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] doesn't 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_poll(&wq->wait,
 320                                 POLLOUT | POLLWRNORM | POLLWRBAND);
 321                 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
 322         }
 323         rcu_read_unlock();
 324 }
 325
 326 /* When dgram socket disconnects (or changes its peer), we clear its receive
 327  * queue of packets arrived from previous peer. First, it allows to do
 328  * flow control based only on wmem_alloc; second, sk connected to peer
 329  * may receive messages only from that peer. */
 330 static void unix_dgram_disconnected(struct sock *sk, struct sock *other)
 331 {
 332         if (!skb_queue_empty(&sk->sk_receive_queue)) {
 333                 skb_queue_purge(&sk->sk_receive_queue);
 334                 wake_up_interruptible_all(&unix_sk(sk)->peer_wait);
 335
 336                 /* If one link of bidirectional dgram pipe is disconnected,
 337                  * we signal error. Messages are lost. Do not make this,
 338                  * when peer was not connected to us.
 339                  */
 340                 if (!sock_flag(other, SOCK_DEAD) && unix_peer(other) == sk) {
 341                         other->sk_err = ECONNRESET;
 342                         other->sk_error_report(other);
 343                 }
 344         }
 345 }
 346
 347 static void unix_sock_destructor(struct sock *sk)
 348 {
 349         struct unix_sock *u = unix_sk(sk);
 350
 351         skb_queue_purge(&sk->sk_receive_queue);
 352
 353         WARN_ON(atomic_read(&sk->sk_wmem_alloc));
 354         WARN_ON(!sk_unhashed(sk));
 355         WARN_ON(sk->sk_socket);
 356         if (!sock_flag(sk, SOCK_DEAD)) {
 357                 printk(KERN_INFO "Attempt to release alive unix socket: %p\n", sk);
 358                 return;
 359         }
 360
 361         if (u->addr)
 362                 unix_release_addr(u->addr);
 363
 364         atomic_long_dec(&unix_nr_socks);
 365         local_bh_disable();
 366         sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
 367         local_bh_enable();
 368 #ifdef UNIX_REFCNT_DEBUG
 369         printk(KERN_DEBUG "UNIX %p is destroyed, %ld are still alive.\n", sk,
 370                 atomic_long_read(&unix_nr_socks));
 371 #endif
 372 }
 373
 374 static int unix_release_sock(struct sock *sk, int embrion)
 375 {
 376         struct unix_sock *u = unix_sk(sk);
 377         struct dentry *dentry;
 378         struct vfsmount *mnt;
 379         struct sock *skpair;
 380         struct sk_buff *skb;
 381         int state;
 382
 383         unix_remove_socket(sk);
 384
 385         /* Clear state */
 386         unix_state_lock(sk);
 387         sock_orphan(sk);
 388         sk->sk_shutdown = SHUTDOWN_MASK;
 389         dentry       = u->dentry;
 390         u->dentry    = NULL;
 391         mnt          = u->mnt;
 392         u->mnt       = NULL;
 393         state = sk->sk_state;
 394         sk->sk_state = TCP_CLOSE;
 395         unix_state_unlock(sk);
 396
 397         wake_up_interruptible_all(&u->peer_wait);
 398
 399         skpair = unix_peer(sk);
 400
 401         if (skpair != NULL) {
 402                 if (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) {
 403                         unix_state_lock(skpair);
 404                         /* No more writes */
 405                         skpair->sk_shutdown = SHUTDOWN_MASK;
 406                         if (!skb_queue_empty(&sk->sk_receive_queue) || embrion)
 407                                 skpair->sk_err = ECONNRESET;
 408                         unix_state_unlock(skpair);
 409                         skpair->sk_state_change(skpair);
 410                         sk_wake_async(skpair, SOCK_WAKE_WAITD, POLL_HUP);
 411                 }
 412                 sock_put(skpair); /* It may now die */
 413                 unix_peer(sk) = NULL;
 414         }
 415
 416         /* Try to flush out this socket. Throw out buffers at least */
 417
 418         while ((skb = skb_dequeue(&sk->sk_receive_queue)) != NULL) {
 419                 if (state == TCP_LISTEN)
 420                         unix_release_sock(skb->sk, 1);
 421                 /* passed fds are erased in the kfree_skb hook        */
 422                 kfree_skb(skb);
 423         }
 424
 425         if (dentry) {
 426                 dput(dentry);
 427                 mntput(mnt);
 428         }
 429
 430         sock_put(sk);
 431
 432         /* ---- Socket is dead now and most probably destroyed ---- */
 433
 434         /*
 435          * Fixme: BSD difference: In BSD all sockets connected to use get
 436          *        ECONNRESET and we die on the spot. In Linux we behave
 437          *        like files and pipes do and wait for the last
 438          *        dereference.
 439          *
 440          * Can't we simply set sock->err?
 441          *
 442          *        What the above comment does talk about? --ANK(980817)
 443          */
 444
 445         if (unix_tot_inflight)
 446                 unix_gc();              /* Garbage collect fds */
 447
 448         return 0;
 449 }
 450
 451 static void init_peercred(struct sock *sk)
 452 {
 453         put_pid(sk->sk_peer_pid);
 454         if (sk->sk_peer_cred)
 455                 put_cred(sk->sk_peer_cred);
 456         sk->sk_peer_pid  = get_pid(task_tgid(current));
 457         sk->sk_peer_cred = get_current_cred();
 458 }
 459
 460 static void copy_peercred(struct sock *sk, struct sock *peersk)
 461 {
 462         put_pid(sk->sk_peer_pid);
 463         if (sk->sk_peer_cred)
 464                 put_cred(sk->sk_peer_cred);
 465         sk->sk_peer_pid  = get_pid(peersk->sk_peer_pid);
 466         sk->sk_peer_cred = get_cred(peersk->sk_peer_cred);
 467 }
 468
 469 static int unix_listen(struct socket *sock, int backlog)
 470 {
 471         int err;
 472         struct sock *sk = sock->sk;
 473         struct unix_sock *u = unix_sk(sk);
 474         struct pid *old_pid = NULL;
 475         const struct cred *old_cred = NULL;
 476
 477         err = -EOPNOTSUPP;
 478         if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
 479                 goto out;       /* Only stream/seqpacket sockets accept */
 480         err = -EINVAL;
 481         if (!u->addr)
 482                 goto out;       /* No listens on an unbound socket */
 483         unix_state_lock(sk);
 484         if (sk->sk_state != TCP_CLOSE && sk->sk_state != TCP_LISTEN)
 485                 goto out_unlock;
 486         if (backlog > sk->sk_max_ack_backlog)
 487                 wake_up_interruptible_all(&u->peer_wait);
 488         sk->sk_max_ack_backlog  = backlog;
 489         sk->sk_state            = TCP_LISTEN;
 490         /* set credentials so connect can copy them */
 491         init_peercred(sk);
 492         err = 0;
 493
 494 out_unlock:
 495         unix_state_unlock(sk);
 496         put_pid(old_pid);
 497         if (old_cred)
 498                 put_cred(old_cred);
 499 out:
 500         return err;
 501 }
 502
 503 static int unix_release(struct socket *);
 504 static int unix_bind(struct socket *, struct sockaddr *, int);
 505 static int unix_stream_connect(struct socket *, struct sockaddr *,
 506                                int addr_len, int flags);
 507 static int unix_socketpair(struct socket *, struct socket *);
 508 static int unix_accept(struct socket *, struct socket *, int);
 509 static int unix_getname(struct socket *, struct sockaddr *, int *, int);
 510 static unsigned int unix_poll(struct file *, struct socket *, poll_table *);
 511 static unsigned int unix_dgram_poll(struct file *, struct socket *,
 512                                     poll_table *);
 513 static int unix_ioctl(struct socket *, unsigned int, unsigned long);
 514 static int unix_shutdown(struct socket *, int);
 515 static int unix_stream_sendmsg(struct kiocb *, struct socket *,
 516                                struct msghdr *, size_t);
 517 static int unix_stream_recvmsg(struct kiocb *, struct socket *,
 518                                struct msghdr *, size_t, int);
 519 static int unix_dgram_sendmsg(struct kiocb *, struct socket *,
 520                               struct msghdr *, size_t);
 521 static int unix_dgram_recvmsg(struct kiocb *, struct socket *,
 522                               struct msghdr *, size_t, int);
 523 static int unix_dgram_connect(struct socket *, struct sockaddr *,
 524                               int, int);
 525 static int unix_seqpacket_sendmsg(struct kiocb *, struct socket *,
 526                                   struct msghdr *, size_t);
 527 static int unix_seqpacket_recvmsg(struct kiocb *, struct socket *,
 528                                   struct msghdr *, size_t, int);
 529
 530 static const struct proto_ops unix_stream_ops = {
 531         .family =       PF_UNIX,
 532         .owner =        THIS_MODULE,
 533         .release =      unix_release,
 534         .bind =         unix_bind,
 535         .connect =      unix_stream_connect,
 536         .socketpair =   unix_socketpair,
 537         .accept =       unix_accept,
 538         .getname =      unix_getname,
 539         .poll =         unix_poll,
 540         .ioctl =        unix_ioctl,
 541         .listen =       unix_listen,
 542         .shutdown =     unix_shutdown,
 543         .setsockopt =   sock_no_setsockopt,
 544         .getsockopt =   sock_no_getsockopt,
 545         .sendmsg =      unix_stream_sendmsg,
 546         .recvmsg =      unix_stream_recvmsg,
 547         .mmap =         sock_no_mmap,
 548         .sendpage =     sock_no_sendpage,
 549 };
 550
 551 static const struct proto_ops unix_dgram_ops = {
 552         .family =       PF_UNIX,
 553         .owner =        THIS_MODULE,
 554         .release =      unix_release,
 555         .bind =         unix_bind,
 556         .connect =      unix_dgram_connect,
 557         .socketpair =   unix_socketpair,
 558         .accept =       sock_no_accept,
 559         .getname =      unix_getname,
 560         .poll =         unix_dgram_poll,
 561         .ioctl =        unix_ioctl,
 562         .listen =       sock_no_listen,
 563         .shutdown =     unix_shutdown,
 564         .setsockopt =   sock_no_setsockopt,
 565         .getsockopt =   sock_no_getsockopt,
 566         .sendmsg =      unix_dgram_sendmsg,
 567         .recvmsg =      unix_dgram_recvmsg,
 568         .mmap =         sock_no_mmap,
 569         .sendpage =     sock_no_sendpage,
 570 };
 571
 572 static const struct proto_ops unix_seqpacket_ops = {
 573         .family =       PF_UNIX,
 574         .owner =        THIS_MODULE,
 575         .release =      unix_release,
 576         .bind =         unix_bind,
 577         .connect =      unix_stream_connect,
 578         .socketpair =   unix_socketpair,
 579         .accept =       unix_accept,
 580         .getname =      unix_getname,
 581         .poll =         unix_dgram_poll,
 582         .ioctl =        unix_ioctl,
 583         .listen =       unix_listen,
 584         .shutdown =     unix_shutdown,
 585         .setsockopt =   sock_no_setsockopt,
 586         .getsockopt =   sock_no_getsockopt,
 587         .sendmsg =      unix_seqpacket_sendmsg,
 588         .recvmsg =      unix_seqpacket_recvmsg,
 589         .mmap =         sock_no_mmap,
 590         .sendpage =     sock_no_sendpage,
 591 };
 592
 593 static struct proto unix_proto = {
 594         .name                   = "UNIX",
 595         .owner                  = THIS_MODULE,
 596         .obj_size               = sizeof(struct unix_sock),
 597 };
 598
 599 /*
 600  * AF_UNIX sockets do not interact with hardware, hence they
 601  * dont trigger interrupts - so it's safe for them to have
 602  * bh-unsafe locking for their sk_receive_queue.lock. Split off
 603  * this special lock-class by reinitializing the spinlock key:
 604  */
 605 static struct lock_class_key af_unix_sk_receive_queue_lock_key;
 606
 607 static struct sock *unix_create1(struct net *net, struct socket *sock)
 608 {
 609         struct sock *sk = NULL;
 610         struct unix_sock *u;
 611
 612         atomic_long_inc(&unix_nr_socks);
 613         if (atomic_long_read(&unix_nr_socks) > 2 * get_max_files())
 614                 goto out;
 615
 616         sk = sk_alloc(net, PF_UNIX, GFP_KERNEL, &unix_proto);
 617         if (!sk)
 618                 goto out;
 619
 620         sock_init_data(sock, sk);
 621         lockdep_set_class(&sk->sk_receive_queue.lock,
 622                                 &af_unix_sk_receive_queue_lock_key);
 623
 624         sk->sk_write_space      = unix_write_space;
 625         sk->sk_max_ack_backlog  = net->unx.sysctl_max_dgram_qlen;
 626         sk->sk_destruct         = unix_sock_destructor;
 627         u         = unix_sk(sk);
 628         u->dentry = NULL;
 629         u->mnt    = NULL;
 630         spin_lock_init(&u->lock);
 631         atomic_long_set(&u->inflight, 0);
 632         INIT_LIST_HEAD(&u->link);
 633         mutex_init(&u->readlock); /* single task reading lock */
 634         init_waitqueue_head(&u->peer_wait);
 635         unix_insert_socket(unix_sockets_unbound, sk);
 636 out:
 637         if (sk == NULL)
 638                 atomic_long_dec(&unix_nr_socks);
 639         else {
 640                 local_bh_disable();
 641                 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
 642                 local_bh_enable();
 643         }
 644         return sk;
 645 }
 646
 647 static int unix_create(struct net *net, struct socket *sock, int protocol,
 648                        int kern)
 649 {
 650         if (protocol && protocol != PF_UNIX)
 651                 return -EPROTONOSUPPORT;
 652
 653         sock->state = SS_UNCONNECTED;
 654
 655         switch (sock->type) {
 656         case SOCK_STREAM:
 657                 sock->ops = &unix_stream_ops;
 658                 break;
 659                 /*
 660                  *      Believe it or not BSD has AF_UNIX, SOCK_RAW though
 661                  *      nothing uses it.
 662                  */
 663         case SOCK_RAW:
 664                 sock->type = SOCK_DGRAM;
 665         case SOCK_DGRAM:
 666                 sock->ops = &unix_dgram_ops;
 667                 break;
 668         case SOCK_SEQPACKET:
 669                 sock->ops = &unix_seqpacket_ops;
 670                 break;
 671         default:
 672                 return -ESOCKTNOSUPPORT;
 673         }
 674
 675         return unix_create1(net, sock) ? 0 : -ENOMEM;
 676 }
 677
 678 static int unix_release(struct socket *sock)
 679 {
 680         struct sock *sk = sock->sk;
 681
 682         if (!sk)
 683                 return 0;
 684
 685         sock->sk = NULL;
 686
 687         return unix_release_sock(sk, 0);
 688 }
 689
 690 static int unix_autobind(struct socket *sock)
 691 {
 692         struct sock *sk = sock->sk;
 693         struct net *net = sock_net(sk);
 694         struct unix_sock *u = unix_sk(sk);
 695         static u32 ordernum = 1;
 696         struct unix_address *addr;
 697         int err;
 698         unsigned int retries = 0;
 699
 700         mutex_lock(&u->readlock);
 701
 702         err = 0;
 703         if (u->addr)
 704                 goto out;
 705
 706         err = -ENOMEM;
 707         addr = kzalloc(sizeof(*addr) + sizeof(short) + 16, GFP_KERNEL);
 708         if (!addr)
 709                 goto out;
 710
 711         addr->name->sun_family = AF_UNIX;
 712         atomic_set(&addr->refcnt, 1);
 713
 714 retry:
 715         addr->len = sprintf(addr->name->sun_path+1, "%05x", ordernum) + 1 + sizeof(short);
 716         addr->hash = unix_hash_fold(csum_partial(addr->name, addr->len, 0));
 717
 718         spin_lock(&unix_table_lock);
 719         ordernum = (ordernum+1)&0xFFFFF;
 720
 721         if (__unix_find_socket_byname(net, addr->name, addr->len, sock->type,
 722                                       addr->hash)) {
 723                 spin_unlock(&unix_table_lock);
 724                 /*
 725                  * __unix_find_socket_byname() may take long time if many names
 726                  * are already in use.
 727                  */
 728                 cond_resched();
 729                 /* Give up if all names seems to be in use. */
 730                 if (retries++ == 0xFFFFF) {
 731                         err = -ENOSPC;
 732                         kfree(addr);
 733                         goto out;
 734                 }
 735                 goto retry;
 736         }
 737         addr->hash ^= sk->sk_type;
 738
 739         __unix_remove_socket(sk);
 740         u->addr = addr;
 741         __unix_insert_socket(&unix_socket_table[addr->hash], sk);
 742         spin_unlock(&unix_table_lock);
 743         err = 0;
 744
 745 out:    mutex_unlock(&u->readlock);
 746         return err;
 747 }
 748
 749 static struct sock *unix_find_other(struct net *net,
 750                                     struct sockaddr_un *sunname, int len,
 751                                     int type, unsigned hash, int *error)
 752 {
 753         struct sock *u;
 754         struct path path;
 755         int err = 0;
 756
 757         if (sunname->sun_path[0]) {
 758                 struct inode *inode;
 759                 err = kern_path(sunname->sun_path, LOOKUP_FOLLOW, &path);
 760                 if (err)
 761                         goto fail;
 762                 inode = path.dentry->d_inode;
 763                 err = inode_permission(inode, MAY_WRITE);
 764                 if (err)
 765                         goto put_fail;
 766
 767                 err = -ECONNREFUSED;
 768                 if (!S_ISSOCK(inode->i_mode))
 769                         goto put_fail;
 770                 u = unix_find_socket_byinode(inode);
 771                 if (!u)
 772                         goto put_fail;
 773
 774                 if (u->sk_type == type)
 775                         touch_atime(path.mnt, path.dentry);
 776
 777                 path_put(&path);
 778
 779                 err = -EPROTOTYPE;
 780                 if (u->sk_type != type) {
 781                         sock_put(u);
 782                         goto fail;
 783                 }
 784         } else {
 785                 err = -ECONNREFUSED;
 786                 u = unix_find_socket_byname(net, sunname, len, type, hash);
 787                 if (u) {
 788                         struct dentry *dentry;
 789                         dentry = unix_sk(u)->dentry;
 790                         if (dentry)
 791                                 touch_atime(unix_sk(u)->mnt, dentry);
 792                 } else
 793                         goto fail;
 794         }
 795         return u;
 796
 797 put_fail:
 798         path_put(&path);
 799 fail:
 800         *error = err;
 801         return NULL;
 802 }
 803
 804
 805 static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
 806 {
 807         struct sock *sk = sock->sk;
 808         struct net *net = sock_net(sk);
 809         struct unix_sock *u = unix_sk(sk);
 810         struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
 811         char *sun_path = sunaddr->sun_path;
 812         struct dentry *dentry = NULL;
 813         struct path path;
 814         int err;
 815         unsigned hash;
 816         struct unix_address *addr;
 817         struct hlist_head *list;
 818
 819         err = -EINVAL;
 820         if (sunaddr->sun_family != AF_UNIX)
 821                 goto out;
 822
 823         if (addr_len == sizeof(short)) {
 824                 err = unix_autobind(sock);
 825                 goto out;
 826         }
 827
 828         err = unix_mkname(sunaddr, addr_len, &hash);
 829         if (err < 0)
 830                 goto out;
 831         addr_len = err;
 832
 833         mutex_lock(&u->readlock);
 834
 835         err = -EINVAL;
 836         if (u->addr)
 837                 goto out_up;
 838
 839         err = -ENOMEM;
 840         addr = kmalloc(sizeof(*addr)+addr_len, GFP_KERNEL);
 841         if (!addr)
 842                 goto out_up;
 843
 844         memcpy(addr->name, sunaddr, addr_len);
 845         addr->len = addr_len;
 846         addr->hash = hash ^ sk->sk_type;
 847         atomic_set(&addr->refcnt, 1);
 848
 849         if (sun_path[0]) {
 850                 unsigned int mode;
 851                 err = 0;
 852                 /*
 853                  * Get the parent directory, calculate the hash for last
 854                  * component.
 855                  */
 856                 dentry = kern_path_create(AT_FDCWD, sun_path, &path, 0);
 857                 err = PTR_ERR(dentry);
 858                 if (IS_ERR(dentry))
 859                         goto out_mknod_parent;
 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(path.mnt);
 867                 if (err)
 868                         goto out_mknod_dput;
 869                 err = security_path_mknod(&path, dentry, mode, 0);
 870                 if (err)
 871                         goto out_mknod_drop_write;
 872                 err = vfs_mknod(path.dentry->d_inode, dentry, mode, 0);
 873 out_mknod_drop_write:
 874                 mnt_drop_write(path.mnt);
 875                 if (err)
 876                         goto out_mknod_dput;
 877                 mutex_unlock(&path.dentry->d_inode->i_mutex);
 878                 dput(path.dentry);
 879                 path.dentry = dentry;
 880
 881                 addr->hash = UNIX_HASH_SIZE;
 882         }
 883
 884         spin_lock(&unix_table_lock);
 885
 886         if (!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 = path.dentry;
 898                 u->mnt    = 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         mutex_unlock(&path.dentry->d_inode->i_mutex);
 916         path_put(&path);
 917 out_mknod_parent:
 918         if (err == -EEXIST)
 919                 err = -EADDRINUSE;
 920         unix_release_addr(addr);
 921         goto out_up;
 922 }
 923
 924 static void unix_state_double_lock(struct sock *sk1, struct sock *sk2)
 925 {
 926         if (unlikely(sk1 == sk2) || !sk2) {
 927                 unix_state_lock(sk1);
 928                 return;
 929         }
 930         if (sk1 < sk2) {
 931                 unix_state_lock(sk1);
 932                 unix_state_lock_nested(sk2);
 933         } else {
 934                 unix_state_lock(sk2);
 935                 unix_state_lock_nested(sk1);
 936         }
 937 }
 938
 939 static void unix_state_double_unlock(struct sock *sk1, struct sock *sk2)
 940 {
 941         if (unlikely(sk1 == sk2) || !sk2) {
 942                 unix_state_unlock(sk1);
 943                 return;
 944         }
 945         unix_state_unlock(sk1);
 946         unix_state_unlock(sk2);
 947 }
 948
 949 static int unix_dgram_connect(struct socket *sock, struct sockaddr *addr,
 950                               int alen, int flags)
 951 {
 952         struct sock *sk = sock->sk;
 953         struct net *net = sock_net(sk);
 954         struct sockaddr_un *sunaddr = (struct sockaddr_un *)addr;
 955         struct sock *other;
 956         unsigned hash;
 957         int err;
 958
 959         if (addr->sa_family != AF_UNSPEC) {
 960                 err = unix_mkname(sunaddr, alen, &hash);
 961                 if (err < 0)
 962                         goto out;
 963                 alen = err;
 964
 965                 if (test_bit(SOCK_PASSCRED, &sock->flags) &&
 966                     !unix_sk(sk)->addr && (err = unix_autobind(sock)) != 0)
 967                         goto out;
 968
 969 restart:
 970                 other = unix_find_other(net, sunaddr, alen, sock->type, hash, &err);
 971                 if (!other)
 972                         goto out;
 973
 974                 unix_state_double_lock(sk, other);
 975
 976                 /* Apparently VFS overslept socket death. Retry. */
 977                 if (sock_flag(other, SOCK_DEAD)) {
 978                         unix_state_double_unlock(sk, other);
 979                         sock_put(other);
 980                         goto restart;
 981                 }
 982
 983                 err = -EPERM;
 984                 if (!unix_may_send(sk, other))
 985                         goto out_unlock;
 986
 987                 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
 988                 if (err)
 989                         goto out_unlock;
 990
 991         } else {
 992                 /*
 993                  *      1003.1g breaking connected state with AF_UNSPEC
 994                  */
 995                 other = NULL;
 996                 unix_state_double_lock(sk, other);
 997         }
 998
 999         /*
1000          * If it was connected, reconnect.
1001          */
1002         if (unix_peer(sk)) {
1003                 struct sock *old_peer = unix_peer(sk);
1004                 unix_peer(sk) = other;
1005                 unix_state_double_unlock(sk, other);
1006
1007                 if (other != old_peer)
1008                         unix_dgram_disconnected(sk, old_peer);
1009                 sock_put(old_peer);
1010         } else {
1011                 unix_peer(sk) = other;
1012                 unix_state_double_unlock(sk, other);
1013         }
1014         return 0;
1015
1016 out_unlock:
1017         unix_state_double_unlock(sk, other);
1018         sock_put(other);
1019 out:
1020         return err;
1021 }
1022
1023 static long unix_wait_for_peer(struct sock *other, long timeo)
1024 {
1025         struct unix_sock *u = unix_sk(other);
1026         int sched;
1027         DEFINE_WAIT(wait);
1028
1029         prepare_to_wait_exclusive(&u->peer_wait, &wait, TASK_INTERRUPTIBLE);
1030
1031         sched = !sock_flag(other, SOCK_DEAD) &&
1032                 !(other->sk_shutdown & RCV_SHUTDOWN) &&
1033                 unix_recvq_full(other);
1034
1035         unix_state_unlock(other);
1036
1037         if (sched)
1038                 timeo = schedule_timeout(timeo);
1039
1040         finish_wait(&u->peer_wait, &wait);
1041         return timeo;
1042 }
1043
1044 static int unix_stream_connect(struct socket *sock, struct sockaddr *uaddr,
1045                                int addr_len, int flags)
1046 {
1047         struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
1048         struct sock *sk = sock->sk;
1049         struct net *net = sock_net(sk);
1050         struct unix_sock *u = unix_sk(sk), *newu, *otheru;
1051         struct sock *newsk = NULL;
1052         struct sock *other = NULL;
1053         struct sk_buff *skb = NULL;
1054         unsigned hash;
1055         int st;
1056         int err;
1057         long timeo;
1058
1059         err = unix_mkname(sunaddr, addr_len, &hash);
1060         if (err < 0)
1061                 goto out;
1062         addr_len = err;
1063
1064         if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr &&
1065             (err = unix_autobind(sock)) != 0)
1066                 goto out;
1067
1068         timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
1069
1070         /* First of all allocate resources.
1071            If we will make it after state is locked,
1072            we will have to recheck all again in any case.
1073          */
1074
1075         err = -ENOMEM;
1076
1077         /* create new sock for complete connection */
1078         newsk = unix_create1(sock_net(sk), NULL);
1079         if (newsk == NULL)
1080                 goto out;
1081
1082         /* Allocate skb for sending to listening sock */
1083         skb = sock_wmalloc(newsk, 1, 0, GFP_KERNEL);
1084         if (skb == NULL)
1085                 goto out;
1086
1087 restart:
1088         /*  Find listening sock. */
1089         other = unix_find_other(net, sunaddr, addr_len, sk->sk_type, hash, &err);
1090         if (!other)
1091                 goto out;
1092
1093         /* Latch state of peer */
1094         unix_state_lock(other);
1095
1096         /* Apparently VFS overslept socket death. Retry. */
1097         if (sock_flag(other, SOCK_DEAD)) {
1098                 unix_state_unlock(other);
1099                 sock_put(other);
1100                 goto restart;
1101         }
1102
1103         err = -ECONNREFUSED;
1104         if (other->sk_state != TCP_LISTEN)
1105                 goto out_unlock;
1106         if (other->sk_shutdown & RCV_SHUTDOWN)
1107                 goto out_unlock;
1108
1109         if (unix_recvq_full(other)) {
1110                 err = -EAGAIN;
1111                 if (!timeo)
1112                         goto out_unlock;
1113
1114                 timeo = unix_wait_for_peer(other, timeo);
1115
1116                 err = sock_intr_errno(timeo);
1117                 if (signal_pending(current))
1118                         goto out;
1119                 sock_put(other);
1120                 goto restart;
1121         }
1122
1123         /* Latch our state.
1124
1125            It is tricky place. We need to grab our state lock and cannot
1126            drop lock on peer. It is dangerous because deadlock is
1127            possible. Connect to self case and simultaneous
1128            attempt to connect are eliminated by checking socket
1129            state. other is TCP_LISTEN, if sk is TCP_LISTEN we
1130            check this before attempt to grab lock.
1131
1132            Well, and we have to recheck the state after socket locked.
1133          */
1134         st = sk->sk_state;
1135
1136         switch (st) {
1137         case TCP_CLOSE:
1138                 /* This is ok... continue with connect */
1139                 break;
1140         case TCP_ESTABLISHED:
1141                 /* Socket is already connected */
1142                 err = -EISCONN;
1143                 goto out_unlock;
1144         default:
1145                 err = -EINVAL;
1146                 goto out_unlock;
1147         }
1148
1149         unix_state_lock_nested(sk);
1150
1151         if (sk->sk_state != st) {
1152                 unix_state_unlock(sk);
1153                 unix_state_unlock(other);
1154                 sock_put(other);
1155                 goto restart;
1156         }
1157
1158         err = security_unix_stream_connect(sk, other, newsk);
1159         if (err) {
1160                 unix_state_unlock(sk);
1161                 goto out_unlock;
1162         }
1163
1164         /* The way is open! Fastly set all the necessary fields... */
1165
1166         sock_hold(sk);
1167         unix_peer(newsk)        = sk;
1168         newsk->sk_state         = TCP_ESTABLISHED;
1169         newsk->sk_type          = sk->sk_type;
1170         init_peercred(newsk);
1171         newu = unix_sk(newsk);
1172         RCU_INIT_POINTER(newsk->sk_wq, &newu->peer_wq);
1173         otheru = unix_sk(other);
1174
1175         /* copy address information from listening to new sock*/
1176         if (otheru->addr) {
1177                 atomic_inc(&otheru->addr->refcnt);
1178                 newu->addr = otheru->addr;
1179         }
1180         if (otheru->dentry) {
1181                 newu->dentry    = dget(otheru->dentry);
1182                 newu->mnt       = mntget(otheru->mnt);
1183         }
1184
1185         /* Set credentials */
1186         copy_peercred(sk, other);
1187
1188         sock->state     = SS_CONNECTED;
1189         sk->sk_state    = TCP_ESTABLISHED;
1190         sock_hold(newsk);
1191
1192         smp_mb__after_atomic_inc();     /* sock_hold() does an atomic_inc() */
1193         unix_peer(sk)   = newsk;
1194
1195         unix_state_unlock(sk);
1196
1197         /* take ten and and send info to listening sock */
1198         spin_lock(&other->sk_receive_queue.lock);
1199         __skb_queue_tail(&other->sk_receive_queue, skb);
1200         spin_unlock(&other->sk_receive_queue.lock);
1201         unix_state_unlock(other);
1202         other->sk_data_ready(other, 0);
1203         sock_put(other);
1204         return 0;
1205
1206 out_unlock:
1207         if (other)
1208                 unix_state_unlock(other);
1209
1210 out:
1211         kfree_skb(skb);
1212         if (newsk)
1213                 unix_release_sock(newsk, 0);
1214         if (other)
1215                 sock_put(other);
1216         return err;
1217 }
1218
1219 static int unix_socketpair(struct socket *socka, struct socket *sockb)
1220 {
1221         struct sock *ska = socka->sk, *skb = sockb->sk;
1222
1223         /* Join our sockets back to back */
1224         sock_hold(ska);
1225         sock_hold(skb);
1226         unix_peer(ska) = skb;
1227         unix_peer(skb) = ska;
1228         init_peercred(ska);
1229         init_peercred(skb);
1230
1231         if (ska->sk_type != SOCK_DGRAM) {
1232                 ska->sk_state = TCP_ESTABLISHED;
1233                 skb->sk_state = TCP_ESTABLISHED;
1234                 socka->state  = SS_CONNECTED;
1235                 sockb->state  = SS_CONNECTED;
1236         }
1237         return 0;
1238 }
1239
1240 static int unix_accept(struct socket *sock, struct socket *newsock, int flags)
1241 {
1242         struct sock *sk = sock->sk;
1243         struct sock *tsk;
1244         struct sk_buff *skb;
1245         int err;
1246
1247         err = -EOPNOTSUPP;
1248         if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
1249                 goto out;
1250
1251         err = -EINVAL;
1252         if (sk->sk_state != TCP_LISTEN)
1253                 goto out;
1254
1255         /* If socket state is TCP_LISTEN it cannot change (for now...),
1256          * so that no locks are necessary.
1257          */
1258
1259         skb = skb_recv_datagram(sk, 0, flags&O_NONBLOCK, &err);
1260         if (!skb) {
1261                 /* This means receive shutdown. */
1262                 if (err == 0)
1263                         err = -EINVAL;
1264                 goto out;
1265         }
1266
1267         tsk = skb->sk;
1268         skb_free_datagram(sk, skb);
1269         wake_up_interruptible(&unix_sk(sk)->peer_wait);
1270
1271         /* attach accepted sock to socket */
1272         unix_state_lock(tsk);
1273         newsock->state = SS_CONNECTED;
1274         sock_graft(tsk, newsock);
1275         unix_state_unlock(tsk);
1276         return 0;
1277
1278 out:
1279         return err;
1280 }
1281
1282
1283 static int unix_getname(struct socket *sock, struct sockaddr *uaddr, int *uaddr_len, int peer)
1284 {
1285         struct sock *sk = sock->sk;
1286         struct unix_sock *u;
1287         DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr, uaddr);
1288         int err = 0;
1289
1290         if (peer) {
1291                 sk = unix_peer_get(sk);
1292
1293                 err = -ENOTCONN;
1294                 if (!sk)
1295                         goto out;
1296                 err = 0;
1297         } else {
1298                 sock_hold(sk);
1299         }
1300
1301         u = unix_sk(sk);
1302         unix_state_lock(sk);
1303         if (!u->addr) {
1304                 sunaddr->sun_family = AF_UNIX;
1305                 sunaddr->sun_path[0] = 0;
1306                 *uaddr_len = sizeof(short);
1307         } else {
1308                 struct unix_address *addr = u->addr;
1309
1310                 *uaddr_len = addr->len;
1311                 memcpy(sunaddr, addr->name, *uaddr_len);
1312         }
1313         unix_state_unlock(sk);
1314         sock_put(sk);
1315 out:
1316         return err;
1317 }
1318
1319 static void unix_detach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1320 {
1321         int i;
1322
1323         scm->fp = UNIXCB(skb).fp;
1324         UNIXCB(skb).fp = NULL;
1325
1326         for (i = scm->fp->count-1; i >= 0; i--)
1327                 unix_notinflight(scm->fp->fp[i]);
1328 }
1329
1330 static void unix_destruct_scm(struct sk_buff *skb)
1331 {
1332         struct scm_cookie scm;
1333         memset(&scm, 0, sizeof(scm));
1334         scm.pid  = UNIXCB(skb).pid;
1335         scm.cred = UNIXCB(skb).cred;
1336         if (UNIXCB(skb).fp)
1337                 unix_detach_fds(&scm, skb);
1338
1339         /* Alas, it calls VFS */
1340         /* So fscking what? fput() had been SMP-safe since the last Summer */
1341         scm_destroy(&scm);
1342         sock_wfree(skb);
1343 }
1344
1345 #define MAX_RECURSION_LEVEL 4
1346
1347 static int unix_attach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1348 {
1349         int i;
1350         unsigned char max_level = 0;
1351         int unix_sock_count = 0;
1352
1353         for (i = scm->fp->count - 1; i >= 0; i--) {
1354                 struct sock *sk = unix_get_socket(scm->fp->fp[i]);
1355
1356                 if (sk) {
1357                         unix_sock_count++;
1358                         max_level = max(max_level,
1359                                         unix_sk(sk)->recursion_level);
1360                 }
1361         }
1362         if (unlikely(max_level > MAX_RECURSION_LEVEL))
1363                 return -ETOOMANYREFS;
1364
1365         /*
1366          * Need to duplicate file references for the sake of garbage
1367          * collection.  Otherwise a socket in the fps might become a
1368          * candidate for GC while the skb is not yet queued.
1369          */
1370         UNIXCB(skb).fp = scm_fp_dup(scm->fp);
1371         if (!UNIXCB(skb).fp)
1372                 return -ENOMEM;
1373
1374         if (unix_sock_count) {
1375                 for (i = scm->fp->count - 1; i >= 0; i--)
1376                         unix_inflight(scm->fp->fp[i]);
1377         }
1378         return max_level;
1379 }
1380
1381 static int unix_scm_to_skb(struct scm_cookie *scm, struct sk_buff *skb, bool send_fds)
1382 {
1383         int err = 0;
1384
1385         UNIXCB(skb).pid  = get_pid(scm->pid);
1386         if (scm->cred)
1387                 UNIXCB(skb).cred = get_cred(scm->cred);
1388         UNIXCB(skb).fp = NULL;
1389         if (scm->fp && send_fds)
1390                 err = unix_attach_fds(scm, skb);
1391
1392         skb->destructor = unix_destruct_scm;
1393         return err;
1394 }
1395
1396 /*
1397  * Some apps rely on write() giving SCM_CREDENTIALS
1398  * We include credentials if source or destination socket
1399  * asserted SOCK_PASSCRED.
1400  */
1401 static void maybe_add_creds(struct sk_buff *skb, const struct socket *sock,
1402                             const struct sock *other)
1403 {
1404         if (UNIXCB(skb).cred)
1405                 return;
1406         if (test_bit(SOCK_PASSCRED, &sock->flags) ||
1407             !other->sk_socket ||
1408             test_bit(SOCK_PASSCRED, &other->sk_socket->flags)) {
1409                 UNIXCB(skb).pid  = get_pid(task_tgid(current));
1410                 UNIXCB(skb).cred = get_current_cred();
1411         }
1412 }
1413
1414 /*
1415  *      Send AF_UNIX data.
1416  */
1417
1418 static int unix_dgram_sendmsg(struct kiocb *kiocb, struct socket *sock,
1419                               struct msghdr *msg, size_t len)
1420 {
1421         struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1422         struct sock *sk = sock->sk;
1423         struct net *net = sock_net(sk);
1424         struct unix_sock *u = unix_sk(sk);
1425         struct sockaddr_un *sunaddr = msg->msg_name;
1426         struct sock *other = NULL;
1427         int namelen = 0; /* fake GCC */
1428         int err;
1429         unsigned hash;
1430         struct sk_buff *skb;
1431         long timeo;
1432         struct scm_cookie tmp_scm;
1433         int max_level;
1434
1435         if (NULL == siocb->scm)
1436                 siocb->scm = &tmp_scm;
1437         wait_for_unix_gc();
1438         err = scm_send(sock, msg, siocb->scm);
1439         if (err < 0)
1440                 return err;
1441
1442         err = -EOPNOTSUPP;
1443         if (msg->msg_flags&MSG_OOB)
1444                 goto out;
1445
1446         if (msg->msg_namelen) {
1447                 err = unix_mkname(sunaddr, msg->msg_namelen, &hash);
1448                 if (err < 0)
1449                         goto out;
1450                 namelen = err;
1451         } else {
1452                 sunaddr = NULL;
1453                 err = -ENOTCONN;
1454                 other = unix_peer_get(sk);
1455                 if (!other)
1456                         goto out;
1457         }
1458
1459         if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr
1460             && (err = unix_autobind(sock)) != 0)
1461                 goto out;
1462
1463         err = -EMSGSIZE;
1464         if (len > sk->sk_sndbuf - 32)
1465                 goto out;
1466
1467         skb = sock_alloc_send_skb(sk, len, msg->msg_flags&MSG_DONTWAIT, &err);
1468         if (skb == NULL)
1469                 goto out;
1470
1471         err = unix_scm_to_skb(siocb->scm, skb, true);
1472         if (err < 0)
1473                 goto out_free;
1474         max_level = err + 1;
1475         unix_get_secdata(siocb->scm, skb);
1476
1477         skb_reset_transport_header(skb);
1478         err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
1479         if (err)
1480                 goto out_free;
1481
1482         timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1483
1484 restart:
1485         if (!other) {
1486                 err = -ECONNRESET;
1487                 if (sunaddr == NULL)
1488                         goto out_free;
1489
1490                 other = unix_find_other(net, sunaddr, namelen, sk->sk_type,
1491                                         hash, &err);
1492                 if (other == NULL)
1493                         goto out_free;
1494         }
1495
1496         if (sk_filter(other, skb) < 0) {
1497                 /* Toss the packet but do not return any error to the sender */
1498                 err = len;
1499                 goto out_free;
1500         }
1501
1502         unix_state_lock(other);
1503         err = -EPERM;
1504         if (!unix_may_send(sk, other))
1505                 goto out_unlock;
1506
1507         if (sock_flag(other, SOCK_DEAD)) {
1508                 /*
1509                  *      Check with 1003.1g - what should
1510                  *      datagram error
1511                  */
1512                 unix_state_unlock(other);
1513                 sock_put(other);
1514
1515                 err = 0;
1516                 unix_state_lock(sk);
1517                 if (unix_peer(sk) == other) {
1518                         unix_peer(sk) = NULL;
1519                         unix_state_unlock(sk);
1520
1521                         unix_dgram_disconnected(sk, other);
1522                         sock_put(other);
1523                         err = -ECONNREFUSED;
1524                 } else {
1525                         unix_state_unlock(sk);
1526                 }
1527
1528                 other = NULL;
1529                 if (err)
1530                         goto out_free;
1531                 goto restart;
1532         }
1533
1534         err = -EPIPE;
1535         if (other->sk_shutdown & RCV_SHUTDOWN)
1536                 goto out_unlock;
1537
1538         if (sk->sk_type != SOCK_SEQPACKET) {
1539                 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1540                 if (err)
1541                         goto out_unlock;
1542         }
1543
1544         if (unix_peer(other) != sk && unix_recvq_full(other)) {
1545                 if (!timeo) {
1546                         err = -EAGAIN;
1547                         goto out_unlock;
1548                 }
1549
1550                 timeo = unix_wait_for_peer(other, timeo);
1551
1552                 err = sock_intr_errno(timeo);
1553                 if (signal_pending(current))
1554                         goto out_free;
1555
1556                 goto restart;
1557         }
1558
1559         if (sock_flag(other, SOCK_RCVTSTAMP))
1560                 __net_timestamp(skb);
1561         maybe_add_creds(skb, sock, other);
1562         skb_queue_tail(&other->sk_receive_queue, skb);
1563         if (max_level > unix_sk(other)->recursion_level)
1564                 unix_sk(other)->recursion_level = max_level;
1565         unix_state_unlock(other);
1566         other->sk_data_ready(other, len);
1567         sock_put(other);
1568         scm_destroy(siocb->scm);
1569         return len;
1570
1571 out_unlock:
1572         unix_state_unlock(other);
1573 out_free:
1574         kfree_skb(skb);
1575 out:
1576         if (other)
1577                 sock_put(other);
1578         scm_destroy(siocb->scm);
1579         return err;
1580 }
1581
1582
1583 static int unix_stream_sendmsg(struct kiocb *kiocb, struct socket *sock,
1584                                struct msghdr *msg, size_t len)
1585 {
1586         struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1587         struct sock *sk = sock->sk;
1588         struct sock *other = NULL;
1589         int err, size;
1590         struct sk_buff *skb;
1591         int sent = 0;
1592         struct scm_cookie tmp_scm;
1593         bool fds_sent = false;
1594         int max_level;
1595
1596         if (NULL == siocb->scm)
1597                 siocb->scm = &tmp_scm;
1598         wait_for_unix_gc();
1599         err = scm_send(sock, msg, siocb->scm);
1600         if (err < 0)
1601                 return err;
1602
1603         err = -EOPNOTSUPP;
1604         if (msg->msg_flags&MSG_OOB)
1605                 goto out_err;
1606
1607         if (msg->msg_namelen) {
1608                 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
1609                 goto out_err;
1610         } else {
1611                 err = -ENOTCONN;
1612                 other = unix_peer(sk);
1613                 if (!other)
1614                         goto out_err;
1615         }
1616
1617         if (sk->sk_shutdown & SEND_SHUTDOWN)
1618                 goto pipe_err;
1619
1620         while (sent < len) {
1621                 /*
1622                  *      Optimisation for the fact that under 0.01% of X
1623                  *      messages typically need breaking up.
1624                  */
1625
1626                 size = len-sent;
1627
1628                 /* Keep two messages in the pipe so it schedules better */
1629                 if (size > ((sk->sk_sndbuf >> 1) - 64))
1630                         size = (sk->sk_sndbuf >> 1) - 64;
1631
1632                 if (size > SKB_MAX_ALLOC)
1633                         size = SKB_MAX_ALLOC;
1634
1635                 /*
1636                  *      Grab a buffer
1637                  */
1638
1639                 skb = sock_alloc_send_skb(sk, size, msg->msg_flags&MSG_DONTWAIT,
1640                                           &err);
1641
1642                 if (skb == NULL)
1643                         goto out_err;
1644
1645                 /*
1646                  *      If you pass two values to the sock_alloc_send_skb
1647                  *      it tries to grab the large buffer with GFP_NOFS
1648                  *      (which can fail easily), and if it fails grab the
1649                  *      fallback size buffer which is under a page and will
1650                  *      succeed. [Alan]
1651                  */
1652                 size = min_t(int, size, skb_tailroom(skb));
1653
1654
1655                 /* Only send the fds in the first buffer */
1656                 err = unix_scm_to_skb(siocb->scm, skb, !fds_sent);
1657                 if (err < 0) {
1658                         kfree_skb(skb);
1659                         goto out_err;
1660                 }
1661                 max_level = err + 1;
1662                 fds_sent = true;
1663
1664                 err = memcpy_fromiovec(skb_put(skb, size), msg->msg_iov, size);
1665                 if (err) {
1666                         kfree_skb(skb);
1667                         goto out_err;
1668                 }
1669
1670                 unix_state_lock(other);
1671
1672                 if (sock_flag(other, SOCK_DEAD) ||
1673                     (other->sk_shutdown & RCV_SHUTDOWN))
1674                         goto pipe_err_free;
1675
1676                 maybe_add_creds(skb, sock, other);
1677                 skb_queue_tail(&other->sk_receive_queue, skb);
1678                 if (max_level > unix_sk(other)->recursion_level)
1679                         unix_sk(other)->recursion_level = max_level;
1680                 unix_state_unlock(other);
1681                 other->sk_data_ready(other, size);
1682                 sent += size;
1683         }
1684
1685         scm_destroy(siocb->scm);
1686         siocb->scm = NULL;
1687
1688         return sent;
1689
1690 pipe_err_free:
1691         unix_state_unlock(other);
1692         kfree_skb(skb);
1693 pipe_err:
1694         if (sent == 0 && !(msg->msg_flags&MSG_NOSIGNAL))
1695                 send_sig(SIGPIPE, current, 0);
1696         err = -EPIPE;
1697 out_err:
1698         scm_destroy(siocb->scm);
1699         siocb->scm = NULL;
1700         return sent ? : err;
1701 }
1702
1703 static int unix_seqpacket_sendmsg(struct kiocb *kiocb, struct socket *sock,
1704                                   struct msghdr *msg, size_t len)
1705 {
1706         int err;
1707         struct sock *sk = sock->sk;
1708
1709         err = sock_error(sk);
1710         if (err)
1711                 return err;
1712
1713         if (sk->sk_state != TCP_ESTABLISHED)
1714                 return -ENOTCONN;
1715
1716         if (msg->msg_namelen)
1717                 msg->msg_namelen = 0;
1718
1719         return unix_dgram_sendmsg(kiocb, sock, msg, len);
1720 }
1721
1722 static int unix_seqpacket_recvmsg(struct kiocb *iocb, struct socket *sock,
1723                               struct msghdr *msg, size_t size,
1724                               int flags)
1725 {
1726         struct sock *sk = sock->sk;
1727
1728         if (sk->sk_state != TCP_ESTABLISHED)
1729                 return -ENOTCONN;
1730
1731         return unix_dgram_recvmsg(iocb, sock, msg, size, flags);
1732 }
1733
1734 static void unix_copy_addr(struct msghdr *msg, struct sock *sk)
1735 {
1736         struct unix_sock *u = unix_sk(sk);
1737
1738         msg->msg_namelen = 0;
1739         if (u->addr) {
1740                 msg->msg_namelen = u->addr->len;
1741                 memcpy(msg->msg_name, u->addr->name, u->addr->len);
1742         }
1743 }
1744
1745 static int unix_dgram_recvmsg(struct kiocb *iocb, struct socket *sock,
1746                               struct msghdr *msg, size_t size,
1747                               int flags)
1748 {
1749         struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1750         struct scm_cookie tmp_scm;
1751         struct sock *sk = sock->sk;
1752         struct unix_sock *u = unix_sk(sk);
1753         int noblock = flags & MSG_DONTWAIT;
1754         struct sk_buff *skb;
1755         int err;
1756
1757         err = -EOPNOTSUPP;
1758         if (flags&MSG_OOB)
1759                 goto out;
1760
1761         msg->msg_namelen = 0;
1762
1763         err = mutex_lock_interruptible(&u->readlock);
1764         if (err) {
1765                 err = sock_intr_errno(sock_rcvtimeo(sk, noblock));
1766                 goto out;
1767         }
1768
1769         skb = skb_recv_datagram(sk, flags, noblock, &err);
1770         if (!skb) {
1771                 unix_state_lock(sk);
1772                 /* Signal EOF on disconnected non-blocking SEQPACKET socket. */
1773                 if (sk->sk_type == SOCK_SEQPACKET && err == -EAGAIN &&
1774                     (sk->sk_shutdown & RCV_SHUTDOWN))
1775                         err = 0;
1776                 unix_state_unlock(sk);
1777                 goto out_unlock;
1778         }
1779
1780         wake_up_interruptible_sync_poll(&u->peer_wait,
1781                                         POLLOUT | POLLWRNORM | POLLWRBAND);
1782
1783         if (msg->msg_name)
1784                 unix_copy_addr(msg, skb->sk);
1785
1786         if (size > skb->len)
1787                 size = skb->len;
1788         else if (size < skb->len)
1789                 msg->msg_flags |= MSG_TRUNC;
1790
1791         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, size);
1792         if (err)
1793                 goto out_free;
1794
1795         if (sock_flag(sk, SOCK_RCVTSTAMP))
1796                 __sock_recv_timestamp(msg, sk, skb);
1797
1798         if (!siocb->scm) {
1799                 siocb->scm = &tmp_scm;
1800                 memset(&tmp_scm, 0, sizeof(tmp_scm));
1801         }
1802         scm_set_cred(siocb->scm, UNIXCB(skb).pid, UNIXCB(skb).cred);
1803         unix_set_secdata(siocb->scm, skb);
1804
1805         if (!(flags & MSG_PEEK)) {
1806                 if (UNIXCB(skb).fp)
1807                         unix_detach_fds(siocb->scm, skb);
1808         } else {
1809                 /* It is questionable: on PEEK we could:
1810                    - do not return fds - good, but too simple 8)
1811                    - return fds, and do not return them on read (old strategy,
1812                      apparently wrong)
1813                    - clone fds (I chose it for now, it is the most universal
1814                      solution)
1815
1816                    POSIX 1003.1g does not actually define this clearly
1817                    at all. POSIX 1003.1g doesn't define a lot of things
1818                    clearly however!
1819
1820                 */
1821                 if (UNIXCB(skb).fp)
1822                         siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1823         }
1824         err = size;
1825
1826         scm_recv(sock, msg, siocb->scm, flags);
1827
1828 out_free:
1829         skb_free_datagram(sk, skb);
1830 out_unlock:
1831         mutex_unlock(&u->readlock);
1832 out:
1833         return err;
1834 }
1835
1836 /*
1837  *      Sleep until data has arrive. But check for races..
1838  */
1839
1840 static long unix_stream_data_wait(struct sock *sk, long timeo)
1841 {
1842         DEFINE_WAIT(wait);
1843
1844         unix_state_lock(sk);
1845
1846         for (;;) {
1847                 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1848
1849                 if (!skb_queue_empty(&sk->sk_receive_queue) ||
1850                     sk->sk_err ||
1851                     (sk->sk_shutdown & RCV_SHUTDOWN) ||
1852                     signal_pending(current) ||
1853                     !timeo)
1854                         break;
1855
1856                 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1857                 unix_state_unlock(sk);
1858                 timeo = schedule_timeout(timeo);
1859                 unix_state_lock(sk);
1860                 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1861         }
1862
1863         finish_wait(sk_sleep(sk), &wait);
1864         unix_state_unlock(sk);
1865         return timeo;
1866 }
1867
1868
1869
1870 static int unix_stream_recvmsg(struct kiocb *iocb, struct socket *sock,
1871                                struct msghdr *msg, size_t size,
1872                                int flags)
1873 {
1874         struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1875         struct scm_cookie tmp_scm;
1876         struct sock *sk = sock->sk;
1877         struct unix_sock *u = unix_sk(sk);
1878         struct sockaddr_un *sunaddr = msg->msg_name;
1879         int copied = 0;
1880         int check_creds = 0;
1881         int target;
1882         int err = 0;
1883         long timeo;
1884
1885         err = -EINVAL;
1886         if (sk->sk_state != TCP_ESTABLISHED)
1887                 goto out;
1888
1889         err = -EOPNOTSUPP;
1890         if (flags&MSG_OOB)
1891                 goto out;
1892
1893         target = sock_rcvlowat(sk, flags&MSG_WAITALL, size);
1894         timeo = sock_rcvtimeo(sk, flags&MSG_DONTWAIT);
1895
1896         msg->msg_namelen = 0;
1897
1898         /* Lock the socket to prevent queue disordering
1899          * while sleeps in memcpy_tomsg
1900          */
1901
1902         if (!siocb->scm) {
1903                 siocb->scm = &tmp_scm;
1904                 memset(&tmp_scm, 0, sizeof(tmp_scm));
1905         }
1906
1907         err = mutex_lock_interruptible(&u->readlock);
1908         if (err) {
1909                 err = sock_intr_errno(timeo);
1910                 goto out;
1911         }
1912
1913         do {
1914                 int chunk;
1915                 struct sk_buff *skb;
1916
1917                 unix_state_lock(sk);
1918                 skb = skb_dequeue(&sk->sk_receive_queue);
1919                 if (skb == NULL) {
1920                         unix_sk(sk)->recursion_level = 0;
1921                         if (copied >= target)
1922                                 goto unlock;
1923
1924                         /*
1925                          *      POSIX 1003.1g mandates this order.
1926                          */
1927
1928                         err = sock_error(sk);
1929                         if (err)
1930                                 goto unlock;
1931                         if (sk->sk_shutdown & RCV_SHUTDOWN)
1932                                 goto unlock;
1933
1934                         unix_state_unlock(sk);
1935                         err = -EAGAIN;
1936                         if (!timeo)
1937                                 break;
1938                         mutex_unlock(&u->readlock);
1939
1940                         timeo = unix_stream_data_wait(sk, timeo);
1941
1942                         if (signal_pending(current)
1943                             ||  mutex_lock_interruptible(&u->readlock)) {
1944                                 err = sock_intr_errno(timeo);
1945                                 goto out;
1946                         }
1947
1948                         continue;
1949  unlock:
1950                         unix_state_unlock(sk);
1951                         break;
1952                 }
1953                 unix_state_unlock(sk);
1954
1955                 if (check_creds) {
1956                         /* Never glue messages from different writers */
1957                         if ((UNIXCB(skb).pid  != siocb->scm->pid) ||
1958                             (UNIXCB(skb).cred != siocb->scm->cred)) {
1959                                 skb_queue_head(&sk->sk_receive_queue, skb);
1960                                 sk->sk_data_ready(sk, skb->len);
1961                                 break;
1962                         }
1963                 } else {
1964                         /* Copy credentials */
1965                         scm_set_cred(siocb->scm, UNIXCB(skb).pid, UNIXCB(skb).cred);
1966                         check_creds = 1;
1967                 }
1968
1969                 /* Copy address just once */
1970                 if (sunaddr) {
1971                         unix_copy_addr(msg, skb->sk);
1972                         sunaddr = NULL;
1973                 }
1974
1975                 chunk = min_t(unsigned int, skb->len, size);
1976                 if (memcpy_toiovec(msg->msg_iov, skb->data, chunk)) {
1977                         skb_queue_head(&sk->sk_receive_queue, skb);
1978                         sk->sk_data_ready(sk, skb->len);
1979                         if (copied == 0)
1980                                 copied = -EFAULT;
1981                         break;
1982                 }
1983                 copied += chunk;
1984                 size -= chunk;
1985
1986                 /* Mark read part of skb as used */
1987                 if (!(flags & MSG_PEEK)) {
1988                         skb_pull(skb, chunk);
1989
1990                         if (UNIXCB(skb).fp)
1991                                 unix_detach_fds(siocb->scm, skb);
1992
1993                         /* put the skb back if we didn't use it up.. */
1994                         if (skb->len) {
1995                                 skb_queue_head(&sk->sk_receive_queue, skb);
1996                                 sk->sk_data_ready(sk, skb->len);
1997                                 break;
1998                         }
1999
2000                         consume_skb(skb);
2001
2002                         if (siocb->scm->fp)
2003                                 break;
2004                 } else {
2005                         /* It is questionable, see note in unix_dgram_recvmsg.
2006                          */
2007                         if (UNIXCB(skb).fp)
2008                                 siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
2009
2010                         /* put message back and return */
2011                         skb_queue_head(&sk->sk_receive_queue, skb);
2012                         sk->sk_data_ready(sk, skb->len);
2013                         break;
2014                 }
2015         } while (size);
2016
2017         mutex_unlock(&u->readlock);
2018         scm_recv(sock, msg, siocb->scm, flags);
2019 out:
2020         return copied ? : err;
2021 }
2022
2023 static int unix_shutdown(struct socket *sock, int mode)
2024 {
2025         struct sock *sk = sock->sk;
2026         struct sock *other;
2027
2028         mode = (mode+1)&(RCV_SHUTDOWN|SEND_SHUTDOWN);
2029
2030         if (!mode)
2031                 return 0;
2032
2033         unix_state_lock(sk);
2034         sk->sk_shutdown |= mode;
2035         other = unix_peer(sk);
2036         if (other)
2037                 sock_hold(other);
2038         unix_state_unlock(sk);
2039         sk->sk_state_change(sk);
2040
2041         if (other &&
2042                 (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET)) {
2043
2044                 int peer_mode = 0;
2045
2046                 if (mode&RCV_SHUTDOWN)
2047                         peer_mode |= SEND_SHUTDOWN;
2048                 if (mode&SEND_SHUTDOWN)
2049                         peer_mode |= RCV_SHUTDOWN;
2050                 unix_state_lock(other);
2051                 other->sk_shutdown |= peer_mode;
2052                 unix_state_unlock(other);
2053                 other->sk_state_change(other);
2054                 if (peer_mode == SHUTDOWN_MASK)
2055                         sk_wake_async(other, SOCK_WAKE_WAITD, POLL_HUP);
2056                 else if (peer_mode & RCV_SHUTDOWN)
2057                         sk_wake_async(other, SOCK_WAKE_WAITD, POLL_IN);
2058         }
2059         if (other)
2060                 sock_put(other);
2061
2062         return 0;
2063 }
2064
2065 static int unix_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2066 {
2067         struct sock *sk = sock->sk;
2068         long amount = 0;
2069         int err;
2070
2071         switch (cmd) {
2072         case SIOCOUTQ:
2073                 amount = sk_wmem_alloc_get(sk);
2074                 err = put_user(amount, (int __user *)arg);
2075                 break;
2076         case SIOCINQ:
2077                 {
2078                         struct sk_buff *skb;
2079
2080                         if (sk->sk_state == TCP_LISTEN) {
2081                                 err = -EINVAL;
2082                                 break;
2083                         }
2084
2085                         spin_lock(&sk->sk_receive_queue.lock);
2086                         if (sk->sk_type == SOCK_STREAM ||
2087                             sk->sk_type == SOCK_SEQPACKET) {
2088                                 skb_queue_walk(&sk->sk_receive_queue, skb)
2089                                         amount += skb->len;
2090                         } else {
2091                                 skb = skb_peek(&sk->sk_receive_queue);
2092                                 if (skb)
2093                                         amount = skb->len;
2094                         }
2095                         spin_unlock(&sk->sk_receive_queue.lock);
2096                         err = put_user(amount, (int __user *)arg);
2097                         break;
2098                 }
2099
2100         default:
2101                 err = -ENOIOCTLCMD;
2102                 break;
2103         }
2104         return err;
2105 }
2106
2107 static unsigned int unix_poll(struct file *file, struct socket *sock, poll_table *wait)
2108 {
2109         struct sock *sk = sock->sk;
2110         unsigned int mask;
2111
2112         sock_poll_wait(file, sk_sleep(sk), wait);
2113         mask = 0;
2114
2115         /* exceptional events? */
2116         if (sk->sk_err)
2117                 mask |= POLLERR;
2118         if (sk->sk_shutdown == SHUTDOWN_MASK)
2119                 mask |= POLLHUP;
2120         if (sk->sk_shutdown & RCV_SHUTDOWN)
2121                 mask |= POLLRDHUP | POLLIN | POLLRDNORM;
2122
2123         /* readable? */
2124         if (!skb_queue_empty(&sk->sk_receive_queue))
2125                 mask |= POLLIN | POLLRDNORM;
2126
2127         /* Connection-based need to check for termination and startup */
2128         if ((sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) &&
2129             sk->sk_state == TCP_CLOSE)
2130                 mask |= POLLHUP;
2131
2132         /*
2133          * we set writable also when the other side has shut down the
2134          * connection. This prevents stuck sockets.
2135          */
2136         if (unix_writable(sk))
2137                 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2138
2139         return mask;
2140 }
2141
2142 static unsigned int unix_dgram_poll(struct file *file, struct socket *sock,
2143                                     poll_table *wait)
2144 {
2145         struct sock *sk = sock->sk, *other;
2146         unsigned int mask, writable;
2147
2148         sock_poll_wait(file, sk_sleep(sk), wait);
2149         mask = 0;
2150
2151         /* exceptional events? */
2152         if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue))
2153                 mask |= POLLERR;
2154         if (sk->sk_shutdown & RCV_SHUTDOWN)
2155                 mask |= POLLRDHUP | POLLIN | POLLRDNORM;
2156         if (sk->sk_shutdown == SHUTDOWN_MASK)
2157                 mask |= POLLHUP;
2158
2159         /* readable? */
2160         if (!skb_queue_empty(&sk->sk_receive_queue))
2161                 mask |= POLLIN | POLLRDNORM;
2162
2163         /* Connection-based need to check for termination and startup */
2164         if (sk->sk_type == SOCK_SEQPACKET) {
2165                 if (sk->sk_state == TCP_CLOSE)
2166                         mask |= POLLHUP;
2167                 /* connection hasn't started yet? */
2168                 if (sk->sk_state == TCP_SYN_SENT)
2169                         return mask;
2170         }
2171
2172         /* No write status requested, avoid expensive OUT tests. */
2173         if (wait && !(wait->key & (POLLWRBAND | POLLWRNORM | POLLOUT)))
2174                 return mask;
2175
2176         writable = unix_writable(sk);
2177         other = unix_peer_get(sk);
2178         if (other) {
2179                 if (unix_peer(other) != sk) {
2180                         sock_poll_wait(file, &unix_sk(other)->peer_wait, wait);
2181                         if (unix_recvq_full(other))
2182                                 writable = 0;
2183                 }
2184                 sock_put(other);
2185         }
2186
2187         if (writable)
2188                 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2189         else
2190                 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
2191
2192         return mask;
2193 }
2194
2195 #ifdef CONFIG_PROC_FS
2196 static struct sock *first_unix_socket(int *i)
2197 {
2198         for (*i = 0; *i <= UNIX_HASH_SIZE; (*i)++) {
2199                 if (!hlist_empty(&unix_socket_table[*i]))
2200                         return __sk_head(&unix_socket_table[*i]);
2201         }
2202         return NULL;
2203 }
2204
2205 static struct sock *next_unix_socket(int *i, struct sock *s)
2206 {
2207         struct sock *next = sk_next(s);
2208         /* More in this chain? */
2209         if (next)
2210                 return next;
2211         /* Look for next non-empty chain. */
2212         for ((*i)++; *i <= UNIX_HASH_SIZE; (*i)++) {
2213                 if (!hlist_empty(&unix_socket_table[*i]))
2214                         return __sk_head(&unix_socket_table[*i]);
2215         }
2216         return NULL;
2217 }
2218
2219 struct unix_iter_state {
2220         struct seq_net_private p;
2221         int i;
2222 };
2223
2224 static struct sock *unix_seq_idx(struct seq_file *seq, loff_t pos)
2225 {
2226         struct unix_iter_state *iter = seq->private;
2227         loff_t off = 0;
2228         struct sock *s;
2229
2230         for (s = first_unix_socket(&iter->i); s; s = next_unix_socket(&iter->i, s)) {
2231                 if (sock_net(s) != seq_file_net(seq))
2232                         continue;
2233                 if (off == pos)
2234                         return s;
2235                 ++off;
2236         }
2237         return NULL;
2238 }
2239
2240 static void *unix_seq_start(struct seq_file *seq, loff_t *pos)
2241         __acquires(unix_table_lock)
2242 {
2243         spin_lock(&unix_table_lock);
2244         return *pos ? unix_seq_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2245 }
2246
2247 static void *unix_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2248 {
2249         struct unix_iter_state *iter = seq->private;
2250         struct sock *sk = v;
2251         ++*pos;
2252
2253         if (v == SEQ_START_TOKEN)
2254                 sk = first_unix_socket(&iter->i);
2255         else
2256                 sk = next_unix_socket(&iter->i, sk);
2257         while (sk && (sock_net(sk) != seq_file_net(seq)))
2258                 sk = next_unix_socket(&iter->i, sk);
2259         return sk;
2260 }
2261
2262 static void unix_seq_stop(struct seq_file *seq, void *v)
2263         __releases(unix_table_lock)
2264 {
2265         spin_unlock(&unix_table_lock);
2266 }
2267
2268 static int unix_seq_show(struct seq_file *seq, void *v)
2269 {
2270
2271         if (v == SEQ_START_TOKEN)
2272                 seq_puts(seq, "Num       RefCount Protocol Flags    Type St "
2273                          "Inode Path\n");
2274         else {
2275                 struct sock *s = v;
2276                 struct unix_sock *u = unix_sk(s);
2277                 unix_state_lock(s);
2278
2279                 seq_printf(seq, "%pK: %08X %08X %08X %04X %02X %5lu",
2280                         s,
2281                         atomic_read(&s->sk_refcnt),
2282                         0,
2283                         s->sk_state == TCP_LISTEN ? __SO_ACCEPTCON : 0,
2284                         s->sk_type,
2285                         s->sk_socket ?
2286                         (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTED : SS_UNCONNECTED) :
2287                         (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTING : SS_DISCONNECTING),
2288                         sock_i_ino(s));
2289
2290                 if (u->addr) {
2291                         int i, len;
2292                         seq_putc(seq, ' ');
2293
2294                         i = 0;
2295                         len = u->addr->len - sizeof(short);
2296                         if (!UNIX_ABSTRACT(s))
2297                                 len--;
2298                         else {
2299                                 seq_putc(seq, '@');
2300                                 i++;
2301                         }
2302                         for ( ; i < len; i++)
2303                                 seq_putc(seq, u->addr->name->sun_path[i]);
2304                 }
2305                 unix_state_unlock(s);
2306                 seq_putc(seq, '\n');
2307         }
2308
2309         return 0;
2310 }
2311
2312 static const struct seq_operations unix_seq_ops = {
2313         .start  = unix_seq_start,
2314         .next   = unix_seq_next,
2315         .stop   = unix_seq_stop,
2316         .show   = unix_seq_show,
2317 };
2318
2319 static int unix_seq_open(struct inode *inode, struct file *file)
2320 {
2321         return seq_open_net(inode, file, &unix_seq_ops,
2322                             sizeof(struct unix_iter_state));
2323 }
2324
2325 static const struct file_operations unix_seq_fops = {
2326         .owner          = THIS_MODULE,
2327         .open           = unix_seq_open,
2328         .read           = seq_read,
2329         .llseek         = seq_lseek,
2330         .release        = seq_release_net,
2331 };
2332
2333 #endif
2334
2335 static const struct net_proto_family unix_family_ops = {
2336         .family = PF_UNIX,
2337         .create = unix_create,
2338         .owner  = THIS_MODULE,
2339 };
2340
2341
2342 static int __net_init unix_net_init(struct net *net)
2343 {
2344         int error = -ENOMEM;
2345
2346         net->unx.sysctl_max_dgram_qlen = 10;
2347         if (unix_sysctl_register(net))
2348                 goto out;
2349
2350 #ifdef CONFIG_PROC_FS
2351         if (!proc_net_fops_create(net, "unix", 0, &unix_seq_fops)) {
2352                 unix_sysctl_unregister(net);
2353                 goto out;
2354         }
2355 #endif
2356         error = 0;
2357 out:
2358         return error;
2359 }
2360
2361 static void __net_exit unix_net_exit(struct net *net)
2362 {
2363         unix_sysctl_unregister(net);
2364         proc_net_remove(net, "unix");
2365 }
2366
2367 static struct pernet_operations unix_net_ops = {
2368         .init = unix_net_init,
2369         .exit = unix_net_exit,
2370 };
2371
2372 static int __init af_unix_init(void)
2373 {
2374         int rc = -1;
2375         struct sk_buff *dummy_skb;
2376
2377         BUILD_BUG_ON(sizeof(struct unix_skb_parms) > sizeof(dummy_skb->cb));
2378
2379         rc = proto_register(&unix_proto, 1);
2380         if (rc != 0) {
2381                 printk(KERN_CRIT "%s: Cannot create unix_sock SLAB cache!\n",
2382                        __func__);
2383                 goto out;
2384         }
2385
2386         sock_register(&unix_family_ops);
2387         register_pernet_subsys(&unix_net_ops);
2388 out:
2389         return rc;
2390 }
2391
2392 static void __exit af_unix_exit(void)
2393 {
2394         sock_unregister(PF_UNIX);
2395         proto_unregister(&unix_proto);
2396         unregister_pernet_subsys(&unix_net_ops);
2397 }
2398
2399 /* Earlier than device_initcall() so that other drivers invoking
2400    request_module() don't end up in a loop when modprobe tries
2401    to use a UNIX socket. But later than subsys_initcall() because
2402    we depend on stuff initialised there */
2403 fs_initcall(af_unix_init);
2404 module_exit(af_unix_exit);
2405
2406 MODULE_LICENSE("GPL");
2407 MODULE_ALIAS_NETPROTO(PF_UNIX);