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_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         struct dentry *dentry = NULL;
 812         struct nameidata nd;
 813         int err;
 814         unsigned hash;
 815         struct unix_address *addr;
 816         struct hlist_head *list;
 817
 818         err = -EINVAL;
 819         if (sunaddr->sun_family != AF_UNIX)
 820                 goto out;
 821
 822         if (addr_len == sizeof(short)) {
 823                 err = unix_autobind(sock);
 824                 goto out;
 825         }
 826
 827         err = unix_mkname(sunaddr, addr_len, &hash);
 828         if (err < 0)
 829                 goto out;
 830         addr_len = err;
 831
 832         mutex_lock(&u->readlock);
 833
 834         err = -EINVAL;
 835         if (u->addr)
 836                 goto out_up;
 837
 838         err = -ENOMEM;
 839         addr = kmalloc(sizeof(*addr)+addr_len, GFP_KERNEL);
 840         if (!addr)
 841                 goto out_up;
 842
 843         memcpy(addr->name, sunaddr, addr_len);
 844         addr->len = addr_len;
 845         addr->hash = hash ^ sk->sk_type;
 846         atomic_set(&addr->refcnt, 1);
 847
 848         if (sunaddr->sun_path[0]) {
 849                 unsigned int mode;
 850                 err = 0;
 851                 /*
 852                  * Get the parent directory, calculate the hash for last
 853                  * component.
 854                  */
 855                 err = path_lookup(sunaddr->sun_path, LOOKUP_PARENT, &nd);
 856                 if (err)
 857                         goto out_mknod_parent;
 858
 859                 dentry = lookup_create(&nd, 0);
 860                 err = PTR_ERR(dentry);
 861                 if (IS_ERR(dentry))
 862                         goto out_mknod_unlock;
 863
 864                 /*
 865                  * All right, let's create it.
 866                  */
 867                 mode = S_IFSOCK |
 868                        (SOCK_INODE(sock)->i_mode & ~current_umask());
 869                 err = mnt_want_write(nd.path.mnt);
 870                 if (err)
 871                         goto out_mknod_dput;
 872                 err = security_path_mknod(&nd.path, dentry, mode, 0);
 873                 if (err)
 874                         goto out_mknod_drop_write;
 875                 err = vfs_mknod(nd.path.dentry->d_inode, dentry, mode, 0);
 876 out_mknod_drop_write:
 877                 mnt_drop_write(nd.path.mnt);
 878                 if (err)
 879                         goto out_mknod_dput;
 880                 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
 881                 dput(nd.path.dentry);
 882                 nd.path.dentry = dentry;
 883
 884                 addr->hash = UNIX_HASH_SIZE;
 885         }
 886
 887         spin_lock(&unix_table_lock);
 888
 889         if (!sunaddr->sun_path[0]) {
 890                 err = -EADDRINUSE;
 891                 if (__unix_find_socket_byname(net, sunaddr, addr_len,
 892                                               sk->sk_type, hash)) {
 893                         unix_release_addr(addr);
 894                         goto out_unlock;
 895                 }
 896
 897                 list = &unix_socket_table[addr->hash];
 898         } else {
 899                 list = &unix_socket_table[dentry->d_inode->i_ino & (UNIX_HASH_SIZE-1)];
 900                 u->dentry = nd.path.dentry;
 901                 u->mnt    = nd.path.mnt;
 902         }
 903
 904         err = 0;
 905         __unix_remove_socket(sk);
 906         u->addr = addr;
 907         __unix_insert_socket(list, sk);
 908
 909 out_unlock:
 910         spin_unlock(&unix_table_lock);
 911 out_up:
 912         mutex_unlock(&u->readlock);
 913 out:
 914         return err;
 915
 916 out_mknod_dput:
 917         dput(dentry);
 918 out_mknod_unlock:
 919         mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
 920         path_put(&nd.path);
 921 out_mknod_parent:
 922         if (err == -EEXIST)
 923                 err = -EADDRINUSE;
 924         unix_release_addr(addr);
 925         goto out_up;
 926 }
 927
 928 static void unix_state_double_lock(struct sock *sk1, struct sock *sk2)
 929 {
 930         if (unlikely(sk1 == sk2) || !sk2) {
 931                 unix_state_lock(sk1);
 932                 return;
 933         }
 934         if (sk1 < sk2) {
 935                 unix_state_lock(sk1);
 936                 unix_state_lock_nested(sk2);
 937         } else {
 938                 unix_state_lock(sk2);
 939                 unix_state_lock_nested(sk1);
 940         }
 941 }
 942
 943 static void unix_state_double_unlock(struct sock *sk1, struct sock *sk2)
 944 {
 945         if (unlikely(sk1 == sk2) || !sk2) {
 946                 unix_state_unlock(sk1);
 947                 return;
 948         }
 949         unix_state_unlock(sk1);
 950         unix_state_unlock(sk2);
 951 }
 952
 953 static int unix_dgram_connect(struct socket *sock, struct sockaddr *addr,
 954                               int alen, int flags)
 955 {
 956         struct sock *sk = sock->sk;
 957         struct net *net = sock_net(sk);
 958         struct sockaddr_un *sunaddr = (struct sockaddr_un *)addr;
 959         struct sock *other;
 960         unsigned hash;
 961         int err;
 962
 963         if (addr->sa_family != AF_UNSPEC) {
 964                 err = unix_mkname(sunaddr, alen, &hash);
 965                 if (err < 0)
 966                         goto out;
 967                 alen = err;
 968
 969                 if (test_bit(SOCK_PASSCRED, &sock->flags) &&
 970                     !unix_sk(sk)->addr && (err = unix_autobind(sock)) != 0)
 971                         goto out;
 972
 973 restart:
 974                 other = unix_find_other(net, sunaddr, alen, sock->type, hash, &err);
 975                 if (!other)
 976                         goto out;
 977
 978                 unix_state_double_lock(sk, other);
 979
 980                 /* Apparently VFS overslept socket death. Retry. */
 981                 if (sock_flag(other, SOCK_DEAD)) {
 982                         unix_state_double_unlock(sk, other);
 983                         sock_put(other);
 984                         goto restart;
 985                 }
 986
 987                 err = -EPERM;
 988                 if (!unix_may_send(sk, other))
 989                         goto out_unlock;
 990
 991                 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
 992                 if (err)
 993                         goto out_unlock;
 994
 995         } else {
 996                 /*
 997                  *      1003.1g breaking connected state with AF_UNSPEC
 998                  */
 999                 other = NULL;
1000                 unix_state_double_lock(sk, other);
1001         }
1002
1003         /*
1004          * If it was connected, reconnect.
1005          */
1006         if (unix_peer(sk)) {
1007                 struct sock *old_peer = unix_peer(sk);
1008                 unix_peer(sk) = other;
1009                 unix_state_double_unlock(sk, other);
1010
1011                 if (other != old_peer)
1012                         unix_dgram_disconnected(sk, old_peer);
1013                 sock_put(old_peer);
1014         } else {
1015                 unix_peer(sk) = other;
1016                 unix_state_double_unlock(sk, other);
1017         }
1018         return 0;
1019
1020 out_unlock:
1021         unix_state_double_unlock(sk, other);
1022         sock_put(other);
1023 out:
1024         return err;
1025 }
1026
1027 static long unix_wait_for_peer(struct sock *other, long timeo)
1028 {
1029         struct unix_sock *u = unix_sk(other);
1030         int sched;
1031         DEFINE_WAIT(wait);
1032
1033         prepare_to_wait_exclusive(&u->peer_wait, &wait, TASK_INTERRUPTIBLE);
1034
1035         sched = !sock_flag(other, SOCK_DEAD) &&
1036                 !(other->sk_shutdown & RCV_SHUTDOWN) &&
1037                 unix_recvq_full(other);
1038
1039         unix_state_unlock(other);
1040
1041         if (sched)
1042                 timeo = schedule_timeout(timeo);
1043
1044         finish_wait(&u->peer_wait, &wait);
1045         return timeo;
1046 }
1047
1048 static int unix_stream_connect(struct socket *sock, struct sockaddr *uaddr,
1049                                int addr_len, int flags)
1050 {
1051         struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
1052         struct sock *sk = sock->sk;
1053         struct net *net = sock_net(sk);
1054         struct unix_sock *u = unix_sk(sk), *newu, *otheru;
1055         struct sock *newsk = NULL;
1056         struct sock *other = NULL;
1057         struct sk_buff *skb = NULL;
1058         unsigned hash;
1059         int st;
1060         int err;
1061         long timeo;
1062
1063         err = unix_mkname(sunaddr, addr_len, &hash);
1064         if (err < 0)
1065                 goto out;
1066         addr_len = err;
1067
1068         if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr &&
1069             (err = unix_autobind(sock)) != 0)
1070                 goto out;
1071
1072         timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
1073
1074         /* First of all allocate resources.
1075            If we will make it after state is locked,
1076            we will have to recheck all again in any case.
1077          */
1078
1079         err = -ENOMEM;
1080
1081         /* create new sock for complete connection */
1082         newsk = unix_create1(sock_net(sk), NULL);
1083         if (newsk == NULL)
1084                 goto out;
1085
1086         /* Allocate skb for sending to listening sock */
1087         skb = sock_wmalloc(newsk, 1, 0, GFP_KERNEL);
1088         if (skb == NULL)
1089                 goto out;
1090
1091 restart:
1092         /*  Find listening sock. */
1093         other = unix_find_other(net, sunaddr, addr_len, sk->sk_type, hash, &err);
1094         if (!other)
1095                 goto out;
1096
1097         /* Latch state of peer */
1098         unix_state_lock(other);
1099
1100         /* Apparently VFS overslept socket death. Retry. */
1101         if (sock_flag(other, SOCK_DEAD)) {
1102                 unix_state_unlock(other);
1103                 sock_put(other);
1104                 goto restart;
1105         }
1106
1107         err = -ECONNREFUSED;
1108         if (other->sk_state != TCP_LISTEN)
1109                 goto out_unlock;
1110         if (other->sk_shutdown & RCV_SHUTDOWN)
1111                 goto out_unlock;
1112
1113         if (unix_recvq_full(other)) {
1114                 err = -EAGAIN;
1115                 if (!timeo)
1116                         goto out_unlock;
1117
1118                 timeo = unix_wait_for_peer(other, timeo);
1119
1120                 err = sock_intr_errno(timeo);
1121                 if (signal_pending(current))
1122                         goto out;
1123                 sock_put(other);
1124                 goto restart;
1125         }
1126
1127         /* Latch our state.
1128
1129            It is tricky place. We need to grab write lock and cannot
1130            drop lock on peer. It is dangerous because deadlock is
1131            possible. Connect to self case and simultaneous
1132            attempt to connect are eliminated by checking socket
1133            state. other is TCP_LISTEN, if sk is TCP_LISTEN we
1134            check this before attempt to grab lock.
1135
1136            Well, and we have to recheck the state after socket locked.
1137          */
1138         st = sk->sk_state;
1139
1140         switch (st) {
1141         case TCP_CLOSE:
1142                 /* This is ok... continue with connect */
1143                 break;
1144         case TCP_ESTABLISHED:
1145                 /* Socket is already connected */
1146                 err = -EISCONN;
1147                 goto out_unlock;
1148         default:
1149                 err = -EINVAL;
1150                 goto out_unlock;
1151         }
1152
1153         unix_state_lock_nested(sk);
1154
1155         if (sk->sk_state != st) {
1156                 unix_state_unlock(sk);
1157                 unix_state_unlock(other);
1158                 sock_put(other);
1159                 goto restart;
1160         }
1161
1162         err = security_unix_stream_connect(sk, other, newsk);
1163         if (err) {
1164                 unix_state_unlock(sk);
1165                 goto out_unlock;
1166         }
1167
1168         /* The way is open! Fastly set all the necessary fields... */
1169
1170         sock_hold(sk);
1171         unix_peer(newsk)        = sk;
1172         newsk->sk_state         = TCP_ESTABLISHED;
1173         newsk->sk_type          = sk->sk_type;
1174         init_peercred(newsk);
1175         newu = unix_sk(newsk);
1176         newsk->sk_wq            = &newu->peer_wq;
1177         otheru = unix_sk(other);
1178
1179         /* copy address information from listening to new sock*/
1180         if (otheru->addr) {
1181                 atomic_inc(&otheru->addr->refcnt);
1182                 newu->addr = otheru->addr;
1183         }
1184         if (otheru->dentry) {
1185                 newu->dentry    = dget(otheru->dentry);
1186                 newu->mnt       = mntget(otheru->mnt);
1187         }
1188
1189         /* Set credentials */
1190         copy_peercred(sk, other);
1191
1192         sock->state     = SS_CONNECTED;
1193         sk->sk_state    = TCP_ESTABLISHED;
1194         sock_hold(newsk);
1195
1196         smp_mb__after_atomic_inc();     /* sock_hold() does an atomic_inc() */
1197         unix_peer(sk)   = newsk;
1198
1199         unix_state_unlock(sk);
1200
1201         /* take ten and and send info to listening sock */
1202         spin_lock(&other->sk_receive_queue.lock);
1203         __skb_queue_tail(&other->sk_receive_queue, skb);
1204         spin_unlock(&other->sk_receive_queue.lock);
1205         unix_state_unlock(other);
1206         other->sk_data_ready(other, 0);
1207         sock_put(other);
1208         return 0;
1209
1210 out_unlock:
1211         if (other)
1212                 unix_state_unlock(other);
1213
1214 out:
1215         kfree_skb(skb);
1216         if (newsk)
1217                 unix_release_sock(newsk, 0);
1218         if (other)
1219                 sock_put(other);
1220         return err;
1221 }
1222
1223 static int unix_socketpair(struct socket *socka, struct socket *sockb)
1224 {
1225         struct sock *ska = socka->sk, *skb = sockb->sk;
1226
1227         /* Join our sockets back to back */
1228         sock_hold(ska);
1229         sock_hold(skb);
1230         unix_peer(ska) = skb;
1231         unix_peer(skb) = ska;
1232         init_peercred(ska);
1233         init_peercred(skb);
1234
1235         if (ska->sk_type != SOCK_DGRAM) {
1236                 ska->sk_state = TCP_ESTABLISHED;
1237                 skb->sk_state = TCP_ESTABLISHED;
1238                 socka->state  = SS_CONNECTED;
1239                 sockb->state  = SS_CONNECTED;
1240         }
1241         return 0;
1242 }
1243
1244 static int unix_accept(struct socket *sock, struct socket *newsock, int flags)
1245 {
1246         struct sock *sk = sock->sk;
1247         struct sock *tsk;
1248         struct sk_buff *skb;
1249         int err;
1250
1251         err = -EOPNOTSUPP;
1252         if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
1253                 goto out;
1254
1255         err = -EINVAL;
1256         if (sk->sk_state != TCP_LISTEN)
1257                 goto out;
1258
1259         /* If socket state is TCP_LISTEN it cannot change (for now...),
1260          * so that no locks are necessary.
1261          */
1262
1263         skb = skb_recv_datagram(sk, 0, flags&O_NONBLOCK, &err);
1264         if (!skb) {
1265                 /* This means receive shutdown. */
1266                 if (err == 0)
1267                         err = -EINVAL;
1268                 goto out;
1269         }
1270
1271         tsk = skb->sk;
1272         skb_free_datagram(sk, skb);
1273         wake_up_interruptible(&unix_sk(sk)->peer_wait);
1274
1275         /* attach accepted sock to socket */
1276         unix_state_lock(tsk);
1277         newsock->state = SS_CONNECTED;
1278         sock_graft(tsk, newsock);
1279         unix_state_unlock(tsk);
1280         return 0;
1281
1282 out:
1283         return err;
1284 }
1285
1286
1287 static int unix_getname(struct socket *sock, struct sockaddr *uaddr, int *uaddr_len, int peer)
1288 {
1289         struct sock *sk = sock->sk;
1290         struct unix_sock *u;
1291         DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr, uaddr);
1292         int err = 0;
1293
1294         if (peer) {
1295                 sk = unix_peer_get(sk);
1296
1297                 err = -ENOTCONN;
1298                 if (!sk)
1299                         goto out;
1300                 err = 0;
1301         } else {
1302                 sock_hold(sk);
1303         }
1304
1305         u = unix_sk(sk);
1306         unix_state_lock(sk);
1307         if (!u->addr) {
1308                 sunaddr->sun_family = AF_UNIX;
1309                 sunaddr->sun_path[0] = 0;
1310                 *uaddr_len = sizeof(short);
1311         } else {
1312                 struct unix_address *addr = u->addr;
1313
1314                 *uaddr_len = addr->len;
1315                 memcpy(sunaddr, addr->name, *uaddr_len);
1316         }
1317         unix_state_unlock(sk);
1318         sock_put(sk);
1319 out:
1320         return err;
1321 }
1322
1323 static void unix_detach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1324 {
1325         int i;
1326
1327         scm->fp = UNIXCB(skb).fp;
1328         UNIXCB(skb).fp = NULL;
1329
1330         for (i = scm->fp->count-1; i >= 0; i--)
1331                 unix_notinflight(scm->fp->fp[i]);
1332 }
1333
1334 static void unix_destruct_scm(struct sk_buff *skb)
1335 {
1336         struct scm_cookie scm;
1337         memset(&scm, 0, sizeof(scm));
1338         scm.pid  = UNIXCB(skb).pid;
1339         scm.cred = UNIXCB(skb).cred;
1340         if (UNIXCB(skb).fp)
1341                 unix_detach_fds(&scm, skb);
1342
1343         /* Alas, it calls VFS */
1344         /* So fscking what? fput() had been SMP-safe since the last Summer */
1345         scm_destroy(&scm);
1346         sock_wfree(skb);
1347 }
1348
1349 #define MAX_RECURSION_LEVEL 4
1350
1351 static int unix_attach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1352 {
1353         int i;
1354         unsigned char max_level = 0;
1355         int unix_sock_count = 0;
1356
1357         for (i = scm->fp->count - 1; i >= 0; i--) {
1358                 struct sock *sk = unix_get_socket(scm->fp->fp[i]);
1359
1360                 if (sk) {
1361                         unix_sock_count++;
1362                         max_level = max(max_level,
1363                                         unix_sk(sk)->recursion_level);
1364                 }
1365         }
1366         if (unlikely(max_level > MAX_RECURSION_LEVEL))
1367                 return -ETOOMANYREFS;
1368
1369         /*
1370          * Need to duplicate file references for the sake of garbage
1371          * collection.  Otherwise a socket in the fps might become a
1372          * candidate for GC while the skb is not yet queued.
1373          */
1374         UNIXCB(skb).fp = scm_fp_dup(scm->fp);
1375         if (!UNIXCB(skb).fp)
1376                 return -ENOMEM;
1377
1378         if (unix_sock_count) {
1379                 for (i = scm->fp->count - 1; i >= 0; i--)
1380                         unix_inflight(scm->fp->fp[i]);
1381         }
1382         return max_level;
1383 }
1384
1385 static int unix_scm_to_skb(struct scm_cookie *scm, struct sk_buff *skb, bool send_fds)
1386 {
1387         int err = 0;
1388         UNIXCB(skb).pid  = get_pid(scm->pid);
1389         UNIXCB(skb).cred = get_cred(scm->cred);
1390         UNIXCB(skb).fp = NULL;
1391         if (scm->fp && send_fds)
1392                 err = unix_attach_fds(scm, skb);
1393
1394         skb->destructor = unix_destruct_scm;
1395         return err;
1396 }
1397
1398 /*
1399  *      Send AF_UNIX data.
1400  */
1401
1402 static int unix_dgram_sendmsg(struct kiocb *kiocb, struct socket *sock,
1403                               struct msghdr *msg, size_t len)
1404 {
1405         struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1406         struct sock *sk = sock->sk;
1407         struct net *net = sock_net(sk);
1408         struct unix_sock *u = unix_sk(sk);
1409         struct sockaddr_un *sunaddr = msg->msg_name;
1410         struct sock *other = NULL;
1411         int namelen = 0; /* fake GCC */
1412         int err;
1413         unsigned hash;
1414         struct sk_buff *skb;
1415         long timeo;
1416         struct scm_cookie tmp_scm;
1417         int max_level;
1418
1419         if (NULL == siocb->scm)
1420                 siocb->scm = &tmp_scm;
1421         wait_for_unix_gc();
1422         err = scm_send(sock, msg, siocb->scm);
1423         if (err < 0)
1424                 return err;
1425
1426         err = -EOPNOTSUPP;
1427         if (msg->msg_flags&MSG_OOB)
1428                 goto out;
1429
1430         if (msg->msg_namelen) {
1431                 err = unix_mkname(sunaddr, msg->msg_namelen, &hash);
1432                 if (err < 0)
1433                         goto out;
1434                 namelen = err;
1435         } else {
1436                 sunaddr = NULL;
1437                 err = -ENOTCONN;
1438                 other = unix_peer_get(sk);
1439                 if (!other)
1440                         goto out;
1441         }
1442
1443         if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr
1444             && (err = unix_autobind(sock)) != 0)
1445                 goto out;
1446
1447         err = -EMSGSIZE;
1448         if (len > sk->sk_sndbuf - 32)
1449                 goto out;
1450
1451         skb = sock_alloc_send_skb(sk, len, msg->msg_flags&MSG_DONTWAIT, &err);
1452         if (skb == NULL)
1453                 goto out;
1454
1455         err = unix_scm_to_skb(siocb->scm, skb, true);
1456         if (err < 0)
1457                 goto out_free;
1458         max_level = err + 1;
1459         unix_get_secdata(siocb->scm, skb);
1460
1461         skb_reset_transport_header(skb);
1462         err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
1463         if (err)
1464                 goto out_free;
1465
1466         timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1467
1468 restart:
1469         if (!other) {
1470                 err = -ECONNRESET;
1471                 if (sunaddr == NULL)
1472                         goto out_free;
1473
1474                 other = unix_find_other(net, sunaddr, namelen, sk->sk_type,
1475                                         hash, &err);
1476                 if (other == NULL)
1477                         goto out_free;
1478         }
1479
1480         unix_state_lock(other);
1481         err = -EPERM;
1482         if (!unix_may_send(sk, other))
1483                 goto out_unlock;
1484
1485         if (sock_flag(other, SOCK_DEAD)) {
1486                 /*
1487                  *      Check with 1003.1g - what should
1488                  *      datagram error
1489                  */
1490                 unix_state_unlock(other);
1491                 sock_put(other);
1492
1493                 err = 0;
1494                 unix_state_lock(sk);
1495                 if (unix_peer(sk) == other) {
1496                         unix_peer(sk) = NULL;
1497                         unix_state_unlock(sk);
1498
1499                         unix_dgram_disconnected(sk, other);
1500                         sock_put(other);
1501                         err = -ECONNREFUSED;
1502                 } else {
1503                         unix_state_unlock(sk);
1504                 }
1505
1506                 other = NULL;
1507                 if (err)
1508                         goto out_free;
1509                 goto restart;
1510         }
1511
1512         err = -EPIPE;
1513         if (other->sk_shutdown & RCV_SHUTDOWN)
1514                 goto out_unlock;
1515
1516         if (sk->sk_type != SOCK_SEQPACKET) {
1517                 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1518                 if (err)
1519                         goto out_unlock;
1520         }
1521
1522         if (unix_peer(other) != sk && unix_recvq_full(other)) {
1523                 if (!timeo) {
1524                         err = -EAGAIN;
1525                         goto out_unlock;
1526                 }
1527
1528                 timeo = unix_wait_for_peer(other, timeo);
1529
1530                 err = sock_intr_errno(timeo);
1531                 if (signal_pending(current))
1532                         goto out_free;
1533
1534                 goto restart;
1535         }
1536
1537         if (sock_flag(other, SOCK_RCVTSTAMP))
1538                 __net_timestamp(skb);
1539         skb_queue_tail(&other->sk_receive_queue, skb);
1540         if (max_level > unix_sk(other)->recursion_level)
1541                 unix_sk(other)->recursion_level = max_level;
1542         unix_state_unlock(other);
1543         other->sk_data_ready(other, len);
1544         sock_put(other);
1545         scm_destroy(siocb->scm);
1546         return len;
1547
1548 out_unlock:
1549         unix_state_unlock(other);
1550 out_free:
1551         kfree_skb(skb);
1552 out:
1553         if (other)
1554                 sock_put(other);
1555         scm_destroy(siocb->scm);
1556         return err;
1557 }
1558
1559
1560 static int unix_stream_sendmsg(struct kiocb *kiocb, struct socket *sock,
1561                                struct msghdr *msg, size_t len)
1562 {
1563         struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1564         struct sock *sk = sock->sk;
1565         struct sock *other = NULL;
1566         struct sockaddr_un *sunaddr = msg->msg_name;
1567         int err, size;
1568         struct sk_buff *skb;
1569         int sent = 0;
1570         struct scm_cookie tmp_scm;
1571         bool fds_sent = false;
1572         int max_level;
1573
1574         if (NULL == siocb->scm)
1575                 siocb->scm = &tmp_scm;
1576         wait_for_unix_gc();
1577         err = scm_send(sock, msg, siocb->scm);
1578         if (err < 0)
1579                 return err;
1580
1581         err = -EOPNOTSUPP;
1582         if (msg->msg_flags&MSG_OOB)
1583                 goto out_err;
1584
1585         if (msg->msg_namelen) {
1586                 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
1587                 goto out_err;
1588         } else {
1589                 sunaddr = NULL;
1590                 err = -ENOTCONN;
1591                 other = unix_peer(sk);
1592                 if (!other)
1593                         goto out_err;
1594         }
1595
1596         if (sk->sk_shutdown & SEND_SHUTDOWN)
1597                 goto pipe_err;
1598
1599         while (sent < len) {
1600                 /*
1601                  *      Optimisation for the fact that under 0.01% of X
1602                  *      messages typically need breaking up.
1603                  */
1604
1605                 size = len-sent;
1606
1607                 /* Keep two messages in the pipe so it schedules better */
1608                 if (size > ((sk->sk_sndbuf >> 1) - 64))
1609                         size = (sk->sk_sndbuf >> 1) - 64;
1610
1611                 if (size > SKB_MAX_ALLOC)
1612                         size = SKB_MAX_ALLOC;
1613
1614                 /*
1615                  *      Grab a buffer
1616                  */
1617
1618                 skb = sock_alloc_send_skb(sk, size, msg->msg_flags&MSG_DONTWAIT,
1619                                           &err);
1620
1621                 if (skb == NULL)
1622                         goto out_err;
1623
1624                 /*
1625                  *      If you pass two values to the sock_alloc_send_skb
1626                  *      it tries to grab the large buffer with GFP_NOFS
1627                  *      (which can fail easily), and if it fails grab the
1628                  *      fallback size buffer which is under a page and will
1629                  *      succeed. [Alan]
1630                  */
1631                 size = min_t(int, size, skb_tailroom(skb));
1632
1633
1634                 /* Only send the fds in the first buffer */
1635                 err = unix_scm_to_skb(siocb->scm, skb, !fds_sent);
1636                 if (err < 0) {
1637                         kfree_skb(skb);
1638                         goto out_err;
1639                 }
1640                 max_level = err + 1;
1641                 fds_sent = true;
1642
1643                 err = memcpy_fromiovec(skb_put(skb, size), msg->msg_iov, size);
1644                 if (err) {
1645                         kfree_skb(skb);
1646                         goto out_err;
1647                 }
1648
1649                 unix_state_lock(other);
1650
1651                 if (sock_flag(other, SOCK_DEAD) ||
1652                     (other->sk_shutdown & RCV_SHUTDOWN))
1653                         goto pipe_err_free;
1654
1655                 skb_queue_tail(&other->sk_receive_queue, skb);
1656                 if (max_level > unix_sk(other)->recursion_level)
1657                         unix_sk(other)->recursion_level = max_level;
1658                 unix_state_unlock(other);
1659                 other->sk_data_ready(other, size);
1660                 sent += size;
1661         }
1662
1663         scm_destroy(siocb->scm);
1664         siocb->scm = NULL;
1665
1666         return sent;
1667
1668 pipe_err_free:
1669         unix_state_unlock(other);
1670         kfree_skb(skb);
1671 pipe_err:
1672         if (sent == 0 && !(msg->msg_flags&MSG_NOSIGNAL))
1673                 send_sig(SIGPIPE, current, 0);
1674         err = -EPIPE;
1675 out_err:
1676         scm_destroy(siocb->scm);
1677         siocb->scm = NULL;
1678         return sent ? : err;
1679 }
1680
1681 static int unix_seqpacket_sendmsg(struct kiocb *kiocb, struct socket *sock,
1682                                   struct msghdr *msg, size_t len)
1683 {
1684         int err;
1685         struct sock *sk = sock->sk;
1686
1687         err = sock_error(sk);
1688         if (err)
1689                 return err;
1690
1691         if (sk->sk_state != TCP_ESTABLISHED)
1692                 return -ENOTCONN;
1693
1694         if (msg->msg_namelen)
1695                 msg->msg_namelen = 0;
1696
1697         return unix_dgram_sendmsg(kiocb, sock, msg, len);
1698 }
1699
1700 static int unix_seqpacket_recvmsg(struct kiocb *iocb, struct socket *sock,
1701                               struct msghdr *msg, size_t size,
1702                               int flags)
1703 {
1704         struct sock *sk = sock->sk;
1705
1706         if (sk->sk_state != TCP_ESTABLISHED)
1707                 return -ENOTCONN;
1708
1709         return unix_dgram_recvmsg(iocb, sock, msg, size, flags);
1710 }
1711
1712 static void unix_copy_addr(struct msghdr *msg, struct sock *sk)
1713 {
1714         struct unix_sock *u = unix_sk(sk);
1715
1716         msg->msg_namelen = 0;
1717         if (u->addr) {
1718                 msg->msg_namelen = u->addr->len;
1719                 memcpy(msg->msg_name, u->addr->name, u->addr->len);
1720         }
1721 }
1722
1723 static int unix_dgram_recvmsg(struct kiocb *iocb, struct socket *sock,
1724                               struct msghdr *msg, size_t size,
1725                               int flags)
1726 {
1727         struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1728         struct scm_cookie tmp_scm;
1729         struct sock *sk = sock->sk;
1730         struct unix_sock *u = unix_sk(sk);
1731         int noblock = flags & MSG_DONTWAIT;
1732         struct sk_buff *skb;
1733         int err;
1734
1735         err = -EOPNOTSUPP;
1736         if (flags&MSG_OOB)
1737                 goto out;
1738
1739         msg->msg_namelen = 0;
1740
1741         err = mutex_lock_interruptible(&u->readlock);
1742         if (err) {
1743                 err = sock_intr_errno(sock_rcvtimeo(sk, noblock));
1744                 goto out;
1745         }
1746
1747         skb = skb_recv_datagram(sk, flags, noblock, &err);
1748         if (!skb) {
1749                 unix_state_lock(sk);
1750                 /* Signal EOF on disconnected non-blocking SEQPACKET socket. */
1751                 if (sk->sk_type == SOCK_SEQPACKET && err == -EAGAIN &&
1752                     (sk->sk_shutdown & RCV_SHUTDOWN))
1753                         err = 0;
1754                 unix_state_unlock(sk);
1755                 goto out_unlock;
1756         }
1757
1758         wake_up_interruptible_sync_poll(&u->peer_wait,
1759                                         POLLOUT | POLLWRNORM | POLLWRBAND);
1760
1761         if (msg->msg_name)
1762                 unix_copy_addr(msg, skb->sk);
1763
1764         if (size > skb->len)
1765                 size = skb->len;
1766         else if (size < skb->len)
1767                 msg->msg_flags |= MSG_TRUNC;
1768
1769         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, size);
1770         if (err)
1771                 goto out_free;
1772
1773         if (sock_flag(sk, SOCK_RCVTSTAMP))
1774                 __sock_recv_timestamp(msg, sk, skb);
1775
1776         if (!siocb->scm) {
1777                 siocb->scm = &tmp_scm;
1778                 memset(&tmp_scm, 0, sizeof(tmp_scm));
1779         }
1780         scm_set_cred(siocb->scm, UNIXCB(skb).pid, UNIXCB(skb).cred);
1781         unix_set_secdata(siocb->scm, skb);
1782
1783         if (!(flags & MSG_PEEK)) {
1784                 if (UNIXCB(skb).fp)
1785                         unix_detach_fds(siocb->scm, skb);
1786         } else {
1787                 /* It is questionable: on PEEK we could:
1788                    - do not return fds - good, but too simple 8)
1789                    - return fds, and do not return them on read (old strategy,
1790                      apparently wrong)
1791                    - clone fds (I chose it for now, it is the most universal
1792                      solution)
1793
1794                    POSIX 1003.1g does not actually define this clearly
1795                    at all. POSIX 1003.1g doesn't define a lot of things
1796                    clearly however!
1797
1798                 */
1799                 if (UNIXCB(skb).fp)
1800                         siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1801         }
1802         err = size;
1803
1804         scm_recv(sock, msg, siocb->scm, flags);
1805
1806 out_free:
1807         skb_free_datagram(sk, skb);
1808 out_unlock:
1809         mutex_unlock(&u->readlock);
1810 out:
1811         return err;
1812 }
1813
1814 /*
1815  *      Sleep until data has arrive. But check for races..
1816  */
1817
1818 static long unix_stream_data_wait(struct sock *sk, long timeo)
1819 {
1820         DEFINE_WAIT(wait);
1821
1822         unix_state_lock(sk);
1823
1824         for (;;) {
1825                 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1826
1827                 if (!skb_queue_empty(&sk->sk_receive_queue) ||
1828                     sk->sk_err ||
1829                     (sk->sk_shutdown & RCV_SHUTDOWN) ||
1830                     signal_pending(current) ||
1831                     !timeo)
1832                         break;
1833
1834                 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1835                 unix_state_unlock(sk);
1836                 timeo = schedule_timeout(timeo);
1837                 unix_state_lock(sk);
1838                 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1839         }
1840
1841         finish_wait(sk_sleep(sk), &wait);
1842         unix_state_unlock(sk);
1843         return timeo;
1844 }
1845
1846
1847
1848 static int unix_stream_recvmsg(struct kiocb *iocb, struct socket *sock,
1849                                struct msghdr *msg, size_t size,
1850                                int flags)
1851 {
1852         struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1853         struct scm_cookie tmp_scm;
1854         struct sock *sk = sock->sk;
1855         struct unix_sock *u = unix_sk(sk);
1856         struct sockaddr_un *sunaddr = msg->msg_name;
1857         int copied = 0;
1858         int check_creds = 0;
1859         int target;
1860         int err = 0;
1861         long timeo;
1862
1863         err = -EINVAL;
1864         if (sk->sk_state != TCP_ESTABLISHED)
1865                 goto out;
1866
1867         err = -EOPNOTSUPP;
1868         if (flags&MSG_OOB)
1869                 goto out;
1870
1871         target = sock_rcvlowat(sk, flags&MSG_WAITALL, size);
1872         timeo = sock_rcvtimeo(sk, flags&MSG_DONTWAIT);
1873
1874         msg->msg_namelen = 0;
1875
1876         /* Lock the socket to prevent queue disordering
1877          * while sleeps in memcpy_tomsg
1878          */
1879
1880         if (!siocb->scm) {
1881                 siocb->scm = &tmp_scm;
1882                 memset(&tmp_scm, 0, sizeof(tmp_scm));
1883         }
1884
1885         err = mutex_lock_interruptible(&u->readlock);
1886         if (err) {
1887                 err = sock_intr_errno(timeo);
1888                 goto out;
1889         }
1890
1891         do {
1892                 int chunk;
1893                 struct sk_buff *skb;
1894
1895                 unix_state_lock(sk);
1896                 skb = skb_dequeue(&sk->sk_receive_queue);
1897                 if (skb == NULL) {
1898                         unix_sk(sk)->recursion_level = 0;
1899                         if (copied >= target)
1900                                 goto unlock;
1901
1902                         /*
1903                          *      POSIX 1003.1g mandates this order.
1904                          */
1905
1906                         err = sock_error(sk);
1907                         if (err)
1908                                 goto unlock;
1909                         if (sk->sk_shutdown & RCV_SHUTDOWN)
1910                                 goto unlock;
1911
1912                         unix_state_unlock(sk);
1913                         err = -EAGAIN;
1914                         if (!timeo)
1915                                 break;
1916                         mutex_unlock(&u->readlock);
1917
1918                         timeo = unix_stream_data_wait(sk, timeo);
1919
1920                         if (signal_pending(current)
1921                             ||  mutex_lock_interruptible(&u->readlock)) {
1922                                 err = sock_intr_errno(timeo);
1923                                 goto out;
1924                         }
1925
1926                         continue;
1927  unlock:
1928                         unix_state_unlock(sk);
1929                         break;
1930                 }
1931                 unix_state_unlock(sk);
1932
1933                 if (check_creds) {
1934                         /* Never glue messages from different writers */
1935                         if ((UNIXCB(skb).pid  != siocb->scm->pid) ||
1936                             (UNIXCB(skb).cred != siocb->scm->cred)) {
1937                                 skb_queue_head(&sk->sk_receive_queue, skb);
1938                                 break;
1939                         }
1940                 } else {
1941                         /* Copy credentials */
1942                         scm_set_cred(siocb->scm, UNIXCB(skb).pid, UNIXCB(skb).cred);
1943                         check_creds = 1;
1944                 }
1945
1946                 /* Copy address just once */
1947                 if (sunaddr) {
1948                         unix_copy_addr(msg, skb->sk);
1949                         sunaddr = NULL;
1950                 }
1951
1952                 chunk = min_t(unsigned int, skb->len, size);
1953                 if (memcpy_toiovec(msg->msg_iov, skb->data, chunk)) {
1954                         skb_queue_head(&sk->sk_receive_queue, skb);
1955                         if (copied == 0)
1956                                 copied = -EFAULT;
1957                         break;
1958                 }
1959                 copied += chunk;
1960                 size -= chunk;
1961
1962                 /* Mark read part of skb as used */
1963                 if (!(flags & MSG_PEEK)) {
1964                         skb_pull(skb, chunk);
1965
1966                         if (UNIXCB(skb).fp)
1967                                 unix_detach_fds(siocb->scm, skb);
1968
1969                         /* put the skb back if we didn't use it up.. */
1970                         if (skb->len) {
1971                                 skb_queue_head(&sk->sk_receive_queue, skb);
1972                                 break;
1973                         }
1974
1975                         consume_skb(skb);
1976
1977                         if (siocb->scm->fp)
1978                                 break;
1979                 } else {
1980                         /* It is questionable, see note in unix_dgram_recvmsg.
1981                          */
1982                         if (UNIXCB(skb).fp)
1983                                 siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1984
1985                         /* put message back and return */
1986                         skb_queue_head(&sk->sk_receive_queue, skb);
1987                         break;
1988                 }
1989         } while (size);
1990
1991         mutex_unlock(&u->readlock);
1992         scm_recv(sock, msg, siocb->scm, flags);
1993 out:
1994         return copied ? : err;
1995 }
1996
1997 static int unix_shutdown(struct socket *sock, int mode)
1998 {
1999         struct sock *sk = sock->sk;
2000         struct sock *other;
2001
2002         mode = (mode+1)&(RCV_SHUTDOWN|SEND_SHUTDOWN);
2003
2004         if (mode) {
2005                 unix_state_lock(sk);
2006                 sk->sk_shutdown |= mode;
2007                 other = unix_peer(sk);
2008                 if (other)
2009                         sock_hold(other);
2010                 unix_state_unlock(sk);
2011                 sk->sk_state_change(sk);
2012
2013                 if (other &&
2014                         (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET)) {
2015
2016                         int peer_mode = 0;
2017
2018                         if (mode&RCV_SHUTDOWN)
2019                                 peer_mode |= SEND_SHUTDOWN;
2020                         if (mode&SEND_SHUTDOWN)
2021                                 peer_mode |= RCV_SHUTDOWN;
2022                         unix_state_lock(other);
2023                         other->sk_shutdown |= peer_mode;
2024                         unix_state_unlock(other);
2025                         other->sk_state_change(other);
2026                         if (peer_mode == SHUTDOWN_MASK)
2027                                 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_HUP);
2028                         else if (peer_mode & RCV_SHUTDOWN)
2029                                 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_IN);
2030                 }
2031                 if (other)
2032                         sock_put(other);
2033         }
2034         return 0;
2035 }
2036
2037 static int unix_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2038 {
2039         struct sock *sk = sock->sk;
2040         long amount = 0;
2041         int err;
2042
2043         switch (cmd) {
2044         case SIOCOUTQ:
2045                 amount = sk_wmem_alloc_get(sk);
2046                 err = put_user(amount, (int __user *)arg);
2047                 break;
2048         case SIOCINQ:
2049                 {
2050                         struct sk_buff *skb;
2051
2052                         if (sk->sk_state == TCP_LISTEN) {
2053                                 err = -EINVAL;
2054                                 break;
2055                         }
2056
2057                         spin_lock(&sk->sk_receive_queue.lock);
2058                         if (sk->sk_type == SOCK_STREAM ||
2059                             sk->sk_type == SOCK_SEQPACKET) {
2060                                 skb_queue_walk(&sk->sk_receive_queue, skb)
2061                                         amount += skb->len;
2062                         } else {
2063                                 skb = skb_peek(&sk->sk_receive_queue);
2064                                 if (skb)
2065                                         amount = skb->len;
2066                         }
2067                         spin_unlock(&sk->sk_receive_queue.lock);
2068                         err = put_user(amount, (int __user *)arg);
2069                         break;
2070                 }
2071
2072         default:
2073                 err = -ENOIOCTLCMD;
2074                 break;
2075         }
2076         return err;
2077 }
2078
2079 static unsigned int unix_poll(struct file *file, struct socket *sock, poll_table *wait)
2080 {
2081         struct sock *sk = sock->sk;
2082         unsigned int mask;
2083
2084         sock_poll_wait(file, sk_sleep(sk), wait);
2085         mask = 0;
2086
2087         /* exceptional events? */
2088         if (sk->sk_err)
2089                 mask |= POLLERR;
2090         if (sk->sk_shutdown == SHUTDOWN_MASK)
2091                 mask |= POLLHUP;
2092         if (sk->sk_shutdown & RCV_SHUTDOWN)
2093                 mask |= POLLRDHUP | POLLIN | POLLRDNORM;
2094
2095         /* readable? */
2096         if (!skb_queue_empty(&sk->sk_receive_queue))
2097                 mask |= POLLIN | POLLRDNORM;
2098
2099         /* Connection-based need to check for termination and startup */
2100         if ((sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) &&
2101             sk->sk_state == TCP_CLOSE)
2102                 mask |= POLLHUP;
2103
2104         /*
2105          * we set writable also when the other side has shut down the
2106          * connection. This prevents stuck sockets.
2107          */
2108         if (unix_writable(sk))
2109                 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2110
2111         return mask;
2112 }
2113
2114 static unsigned int unix_dgram_poll(struct file *file, struct socket *sock,
2115                                     poll_table *wait)
2116 {
2117         struct sock *sk = sock->sk, *other;
2118         unsigned int mask, writable;
2119
2120         sock_poll_wait(file, sk_sleep(sk), wait);
2121         mask = 0;
2122
2123         /* exceptional events? */
2124         if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue))
2125                 mask |= POLLERR;
2126         if (sk->sk_shutdown & RCV_SHUTDOWN)
2127                 mask |= POLLRDHUP | POLLIN | POLLRDNORM;
2128         if (sk->sk_shutdown == SHUTDOWN_MASK)
2129                 mask |= POLLHUP;
2130
2131         /* readable? */
2132         if (!skb_queue_empty(&sk->sk_receive_queue))
2133                 mask |= POLLIN | POLLRDNORM;
2134
2135         /* Connection-based need to check for termination and startup */
2136         if (sk->sk_type == SOCK_SEQPACKET) {
2137                 if (sk->sk_state == TCP_CLOSE)
2138                         mask |= POLLHUP;
2139                 /* connection hasn't started yet? */
2140                 if (sk->sk_state == TCP_SYN_SENT)
2141                         return mask;
2142         }
2143
2144         /* No write status requested, avoid expensive OUT tests. */
2145         if (wait && !(wait->key & (POLLWRBAND | POLLWRNORM | POLLOUT)))
2146                 return mask;
2147
2148         writable = unix_writable(sk);
2149         other = unix_peer_get(sk);
2150         if (other) {
2151                 if (unix_peer(other) != sk) {
2152                         sock_poll_wait(file, &unix_sk(other)->peer_wait, wait);
2153                         if (unix_recvq_full(other))
2154                                 writable = 0;
2155                 }
2156                 sock_put(other);
2157         }
2158
2159         if (writable)
2160                 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2161         else
2162                 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
2163
2164         return mask;
2165 }
2166
2167 #ifdef CONFIG_PROC_FS
2168 static struct sock *first_unix_socket(int *i)
2169 {
2170         for (*i = 0; *i <= UNIX_HASH_SIZE; (*i)++) {
2171                 if (!hlist_empty(&unix_socket_table[*i]))
2172                         return __sk_head(&unix_socket_table[*i]);
2173         }
2174         return NULL;
2175 }
2176
2177 static struct sock *next_unix_socket(int *i, struct sock *s)
2178 {
2179         struct sock *next = sk_next(s);
2180         /* More in this chain? */
2181         if (next)
2182                 return next;
2183         /* Look for next non-empty chain. */
2184         for ((*i)++; *i <= UNIX_HASH_SIZE; (*i)++) {
2185                 if (!hlist_empty(&unix_socket_table[*i]))
2186                         return __sk_head(&unix_socket_table[*i]);
2187         }
2188         return NULL;
2189 }
2190
2191 struct unix_iter_state {
2192         struct seq_net_private p;
2193         int i;
2194 };
2195
2196 static struct sock *unix_seq_idx(struct seq_file *seq, loff_t pos)
2197 {
2198         struct unix_iter_state *iter = seq->private;
2199         loff_t off = 0;
2200         struct sock *s;
2201
2202         for (s = first_unix_socket(&iter->i); s; s = next_unix_socket(&iter->i, s)) {
2203                 if (sock_net(s) != seq_file_net(seq))
2204                         continue;
2205                 if (off == pos)
2206                         return s;
2207                 ++off;
2208         }
2209         return NULL;
2210 }
2211
2212 static void *unix_seq_start(struct seq_file *seq, loff_t *pos)
2213         __acquires(unix_table_lock)
2214 {
2215         spin_lock(&unix_table_lock);
2216         return *pos ? unix_seq_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2217 }
2218
2219 static void *unix_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2220 {
2221         struct unix_iter_state *iter = seq->private;
2222         struct sock *sk = v;
2223         ++*pos;
2224
2225         if (v == SEQ_START_TOKEN)
2226                 sk = first_unix_socket(&iter->i);
2227         else
2228                 sk = next_unix_socket(&iter->i, sk);
2229         while (sk && (sock_net(sk) != seq_file_net(seq)))
2230                 sk = next_unix_socket(&iter->i, sk);
2231         return sk;
2232 }
2233
2234 static void unix_seq_stop(struct seq_file *seq, void *v)
2235         __releases(unix_table_lock)
2236 {
2237         spin_unlock(&unix_table_lock);
2238 }
2239
2240 static int unix_seq_show(struct seq_file *seq, void *v)
2241 {
2242
2243         if (v == SEQ_START_TOKEN)
2244                 seq_puts(seq, "Num       RefCount Protocol Flags    Type St "
2245                          "Inode Path\n");
2246         else {
2247                 struct sock *s = v;
2248                 struct unix_sock *u = unix_sk(s);
2249                 unix_state_lock(s);
2250
2251                 seq_printf(seq, "%pK: %08X %08X %08X %04X %02X %5lu",
2252                         s,
2253                         atomic_read(&s->sk_refcnt),
2254                         0,
2255                         s->sk_state == TCP_LISTEN ? __SO_ACCEPTCON : 0,
2256                         s->sk_type,
2257                         s->sk_socket ?
2258                         (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTED : SS_UNCONNECTED) :
2259                         (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTING : SS_DISCONNECTING),
2260                         sock_i_ino(s));
2261
2262                 if (u->addr) {
2263                         int i, len;
2264                         seq_putc(seq, ' ');
2265
2266                         i = 0;
2267                         len = u->addr->len - sizeof(short);
2268                         if (!UNIX_ABSTRACT(s))
2269                                 len--;
2270                         else {
2271                                 seq_putc(seq, '@');
2272                                 i++;
2273                         }
2274                         for ( ; i < len; i++)
2275                                 seq_putc(seq, u->addr->name->sun_path[i]);
2276                 }
2277                 unix_state_unlock(s);
2278                 seq_putc(seq, '\n');
2279         }
2280
2281         return 0;
2282 }
2283
2284 static const struct seq_operations unix_seq_ops = {
2285         .start  = unix_seq_start,
2286         .next   = unix_seq_next,
2287         .stop   = unix_seq_stop,
2288         .show   = unix_seq_show,
2289 };
2290
2291 static int unix_seq_open(struct inode *inode, struct file *file)
2292 {
2293         return seq_open_net(inode, file, &unix_seq_ops,
2294                             sizeof(struct unix_iter_state));
2295 }
2296
2297 static const struct file_operations unix_seq_fops = {
2298         .owner          = THIS_MODULE,
2299         .open           = unix_seq_open,
2300         .read           = seq_read,
2301         .llseek         = seq_lseek,
2302         .release        = seq_release_net,
2303 };
2304
2305 #endif
2306
2307 static const struct net_proto_family unix_family_ops = {
2308         .family = PF_UNIX,
2309         .create = unix_create,
2310         .owner  = THIS_MODULE,
2311 };
2312
2313
2314 static int __net_init unix_net_init(struct net *net)
2315 {
2316         int error = -ENOMEM;
2317
2318         net->unx.sysctl_max_dgram_qlen = 10;
2319         if (unix_sysctl_register(net))
2320                 goto out;
2321
2322 #ifdef CONFIG_PROC_FS
2323         if (!proc_net_fops_create(net, "unix", 0, &unix_seq_fops)) {
2324                 unix_sysctl_unregister(net);
2325                 goto out;
2326         }
2327 #endif
2328         error = 0;
2329 out:
2330         return error;
2331 }
2332
2333 static void __net_exit unix_net_exit(struct net *net)
2334 {
2335         unix_sysctl_unregister(net);
2336         proc_net_remove(net, "unix");
2337 }
2338
2339 static struct pernet_operations unix_net_ops = {
2340         .init = unix_net_init,
2341         .exit = unix_net_exit,
2342 };
2343
2344 static int __init af_unix_init(void)
2345 {
2346         int rc = -1;
2347         struct sk_buff *dummy_skb;
2348
2349         BUILD_BUG_ON(sizeof(struct unix_skb_parms) > sizeof(dummy_skb->cb));
2350
2351         rc = proto_register(&unix_proto, 1);
2352         if (rc != 0) {
2353                 printk(KERN_CRIT "%s: Cannot create unix_sock SLAB cache!\n",
2354                        __func__);
2355                 goto out;
2356         }
2357
2358         sock_register(&unix_family_ops);
2359         register_pernet_subsys(&unix_net_ops);
2360 out:
2361         return rc;
2362 }
2363
2364 static void __exit af_unix_exit(void)
2365 {
2366         sock_unregister(PF_UNIX);
2367         proto_unregister(&unix_proto);
2368         unregister_pernet_subsys(&unix_net_ops);
2369 }
2370
2371 /* Earlier than device_initcall() so that other drivers invoking
2372    request_module() don't end up in a loop when modprobe tries
2373    to use a UNIX socket. But later than subsys_initcall() because
2374    we depend on stuff initialised there */
2375 fs_initcall(af_unix_init);
2376 module_exit(af_unix_exit);
2377
2378 MODULE_LICENSE("GPL");
2379 MODULE_ALIAS_NETPROTO(PF_UNIX);