net/unix/af_unix.c

   1 /*
   2  * NET4:        Implementation of BSD Unix domain sockets.
   3  *
   4  * Authors:     Alan Cox, <alan@lxorguk.ukuu.org.uk>
   5  *
   6  *              This program is free software; you can redistribute it and/or
   7  *              modify it under the terms of the GNU General Public License
   8  *              as published by the Free Software Foundation; either version
   9  *              2 of the License, or (at your option) any later version.
  10  *
  11  * Fixes:
  12  *              Linus Torvalds  :       Assorted bug cures.
  13  *              Niibe Yutaka    :       async I/O support.
  14  *              Carsten Paeth   :       PF_UNIX check, address fixes.
  15  *              Alan Cox        :       Limit size of allocated blocks.
  16  *              Alan Cox        :       Fixed the stupid socketpair bug.
  17  *              Alan Cox        :       BSD compatibility fine tuning.
  18  *              Alan Cox        :       Fixed a bug in connect when interrupted.
  19  *              Alan Cox        :       Sorted out a proper draft version of
  20  *                                      file descriptor passing hacked up from
  21  *                                      Mike Shaver's work.
  22  *              Marty Leisner   :       Fixes to fd passing
  23  *              Nick Nevin      :       recvmsg bugfix.
  24  *              Alan Cox        :       Started proper garbage collector
  25  *              Heiko EiBfeldt  :       Missing verify_area check
  26  *              Alan Cox        :       Started POSIXisms
  27  *              Andreas Schwab  :       Replace inode by dentry for proper
  28  *                                      reference counting
  29  *              Kirk Petersen   :       Made this a module
  30  *          Christoph Rohland   :       Elegant non-blocking accept/connect algorithm.
  31  *                                      Lots of bug fixes.
  32  *           Alexey Kuznetosv   :       Repaired (I hope) bugs introduces
  33  *                                      by above two patches.
  34  *           Andrea Arcangeli   :       If possible we block in connect(2)
  35  *                                      if the max backlog of the listen socket
  36  *                                      is been reached. This won't break
  37  *                                      old apps and it will avoid huge amount
  38  *                                      of socks hashed (this for unix_gc()
  39  *                                      performances reasons).
  40  *                                      Security fix that limits the max
  41  *                                      number of socks to 2*max_files and
  42  *                                      the number of skb queueable in the
  43  *                                      dgram receiver.
  44  *              Artur Skawina   :       Hash function optimizations
  45  *           Alexey Kuznetsov   :       Full scale SMP. Lot of bugs are introduced 8)
  46  *            Malcolm Beattie   :       Set peercred for socketpair
  47  *           Michal Ostrowski   :       Module initialization cleanup.
  48  *           Arnaldo C. Melo    :       Remove MOD_{INC,DEC}_USE_COUNT,
  49  *                                      the core infrastructure is doing that
  50  *                                      for all net proto families now (2.5.69+)
  51  *
  52  *
  53  * Known differences from reference BSD that was tested:
  54  *
  55  *      [TO FIX]
  56  *      ECONNREFUSED is not returned from one end of a connected() socket to the
  57  *              other the moment one end closes.
  58  *      fstat() doesn't return st_dev=0, and give the blksize as high water mark
  59  *              and a fake inode identifier (nor the BSD first socket fstat twice bug).
  60  *      [NOT TO FIX]
  61  *      accept() returns a path name even if the connecting socket has closed
  62  *              in the meantime (BSD loses the path and gives up).
  63  *      accept() returns 0 length path for an unbound connector. BSD returns 16
  64  *              and a null first byte in the path (but not for gethost/peername - BSD bug ??)
  65  *      socketpair(...SOCK_RAW..) doesn't panic the kernel.
  66  *      BSD af_unix apparently has connect forgetting to block properly.
  67  *              (need to check this with the POSIX spec in detail)
  68  *
  69  * Differences from 2.0.0-11-... (ANK)
  70  *      Bug fixes and improvements.
  71  *              - client shutdown killed server socket.
  72  *              - removed all useless cli/sti pairs.
  73  *
  74  *      Semantic changes/extensions.
  75  *              - generic control message passing.
  76  *              - SCM_CREDENTIALS control message.
  77  *              - "Abstract" (not FS based) socket bindings.
  78  *                Abstract names are sequences of bytes (not zero terminated)
  79  *                started by 0, so that this name space does not intersect
  80  *                with BSD names.
  81  */
  82
  83 #include <linux/module.h>
  84 #include <linux/kernel.h>
  85 #include <linux/signal.h>
  86 #include <linux/sched.h>
  87 #include <linux/errno.h>
  88 #include <linux/string.h>
  89 #include <linux/stat.h>
  90 #include <linux/dcache.h>
  91 #include <linux/namei.h>
  92 #include <linux/socket.h>
  93 #include <linux/un.h>
  94 #include <linux/fcntl.h>
  95 #include <linux/termios.h>
  96 #include <linux/sockios.h>
  97 #include <linux/net.h>
  98 #include <linux/in.h>
  99 #include <linux/fs.h>
 100 #include <linux/slab.h>
 101 #include <asm/uaccess.h>
 102 #include <linux/skbuff.h>
 103 #include <linux/netdevice.h>
 104 #include <net/net_namespace.h>
 105 #include <net/sock.h>
 106 #include <net/tcp_states.h>
 107 #include <net/af_unix.h>
 108 #include <linux/proc_fs.h>
 109 #include <linux/seq_file.h>
 110 #include <net/scm.h>
 111 #include <linux/init.h>
 112 #include <linux/poll.h>
 113 #include <linux/rtnetlink.h>
 114 #include <linux/mount.h>
 115 #include <net/checksum.h>
 116 #include <linux/security.h>
 117
 118 static struct hlist_head unix_socket_table[UNIX_HASH_SIZE + 1];
 119 static DEFINE_SPINLOCK(unix_table_lock);
 120 static atomic_t unix_nr_socks = ATOMIC_INIT(0);
 121
 122 #define unix_sockets_unbound    (&unix_socket_table[UNIX_HASH_SIZE])
 123
 124 #define UNIX_ABSTRACT(sk)       (unix_sk(sk)->addr->hash != UNIX_HASH_SIZE)
 125
 126 #ifdef CONFIG_SECURITY_NETWORK
 127 static void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb)
 128 {
 129         memcpy(UNIXSID(skb), &scm->secid, sizeof(u32));
 130 }
 131
 132 static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb)
 133 {
 134         scm->secid = *UNIXSID(skb);
 135 }
 136 #else
 137 static inline void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb)
 138 { }
 139
 140 static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb)
 141 { }
 142 #endif /* CONFIG_SECURITY_NETWORK */
 143
 144 /*
 145  *  SMP locking strategy:
 146  *    hash table is protected with spinlock unix_table_lock
 147  *    each socket state is protected by separate rwlock.
 148  */
 149
 150 static inline unsigned unix_hash_fold(__wsum n)
 151 {
 152         unsigned hash = (__force unsigned)n;
 153         hash ^= hash>>16;
 154         hash ^= hash>>8;
 155         return hash&(UNIX_HASH_SIZE-1);
 156 }
 157
 158 #define unix_peer(sk) (unix_sk(sk)->peer)
 159
 160 static inline int unix_our_peer(struct sock *sk, struct sock *osk)
 161 {
 162         return unix_peer(osk) == sk;
 163 }
 164
 165 static inline int unix_may_send(struct sock *sk, struct sock *osk)
 166 {
 167         return unix_peer(osk) == NULL || unix_our_peer(sk, osk);
 168 }
 169
 170 static inline int unix_recvq_full(struct sock const *sk)
 171 {
 172         return skb_queue_len(&sk->sk_receive_queue) > sk->sk_max_ack_backlog;
 173 }
 174
 175 static struct sock *unix_peer_get(struct sock *s)
 176 {
 177         struct sock *peer;
 178
 179         unix_state_lock(s);
 180         peer = unix_peer(s);
 181         if (peer)
 182                 sock_hold(peer);
 183         unix_state_unlock(s);
 184         return peer;
 185 }
 186
 187 static inline void unix_release_addr(struct unix_address *addr)
 188 {
 189         if (atomic_dec_and_test(&addr->refcnt))
 190                 kfree(addr);
 191 }
 192
 193 /*
 194  *      Check unix socket name:
 195  *              - should be not zero length.
 196  *              - if started by not zero, should be NULL terminated (FS object)
 197  *              - if started by zero, it is abstract name.
 198  */
 199
 200 static int unix_mkname(struct sockaddr_un *sunaddr, int len, unsigned *hashp)
 201 {
 202         if (len <= sizeof(short) || len > sizeof(*sunaddr))
 203                 return -EINVAL;
 204         if (!sunaddr || sunaddr->sun_family != AF_UNIX)
 205                 return -EINVAL;
 206         if (sunaddr->sun_path[0]) {
 207                 /*
 208                  * This may look like an off by one error but it is a bit more
 209                  * subtle. 108 is the longest valid AF_UNIX path for a binding.
 210                  * sun_path[108] doesnt as such exist.  However in kernel space
 211                  * we are guaranteed that it is a valid memory location in our
 212                  * kernel address buffer.
 213                  */
 214                 ((char *)sunaddr)[len] = 0;
 215                 len = strlen(sunaddr->sun_path)+1+sizeof(short);
 216                 return len;
 217         }
 218
 219         *hashp = unix_hash_fold(csum_partial(sunaddr, len, 0));
 220         return len;
 221 }
 222
 223 static void __unix_remove_socket(struct sock *sk)
 224 {
 225         sk_del_node_init(sk);
 226 }
 227
 228 static void __unix_insert_socket(struct hlist_head *list, struct sock *sk)
 229 {
 230         WARN_ON(!sk_unhashed(sk));
 231         sk_add_node(sk, list);
 232 }
 233
 234 static inline void unix_remove_socket(struct sock *sk)
 235 {
 236         spin_lock(&unix_table_lock);
 237         __unix_remove_socket(sk);
 238         spin_unlock(&unix_table_lock);
 239 }
 240
 241 static inline void unix_insert_socket(struct hlist_head *list, struct sock *sk)
 242 {
 243         spin_lock(&unix_table_lock);
 244         __unix_insert_socket(list, sk);
 245         spin_unlock(&unix_table_lock);
 246 }
 247
 248 static struct sock *__unix_find_socket_byname(struct net *net,
 249                                               struct sockaddr_un *sunname,
 250                                               int len, int type, unsigned hash)
 251 {
 252         struct sock *s;
 253         struct hlist_node *node;
 254
 255         sk_for_each(s, node, &unix_socket_table[hash ^ type]) {
 256                 struct unix_sock *u = unix_sk(s);
 257
 258                 if (!net_eq(sock_net(s), net))
 259                         continue;
 260
 261                 if (u->addr->len == len &&
 262                     !memcmp(u->addr->name, sunname, len))
 263                         goto found;
 264         }
 265         s = NULL;
 266 found:
 267         return s;
 268 }
 269
 270 static inline struct sock *unix_find_socket_byname(struct net *net,
 271                                                    struct sockaddr_un *sunname,
 272                                                    int len, int type,
 273                                                    unsigned hash)
 274 {
 275         struct sock *s;
 276
 277         spin_lock(&unix_table_lock);
 278         s = __unix_find_socket_byname(net, sunname, len, type, hash);
 279         if (s)
 280                 sock_hold(s);
 281         spin_unlock(&unix_table_lock);
 282         return s;
 283 }
 284
 285 static struct sock *unix_find_socket_byinode(struct net *net, struct inode *i)
 286 {
 287         struct sock *s;
 288         struct hlist_node *node;
 289
 290         spin_lock(&unix_table_lock);
 291         sk_for_each(s, node,
 292                     &unix_socket_table[i->i_ino & (UNIX_HASH_SIZE - 1)]) {
 293                 struct dentry *dentry = unix_sk(s)->dentry;
 294
 295                 if (!net_eq(sock_net(s), net))
 296                         continue;
 297
 298                 if (dentry && dentry->d_inode == i) {
 299                         sock_hold(s);
 300                         goto found;
 301                 }
 302         }
 303         s = NULL;
 304 found:
 305         spin_unlock(&unix_table_lock);
 306         return s;
 307 }
 308
 309 static inline int unix_writable(struct sock *sk)
 310 {
 311         return (atomic_read(&sk->sk_wmem_alloc) << 2) <= sk->sk_sndbuf;
 312 }
 313
 314 static void unix_write_space(struct sock *sk)
 315 {
 316         read_lock(&sk->sk_callback_lock);
 317         if (unix_writable(sk)) {
 318                 if (sk_has_sleeper(sk))
 319                         wake_up_interruptible_sync(sk->sk_sleep);
 320                 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
 321         }
 322         read_unlock(&sk->sk_callback_lock);
 323 }
 324
 325 /* When dgram socket disconnects (or changes its peer), we clear its receive
 326  * queue of packets arrived from previous peer. First, it allows to do
 327  * flow control based only on wmem_alloc; second, sk connected to peer
 328  * may receive messages only from that peer. */
 329 static void unix_dgram_disconnected(struct sock *sk, struct sock *other)
 330 {
 331         if (!skb_queue_empty(&sk->sk_receive_queue)) {
 332                 skb_queue_purge(&sk->sk_receive_queue);
 333                 wake_up_interruptible_all(&unix_sk(sk)->peer_wait);
 334
 335                 /* If one link of bidirectional dgram pipe is disconnected,
 336                  * we signal error. Messages are lost. Do not make this,
 337                  * when peer was not connected to us.
 338                  */
 339                 if (!sock_flag(other, SOCK_DEAD) && unix_peer(other) == sk) {
 340                         other->sk_err = ECONNRESET;
 341                         other->sk_error_report(other);
 342                 }
 343         }
 344 }
 345
 346 static void unix_sock_destructor(struct sock *sk)
 347 {
 348         struct unix_sock *u = unix_sk(sk);
 349
 350         skb_queue_purge(&sk->sk_receive_queue);
 351
 352         WARN_ON(atomic_read(&sk->sk_wmem_alloc));
 353         WARN_ON(!sk_unhashed(sk));
 354         WARN_ON(sk->sk_socket);
 355         if (!sock_flag(sk, SOCK_DEAD)) {
 356                 printk(KERN_INFO "Attempt to release alive unix socket: %p\n", sk);
 357                 return;
 358         }
 359
 360         if (u->addr)
 361                 unix_release_addr(u->addr);
 362
 363         atomic_dec(&unix_nr_socks);
 364         local_bh_disable();
 365         sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
 366         local_bh_enable();
 367 #ifdef UNIX_REFCNT_DEBUG
 368         printk(KERN_DEBUG "UNIX %p is destroyed, %d are still alive.\n", sk,
 369                 atomic_read(&unix_nr_socks));
 370 #endif
 371 }
 372
 373 static int unix_release_sock(struct sock *sk, int embrion)
 374 {
 375         struct unix_sock *u = unix_sk(sk);
 376         struct dentry *dentry;
 377         struct vfsmount *mnt;
 378         struct sock *skpair;
 379         struct sk_buff *skb;
 380         int state;
 381
 382         unix_remove_socket(sk);
 383
 384         /* Clear state */
 385         unix_state_lock(sk);
 386         sock_orphan(sk);
 387         sk->sk_shutdown = SHUTDOWN_MASK;
 388         dentry       = u->dentry;
 389         u->dentry    = NULL;
 390         mnt          = u->mnt;
 391         u->mnt       = NULL;
 392         state = sk->sk_state;
 393         sk->sk_state = TCP_CLOSE;
 394         unix_state_unlock(sk);
 395
 396         wake_up_interruptible_all(&u->peer_wait);
 397
 398         skpair = unix_peer(sk);
 399
 400         if (skpair != NULL) {
 401                 if (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) {
 402                         unix_state_lock(skpair);
 403                         /* No more writes */
 404                         skpair->sk_shutdown = SHUTDOWN_MASK;
 405                         if (!skb_queue_empty(&sk->sk_receive_queue) || embrion)
 406                                 skpair->sk_err = ECONNRESET;
 407                         unix_state_unlock(skpair);
 408                         skpair->sk_state_change(skpair);
 409                         read_lock(&skpair->sk_callback_lock);
 410                         sk_wake_async(skpair, SOCK_WAKE_WAITD, POLL_HUP);
 411                         read_unlock(&skpair->sk_callback_lock);
 412                 }
 413                 sock_put(skpair); /* It may now die */
 414                 unix_peer(sk) = NULL;
 415         }
 416
 417         /* Try to flush out this socket. Throw out buffers at least */
 418
 419         while ((skb = skb_dequeue(&sk->sk_receive_queue)) != NULL) {
 420                 if (state == TCP_LISTEN)
 421                         unix_release_sock(skb->sk, 1);
 422                 /* passed fds are erased in the kfree_skb hook        */
 423                 kfree_skb(skb);
 424         }
 425
 426         if (dentry) {
 427                 dput(dentry);
 428                 mntput(mnt);
 429         }
 430
 431         sock_put(sk);
 432
 433         /* ---- Socket is dead now and most probably destroyed ---- */
 434
 435         /*
 436          * Fixme: BSD difference: In BSD all sockets connected to use get
 437          *        ECONNRESET and we die on the spot. In Linux we behave
 438          *        like files and pipes do and wait for the last
 439          *        dereference.
 440          *
 441          * Can't we simply set sock->err?
 442          *
 443          *        What the above comment does talk about? --ANK(980817)
 444          */
 445
 446         if (unix_tot_inflight)
 447                 unix_gc();              /* Garbage collect fds */
 448
 449         return 0;
 450 }
 451
 452 static int unix_listen(struct socket *sock, int backlog)
 453 {
 454         int err;
 455         struct sock *sk = sock->sk;
 456         struct unix_sock *u = unix_sk(sk);
 457
 458         err = -EOPNOTSUPP;
 459         if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
 460                 goto out;       /* Only stream/seqpacket sockets accept */
 461         err = -EINVAL;
 462         if (!u->addr)
 463                 goto out;       /* No listens on an unbound socket */
 464         unix_state_lock(sk);
 465         if (sk->sk_state != TCP_CLOSE && sk->sk_state != TCP_LISTEN)
 466                 goto out_unlock;
 467         if (backlog > sk->sk_max_ack_backlog)
 468                 wake_up_interruptible_all(&u->peer_wait);
 469         sk->sk_max_ack_backlog  = backlog;
 470         sk->sk_state            = TCP_LISTEN;
 471         /* set credentials so connect can copy them */
 472         sk->sk_peercred.pid     = task_tgid_vnr(current);
 473         current_euid_egid(&sk->sk_peercred.uid, &sk->sk_peercred.gid);
 474         err = 0;
 475
 476 out_unlock:
 477         unix_state_unlock(sk);
 478 out:
 479         return err;
 480 }
 481
 482 static int unix_release(struct socket *);
 483 static int unix_bind(struct socket *, struct sockaddr *, int);
 484 static int unix_stream_connect(struct socket *, struct sockaddr *,
 485                                int addr_len, int flags);
 486 static int unix_socketpair(struct socket *, struct socket *);
 487 static int unix_accept(struct socket *, struct socket *, int);
 488 static int unix_getname(struct socket *, struct sockaddr *, int *, int);
 489 static unsigned int unix_poll(struct file *, struct socket *, poll_table *);
 490 static unsigned int unix_dgram_poll(struct file *, struct socket *,
 491                                     poll_table *);
 492 static int unix_ioctl(struct socket *, unsigned int, unsigned long);
 493 static int unix_shutdown(struct socket *, int);
 494 static int unix_stream_sendmsg(struct kiocb *, struct socket *,
 495                                struct msghdr *, size_t);
 496 static int unix_stream_recvmsg(struct kiocb *, struct socket *,
 497                                struct msghdr *, size_t, int);
 498 static int unix_dgram_sendmsg(struct kiocb *, struct socket *,
 499                               struct msghdr *, size_t);
 500 static int unix_dgram_recvmsg(struct kiocb *, struct socket *,
 501                               struct msghdr *, size_t, int);
 502 static int unix_dgram_connect(struct socket *, struct sockaddr *,
 503                               int, int);
 504 static int unix_seqpacket_sendmsg(struct kiocb *, struct socket *,
 505                                   struct msghdr *, size_t);
 506
 507 static const struct proto_ops unix_stream_ops = {
 508         .family =       PF_UNIX,
 509         .owner =        THIS_MODULE,
 510         .release =      unix_release,
 511         .bind =         unix_bind,
 512         .connect =      unix_stream_connect,
 513         .socketpair =   unix_socketpair,
 514         .accept =       unix_accept,
 515         .getname =      unix_getname,
 516         .poll =         unix_poll,
 517         .ioctl =        unix_ioctl,
 518         .listen =       unix_listen,
 519         .shutdown =     unix_shutdown,
 520         .setsockopt =   sock_no_setsockopt,
 521         .getsockopt =   sock_no_getsockopt,
 522         .sendmsg =      unix_stream_sendmsg,
 523         .recvmsg =      unix_stream_recvmsg,
 524         .mmap =         sock_no_mmap,
 525         .sendpage =     sock_no_sendpage,
 526 };
 527
 528 static const struct proto_ops unix_dgram_ops = {
 529         .family =       PF_UNIX,
 530         .owner =        THIS_MODULE,
 531         .release =      unix_release,
 532         .bind =         unix_bind,
 533         .connect =      unix_dgram_connect,
 534         .socketpair =   unix_socketpair,
 535         .accept =       sock_no_accept,
 536         .getname =      unix_getname,
 537         .poll =         unix_dgram_poll,
 538         .ioctl =        unix_ioctl,
 539         .listen =       sock_no_listen,
 540         .shutdown =     unix_shutdown,
 541         .setsockopt =   sock_no_setsockopt,
 542         .getsockopt =   sock_no_getsockopt,
 543         .sendmsg =      unix_dgram_sendmsg,
 544         .recvmsg =      unix_dgram_recvmsg,
 545         .mmap =         sock_no_mmap,
 546         .sendpage =     sock_no_sendpage,
 547 };
 548
 549 static const struct proto_ops unix_seqpacket_ops = {
 550         .family =       PF_UNIX,
 551         .owner =        THIS_MODULE,
 552         .release =      unix_release,
 553         .bind =         unix_bind,
 554         .connect =      unix_stream_connect,
 555         .socketpair =   unix_socketpair,
 556         .accept =       unix_accept,
 557         .getname =      unix_getname,
 558         .poll =         unix_dgram_poll,
 559         .ioctl =        unix_ioctl,
 560         .listen =       unix_listen,
 561         .shutdown =     unix_shutdown,
 562         .setsockopt =   sock_no_setsockopt,
 563         .getsockopt =   sock_no_getsockopt,
 564         .sendmsg =      unix_seqpacket_sendmsg,
 565         .recvmsg =      unix_dgram_recvmsg,
 566         .mmap =         sock_no_mmap,
 567         .sendpage =     sock_no_sendpage,
 568 };
 569
 570 static struct proto unix_proto = {
 571         .name                   = "UNIX",
 572         .owner                  = THIS_MODULE,
 573         .obj_size               = sizeof(struct unix_sock),
 574 };
 575
 576 /*
 577  * AF_UNIX sockets do not interact with hardware, hence they
 578  * dont trigger interrupts - so it's safe for them to have
 579  * bh-unsafe locking for their sk_receive_queue.lock. Split off
 580  * this special lock-class by reinitializing the spinlock key:
 581  */
 582 static struct lock_class_key af_unix_sk_receive_queue_lock_key;
 583
 584 static struct sock *unix_create1(struct net *net, struct socket *sock)
 585 {
 586         struct sock *sk = NULL;
 587         struct unix_sock *u;
 588
 589         atomic_inc(&unix_nr_socks);
 590         if (atomic_read(&unix_nr_socks) > 2 * get_max_files())
 591                 goto out;
 592
 593         sk = sk_alloc(net, PF_UNIX, GFP_KERNEL, &unix_proto);
 594         if (!sk)
 595                 goto out;
 596
 597         sock_init_data(sock, sk);
 598         lockdep_set_class(&sk->sk_receive_queue.lock,
 599                                 &af_unix_sk_receive_queue_lock_key);
 600
 601         sk->sk_write_space      = unix_write_space;
 602         sk->sk_max_ack_backlog  = net->unx.sysctl_max_dgram_qlen;
 603         sk->sk_destruct         = unix_sock_destructor;
 604         u         = unix_sk(sk);
 605         u->dentry = NULL;
 606         u->mnt    = NULL;
 607         spin_lock_init(&u->lock);
 608         atomic_long_set(&u->inflight, 0);
 609         INIT_LIST_HEAD(&u->link);
 610         mutex_init(&u->readlock); /* single task reading lock */
 611         init_waitqueue_head(&u->peer_wait);
 612         unix_insert_socket(unix_sockets_unbound, sk);
 613 out:
 614         if (sk == NULL)
 615                 atomic_dec(&unix_nr_socks);
 616         else {
 617                 local_bh_disable();
 618                 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
 619                 local_bh_enable();
 620         }
 621         return sk;
 622 }
 623
 624 static int unix_create(struct net *net, struct socket *sock, int protocol)
 625 {
 626         if (protocol && protocol != PF_UNIX)
 627                 return -EPROTONOSUPPORT;
 628
 629         sock->state = SS_UNCONNECTED;
 630
 631         switch (sock->type) {
 632         case SOCK_STREAM:
 633                 sock->ops = &unix_stream_ops;
 634                 break;
 635                 /*
 636                  *      Believe it or not BSD has AF_UNIX, SOCK_RAW though
 637                  *      nothing uses it.
 638                  */
 639         case SOCK_RAW:
 640                 sock->type = SOCK_DGRAM;
 641         case SOCK_DGRAM:
 642                 sock->ops = &unix_dgram_ops;
 643                 break;
 644         case SOCK_SEQPACKET:
 645                 sock->ops = &unix_seqpacket_ops;
 646                 break;
 647         default:
 648                 return -ESOCKTNOSUPPORT;
 649         }
 650
 651         return unix_create1(net, sock) ? 0 : -ENOMEM;
 652 }
 653
 654 static int unix_release(struct socket *sock)
 655 {
 656         struct sock *sk = sock->sk;
 657
 658         if (!sk)
 659                 return 0;
 660
 661         sock->sk = NULL;
 662
 663         return unix_release_sock(sk, 0);
 664 }
 665
 666 static int unix_autobind(struct socket *sock)
 667 {
 668         struct sock *sk = sock->sk;
 669         struct net *net = sock_net(sk);
 670         struct unix_sock *u = unix_sk(sk);
 671         static u32 ordernum = 1;
 672         struct unix_address *addr;
 673         int err;
 674         unsigned int retries = 0;
 675
 676         mutex_lock(&u->readlock);
 677
 678         err = 0;
 679         if (u->addr)
 680                 goto out;
 681
 682         err = -ENOMEM;
 683         addr = kzalloc(sizeof(*addr) + sizeof(short) + 16, GFP_KERNEL);
 684         if (!addr)
 685                 goto out;
 686
 687         addr->name->sun_family = AF_UNIX;
 688         atomic_set(&addr->refcnt, 1);
 689
 690 retry:
 691         addr->len = sprintf(addr->name->sun_path+1, "%05x", ordernum) + 1 + sizeof(short);
 692         addr->hash = unix_hash_fold(csum_partial(addr->name, addr->len, 0));
 693
 694         spin_lock(&unix_table_lock);
 695         ordernum = (ordernum+1)&0xFFFFF;
 696
 697         if (__unix_find_socket_byname(net, addr->name, addr->len, sock->type,
 698                                       addr->hash)) {
 699                 spin_unlock(&unix_table_lock);
 700                 /*
 701                  * __unix_find_socket_byname() may take long time if many names
 702                  * are already in use.
 703                  */
 704                 cond_resched();
 705                 /* Give up if all names seems to be in use. */
 706                 if (retries++ == 0xFFFFF) {
 707                         err = -ENOSPC;
 708                         kfree(addr);
 709                         goto out;
 710                 }
 711                 goto retry;
 712         }
 713         addr->hash ^= sk->sk_type;
 714
 715         __unix_remove_socket(sk);
 716         u->addr = addr;
 717         __unix_insert_socket(&unix_socket_table[addr->hash], sk);
 718         spin_unlock(&unix_table_lock);
 719         err = 0;
 720
 721 out:    mutex_unlock(&u->readlock);
 722         return err;
 723 }
 724
 725 static struct sock *unix_find_other(struct net *net,
 726                                     struct sockaddr_un *sunname, int len,
 727                                     int type, unsigned hash, int *error)
 728 {
 729         struct sock *u;
 730         struct path path;
 731         int err = 0;
 732
 733         if (sunname->sun_path[0]) {
 734                 struct inode *inode;
 735                 err = kern_path(sunname->sun_path, LOOKUP_FOLLOW, &path);
 736                 if (err)
 737                         goto fail;
 738                 inode = path.dentry->d_inode;
 739                 err = inode_permission(inode, MAY_WRITE);
 740                 if (err)
 741                         goto put_fail;
 742
 743                 err = -ECONNREFUSED;
 744                 if (!S_ISSOCK(inode->i_mode))
 745                         goto put_fail;
 746                 u = unix_find_socket_byinode(net, inode);
 747                 if (!u)
 748                         goto put_fail;
 749
 750                 if (u->sk_type == type)
 751                         touch_atime(path.mnt, path.dentry);
 752
 753                 path_put(&path);
 754
 755                 err = -EPROTOTYPE;
 756                 if (u->sk_type != type) {
 757                         sock_put(u);
 758                         goto fail;
 759                 }
 760         } else {
 761                 err = -ECONNREFUSED;
 762                 u = unix_find_socket_byname(net, sunname, len, type, hash);
 763                 if (u) {
 764                         struct dentry *dentry;
 765                         dentry = unix_sk(u)->dentry;
 766                         if (dentry)
 767                                 touch_atime(unix_sk(u)->mnt, dentry);
 768                 } else
 769                         goto fail;
 770         }
 771         return u;
 772
 773 put_fail:
 774         path_put(&path);
 775 fail:
 776         *error = err;
 777         return NULL;
 778 }
 779
 780
 781 static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
 782 {
 783         struct sock *sk = sock->sk;
 784         struct net *net = sock_net(sk);
 785         struct unix_sock *u = unix_sk(sk);
 786         struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
 787         struct dentry *dentry = NULL;
 788         struct nameidata nd;
 789         int err;
 790         unsigned hash;
 791         struct unix_address *addr;
 792         struct hlist_head *list;
 793
 794         err = -EINVAL;
 795         if (sunaddr->sun_family != AF_UNIX)
 796                 goto out;
 797
 798         if (addr_len == sizeof(short)) {
 799                 err = unix_autobind(sock);
 800                 goto out;
 801         }
 802
 803         err = unix_mkname(sunaddr, addr_len, &hash);
 804         if (err < 0)
 805                 goto out;
 806         addr_len = err;
 807
 808         mutex_lock(&u->readlock);
 809
 810         err = -EINVAL;
 811         if (u->addr)
 812                 goto out_up;
 813
 814         err = -ENOMEM;
 815         addr = kmalloc(sizeof(*addr)+addr_len, GFP_KERNEL);
 816         if (!addr)
 817                 goto out_up;
 818
 819         memcpy(addr->name, sunaddr, addr_len);
 820         addr->len = addr_len;
 821         addr->hash = hash ^ sk->sk_type;
 822         atomic_set(&addr->refcnt, 1);
 823
 824         if (sunaddr->sun_path[0]) {
 825                 unsigned int mode;
 826                 err = 0;
 827                 /*
 828                  * Get the parent directory, calculate the hash for last
 829                  * component.
 830                  */
 831                 err = path_lookup(sunaddr->sun_path, LOOKUP_PARENT, &nd);
 832                 if (err)
 833                         goto out_mknod_parent;
 834
 835                 dentry = lookup_create(&nd, 0);
 836                 err = PTR_ERR(dentry);
 837                 if (IS_ERR(dentry))
 838                         goto out_mknod_unlock;
 839
 840                 /*
 841                  * All right, let's create it.
 842                  */
 843                 mode = S_IFSOCK |
 844                        (SOCK_INODE(sock)->i_mode & ~current_umask());
 845                 err = mnt_want_write(nd.path.mnt);
 846                 if (err)
 847                         goto out_mknod_dput;
 848                 err = security_path_mknod(&nd.path, dentry, mode, 0);
 849                 if (err)
 850                         goto out_mknod_drop_write;
 851                 err = vfs_mknod(nd.path.dentry->d_inode, dentry, mode, 0);
 852 out_mknod_drop_write:
 853                 mnt_drop_write(nd.path.mnt);
 854                 if (err)
 855                         goto out_mknod_dput;
 856                 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
 857                 dput(nd.path.dentry);
 858                 nd.path.dentry = dentry;
 859
 860                 addr->hash = UNIX_HASH_SIZE;
 861         }
 862
 863         spin_lock(&unix_table_lock);
 864
 865         if (!sunaddr->sun_path[0]) {
 866                 err = -EADDRINUSE;
 867                 if (__unix_find_socket_byname(net, sunaddr, addr_len,
 868                                               sk->sk_type, hash)) {
 869                         unix_release_addr(addr);
 870                         goto out_unlock;
 871                 }
 872
 873                 list = &unix_socket_table[addr->hash];
 874         } else {
 875                 list = &unix_socket_table[dentry->d_inode->i_ino & (UNIX_HASH_SIZE-1)];
 876                 u->dentry = nd.path.dentry;
 877                 u->mnt    = nd.path.mnt;
 878         }
 879
 880         err = 0;
 881         __unix_remove_socket(sk);
 882         u->addr = addr;
 883         __unix_insert_socket(list, sk);
 884
 885 out_unlock:
 886         spin_unlock(&unix_table_lock);
 887 out_up:
 888         mutex_unlock(&u->readlock);
 889 out:
 890         return err;
 891
 892 out_mknod_dput:
 893         dput(dentry);
 894 out_mknod_unlock:
 895         mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
 896         path_put(&nd.path);
 897 out_mknod_parent:
 898         if (err == -EEXIST)
 899                 err = -EADDRINUSE;
 900         unix_release_addr(addr);
 901         goto out_up;
 902 }
 903
 904 static void unix_state_double_lock(struct sock *sk1, struct sock *sk2)
 905 {
 906         if (unlikely(sk1 == sk2) || !sk2) {
 907                 unix_state_lock(sk1);
 908                 return;
 909         }
 910         if (sk1 < sk2) {
 911                 unix_state_lock(sk1);
 912                 unix_state_lock_nested(sk2);
 913         } else {
 914                 unix_state_lock(sk2);
 915                 unix_state_lock_nested(sk1);
 916         }
 917 }
 918
 919 static void unix_state_double_unlock(struct sock *sk1, struct sock *sk2)
 920 {
 921         if (unlikely(sk1 == sk2) || !sk2) {
 922                 unix_state_unlock(sk1);
 923                 return;
 924         }
 925         unix_state_unlock(sk1);
 926         unix_state_unlock(sk2);
 927 }
 928
 929 static int unix_dgram_connect(struct socket *sock, struct sockaddr *addr,
 930                               int alen, int flags)
 931 {
 932         struct sock *sk = sock->sk;
 933         struct net *net = sock_net(sk);
 934         struct sockaddr_un *sunaddr = (struct sockaddr_un *)addr;
 935         struct sock *other;
 936         unsigned hash;
 937         int err;
 938
 939         if (addr->sa_family != AF_UNSPEC) {
 940                 err = unix_mkname(sunaddr, alen, &hash);
 941                 if (err < 0)
 942                         goto out;
 943                 alen = err;
 944
 945                 if (test_bit(SOCK_PASSCRED, &sock->flags) &&
 946                     !unix_sk(sk)->addr && (err = unix_autobind(sock)) != 0)
 947                         goto out;
 948
 949 restart:
 950                 other = unix_find_other(net, sunaddr, alen, sock->type, hash, &err);
 951                 if (!other)
 952                         goto out;
 953
 954                 unix_state_double_lock(sk, other);
 955
 956                 /* Apparently VFS overslept socket death. Retry. */
 957                 if (sock_flag(other, SOCK_DEAD)) {
 958                         unix_state_double_unlock(sk, other);
 959                         sock_put(other);
 960                         goto restart;
 961                 }
 962
 963                 err = -EPERM;
 964                 if (!unix_may_send(sk, other))
 965                         goto out_unlock;
 966
 967                 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
 968                 if (err)
 969                         goto out_unlock;
 970
 971         } else {
 972                 /*
 973                  *      1003.1g breaking connected state with AF_UNSPEC
 974                  */
 975                 other = NULL;
 976                 unix_state_double_lock(sk, other);
 977         }
 978
 979         /*
 980          * If it was connected, reconnect.
 981          */
 982         if (unix_peer(sk)) {
 983                 struct sock *old_peer = unix_peer(sk);
 984                 unix_peer(sk) = other;
 985                 unix_state_double_unlock(sk, other);
 986
 987                 if (other != old_peer)
 988                         unix_dgram_disconnected(sk, old_peer);
 989                 sock_put(old_peer);
 990         } else {
 991                 unix_peer(sk) = other;
 992                 unix_state_double_unlock(sk, other);
 993         }
 994         return 0;
 995
 996 out_unlock:
 997         unix_state_double_unlock(sk, other);
 998         sock_put(other);
 999 out:
1000         return err;
1001 }
1002
1003 static long unix_wait_for_peer(struct sock *other, long timeo)
1004 {
1005         struct unix_sock *u = unix_sk(other);
1006         int sched;
1007         DEFINE_WAIT(wait);
1008
1009         prepare_to_wait_exclusive(&u->peer_wait, &wait, TASK_INTERRUPTIBLE);
1010
1011         sched = !sock_flag(other, SOCK_DEAD) &&
1012                 !(other->sk_shutdown & RCV_SHUTDOWN) &&
1013                 unix_recvq_full(other);
1014
1015         unix_state_unlock(other);
1016
1017         if (sched)
1018                 timeo = schedule_timeout(timeo);
1019
1020         finish_wait(&u->peer_wait, &wait);
1021         return timeo;
1022 }
1023
1024 static int unix_stream_connect(struct socket *sock, struct sockaddr *uaddr,
1025                                int addr_len, int flags)
1026 {
1027         struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
1028         struct sock *sk = sock->sk;
1029         struct net *net = sock_net(sk);
1030         struct unix_sock *u = unix_sk(sk), *newu, *otheru;
1031         struct sock *newsk = NULL;
1032         struct sock *other = NULL;
1033         struct sk_buff *skb = NULL;
1034         unsigned hash;
1035         int st;
1036         int err;
1037         long timeo;
1038
1039         err = unix_mkname(sunaddr, addr_len, &hash);
1040         if (err < 0)
1041                 goto out;
1042         addr_len = err;
1043
1044         if (test_bit(SOCK_PASSCRED, &sock->flags)
1045                 && !u->addr && (err = unix_autobind(sock)) != 0)
1046                 goto out;
1047
1048         timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
1049
1050         /* First of all allocate resources.
1051            If we will make it after state is locked,
1052            we will have to recheck all again in any case.
1053          */
1054
1055         err = -ENOMEM;
1056
1057         /* create new sock for complete connection */
1058         newsk = unix_create1(sock_net(sk), NULL);
1059         if (newsk == NULL)
1060                 goto out;
1061
1062         /* Allocate skb for sending to listening sock */
1063         skb = sock_wmalloc(newsk, 1, 0, GFP_KERNEL);
1064         if (skb == NULL)
1065                 goto out;
1066
1067 restart:
1068         /*  Find listening sock. */
1069         other = unix_find_other(net, sunaddr, addr_len, sk->sk_type, hash, &err);
1070         if (!other)
1071                 goto out;
1072
1073         /* Latch state of peer */
1074         unix_state_lock(other);
1075
1076         /* Apparently VFS overslept socket death. Retry. */
1077         if (sock_flag(other, SOCK_DEAD)) {
1078                 unix_state_unlock(other);
1079                 sock_put(other);
1080                 goto restart;
1081         }
1082
1083         err = -ECONNREFUSED;
1084         if (other->sk_state != TCP_LISTEN)
1085                 goto out_unlock;
1086         if (other->sk_shutdown & RCV_SHUTDOWN)
1087                 goto out_unlock;
1088
1089         if (unix_recvq_full(other)) {
1090                 err = -EAGAIN;
1091                 if (!timeo)
1092                         goto out_unlock;
1093
1094                 timeo = unix_wait_for_peer(other, timeo);
1095
1096                 err = sock_intr_errno(timeo);
1097                 if (signal_pending(current))
1098                         goto out;
1099                 sock_put(other);
1100                 goto restart;
1101         }
1102
1103         /* Latch our state.
1104
1105            It is tricky place. We need to grab write lock and cannot
1106            drop lock on peer. It is dangerous because deadlock is
1107            possible. Connect to self case and simultaneous
1108            attempt to connect are eliminated by checking socket
1109            state. other is TCP_LISTEN, if sk is TCP_LISTEN we
1110            check this before attempt to grab lock.
1111
1112            Well, and we have to recheck the state after socket locked.
1113          */
1114         st = sk->sk_state;
1115
1116         switch (st) {
1117         case TCP_CLOSE:
1118                 /* This is ok... continue with connect */
1119                 break;
1120         case TCP_ESTABLISHED:
1121                 /* Socket is already connected */
1122                 err = -EISCONN;
1123                 goto out_unlock;
1124         default:
1125                 err = -EINVAL;
1126                 goto out_unlock;
1127         }
1128
1129         unix_state_lock_nested(sk);
1130
1131         if (sk->sk_state != st) {
1132                 unix_state_unlock(sk);
1133                 unix_state_unlock(other);
1134                 sock_put(other);
1135                 goto restart;
1136         }
1137
1138         err = security_unix_stream_connect(sock, other->sk_socket, newsk);
1139         if (err) {
1140                 unix_state_unlock(sk);
1141                 goto out_unlock;
1142         }
1143
1144         /* The way is open! Fastly set all the necessary fields... */
1145
1146         sock_hold(sk);
1147         unix_peer(newsk)        = sk;
1148         newsk->sk_state         = TCP_ESTABLISHED;
1149         newsk->sk_type          = sk->sk_type;
1150         newsk->sk_peercred.pid  = task_tgid_vnr(current);
1151         current_euid_egid(&newsk->sk_peercred.uid, &newsk->sk_peercred.gid);
1152         newu = unix_sk(newsk);
1153         newsk->sk_sleep         = &newu->peer_wait;
1154         otheru = unix_sk(other);
1155
1156         /* copy address information from listening to new sock*/
1157         if (otheru->addr) {
1158                 atomic_inc(&otheru->addr->refcnt);
1159                 newu->addr = otheru->addr;
1160         }
1161         if (otheru->dentry) {
1162                 newu->dentry    = dget(otheru->dentry);
1163                 newu->mnt       = mntget(otheru->mnt);
1164         }
1165
1166         /* Set credentials */
1167         sk->sk_peercred = other->sk_peercred;
1168
1169         sock->state     = SS_CONNECTED;
1170         sk->sk_state    = TCP_ESTABLISHED;
1171         sock_hold(newsk);
1172
1173         smp_mb__after_atomic_inc();     /* sock_hold() does an atomic_inc() */
1174         unix_peer(sk)   = newsk;
1175
1176         unix_state_unlock(sk);
1177
1178         /* take ten and and send info to listening sock */
1179         spin_lock(&other->sk_receive_queue.lock);
1180         __skb_queue_tail(&other->sk_receive_queue, skb);
1181         spin_unlock(&other->sk_receive_queue.lock);
1182         unix_state_unlock(other);
1183         other->sk_data_ready(other, 0);
1184         sock_put(other);
1185         return 0;
1186
1187 out_unlock:
1188         if (other)
1189                 unix_state_unlock(other);
1190
1191 out:
1192         kfree_skb(skb);
1193         if (newsk)
1194                 unix_release_sock(newsk, 0);
1195         if (other)
1196                 sock_put(other);
1197         return err;
1198 }
1199
1200 static int unix_socketpair(struct socket *socka, struct socket *sockb)
1201 {
1202         struct sock *ska = socka->sk, *skb = sockb->sk;
1203
1204         /* Join our sockets back to back */
1205         sock_hold(ska);
1206         sock_hold(skb);
1207         unix_peer(ska) = skb;
1208         unix_peer(skb) = ska;
1209         ska->sk_peercred.pid = skb->sk_peercred.pid = task_tgid_vnr(current);
1210         current_euid_egid(&skb->sk_peercred.uid, &skb->sk_peercred.gid);
1211         ska->sk_peercred.uid = skb->sk_peercred.uid;
1212         ska->sk_peercred.gid = skb->sk_peercred.gid;
1213
1214         if (ska->sk_type != SOCK_DGRAM) {
1215                 ska->sk_state = TCP_ESTABLISHED;
1216                 skb->sk_state = TCP_ESTABLISHED;
1217                 socka->state  = SS_CONNECTED;
1218                 sockb->state  = SS_CONNECTED;
1219         }
1220         return 0;
1221 }
1222
1223 static int unix_accept(struct socket *sock, struct socket *newsock, int flags)
1224 {
1225         struct sock *sk = sock->sk;
1226         struct sock *tsk;
1227         struct sk_buff *skb;
1228         int err;
1229
1230         err = -EOPNOTSUPP;
1231         if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
1232                 goto out;
1233
1234         err = -EINVAL;
1235         if (sk->sk_state != TCP_LISTEN)
1236                 goto out;
1237
1238         /* If socket state is TCP_LISTEN it cannot change (for now...),
1239          * so that no locks are necessary.
1240          */
1241
1242         skb = skb_recv_datagram(sk, 0, flags&O_NONBLOCK, &err);
1243         if (!skb) {
1244                 /* This means receive shutdown. */
1245                 if (err == 0)
1246                         err = -EINVAL;
1247                 goto out;
1248         }
1249
1250         tsk = skb->sk;
1251         skb_free_datagram(sk, skb);
1252         wake_up_interruptible(&unix_sk(sk)->peer_wait);
1253
1254         /* attach accepted sock to socket */
1255         unix_state_lock(tsk);
1256         newsock->state = SS_CONNECTED;
1257         sock_graft(tsk, newsock);
1258         unix_state_unlock(tsk);
1259         return 0;
1260
1261 out:
1262         return err;
1263 }
1264
1265
1266 static int unix_getname(struct socket *sock, struct sockaddr *uaddr, int *uaddr_len, int peer)
1267 {
1268         struct sock *sk = sock->sk;
1269         struct unix_sock *u;
1270         struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
1271         int err = 0;
1272
1273         if (peer) {
1274                 sk = unix_peer_get(sk);
1275
1276                 err = -ENOTCONN;
1277                 if (!sk)
1278                         goto out;
1279                 err = 0;
1280         } else {
1281                 sock_hold(sk);
1282         }
1283
1284         u = unix_sk(sk);
1285         unix_state_lock(sk);
1286         if (!u->addr) {
1287                 sunaddr->sun_family = AF_UNIX;
1288                 sunaddr->sun_path[0] = 0;
1289                 *uaddr_len = sizeof(short);
1290         } else {
1291                 struct unix_address *addr = u->addr;
1292
1293                 *uaddr_len = addr->len;
1294                 memcpy(sunaddr, addr->name, *uaddr_len);
1295         }
1296         unix_state_unlock(sk);
1297         sock_put(sk);
1298 out:
1299         return err;
1300 }
1301
1302 static void unix_detach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1303 {
1304         int i;
1305
1306         scm->fp = UNIXCB(skb).fp;
1307         skb->destructor = sock_wfree;
1308         UNIXCB(skb).fp = NULL;
1309
1310         for (i = scm->fp->count-1; i >= 0; i--)
1311                 unix_notinflight(scm->fp->fp[i]);
1312 }
1313
1314 static void unix_destruct_fds(struct sk_buff *skb)
1315 {
1316         struct scm_cookie scm;
1317         memset(&scm, 0, sizeof(scm));
1318         unix_detach_fds(&scm, skb);
1319
1320         /* Alas, it calls VFS */
1321         /* So fscking what? fput() had been SMP-safe since the last Summer */
1322         scm_destroy(&scm);
1323         sock_wfree(skb);
1324 }
1325
1326 static int unix_attach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1327 {
1328         int i;
1329
1330         /*
1331          * Need to duplicate file references for the sake of garbage
1332          * collection.  Otherwise a socket in the fps might become a
1333          * candidate for GC while the skb is not yet queued.
1334          */
1335         UNIXCB(skb).fp = scm_fp_dup(scm->fp);
1336         if (!UNIXCB(skb).fp)
1337                 return -ENOMEM;
1338
1339         for (i = scm->fp->count-1; i >= 0; i--)
1340                 unix_inflight(scm->fp->fp[i]);
1341         skb->destructor = unix_destruct_fds;
1342         return 0;
1343 }
1344
1345 /*
1346  *      Send AF_UNIX data.
1347  */
1348
1349 static int unix_dgram_sendmsg(struct kiocb *kiocb, struct socket *sock,
1350                               struct msghdr *msg, size_t len)
1351 {
1352         struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1353         struct sock *sk = sock->sk;
1354         struct net *net = sock_net(sk);
1355         struct unix_sock *u = unix_sk(sk);
1356         struct sockaddr_un *sunaddr = msg->msg_name;
1357         struct sock *other = NULL;
1358         int namelen = 0; /* fake GCC */
1359         int err;
1360         unsigned hash;
1361         struct sk_buff *skb;
1362         long timeo;
1363         struct scm_cookie tmp_scm;
1364
1365         if (NULL == siocb->scm)
1366                 siocb->scm = &tmp_scm;
1367         wait_for_unix_gc();
1368         err = scm_send(sock, msg, siocb->scm);
1369         if (err < 0)
1370                 return err;
1371
1372         err = -EOPNOTSUPP;
1373         if (msg->msg_flags&MSG_OOB)
1374                 goto out;
1375
1376         if (msg->msg_namelen) {
1377                 err = unix_mkname(sunaddr, msg->msg_namelen, &hash);
1378                 if (err < 0)
1379                         goto out;
1380                 namelen = err;
1381         } else {
1382                 sunaddr = NULL;
1383                 err = -ENOTCONN;
1384                 other = unix_peer_get(sk);
1385                 if (!other)
1386                         goto out;
1387         }
1388
1389         if (test_bit(SOCK_PASSCRED, &sock->flags)
1390                 && !u->addr && (err = unix_autobind(sock)) != 0)
1391                 goto out;
1392
1393         err = -EMSGSIZE;
1394         if (len > sk->sk_sndbuf - 32)
1395                 goto out;
1396
1397         skb = sock_alloc_send_skb(sk, len, msg->msg_flags&MSG_DONTWAIT, &err);
1398         if (skb == NULL)
1399                 goto out;
1400
1401         memcpy(UNIXCREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1402         if (siocb->scm->fp) {
1403                 err = unix_attach_fds(siocb->scm, skb);
1404                 if (err)
1405                         goto out_free;
1406         }
1407         unix_get_secdata(siocb->scm, skb);
1408
1409         skb_reset_transport_header(skb);
1410         err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
1411         if (err)
1412                 goto out_free;
1413
1414         timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1415
1416 restart:
1417         if (!other) {
1418                 err = -ECONNRESET;
1419                 if (sunaddr == NULL)
1420                         goto out_free;
1421
1422                 other = unix_find_other(net, sunaddr, namelen, sk->sk_type,
1423                                         hash, &err);
1424                 if (other == NULL)
1425                         goto out_free;
1426         }
1427
1428         unix_state_lock(other);
1429         err = -EPERM;
1430         if (!unix_may_send(sk, other))
1431                 goto out_unlock;
1432
1433         if (sock_flag(other, SOCK_DEAD)) {
1434                 /*
1435                  *      Check with 1003.1g - what should
1436                  *      datagram error
1437                  */
1438                 unix_state_unlock(other);
1439                 sock_put(other);
1440
1441                 err = 0;
1442                 unix_state_lock(sk);
1443                 if (unix_peer(sk) == other) {
1444                         unix_peer(sk) = NULL;
1445                         unix_state_unlock(sk);
1446
1447                         unix_dgram_disconnected(sk, other);
1448                         sock_put(other);
1449                         err = -ECONNREFUSED;
1450                 } else {
1451                         unix_state_unlock(sk);
1452                 }
1453
1454                 other = NULL;
1455                 if (err)
1456                         goto out_free;
1457                 goto restart;
1458         }
1459
1460         err = -EPIPE;
1461         if (other->sk_shutdown & RCV_SHUTDOWN)
1462                 goto out_unlock;
1463
1464         if (sk->sk_type != SOCK_SEQPACKET) {
1465                 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1466                 if (err)
1467                         goto out_unlock;
1468         }
1469
1470         if (unix_peer(other) != sk && unix_recvq_full(other)) {
1471                 if (!timeo) {
1472                         err = -EAGAIN;
1473                         goto out_unlock;
1474                 }
1475
1476                 timeo = unix_wait_for_peer(other, timeo);
1477
1478                 err = sock_intr_errno(timeo);
1479                 if (signal_pending(current))
1480                         goto out_free;
1481
1482                 goto restart;
1483         }
1484
1485         skb_queue_tail(&other->sk_receive_queue, skb);
1486         unix_state_unlock(other);
1487         other->sk_data_ready(other, len);
1488         sock_put(other);
1489         scm_destroy(siocb->scm);
1490         return len;
1491
1492 out_unlock:
1493         unix_state_unlock(other);
1494 out_free:
1495         kfree_skb(skb);
1496 out:
1497         if (other)
1498                 sock_put(other);
1499         scm_destroy(siocb->scm);
1500         return err;
1501 }
1502
1503
1504 static int unix_stream_sendmsg(struct kiocb *kiocb, struct socket *sock,
1505                                struct msghdr *msg, size_t len)
1506 {
1507         struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1508         struct sock *sk = sock->sk;
1509         struct sock *other = NULL;
1510         struct sockaddr_un *sunaddr = msg->msg_name;
1511         int err, size;
1512         struct sk_buff *skb;
1513         int sent = 0;
1514         struct scm_cookie tmp_scm;
1515         bool fds_sent = false;
1516
1517         if (NULL == siocb->scm)
1518                 siocb->scm = &tmp_scm;
1519         wait_for_unix_gc();
1520         err = scm_send(sock, msg, siocb->scm);
1521         if (err < 0)
1522                 return err;
1523
1524         err = -EOPNOTSUPP;
1525         if (msg->msg_flags&MSG_OOB)
1526                 goto out_err;
1527
1528         if (msg->msg_namelen) {
1529                 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
1530                 goto out_err;
1531         } else {
1532                 sunaddr = NULL;
1533                 err = -ENOTCONN;
1534                 other = unix_peer(sk);
1535                 if (!other)
1536                         goto out_err;
1537         }
1538
1539         if (sk->sk_shutdown & SEND_SHUTDOWN)
1540                 goto pipe_err;
1541
1542         while (sent < len) {
1543                 /*
1544                  *      Optimisation for the fact that under 0.01% of X
1545                  *      messages typically need breaking up.
1546                  */
1547
1548                 size = len-sent;
1549
1550                 /* Keep two messages in the pipe so it schedules better */
1551                 if (size > ((sk->sk_sndbuf >> 1) - 64))
1552                         size = (sk->sk_sndbuf >> 1) - 64;
1553
1554                 if (size > SKB_MAX_ALLOC)
1555                         size = SKB_MAX_ALLOC;
1556
1557                 /*
1558                  *      Grab a buffer
1559                  */
1560
1561                 skb = sock_alloc_send_skb(sk, size, msg->msg_flags&MSG_DONTWAIT,
1562                                           &err);
1563
1564                 if (skb == NULL)
1565                         goto out_err;
1566
1567                 /*
1568                  *      If you pass two values to the sock_alloc_send_skb
1569                  *      it tries to grab the large buffer with GFP_NOFS
1570                  *      (which can fail easily), and if it fails grab the
1571                  *      fallback size buffer which is under a page and will
1572                  *      succeed. [Alan]
1573                  */
1574                 size = min_t(int, size, skb_tailroom(skb));
1575
1576                 memcpy(UNIXCREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1577                 /* Only send the fds in the first buffer */
1578                 if (siocb->scm->fp && !fds_sent) {
1579                         err = unix_attach_fds(siocb->scm, skb);
1580                         if (err) {
1581                                 kfree_skb(skb);
1582                                 goto out_err;
1583                         }
1584                         fds_sent = true;
1585                 }
1586
1587                 err = memcpy_fromiovec(skb_put(skb, size), msg->msg_iov, size);
1588                 if (err) {
1589                         kfree_skb(skb);
1590                         goto out_err;
1591                 }
1592
1593                 unix_state_lock(other);
1594
1595                 if (sock_flag(other, SOCK_DEAD) ||
1596                     (other->sk_shutdown & RCV_SHUTDOWN))
1597                         goto pipe_err_free;
1598
1599                 skb_queue_tail(&other->sk_receive_queue, skb);
1600                 unix_state_unlock(other);
1601                 other->sk_data_ready(other, size);
1602                 sent += size;
1603         }
1604
1605         scm_destroy(siocb->scm);
1606         siocb->scm = NULL;
1607
1608         return sent;
1609
1610 pipe_err_free:
1611         unix_state_unlock(other);
1612         kfree_skb(skb);
1613 pipe_err:
1614         if (sent == 0 && !(msg->msg_flags&MSG_NOSIGNAL))
1615                 send_sig(SIGPIPE, current, 0);
1616         err = -EPIPE;
1617 out_err:
1618         scm_destroy(siocb->scm);
1619         siocb->scm = NULL;
1620         return sent ? : err;
1621 }
1622
1623 static int unix_seqpacket_sendmsg(struct kiocb *kiocb, struct socket *sock,
1624                                   struct msghdr *msg, size_t len)
1625 {
1626         int err;
1627         struct sock *sk = sock->sk;
1628
1629         err = sock_error(sk);
1630         if (err)
1631                 return err;
1632
1633         if (sk->sk_state != TCP_ESTABLISHED)
1634                 return -ENOTCONN;
1635
1636         if (msg->msg_namelen)
1637                 msg->msg_namelen = 0;
1638
1639         return unix_dgram_sendmsg(kiocb, sock, msg, len);
1640 }
1641
1642 static void unix_copy_addr(struct msghdr *msg, struct sock *sk)
1643 {
1644         struct unix_sock *u = unix_sk(sk);
1645
1646         msg->msg_namelen = 0;
1647         if (u->addr) {
1648                 msg->msg_namelen = u->addr->len;
1649                 memcpy(msg->msg_name, u->addr->name, u->addr->len);
1650         }
1651 }
1652
1653 static int unix_dgram_recvmsg(struct kiocb *iocb, struct socket *sock,
1654                               struct msghdr *msg, size_t size,
1655                               int flags)
1656 {
1657         struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1658         struct scm_cookie tmp_scm;
1659         struct sock *sk = sock->sk;
1660         struct unix_sock *u = unix_sk(sk);
1661         int noblock = flags & MSG_DONTWAIT;
1662         struct sk_buff *skb;
1663         int err;
1664
1665         err = -EOPNOTSUPP;
1666         if (flags&MSG_OOB)
1667                 goto out;
1668
1669         msg->msg_namelen = 0;
1670
1671         mutex_lock(&u->readlock);
1672
1673         skb = skb_recv_datagram(sk, flags, noblock, &err);
1674         if (!skb) {
1675                 unix_state_lock(sk);
1676                 /* Signal EOF on disconnected non-blocking SEQPACKET socket. */
1677                 if (sk->sk_type == SOCK_SEQPACKET && err == -EAGAIN &&
1678                     (sk->sk_shutdown & RCV_SHUTDOWN))
1679                         err = 0;
1680                 unix_state_unlock(sk);
1681                 goto out_unlock;
1682         }
1683
1684         wake_up_interruptible_sync(&u->peer_wait);
1685
1686         if (msg->msg_name)
1687                 unix_copy_addr(msg, skb->sk);
1688
1689         if (size > skb->len)
1690                 size = skb->len;
1691         else if (size < skb->len)
1692                 msg->msg_flags |= MSG_TRUNC;
1693
1694         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, size);
1695         if (err)
1696                 goto out_free;
1697
1698         if (!siocb->scm) {
1699                 siocb->scm = &tmp_scm;
1700                 memset(&tmp_scm, 0, sizeof(tmp_scm));
1701         }
1702         siocb->scm->creds = *UNIXCREDS(skb);
1703         unix_set_secdata(siocb->scm, skb);
1704
1705         if (!(flags & MSG_PEEK)) {
1706                 if (UNIXCB(skb).fp)
1707                         unix_detach_fds(siocb->scm, skb);
1708         } else {
1709                 /* It is questionable: on PEEK we could:
1710                    - do not return fds - good, but too simple 8)
1711                    - return fds, and do not return them on read (old strategy,
1712                      apparently wrong)
1713                    - clone fds (I chose it for now, it is the most universal
1714                      solution)
1715
1716                    POSIX 1003.1g does not actually define this clearly
1717                    at all. POSIX 1003.1g doesn't define a lot of things
1718                    clearly however!
1719
1720                 */
1721                 if (UNIXCB(skb).fp)
1722                         siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1723         }
1724         err = size;
1725
1726         scm_recv(sock, msg, siocb->scm, flags);
1727
1728 out_free:
1729         skb_free_datagram(sk, skb);
1730 out_unlock:
1731         mutex_unlock(&u->readlock);
1732 out:
1733         return err;
1734 }
1735
1736 /*
1737  *      Sleep until data has arrive. But check for races..
1738  */
1739
1740 static long unix_stream_data_wait(struct sock *sk, long timeo)
1741 {
1742         DEFINE_WAIT(wait);
1743
1744         unix_state_lock(sk);
1745
1746         for (;;) {
1747                 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1748
1749                 if (!skb_queue_empty(&sk->sk_receive_queue) ||
1750                     sk->sk_err ||
1751                     (sk->sk_shutdown & RCV_SHUTDOWN) ||
1752                     signal_pending(current) ||
1753                     !timeo)
1754                         break;
1755
1756                 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1757                 unix_state_unlock(sk);
1758                 timeo = schedule_timeout(timeo);
1759                 unix_state_lock(sk);
1760                 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1761         }
1762
1763         finish_wait(sk->sk_sleep, &wait);
1764         unix_state_unlock(sk);
1765         return timeo;
1766 }
1767
1768
1769
1770 static int unix_stream_recvmsg(struct kiocb *iocb, struct socket *sock,
1771                                struct msghdr *msg, size_t size,
1772                                int flags)
1773 {
1774         struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1775         struct scm_cookie tmp_scm;
1776         struct sock *sk = sock->sk;
1777         struct unix_sock *u = unix_sk(sk);
1778         struct sockaddr_un *sunaddr = msg->msg_name;
1779         int copied = 0;
1780         int check_creds = 0;
1781         int target;
1782         int err = 0;
1783         long timeo;
1784
1785         err = -EINVAL;
1786         if (sk->sk_state != TCP_ESTABLISHED)
1787                 goto out;
1788
1789         err = -EOPNOTSUPP;
1790         if (flags&MSG_OOB)
1791                 goto out;
1792
1793         target = sock_rcvlowat(sk, flags&MSG_WAITALL, size);
1794         timeo = sock_rcvtimeo(sk, flags&MSG_DONTWAIT);
1795
1796         msg->msg_namelen = 0;
1797
1798         /* Lock the socket to prevent queue disordering
1799          * while sleeps in memcpy_tomsg
1800          */
1801
1802         if (!siocb->scm) {
1803                 siocb->scm = &tmp_scm;
1804                 memset(&tmp_scm, 0, sizeof(tmp_scm));
1805         }
1806
1807         mutex_lock(&u->readlock);
1808
1809         do {
1810                 int chunk;
1811                 struct sk_buff *skb;
1812
1813                 unix_state_lock(sk);
1814                 skb = skb_dequeue(&sk->sk_receive_queue);
1815                 if (skb == NULL) {
1816                         if (copied >= target)
1817                                 goto unlock;
1818
1819                         /*
1820                          *      POSIX 1003.1g mandates this order.
1821                          */
1822
1823                         err = sock_error(sk);
1824                         if (err)
1825                                 goto unlock;
1826                         if (sk->sk_shutdown & RCV_SHUTDOWN)
1827                                 goto unlock;
1828
1829                         unix_state_unlock(sk);
1830                         err = -EAGAIN;
1831                         if (!timeo)
1832                                 break;
1833                         mutex_unlock(&u->readlock);
1834
1835                         timeo = unix_stream_data_wait(sk, timeo);
1836
1837                         if (signal_pending(current)) {
1838                                 err = sock_intr_errno(timeo);
1839                                 goto out;
1840                         }
1841                         mutex_lock(&u->readlock);
1842                         continue;
1843  unlock:
1844                         unix_state_unlock(sk);
1845                         break;
1846                 }
1847                 unix_state_unlock(sk);
1848
1849                 if (check_creds) {
1850                         /* Never glue messages from different writers */
1851                         if (memcmp(UNIXCREDS(skb), &siocb->scm->creds,
1852                                    sizeof(siocb->scm->creds)) != 0) {
1853                                 skb_queue_head(&sk->sk_receive_queue, skb);
1854                                 break;
1855                         }
1856                 } else {
1857                         /* Copy credentials */
1858                         siocb->scm->creds = *UNIXCREDS(skb);
1859                         check_creds = 1;
1860                 }
1861
1862                 /* Copy address just once */
1863                 if (sunaddr) {
1864                         unix_copy_addr(msg, skb->sk);
1865                         sunaddr = NULL;
1866                 }
1867
1868                 chunk = min_t(unsigned int, skb->len, size);
1869                 if (memcpy_toiovec(msg->msg_iov, skb->data, chunk)) {
1870                         skb_queue_head(&sk->sk_receive_queue, skb);
1871                         if (copied == 0)
1872                                 copied = -EFAULT;
1873                         break;
1874                 }
1875                 copied += chunk;
1876                 size -= chunk;
1877
1878                 /* Mark read part of skb as used */
1879                 if (!(flags & MSG_PEEK)) {
1880                         skb_pull(skb, chunk);
1881
1882                         if (UNIXCB(skb).fp)
1883                                 unix_detach_fds(siocb->scm, skb);
1884
1885                         /* put the skb back if we didn't use it up.. */
1886                         if (skb->len) {
1887                                 skb_queue_head(&sk->sk_receive_queue, skb);
1888                                 break;
1889                         }
1890
1891                         kfree_skb(skb);
1892
1893                         if (siocb->scm->fp)
1894                                 break;
1895                 } else {
1896                         /* It is questionable, see note in unix_dgram_recvmsg.
1897                          */
1898                         if (UNIXCB(skb).fp)
1899                                 siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1900
1901                         /* put message back and return */
1902                         skb_queue_head(&sk->sk_receive_queue, skb);
1903                         break;
1904                 }
1905         } while (size);
1906
1907         mutex_unlock(&u->readlock);
1908         scm_recv(sock, msg, siocb->scm, flags);
1909 out:
1910         return copied ? : err;
1911 }
1912
1913 static int unix_shutdown(struct socket *sock, int mode)
1914 {
1915         struct sock *sk = sock->sk;
1916         struct sock *other;
1917
1918         mode = (mode+1)&(RCV_SHUTDOWN|SEND_SHUTDOWN);
1919
1920         if (mode) {
1921                 unix_state_lock(sk);
1922                 sk->sk_shutdown |= mode;
1923                 other = unix_peer(sk);
1924                 if (other)
1925                         sock_hold(other);
1926                 unix_state_unlock(sk);
1927                 sk->sk_state_change(sk);
1928
1929                 if (other &&
1930                         (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET)) {
1931
1932                         int peer_mode = 0;
1933
1934                         if (mode&RCV_SHUTDOWN)
1935                                 peer_mode |= SEND_SHUTDOWN;
1936                         if (mode&SEND_SHUTDOWN)
1937                                 peer_mode |= RCV_SHUTDOWN;
1938                         unix_state_lock(other);
1939                         other->sk_shutdown |= peer_mode;
1940                         unix_state_unlock(other);
1941                         other->sk_state_change(other);
1942                         read_lock(&other->sk_callback_lock);
1943                         if (peer_mode == SHUTDOWN_MASK)
1944                                 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_HUP);
1945                         else if (peer_mode & RCV_SHUTDOWN)
1946                                 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_IN);
1947                         read_unlock(&other->sk_callback_lock);
1948                 }
1949                 if (other)
1950                         sock_put(other);
1951         }
1952         return 0;
1953 }
1954
1955 static int unix_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1956 {
1957         struct sock *sk = sock->sk;
1958         long amount = 0;
1959         int err;
1960
1961         switch (cmd) {
1962         case SIOCOUTQ:
1963                 amount = sk_wmem_alloc_get(sk);
1964                 err = put_user(amount, (int __user *)arg);
1965                 break;
1966         case SIOCINQ:
1967                 {
1968                         struct sk_buff *skb;
1969
1970                         if (sk->sk_state == TCP_LISTEN) {
1971                                 err = -EINVAL;
1972                                 break;
1973                         }
1974
1975                         spin_lock(&sk->sk_receive_queue.lock);
1976                         if (sk->sk_type == SOCK_STREAM ||
1977                             sk->sk_type == SOCK_SEQPACKET) {
1978                                 skb_queue_walk(&sk->sk_receive_queue, skb)
1979                                         amount += skb->len;
1980                         } else {
1981                                 skb = skb_peek(&sk->sk_receive_queue);
1982                                 if (skb)
1983                                         amount = skb->len;
1984                         }
1985                         spin_unlock(&sk->sk_receive_queue.lock);
1986                         err = put_user(amount, (int __user *)arg);
1987                         break;
1988                 }
1989
1990         default:
1991                 err = -ENOIOCTLCMD;
1992                 break;
1993         }
1994         return err;
1995 }
1996
1997 static unsigned int unix_poll(struct file *file, struct socket *sock, poll_table *wait)
1998 {
1999         struct sock *sk = sock->sk;
2000         unsigned int mask;
2001
2002         sock_poll_wait(file, sk->sk_sleep, wait);
2003         mask = 0;
2004
2005         /* exceptional events? */
2006         if (sk->sk_err)
2007                 mask |= POLLERR;
2008         if (sk->sk_shutdown == SHUTDOWN_MASK)
2009                 mask |= POLLHUP;
2010         if (sk->sk_shutdown & RCV_SHUTDOWN)
2011                 mask |= POLLRDHUP;
2012
2013         /* readable? */
2014         if (!skb_queue_empty(&sk->sk_receive_queue) ||
2015             (sk->sk_shutdown & RCV_SHUTDOWN))
2016                 mask |= POLLIN | POLLRDNORM;
2017
2018         /* Connection-based need to check for termination and startup */
2019         if ((sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) &&
2020             sk->sk_state == TCP_CLOSE)
2021                 mask |= POLLHUP;
2022
2023         /*
2024          * we set writable also when the other side has shut down the
2025          * connection. This prevents stuck sockets.
2026          */
2027         if (unix_writable(sk))
2028                 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2029
2030         return mask;
2031 }
2032
2033 static unsigned int unix_dgram_poll(struct file *file, struct socket *sock,
2034                                     poll_table *wait)
2035 {
2036         struct sock *sk = sock->sk, *other;
2037         unsigned int mask, writable;
2038
2039         sock_poll_wait(file, sk->sk_sleep, wait);
2040         mask = 0;
2041
2042         /* exceptional events? */
2043         if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue))
2044                 mask |= POLLERR;
2045         if (sk->sk_shutdown & RCV_SHUTDOWN)
2046                 mask |= POLLRDHUP;
2047         if (sk->sk_shutdown == SHUTDOWN_MASK)
2048                 mask |= POLLHUP;
2049
2050         /* readable? */
2051         if (!skb_queue_empty(&sk->sk_receive_queue) ||
2052             (sk->sk_shutdown & RCV_SHUTDOWN))
2053                 mask |= POLLIN | POLLRDNORM;
2054
2055         /* Connection-based need to check for termination and startup */
2056         if (sk->sk_type == SOCK_SEQPACKET) {
2057                 if (sk->sk_state == TCP_CLOSE)
2058                         mask |= POLLHUP;
2059                 /* connection hasn't started yet? */
2060                 if (sk->sk_state == TCP_SYN_SENT)
2061                         return mask;
2062         }
2063
2064         /* writable? */
2065         writable = unix_writable(sk);
2066         if (writable) {
2067                 other = unix_peer_get(sk);
2068                 if (other) {
2069                         if (unix_peer(other) != sk) {
2070                                 sock_poll_wait(file, &unix_sk(other)->peer_wait,
2071                                           wait);
2072                                 if (unix_recvq_full(other))
2073                                         writable = 0;
2074                         }
2075
2076                         sock_put(other);
2077                 }
2078         }
2079
2080         if (writable)
2081                 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2082         else
2083                 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
2084
2085         return mask;
2086 }
2087
2088 #ifdef CONFIG_PROC_FS
2089 static struct sock *first_unix_socket(int *i)
2090 {
2091         for (*i = 0; *i <= UNIX_HASH_SIZE; (*i)++) {
2092                 if (!hlist_empty(&unix_socket_table[*i]))
2093                         return __sk_head(&unix_socket_table[*i]);
2094         }
2095         return NULL;
2096 }
2097
2098 static struct sock *next_unix_socket(int *i, struct sock *s)
2099 {
2100         struct sock *next = sk_next(s);
2101         /* More in this chain? */
2102         if (next)
2103                 return next;
2104         /* Look for next non-empty chain. */
2105         for ((*i)++; *i <= UNIX_HASH_SIZE; (*i)++) {
2106                 if (!hlist_empty(&unix_socket_table[*i]))
2107                         return __sk_head(&unix_socket_table[*i]);
2108         }
2109         return NULL;
2110 }
2111
2112 struct unix_iter_state {
2113         struct seq_net_private p;
2114         int i;
2115 };
2116
2117 static struct sock *unix_seq_idx(struct seq_file *seq, loff_t pos)
2118 {
2119         struct unix_iter_state *iter = seq->private;
2120         loff_t off = 0;
2121         struct sock *s;
2122
2123         for (s = first_unix_socket(&iter->i); s; s = next_unix_socket(&iter->i, s)) {
2124                 if (sock_net(s) != seq_file_net(seq))
2125                         continue;
2126                 if (off == pos)
2127                         return s;
2128                 ++off;
2129         }
2130         return NULL;
2131 }
2132
2133 static void *unix_seq_start(struct seq_file *seq, loff_t *pos)
2134         __acquires(unix_table_lock)
2135 {
2136         spin_lock(&unix_table_lock);
2137         return *pos ? unix_seq_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2138 }
2139
2140 static void *unix_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2141 {
2142         struct unix_iter_state *iter = seq->private;
2143         struct sock *sk = v;
2144         ++*pos;
2145
2146         if (v == SEQ_START_TOKEN)
2147                 sk = first_unix_socket(&iter->i);
2148         else
2149                 sk = next_unix_socket(&iter->i, sk);
2150         while (sk && (sock_net(sk) != seq_file_net(seq)))
2151                 sk = next_unix_socket(&iter->i, sk);
2152         return sk;
2153 }
2154
2155 static void unix_seq_stop(struct seq_file *seq, void *v)
2156         __releases(unix_table_lock)
2157 {
2158         spin_unlock(&unix_table_lock);
2159 }
2160
2161 static int unix_seq_show(struct seq_file *seq, void *v)
2162 {
2163
2164         if (v == SEQ_START_TOKEN)
2165                 seq_puts(seq, "Num       RefCount Protocol Flags    Type St "
2166                          "Inode Path\n");
2167         else {
2168                 struct sock *s = v;
2169                 struct unix_sock *u = unix_sk(s);
2170                 unix_state_lock(s);
2171
2172                 seq_printf(seq, "%p: %08X %08X %08X %04X %02X %5lu",
2173                         s,
2174                         atomic_read(&s->sk_refcnt),
2175                         0,
2176                         s->sk_state == TCP_LISTEN ? __SO_ACCEPTCON : 0,
2177                         s->sk_type,
2178                         s->sk_socket ?
2179                         (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTED : SS_UNCONNECTED) :
2180                         (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTING : SS_DISCONNECTING),
2181                         sock_i_ino(s));
2182
2183                 if (u->addr) {
2184                         int i, len;
2185                         seq_putc(seq, ' ');
2186
2187                         i = 0;
2188                         len = u->addr->len - sizeof(short);
2189                         if (!UNIX_ABSTRACT(s))
2190                                 len--;
2191                         else {
2192                                 seq_putc(seq, '@');
2193                                 i++;
2194                         }
2195                         for ( ; i < len; i++)
2196                                 seq_putc(seq, u->addr->name->sun_path[i]);
2197                 }
2198                 unix_state_unlock(s);
2199                 seq_putc(seq, '\n');
2200         }
2201
2202         return 0;
2203 }
2204
2205 static const struct seq_operations unix_seq_ops = {
2206         .start  = unix_seq_start,
2207         .next   = unix_seq_next,
2208         .stop   = unix_seq_stop,
2209         .show   = unix_seq_show,
2210 };
2211
2212 static int unix_seq_open(struct inode *inode, struct file *file)
2213 {
2214         return seq_open_net(inode, file, &unix_seq_ops,
2215                             sizeof(struct unix_iter_state));
2216 }
2217
2218 static const struct file_operations unix_seq_fops = {
2219         .owner          = THIS_MODULE,
2220         .open           = unix_seq_open,
2221         .read           = seq_read,
2222         .llseek         = seq_lseek,
2223         .release        = seq_release_net,
2224 };
2225
2226 #endif
2227
2228 static struct net_proto_family unix_family_ops = {
2229         .family = PF_UNIX,
2230         .create = unix_create,
2231         .owner  = THIS_MODULE,
2232 };
2233
2234
2235 static int unix_net_init(struct net *net)
2236 {
2237         int error = -ENOMEM;
2238
2239         net->unx.sysctl_max_dgram_qlen = 10;
2240         if (unix_sysctl_register(net))
2241                 goto out;
2242
2243 #ifdef CONFIG_PROC_FS
2244         if (!proc_net_fops_create(net, "unix", 0, &unix_seq_fops)) {
2245                 unix_sysctl_unregister(net);
2246                 goto out;
2247         }
2248 #endif
2249         error = 0;
2250 out:
2251         return error;
2252 }
2253
2254 static void unix_net_exit(struct net *net)
2255 {
2256         unix_sysctl_unregister(net);
2257         proc_net_remove(net, "unix");
2258 }
2259
2260 static struct pernet_operations unix_net_ops = {
2261         .init = unix_net_init,
2262         .exit = unix_net_exit,
2263 };
2264
2265 static int __init af_unix_init(void)
2266 {
2267         int rc = -1;
2268         struct sk_buff *dummy_skb;
2269
2270         BUILD_BUG_ON(sizeof(struct unix_skb_parms) > sizeof(dummy_skb->cb));
2271
2272         rc = proto_register(&unix_proto, 1);
2273         if (rc != 0) {
2274                 printk(KERN_CRIT "%s: Cannot create unix_sock SLAB cache!\n",
2275                        __func__);
2276                 goto out;
2277         }
2278
2279         sock_register(&unix_family_ops);
2280         register_pernet_subsys(&unix_net_ops);
2281 out:
2282         return rc;
2283 }
2284
2285 static void __exit af_unix_exit(void)
2286 {
2287         sock_unregister(PF_UNIX);
2288         proto_unregister(&unix_proto);
2289         unregister_pernet_subsys(&unix_net_ops);
2290 }
2291
2292 /* Earlier than device_initcall() so that other drivers invoking
2293    request_module() don't end up in a loop when modprobe tries
2294    to use a UNIX socket. But later than subsys_initcall() because
2295    we depend on stuff initialised there */
2296 fs_initcall(af_unix_init);
2297 module_exit(af_unix_exit);
2298
2299 MODULE_LICENSE("GPL");
2300 MODULE_ALIAS_NETPROTO(PF_UNIX);