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 spin lock.
 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 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 (dentry && dentry->d_inode == i) {
 296                         sock_hold(s);
 297                         goto found;
 298                 }
 299         }
 300         s = NULL;
 301 found:
 302         spin_unlock(&unix_table_lock);
 303         return s;
 304 }
 305
 306 static inline int unix_writable(struct sock *sk)
 307 {
 308         return (atomic_read(&sk->sk_wmem_alloc) << 2) <= sk->sk_sndbuf;
 309 }
 310
 311 static void unix_write_space(struct sock *sk)
 312 {
 313         struct socket_wq *wq;
 314
 315         rcu_read_lock();
 316         if (unix_writable(sk)) {
 317                 wq = rcu_dereference(sk->sk_wq);
 318                 if (wq_has_sleeper(wq))
 319                         wake_up_interruptible_sync(&wq->wait);
 320                 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
 321         }
 322         rcu_read_unlock();
 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                         sk_wake_async(skpair, SOCK_WAKE_WAITD, POLL_HUP);
 410                 }
 411                 sock_put(skpair); /* It may now die */
 412                 unix_peer(sk) = NULL;
 413         }
 414
 415         /* Try to flush out this socket. Throw out buffers at least */
 416
 417         while ((skb = skb_dequeue(&sk->sk_receive_queue)) != NULL) {
 418                 if (state == TCP_LISTEN)
 419                         unix_release_sock(skb->sk, 1);
 420                 /* passed fds are erased in the kfree_skb hook        */
 421                 kfree_skb(skb);
 422         }
 423
 424         if (dentry) {
 425                 dput(dentry);
 426                 mntput(mnt);
 427         }
 428
 429         sock_put(sk);
 430
 431         /* ---- Socket is dead now and most probably destroyed ---- */
 432
 433         /*
 434          * Fixme: BSD difference: In BSD all sockets connected to use get
 435          *        ECONNRESET and we die on the spot. In Linux we behave
 436          *        like files and pipes do and wait for the last
 437          *        dereference.
 438          *
 439          * Can't we simply set sock->err?
 440          *
 441          *        What the above comment does talk about? --ANK(980817)
 442          */
 443
 444         if (unix_tot_inflight)
 445                 unix_gc();              /* Garbage collect fds */
 446
 447         return 0;
 448 }
 449
 450 static void init_peercred(struct sock *sk)
 451 {
 452         put_pid(sk->sk_peer_pid);
 453         if (sk->sk_peer_cred)
 454                 put_cred(sk->sk_peer_cred);
 455         sk->sk_peer_pid  = get_pid(task_tgid(current));
 456         sk->sk_peer_cred = get_current_cred();
 457 }
 458
 459 static void copy_peercred(struct sock *sk, struct sock *peersk)
 460 {
 461         put_pid(sk->sk_peer_pid);
 462         if (sk->sk_peer_cred)
 463                 put_cred(sk->sk_peer_cred);
 464         sk->sk_peer_pid  = get_pid(peersk->sk_peer_pid);
 465         sk->sk_peer_cred = get_cred(peersk->sk_peer_cred);
 466 }
 467
 468 static int unix_listen(struct socket *sock, int backlog)
 469 {
 470         int err;
 471         struct sock *sk = sock->sk;
 472         struct unix_sock *u = unix_sk(sk);
 473         struct pid *old_pid = NULL;
 474         const struct cred *old_cred = NULL;
 475
 476         err = -EOPNOTSUPP;
 477         if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
 478                 goto out;       /* Only stream/seqpacket sockets accept */
 479         err = -EINVAL;
 480         if (!u->addr)
 481                 goto out;       /* No listens on an unbound socket */
 482         unix_state_lock(sk);
 483         if (sk->sk_state != TCP_CLOSE && sk->sk_state != TCP_LISTEN)
 484                 goto out_unlock;
 485         if (backlog > sk->sk_max_ack_backlog)
 486                 wake_up_interruptible_all(&u->peer_wait);
 487         sk->sk_max_ack_backlog  = backlog;
 488         sk->sk_state            = TCP_LISTEN;
 489         /* set credentials so connect can copy them */
 490         init_peercred(sk);
 491         err = 0;
 492
 493 out_unlock:
 494         unix_state_unlock(sk);
 495         put_pid(old_pid);
 496         if (old_cred)
 497                 put_cred(old_cred);
 498 out:
 499         return err;
 500 }
 501
 502 static int unix_release(struct socket *);
 503 static int unix_bind(struct socket *, struct sockaddr *, int);
 504 static int unix_stream_connect(struct socket *, struct sockaddr *,
 505                                int addr_len, int flags);
 506 static int unix_socketpair(struct socket *, struct socket *);
 507 static int unix_accept(struct socket *, struct socket *, int);
 508 static int unix_getname(struct socket *, struct sockaddr *, int *, int);
 509 static unsigned int unix_poll(struct file *, struct socket *, poll_table *);
 510 static unsigned int unix_dgram_poll(struct file *, struct socket *,
 511                                     poll_table *);
 512 static int unix_ioctl(struct socket *, unsigned int, unsigned long);
 513 static int unix_shutdown(struct socket *, int);
 514 static int unix_stream_sendmsg(struct kiocb *, struct socket *,
 515                                struct msghdr *, size_t);
 516 static int unix_stream_recvmsg(struct kiocb *, struct socket *,
 517                                struct msghdr *, size_t, int);
 518 static int unix_dgram_sendmsg(struct kiocb *, struct socket *,
 519                               struct msghdr *, size_t);
 520 static int unix_dgram_recvmsg(struct kiocb *, struct socket *,
 521                               struct msghdr *, size_t, int);
 522 static int unix_dgram_connect(struct socket *, struct sockaddr *,
 523                               int, int);
 524 static int unix_seqpacket_sendmsg(struct kiocb *, struct socket *,
 525                                   struct msghdr *, size_t);
 526
 527 static const struct proto_ops unix_stream_ops = {
 528         .family =       PF_UNIX,
 529         .owner =        THIS_MODULE,
 530         .release =      unix_release,
 531         .bind =         unix_bind,
 532         .connect =      unix_stream_connect,
 533         .socketpair =   unix_socketpair,
 534         .accept =       unix_accept,
 535         .getname =      unix_getname,
 536         .poll =         unix_poll,
 537         .ioctl =        unix_ioctl,
 538         .listen =       unix_listen,
 539         .shutdown =     unix_shutdown,
 540         .setsockopt =   sock_no_setsockopt,
 541         .getsockopt =   sock_no_getsockopt,
 542         .sendmsg =      unix_stream_sendmsg,
 543         .recvmsg =      unix_stream_recvmsg,
 544         .mmap =         sock_no_mmap,
 545         .sendpage =     sock_no_sendpage,
 546 };
 547
 548 static const struct proto_ops unix_dgram_ops = {
 549         .family =       PF_UNIX,
 550         .owner =        THIS_MODULE,
 551         .release =      unix_release,
 552         .bind =         unix_bind,
 553         .connect =      unix_dgram_connect,
 554         .socketpair =   unix_socketpair,
 555         .accept =       sock_no_accept,
 556         .getname =      unix_getname,
 557         .poll =         unix_dgram_poll,
 558         .ioctl =        unix_ioctl,
 559         .listen =       sock_no_listen,
 560         .shutdown =     unix_shutdown,
 561         .setsockopt =   sock_no_setsockopt,
 562         .getsockopt =   sock_no_getsockopt,
 563         .sendmsg =      unix_dgram_sendmsg,
 564         .recvmsg =      unix_dgram_recvmsg,
 565         .mmap =         sock_no_mmap,
 566         .sendpage =     sock_no_sendpage,
 567 };
 568
 569 static const struct proto_ops unix_seqpacket_ops = {
 570         .family =       PF_UNIX,
 571         .owner =        THIS_MODULE,
 572         .release =      unix_release,
 573         .bind =         unix_bind,
 574         .connect =      unix_stream_connect,
 575         .socketpair =   unix_socketpair,
 576         .accept =       unix_accept,
 577         .getname =      unix_getname,
 578         .poll =         unix_dgram_poll,
 579         .ioctl =        unix_ioctl,
 580         .listen =       unix_listen,
 581         .shutdown =     unix_shutdown,
 582         .setsockopt =   sock_no_setsockopt,
 583         .getsockopt =   sock_no_getsockopt,
 584         .sendmsg =      unix_seqpacket_sendmsg,
 585         .recvmsg =      unix_dgram_recvmsg,
 586         .mmap =         sock_no_mmap,
 587         .sendpage =     sock_no_sendpage,
 588 };
 589
 590 static struct proto unix_proto = {
 591         .name                   = "UNIX",
 592         .owner                  = THIS_MODULE,
 593         .obj_size               = sizeof(struct unix_sock),
 594 };
 595
 596 /*
 597  * AF_UNIX sockets do not interact with hardware, hence they
 598  * dont trigger interrupts - so it's safe for them to have
 599  * bh-unsafe locking for their sk_receive_queue.lock. Split off
 600  * this special lock-class by reinitializing the spinlock key:
 601  */
 602 static struct lock_class_key af_unix_sk_receive_queue_lock_key;
 603
 604 static struct sock *unix_create1(struct net *net, struct socket *sock)
 605 {
 606         struct sock *sk = NULL;
 607         struct unix_sock *u;
 608
 609         atomic_inc(&unix_nr_socks);
 610         if (atomic_read(&unix_nr_socks) > 2 * get_max_files())
 611                 goto out;
 612
 613         sk = sk_alloc(net, PF_UNIX, GFP_KERNEL, &unix_proto);
 614         if (!sk)
 615                 goto out;
 616
 617         sock_init_data(sock, sk);
 618         lockdep_set_class(&sk->sk_receive_queue.lock,
 619                                 &af_unix_sk_receive_queue_lock_key);
 620
 621         sk->sk_write_space      = unix_write_space;
 622         sk->sk_max_ack_backlog  = net->unx.sysctl_max_dgram_qlen;
 623         sk->sk_destruct         = unix_sock_destructor;
 624         u         = unix_sk(sk);
 625         u->dentry = NULL;
 626         u->mnt    = NULL;
 627         spin_lock_init(&u->lock);
 628         atomic_long_set(&u->inflight, 0);
 629         INIT_LIST_HEAD(&u->link);
 630         mutex_init(&u->readlock); /* single task reading lock */
 631         init_waitqueue_head(&u->peer_wait);
 632         unix_insert_socket(unix_sockets_unbound, sk);
 633 out:
 634         if (sk == NULL)
 635                 atomic_dec(&unix_nr_socks);
 636         else {
 637                 local_bh_disable();
 638                 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
 639                 local_bh_enable();
 640         }
 641         return sk;
 642 }
 643
 644 static int unix_create(struct net *net, struct socket *sock, int protocol,
 645                        int kern)
 646 {
 647         if (protocol && protocol != PF_UNIX)
 648                 return -EPROTONOSUPPORT;
 649
 650         sock->state = SS_UNCONNECTED;
 651
 652         switch (sock->type) {
 653         case SOCK_STREAM:
 654                 sock->ops = &unix_stream_ops;
 655                 break;
 656                 /*
 657                  *      Believe it or not BSD has AF_UNIX, SOCK_RAW though
 658                  *      nothing uses it.
 659                  */
 660         case SOCK_RAW:
 661                 sock->type = SOCK_DGRAM;
 662         case SOCK_DGRAM:
 663                 sock->ops = &unix_dgram_ops;
 664                 break;
 665         case SOCK_SEQPACKET:
 666                 sock->ops = &unix_seqpacket_ops;
 667                 break;
 668         default:
 669                 return -ESOCKTNOSUPPORT;
 670         }
 671
 672         return unix_create1(net, sock) ? 0 : -ENOMEM;
 673 }
 674
 675 static int unix_release(struct socket *sock)
 676 {
 677         struct sock *sk = sock->sk;
 678
 679         if (!sk)
 680                 return 0;
 681
 682         sock->sk = NULL;
 683
 684         return unix_release_sock(sk, 0);
 685 }
 686
 687 static int unix_autobind(struct socket *sock)
 688 {
 689         struct sock *sk = sock->sk;
 690         struct net *net = sock_net(sk);
 691         struct unix_sock *u = unix_sk(sk);
 692         static u32 ordernum = 1;
 693         struct unix_address *addr;
 694         int err;
 695
 696         mutex_lock(&u->readlock);
 697
 698         err = 0;
 699         if (u->addr)
 700                 goto out;
 701
 702         err = -ENOMEM;
 703         addr = kzalloc(sizeof(*addr) + sizeof(short) + 16, GFP_KERNEL);
 704         if (!addr)
 705                 goto out;
 706
 707         addr->name->sun_family = AF_UNIX;
 708         atomic_set(&addr->refcnt, 1);
 709
 710 retry:
 711         addr->len = sprintf(addr->name->sun_path+1, "%05x", ordernum) + 1 + sizeof(short);
 712         addr->hash = unix_hash_fold(csum_partial(addr->name, addr->len, 0));
 713
 714         spin_lock(&unix_table_lock);
 715         ordernum = (ordernum+1)&0xFFFFF;
 716
 717         if (__unix_find_socket_byname(net, addr->name, addr->len, sock->type,
 718                                       addr->hash)) {
 719                 spin_unlock(&unix_table_lock);
 720                 /* Sanity yield. It is unusual case, but yet... */
 721                 if (!(ordernum&0xFF))
 722                         yield();
 723                 goto retry;
 724         }
 725         addr->hash ^= sk->sk_type;
 726
 727         __unix_remove_socket(sk);
 728         u->addr = addr;
 729         __unix_insert_socket(&unix_socket_table[addr->hash], sk);
 730         spin_unlock(&unix_table_lock);
 731         err = 0;
 732
 733 out:    mutex_unlock(&u->readlock);
 734         return err;
 735 }
 736
 737 static struct sock *unix_find_other(struct net *net,
 738                                     struct sockaddr_un *sunname, int len,
 739                                     int type, unsigned hash, int *error)
 740 {
 741         struct sock *u;
 742         struct path path;
 743         int err = 0;
 744
 745         if (sunname->sun_path[0]) {
 746                 struct inode *inode;
 747                 err = kern_path(sunname->sun_path, LOOKUP_FOLLOW, &path);
 748                 if (err)
 749                         goto fail;
 750                 inode = path.dentry->d_inode;
 751                 err = inode_permission(inode, MAY_WRITE);
 752                 if (err)
 753                         goto put_fail;
 754
 755                 err = -ECONNREFUSED;
 756                 if (!S_ISSOCK(inode->i_mode))
 757                         goto put_fail;
 758                 u = unix_find_socket_byinode(inode);
 759                 if (!u)
 760                         goto put_fail;
 761
 762                 if (u->sk_type == type)
 763                         touch_atime(path.mnt, path.dentry);
 764
 765                 path_put(&path);
 766
 767                 err = -EPROTOTYPE;
 768                 if (u->sk_type != type) {
 769                         sock_put(u);
 770                         goto fail;
 771                 }
 772         } else {
 773                 err = -ECONNREFUSED;
 774                 u = unix_find_socket_byname(net, sunname, len, type, hash);
 775                 if (u) {
 776                         struct dentry *dentry;
 777                         dentry = unix_sk(u)->dentry;
 778                         if (dentry)
 779                                 touch_atime(unix_sk(u)->mnt, dentry);
 780                 } else
 781                         goto fail;
 782         }
 783         return u;
 784
 785 put_fail:
 786         path_put(&path);
 787 fail:
 788         *error = err;
 789         return NULL;
 790 }
 791
 792
 793 static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
 794 {
 795         struct sock *sk = sock->sk;
 796         struct net *net = sock_net(sk);
 797         struct unix_sock *u = unix_sk(sk);
 798         struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
 799         struct dentry *dentry = NULL;
 800         struct nameidata nd;
 801         int err;
 802         unsigned hash;
 803         struct unix_address *addr;
 804         struct hlist_head *list;
 805
 806         err = -EINVAL;
 807         if (sunaddr->sun_family != AF_UNIX)
 808                 goto out;
 809
 810         if (addr_len == sizeof(short)) {
 811                 err = unix_autobind(sock);
 812                 goto out;
 813         }
 814
 815         err = unix_mkname(sunaddr, addr_len, &hash);
 816         if (err < 0)
 817                 goto out;
 818         addr_len = err;
 819
 820         mutex_lock(&u->readlock);
 821
 822         err = -EINVAL;
 823         if (u->addr)
 824                 goto out_up;
 825
 826         err = -ENOMEM;
 827         addr = kmalloc(sizeof(*addr)+addr_len, GFP_KERNEL);
 828         if (!addr)
 829                 goto out_up;
 830
 831         memcpy(addr->name, sunaddr, addr_len);
 832         addr->len = addr_len;
 833         addr->hash = hash ^ sk->sk_type;
 834         atomic_set(&addr->refcnt, 1);
 835
 836         if (sunaddr->sun_path[0]) {
 837                 unsigned int mode;
 838                 err = 0;
 839                 /*
 840                  * Get the parent directory, calculate the hash for last
 841                  * component.
 842                  */
 843                 err = path_lookup(sunaddr->sun_path, LOOKUP_PARENT, &nd);
 844                 if (err)
 845                         goto out_mknod_parent;
 846
 847                 dentry = lookup_create(&nd, 0);
 848                 err = PTR_ERR(dentry);
 849                 if (IS_ERR(dentry))
 850                         goto out_mknod_unlock;
 851
 852                 /*
 853                  * All right, let's create it.
 854                  */
 855                 mode = S_IFSOCK |
 856                        (SOCK_INODE(sock)->i_mode & ~current_umask());
 857                 err = mnt_want_write(nd.path.mnt);
 858                 if (err)
 859                         goto out_mknod_dput;
 860                 err = security_path_mknod(&nd.path, dentry, mode, 0);
 861                 if (err)
 862                         goto out_mknod_drop_write;
 863                 err = vfs_mknod(nd.path.dentry->d_inode, dentry, mode, 0);
 864 out_mknod_drop_write:
 865                 mnt_drop_write(nd.path.mnt);
 866                 if (err)
 867                         goto out_mknod_dput;
 868                 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
 869                 dput(nd.path.dentry);
 870                 nd.path.dentry = dentry;
 871
 872                 addr->hash = UNIX_HASH_SIZE;
 873         }
 874
 875         spin_lock(&unix_table_lock);
 876
 877         if (!sunaddr->sun_path[0]) {
 878                 err = -EADDRINUSE;
 879                 if (__unix_find_socket_byname(net, sunaddr, addr_len,
 880                                               sk->sk_type, hash)) {
 881                         unix_release_addr(addr);
 882                         goto out_unlock;
 883                 }
 884
 885                 list = &unix_socket_table[addr->hash];
 886         } else {
 887                 list = &unix_socket_table[dentry->d_inode->i_ino & (UNIX_HASH_SIZE-1)];
 888                 u->dentry = nd.path.dentry;
 889                 u->mnt    = nd.path.mnt;
 890         }
 891
 892         err = 0;
 893         __unix_remove_socket(sk);
 894         u->addr = addr;
 895         __unix_insert_socket(list, sk);
 896
 897 out_unlock:
 898         spin_unlock(&unix_table_lock);
 899 out_up:
 900         mutex_unlock(&u->readlock);
 901 out:
 902         return err;
 903
 904 out_mknod_dput:
 905         dput(dentry);
 906 out_mknod_unlock:
 907         mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
 908         path_put(&nd.path);
 909 out_mknod_parent:
 910         if (err == -EEXIST)
 911                 err = -EADDRINUSE;
 912         unix_release_addr(addr);
 913         goto out_up;
 914 }
 915
 916 static void unix_state_double_lock(struct sock *sk1, struct sock *sk2)
 917 {
 918         if (unlikely(sk1 == sk2) || !sk2) {
 919                 unix_state_lock(sk1);
 920                 return;
 921         }
 922         if (sk1 < sk2) {
 923                 unix_state_lock(sk1);
 924                 unix_state_lock_nested(sk2);
 925         } else {
 926                 unix_state_lock(sk2);
 927                 unix_state_lock_nested(sk1);
 928         }
 929 }
 930
 931 static void unix_state_double_unlock(struct sock *sk1, struct sock *sk2)
 932 {
 933         if (unlikely(sk1 == sk2) || !sk2) {
 934                 unix_state_unlock(sk1);
 935                 return;
 936         }
 937         unix_state_unlock(sk1);
 938         unix_state_unlock(sk2);
 939 }
 940
 941 static int unix_dgram_connect(struct socket *sock, struct sockaddr *addr,
 942                               int alen, int flags)
 943 {
 944         struct sock *sk = sock->sk;
 945         struct net *net = sock_net(sk);
 946         struct sockaddr_un *sunaddr = (struct sockaddr_un *)addr;
 947         struct sock *other;
 948         unsigned hash;
 949         int err;
 950
 951         if (addr->sa_family != AF_UNSPEC) {
 952                 err = unix_mkname(sunaddr, alen, &hash);
 953                 if (err < 0)
 954                         goto out;
 955                 alen = err;
 956
 957                 if (test_bit(SOCK_PASSCRED, &sock->flags) &&
 958                     !unix_sk(sk)->addr && (err = unix_autobind(sock)) != 0)
 959                         goto out;
 960
 961 restart:
 962                 other = unix_find_other(net, sunaddr, alen, sock->type, hash, &err);
 963                 if (!other)
 964                         goto out;
 965
 966                 unix_state_double_lock(sk, other);
 967
 968                 /* Apparently VFS overslept socket death. Retry. */
 969                 if (sock_flag(other, SOCK_DEAD)) {
 970                         unix_state_double_unlock(sk, other);
 971                         sock_put(other);
 972                         goto restart;
 973                 }
 974
 975                 err = -EPERM;
 976                 if (!unix_may_send(sk, other))
 977                         goto out_unlock;
 978
 979                 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
 980                 if (err)
 981                         goto out_unlock;
 982
 983         } else {
 984                 /*
 985                  *      1003.1g breaking connected state with AF_UNSPEC
 986                  */
 987                 other = NULL;
 988                 unix_state_double_lock(sk, other);
 989         }
 990
 991         /*
 992          * If it was connected, reconnect.
 993          */
 994         if (unix_peer(sk)) {
 995                 struct sock *old_peer = unix_peer(sk);
 996                 unix_peer(sk) = other;
 997                 unix_state_double_unlock(sk, other);
 998
 999                 if (other != old_peer)
1000                         unix_dgram_disconnected(sk, old_peer);
1001                 sock_put(old_peer);
1002         } else {
1003                 unix_peer(sk) = other;
1004                 unix_state_double_unlock(sk, other);
1005         }
1006         return 0;
1007
1008 out_unlock:
1009         unix_state_double_unlock(sk, other);
1010         sock_put(other);
1011 out:
1012         return err;
1013 }
1014
1015 static long unix_wait_for_peer(struct sock *other, long timeo)
1016 {
1017         struct unix_sock *u = unix_sk(other);
1018         int sched;
1019         DEFINE_WAIT(wait);
1020
1021         prepare_to_wait_exclusive(&u->peer_wait, &wait, TASK_INTERRUPTIBLE);
1022
1023         sched = !sock_flag(other, SOCK_DEAD) &&
1024                 !(other->sk_shutdown & RCV_SHUTDOWN) &&
1025                 unix_recvq_full(other);
1026
1027         unix_state_unlock(other);
1028
1029         if (sched)
1030                 timeo = schedule_timeout(timeo);
1031
1032         finish_wait(&u->peer_wait, &wait);
1033         return timeo;
1034 }
1035
1036 static int unix_stream_connect(struct socket *sock, struct sockaddr *uaddr,
1037                                int addr_len, int flags)
1038 {
1039         struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
1040         struct sock *sk = sock->sk;
1041         struct net *net = sock_net(sk);
1042         struct unix_sock *u = unix_sk(sk), *newu, *otheru;
1043         struct sock *newsk = NULL;
1044         struct sock *other = NULL;
1045         struct sk_buff *skb = NULL;
1046         unsigned hash;
1047         int st;
1048         int err;
1049         long timeo;
1050
1051         err = unix_mkname(sunaddr, addr_len, &hash);
1052         if (err < 0)
1053                 goto out;
1054         addr_len = err;
1055
1056         if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr &&
1057             (err = unix_autobind(sock)) != 0)
1058                 goto out;
1059
1060         timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
1061
1062         /* First of all allocate resources.
1063            If we will make it after state is locked,
1064            we will have to recheck all again in any case.
1065          */
1066
1067         err = -ENOMEM;
1068
1069         /* create new sock for complete connection */
1070         newsk = unix_create1(sock_net(sk), NULL);
1071         if (newsk == NULL)
1072                 goto out;
1073
1074         /* Allocate skb for sending to listening sock */
1075         skb = sock_wmalloc(newsk, 1, 0, GFP_KERNEL);
1076         if (skb == NULL)
1077                 goto out;
1078
1079 restart:
1080         /*  Find listening sock. */
1081         other = unix_find_other(net, sunaddr, addr_len, sk->sk_type, hash, &err);
1082         if (!other)
1083                 goto out;
1084
1085         /* Latch state of peer */
1086         unix_state_lock(other);
1087
1088         /* Apparently VFS overslept socket death. Retry. */
1089         if (sock_flag(other, SOCK_DEAD)) {
1090                 unix_state_unlock(other);
1091                 sock_put(other);
1092                 goto restart;
1093         }
1094
1095         err = -ECONNREFUSED;
1096         if (other->sk_state != TCP_LISTEN)
1097                 goto out_unlock;
1098         if (other->sk_shutdown & RCV_SHUTDOWN)
1099                 goto out_unlock;
1100
1101         if (unix_recvq_full(other)) {
1102                 err = -EAGAIN;
1103                 if (!timeo)
1104                         goto out_unlock;
1105
1106                 timeo = unix_wait_for_peer(other, timeo);
1107
1108                 err = sock_intr_errno(timeo);
1109                 if (signal_pending(current))
1110                         goto out;
1111                 sock_put(other);
1112                 goto restart;
1113         }
1114
1115         /* Latch our state.
1116
1117            It is tricky place. We need to grab write lock and cannot
1118            drop lock on peer. It is dangerous because deadlock is
1119            possible. Connect to self case and simultaneous
1120            attempt to connect are eliminated by checking socket
1121            state. other is TCP_LISTEN, if sk is TCP_LISTEN we
1122            check this before attempt to grab lock.
1123
1124            Well, and we have to recheck the state after socket locked.
1125          */
1126         st = sk->sk_state;
1127
1128         switch (st) {
1129         case TCP_CLOSE:
1130                 /* This is ok... continue with connect */
1131                 break;
1132         case TCP_ESTABLISHED:
1133                 /* Socket is already connected */
1134                 err = -EISCONN;
1135                 goto out_unlock;
1136         default:
1137                 err = -EINVAL;
1138                 goto out_unlock;
1139         }
1140
1141         unix_state_lock_nested(sk);
1142
1143         if (sk->sk_state != st) {
1144                 unix_state_unlock(sk);
1145                 unix_state_unlock(other);
1146                 sock_put(other);
1147                 goto restart;
1148         }
1149
1150         err = security_unix_stream_connect(sock, other->sk_socket, newsk);
1151         if (err) {
1152                 unix_state_unlock(sk);
1153                 goto out_unlock;
1154         }
1155
1156         /* The way is open! Fastly set all the necessary fields... */
1157
1158         sock_hold(sk);
1159         unix_peer(newsk)        = sk;
1160         newsk->sk_state         = TCP_ESTABLISHED;
1161         newsk->sk_type          = sk->sk_type;
1162         init_peercred(newsk);
1163         newu = unix_sk(newsk);
1164         newsk->sk_wq            = &newu->peer_wq;
1165         otheru = unix_sk(other);
1166
1167         /* copy address information from listening to new sock*/
1168         if (otheru->addr) {
1169                 atomic_inc(&otheru->addr->refcnt);
1170                 newu->addr = otheru->addr;
1171         }
1172         if (otheru->dentry) {
1173                 newu->dentry    = dget(otheru->dentry);
1174                 newu->mnt       = mntget(otheru->mnt);
1175         }
1176
1177         /* Set credentials */
1178         copy_peercred(sk, other);
1179
1180         sock->state     = SS_CONNECTED;
1181         sk->sk_state    = TCP_ESTABLISHED;
1182         sock_hold(newsk);
1183
1184         smp_mb__after_atomic_inc();     /* sock_hold() does an atomic_inc() */
1185         unix_peer(sk)   = newsk;
1186
1187         unix_state_unlock(sk);
1188
1189         /* take ten and and send info to listening sock */
1190         spin_lock(&other->sk_receive_queue.lock);
1191         __skb_queue_tail(&other->sk_receive_queue, skb);
1192         spin_unlock(&other->sk_receive_queue.lock);
1193         unix_state_unlock(other);
1194         other->sk_data_ready(other, 0);
1195         sock_put(other);
1196         return 0;
1197
1198 out_unlock:
1199         if (other)
1200                 unix_state_unlock(other);
1201
1202 out:
1203         kfree_skb(skb);
1204         if (newsk)
1205                 unix_release_sock(newsk, 0);
1206         if (other)
1207                 sock_put(other);
1208         return err;
1209 }
1210
1211 static int unix_socketpair(struct socket *socka, struct socket *sockb)
1212 {
1213         struct sock *ska = socka->sk, *skb = sockb->sk;
1214
1215         /* Join our sockets back to back */
1216         sock_hold(ska);
1217         sock_hold(skb);
1218         unix_peer(ska) = skb;
1219         unix_peer(skb) = ska;
1220         init_peercred(ska);
1221         init_peercred(skb);
1222
1223         if (ska->sk_type != SOCK_DGRAM) {
1224                 ska->sk_state = TCP_ESTABLISHED;
1225                 skb->sk_state = TCP_ESTABLISHED;
1226                 socka->state  = SS_CONNECTED;
1227                 sockb->state  = SS_CONNECTED;
1228         }
1229         return 0;
1230 }
1231
1232 static int unix_accept(struct socket *sock, struct socket *newsock, int flags)
1233 {
1234         struct sock *sk = sock->sk;
1235         struct sock *tsk;
1236         struct sk_buff *skb;
1237         int err;
1238
1239         err = -EOPNOTSUPP;
1240         if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
1241                 goto out;
1242
1243         err = -EINVAL;
1244         if (sk->sk_state != TCP_LISTEN)
1245                 goto out;
1246
1247         /* If socket state is TCP_LISTEN it cannot change (for now...),
1248          * so that no locks are necessary.
1249          */
1250
1251         skb = skb_recv_datagram(sk, 0, flags&O_NONBLOCK, &err);
1252         if (!skb) {
1253                 /* This means receive shutdown. */
1254                 if (err == 0)
1255                         err = -EINVAL;
1256                 goto out;
1257         }
1258
1259         tsk = skb->sk;
1260         skb_free_datagram(sk, skb);
1261         wake_up_interruptible(&unix_sk(sk)->peer_wait);
1262
1263         /* attach accepted sock to socket */
1264         unix_state_lock(tsk);
1265         newsock->state = SS_CONNECTED;
1266         sock_graft(tsk, newsock);
1267         unix_state_unlock(tsk);
1268         return 0;
1269
1270 out:
1271         return err;
1272 }
1273
1274
1275 static int unix_getname(struct socket *sock, struct sockaddr *uaddr, int *uaddr_len, int peer)
1276 {
1277         struct sock *sk = sock->sk;
1278         struct unix_sock *u;
1279         DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr, uaddr);
1280         int err = 0;
1281
1282         if (peer) {
1283                 sk = unix_peer_get(sk);
1284
1285                 err = -ENOTCONN;
1286                 if (!sk)
1287                         goto out;
1288                 err = 0;
1289         } else {
1290                 sock_hold(sk);
1291         }
1292
1293         u = unix_sk(sk);
1294         unix_state_lock(sk);
1295         if (!u->addr) {
1296                 sunaddr->sun_family = AF_UNIX;
1297                 sunaddr->sun_path[0] = 0;
1298                 *uaddr_len = sizeof(short);
1299         } else {
1300                 struct unix_address *addr = u->addr;
1301
1302                 *uaddr_len = addr->len;
1303                 memcpy(sunaddr, addr->name, *uaddr_len);
1304         }
1305         unix_state_unlock(sk);
1306         sock_put(sk);
1307 out:
1308         return err;
1309 }
1310
1311 static void unix_detach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1312 {
1313         int i;
1314
1315         scm->fp = UNIXCB(skb).fp;
1316         UNIXCB(skb).fp = NULL;
1317
1318         for (i = scm->fp->count-1; i >= 0; i--)
1319                 unix_notinflight(scm->fp->fp[i]);
1320 }
1321
1322 static void unix_destruct_scm(struct sk_buff *skb)
1323 {
1324         struct scm_cookie scm;
1325         memset(&scm, 0, sizeof(scm));
1326         scm.pid  = UNIXCB(skb).pid;
1327         scm.cred = UNIXCB(skb).cred;
1328         if (UNIXCB(skb).fp)
1329                 unix_detach_fds(&scm, skb);
1330
1331         /* Alas, it calls VFS */
1332         /* So fscking what? fput() had been SMP-safe since the last Summer */
1333         scm_destroy(&scm);
1334         sock_wfree(skb);
1335 }
1336
1337 static int unix_attach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1338 {
1339         int i;
1340
1341         /*
1342          * Need to duplicate file references for the sake of garbage
1343          * collection.  Otherwise a socket in the fps might become a
1344          * candidate for GC while the skb is not yet queued.
1345          */
1346         UNIXCB(skb).fp = scm_fp_dup(scm->fp);
1347         if (!UNIXCB(skb).fp)
1348                 return -ENOMEM;
1349
1350         for (i = scm->fp->count-1; i >= 0; i--)
1351                 unix_inflight(scm->fp->fp[i]);
1352         return 0;
1353 }
1354
1355 static int unix_scm_to_skb(struct scm_cookie *scm, struct sk_buff *skb, bool send_fds)
1356 {
1357         int err = 0;
1358         UNIXCB(skb).pid  = get_pid(scm->pid);
1359         UNIXCB(skb).cred = get_cred(scm->cred);
1360         UNIXCB(skb).fp = NULL;
1361         if (scm->fp && send_fds)
1362                 err = unix_attach_fds(scm, skb);
1363
1364         skb->destructor = unix_destruct_scm;
1365         return err;
1366 }
1367
1368 /*
1369  *      Send AF_UNIX data.
1370  */
1371
1372 static int unix_dgram_sendmsg(struct kiocb *kiocb, struct socket *sock,
1373                               struct msghdr *msg, size_t len)
1374 {
1375         struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1376         struct sock *sk = sock->sk;
1377         struct net *net = sock_net(sk);
1378         struct unix_sock *u = unix_sk(sk);
1379         struct sockaddr_un *sunaddr = msg->msg_name;
1380         struct sock *other = NULL;
1381         int namelen = 0; /* fake GCC */
1382         int err;
1383         unsigned hash;
1384         struct sk_buff *skb;
1385         long timeo;
1386         struct scm_cookie tmp_scm;
1387
1388         if (NULL == siocb->scm)
1389                 siocb->scm = &tmp_scm;
1390         wait_for_unix_gc();
1391         err = scm_send(sock, msg, siocb->scm);
1392         if (err < 0)
1393                 return err;
1394
1395         err = -EOPNOTSUPP;
1396         if (msg->msg_flags&MSG_OOB)
1397                 goto out;
1398
1399         if (msg->msg_namelen) {
1400                 err = unix_mkname(sunaddr, msg->msg_namelen, &hash);
1401                 if (err < 0)
1402                         goto out;
1403                 namelen = err;
1404         } else {
1405                 sunaddr = NULL;
1406                 err = -ENOTCONN;
1407                 other = unix_peer_get(sk);
1408                 if (!other)
1409                         goto out;
1410         }
1411
1412         if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr
1413             && (err = unix_autobind(sock)) != 0)
1414                 goto out;
1415
1416         err = -EMSGSIZE;
1417         if (len > sk->sk_sndbuf - 32)
1418                 goto out;
1419
1420         skb = sock_alloc_send_skb(sk, len, msg->msg_flags&MSG_DONTWAIT, &err);
1421         if (skb == NULL)
1422                 goto out;
1423
1424         err = unix_scm_to_skb(siocb->scm, skb, true);
1425         if (err)
1426                 goto out_free;
1427         unix_get_secdata(siocb->scm, skb);
1428
1429         skb_reset_transport_header(skb);
1430         err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
1431         if (err)
1432                 goto out_free;
1433
1434         timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1435
1436 restart:
1437         if (!other) {
1438                 err = -ECONNRESET;
1439                 if (sunaddr == NULL)
1440                         goto out_free;
1441
1442                 other = unix_find_other(net, sunaddr, namelen, sk->sk_type,
1443                                         hash, &err);
1444                 if (other == NULL)
1445                         goto out_free;
1446         }
1447
1448         unix_state_lock(other);
1449         err = -EPERM;
1450         if (!unix_may_send(sk, other))
1451                 goto out_unlock;
1452
1453         if (sock_flag(other, SOCK_DEAD)) {
1454                 /*
1455                  *      Check with 1003.1g - what should
1456                  *      datagram error
1457                  */
1458                 unix_state_unlock(other);
1459                 sock_put(other);
1460
1461                 err = 0;
1462                 unix_state_lock(sk);
1463                 if (unix_peer(sk) == other) {
1464                         unix_peer(sk) = NULL;
1465                         unix_state_unlock(sk);
1466
1467                         unix_dgram_disconnected(sk, other);
1468                         sock_put(other);
1469                         err = -ECONNREFUSED;
1470                 } else {
1471                         unix_state_unlock(sk);
1472                 }
1473
1474                 other = NULL;
1475                 if (err)
1476                         goto out_free;
1477                 goto restart;
1478         }
1479
1480         err = -EPIPE;
1481         if (other->sk_shutdown & RCV_SHUTDOWN)
1482                 goto out_unlock;
1483
1484         if (sk->sk_type != SOCK_SEQPACKET) {
1485                 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1486                 if (err)
1487                         goto out_unlock;
1488         }
1489
1490         if (unix_peer(other) != sk && unix_recvq_full(other)) {
1491                 if (!timeo) {
1492                         err = -EAGAIN;
1493                         goto out_unlock;
1494                 }
1495
1496                 timeo = unix_wait_for_peer(other, timeo);
1497
1498                 err = sock_intr_errno(timeo);
1499                 if (signal_pending(current))
1500                         goto out_free;
1501
1502                 goto restart;
1503         }
1504
1505         skb_queue_tail(&other->sk_receive_queue, skb);
1506         unix_state_unlock(other);
1507         other->sk_data_ready(other, len);
1508         sock_put(other);
1509         scm_destroy(siocb->scm);
1510         return len;
1511
1512 out_unlock:
1513         unix_state_unlock(other);
1514 out_free:
1515         kfree_skb(skb);
1516 out:
1517         if (other)
1518                 sock_put(other);
1519         scm_destroy(siocb->scm);
1520         return err;
1521 }
1522
1523
1524 static int unix_stream_sendmsg(struct kiocb *kiocb, struct socket *sock,
1525                                struct msghdr *msg, size_t len)
1526 {
1527         struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1528         struct sock *sk = sock->sk;
1529         struct sock *other = NULL;
1530         struct sockaddr_un *sunaddr = msg->msg_name;
1531         int err, size;
1532         struct sk_buff *skb;
1533         int sent = 0;
1534         struct scm_cookie tmp_scm;
1535         bool fds_sent = false;
1536
1537         if (NULL == siocb->scm)
1538                 siocb->scm = &tmp_scm;
1539         wait_for_unix_gc();
1540         err = scm_send(sock, msg, siocb->scm);
1541         if (err < 0)
1542                 return err;
1543
1544         err = -EOPNOTSUPP;
1545         if (msg->msg_flags&MSG_OOB)
1546                 goto out_err;
1547
1548         if (msg->msg_namelen) {
1549                 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
1550                 goto out_err;
1551         } else {
1552                 sunaddr = NULL;
1553                 err = -ENOTCONN;
1554                 other = unix_peer(sk);
1555                 if (!other)
1556                         goto out_err;
1557         }
1558
1559         if (sk->sk_shutdown & SEND_SHUTDOWN)
1560                 goto pipe_err;
1561
1562         while (sent < len) {
1563                 /*
1564                  *      Optimisation for the fact that under 0.01% of X
1565                  *      messages typically need breaking up.
1566                  */
1567
1568                 size = len-sent;
1569
1570                 /* Keep two messages in the pipe so it schedules better */
1571                 if (size > ((sk->sk_sndbuf >> 1) - 64))
1572                         size = (sk->sk_sndbuf >> 1) - 64;
1573
1574                 if (size > SKB_MAX_ALLOC)
1575                         size = SKB_MAX_ALLOC;
1576
1577                 /*
1578                  *      Grab a buffer
1579                  */
1580
1581                 skb = sock_alloc_send_skb(sk, size, msg->msg_flags&MSG_DONTWAIT,
1582                                           &err);
1583
1584                 if (skb == NULL)
1585                         goto out_err;
1586
1587                 /*
1588                  *      If you pass two values to the sock_alloc_send_skb
1589                  *      it tries to grab the large buffer with GFP_NOFS
1590                  *      (which can fail easily), and if it fails grab the
1591                  *      fallback size buffer which is under a page and will
1592                  *      succeed. [Alan]
1593                  */
1594                 size = min_t(int, size, skb_tailroom(skb));
1595
1596
1597                 /* Only send the fds in the first buffer */
1598                 err = unix_scm_to_skb(siocb->scm, skb, !fds_sent);
1599                 if (err) {
1600                         kfree_skb(skb);
1601                         goto out_err;
1602                 }
1603                 fds_sent = true;
1604
1605                 err = memcpy_fromiovec(skb_put(skb, size), msg->msg_iov, size);
1606                 if (err) {
1607                         kfree_skb(skb);
1608                         goto out_err;
1609                 }
1610
1611                 unix_state_lock(other);
1612
1613                 if (sock_flag(other, SOCK_DEAD) ||
1614                     (other->sk_shutdown & RCV_SHUTDOWN))
1615                         goto pipe_err_free;
1616
1617                 skb_queue_tail(&other->sk_receive_queue, skb);
1618                 unix_state_unlock(other);
1619                 other->sk_data_ready(other, size);
1620                 sent += size;
1621         }
1622
1623         scm_destroy(siocb->scm);
1624         siocb->scm = NULL;
1625
1626         return sent;
1627
1628 pipe_err_free:
1629         unix_state_unlock(other);
1630         kfree_skb(skb);
1631 pipe_err:
1632         if (sent == 0 && !(msg->msg_flags&MSG_NOSIGNAL))
1633                 send_sig(SIGPIPE, current, 0);
1634         err = -EPIPE;
1635 out_err:
1636         scm_destroy(siocb->scm);
1637         siocb->scm = NULL;
1638         return sent ? : err;
1639 }
1640
1641 static int unix_seqpacket_sendmsg(struct kiocb *kiocb, struct socket *sock,
1642                                   struct msghdr *msg, size_t len)
1643 {
1644         int err;
1645         struct sock *sk = sock->sk;
1646
1647         err = sock_error(sk);
1648         if (err)
1649                 return err;
1650
1651         if (sk->sk_state != TCP_ESTABLISHED)
1652                 return -ENOTCONN;
1653
1654         if (msg->msg_namelen)
1655                 msg->msg_namelen = 0;
1656
1657         return unix_dgram_sendmsg(kiocb, sock, msg, len);
1658 }
1659
1660 static void unix_copy_addr(struct msghdr *msg, struct sock *sk)
1661 {
1662         struct unix_sock *u = unix_sk(sk);
1663
1664         msg->msg_namelen = 0;
1665         if (u->addr) {
1666                 msg->msg_namelen = u->addr->len;
1667                 memcpy(msg->msg_name, u->addr->name, u->addr->len);
1668         }
1669 }
1670
1671 static int unix_dgram_recvmsg(struct kiocb *iocb, struct socket *sock,
1672                               struct msghdr *msg, size_t size,
1673                               int flags)
1674 {
1675         struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1676         struct scm_cookie tmp_scm;
1677         struct sock *sk = sock->sk;
1678         struct unix_sock *u = unix_sk(sk);
1679         int noblock = flags & MSG_DONTWAIT;
1680         struct sk_buff *skb;
1681         int err;
1682
1683         err = -EOPNOTSUPP;
1684         if (flags&MSG_OOB)
1685                 goto out;
1686
1687         msg->msg_namelen = 0;
1688
1689         mutex_lock(&u->readlock);
1690
1691         skb = skb_recv_datagram(sk, flags, noblock, &err);
1692         if (!skb) {
1693                 unix_state_lock(sk);
1694                 /* Signal EOF on disconnected non-blocking SEQPACKET socket. */
1695                 if (sk->sk_type == SOCK_SEQPACKET && err == -EAGAIN &&
1696                     (sk->sk_shutdown & RCV_SHUTDOWN))
1697                         err = 0;
1698                 unix_state_unlock(sk);
1699                 goto out_unlock;
1700         }
1701
1702         wake_up_interruptible_sync(&u->peer_wait);
1703
1704         if (msg->msg_name)
1705                 unix_copy_addr(msg, skb->sk);
1706
1707         if (size > skb->len)
1708                 size = skb->len;
1709         else if (size < skb->len)
1710                 msg->msg_flags |= MSG_TRUNC;
1711
1712         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, size);
1713         if (err)
1714                 goto out_free;
1715
1716         if (!siocb->scm) {
1717                 siocb->scm = &tmp_scm;
1718                 memset(&tmp_scm, 0, sizeof(tmp_scm));
1719         }
1720         scm_set_cred(siocb->scm, UNIXCB(skb).pid, UNIXCB(skb).cred);
1721         unix_set_secdata(siocb->scm, skb);
1722
1723         if (!(flags & MSG_PEEK)) {
1724                 if (UNIXCB(skb).fp)
1725                         unix_detach_fds(siocb->scm, skb);
1726         } else {
1727                 /* It is questionable: on PEEK we could:
1728                    - do not return fds - good, but too simple 8)
1729                    - return fds, and do not return them on read (old strategy,
1730                      apparently wrong)
1731                    - clone fds (I chose it for now, it is the most universal
1732                      solution)
1733
1734                    POSIX 1003.1g does not actually define this clearly
1735                    at all. POSIX 1003.1g doesn't define a lot of things
1736                    clearly however!
1737
1738                 */
1739                 if (UNIXCB(skb).fp)
1740                         siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1741         }
1742         err = size;
1743
1744         scm_recv(sock, msg, siocb->scm, flags);
1745
1746 out_free:
1747         skb_free_datagram(sk, skb);
1748 out_unlock:
1749         mutex_unlock(&u->readlock);
1750 out:
1751         return err;
1752 }
1753
1754 /*
1755  *      Sleep until data has arrive. But check for races..
1756  */
1757
1758 static long unix_stream_data_wait(struct sock *sk, long timeo)
1759 {
1760         DEFINE_WAIT(wait);
1761
1762         unix_state_lock(sk);
1763
1764         for (;;) {
1765                 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1766
1767                 if (!skb_queue_empty(&sk->sk_receive_queue) ||
1768                     sk->sk_err ||
1769                     (sk->sk_shutdown & RCV_SHUTDOWN) ||
1770                     signal_pending(current) ||
1771                     !timeo)
1772                         break;
1773
1774                 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1775                 unix_state_unlock(sk);
1776                 timeo = schedule_timeout(timeo);
1777                 unix_state_lock(sk);
1778                 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1779         }
1780
1781         finish_wait(sk_sleep(sk), &wait);
1782         unix_state_unlock(sk);
1783         return timeo;
1784 }
1785
1786
1787
1788 static int unix_stream_recvmsg(struct kiocb *iocb, struct socket *sock,
1789                                struct msghdr *msg, size_t size,
1790                                int flags)
1791 {
1792         struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1793         struct scm_cookie tmp_scm;
1794         struct sock *sk = sock->sk;
1795         struct unix_sock *u = unix_sk(sk);
1796         struct sockaddr_un *sunaddr = msg->msg_name;
1797         int copied = 0;
1798         int check_creds = 0;
1799         int target;
1800         int err = 0;
1801         long timeo;
1802
1803         err = -EINVAL;
1804         if (sk->sk_state != TCP_ESTABLISHED)
1805                 goto out;
1806
1807         err = -EOPNOTSUPP;
1808         if (flags&MSG_OOB)
1809                 goto out;
1810
1811         target = sock_rcvlowat(sk, flags&MSG_WAITALL, size);
1812         timeo = sock_rcvtimeo(sk, flags&MSG_DONTWAIT);
1813
1814         msg->msg_namelen = 0;
1815
1816         /* Lock the socket to prevent queue disordering
1817          * while sleeps in memcpy_tomsg
1818          */
1819
1820         if (!siocb->scm) {
1821                 siocb->scm = &tmp_scm;
1822                 memset(&tmp_scm, 0, sizeof(tmp_scm));
1823         }
1824
1825         mutex_lock(&u->readlock);
1826
1827         do {
1828                 int chunk;
1829                 struct sk_buff *skb;
1830
1831                 unix_state_lock(sk);
1832                 skb = skb_dequeue(&sk->sk_receive_queue);
1833                 if (skb == NULL) {
1834                         if (copied >= target)
1835                                 goto unlock;
1836
1837                         /*
1838                          *      POSIX 1003.1g mandates this order.
1839                          */
1840
1841                         err = sock_error(sk);
1842                         if (err)
1843                                 goto unlock;
1844                         if (sk->sk_shutdown & RCV_SHUTDOWN)
1845                                 goto unlock;
1846
1847                         unix_state_unlock(sk);
1848                         err = -EAGAIN;
1849                         if (!timeo)
1850                                 break;
1851                         mutex_unlock(&u->readlock);
1852
1853                         timeo = unix_stream_data_wait(sk, timeo);
1854
1855                         if (signal_pending(current)) {
1856                                 err = sock_intr_errno(timeo);
1857                                 goto out;
1858                         }
1859                         mutex_lock(&u->readlock);
1860                         continue;
1861  unlock:
1862                         unix_state_unlock(sk);
1863                         break;
1864                 }
1865                 unix_state_unlock(sk);
1866
1867                 if (check_creds) {
1868                         /* Never glue messages from different writers */
1869                         if ((UNIXCB(skb).pid  != siocb->scm->pid) ||
1870                             (UNIXCB(skb).cred != siocb->scm->cred)) {
1871                                 skb_queue_head(&sk->sk_receive_queue, skb);
1872                                 break;
1873                         }
1874                 } else {
1875                         /* Copy credentials */
1876                         scm_set_cred(siocb->scm, UNIXCB(skb).pid, UNIXCB(skb).cred);
1877                         check_creds = 1;
1878                 }
1879
1880                 /* Copy address just once */
1881                 if (sunaddr) {
1882                         unix_copy_addr(msg, skb->sk);
1883                         sunaddr = NULL;
1884                 }
1885
1886                 chunk = min_t(unsigned int, skb->len, size);
1887                 if (memcpy_toiovec(msg->msg_iov, skb->data, chunk)) {
1888                         skb_queue_head(&sk->sk_receive_queue, skb);
1889                         if (copied == 0)
1890                                 copied = -EFAULT;
1891                         break;
1892                 }
1893                 copied += chunk;
1894                 size -= chunk;
1895
1896                 /* Mark read part of skb as used */
1897                 if (!(flags & MSG_PEEK)) {
1898                         skb_pull(skb, chunk);
1899
1900                         if (UNIXCB(skb).fp)
1901                                 unix_detach_fds(siocb->scm, skb);
1902
1903                         /* put the skb back if we didn't use it up.. */
1904                         if (skb->len) {
1905                                 skb_queue_head(&sk->sk_receive_queue, skb);
1906                                 break;
1907                         }
1908
1909                         consume_skb(skb);
1910
1911                         if (siocb->scm->fp)
1912                                 break;
1913                 } else {
1914                         /* It is questionable, see note in unix_dgram_recvmsg.
1915                          */
1916                         if (UNIXCB(skb).fp)
1917                                 siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1918
1919                         /* put message back and return */
1920                         skb_queue_head(&sk->sk_receive_queue, skb);
1921                         break;
1922                 }
1923         } while (size);
1924
1925         mutex_unlock(&u->readlock);
1926         scm_recv(sock, msg, siocb->scm, flags);
1927 out:
1928         return copied ? : err;
1929 }
1930
1931 static int unix_shutdown(struct socket *sock, int mode)
1932 {
1933         struct sock *sk = sock->sk;
1934         struct sock *other;
1935
1936         mode = (mode+1)&(RCV_SHUTDOWN|SEND_SHUTDOWN);
1937
1938         if (mode) {
1939                 unix_state_lock(sk);
1940                 sk->sk_shutdown |= mode;
1941                 other = unix_peer(sk);
1942                 if (other)
1943                         sock_hold(other);
1944                 unix_state_unlock(sk);
1945                 sk->sk_state_change(sk);
1946
1947                 if (other &&
1948                         (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET)) {
1949
1950                         int peer_mode = 0;
1951
1952                         if (mode&RCV_SHUTDOWN)
1953                                 peer_mode |= SEND_SHUTDOWN;
1954                         if (mode&SEND_SHUTDOWN)
1955                                 peer_mode |= RCV_SHUTDOWN;
1956                         unix_state_lock(other);
1957                         other->sk_shutdown |= peer_mode;
1958                         unix_state_unlock(other);
1959                         other->sk_state_change(other);
1960                         if (peer_mode == SHUTDOWN_MASK)
1961                                 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_HUP);
1962                         else if (peer_mode & RCV_SHUTDOWN)
1963                                 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_IN);
1964                 }
1965                 if (other)
1966                         sock_put(other);
1967         }
1968         return 0;
1969 }
1970
1971 static int unix_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1972 {
1973         struct sock *sk = sock->sk;
1974         long amount = 0;
1975         int err;
1976
1977         switch (cmd) {
1978         case SIOCOUTQ:
1979                 amount = sk_wmem_alloc_get(sk);
1980                 err = put_user(amount, (int __user *)arg);
1981                 break;
1982         case SIOCINQ:
1983                 {
1984                         struct sk_buff *skb;
1985
1986                         if (sk->sk_state == TCP_LISTEN) {
1987                                 err = -EINVAL;
1988                                 break;
1989                         }
1990
1991                         spin_lock(&sk->sk_receive_queue.lock);
1992                         if (sk->sk_type == SOCK_STREAM ||
1993                             sk->sk_type == SOCK_SEQPACKET) {
1994                                 skb_queue_walk(&sk->sk_receive_queue, skb)
1995                                         amount += skb->len;
1996                         } else {
1997                                 skb = skb_peek(&sk->sk_receive_queue);
1998                                 if (skb)
1999                                         amount = skb->len;
2000                         }
2001                         spin_unlock(&sk->sk_receive_queue.lock);
2002                         err = put_user(amount, (int __user *)arg);
2003                         break;
2004                 }
2005
2006         default:
2007                 err = -ENOIOCTLCMD;
2008                 break;
2009         }
2010         return err;
2011 }
2012
2013 static unsigned int unix_poll(struct file *file, struct socket *sock, poll_table *wait)
2014 {
2015         struct sock *sk = sock->sk;
2016         unsigned int mask;
2017
2018         sock_poll_wait(file, sk_sleep(sk), wait);
2019         mask = 0;
2020
2021         /* exceptional events? */
2022         if (sk->sk_err)
2023                 mask |= POLLERR;
2024         if (sk->sk_shutdown == SHUTDOWN_MASK)
2025                 mask |= POLLHUP;
2026         if (sk->sk_shutdown & RCV_SHUTDOWN)
2027                 mask |= POLLRDHUP;
2028
2029         /* readable? */
2030         if (!skb_queue_empty(&sk->sk_receive_queue) ||
2031             (sk->sk_shutdown & RCV_SHUTDOWN))
2032                 mask |= POLLIN | POLLRDNORM;
2033
2034         /* Connection-based need to check for termination and startup */
2035         if ((sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) &&
2036             sk->sk_state == TCP_CLOSE)
2037                 mask |= POLLHUP;
2038
2039         /*
2040          * we set writable also when the other side has shut down the
2041          * connection. This prevents stuck sockets.
2042          */
2043         if (unix_writable(sk))
2044                 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2045
2046         return mask;
2047 }
2048
2049 static unsigned int unix_dgram_poll(struct file *file, struct socket *sock,
2050                                     poll_table *wait)
2051 {
2052         struct sock *sk = sock->sk, *other;
2053         unsigned int mask, writable;
2054
2055         sock_poll_wait(file, sk_sleep(sk), wait);
2056         mask = 0;
2057
2058         /* exceptional events? */
2059         if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue))
2060                 mask |= POLLERR;
2061         if (sk->sk_shutdown & RCV_SHUTDOWN)
2062                 mask |= POLLRDHUP;
2063         if (sk->sk_shutdown == SHUTDOWN_MASK)
2064                 mask |= POLLHUP;
2065
2066         /* readable? */
2067         if (!skb_queue_empty(&sk->sk_receive_queue) ||
2068             (sk->sk_shutdown & RCV_SHUTDOWN))
2069                 mask |= POLLIN | POLLRDNORM;
2070
2071         /* Connection-based need to check for termination and startup */
2072         if (sk->sk_type == SOCK_SEQPACKET) {
2073                 if (sk->sk_state == TCP_CLOSE)
2074                         mask |= POLLHUP;
2075                 /* connection hasn't started yet? */
2076                 if (sk->sk_state == TCP_SYN_SENT)
2077                         return mask;
2078         }
2079
2080         /* writable? */
2081         writable = unix_writable(sk);
2082         if (writable) {
2083                 other = unix_peer_get(sk);
2084                 if (other) {
2085                         if (unix_peer(other) != sk) {
2086                                 sock_poll_wait(file, &unix_sk(other)->peer_wait,
2087                                           wait);
2088                                 if (unix_recvq_full(other))
2089                                         writable = 0;
2090                         }
2091
2092                         sock_put(other);
2093                 }
2094         }
2095
2096         if (writable)
2097                 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2098         else
2099                 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
2100
2101         return mask;
2102 }
2103
2104 #ifdef CONFIG_PROC_FS
2105 static struct sock *first_unix_socket(int *i)
2106 {
2107         for (*i = 0; *i <= UNIX_HASH_SIZE; (*i)++) {
2108                 if (!hlist_empty(&unix_socket_table[*i]))
2109                         return __sk_head(&unix_socket_table[*i]);
2110         }
2111         return NULL;
2112 }
2113
2114 static struct sock *next_unix_socket(int *i, struct sock *s)
2115 {
2116         struct sock *next = sk_next(s);
2117         /* More in this chain? */
2118         if (next)
2119                 return next;
2120         /* Look for next non-empty chain. */
2121         for ((*i)++; *i <= UNIX_HASH_SIZE; (*i)++) {
2122                 if (!hlist_empty(&unix_socket_table[*i]))
2123                         return __sk_head(&unix_socket_table[*i]);
2124         }
2125         return NULL;
2126 }
2127
2128 struct unix_iter_state {
2129         struct seq_net_private p;
2130         int i;
2131 };
2132
2133 static struct sock *unix_seq_idx(struct seq_file *seq, loff_t pos)
2134 {
2135         struct unix_iter_state *iter = seq->private;
2136         loff_t off = 0;
2137         struct sock *s;
2138
2139         for (s = first_unix_socket(&iter->i); s; s = next_unix_socket(&iter->i, s)) {
2140                 if (sock_net(s) != seq_file_net(seq))
2141                         continue;
2142                 if (off == pos)
2143                         return s;
2144                 ++off;
2145         }
2146         return NULL;
2147 }
2148
2149 static void *unix_seq_start(struct seq_file *seq, loff_t *pos)
2150         __acquires(unix_table_lock)
2151 {
2152         spin_lock(&unix_table_lock);
2153         return *pos ? unix_seq_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2154 }
2155
2156 static void *unix_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2157 {
2158         struct unix_iter_state *iter = seq->private;
2159         struct sock *sk = v;
2160         ++*pos;
2161
2162         if (v == SEQ_START_TOKEN)
2163                 sk = first_unix_socket(&iter->i);
2164         else
2165                 sk = next_unix_socket(&iter->i, sk);
2166         while (sk && (sock_net(sk) != seq_file_net(seq)))
2167                 sk = next_unix_socket(&iter->i, sk);
2168         return sk;
2169 }
2170
2171 static void unix_seq_stop(struct seq_file *seq, void *v)
2172         __releases(unix_table_lock)
2173 {
2174         spin_unlock(&unix_table_lock);
2175 }
2176
2177 static int unix_seq_show(struct seq_file *seq, void *v)
2178 {
2179
2180         if (v == SEQ_START_TOKEN)
2181                 seq_puts(seq, "Num       RefCount Protocol Flags    Type St "
2182                          "Inode Path\n");
2183         else {
2184                 struct sock *s = v;
2185                 struct unix_sock *u = unix_sk(s);
2186                 unix_state_lock(s);
2187
2188                 seq_printf(seq, "%p: %08X %08X %08X %04X %02X %5lu",
2189                         s,
2190                         atomic_read(&s->sk_refcnt),
2191                         0,
2192                         s->sk_state == TCP_LISTEN ? __SO_ACCEPTCON : 0,
2193                         s->sk_type,
2194                         s->sk_socket ?
2195                         (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTED : SS_UNCONNECTED) :
2196                         (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTING : SS_DISCONNECTING),
2197                         sock_i_ino(s));
2198
2199                 if (u->addr) {
2200                         int i, len;
2201                         seq_putc(seq, ' ');
2202
2203                         i = 0;
2204                         len = u->addr->len - sizeof(short);
2205                         if (!UNIX_ABSTRACT(s))
2206                                 len--;
2207                         else {
2208                                 seq_putc(seq, '@');
2209                                 i++;
2210                         }
2211                         for ( ; i < len; i++)
2212                                 seq_putc(seq, u->addr->name->sun_path[i]);
2213                 }
2214                 unix_state_unlock(s);
2215                 seq_putc(seq, '\n');
2216         }
2217
2218         return 0;
2219 }
2220
2221 static const struct seq_operations unix_seq_ops = {
2222         .start  = unix_seq_start,
2223         .next   = unix_seq_next,
2224         .stop   = unix_seq_stop,
2225         .show   = unix_seq_show,
2226 };
2227
2228 static int unix_seq_open(struct inode *inode, struct file *file)
2229 {
2230         return seq_open_net(inode, file, &unix_seq_ops,
2231                             sizeof(struct unix_iter_state));
2232 }
2233
2234 static const struct file_operations unix_seq_fops = {
2235         .owner          = THIS_MODULE,
2236         .open           = unix_seq_open,
2237         .read           = seq_read,
2238         .llseek         = seq_lseek,
2239         .release        = seq_release_net,
2240 };
2241
2242 #endif
2243
2244 static const struct net_proto_family unix_family_ops = {
2245         .family = PF_UNIX,
2246         .create = unix_create,
2247         .owner  = THIS_MODULE,
2248 };
2249
2250
2251 static int __net_init unix_net_init(struct net *net)
2252 {
2253         int error = -ENOMEM;
2254
2255         net->unx.sysctl_max_dgram_qlen = 10;
2256         if (unix_sysctl_register(net))
2257                 goto out;
2258
2259 #ifdef CONFIG_PROC_FS
2260         if (!proc_net_fops_create(net, "unix", 0, &unix_seq_fops)) {
2261                 unix_sysctl_unregister(net);
2262                 goto out;
2263         }
2264 #endif
2265         error = 0;
2266 out:
2267         return error;
2268 }
2269
2270 static void __net_exit unix_net_exit(struct net *net)
2271 {
2272         unix_sysctl_unregister(net);
2273         proc_net_remove(net, "unix");
2274 }
2275
2276 static struct pernet_operations unix_net_ops = {
2277         .init = unix_net_init,
2278         .exit = unix_net_exit,
2279 };
2280
2281 static int __init af_unix_init(void)
2282 {
2283         int rc = -1;
2284         struct sk_buff *dummy_skb;
2285
2286         BUILD_BUG_ON(sizeof(struct unix_skb_parms) > sizeof(dummy_skb->cb));
2287
2288         rc = proto_register(&unix_proto, 1);
2289         if (rc != 0) {
2290                 printk(KERN_CRIT "%s: Cannot create unix_sock SLAB cache!\n",
2291                        __func__);
2292                 goto out;
2293         }
2294
2295         sock_register(&unix_family_ops);
2296         register_pernet_subsys(&unix_net_ops);
2297 out:
2298         return rc;
2299 }
2300
2301 static void __exit af_unix_exit(void)
2302 {
2303         sock_unregister(PF_UNIX);
2304         proto_unregister(&unix_proto);
2305         unregister_pernet_subsys(&unix_net_ops);
2306 }
2307
2308 /* Earlier than device_initcall() so that other drivers invoking
2309    request_module() don't end up in a loop when modprobe tries
2310    to use a UNIX socket. But later than subsys_initcall() because
2311    we depend on stuff initialised there */
2312 fs_initcall(af_unix_init);
2313 module_exit(af_unix_exit);
2314
2315 MODULE_LICENSE("GPL");
2316 MODULE_ALIAS_NETPROTO(PF_UNIX);