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