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

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