
1697                    - do not return fds - good, but too simple 8)
1698                    - return fds, and do not return them on read (old strategy,
1699                      apparently wrong)
1700                    - clone fds (I chose it for now, it is the most universal
1701                      solution)
1702
1703                    POSIX 1003.1g does not actually define this clearly
1704                    at all. POSIX 1003.1g doesn't define a lot of things
1705                    clearly however!
1706
1707                 */
1708                 if (UNIXCB(skb).fp)
1709                         siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1710         }
1711         err = size;
1712
1713         scm_recv(sock, msg, siocb->scm, flags);
1714
1715 out_free:
1716         skb_free_datagram(sk, skb);
1717 out_unlock:
1718         mutex_unlock(&u->readlock);
1719 out:
1720         return err;
1721 }
1722
1723 /*
1724  *      Sleep until data has arrive. But check for races..
1725  */
1726
1727 static long unix_stream_data_wait(struct sock *sk, long timeo)
1728 {
1729         DEFINE_WAIT(wait);
1730
1731         unix_state_lock(sk);
1732
1733         for (;;) {
1734                 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1735
1736                 if (!skb_queue_empty(&sk->sk_receive_queue) ||
1737                     sk->sk_err ||
1738                     (sk->sk_shutdown & RCV_SHUTDOWN) ||
1739                     signal_pending(current) ||
1740                     !timeo)
1741                         break;
1742
1743                 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1744                 unix_state_unlock(sk);
1745                 timeo = schedule_timeout(timeo);
1746                 unix_state_lock(sk);
1747                 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1748         }
1749
1750         finish_wait(sk->sk_sleep, &wait);
1751         unix_state_unlock(sk);
1752         return timeo;
1753 }
1754
1755
1756
1757 static int unix_stream_recvmsg(struct kiocb *iocb, struct socket *sock,
1758                                struct msghdr *msg, size_t size,
1759                                int flags)
1760 {
1761         struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1762         struct scm_cookie tmp_scm;
1763         struct sock *sk = sock->sk;
1764         struct unix_sock *u = unix_sk(sk);
1765         struct sockaddr_un *sunaddr = msg->msg_name;
1766         int copied = 0;
1767         int check_creds = 0;
1768         int target;
1769         int err = 0;
1770         long timeo;
1771
1772         err = -EINVAL;
1773         if (sk->sk_state != TCP_ESTABLISHED)
1774                 goto out;
1775
1776         err = -EOPNOTSUPP;
1777         if (flags&MSG_OOB)
1778                 goto out;
1779
1780         target = sock_rcvlowat(sk, flags&MSG_WAITALL, size);
1781         timeo = sock_rcvtimeo(sk, flags&MSG_DONTWAIT);
1782
1783         msg->msg_namelen = 0;
1784
1785         /* Lock the socket to prevent queue disordering
1786          * while sleeps in memcpy_tomsg
1787          */
1788
1789         if (!siocb->scm) {
1790                 siocb->scm = &tmp_scm;
1791                 memset(&tmp_scm, 0, sizeof(tmp_scm));
1792         }
1793
1794         mutex_lock(&u->readlock);
1795
1796         do {
1797                 int chunk;
1798                 struct sk_buff *skb;
1799
1800                 unix_state_lock(sk);
1801                 skb = skb_dequeue(&sk->sk_receive_queue);
1802                 if (skb == NULL) {
1803                         if (copied >= target)
1804                                 goto unlock;
1805
1806                         /*
1807                          *      POSIX 1003.1g mandates this order.
1808                          */
1809
1810                         err = sock_error(sk);
1811                         if (err)
1812                                 goto unlock;
1813                         if (sk->sk_shutdown & RCV_SHUTDOWN)
1814                                 goto unlock;
1815
1816                         unix_state_unlock(sk);
1817                         err = -EAGAIN;
1818                         if (!timeo)
1819                                 break;
1820                         mutex_unlock(&u->readlock);
1821
1822                         timeo = unix_stream_data_wait(sk, timeo);
1823
1824                         if (signal_pending(current)) {
1825                                 err = sock_intr_errno(timeo);
1826                                 goto out;
1827                         }
1828                         mutex_lock(&u->readlock);
1829                         continue;
1830  unlock:
1831                         unix_state_unlock(sk);
1832                         break;
1833                 }
1834                 unix_state_unlock(sk);
1835
1836                 if (check_creds) {
1837                         /* Never glue messages from different writers */
1838                         if (memcmp(UNIXCREDS(skb), &siocb->scm->creds,
1839                                    sizeof(siocb->scm->creds)) != 0) {
1840                                 skb_queue_head(&sk->sk_receive_queue, skb);
1841                                 break;
1842                         }
1843                 } else {
1844                         /* Copy credentials */
1845                         siocb->scm->creds = *UNIXCREDS(skb);
1846                         check_creds = 1;
1847                 }
1848
1849                 /* Copy address just once */
1850                 if (sunaddr) {
1851                         unix_copy_addr(msg, skb->sk);
1852                         sunaddr = NULL;
1853                 }
1854
1855                 chunk = min_t(unsigned int, skb->len, size);
1856                 if (memcpy_toiovec(msg->msg_iov, skb->data, chunk)) {
1857                         skb_queue_head(&sk->sk_receive_queue, skb);
1858                         if (copied == 0)
1859                                 copied = -EFAULT;
1860                         break;
1861                 }
1862                 copied += chunk;
1863                 size -= chunk;
1864
1865                 /* Mark read part of skb as used */
1866                 if (!(flags & MSG_PEEK)) {
1867                         skb_pull(skb, chunk);
1868
1869                         if (UNIXCB(skb).fp)
1870                                 unix_detach_fds(siocb->scm, skb);
1871
1872                         /* put the skb back if we didn't use it up.. */
1873                         if (skb->len) {
1874                                 skb_queue_head(&sk->sk_receive_queue, skb);
1875                                 break;
1876                         }
1877
1878                         kfree_skb(skb);
1879
1880                         if (siocb->scm->fp)
1881                                 break;
1882                 } else {
1883                         /* It is questionable, see note in unix_dgram_recvmsg.
1884                          */
1885                         if (UNIXCB(skb).fp)
1886                                 siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1887
1888                         /* put message back and return */
1889                         skb_queue_head(&sk->sk_receive_queue, skb);
1890                         break;
1891                 }
1892         } while (size);
1893
1894         mutex_unlock(&u->readlock);
1895         scm_recv(sock, msg, siocb->scm, flags);
1896 out:
1897         return copied ? : err;
1898 }
1899
1900 static int unix_shutdown(struct socket *sock, int mode)
1901 {
1902         struct sock *sk = sock->sk;
1903         struct sock *other;
1904
1905         mode = (mode+1)&(RCV_SHUTDOWN|SEND_SHUTDOWN);
1906
1907         if (mode) {
1908                 unix_state_lock(sk);
1909                 sk->sk_shutdown |= mode;
1910                 other = unix_peer(sk);
1911                 if (other)
1912                         sock_hold(other);
1913                 unix_state_unlock(sk);
1914                 sk->sk_state_change(sk);
1915
1916                 if (other &&
1917                         (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET)) {
1918
1919                         int peer_mode = 0;
1920
1921                         if (mode&RCV_SHUTDOWN)
1922                                 peer_mode |= SEND_SHUTDOWN;
1923                         if (mode&SEND_SHUTDOWN)
1924                                 peer_mode |= RCV_SHUTDOWN;
1925                         unix_state_lock(other);
1926                         other->sk_shutdown |= peer_mode;
1927                         unix_state_unlock(other);
1928                         other->sk_state_change(other);
1929                         read_lock(&other->sk_callback_lock);
1930                         if (peer_mode == SHUTDOWN_MASK)
1931                                 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_HUP);
1932                         else if (peer_mode & RCV_SHUTDOWN)
1933                                 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_IN);
1934                         read_unlock(&other->sk_callback_lock);
1935                 }
1936                 if (other)
1937                         sock_put(other);
1938         }
1939         return 0;
1940 }
1941
1942 static int unix_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1943 {
1944         struct sock *sk = sock->sk;
1945         long amount = 0;
1946         int err;
1947
1948         switch (cmd) {
1949         case SIOCOUTQ:
1950                 amount = atomic_read(&sk->sk_wmem_alloc);
1951                 err = put_user(amount, (int __user *)arg);
1952                 break;
1953         case SIOCINQ:
1954                 {
1955                         struct sk_buff *skb;
1956
1957                         if (sk->sk_state == TCP_LISTEN) {
1958                                 err = -EINVAL;
1959                                 break;
1960                         }
1961
1962                         spin_lock(&sk->sk_receive_queue.lock);
1963                         if (sk->sk_type == SOCK_STREAM ||
1964                             sk->sk_type == SOCK_SEQPACKET) {
1965                                 skb_queue_walk(&sk->sk_receive_queue, skb)
1966                                         amount += skb->len;
1967                         } else {
1968                                 skb = skb_peek(&sk->sk_receive_queue);
1969                                 if (skb)
1970                                         amount = skb->len;
1971                         }
1972                         spin_unlock(&sk->sk_receive_queue.lock);
1973                         err = put_user(amount, (int __user *)arg);
1974                         break;
1975                 }
1976
1977         default:
1978                 err = -ENOIOCTLCMD;
1979                 break;
1980         }
1981         return err;
1982 }
1983
1984 static unsigned int unix_poll(struct file *file, struct socket *sock, poll_table *wait)
1985 {
1986         struct sock *sk = sock->sk;
1987         unsigned int mask;
1988
1989         poll_wait(file, sk->sk_sleep, wait);
1990         mask = 0;
1991
1992         /* exceptional events? */
1993         if (sk->sk_err)
1994                 mask |= POLLERR;
1995         if (sk->sk_shutdown == SHUTDOWN_MASK)
1996                 mask |= POLLHUP;
1997         if (sk->sk_shutdown & RCV_SHUTDOWN)
1998                 mask |= POLLRDHUP;
1999
2000         /* readable? */
2001         if (!skb_queue_empty(&sk->sk_receive_queue) ||
2002             (sk->sk_shutdown & RCV_SHUTDOWN))
2003                 mask |= POLLIN | POLLRDNORM;
2004
2005         /* Connection-based need to check for termination and startup */
2006         if ((sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) &&
2007             sk->sk_state == TCP_CLOSE)
2008                 mask |= POLLHUP;
2009
2010         /*
2011          * we set writable also when the other side has shut down the
2012          * connection. This prevents stuck sockets.
2013          */
2014         if (unix_writable(sk))
2015                 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2016
2017         return mask;
2018 }
2019
2020 static unsigned int unix_dgram_poll(struct file *file, struct socket *sock,
2021                                     poll_table *wait)
2022 {
2023         struct sock *sk = sock->sk, *other;
2024         unsigned int mask, writable;
2025
2026         poll_wait(file, sk->sk_sleep, wait);
2027         mask = 0;
2028
2029         /* exceptional events? */
2030         if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue))
2031                 mask |= POLLERR;
2032         if (sk->sk_shutdown & RCV_SHUTDOWN)
2033                 mask |= POLLRDHUP;
2034         if (sk->sk_shutdown == SHUTDOWN_MASK)
2035                 mask |= POLLHUP;
2036
2037         /* readable? */
2038         if (!skb_queue_empty(&sk->sk_receive_queue) ||
2039             (sk->sk_shutdown & RCV_SHUTDOWN))
2040                 mask |= POLLIN | POLLRDNORM;
2041
2042         /* Connection-based need to check for termination and startup */
2043         if (sk->sk_type == SOCK_SEQPACKET) {
2044                 if (sk->sk_state == TCP_CLOSE)
2045                         mask |= POLLHUP;
2046                 /* connection hasn't started yet? */
2047                 if (sk->sk_state == TCP_SYN_SENT)
2048                         return mask;
2049         }
2050
2051         /* writable? */
2052         writable = unix_writable(sk);
2053         if (writable) {
2054                 other = unix_peer_get(sk);
2055                 if (other) {
2056                         if (unix_peer(other) != sk) {
2057                                 poll_wait(file, &unix_sk(other)->peer_wait,
2058                                           wait);
2059                                 if (unix_recvq_full(other))
2060                                         writable = 0;
2061                         }
2062
2063                         sock_put(other);
2064                 }
2065         }
2066
2067         if (writable)
2068                 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2069         else
2070                 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
2071
2072         return mask;
2073 }
2074
2075 #ifdef CONFIG_PROC_FS
2076 static struct sock *first_unix_socket(int *i)
2077 {
2078         for (*i = 0; *i <= UNIX_HASH_SIZE; (*i)++) {
2079                 if (!hlist_empty(&unix_socket_table[*i]))
2080                         return __sk_head(&unix_socket_table[*i]);
2081         }
2082         return NULL;
2083 }
2084
2085 static struct sock *next_unix_socket(int *i, struct sock *s)
2086 {
2087         struct sock *next = sk_next(s);
2088         /* More in this chain? */
2089         if (next)
2090                 return next;
2091         /* Look for next non-empty chain. */
2092         for ((*i)++; *i <= UNIX_HASH_SIZE; (*i)++) {
2093                 if (!hlist_empty(&unix_socket_table[*i]))
2094                         return __sk_head(&unix_socket_table[*i]);
2095         }
2096         return NULL;
2097 }
2098
2099 struct unix_iter_state {
2100         struct seq_net_private p;
2101         int i;
2102 };
2103
2104 static struct sock *unix_seq_idx(struct seq_file *seq, loff_t pos)
2105 {
2106         struct unix_iter_state *iter = seq->private;
2107         loff_t off = 0;
2108         struct sock *s;
2109
2110         for (s = first_unix_socket(&iter->i); s; s = next_unix_socket(&iter->i, s)) {
2111                 if (sock_net(s) != seq_file_net(seq))
2112                         continue;
2113                 if (off == pos)
2114                         return s;
2115                 ++off;
2116         }
2117         return NULL;
2118 }
2119
2120 static void *unix_seq_start(struct seq_file *seq, loff_t *pos)
2121         __acquires(unix_table_lock)
2122 {
2123         spin_lock(&unix_table_lock);
2124         return *pos ? unix_seq_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2125 }
2126
2127 static void *unix_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2128 {
2129         struct unix_iter_state *iter = seq->private;
2130         struct sock *sk = v;
2131         ++*pos;
2132
2133         if (v == SEQ_START_TOKEN)
2134                 sk = first_unix_socket(&iter->i);
2135         else
2136                 sk = next_unix_socket(&iter->i, sk);
2137         while (sk && (sock_net(sk) != seq_file_net(seq)))
2138                 sk = next_unix_socket(&iter->i, sk);
2139         return sk;
2140 }
2141
2142 static void unix_seq_stop(struct seq_file *seq, void *v)
2143         __releases(unix_table_lock)
2144 {
2145         spin_unlock(&unix_table_lock);
2146 }
2147
2148 static int unix_seq_show(struct seq_file *seq, void *v)
2149 {
2150
2151         if (v == SEQ_START_TOKEN)
2152                 seq_puts(seq, "Num       RefCount Protocol Flags    Type St "
2153                          "Inode Path\n");
2154         else {
2155                 struct sock *s = v;
2156                 struct unix_sock *u = unix_sk(s);
2157                 unix_state_lock(s);
2158
2159                 seq_printf(seq, "%p: %08X %08X %08X %04X %02X %5lu",
2160                         s,
2161                         atomic_read(&s->sk_refcnt),
2162                         0,
2163                         s->sk_state == TCP_LISTEN ? __SO_ACCEPTCON : 0,
2164                         s->sk_type,
2165                         s->sk_socket ?
2166                         (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTED : SS_UNCONNECTED) :
2167                         (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTING : SS_DISCONNECTING),
2168                         sock_i_ino(s));
2169
2170                 if (u->addr) {
2171                         int i, len;
2172                         seq_putc(seq, ' ');
2173
2174                         i = 0;
2175                         len = u->addr->len - sizeof(short);
2176                         if (!UNIX_ABSTRACT(s))
2177                                 len--;
2178                         else {
2179                                 seq_putc(seq, '@');
2180                                 i++;
2181                         }
2182                         for ( ; i < len; i++)
2183                                 seq_putc(seq, u->addr->name->sun_path[i]);
2184                 }
2185                 unix_state_unlock(s);
2186                 seq_putc(seq, '\n');
2187         }
2188
2189         return 0;
2190 }
2191
2192 static const struct seq_operations unix_seq_ops = {
2193         .start  = unix_seq_start,
2194         .next   = unix_seq_next,
2195         .stop   = unix_seq_stop,
2196         .show   = unix_seq_show,
2197 };
2198
2199 static int unix_seq_open(struct inode *inode, struct file *file)
2200 {
2201         return seq_open_net(inode, file, &unix_seq_ops,
2202                             sizeof(struct unix_iter_state));
2203 }
2204
2205 static const struct file_operations unix_seq_fops = {
2206         .owner          = THIS_MODULE,
2207         .open           = unix_seq_open,
2208         .read           = seq_read,
2209         .llseek         = seq_lseek,
2210         .release        = seq_release_net,
2211 };
2212
2213 #endif
2214
2215 static struct net_proto_family unix_family_ops = {
2216         .family = PF_UNIX,
2217         .create = unix_create,
2218         .owner  = THIS_MODULE,
2219 };
2220
2221
2222 static int unix_net_init(struct net *net)
2223 {
2224         int error = -ENOMEM;
2225
2226         net->unx.sysctl_max_dgram_qlen = 10;
2227         if (unix_sysctl_register(net))
2228                 goto out;
2229
2230 #ifdef CONFIG_PROC_FS
2231         if (!proc_net_fops_create(net, "unix", 0, &unix_seq_fops)) {
2232                 unix_sysctl_unregister(net);
2233                 goto out;
2234         }
2235 #endif
2236         error = 0;
2237 out:
2238         return error;
2239 }
2240
2241 static void unix_net_exit(struct net *net)
2242 {
2243         unix_sysctl_unregister(net);
2244         proc_net_remove(net, "unix");
2245 }
2246
2247 static struct pernet_operations unix_net_ops = {
2248         .init = unix_net_init,
2249         .exit = unix_net_exit,
2250 };
2251
2252 static int __init af_unix_init(void)
2253 {
2254         int rc = -1;
2255         struct sk_buff *dummy_skb;
2256
2257         BUILD_BUG_ON(sizeof(struct unix_skb_parms) > sizeof(dummy_skb->cb));
2258
2259         rc = proto_register(&unix_proto, 1);
2260         if (rc != 0) {
2261                 printk(KERN_CRIT "%s: Cannot create unix_sock SLAB cache!\n",
2262                        __func__);
2263                 goto out;
2264         }
2265
2266         sock_register(&unix_family_ops);
2267         register_pernet_subsys(&unix_net_ops);
2268 out:
2269         return rc;
2270 }
2271
2272 static void __exit af_unix_exit(void)
2273 {
2274         sock_unregister(PF_UNIX);
2275         proto_unregister(&unix_proto);
2276         unregister_pernet_subsys(&unix_net_ops);
2277 }
2278
2279 /* Earlier than device_initcall() so that other drivers invoking
2280    request_module() don't end up in a loop when modprobe tries
2281    to use a UNIX socket. But later than subsys_initcall() because
2282    we depend on stuff initialised there */
2283 fs_initcall(af_unix_init);
2284 module_exit(af_unix_exit);
2285
2286 MODULE_LICENSE("GPL");
2287 MODULE_ALIAS_NETPROTO(PF_UNIX);