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_long_t unix_nr_socks;
 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_long_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, %ld are still alive.\n", sk,
 369                 atomic_long_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_long_inc(&unix_nr_socks);
 610         if (atomic_long_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_long_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         unsigned int retries = 0;
 696
 697         mutex_lock(&u->readlock);
 698
 699         err = 0;
 700         if (u->addr)
 701                 goto out;
 702
 703         err = -ENOMEM;
 704         addr = kzalloc(sizeof(*addr) + sizeof(short) + 16, GFP_KERNEL);
 705         if (!addr)
 706                 goto out;
 707
 708         addr->name->sun_family = AF_UNIX;
 709         atomic_set(&addr->refcnt, 1);
 710
 711 retry:
 712         addr->len = sprintf(addr->name->sun_path+1, "%05x", ordernum) + 1 + sizeof(short);
 713         addr->hash = unix_hash_fold(csum_partial(addr->name, addr->len, 0));
 714
 715         spin_lock(&unix_table_lock);
 716         ordernum = (ordernum+1)&0xFFFFF;
 717
 718         if (__unix_find_socket_byname(net, addr->name, addr->len, sock->type,
 719                                       addr->hash)) {
 720                 spin_unlock(&unix_table_lock);
 721                 /*
 722                  * __unix_find_socket_byname() may take long time if many names
 723                  * are already in use.
 724                  */
 725                 cond_resched();
 726                 /* Give up if all names seems to be in use. */
 727                 if (retries++ == 0xFFFFF) {
 728                         err = -ENOSPC;
 729                         kfree(addr);
 730                         goto out;
 731                 }
 732                 goto retry;
 733         }
 734         addr->hash ^= sk->sk_type;
 735
 736         __unix_remove_socket(sk);
 737         u->addr = addr;
 738         __unix_insert_socket(&unix_socket_table[addr->hash], sk);
 739         spin_unlock(&unix_table_lock);
 740         err = 0;
 741
 742 out:    mutex_unlock(&u->readlock);
 743         return err;
 744 }
 745
 746 static struct sock *unix_find_other(struct net *net,
 747                                     struct sockaddr_un *sunname, int len,
 748                                     int type, unsigned hash, int *error)
 749 {
 750         struct sock *u;
 751         struct path path;
 752         int err = 0;
 753
 754         if (sunname->sun_path[0]) {
 755                 struct inode *inode;
 756                 err = kern_path(sunname->sun_path, LOOKUP_FOLLOW, &path);
 757                 if (err)
 758                         goto fail;
 759                 inode = path.dentry->d_inode;
 760                 err = inode_permission(inode, MAY_WRITE);
 761                 if (err)
 762                         goto put_fail;
 763
 764                 err = -ECONNREFUSED;
 765                 if (!S_ISSOCK(inode->i_mode))
 766                         goto put_fail;
 767                 u = unix_find_socket_byinode(inode);
 768                 if (!u)
 769                         goto put_fail;
 770
 771                 if (u->sk_type == type)
 772                         touch_atime(path.mnt, path.dentry);
 773
 774                 path_put(&path);
 775
 776                 err = -EPROTOTYPE;
 777                 if (u->sk_type != type) {
 778                         sock_put(u);
 779                         goto fail;
 780                 }
 781         } else {
 782                 err = -ECONNREFUSED;
 783                 u = unix_find_socket_byname(net, sunname, len, type, hash);
 784                 if (u) {
 785                         struct dentry *dentry;
 786                         dentry = unix_sk(u)->dentry;
 787                         if (dentry)
 788                                 touch_atime(unix_sk(u)->mnt, dentry);
 789                 } else
 790                         goto fail;
 791         }
 792         return u;
 793
 794 put_fail:
 795         path_put(&path);
 796 fail:
 797         *error = err;
 798         return NULL;
 799 }
 800
 801
 802 static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
 803 {
 804         struct sock *sk = sock->sk;
 805         struct net *net = sock_net(sk);
 806         struct unix_sock *u = unix_sk(sk);
 807         struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
 808         struct dentry *dentry = NULL;
 809         struct nameidata nd;
 810         int err;
 811         unsigned hash;
 812         struct unix_address *addr;
 813         struct hlist_head *list;
 814
 815         err = -EINVAL;
 816         if (sunaddr->sun_family != AF_UNIX)
 817                 goto out;
 818
 819         if (addr_len == sizeof(short)) {
 820                 err = unix_autobind(sock);
 821                 goto out;
 822         }
 823
 824         err = unix_mkname(sunaddr, addr_len, &hash);
 825         if (err < 0)
 826                 goto out;
 827         addr_len = err;
 828
 829         mutex_lock(&u->readlock);
 830
 831         err = -EINVAL;
 832         if (u->addr)
 833                 goto out_up;
 834
 835         err = -ENOMEM;
 836         addr = kmalloc(sizeof(*addr)+addr_len, GFP_KERNEL);
 837         if (!addr)
 838                 goto out_up;
 839
 840         memcpy(addr->name, sunaddr, addr_len);
 841         addr->len = addr_len;
 842         addr->hash = hash ^ sk->sk_type;
 843         atomic_set(&addr->refcnt, 1);
 844
 845         if (sunaddr->sun_path[0]) {
 846                 unsigned int mode;
 847                 err = 0;
 848                 /*
 849                  * Get the parent directory, calculate the hash for last
 850                  * component.
 851                  */
 852                 err = path_lookup(sunaddr->sun_path, LOOKUP_PARENT, &nd);
 853                 if (err)
 854                         goto out_mknod_parent;
 855
 856                 dentry = lookup_create(&nd, 0);
 857                 err = PTR_ERR(dentry);
 858                 if (IS_ERR(dentry))
 859                         goto out_mknod_unlock;
 860
 861                 /*
 862                  * All right, let's create it.
 863                  */
 864                 mode = S_IFSOCK |
 865                        (SOCK_INODE(sock)->i_mode & ~current_umask());
 866                 err = mnt_want_write(nd.path.mnt);
 867                 if (err)
 868                         goto out_mknod_dput;
 869                 err = security_path_mknod(&nd.path, dentry, mode, 0);
 870                 if (err)
 871                         goto out_mknod_drop_write;
 872                 err = vfs_mknod(nd.path.dentry->d_inode, dentry, mode, 0);
 873 out_mknod_drop_write:
 874                 mnt_drop_write(nd.path.mnt);
 875                 if (err)
 876                         goto out_mknod_dput;
 877                 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
 878                 dput(nd.path.dentry);
 879                 nd.path.dentry = dentry;
 880
 881                 addr->hash = UNIX_HASH_SIZE;
 882         }
 883
 884         spin_lock(&unix_table_lock);
 885
 886         if (!sunaddr->sun_path[0]) {
 887                 err = -EADDRINUSE;
 888                 if (__unix_find_socket_byname(net, sunaddr, addr_len,
 889                                               sk->sk_type, hash)) {
 890                         unix_release_addr(addr);
 891                         goto out_unlock;
 892                 }
 893
 894                 list = &unix_socket_table[addr->hash];
 895         } else {
 896                 list = &unix_socket_table[dentry->d_inode->i_ino & (UNIX_HASH_SIZE-1)];
 897                 u->dentry = nd.path.dentry;
 898                 u->mnt    = nd.path.mnt;
 899         }
 900
 901         err = 0;
 902         __unix_remove_socket(sk);
 903         u->addr = addr;
 904         __unix_insert_socket(list, sk);
 905
 906 out_unlock:
 907         spin_unlock(&unix_table_lock);
 908 out_up:
 909         mutex_unlock(&u->readlock);
 910 out:
 911         return err;
 912
 913 out_mknod_dput:
 914         dput(dentry);
 915 out_mknod_unlock:
 916         mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
 917         path_put(&nd.path);
 918 out_mknod_parent:
 919         if (err == -EEXIST)
 920                 err = -EADDRINUSE;
 921         unix_release_addr(addr);
 922         goto out_up;
 923 }
 924
 925 static void unix_state_double_lock(struct sock *sk1, struct sock *sk2)
 926 {
 927         if (unlikely(sk1 == sk2) || !sk2) {
 928                 unix_state_lock(sk1);
 929                 return;
 930         }
 931         if (sk1 < sk2) {
 932                 unix_state_lock(sk1);
 933                 unix_state_lock_nested(sk2);
 934         } else {
 935                 unix_state_lock(sk2);
 936                 unix_state_lock_nested(sk1);
 937         }
 938 }
 939
 940 static void unix_state_double_unlock(struct sock *sk1, struct sock *sk2)
 941 {
 942         if (unlikely(sk1 == sk2) || !sk2) {
 943                 unix_state_unlock(sk1);
 944                 return;
 945         }
 946         unix_state_unlock(sk1);
 947         unix_state_unlock(sk2);
 948 }
 949
 950 static int unix_dgram_connect(struct socket *sock, struct sockaddr *addr,
 951                               int alen, int flags)
 952 {
 953         struct sock *sk = sock->sk;
 954         struct net *net = sock_net(sk);
 955         struct sockaddr_un *sunaddr = (struct sockaddr_un *)addr;
 956         struct sock *other;
 957         unsigned hash;
 958         int err;
 959
 960         if (addr->sa_family != AF_UNSPEC) {
 961                 err = unix_mkname(sunaddr, alen, &hash);
 962                 if (err < 0)
 963                         goto out;
 964                 alen = err;
 965
 966                 if (test_bit(SOCK_PASSCRED, &sock->flags) &&
 967                     !unix_sk(sk)->addr && (err = unix_autobind(sock)) != 0)
 968                         goto out;
 969
 970 restart:
 971                 other = unix_find_other(net, sunaddr, alen, sock->type, hash, &err);
 972                 if (!other)
 973                         goto out;
 974
 975                 unix_state_double_lock(sk, other);
 976
 977                 /* Apparently VFS overslept socket death. Retry. */
 978                 if (sock_flag(other, SOCK_DEAD)) {
 979                         unix_state_double_unlock(sk, other);
 980                         sock_put(other);
 981                         goto restart;
 982                 }
 983
 984                 err = -EPERM;
 985                 if (!unix_may_send(sk, other))
 986                         goto out_unlock;
 987
 988                 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
 989                 if (err)
 990                         goto out_unlock;
 991
 992         } else {
 993                 /*
 994                  *      1003.1g breaking connected state with AF_UNSPEC
 995                  */
 996                 other = NULL;
 997                 unix_state_double_lock(sk, other);
 998         }
 999
1000         /*
1001          * If it was connected, reconnect.
1002          */
1003         if (unix_peer(sk)) {
1004                 struct sock *old_peer = unix_peer(sk);
1005                 unix_peer(sk) = other;
1006                 unix_state_double_unlock(sk, other);
1007
1008                 if (other != old_peer)
1009                         unix_dgram_disconnected(sk, old_peer);
1010                 sock_put(old_peer);
1011         } else {
1012                 unix_peer(sk) = other;
1013                 unix_state_double_unlock(sk, other);
1014         }
1015         return 0;
1016
1017 out_unlock:
1018         unix_state_double_unlock(sk, other);
1019         sock_put(other);
1020 out:
1021         return err;
1022 }
1023
1024 static long unix_wait_for_peer(struct sock *other, long timeo)
1025 {
1026         struct unix_sock *u = unix_sk(other);
1027         int sched;
1028         DEFINE_WAIT(wait);
1029
1030         prepare_to_wait_exclusive(&u->peer_wait, &wait, TASK_INTERRUPTIBLE);
1031
1032         sched = !sock_flag(other, SOCK_DEAD) &&
1033                 !(other->sk_shutdown & RCV_SHUTDOWN) &&
1034                 unix_recvq_full(other);
1035
1036         unix_state_unlock(other);
1037
1038         if (sched)
1039                 timeo = schedule_timeout(timeo);
1040
1041         finish_wait(&u->peer_wait, &wait);
1042         return timeo;
1043 }
1044
1045 static int unix_stream_connect(struct socket *sock, struct sockaddr *uaddr,
1046                                int addr_len, int flags)
1047 {
1048         struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
1049         struct sock *sk = sock->sk;
1050         struct net *net = sock_net(sk);
1051         struct unix_sock *u = unix_sk(sk), *newu, *otheru;
1052         struct sock *newsk = NULL;
1053         struct sock *other = NULL;
1054         struct sk_buff *skb = NULL;
1055         unsigned hash;
1056         int st;
1057         int err;
1058         long timeo;
1059
1060         err = unix_mkname(sunaddr, addr_len, &hash);
1061         if (err < 0)
1062                 goto out;
1063         addr_len = err;
1064
1065         if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr &&
1066             (err = unix_autobind(sock)) != 0)
1067                 goto out;
1068
1069         timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
1070
1071         /* First of all allocate resources.
1072            If we will make it after state is locked,
1073            we will have to recheck all again in any case.
1074          */
1075
1076         err = -ENOMEM;
1077
1078         /* create new sock for complete connection */
1079         newsk = unix_create1(sock_net(sk), NULL);
1080         if (newsk == NULL)
1081                 goto out;
1082
1083         /* Allocate skb for sending to listening sock */
1084         skb = sock_wmalloc(newsk, 1, 0, GFP_KERNEL);
1085         if (skb == NULL)
1086                 goto out;
1087
1088 restart:
1089         /*  Find listening sock. */
1090         other = unix_find_other(net, sunaddr, addr_len, sk->sk_type, hash, &err);
1091         if (!other)
1092                 goto out;
1093
1094         /* Latch state of peer */
1095         unix_state_lock(other);
1096
1097         /* Apparently VFS overslept socket death. Retry. */
1098         if (sock_flag(other, SOCK_DEAD)) {
1099                 unix_state_unlock(other);
1100                 sock_put(other);
1101                 goto restart;
1102         }
1103
1104         err = -ECONNREFUSED;
1105         if (other->sk_state != TCP_LISTEN)
1106                 goto out_unlock;
1107         if (other->sk_shutdown & RCV_SHUTDOWN)
1108                 goto out_unlock;
1109
1110         if (unix_recvq_full(other)) {
1111                 err = -EAGAIN;
1112                 if (!timeo)
1113                         goto out_unlock;
1114
1115                 timeo = unix_wait_for_peer(other, timeo);
1116
1117                 err = sock_intr_errno(timeo);
1118                 if (signal_pending(current))
1119                         goto out;
1120                 sock_put(other);
1121                 goto restart;
1122         }
1123
1124         /* Latch our state.
1125
1126            It is tricky place. We need to grab write lock and cannot
1127            drop lock on peer. It is dangerous because deadlock is
1128            possible. Connect to self case and simultaneous
1129            attempt to connect are eliminated by checking socket
1130            state. other is TCP_LISTEN, if sk is TCP_LISTEN we
1131            check this before attempt to grab lock.
1132
1133            Well, and we have to recheck the state after socket locked.
1134          */
1135         st = sk->sk_state;
1136
1137         switch (st) {
1138         case TCP_CLOSE:
1139                 /* This is ok... continue with connect */
1140                 break;
1141         case TCP_ESTABLISHED:
1142                 /* Socket is already connected */
1143                 err = -EISCONN;
1144                 goto out_unlock;
1145         default:
1146                 err = -EINVAL;
1147                 goto out_unlock;
1148         }
1149
1150         unix_state_lock_nested(sk);
1151
1152         if (sk->sk_state != st) {
1153                 unix_state_unlock(sk);
1154                 unix_state_unlock(other);
1155                 sock_put(other);
1156                 goto restart;
1157         }
1158
1159         err = security_unix_stream_connect(sock, other->sk_socket, newsk);
1160         if (err) {
1161                 unix_state_unlock(sk);
1162                 goto out_unlock;
1163         }
1164
1165         /* The way is open! Fastly set all the necessary fields... */
1166
1167         sock_hold(sk);
1168         unix_peer(newsk)        = sk;
1169         newsk->sk_state         = TCP_ESTABLISHED;
1170         newsk->sk_type          = sk->sk_type;
1171         init_peercred(newsk);
1172         newu = unix_sk(newsk);
1173         newsk->sk_wq            = &newu->peer_wq;
1174         otheru = unix_sk(other);
1175
1176         /* copy address information from listening to new sock*/
1177         if (otheru->addr) {
1178                 atomic_inc(&otheru->addr->refcnt);
1179                 newu->addr = otheru->addr;
1180         }
1181         if (otheru->dentry) {
1182                 newu->dentry    = dget(otheru->dentry);
1183                 newu->mnt       = mntget(otheru->mnt);
1184         }
1185
1186         /* Set credentials */
1187         copy_peercred(sk, other);
1188
1189         sock->state     = SS_CONNECTED;
1190         sk->sk_state    = TCP_ESTABLISHED;
1191         sock_hold(newsk);
1192
1193         smp_mb__after_atomic_inc();     /* sock_hold() does an atomic_inc() */
1194         unix_peer(sk)   = newsk;
1195
1196         unix_state_unlock(sk);
1197
1198         /* take ten and and send info to listening sock */
1199         spin_lock(&other->sk_receive_queue.lock);
1200         __skb_queue_tail(&other->sk_receive_queue, skb);
1201         spin_unlock(&other->sk_receive_queue.lock);
1202         unix_state_unlock(other);
1203         other->sk_data_ready(other, 0);
1204         sock_put(other);
1205         return 0;
1206
1207 out_unlock:
1208         if (other)
1209                 unix_state_unlock(other);
1210
1211 out:
1212         kfree_skb(skb);
1213         if (newsk)
1214                 unix_release_sock(newsk, 0);
1215         if (other)
1216                 sock_put(other);
1217         return err;
1218 }
1219
1220 static int unix_socketpair(struct socket *socka, struct socket *sockb)
1221 {
1222         struct sock *ska = socka->sk, *skb = sockb->sk;
1223
1224         /* Join our sockets back to back */
1225         sock_hold(ska);
1226         sock_hold(skb);
1227         unix_peer(ska) = skb;
1228         unix_peer(skb) = ska;
1229         init_peercred(ska);
1230         init_peercred(skb);
1231
1232         if (ska->sk_type != SOCK_DGRAM) {
1233                 ska->sk_state = TCP_ESTABLISHED;
1234                 skb->sk_state = TCP_ESTABLISHED;
1235                 socka->state  = SS_CONNECTED;
1236                 sockb->state  = SS_CONNECTED;
1237         }
1238         return 0;
1239 }
1240
1241 static int unix_accept(struct socket *sock, struct socket *newsock, int flags)
1242 {
1243         struct sock *sk = sock->sk;
1244         struct sock *tsk;
1245         struct sk_buff *skb;
1246         int err;
1247
1248         err = -EOPNOTSUPP;
1249         if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
1250                 goto out;
1251
1252         err = -EINVAL;
1253         if (sk->sk_state != TCP_LISTEN)
1254                 goto out;
1255
1256         /* If socket state is TCP_LISTEN it cannot change (for now...),
1257          * so that no locks are necessary.
1258          */
1259
1260         skb = skb_recv_datagram(sk, 0, flags&O_NONBLOCK, &err);
1261         if (!skb) {
1262                 /* This means receive shutdown. */
1263                 if (err == 0)
1264                         err = -EINVAL;
1265                 goto out;
1266         }
1267
1268         tsk = skb->sk;
1269         skb_free_datagram(sk, skb);
1270         wake_up_interruptible(&unix_sk(sk)->peer_wait);
1271
1272         /* attach accepted sock to socket */
1273         unix_state_lock(tsk);
1274         newsock->state = SS_CONNECTED;
1275         sock_graft(tsk, newsock);
1276         unix_state_unlock(tsk);
1277         return 0;
1278
1279 out:
1280         return err;
1281 }
1282
1283
1284 static int unix_getname(struct socket *sock, struct sockaddr *uaddr, int *uaddr_len, int peer)
1285 {
1286         struct sock *sk = sock->sk;
1287         struct unix_sock *u;
1288         DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr, uaddr);
1289         int err = 0;
1290
1291         if (peer) {
1292                 sk = unix_peer_get(sk);
1293
1294                 err = -ENOTCONN;
1295                 if (!sk)
1296                         goto out;
1297                 err = 0;
1298         } else {
1299                 sock_hold(sk);
1300         }
1301
1302         u = unix_sk(sk);
1303         unix_state_lock(sk);
1304         if (!u->addr) {
1305                 sunaddr->sun_family = AF_UNIX;
1306                 sunaddr->sun_path[0] = 0;
1307                 *uaddr_len = sizeof(short);
1308         } else {
1309                 struct unix_address *addr = u->addr;
1310
1311                 *uaddr_len = addr->len;
1312                 memcpy(sunaddr, addr->name, *uaddr_len);
1313         }
1314         unix_state_unlock(sk);
1315         sock_put(sk);
1316 out:
1317         return err;
1318 }
1319
1320 static void unix_detach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1321 {
1322         int i;
1323
1324         scm->fp = UNIXCB(skb).fp;
1325         UNIXCB(skb).fp = NULL;
1326
1327         for (i = scm->fp->count-1; i >= 0; i--)
1328                 unix_notinflight(scm->fp->fp[i]);
1329 }
1330
1331 static void unix_destruct_scm(struct sk_buff *skb)
1332 {
1333         struct scm_cookie scm;
1334         memset(&scm, 0, sizeof(scm));
1335         scm.pid  = UNIXCB(skb).pid;
1336         scm.cred = UNIXCB(skb).cred;
1337         if (UNIXCB(skb).fp)
1338                 unix_detach_fds(&scm, skb);
1339
1340         /* Alas, it calls VFS */
1341         /* So fscking what? fput() had been SMP-safe since the last Summer */
1342         scm_destroy(&scm);
1343         sock_wfree(skb);
1344 }
1345
1346 static int unix_attach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1347 {
1348         int i;
1349
1350         /*
1351          * Need to duplicate file references for the sake of garbage
1352          * collection.  Otherwise a socket in the fps might become a
1353          * candidate for GC while the skb is not yet queued.
1354          */
1355         UNIXCB(skb).fp = scm_fp_dup(scm->fp);
1356         if (!UNIXCB(skb).fp)
1357                 return -ENOMEM;
1358
1359         for (i = scm->fp->count-1; i >= 0; i--)
1360                 unix_inflight(scm->fp->fp[i]);
1361         return 0;
1362 }
1363
1364 static int unix_scm_to_skb(struct scm_cookie *scm, struct sk_buff *skb, bool send_fds)
1365 {
1366         int err = 0;
1367         UNIXCB(skb).pid  = get_pid(scm->pid);
1368         UNIXCB(skb).cred = get_cred(scm->cred);
1369         UNIXCB(skb).fp = NULL;
1370         if (scm->fp && send_fds)
1371                 err = unix_attach_fds(scm, skb);
1372
1373         skb->destructor = unix_destruct_scm;
1374         return err;
1375 }
1376
1377 /*
1378  *      Send AF_UNIX data.
1379  */
1380
1381 static int unix_dgram_sendmsg(struct kiocb *kiocb, struct socket *sock,
1382                               struct msghdr *msg, size_t len)
1383 {
1384         struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1385         struct sock *sk = sock->sk;
1386         struct net *net = sock_net(sk);
1387         struct unix_sock *u = unix_sk(sk);
1388         struct sockaddr_un *sunaddr = msg->msg_name;
1389         struct sock *other = NULL;
1390         int namelen = 0; /* fake GCC */
1391         int err;
1392         unsigned hash;
1393         struct sk_buff *skb;
1394         long timeo;
1395         struct scm_cookie tmp_scm;
1396
1397         if (NULL == siocb->scm)
1398                 siocb->scm = &tmp_scm;
1399         wait_for_unix_gc();
1400         err = scm_send(sock, msg, siocb->scm);
1401         if (err < 0)
1402                 return err;
1403
1404         err = -EOPNOTSUPP;
1405         if (msg->msg_flags&MSG_OOB)
1406                 goto out;
1407
1408         if (msg->msg_namelen) {
1409                 err = unix_mkname(sunaddr, msg->msg_namelen, &hash);
1410                 if (err < 0)
1411                         goto out;
1412                 namelen = err;
1413         } else {
1414                 sunaddr = NULL;
1415                 err = -ENOTCONN;
1416                 other = unix_peer_get(sk);
1417                 if (!other)
1418                         goto out;
1419         }
1420
1421         if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr
1422             && (err = unix_autobind(sock)) != 0)
1423                 goto out;
1424
1425         err = -EMSGSIZE;
1426         if (len > sk->sk_sndbuf - 32)
1427                 goto out;
1428
1429         skb = sock_alloc_send_skb(sk, len, msg->msg_flags&MSG_DONTWAIT, &err);
1430         if (skb == NULL)
1431                 goto out;
1432
1433         err = unix_scm_to_skb(siocb->scm, skb, true);
1434         if (err)
1435                 goto out_free;
1436         unix_get_secdata(siocb->scm, skb);
1437
1438         skb_reset_transport_header(skb);
1439         err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
1440         if (err)
1441                 goto out_free;
1442
1443         timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1444
1445 restart:
1446         if (!other) {
1447                 err = -ECONNRESET;
1448                 if (sunaddr == NULL)
1449                         goto out_free;
1450
1451                 other = unix_find_other(net, sunaddr, namelen, sk->sk_type,
1452                                         hash, &err);
1453                 if (other == NULL)
1454                         goto out_free;
1455         }
1456
1457         unix_state_lock(other);
1458         err = -EPERM;
1459         if (!unix_may_send(sk, other))
1460                 goto out_unlock;
1461
1462         if (sock_flag(other, SOCK_DEAD)) {
1463                 /*
1464                  *      Check with 1003.1g - what should
1465                  *      datagram error
1466                  */
1467                 unix_state_unlock(other);
1468                 sock_put(other);
1469
1470                 err = 0;
1471                 unix_state_lock(sk);
1472                 if (unix_peer(sk) == other) {
1473                         unix_peer(sk) = NULL;
1474                         unix_state_unlock(sk);
1475
1476                         unix_dgram_disconnected(sk, other);
1477                         sock_put(other);
1478                         err = -ECONNREFUSED;
1479                 } else {
1480                         unix_state_unlock(sk);
1481                 }
1482
1483                 other = NULL;
1484                 if (err)
1485                         goto out_free;
1486                 goto restart;
1487         }
1488
1489         err = -EPIPE;
1490         if (other->sk_shutdown & RCV_SHUTDOWN)
1491                 goto out_unlock;
1492
1493         if (sk->sk_type != SOCK_SEQPACKET) {
1494                 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1495                 if (err)
1496                         goto out_unlock;
1497         }
1498
1499         if (unix_peer(other) != sk && unix_recvq_full(other)) {
1500                 if (!timeo) {
1501                         err = -EAGAIN;
1502                         goto out_unlock;
1503                 }
1504
1505                 timeo = unix_wait_for_peer(other, timeo);
1506
1507                 err = sock_intr_errno(timeo);
1508                 if (signal_pending(current))
1509                         goto out_free;
1510
1511                 goto restart;
1512         }
1513
1514         if (sock_flag(other, SOCK_RCVTSTAMP))
1515                 __net_timestamp(skb);
1516         skb_queue_tail(&other->sk_receive_queue, skb);
1517         unix_state_unlock(other);
1518         other->sk_data_ready(other, len);
1519         sock_put(other);
1520         scm_destroy(siocb->scm);
1521         return len;
1522
1523 out_unlock:
1524         unix_state_unlock(other);
1525 out_free:
1526         kfree_skb(skb);
1527 out:
1528         if (other)
1529                 sock_put(other);
1530         scm_destroy(siocb->scm);
1531         return err;
1532 }
1533
1534
1535 static int unix_stream_sendmsg(struct kiocb *kiocb, struct socket *sock,
1536                                struct msghdr *msg, size_t len)
1537 {
1538         struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1539         struct sock *sk = sock->sk;
1540         struct sock *other = NULL;
1541         struct sockaddr_un *sunaddr = msg->msg_name;
1542         int err, size;
1543         struct sk_buff *skb;
1544         int sent = 0;
1545         struct scm_cookie tmp_scm;
1546         bool fds_sent = false;
1547
1548         if (NULL == siocb->scm)
1549                 siocb->scm = &tmp_scm;
1550         wait_for_unix_gc();
1551         err = scm_send(sock, msg, siocb->scm);
1552         if (err < 0)
1553                 return err;
1554
1555         err = -EOPNOTSUPP;
1556         if (msg->msg_flags&MSG_OOB)
1557                 goto out_err;
1558
1559         if (msg->msg_namelen) {
1560                 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
1561                 goto out_err;
1562         } else {
1563                 sunaddr = NULL;
1564                 err = -ENOTCONN;
1565                 other = unix_peer(sk);
1566                 if (!other)
1567                         goto out_err;
1568         }
1569
1570         if (sk->sk_shutdown & SEND_SHUTDOWN)
1571                 goto pipe_err;
1572
1573         while (sent < len) {
1574                 /*
1575                  *      Optimisation for the fact that under 0.01% of X
1576                  *      messages typically need breaking up.
1577                  */
1578
1579                 size = len-sent;
1580
1581                 /* Keep two messages in the pipe so it schedules better */
1582                 if (size > ((sk->sk_sndbuf >> 1) - 64))
1583                         size = (sk->sk_sndbuf >> 1) - 64;
1584
1585                 if (size > SKB_MAX_ALLOC)
1586                         size = SKB_MAX_ALLOC;
1587
1588                 /*
1589                  *      Grab a buffer
1590                  */
1591
1592                 skb = sock_alloc_send_skb(sk, size, msg->msg_flags&MSG_DONTWAIT,
1593                                           &err);
1594
1595                 if (skb == NULL)
1596                         goto out_err;
1597
1598                 /*
1599                  *      If you pass two values to the sock_alloc_send_skb
1600                  *      it tries to grab the large buffer with GFP_NOFS
1601                  *      (which can fail easily), and if it fails grab the
1602                  *      fallback size buffer which is under a page and will
1603                  *      succeed. [Alan]
1604                  */
1605                 size = min_t(int, size, skb_tailroom(skb));
1606
1607
1608                 /* Only send the fds in the first buffer */
1609                 err = unix_scm_to_skb(siocb->scm, skb, !fds_sent);
1610                 if (err) {
1611                         kfree_skb(skb);
1612                         goto out_err;
1613                 }
1614                 fds_sent = true;
1615
1616                 err = memcpy_fromiovec(skb_put(skb, size), msg->msg_iov, size);
1617                 if (err) {
1618                         kfree_skb(skb);
1619                         goto out_err;
1620                 }
1621
1622                 unix_state_lock(other);
1623
1624                 if (sock_flag(other, SOCK_DEAD) ||
1625                     (other->sk_shutdown & RCV_SHUTDOWN))
1626                         goto pipe_err_free;
1627
1628                 skb_queue_tail(&other->sk_receive_queue, skb);
1629                 unix_state_unlock(other);
1630                 other->sk_data_ready(other, size);
1631                 sent += size;
1632         }
1633
1634         scm_destroy(siocb->scm);
1635         siocb->scm = NULL;
1636
1637         return sent;
1638
1639 pipe_err_free:
1640         unix_state_unlock(other);
1641         kfree_skb(skb);
1642 pipe_err:
1643         if (sent == 0 && !(msg->msg_flags&MSG_NOSIGNAL))
1644                 send_sig(SIGPIPE, current, 0);
1645         err = -EPIPE;
1646 out_err:
1647         scm_destroy(siocb->scm);
1648         siocb->scm = NULL;
1649         return sent ? : err;
1650 }
1651
1652 static int unix_seqpacket_sendmsg(struct kiocb *kiocb, struct socket *sock,
1653                                   struct msghdr *msg, size_t len)
1654 {
1655         int err;
1656         struct sock *sk = sock->sk;
1657
1658         err = sock_error(sk);
1659         if (err)
1660                 return err;
1661
1662         if (sk->sk_state != TCP_ESTABLISHED)
1663                 return -ENOTCONN;
1664
1665         if (msg->msg_namelen)
1666                 msg->msg_namelen = 0;
1667
1668         return unix_dgram_sendmsg(kiocb, sock, msg, len);
1669 }
1670
1671 static void unix_copy_addr(struct msghdr *msg, struct sock *sk)
1672 {
1673         struct unix_sock *u = unix_sk(sk);
1674
1675         msg->msg_namelen = 0;
1676         if (u->addr) {
1677                 msg->msg_namelen = u->addr->len;
1678                 memcpy(msg->msg_name, u->addr->name, u->addr->len);
1679         }
1680 }
1681
1682 static int unix_dgram_recvmsg(struct kiocb *iocb, struct socket *sock,
1683                               struct msghdr *msg, size_t size,
1684                               int flags)
1685 {
1686         struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1687         struct scm_cookie tmp_scm;
1688         struct sock *sk = sock->sk;
1689         struct unix_sock *u = unix_sk(sk);
1690         int noblock = flags & MSG_DONTWAIT;
1691         struct sk_buff *skb;
1692         int err;
1693
1694         err = -EOPNOTSUPP;
1695         if (flags&MSG_OOB)
1696                 goto out;
1697
1698         msg->msg_namelen = 0;
1699
1700         mutex_lock(&u->readlock);
1701
1702         skb = skb_recv_datagram(sk, flags, noblock, &err);
1703         if (!skb) {
1704                 unix_state_lock(sk);
1705                 /* Signal EOF on disconnected non-blocking SEQPACKET socket. */
1706                 if (sk->sk_type == SOCK_SEQPACKET && err == -EAGAIN &&
1707                     (sk->sk_shutdown & RCV_SHUTDOWN))
1708                         err = 0;
1709                 unix_state_unlock(sk);
1710                 goto out_unlock;
1711         }
1712
1713         wake_up_interruptible_sync(&u->peer_wait);
1714
1715         if (msg->msg_name)
1716                 unix_copy_addr(msg, skb->sk);
1717
1718         if (size > skb->len)
1719                 size = skb->len;
1720         else if (size < skb->len)
1721                 msg->msg_flags |= MSG_TRUNC;
1722
1723         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, size);
1724         if (err)
1725                 goto out_free;
1726
1727         if (sock_flag(sk, SOCK_RCVTSTAMP))
1728                 __sock_recv_timestamp(msg, sk, skb);
1729
1730         if (!siocb->scm) {
1731                 siocb->scm = &tmp_scm;
1732                 memset(&tmp_scm, 0, sizeof(tmp_scm));
1733         }
1734         scm_set_cred(siocb->scm, UNIXCB(skb).pid, UNIXCB(skb).cred);
1735         unix_set_secdata(siocb->scm, skb);
1736
1737         if (!(flags & MSG_PEEK)) {
1738                 if (UNIXCB(skb).fp)
1739                         unix_detach_fds(siocb->scm, skb);
1740         } else {
1741                 /* It is questionable: on PEEK we could:
1742                    - do not return fds - good, but too simple 8)
1743                    - return fds, and do not return them on read (old strategy,
1744                      apparently wrong)
1745                    - clone fds (I chose it for now, it is the most universal
1746                      solution)
1747
1748                    POSIX 1003.1g does not actually define this clearly
1749                    at all. POSIX 1003.1g doesn't define a lot of things
1750                    clearly however!
1751
1752                 */
1753                 if (UNIXCB(skb).fp)
1754                         siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1755         }
1756         err = size;
1757
1758         scm_recv(sock, msg, siocb->scm, flags);
1759
1760 out_free:
1761         skb_free_datagram(sk, skb);
1762 out_unlock:
1763         mutex_unlock(&u->readlock);
1764 out:
1765         return err;
1766 }
1767
1768 /*
1769  *      Sleep until data has arrive. But check for races..
1770  */
1771
1772 static long unix_stream_data_wait(struct sock *sk, long timeo)
1773 {
1774         DEFINE_WAIT(wait);
1775
1776         unix_state_lock(sk);
1777
1778         for (;;) {
1779                 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1780
1781                 if (!skb_queue_empty(&sk->sk_receive_queue) ||
1782                     sk->sk_err ||
1783                     (sk->sk_shutdown & RCV_SHUTDOWN) ||
1784                     signal_pending(current) ||
1785                     !timeo)
1786                         break;
1787
1788                 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1789                 unix_state_unlock(sk);
1790                 timeo = schedule_timeout(timeo);
1791                 unix_state_lock(sk);
1792                 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1793         }
1794
1795         finish_wait(sk_sleep(sk), &wait);
1796         unix_state_unlock(sk);
1797         return timeo;
1798 }
1799
1800
1801
1802 static int unix_stream_recvmsg(struct kiocb *iocb, struct socket *sock,
1803                                struct msghdr *msg, size_t size,
1804                                int flags)
1805 {
1806         struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1807         struct scm_cookie tmp_scm;
1808         struct sock *sk = sock->sk;
1809         struct unix_sock *u = unix_sk(sk);
1810         struct sockaddr_un *sunaddr = msg->msg_name;
1811         int copied = 0;
1812         int check_creds = 0;
1813         int target;
1814         int err = 0;
1815         long timeo;
1816
1817         err = -EINVAL;
1818         if (sk->sk_state != TCP_ESTABLISHED)
1819                 goto out;
1820
1821         err = -EOPNOTSUPP;
1822         if (flags&MSG_OOB)
1823                 goto out;
1824
1825         target = sock_rcvlowat(sk, flags&MSG_WAITALL, size);
1826         timeo = sock_rcvtimeo(sk, flags&MSG_DONTWAIT);
1827
1828         msg->msg_namelen = 0;
1829
1830         /* Lock the socket to prevent queue disordering
1831          * while sleeps in memcpy_tomsg
1832          */
1833
1834         if (!siocb->scm) {
1835                 siocb->scm = &tmp_scm;
1836                 memset(&tmp_scm, 0, sizeof(tmp_scm));
1837         }
1838
1839         mutex_lock(&u->readlock);
1840
1841         do {
1842                 int chunk;
1843                 struct sk_buff *skb;
1844
1845                 unix_state_lock(sk);
1846                 skb = skb_dequeue(&sk->sk_receive_queue);
1847                 if (skb == NULL) {
1848                         if (copied >= target)
1849                                 goto unlock;
1850
1851                         /*
1852                          *      POSIX 1003.1g mandates this order.
1853                          */
1854
1855                         err = sock_error(sk);
1856                         if (err)
1857                                 goto unlock;
1858                         if (sk->sk_shutdown & RCV_SHUTDOWN)
1859                                 goto unlock;
1860
1861                         unix_state_unlock(sk);
1862                         err = -EAGAIN;
1863                         if (!timeo)
1864                                 break;
1865                         mutex_unlock(&u->readlock);
1866
1867                         timeo = unix_stream_data_wait(sk, timeo);
1868
1869                         if (signal_pending(current)) {
1870                                 err = sock_intr_errno(timeo);
1871                                 goto out;
1872                         }
1873                         mutex_lock(&u->readlock);
1874                         continue;
1875  unlock:
1876                         unix_state_unlock(sk);
1877                         break;
1878                 }
1879                 unix_state_unlock(sk);
1880
1881                 if (check_creds) {
1882                         /* Never glue messages from different writers */
1883                         if ((UNIXCB(skb).pid  != siocb->scm->pid) ||
1884                             (UNIXCB(skb).cred != siocb->scm->cred)) {
1885                                 skb_queue_head(&sk->sk_receive_queue, skb);
1886                                 break;
1887                         }
1888                 } else {
1889                         /* Copy credentials */
1890                         scm_set_cred(siocb->scm, UNIXCB(skb).pid, UNIXCB(skb).cred);
1891                         check_creds = 1;
1892                 }
1893
1894                 /* Copy address just once */
1895                 if (sunaddr) {
1896                         unix_copy_addr(msg, skb->sk);
1897                         sunaddr = NULL;
1898                 }
1899
1900                 chunk = min_t(unsigned int, skb->len, size);
1901                 if (memcpy_toiovec(msg->msg_iov, skb->data, chunk)) {
1902                         skb_queue_head(&sk->sk_receive_queue, skb);
1903                         if (copied == 0)
1904                                 copied = -EFAULT;
1905                         break;
1906                 }
1907                 copied += chunk;
1908                 size -= chunk;
1909
1910                 /* Mark read part of skb as used */
1911                 if (!(flags & MSG_PEEK)) {
1912                         skb_pull(skb, chunk);
1913
1914                         if (UNIXCB(skb).fp)
1915                                 unix_detach_fds(siocb->scm, skb);
1916
1917                         /* put the skb back if we didn't use it up.. */
1918                         if (skb->len) {
1919                                 skb_queue_head(&sk->sk_receive_queue, skb);
1920                                 break;
1921                         }
1922
1923                         consume_skb(skb);
1924
1925                         if (siocb->scm->fp)
1926                                 break;
1927                 } else {
1928                         /* It is questionable, see note in unix_dgram_recvmsg.
1929                          */
1930                         if (UNIXCB(skb).fp)
1931                                 siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1932
1933                         /* put message back and return */
1934                         skb_queue_head(&sk->sk_receive_queue, skb);
1935                         break;
1936                 }
1937         } while (size);
1938
1939         mutex_unlock(&u->readlock);
1940         scm_recv(sock, msg, siocb->scm, flags);
1941 out:
1942         return copied ? : err;
1943 }
1944
1945 static int unix_shutdown(struct socket *sock, int mode)
1946 {
1947         struct sock *sk = sock->sk;
1948         struct sock *other;
1949
1950         mode = (mode+1)&(RCV_SHUTDOWN|SEND_SHUTDOWN);
1951
1952         if (mode) {
1953                 unix_state_lock(sk);
1954                 sk->sk_shutdown |= mode;
1955                 other = unix_peer(sk);
1956                 if (other)
1957                         sock_hold(other);
1958                 unix_state_unlock(sk);
1959                 sk->sk_state_change(sk);
1960
1961                 if (other &&
1962                         (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET)) {
1963
1964                         int peer_mode = 0;
1965
1966                         if (mode&RCV_SHUTDOWN)
1967                                 peer_mode |= SEND_SHUTDOWN;
1968                         if (mode&SEND_SHUTDOWN)
1969                                 peer_mode |= RCV_SHUTDOWN;
1970                         unix_state_lock(other);
1971                         other->sk_shutdown |= peer_mode;
1972                         unix_state_unlock(other);
1973                         other->sk_state_change(other);
1974                         if (peer_mode == SHUTDOWN_MASK)
1975                                 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_HUP);
1976                         else if (peer_mode & RCV_SHUTDOWN)
1977                                 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_IN);
1978                 }
1979                 if (other)
1980                         sock_put(other);
1981         }
1982         return 0;
1983 }
1984
1985 static int unix_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1986 {
1987         struct sock *sk = sock->sk;
1988         long amount = 0;
1989         int err;
1990
1991         switch (cmd) {
1992         case SIOCOUTQ:
1993                 amount = sk_wmem_alloc_get(sk);
1994                 err = put_user(amount, (int __user *)arg);
1995                 break;
1996         case SIOCINQ:
1997                 {
1998                         struct sk_buff *skb;
1999
2000                         if (sk->sk_state == TCP_LISTEN) {
2001                                 err = -EINVAL;
2002                                 break;
2003                         }
2004
2005                         spin_lock(&sk->sk_receive_queue.lock);
2006                         if (sk->sk_type == SOCK_STREAM ||
2007                             sk->sk_type == SOCK_SEQPACKET) {
2008                                 skb_queue_walk(&sk->sk_receive_queue, skb)
2009                                         amount += skb->len;
2010                         } else {
2011                                 skb = skb_peek(&sk->sk_receive_queue);
2012                                 if (skb)
2013                                         amount = skb->len;
2014                         }
2015                         spin_unlock(&sk->sk_receive_queue.lock);
2016                         err = put_user(amount, (int __user *)arg);
2017                         break;
2018                 }
2019
2020         default:
2021                 err = -ENOIOCTLCMD;
2022                 break;
2023         }
2024         return err;
2025 }
2026
2027 static unsigned int unix_poll(struct file *file, struct socket *sock, poll_table *wait)
2028 {
2029         struct sock *sk = sock->sk;
2030         unsigned int mask;
2031
2032         sock_poll_wait(file, sk_sleep(sk), wait);
2033         mask = 0;
2034
2035         /* exceptional events? */
2036         if (sk->sk_err)
2037                 mask |= POLLERR;
2038         if (sk->sk_shutdown == SHUTDOWN_MASK)
2039                 mask |= POLLHUP;
2040         if (sk->sk_shutdown & RCV_SHUTDOWN)
2041                 mask |= POLLRDHUP | POLLIN | POLLRDNORM;
2042
2043         /* readable? */
2044         if (!skb_queue_empty(&sk->sk_receive_queue))
2045                 mask |= POLLIN | POLLRDNORM;
2046
2047         /* Connection-based need to check for termination and startup */
2048         if ((sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) &&
2049             sk->sk_state == TCP_CLOSE)
2050                 mask |= POLLHUP;
2051
2052         /*
2053          * we set writable also when the other side has shut down the
2054          * connection. This prevents stuck sockets.
2055          */
2056         if (unix_writable(sk))
2057                 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2058
2059         return mask;
2060 }
2061
2062 static unsigned int unix_dgram_poll(struct file *file, struct socket *sock,
2063                                     poll_table *wait)
2064 {
2065         struct sock *sk = sock->sk, *other;
2066         unsigned int mask, writable;
2067
2068         sock_poll_wait(file, sk_sleep(sk), wait);
2069         mask = 0;
2070
2071         /* exceptional events? */
2072         if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue))
2073                 mask |= POLLERR;
2074         if (sk->sk_shutdown & RCV_SHUTDOWN)
2075                 mask |= POLLRDHUP;
2076         if (sk->sk_shutdown == SHUTDOWN_MASK)
2077                 mask |= POLLHUP;
2078
2079         /* readable? */
2080         if (!skb_queue_empty(&sk->sk_receive_queue) ||
2081             (sk->sk_shutdown & RCV_SHUTDOWN))
2082                 mask |= POLLIN | POLLRDNORM;
2083
2084         /* Connection-based need to check for termination and startup */
2085         if (sk->sk_type == SOCK_SEQPACKET) {
2086                 if (sk->sk_state == TCP_CLOSE)
2087                         mask |= POLLHUP;
2088                 /* connection hasn't started yet? */
2089                 if (sk->sk_state == TCP_SYN_SENT)
2090                         return mask;
2091         }
2092
2093         /* writable? */
2094         writable = unix_writable(sk);
2095         if (writable) {
2096                 other = unix_peer_get(sk);
2097                 if (other) {
2098                         if (unix_peer(other) != sk) {
2099                                 sock_poll_wait(file, &unix_sk(other)->peer_wait,
2100                                           wait);
2101                                 if (unix_recvq_full(other))
2102                                         writable = 0;
2103                         }
2104
2105                         sock_put(other);
2106                 }
2107         }
2108
2109         if (writable)
2110                 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2111         else
2112                 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
2113
2114         return mask;
2115 }
2116
2117 #ifdef CONFIG_PROC_FS
2118 static struct sock *first_unix_socket(int *i)
2119 {
2120         for (*i = 0; *i <= UNIX_HASH_SIZE; (*i)++) {
2121                 if (!hlist_empty(&unix_socket_table[*i]))
2122                         return __sk_head(&unix_socket_table[*i]);
2123         }
2124         return NULL;
2125 }
2126
2127 static struct sock *next_unix_socket(int *i, struct sock *s)
2128 {
2129         struct sock *next = sk_next(s);
2130         /* More in this chain? */
2131         if (next)
2132                 return next;
2133         /* Look for next non-empty chain. */
2134         for ((*i)++; *i <= UNIX_HASH_SIZE; (*i)++) {
2135                 if (!hlist_empty(&unix_socket_table[*i]))
2136                         return __sk_head(&unix_socket_table[*i]);
2137         }
2138         return NULL;
2139 }
2140
2141 struct unix_iter_state {
2142         struct seq_net_private p;
2143         int i;
2144 };
2145
2146 static struct sock *unix_seq_idx(struct seq_file *seq, loff_t pos)
2147 {
2148         struct unix_iter_state *iter = seq->private;
2149         loff_t off = 0;
2150         struct sock *s;
2151
2152         for (s = first_unix_socket(&iter->i); s; s = next_unix_socket(&iter->i, s)) {
2153                 if (sock_net(s) != seq_file_net(seq))
2154                         continue;
2155                 if (off == pos)
2156                         return s;
2157                 ++off;
2158         }
2159         return NULL;
2160 }
2161
2162 static void *unix_seq_start(struct seq_file *seq, loff_t *pos)
2163         __acquires(unix_table_lock)
2164 {
2165         spin_lock(&unix_table_lock);
2166         return *pos ? unix_seq_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2167 }
2168
2169 static void *unix_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2170 {
2171         struct unix_iter_state *iter = seq->private;
2172         struct sock *sk = v;
2173         ++*pos;
2174
2175         if (v == SEQ_START_TOKEN)
2176                 sk = first_unix_socket(&iter->i);
2177         else
2178                 sk = next_unix_socket(&iter->i, sk);
2179         while (sk && (sock_net(sk) != seq_file_net(seq)))
2180                 sk = next_unix_socket(&iter->i, sk);
2181         return sk;
2182 }
2183
2184 static void unix_seq_stop(struct seq_file *seq, void *v)
2185         __releases(unix_table_lock)
2186 {
2187         spin_unlock(&unix_table_lock);
2188 }
2189
2190 static int unix_seq_show(struct seq_file *seq, void *v)
2191 {
2192
2193         if (v == SEQ_START_TOKEN)
2194                 seq_puts(seq, "Num       RefCount Protocol Flags    Type St "
2195                          "Inode Path\n");
2196         else {
2197                 struct sock *s = v;
2198                 struct unix_sock *u = unix_sk(s);
2199                 unix_state_lock(s);
2200
2201                 seq_printf(seq, "%p: %08X %08X %08X %04X %02X %5lu",
2202                         s,
2203                         atomic_read(&s->sk_refcnt),
2204                         0,
2205                         s->sk_state == TCP_LISTEN ? __SO_ACCEPTCON : 0,
2206                         s->sk_type,
2207                         s->sk_socket ?
2208                         (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTED : SS_UNCONNECTED) :
2209                         (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTING : SS_DISCONNECTING),
2210                         sock_i_ino(s));
2211
2212                 if (u->addr) {
2213                         int i, len;
2214                         seq_putc(seq, ' ');
2215
2216                         i = 0;
2217                         len = u->addr->len - sizeof(short);
2218                         if (!UNIX_ABSTRACT(s))
2219                                 len--;
2220                         else {
2221                                 seq_putc(seq, '@');
2222                                 i++;
2223                         }
2224                         for ( ; i < len; i++)
2225                                 seq_putc(seq, u->addr->name->sun_path[i]);
2226                 }
2227                 unix_state_unlock(s);
2228                 seq_putc(seq, '\n');
2229         }
2230
2231         return 0;
2232 }
2233
2234 static const struct seq_operations unix_seq_ops = {
2235         .start  = unix_seq_start,
2236         .next   = unix_seq_next,
2237         .stop   = unix_seq_stop,
2238         .show   = unix_seq_show,
2239 };
2240
2241 static int unix_seq_open(struct inode *inode, struct file *file)
2242 {
2243         return seq_open_net(inode, file, &unix_seq_ops,
2244                             sizeof(struct unix_iter_state));
2245 }
2246
2247 static const struct file_operations unix_seq_fops = {
2248         .owner          = THIS_MODULE,
2249         .open           = unix_seq_open,
2250         .read           = seq_read,
2251         .llseek         = seq_lseek,
2252         .release        = seq_release_net,
2253 };
2254
2255 #endif
2256
2257 static const struct net_proto_family unix_family_ops = {
2258         .family = PF_UNIX,
2259         .create = unix_create,
2260         .owner  = THIS_MODULE,
2261 };
2262
2263
2264 static int __net_init unix_net_init(struct net *net)
2265 {
2266         int error = -ENOMEM;
2267
2268         net->unx.sysctl_max_dgram_qlen = 10;
2269         if (unix_sysctl_register(net))
2270                 goto out;
2271
2272 #ifdef CONFIG_PROC_FS
2273         if (!proc_net_fops_create(net, "unix", 0, &unix_seq_fops)) {
2274                 unix_sysctl_unregister(net);
2275                 goto out;
2276         }
2277 #endif
2278         error = 0;
2279 out:
2280         return error;
2281 }
2282
2283 static void __net_exit unix_net_exit(struct net *net)
2284 {
2285         unix_sysctl_unregister(net);
2286         proc_net_remove(net, "unix");
2287 }
2288
2289 static struct pernet_operations unix_net_ops = {
2290         .init = unix_net_init,
2291         .exit = unix_net_exit,
2292 };
2293
2294 static int __init af_unix_init(void)
2295 {
2296         int rc = -1;
2297         struct sk_buff *dummy_skb;
2298
2299         BUILD_BUG_ON(sizeof(struct unix_skb_parms) > sizeof(dummy_skb->cb));
2300
2301         rc = proto_register(&unix_proto, 1);
2302         if (rc != 0) {
2303                 printk(KERN_CRIT "%s: Cannot create unix_sock SLAB cache!\n",
2304                        __func__);
2305                 goto out;
2306         }
2307
2308         sock_register(&unix_family_ops);
2309         register_pernet_subsys(&unix_net_ops);
2310 out:
2311         return rc;
2312 }
2313
2314 static void __exit af_unix_exit(void)
2315 {
2316         sock_unregister(PF_UNIX);
2317         proto_unregister(&unix_proto);
2318         unregister_pernet_subsys(&unix_net_ops);
2319 }
2320
2321 /* Earlier than device_initcall() so that other drivers invoking
2322    request_module() don't end up in a loop when modprobe tries
2323    to use a UNIX socket. But later than subsys_initcall() because
2324    we depend on stuff initialised there */
2325 fs_initcall(af_unix_init);
2326 module_exit(af_unix_exit);
2327
2328 MODULE_LICENSE("GPL");
2329 MODULE_ALIAS_NETPROTO(PF_UNIX);