net/unix/af_unix.c

   1 /*
   2  * NET4:        Implementation of BSD Unix domain sockets.
   3  *
   4  * Authors:     Alan Cox, <alan@lxorguk.ukuu.org.uk>
   5  *
   6  *              This program is free software; you can redistribute it and/or
   7  *              modify it under the terms of the GNU General Public License
   8  *              as published by the Free Software Foundation; either version
   9  *              2 of the License, or (at your option) any later version.
  10  *
  11  * Fixes:
  12  *              Linus Torvalds  :       Assorted bug cures.
  13  *              Niibe Yutaka    :       async I/O support.
  14  *              Carsten Paeth   :       PF_UNIX check, address fixes.
  15  *              Alan Cox        :       Limit size of allocated blocks.
  16  *              Alan Cox        :       Fixed the stupid socketpair bug.
  17  *              Alan Cox        :       BSD compatibility fine tuning.
  18  *              Alan Cox        :       Fixed a bug in connect when interrupted.
  19  *              Alan Cox        :       Sorted out a proper draft version of
  20  *                                      file descriptor passing hacked up from
  21  *                                      Mike Shaver's work.
  22  *              Marty Leisner   :       Fixes to fd passing
  23  *              Nick Nevin      :       recvmsg bugfix.
  24  *              Alan Cox        :       Started proper garbage collector
  25  *              Heiko EiBfeldt  :       Missing verify_area check
  26  *              Alan Cox        :       Started POSIXisms
  27  *              Andreas Schwab  :       Replace inode by dentry for proper
  28  *                                      reference counting
  29  *              Kirk Petersen   :       Made this a module
  30  *          Christoph Rohland   :       Elegant non-blocking accept/connect algorithm.
  31  *                                      Lots of bug fixes.
  32  *           Alexey Kuznetosv   :       Repaired (I hope) bugs introduces
  33  *                                      by above two patches.
  34  *           Andrea Arcangeli   :       If possible we block in connect(2)
  35  *                                      if the max backlog of the listen socket
  36  *                                      is been reached. This won't break
  37  *                                      old apps and it will avoid huge amount
  38  *                                      of socks hashed (this for unix_gc()
  39  *                                      performances reasons).
  40  *                                      Security fix that limits the max
  41  *                                      number of socks to 2*max_files and
  42  *                                      the number of skb queueable in the
  43  *                                      dgram receiver.
  44  *              Artur Skawina   :       Hash function optimizations
  45  *           Alexey Kuznetsov   :       Full scale SMP. Lot of bugs are introduced 8)
  46  *            Malcolm Beattie   :       Set peercred for socketpair
  47  *           Michal Ostrowski   :       Module initialization cleanup.
  48  *           Arnaldo C. Melo    :       Remove MOD_{INC,DEC}_USE_COUNT,
  49  *                                      the core infrastructure is doing that
  50  *                                      for all net proto families now (2.5.69+)
  51  *
  52  *
  53  * Known differences from reference BSD that was tested:
  54  *
  55  *      [TO FIX]
  56  *      ECONNREFUSED is not returned from one end of a connected() socket to the
  57  *              other the moment one end closes.
  58  *      fstat() doesn't return st_dev=0, and give the blksize as high water mark
  59  *              and a fake inode identifier (nor the BSD first socket fstat twice bug).
  60  *      [NOT TO FIX]
  61  *      accept() returns a path name even if the connecting socket has closed
  62  *              in the meantime (BSD loses the path and gives up).
  63  *      accept() returns 0 length path for an unbound connector. BSD returns 16
  64  *              and a null first byte in the path (but not for gethost/peername - BSD bug ??)
  65  *      socketpair(...SOCK_RAW..) doesn't panic the kernel.
  66  *      BSD af_unix apparently has connect forgetting to block properly.
  67  *              (need to check this with the POSIX spec in detail)
  68  *
  69  * Differences from 2.0.0-11-... (ANK)
  70  *      Bug fixes and improvements.
  71  *              - client shutdown killed server socket.
  72  *              - removed all useless cli/sti pairs.
  73  *
  74  *      Semantic changes/extensions.
  75  *              - generic control message passing.
  76  *              - SCM_CREDENTIALS control message.
  77  *              - "Abstract" (not FS based) socket bindings.
  78  *                Abstract names are sequences of bytes (not zero terminated)
  79  *                started by 0, so that this name space does not intersect
  80  *                with BSD names.
  81  */
  82
  83 #include <linux/module.h>
  84 #include <linux/kernel.h>
  85 #include <linux/signal.h>
  86 #include <linux/sched.h>
  87 #include <linux/errno.h>
  88 #include <linux/string.h>
  89 #include <linux/stat.h>
  90 #include <linux/dcache.h>
  91 #include <linux/namei.h>
  92 #include <linux/socket.h>
  93 #include <linux/un.h>
  94 #include <linux/fcntl.h>
  95 #include <linux/termios.h>
  96 #include <linux/sockios.h>
  97 #include <linux/net.h>
  98 #include <linux/in.h>
  99 #include <linux/fs.h>
 100 #include <linux/slab.h>
 101 #include <asm/uaccess.h>
 102 #include <linux/skbuff.h>
 103 #include <linux/netdevice.h>
 104 #include <net/net_namespace.h>
 105 #include <net/sock.h>
 106 #include <net/tcp_states.h>
 107 #include <net/af_unix.h>
 108 #include <linux/proc_fs.h>
 109 #include <linux/seq_file.h>
 110 #include <net/scm.h>
 111 #include <linux/init.h>
 112 #include <linux/poll.h>
 113 #include <linux/rtnetlink.h>
 114 #include <linux/mount.h>
 115 #include <net/checksum.h>
 116 #include <linux/security.h>
 117
 118 static struct hlist_head unix_socket_table[UNIX_HASH_SIZE + 1];
 119 static DEFINE_SPINLOCK(unix_table_lock);
 120 static atomic_t unix_nr_socks = ATOMIC_INIT(0);
 121
 122 #define unix_sockets_unbound    (&unix_socket_table[UNIX_HASH_SIZE])
 123
 124 #define UNIX_ABSTRACT(sk)       (unix_sk(sk)->addr->hash != UNIX_HASH_SIZE)
 125
 126 #ifdef CONFIG_SECURITY_NETWORK
 127 static void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb)
 128 {
 129         memcpy(UNIXSID(skb), &scm->secid, sizeof(u32));
 130 }
 131
 132 static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb)
 133 {
 134         scm->secid = *UNIXSID(skb);
 135 }
 136 #else
 137 static inline void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb)
 138 { }
 139
 140 static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb)
 141 { }
 142 #endif /* CONFIG_SECURITY_NETWORK */
 143
 144 /*
 145  *  SMP locking strategy:
 146  *    hash table is protected with spinlock unix_table_lock
 147  *    each socket state is protected by separate rwlock.
 148  */
 149
 150 static inline unsigned unix_hash_fold(__wsum n)
 151 {
 152         unsigned hash = (__force unsigned)n;
 153         hash ^= hash>>16;
 154         hash ^= hash>>8;
 155         return hash&(UNIX_HASH_SIZE-1);
 156 }
 157
 158 #define unix_peer(sk) (unix_sk(sk)->peer)
 159
 160 static inline int unix_our_peer(struct sock *sk, struct sock *osk)
 161 {
 162         return unix_peer(osk) == sk;
 163 }
 164
 165 static inline int unix_may_send(struct sock *sk, struct sock *osk)
 166 {
 167         return unix_peer(osk) == NULL || unix_our_peer(sk, osk);
 168 }
 169
 170 static inline int unix_recvq_full(struct sock const *sk)
 171 {
 172         return skb_queue_len(&sk->sk_receive_queue) > sk->sk_max_ack_backlog;
 173 }
 174
 175 static struct sock *unix_peer_get(struct sock *s)
 176 {
 177         struct sock *peer;
 178
 179         unix_state_lock(s);
 180         peer = unix_peer(s);
 181         if (peer)
 182                 sock_hold(peer);
 183         unix_state_unlock(s);
 184         return peer;
 185 }
 186
 187 static inline void unix_release_addr(struct unix_address *addr)
 188 {
 189         if (atomic_dec_and_test(&addr->refcnt))
 190                 kfree(addr);
 191 }
 192
 193 /*
 194  *      Check unix socket name:
 195  *              - should be not zero length.
 196  *              - if started by not zero, should be NULL terminated (FS object)
 197  *              - if started by zero, it is abstract name.
 198  */
 199
 200 static int unix_mkname(struct sockaddr_un *sunaddr, int len, unsigned *hashp)
 201 {
 202         if (len <= sizeof(short) || len > sizeof(*sunaddr))
 203                 return -EINVAL;
 204         if (!sunaddr || sunaddr->sun_family != AF_UNIX)
 205                 return -EINVAL;
 206         if (sunaddr->sun_path[0]) {
 207                 /*
 208                  * This may look like an off by one error but it is a bit more
 209                  * subtle. 108 is the longest valid AF_UNIX path for a binding.
 210                  * sun_path[108] doesnt as such exist.  However in kernel space
 211                  * we are guaranteed that it is a valid memory location in our
 212                  * kernel address buffer.
 213                  */
 214                 ((char *)sunaddr)[len] = 0;
 215                 len = strlen(sunaddr->sun_path)+1+sizeof(short);
 216                 return len;
 217         }
 218
 219         *hashp = unix_hash_fold(csum_partial(sunaddr, len, 0));
 220         return len;
 221 }
 222
 223 static void __unix_remove_socket(struct sock *sk)
 224 {
 225         sk_del_node_init(sk);
 226 }
 227
 228 static void __unix_insert_socket(struct hlist_head *list, struct sock *sk)
 229 {
 230         WARN_ON(!sk_unhashed(sk));
 231         sk_add_node(sk, list);
 232 }
 233
 234 static inline void unix_remove_socket(struct sock *sk)
 235 {
 236         spin_lock(&unix_table_lock);
 237         __unix_remove_socket(sk);
 238         spin_unlock(&unix_table_lock);
 239 }
 240
 241 static inline void unix_insert_socket(struct hlist_head *list, struct sock *sk)
 242 {
 243         spin_lock(&unix_table_lock);
 244         __unix_insert_socket(list, sk);
 245         spin_unlock(&unix_table_lock);
 246 }
 247
 248 static struct sock *__unix_find_socket_byname(struct net *net,
 249                                               struct sockaddr_un *sunname,
 250                                               int len, int type, unsigned hash)
 251 {
 252         struct sock *s;
 253         struct hlist_node *node;
 254
 255         sk_for_each(s, node, &unix_socket_table[hash ^ type]) {
 256                 struct unix_sock *u = unix_sk(s);
 257
 258                 if (!net_eq(sock_net(s), net))
 259                         continue;
 260
 261                 if (u->addr->len == len &&
 262                     !memcmp(u->addr->name, sunname, len))
 263                         goto found;
 264         }
 265         s = NULL;
 266 found:
 267         return s;
 268 }
 269
 270 static inline struct sock *unix_find_socket_byname(struct net *net,
 271                                                    struct sockaddr_un *sunname,
 272                                                    int len, int type,
 273                                                    unsigned hash)
 274 {
 275         struct sock *s;
 276
 277         spin_lock(&unix_table_lock);
 278         s = __unix_find_socket_byname(net, sunname, len, type, hash);
 279         if (s)
 280                 sock_hold(s);
 281         spin_unlock(&unix_table_lock);
 282         return s;
 283 }
 284
 285 static struct sock *unix_find_socket_byinode(struct net *net, struct inode *i)
 286 {
 287         struct sock *s;
 288         struct hlist_node *node;
 289
 290         spin_lock(&unix_table_lock);
 291         sk_for_each(s, node,
 292                     &unix_socket_table[i->i_ino & (UNIX_HASH_SIZE - 1)]) {
 293                 struct dentry *dentry = unix_sk(s)->dentry;
 294
 295                 if (!net_eq(sock_net(s), net))
 296                         continue;
 297
 298                 if (dentry && dentry->d_inode == i) {
 299                         sock_hold(s);
 300                         goto found;
 301                 }
 302         }
 303         s = NULL;
 304 found:
 305         spin_unlock(&unix_table_lock);
 306         return s;
 307 }
 308
 309 static inline int unix_writable(struct sock *sk)
 310 {
 311         return (atomic_read(&sk->sk_wmem_alloc) << 2) <= sk->sk_sndbuf;
 312 }
 313
 314 static void unix_write_space(struct sock *sk)
 315 {
 316         read_lock(&sk->sk_callback_lock);
 317         if (unix_writable(sk)) {
 318                 if (sk_has_sleeper(sk))
 319                         wake_up_interruptible_sync(sk->sk_sleep);
 320                 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
 321         }
 322         read_unlock(&sk->sk_callback_lock);
 323 }
 324
 325 /* When dgram socket disconnects (or changes its peer), we clear its receive
 326  * queue of packets arrived from previous peer. First, it allows to do
 327  * flow control based only on wmem_alloc; second, sk connected to peer
 328  * may receive messages only from that peer. */
 329 static void unix_dgram_disconnected(struct sock *sk, struct sock *other)
 330 {
 331         if (!skb_queue_empty(&sk->sk_receive_queue)) {
 332                 skb_queue_purge(&sk->sk_receive_queue);
 333                 wake_up_interruptible_all(&unix_sk(sk)->peer_wait);
 334
 335                 /* If one link of bidirectional dgram pipe is disconnected,
 336                  * we signal error. Messages are lost. Do not make this,
 337                  * when peer was not connected to us.
 338                  */
 339                 if (!sock_flag(other, SOCK_DEAD) && unix_peer(other) == sk) {
 340                         other->sk_err = ECONNRESET;
 341                         other->sk_error_report(other);
 342                 }
 343         }
 344 }
 345
 346 static void unix_sock_destructor(struct sock *sk)
 347 {
 348         struct unix_sock *u = unix_sk(sk);
 349
 350         skb_queue_purge(&sk->sk_receive_queue);
 351
 352         WARN_ON(atomic_read(&sk->sk_wmem_alloc));
 353         WARN_ON(!sk_unhashed(sk));
 354         WARN_ON(sk->sk_socket);
 355         if (!sock_flag(sk, SOCK_DEAD)) {
 356                 printk(KERN_INFO "Attempt to release alive unix socket: %p\n", sk);
 357                 return;
 358         }
 359
 360         if (u->addr)
 361                 unix_release_addr(u->addr);
 362
 363         atomic_dec(&unix_nr_socks);
 364         local_bh_disable();
 365         sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
 366         local_bh_enable();
 367 #ifdef UNIX_REFCNT_DEBUG
 368         printk(KERN_DEBUG "UNIX %p is destroyed, %d are still alive.\n", sk,
 369                 atomic_read(&unix_nr_socks));
 370 #endif
 371 }
 372
 373 static int unix_release_sock(struct sock *sk, int embrion)
 374 {
 375         struct unix_sock *u = unix_sk(sk);
 376         struct dentry *dentry;
 377         struct vfsmount *mnt;
 378         struct sock *skpair;
 379         struct sk_buff *skb;
 380         int state;
 381
 382         unix_remove_socket(sk);
 383
 384         /* Clear state */
 385         unix_state_lock(sk);
 386         sock_orphan(sk);
 387         sk->sk_shutdown = SHUTDOWN_MASK;
 388         dentry       = u->dentry;
 389         u->dentry    = NULL;
 390         mnt          = u->mnt;
 391         u->mnt       = NULL;
 392         state = sk->sk_state;
 393         sk->sk_state = TCP_CLOSE;
 394         unix_state_unlock(sk);
 395
 396         wake_up_interruptible_all(&u->peer_wait);
 397
 398         skpair = unix_peer(sk);
 399
 400         if (skpair != NULL) {
 401                 if (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) {
 402                         unix_state_lock(skpair);
 403                         /* No more writes */
 404                         skpair->sk_shutdown = SHUTDOWN_MASK;
 405                         if (!skb_queue_empty(&sk->sk_receive_queue) || embrion)
 406                                 skpair->sk_err = ECONNRESET;
 407                         unix_state_unlock(skpair);
 408                         skpair->sk_state_change(skpair);
 409                         read_lock(&skpair->sk_callback_lock);
 410                         sk_wake_async(skpair, SOCK_WAKE_WAITD, POLL_HUP);
 411                         read_unlock(&skpair->sk_callback_lock);
 412                 }
 413                 sock_put(skpair); /* It may now die */
 414                 unix_peer(sk) = NULL;
 415         }
 416
 417         /* Try to flush out this socket. Throw out buffers at least */
 418
 419         while ((skb = skb_dequeue(&sk->sk_receive_queue)) != NULL) {
 420                 if (state == TCP_LISTEN)
 421                         unix_release_sock(skb->sk, 1);
 422                 /* passed fds are erased in the kfree_skb hook        */
 423                 kfree_skb(skb);
 424         }
 425
 426         if (dentry) {
 427                 dput(dentry);
 428                 mntput(mnt);
 429         }
 430
 431         sock_put(sk);
 432
 433         /* ---- Socket is dead now and most probably destroyed ---- */
 434
 435         /*
 436          * Fixme: BSD difference: In BSD all sockets connected to use get
 437          *        ECONNRESET and we die on the spot. In Linux we behave
 438          *        like files and pipes do and wait for the last
 439          *        dereference.
 440          *
 441          * Can't we simply set sock->err?
 442          *
 443          *        What the above comment does talk about? --ANK(980817)
 444          */
 445
 446         if (unix_tot_inflight)
 447                 unix_gc();              /* Garbage collect fds */
 448
 449         return 0;
 450 }
 451
 452 static int unix_listen(struct socket *sock, int backlog)
 453 {
 454         int err;
 455         struct sock *sk = sock->sk;
 456         struct unix_sock *u = unix_sk(sk);
 457
 458         err = -EOPNOTSUPP;
 459         if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
 460                 goto out;       /* Only stream/seqpacket sockets accept */
 461         err = -EINVAL;
 462         if (!u->addr)
 463                 goto out;       /* No listens on an unbound socket */
 464         unix_state_lock(sk);
 465         if (sk->sk_state != TCP_CLOSE && sk->sk_state != TCP_LISTEN)
 466                 goto out_unlock;
 467         if (backlog > sk->sk_max_ack_backlog)
 468                 wake_up_interruptible_all(&u->peer_wait);
 469         sk->sk_max_ack_backlog  = backlog;
 470         sk->sk_state            = TCP_LISTEN;
 471         /* set credentials so connect can copy them */
 472         sk->sk_peercred.pid     = task_tgid_vnr(current);
 473         current_euid_egid(&sk->sk_peercred.uid, &sk->sk_peercred.gid);
 474         err = 0;
 475
 476 out_unlock:
 477         unix_state_unlock(sk);
 478 out:
 479         return err;
 480 }
 481
 482 static int unix_release(struct socket *);
 483 static int unix_bind(struct socket *, struct sockaddr *, int);
 484 static int unix_stream_connect(struct socket *, struct sockaddr *,
 485                                int addr_len, int flags);
 486 static int unix_socketpair(struct socket *, struct socket *);
 487 static int unix_accept(struct socket *, struct socket *, int);
 488 static int unix_getname(struct socket *, struct sockaddr *, int *, int);
 489 static unsigned int unix_poll(struct file *, struct socket *, poll_table *);
 490 static unsigned int unix_dgram_poll(struct file *, struct socket *,
 491                                     poll_table *);
 492 static int unix_ioctl(struct socket *, unsigned int, unsigned long);
 493 static int unix_shutdown(struct socket *, int);
 494 static int unix_stream_sendmsg(struct kiocb *, struct socket *,
 495                                struct msghdr *, size_t);
 496 static int unix_stream_recvmsg(struct kiocb *, struct socket *,
 497                                struct msghdr *, size_t, int);
 498 static int unix_dgram_sendmsg(struct kiocb *, struct socket *,
 499                               struct msghdr *, size_t);
 500 static int unix_dgram_recvmsg(struct kiocb *, struct socket *,
 501                               struct msghdr *, size_t, int);
 502 static int unix_dgram_connect(struct socket *, struct sockaddr *,
 503                               int, int);
 504 static int unix_seqpacket_sendmsg(struct kiocb *, struct socket *,
 505                                   struct msghdr *, size_t);
 506 static int unix_seqpacket_recvmsg(struct kiocb *, struct socket *,
 507                                   struct msghdr *, size_t, int);
 508
 509 static const struct proto_ops unix_stream_ops = {
 510         .family =       PF_UNIX,
 511         .owner =        THIS_MODULE,
 512         .release =      unix_release,
 513         .bind =         unix_bind,
 514         .connect =      unix_stream_connect,
 515         .socketpair =   unix_socketpair,
 516         .accept =       unix_accept,
 517         .getname =      unix_getname,
 518         .poll =         unix_poll,
 519         .ioctl =        unix_ioctl,
 520         .listen =       unix_listen,
 521         .shutdown =     unix_shutdown,
 522         .setsockopt =   sock_no_setsockopt,
 523         .getsockopt =   sock_no_getsockopt,
 524         .sendmsg =      unix_stream_sendmsg,
 525         .recvmsg =      unix_stream_recvmsg,
 526         .mmap =         sock_no_mmap,
 527         .sendpage =     sock_no_sendpage,
 528 };
 529
 530 static const struct proto_ops unix_dgram_ops = {
 531         .family =       PF_UNIX,
 532         .owner =        THIS_MODULE,
 533         .release =      unix_release,
 534         .bind =         unix_bind,
 535         .connect =      unix_dgram_connect,
 536         .socketpair =   unix_socketpair,
 537         .accept =       sock_no_accept,
 538         .getname =      unix_getname,
 539         .poll =         unix_dgram_poll,
 540         .ioctl =        unix_ioctl,
 541         .listen =       sock_no_listen,
 542         .shutdown =     unix_shutdown,
 543         .setsockopt =   sock_no_setsockopt,
 544         .getsockopt =   sock_no_getsockopt,
 545         .sendmsg =      unix_dgram_sendmsg,
 546         .recvmsg =      unix_dgram_recvmsg,
 547         .mmap =         sock_no_mmap,
 548         .sendpage =     sock_no_sendpage,
 549 };
 550
 551 static const struct proto_ops unix_seqpacket_ops = {
 552         .family =       PF_UNIX,
 553         .owner =        THIS_MODULE,
 554         .release =      unix_release,
 555         .bind =         unix_bind,
 556         .connect =      unix_stream_connect,
 557         .socketpair =   unix_socketpair,
 558         .accept =       unix_accept,
 559         .getname =      unix_getname,
 560         .poll =         unix_dgram_poll,
 561         .ioctl =        unix_ioctl,
 562         .listen =       unix_listen,
 563         .shutdown =     unix_shutdown,
 564         .setsockopt =   sock_no_setsockopt,
 565         .getsockopt =   sock_no_getsockopt,
 566         .sendmsg =      unix_seqpacket_sendmsg,
 567         .recvmsg =      unix_seqpacket_recvmsg,
 568         .mmap =         sock_no_mmap,
 569         .sendpage =     sock_no_sendpage,
 570 };
 571
 572 static struct proto unix_proto = {
 573         .name                   = "UNIX",
 574         .owner                  = THIS_MODULE,
 575         .obj_size               = sizeof(struct unix_sock),
 576 };
 577
 578 /*
 579  * AF_UNIX sockets do not interact with hardware, hence they
 580  * dont trigger interrupts - so it's safe for them to have
 581  * bh-unsafe locking for their sk_receive_queue.lock. Split off
 582  * this special lock-class by reinitializing the spinlock key:
 583  */
 584 static struct lock_class_key af_unix_sk_receive_queue_lock_key;
 585
 586 static struct sock *unix_create1(struct net *net, struct socket *sock)
 587 {
 588         struct sock *sk = NULL;
 589         struct unix_sock *u;
 590
 591         atomic_inc(&unix_nr_socks);
 592         if (atomic_read(&unix_nr_socks) > 2 * get_max_files())
 593                 goto out;
 594
 595         sk = sk_alloc(net, PF_UNIX, GFP_KERNEL, &unix_proto);
 596         if (!sk)
 597                 goto out;
 598
 599         sock_init_data(sock, sk);
 600         lockdep_set_class(&sk->sk_receive_queue.lock,
 601                                 &af_unix_sk_receive_queue_lock_key);
 602
 603         sk->sk_write_space      = unix_write_space;
 604         sk->sk_max_ack_backlog  = net->unx.sysctl_max_dgram_qlen;
 605         sk->sk_destruct         = unix_sock_destructor;
 606         u         = unix_sk(sk);
 607         u->dentry = NULL;
 608         u->mnt    = NULL;
 609         spin_lock_init(&u->lock);
 610         atomic_long_set(&u->inflight, 0);
 611         INIT_LIST_HEAD(&u->link);
 612         mutex_init(&u->readlock); /* single task reading lock */
 613         init_waitqueue_head(&u->peer_wait);
 614         unix_insert_socket(unix_sockets_unbound, sk);
 615 out:
 616         if (sk == NULL)
 617                 atomic_dec(&unix_nr_socks);
 618         else {
 619                 local_bh_disable();
 620                 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
 621                 local_bh_enable();
 622         }
 623         return sk;
 624 }
 625
 626 static int unix_create(struct net *net, struct socket *sock, int protocol)
 627 {
 628         if (protocol && protocol != PF_UNIX)
 629                 return -EPROTONOSUPPORT;
 630
 631         sock->state = SS_UNCONNECTED;
 632
 633         switch (sock->type) {
 634         case SOCK_STREAM:
 635                 sock->ops = &unix_stream_ops;
 636                 break;
 637                 /*
 638                  *      Believe it or not BSD has AF_UNIX, SOCK_RAW though
 639                  *      nothing uses it.
 640                  */
 641         case SOCK_RAW:
 642                 sock->type = SOCK_DGRAM;
 643         case SOCK_DGRAM:
 644                 sock->ops = &unix_dgram_ops;
 645                 break;
 646         case SOCK_SEQPACKET:
 647                 sock->ops = &unix_seqpacket_ops;
 648                 break;
 649         default:
 650                 return -ESOCKTNOSUPPORT;
 651         }
 652
 653         return unix_create1(net, sock) ? 0 : -ENOMEM;
 654 }
 655
 656 static int unix_release(struct socket *sock)
 657 {
 658         struct sock *sk = sock->sk;
 659
 660         if (!sk)
 661                 return 0;
 662
 663         sock->sk = NULL;
 664
 665         return unix_release_sock(sk, 0);
 666 }
 667
 668 static int unix_autobind(struct socket *sock)
 669 {
 670         struct sock *sk = sock->sk;
 671         struct net *net = sock_net(sk);
 672         struct unix_sock *u = unix_sk(sk);
 673         static u32 ordernum = 1;
 674         struct unix_address *addr;
 675         int err;
 676         unsigned int retries = 0;
 677
 678         mutex_lock(&u->readlock);
 679
 680         err = 0;
 681         if (u->addr)
 682                 goto out;
 683
 684         err = -ENOMEM;
 685         addr = kzalloc(sizeof(*addr) + sizeof(short) + 16, GFP_KERNEL);
 686         if (!addr)
 687                 goto out;
 688
 689         addr->name->sun_family = AF_UNIX;
 690         atomic_set(&addr->refcnt, 1);
 691
 692 retry:
 693         addr->len = sprintf(addr->name->sun_path+1, "%05x", ordernum) + 1 + sizeof(short);
 694         addr->hash = unix_hash_fold(csum_partial(addr->name, addr->len, 0));
 695
 696         spin_lock(&unix_table_lock);
 697         ordernum = (ordernum+1)&0xFFFFF;
 698
 699         if (__unix_find_socket_byname(net, addr->name, addr->len, sock->type,
 700                                       addr->hash)) {
 701                 spin_unlock(&unix_table_lock);
 702                 /*
 703                  * __unix_find_socket_byname() may take long time if many names
 704                  * are already in use.
 705                  */
 706                 cond_resched();
 707                 /* Give up if all names seems to be in use. */
 708                 if (retries++ == 0xFFFFF) {
 709                         err = -ENOSPC;
 710                         kfree(addr);
 711                         goto out;
 712                 }
 713                 goto retry;
 714         }
 715         addr->hash ^= sk->sk_type;
 716
 717         __unix_remove_socket(sk);
 718         u->addr = addr;
 719         __unix_insert_socket(&unix_socket_table[addr->hash], sk);
 720         spin_unlock(&unix_table_lock);
 721         err = 0;
 722
 723 out:    mutex_unlock(&u->readlock);
 724         return err;
 725 }
 726
 727 static struct sock *unix_find_other(struct net *net,
 728                                     struct sockaddr_un *sunname, int len,
 729                                     int type, unsigned hash, int *error)
 730 {
 731         struct sock *u;
 732         struct path path;
 733         int err = 0;
 734
 735         if (sunname->sun_path[0]) {
 736                 struct inode *inode;
 737                 err = kern_path(sunname->sun_path, LOOKUP_FOLLOW, &path);
 738                 if (err)
 739                         goto fail;
 740                 inode = path.dentry->d_inode;
 741                 err = inode_permission(inode, MAY_WRITE);
 742                 if (err)
 743                         goto put_fail;
 744
 745                 err = -ECONNREFUSED;
 746                 if (!S_ISSOCK(inode->i_mode))
 747                         goto put_fail;
 748                 u = unix_find_socket_byinode(net, inode);
 749                 if (!u)
 750                         goto put_fail;
 751
 752                 if (u->sk_type == type)
 753                         touch_atime(path.mnt, path.dentry);
 754
 755                 path_put(&path);
 756
 757                 err = -EPROTOTYPE;
 758                 if (u->sk_type != type) {
 759                         sock_put(u);
 760                         goto fail;
 761                 }
 762         } else {
 763                 err = -ECONNREFUSED;
 764                 u = unix_find_socket_byname(net, sunname, len, type, hash);
 765                 if (u) {
 766                         struct dentry *dentry;
 767                         dentry = unix_sk(u)->dentry;
 768                         if (dentry)
 769                                 touch_atime(unix_sk(u)->mnt, dentry);
 770                 } else
 771                         goto fail;
 772         }
 773         return u;
 774
 775 put_fail:
 776         path_put(&path);
 777 fail:
 778         *error = err;
 779         return NULL;
 780 }
 781
 782
 783 static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
 784 {
 785         struct sock *sk = sock->sk;
 786         struct net *net = sock_net(sk);
 787         struct unix_sock *u = unix_sk(sk);
 788         struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
 789         struct dentry *dentry = NULL;
 790         struct nameidata nd;
 791         int err;
 792         unsigned hash;
 793         struct unix_address *addr;
 794         struct hlist_head *list;
 795
 796         err = -EINVAL;
 797         if (sunaddr->sun_family != AF_UNIX)
 798                 goto out;
 799
 800         if (addr_len == sizeof(short)) {
 801                 err = unix_autobind(sock);
 802                 goto out;
 803         }
 804
 805         err = unix_mkname(sunaddr, addr_len, &hash);
 806         if (err < 0)
 807                 goto out;
 808         addr_len = err;
 809
 810         mutex_lock(&u->readlock);
 811
 812         err = -EINVAL;
 813         if (u->addr)
 814                 goto out_up;
 815
 816         err = -ENOMEM;
 817         addr = kmalloc(sizeof(*addr)+addr_len, GFP_KERNEL);
 818         if (!addr)
 819                 goto out_up;
 820
 821         memcpy(addr->name, sunaddr, addr_len);
 822         addr->len = addr_len;
 823         addr->hash = hash ^ sk->sk_type;
 824         atomic_set(&addr->refcnt, 1);
 825
 826         if (sunaddr->sun_path[0]) {
 827                 unsigned int mode;
 828                 err = 0;
 829                 /*
 830                  * Get the parent directory, calculate the hash for last
 831                  * component.
 832                  */
 833                 err = path_lookup(sunaddr->sun_path, LOOKUP_PARENT, &nd);
 834                 if (err)
 835                         goto out_mknod_parent;
 836
 837                 dentry = lookup_create(&nd, 0);
 838                 err = PTR_ERR(dentry);
 839                 if (IS_ERR(dentry))
 840                         goto out_mknod_unlock;
 841
 842                 /*
 843                  * All right, let's create it.
 844                  */
 845                 mode = S_IFSOCK |
 846                        (SOCK_INODE(sock)->i_mode & ~current_umask());
 847                 err = mnt_want_write(nd.path.mnt);
 848                 if (err)
 849                         goto out_mknod_dput;
 850                 err = security_path_mknod(&nd.path, dentry, mode, 0);
 851                 if (err)
 852                         goto out_mknod_drop_write;
 853                 err = vfs_mknod(nd.path.dentry->d_inode, dentry, mode, 0);
 854 out_mknod_drop_write:
 855                 mnt_drop_write(nd.path.mnt);
 856                 if (err)
 857                         goto out_mknod_dput;
 858                 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
 859                 dput(nd.path.dentry);
 860                 nd.path.dentry = dentry;
 861
 862                 addr->hash = UNIX_HASH_SIZE;
 863         }
 864
 865         spin_lock(&unix_table_lock);
 866
 867         if (!sunaddr->sun_path[0]) {
 868                 err = -EADDRINUSE;
 869                 if (__unix_find_socket_byname(net, sunaddr, addr_len,
 870                                               sk->sk_type, hash)) {
 871                         unix_release_addr(addr);
 872                         goto out_unlock;
 873                 }
 874
 875                 list = &unix_socket_table[addr->hash];
 876         } else {
 877                 list = &unix_socket_table[dentry->d_inode->i_ino & (UNIX_HASH_SIZE-1)];
 878                 u->dentry = nd.path.dentry;
 879                 u->mnt    = nd.path.mnt;
 880         }
 881
 882         err = 0;
 883         __unix_remove_socket(sk);
 884         u->addr = addr;
 885         __unix_insert_socket(list, sk);
 886
 887 out_unlock:
 888         spin_unlock(&unix_table_lock);
 889 out_up:
 890         mutex_unlock(&u->readlock);
 891 out:
 892         return err;
 893
 894 out_mknod_dput:
 895         dput(dentry);
 896 out_mknod_unlock:
 897         mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
 898         path_put(&nd.path);
 899 out_mknod_parent:
 900         if (err == -EEXIST)
 901                 err = -EADDRINUSE;
 902         unix_release_addr(addr);
 903         goto out_up;
 904 }
 905
 906 static void unix_state_double_lock(struct sock *sk1, struct sock *sk2)
 907 {
 908         if (unlikely(sk1 == sk2) || !sk2) {
 909                 unix_state_lock(sk1);
 910                 return;
 911         }
 912         if (sk1 < sk2) {
 913                 unix_state_lock(sk1);
 914                 unix_state_lock_nested(sk2);
 915         } else {
 916                 unix_state_lock(sk2);
 917                 unix_state_lock_nested(sk1);
 918         }
 919 }
 920
 921 static void unix_state_double_unlock(struct sock *sk1, struct sock *sk2)
 922 {
 923         if (unlikely(sk1 == sk2) || !sk2) {
 924                 unix_state_unlock(sk1);
 925                 return;
 926         }
 927         unix_state_unlock(sk1);
 928         unix_state_unlock(sk2);
 929 }
 930
 931 static int unix_dgram_connect(struct socket *sock, struct sockaddr *addr,
 932                               int alen, int flags)
 933 {
 934         struct sock *sk = sock->sk;
 935         struct net *net = sock_net(sk);
 936         struct sockaddr_un *sunaddr = (struct sockaddr_un *)addr;
 937         struct sock *other;
 938         unsigned hash;
 939         int err;
 940
 941         if (addr->sa_family != AF_UNSPEC) {
 942                 err = unix_mkname(sunaddr, alen, &hash);
 943                 if (err < 0)
 944                         goto out;
 945                 alen = err;
 946
 947                 if (test_bit(SOCK_PASSCRED, &sock->flags) &&
 948                     !unix_sk(sk)->addr && (err = unix_autobind(sock)) != 0)
 949                         goto out;
 950
 951 restart:
 952                 other = unix_find_other(net, sunaddr, alen, sock->type, hash, &err);
 953                 if (!other)
 954                         goto out;
 955
 956                 unix_state_double_lock(sk, other);
 957
 958                 /* Apparently VFS overslept socket death. Retry. */
 959                 if (sock_flag(other, SOCK_DEAD)) {
 960                         unix_state_double_unlock(sk, other);
 961                         sock_put(other);
 962                         goto restart;
 963                 }
 964
 965                 err = -EPERM;
 966                 if (!unix_may_send(sk, other))
 967                         goto out_unlock;
 968
 969                 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
 970                 if (err)
 971                         goto out_unlock;
 972
 973         } else {
 974                 /*
 975                  *      1003.1g breaking connected state with AF_UNSPEC
 976                  */
 977                 other = NULL;
 978                 unix_state_double_lock(sk, other);
 979         }
 980
 981         /*
 982          * If it was connected, reconnect.
 983          */
 984         if (unix_peer(sk)) {
 985                 struct sock *old_peer = unix_peer(sk);
 986                 unix_peer(sk) = other;
 987                 unix_state_double_unlock(sk, other);
 988
 989                 if (other != old_peer)
 990                         unix_dgram_disconnected(sk, old_peer);
 991                 sock_put(old_peer);
 992         } else {
 993                 unix_peer(sk) = other;
 994                 unix_state_double_unlock(sk, other);
 995         }
 996         return 0;
 997
 998 out_unlock:
 999         unix_state_double_unlock(sk, other);
1000         sock_put(other);
1001 out:
1002         return err;
1003 }
1004
1005 static long unix_wait_for_peer(struct sock *other, long timeo)
1006 {
1007         struct unix_sock *u = unix_sk(other);
1008         int sched;
1009         DEFINE_WAIT(wait);
1010
1011         prepare_to_wait_exclusive(&u->peer_wait, &wait, TASK_INTERRUPTIBLE);
1012
1013         sched = !sock_flag(other, SOCK_DEAD) &&
1014                 !(other->sk_shutdown & RCV_SHUTDOWN) &&
1015                 unix_recvq_full(other);
1016
1017         unix_state_unlock(other);
1018
1019         if (sched)
1020                 timeo = schedule_timeout(timeo);
1021
1022         finish_wait(&u->peer_wait, &wait);
1023         return timeo;
1024 }
1025
1026 static int unix_stream_connect(struct socket *sock, struct sockaddr *uaddr,
1027                                int addr_len, int flags)
1028 {
1029         struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
1030         struct sock *sk = sock->sk;
1031         struct net *net = sock_net(sk);
1032         struct unix_sock *u = unix_sk(sk), *newu, *otheru;
1033         struct sock *newsk = NULL;
1034         struct sock *other = NULL;
1035         struct sk_buff *skb = NULL;
1036         unsigned hash;
1037         int st;
1038         int err;
1039         long timeo;
1040
1041         err = unix_mkname(sunaddr, addr_len, &hash);
1042         if (err < 0)
1043                 goto out;
1044         addr_len = err;
1045
1046         if (test_bit(SOCK_PASSCRED, &sock->flags)
1047                 && !u->addr && (err = unix_autobind(sock)) != 0)
1048                 goto out;
1049
1050         timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
1051
1052         /* First of all allocate resources.
1053            If we will make it after state is locked,
1054            we will have to recheck all again in any case.
1055          */
1056
1057         err = -ENOMEM;
1058
1059         /* create new sock for complete connection */
1060         newsk = unix_create1(sock_net(sk), NULL);
1061         if (newsk == NULL)
1062                 goto out;
1063
1064         /* Allocate skb for sending to listening sock */
1065         skb = sock_wmalloc(newsk, 1, 0, GFP_KERNEL);
1066         if (skb == NULL)
1067                 goto out;
1068
1069 restart:
1070         /*  Find listening sock. */
1071         other = unix_find_other(net, sunaddr, addr_len, sk->sk_type, hash, &err);
1072         if (!other)
1073                 goto out;
1074
1075         /* Latch state of peer */
1076         unix_state_lock(other);
1077
1078         /* Apparently VFS overslept socket death. Retry. */
1079         if (sock_flag(other, SOCK_DEAD)) {
1080                 unix_state_unlock(other);
1081                 sock_put(other);
1082                 goto restart;
1083         }
1084
1085         err = -ECONNREFUSED;
1086         if (other->sk_state != TCP_LISTEN)
1087                 goto out_unlock;
1088         if (other->sk_shutdown & RCV_SHUTDOWN)
1089                 goto out_unlock;
1090
1091         if (unix_recvq_full(other)) {
1092                 err = -EAGAIN;
1093                 if (!timeo)
1094                         goto out_unlock;
1095
1096                 timeo = unix_wait_for_peer(other, timeo);
1097
1098                 err = sock_intr_errno(timeo);
1099                 if (signal_pending(current))
1100                         goto out;
1101                 sock_put(other);
1102                 goto restart;
1103         }
1104
1105         /* Latch our state.
1106
1107            It is tricky place. We need to grab write lock and cannot
1108            drop lock on peer. It is dangerous because deadlock is
1109            possible. Connect to self case and simultaneous
1110            attempt to connect are eliminated by checking socket
1111            state. other is TCP_LISTEN, if sk is TCP_LISTEN we
1112            check this before attempt to grab lock.
1113
1114            Well, and we have to recheck the state after socket locked.
1115          */
1116         st = sk->sk_state;
1117
1118         switch (st) {
1119         case TCP_CLOSE:
1120                 /* This is ok... continue with connect */
1121                 break;
1122         case TCP_ESTABLISHED:
1123                 /* Socket is already connected */
1124                 err = -EISCONN;
1125                 goto out_unlock;
1126         default:
1127                 err = -EINVAL;
1128                 goto out_unlock;
1129         }
1130
1131         unix_state_lock_nested(sk);
1132
1133         if (sk->sk_state != st) {
1134                 unix_state_unlock(sk);
1135                 unix_state_unlock(other);
1136                 sock_put(other);
1137                 goto restart;
1138         }
1139
1140         err = security_unix_stream_connect(sock, other->sk_socket, newsk);
1141         if (err) {
1142                 unix_state_unlock(sk);
1143                 goto out_unlock;
1144         }
1145
1146         /* The way is open! Fastly set all the necessary fields... */
1147
1148         sock_hold(sk);
1149         unix_peer(newsk)        = sk;
1150         newsk->sk_state         = TCP_ESTABLISHED;
1151         newsk->sk_type          = sk->sk_type;
1152         newsk->sk_peercred.pid  = task_tgid_vnr(current);
1153         current_euid_egid(&newsk->sk_peercred.uid, &newsk->sk_peercred.gid);
1154         newu = unix_sk(newsk);
1155         newsk->sk_sleep         = &newu->peer_wait;
1156         otheru = unix_sk(other);
1157
1158         /* copy address information from listening to new sock*/
1159         if (otheru->addr) {
1160                 atomic_inc(&otheru->addr->refcnt);
1161                 newu->addr = otheru->addr;
1162         }
1163         if (otheru->dentry) {
1164                 newu->dentry    = dget(otheru->dentry);
1165                 newu->mnt       = mntget(otheru->mnt);
1166         }
1167
1168         /* Set credentials */
1169         sk->sk_peercred = other->sk_peercred;
1170
1171         sock->state     = SS_CONNECTED;
1172         sk->sk_state    = TCP_ESTABLISHED;
1173         sock_hold(newsk);
1174
1175         smp_mb__after_atomic_inc();     /* sock_hold() does an atomic_inc() */
1176         unix_peer(sk)   = newsk;
1177
1178         unix_state_unlock(sk);
1179
1180         /* take ten and and send info to listening sock */
1181         spin_lock(&other->sk_receive_queue.lock);
1182         __skb_queue_tail(&other->sk_receive_queue, skb);
1183         spin_unlock(&other->sk_receive_queue.lock);
1184         unix_state_unlock(other);
1185         other->sk_data_ready(other, 0);
1186         sock_put(other);
1187         return 0;
1188
1189 out_unlock:
1190         if (other)
1191                 unix_state_unlock(other);
1192
1193 out:
1194         kfree_skb(skb);
1195         if (newsk)
1196                 unix_release_sock(newsk, 0);
1197         if (other)
1198                 sock_put(other);
1199         return err;
1200 }
1201
1202 static int unix_socketpair(struct socket *socka, struct socket *sockb)
1203 {
1204         struct sock *ska = socka->sk, *skb = sockb->sk;
1205
1206         /* Join our sockets back to back */
1207         sock_hold(ska);
1208         sock_hold(skb);
1209         unix_peer(ska) = skb;
1210         unix_peer(skb) = ska;
1211         ska->sk_peercred.pid = skb->sk_peercred.pid = task_tgid_vnr(current);
1212         current_euid_egid(&skb->sk_peercred.uid, &skb->sk_peercred.gid);
1213         ska->sk_peercred.uid = skb->sk_peercred.uid;
1214         ska->sk_peercred.gid = skb->sk_peercred.gid;
1215
1216         if (ska->sk_type != SOCK_DGRAM) {
1217                 ska->sk_state = TCP_ESTABLISHED;
1218                 skb->sk_state = TCP_ESTABLISHED;
1219                 socka->state  = SS_CONNECTED;
1220                 sockb->state  = SS_CONNECTED;
1221         }
1222         return 0;
1223 }
1224
1225 static int unix_accept(struct socket *sock, struct socket *newsock, int flags)
1226 {
1227         struct sock *sk = sock->sk;
1228         struct sock *tsk;
1229         struct sk_buff *skb;
1230         int err;
1231
1232         err = -EOPNOTSUPP;
1233         if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET)
1234                 goto out;
1235
1236         err = -EINVAL;
1237         if (sk->sk_state != TCP_LISTEN)
1238                 goto out;
1239
1240         /* If socket state is TCP_LISTEN it cannot change (for now...),
1241          * so that no locks are necessary.
1242          */
1243
1244         skb = skb_recv_datagram(sk, 0, flags&O_NONBLOCK, &err);
1245         if (!skb) {
1246                 /* This means receive shutdown. */
1247                 if (err == 0)
1248                         err = -EINVAL;
1249                 goto out;
1250         }
1251
1252         tsk = skb->sk;
1253         skb_free_datagram(sk, skb);
1254         wake_up_interruptible(&unix_sk(sk)->peer_wait);
1255
1256         /* attach accepted sock to socket */
1257         unix_state_lock(tsk);
1258         newsock->state = SS_CONNECTED;
1259         sock_graft(tsk, newsock);
1260         unix_state_unlock(tsk);
1261         return 0;
1262
1263 out:
1264         return err;
1265 }
1266
1267
1268 static int unix_getname(struct socket *sock, struct sockaddr *uaddr, int *uaddr_len, int peer)
1269 {
1270         struct sock *sk = sock->sk;
1271         struct unix_sock *u;
1272         struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;
1273         int err = 0;
1274
1275         if (peer) {
1276                 sk = unix_peer_get(sk);
1277
1278                 err = -ENOTCONN;
1279                 if (!sk)
1280                         goto out;
1281                 err = 0;
1282         } else {
1283                 sock_hold(sk);
1284         }
1285
1286         u = unix_sk(sk);
1287         unix_state_lock(sk);
1288         if (!u->addr) {
1289                 sunaddr->sun_family = AF_UNIX;
1290                 sunaddr->sun_path[0] = 0;
1291                 *uaddr_len = sizeof(short);
1292         } else {
1293                 struct unix_address *addr = u->addr;
1294
1295                 *uaddr_len = addr->len;
1296                 memcpy(sunaddr, addr->name, *uaddr_len);
1297         }
1298         unix_state_unlock(sk);
1299         sock_put(sk);
1300 out:
1301         return err;
1302 }
1303
1304 static void unix_detach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1305 {
1306         int i;
1307
1308         scm->fp = UNIXCB(skb).fp;
1309         skb->destructor = sock_wfree;
1310         UNIXCB(skb).fp = NULL;
1311
1312         for (i = scm->fp->count-1; i >= 0; i--)
1313                 unix_notinflight(scm->fp->fp[i]);
1314 }
1315
1316 static void unix_destruct_fds(struct sk_buff *skb)
1317 {
1318         struct scm_cookie scm;
1319         memset(&scm, 0, sizeof(scm));
1320         unix_detach_fds(&scm, skb);
1321
1322         /* Alas, it calls VFS */
1323         /* So fscking what? fput() had been SMP-safe since the last Summer */
1324         scm_destroy(&scm);
1325         sock_wfree(skb);
1326 }
1327
1328 #define MAX_RECURSION_LEVEL 4
1329
1330 static int unix_attach_fds(struct scm_cookie *scm, struct sk_buff *skb)
1331 {
1332         int i;
1333         unsigned char max_level = 0;
1334         int unix_sock_count = 0;
1335
1336         for (i = scm->fp->count - 1; i >= 0; i--) {
1337                 struct sock *sk = unix_get_socket(scm->fp->fp[i]);
1338
1339                 if (sk) {
1340                         unix_sock_count++;
1341                         max_level = max(max_level,
1342                                         unix_sk(sk)->recursion_level);
1343                 }
1344         }
1345         if (unlikely(max_level > MAX_RECURSION_LEVEL))
1346                 return -ETOOMANYREFS;
1347
1348         /*
1349          * Need to duplicate file references for the sake of garbage
1350          * collection.  Otherwise a socket in the fps might become a
1351          * candidate for GC while the skb is not yet queued.
1352          */
1353         UNIXCB(skb).fp = scm_fp_dup(scm->fp);
1354         if (!UNIXCB(skb).fp)
1355                 return -ENOMEM;
1356
1357         if (unix_sock_count) {
1358                 for (i = scm->fp->count - 1; i >= 0; i--)
1359                         unix_inflight(scm->fp->fp[i]);
1360         }
1361         skb->destructor = unix_destruct_fds;
1362         return max_level;
1363 }
1364
1365 /*
1366  *      Send AF_UNIX data.
1367  */
1368
1369 static int unix_dgram_sendmsg(struct kiocb *kiocb, struct socket *sock,
1370                               struct msghdr *msg, size_t len)
1371 {
1372         struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1373         struct sock *sk = sock->sk;
1374         struct net *net = sock_net(sk);
1375         struct unix_sock *u = unix_sk(sk);
1376         struct sockaddr_un *sunaddr = msg->msg_name;
1377         struct sock *other = NULL;
1378         int namelen = 0; /* fake GCC */
1379         int err;
1380         unsigned hash;
1381         struct sk_buff *skb;
1382         long timeo;
1383         struct scm_cookie tmp_scm;
1384         int max_level = 0;
1385
1386         if (NULL == siocb->scm)
1387                 siocb->scm = &tmp_scm;
1388         wait_for_unix_gc();
1389         err = scm_send(sock, msg, siocb->scm);
1390         if (err < 0)
1391                 return err;
1392
1393         err = -EOPNOTSUPP;
1394         if (msg->msg_flags&MSG_OOB)
1395                 goto out;
1396
1397         if (msg->msg_namelen) {
1398                 err = unix_mkname(sunaddr, msg->msg_namelen, &hash);
1399                 if (err < 0)
1400                         goto out;
1401                 namelen = err;
1402         } else {
1403                 sunaddr = NULL;
1404                 err = -ENOTCONN;
1405                 other = unix_peer_get(sk);
1406                 if (!other)
1407                         goto out;
1408         }
1409
1410         if (test_bit(SOCK_PASSCRED, &sock->flags)
1411                 && !u->addr && (err = unix_autobind(sock)) != 0)
1412                 goto out;
1413
1414         err = -EMSGSIZE;
1415         if (len > sk->sk_sndbuf - 32)
1416                 goto out;
1417
1418         skb = sock_alloc_send_skb(sk, len, msg->msg_flags&MSG_DONTWAIT, &err);
1419         if (skb == NULL)
1420                 goto out;
1421
1422         memcpy(UNIXCREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1423         if (siocb->scm->fp) {
1424                 err = unix_attach_fds(siocb->scm, skb);
1425                 if (err < 0)
1426                         goto out_free;
1427                 max_level = err + 1;
1428         }
1429         unix_get_secdata(siocb->scm, skb);
1430
1431         skb_reset_transport_header(skb);
1432         err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
1433         if (err)
1434                 goto out_free;
1435
1436         timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1437
1438 restart:
1439         if (!other) {
1440                 err = -ECONNRESET;
1441                 if (sunaddr == NULL)
1442                         goto out_free;
1443
1444                 other = unix_find_other(net, sunaddr, namelen, sk->sk_type,
1445                                         hash, &err);
1446                 if (other == NULL)
1447                         goto out_free;
1448         }
1449
1450         unix_state_lock(other);
1451         err = -EPERM;
1452         if (!unix_may_send(sk, other))
1453                 goto out_unlock;
1454
1455         if (sock_flag(other, SOCK_DEAD)) {
1456                 /*
1457                  *      Check with 1003.1g - what should
1458                  *      datagram error
1459                  */
1460                 unix_state_unlock(other);
1461                 sock_put(other);
1462
1463                 err = 0;
1464                 unix_state_lock(sk);
1465                 if (unix_peer(sk) == other) {
1466                         unix_peer(sk) = NULL;
1467                         unix_state_unlock(sk);
1468
1469                         unix_dgram_disconnected(sk, other);
1470                         sock_put(other);
1471                         err = -ECONNREFUSED;
1472                 } else {
1473                         unix_state_unlock(sk);
1474                 }
1475
1476                 other = NULL;
1477                 if (err)
1478                         goto out_free;
1479                 goto restart;
1480         }
1481
1482         err = -EPIPE;
1483         if (other->sk_shutdown & RCV_SHUTDOWN)
1484                 goto out_unlock;
1485
1486         if (sk->sk_type != SOCK_SEQPACKET) {
1487                 err = security_unix_may_send(sk->sk_socket, other->sk_socket);
1488                 if (err)
1489                         goto out_unlock;
1490         }
1491
1492         if (unix_peer(other) != sk && unix_recvq_full(other)) {
1493                 if (!timeo) {
1494                         err = -EAGAIN;
1495                         goto out_unlock;
1496                 }
1497
1498                 timeo = unix_wait_for_peer(other, timeo);
1499
1500                 err = sock_intr_errno(timeo);
1501                 if (signal_pending(current))
1502                         goto out_free;
1503
1504                 goto restart;
1505         }
1506
1507         skb_queue_tail(&other->sk_receive_queue, skb);
1508         if (max_level > unix_sk(other)->recursion_level)
1509                 unix_sk(other)->recursion_level = max_level;
1510         unix_state_unlock(other);
1511         other->sk_data_ready(other, len);
1512         sock_put(other);
1513         scm_destroy(siocb->scm);
1514         return len;
1515
1516 out_unlock:
1517         unix_state_unlock(other);
1518 out_free:
1519         kfree_skb(skb);
1520 out:
1521         if (other)
1522                 sock_put(other);
1523         scm_destroy(siocb->scm);
1524         return err;
1525 }
1526
1527
1528 static int unix_stream_sendmsg(struct kiocb *kiocb, struct socket *sock,
1529                                struct msghdr *msg, size_t len)
1530 {
1531         struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1532         struct sock *sk = sock->sk;
1533         struct sock *other = NULL;
1534         struct sockaddr_un *sunaddr = msg->msg_name;
1535         int err, size;
1536         struct sk_buff *skb;
1537         int sent = 0;
1538         struct scm_cookie tmp_scm;
1539         bool fds_sent = false;
1540         int max_level = 0;
1541
1542         if (NULL == siocb->scm)
1543                 siocb->scm = &tmp_scm;
1544         wait_for_unix_gc();
1545         err = scm_send(sock, msg, siocb->scm);
1546         if (err < 0)
1547                 return err;
1548
1549         err = -EOPNOTSUPP;
1550         if (msg->msg_flags&MSG_OOB)
1551                 goto out_err;
1552
1553         if (msg->msg_namelen) {
1554                 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
1555                 goto out_err;
1556         } else {
1557                 sunaddr = NULL;
1558                 err = -ENOTCONN;
1559                 other = unix_peer(sk);
1560                 if (!other)
1561                         goto out_err;
1562         }
1563
1564         if (sk->sk_shutdown & SEND_SHUTDOWN)
1565                 goto pipe_err;
1566
1567         while (sent < len) {
1568                 /*
1569                  *      Optimisation for the fact that under 0.01% of X
1570                  *      messages typically need breaking up.
1571                  */
1572
1573                 size = len-sent;
1574
1575                 /* Keep two messages in the pipe so it schedules better */
1576                 if (size > ((sk->sk_sndbuf >> 1) - 64))
1577                         size = (sk->sk_sndbuf >> 1) - 64;
1578
1579                 if (size > SKB_MAX_ALLOC)
1580                         size = SKB_MAX_ALLOC;
1581
1582                 /*
1583                  *      Grab a buffer
1584                  */
1585
1586                 skb = sock_alloc_send_skb(sk, size, msg->msg_flags&MSG_DONTWAIT,
1587                                           &err);
1588
1589                 if (skb == NULL)
1590                         goto out_err;
1591
1592                 /*
1593                  *      If you pass two values to the sock_alloc_send_skb
1594                  *      it tries to grab the large buffer with GFP_NOFS
1595                  *      (which can fail easily), and if it fails grab the
1596                  *      fallback size buffer which is under a page and will
1597                  *      succeed. [Alan]
1598                  */
1599                 size = min_t(int, size, skb_tailroom(skb));
1600
1601                 memcpy(UNIXCREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1602                 /* Only send the fds in the first buffer */
1603                 if (siocb->scm->fp && !fds_sent) {
1604                         err = unix_attach_fds(siocb->scm, skb);
1605                         if (err < 0) {
1606                                 kfree_skb(skb);
1607                                 goto out_err;
1608                         }
1609                         max_level = err + 1;
1610                         fds_sent = true;
1611                 }
1612
1613                 err = memcpy_fromiovec(skb_put(skb, size), msg->msg_iov, size);
1614                 if (err) {
1615                         kfree_skb(skb);
1616                         goto out_err;
1617                 }
1618
1619                 unix_state_lock(other);
1620
1621                 if (sock_flag(other, SOCK_DEAD) ||
1622                     (other->sk_shutdown & RCV_SHUTDOWN))
1623                         goto pipe_err_free;
1624
1625                 skb_queue_tail(&other->sk_receive_queue, skb);
1626                 if (max_level > unix_sk(other)->recursion_level)
1627                         unix_sk(other)->recursion_level = max_level;
1628                 unix_state_unlock(other);
1629                 other->sk_data_ready(other, size);
1630                 sent += size;
1631         }
1632
1633         scm_destroy(siocb->scm);
1634         siocb->scm = NULL;
1635
1636         return sent;
1637
1638 pipe_err_free:
1639         unix_state_unlock(other);
1640         kfree_skb(skb);
1641 pipe_err:
1642         if (sent == 0 && !(msg->msg_flags&MSG_NOSIGNAL))
1643                 send_sig(SIGPIPE, current, 0);
1644         err = -EPIPE;
1645 out_err:
1646         scm_destroy(siocb->scm);
1647         siocb->scm = NULL;
1648         return sent ? : err;
1649 }
1650
1651 static int unix_seqpacket_sendmsg(struct kiocb *kiocb, struct socket *sock,
1652                                   struct msghdr *msg, size_t len)
1653 {
1654         int err;
1655         struct sock *sk = sock->sk;
1656
1657         err = sock_error(sk);
1658         if (err)
1659                 return err;
1660
1661         if (sk->sk_state != TCP_ESTABLISHED)
1662                 return -ENOTCONN;
1663
1664         if (msg->msg_namelen)
1665                 msg->msg_namelen = 0;
1666
1667         return unix_dgram_sendmsg(kiocb, sock, msg, len);
1668 }
1669
1670 static int unix_seqpacket_recvmsg(struct kiocb *iocb, struct socket *sock,
1671                               struct msghdr *msg, size_t size,
1672                               int flags)
1673 {
1674         struct sock *sk = sock->sk;
1675
1676         if (sk->sk_state != TCP_ESTABLISHED)
1677                 return -ENOTCONN;
1678
1679         return unix_dgram_recvmsg(iocb, sock, msg, size, flags);
1680 }
1681
1682 static void unix_copy_addr(struct msghdr *msg, struct sock *sk)
1683 {
1684         struct unix_sock *u = unix_sk(sk);
1685
1686         msg->msg_namelen = 0;
1687         if (u->addr) {
1688                 msg->msg_namelen = u->addr->len;
1689                 memcpy(msg->msg_name, u->addr->name, u->addr->len);
1690         }
1691 }
1692
1693 static int unix_dgram_recvmsg(struct kiocb *iocb, struct socket *sock,
1694                               struct msghdr *msg, size_t size,
1695                               int flags)
1696 {
1697         struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1698         struct scm_cookie tmp_scm;
1699         struct sock *sk = sock->sk;
1700         struct unix_sock *u = unix_sk(sk);
1701         int noblock = flags & MSG_DONTWAIT;
1702         struct sk_buff *skb;
1703         int err;
1704
1705         err = -EOPNOTSUPP;
1706         if (flags&MSG_OOB)
1707                 goto out;
1708
1709         msg->msg_namelen = 0;
1710
1711         mutex_lock(&u->readlock);
1712
1713         skb = skb_recv_datagram(sk, flags, noblock, &err);
1714         if (!skb) {
1715                 unix_state_lock(sk);
1716                 /* Signal EOF on disconnected non-blocking SEQPACKET socket. */
1717                 if (sk->sk_type == SOCK_SEQPACKET && err == -EAGAIN &&
1718                     (sk->sk_shutdown & RCV_SHUTDOWN))
1719                         err = 0;
1720                 unix_state_unlock(sk);
1721                 goto out_unlock;
1722         }
1723
1724         wake_up_interruptible_sync(&u->peer_wait);
1725
1726         if (msg->msg_name)
1727                 unix_copy_addr(msg, skb->sk);
1728
1729         if (size > skb->len)
1730                 size = skb->len;
1731         else if (size < skb->len)
1732                 msg->msg_flags |= MSG_TRUNC;
1733
1734         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, size);
1735         if (err)
1736                 goto out_free;
1737
1738         if (!siocb->scm) {
1739                 siocb->scm = &tmp_scm;
1740                 memset(&tmp_scm, 0, sizeof(tmp_scm));
1741         }
1742         siocb->scm->creds = *UNIXCREDS(skb);
1743         unix_set_secdata(siocb->scm, skb);
1744
1745         if (!(flags & MSG_PEEK)) {
1746                 if (UNIXCB(skb).fp)
1747                         unix_detach_fds(siocb->scm, skb);
1748         } else {
1749                 /* It is questionable: on PEEK we could:
1750                    - do not return fds - good, but too simple 8)
1751                    - return fds, and do not return them on read (old strategy,
1752                      apparently wrong)
1753                    - clone fds (I chose it for now, it is the most universal
1754                      solution)
1755
1756                    POSIX 1003.1g does not actually define this clearly
1757                    at all. POSIX 1003.1g doesn't define a lot of things
1758                    clearly however!
1759
1760                 */
1761                 if (UNIXCB(skb).fp)
1762                         siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1763         }
1764         err = size;
1765
1766         scm_recv(sock, msg, siocb->scm, flags);
1767
1768 out_free:
1769         skb_free_datagram(sk, skb);
1770 out_unlock:
1771         mutex_unlock(&u->readlock);
1772 out:
1773         return err;
1774 }
1775
1776 /*
1777  *      Sleep until data has arrive. But check for races..
1778  */
1779
1780 static long unix_stream_data_wait(struct sock *sk, long timeo)
1781 {
1782         DEFINE_WAIT(wait);
1783
1784         unix_state_lock(sk);
1785
1786         for (;;) {
1787                 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1788
1789                 if (!skb_queue_empty(&sk->sk_receive_queue) ||
1790                     sk->sk_err ||
1791                     (sk->sk_shutdown & RCV_SHUTDOWN) ||
1792                     signal_pending(current) ||
1793                     !timeo)
1794                         break;
1795
1796                 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1797                 unix_state_unlock(sk);
1798                 timeo = schedule_timeout(timeo);
1799                 unix_state_lock(sk);
1800                 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1801         }
1802
1803         finish_wait(sk->sk_sleep, &wait);
1804         unix_state_unlock(sk);
1805         return timeo;
1806 }
1807
1808
1809
1810 static int unix_stream_recvmsg(struct kiocb *iocb, struct socket *sock,
1811                                struct msghdr *msg, size_t size,
1812                                int flags)
1813 {
1814         struct sock_iocb *siocb = kiocb_to_siocb(iocb);
1815         struct scm_cookie tmp_scm;
1816         struct sock *sk = sock->sk;
1817         struct unix_sock *u = unix_sk(sk);
1818         struct sockaddr_un *sunaddr = msg->msg_name;
1819         int copied = 0;
1820         int check_creds = 0;
1821         int target;
1822         int err = 0;
1823         long timeo;
1824
1825         err = -EINVAL;
1826         if (sk->sk_state != TCP_ESTABLISHED)
1827                 goto out;
1828
1829         err = -EOPNOTSUPP;
1830         if (flags&MSG_OOB)
1831                 goto out;
1832
1833         target = sock_rcvlowat(sk, flags&MSG_WAITALL, size);
1834         timeo = sock_rcvtimeo(sk, flags&MSG_DONTWAIT);
1835
1836         msg->msg_namelen = 0;
1837
1838         /* Lock the socket to prevent queue disordering
1839          * while sleeps in memcpy_tomsg
1840          */
1841
1842         if (!siocb->scm) {
1843                 siocb->scm = &tmp_scm;
1844                 memset(&tmp_scm, 0, sizeof(tmp_scm));
1845         }
1846
1847         mutex_lock(&u->readlock);
1848
1849         do {
1850                 int chunk;
1851                 struct sk_buff *skb;
1852
1853                 unix_state_lock(sk);
1854                 skb = skb_dequeue(&sk->sk_receive_queue);
1855                 if (skb == NULL) {
1856                         unix_sk(sk)->recursion_level = 0;
1857                         if (copied >= target)
1858                                 goto unlock;
1859
1860                         /*
1861                          *      POSIX 1003.1g mandates this order.
1862                          */
1863
1864                         err = sock_error(sk);
1865                         if (err)
1866                                 goto unlock;
1867                         if (sk->sk_shutdown & RCV_SHUTDOWN)
1868                                 goto unlock;
1869
1870                         unix_state_unlock(sk);
1871                         err = -EAGAIN;
1872                         if (!timeo)
1873                                 break;
1874                         mutex_unlock(&u->readlock);
1875
1876                         timeo = unix_stream_data_wait(sk, timeo);
1877
1878                         if (signal_pending(current)) {
1879                                 err = sock_intr_errno(timeo);
1880                                 goto out;
1881                         }
1882                         mutex_lock(&u->readlock);
1883                         continue;
1884  unlock:
1885                         unix_state_unlock(sk);
1886                         break;
1887                 }
1888                 unix_state_unlock(sk);
1889
1890                 if (check_creds) {
1891                         /* Never glue messages from different writers */
1892                         if (memcmp(UNIXCREDS(skb), &siocb->scm->creds,
1893                                    sizeof(siocb->scm->creds)) != 0) {
1894                                 skb_queue_head(&sk->sk_receive_queue, skb);
1895                                 break;
1896                         }
1897                 } else {
1898                         /* Copy credentials */
1899                         siocb->scm->creds = *UNIXCREDS(skb);
1900                         check_creds = 1;
1901                 }
1902
1903                 /* Copy address just once */
1904                 if (sunaddr) {
1905                         unix_copy_addr(msg, skb->sk);
1906                         sunaddr = NULL;
1907                 }
1908
1909                 chunk = min_t(unsigned int, skb->len, size);
1910                 if (memcpy_toiovec(msg->msg_iov, skb->data, chunk)) {
1911                         skb_queue_head(&sk->sk_receive_queue, skb);
1912                         if (copied == 0)
1913                                 copied = -EFAULT;
1914                         break;
1915                 }
1916                 copied += chunk;
1917                 size -= chunk;
1918
1919                 /* Mark read part of skb as used */
1920                 if (!(flags & MSG_PEEK)) {
1921                         skb_pull(skb, chunk);
1922
1923                         if (UNIXCB(skb).fp)
1924                                 unix_detach_fds(siocb->scm, skb);
1925
1926                         /* put the skb back if we didn't use it up.. */
1927                         if (skb->len) {
1928                                 skb_queue_head(&sk->sk_receive_queue, skb);
1929                                 break;
1930                         }
1931
1932                         kfree_skb(skb);
1933
1934                         if (siocb->scm->fp)
1935                                 break;
1936                 } else {
1937                         /* It is questionable, see note in unix_dgram_recvmsg.
1938                          */
1939                         if (UNIXCB(skb).fp)
1940                                 siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1941
1942                         /* put message back and return */
1943                         skb_queue_head(&sk->sk_receive_queue, skb);
1944                         break;
1945                 }
1946         } while (size);
1947
1948         mutex_unlock(&u->readlock);
1949         scm_recv(sock, msg, siocb->scm, flags);
1950 out:
1951         return copied ? : err;
1952 }
1953
1954 static int unix_shutdown(struct socket *sock, int mode)
1955 {
1956         struct sock *sk = sock->sk;
1957         struct sock *other;
1958
1959         mode = (mode+1)&(RCV_SHUTDOWN|SEND_SHUTDOWN);
1960
1961         if (mode) {
1962                 unix_state_lock(sk);
1963                 sk->sk_shutdown |= mode;
1964                 other = unix_peer(sk);
1965                 if (other)
1966                         sock_hold(other);
1967                 unix_state_unlock(sk);
1968                 sk->sk_state_change(sk);
1969
1970                 if (other &&
1971                         (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET)) {
1972
1973                         int peer_mode = 0;
1974
1975                         if (mode&RCV_SHUTDOWN)
1976                                 peer_mode |= SEND_SHUTDOWN;
1977                         if (mode&SEND_SHUTDOWN)
1978                                 peer_mode |= RCV_SHUTDOWN;
1979                         unix_state_lock(other);
1980                         other->sk_shutdown |= peer_mode;
1981                         unix_state_unlock(other);
1982                         other->sk_state_change(other);
1983                         read_lock(&other->sk_callback_lock);
1984                         if (peer_mode == SHUTDOWN_MASK)
1985                                 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_HUP);
1986                         else if (peer_mode & RCV_SHUTDOWN)
1987                                 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_IN);
1988                         read_unlock(&other->sk_callback_lock);
1989                 }
1990                 if (other)
1991                         sock_put(other);
1992         }
1993         return 0;
1994 }
1995
1996 static int unix_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1997 {
1998         struct sock *sk = sock->sk;
1999         long amount = 0;
2000         int err;
2001
2002         switch (cmd) {
2003         case SIOCOUTQ:
2004                 amount = sk_wmem_alloc_get(sk);
2005                 err = put_user(amount, (int __user *)arg);
2006                 break;
2007         case SIOCINQ:
2008                 {
2009                         struct sk_buff *skb;
2010
2011                         if (sk->sk_state == TCP_LISTEN) {
2012                                 err = -EINVAL;
2013                                 break;
2014                         }
2015
2016                         spin_lock(&sk->sk_receive_queue.lock);
2017                         if (sk->sk_type == SOCK_STREAM ||
2018                             sk->sk_type == SOCK_SEQPACKET) {
2019                                 skb_queue_walk(&sk->sk_receive_queue, skb)
2020                                         amount += skb->len;
2021                         } else {
2022                                 skb = skb_peek(&sk->sk_receive_queue);
2023                                 if (skb)
2024                                         amount = skb->len;
2025                         }
2026                         spin_unlock(&sk->sk_receive_queue.lock);
2027                         err = put_user(amount, (int __user *)arg);
2028                         break;
2029                 }
2030
2031         default:
2032                 err = -ENOIOCTLCMD;
2033                 break;
2034         }
2035         return err;
2036 }
2037
2038 static unsigned int unix_poll(struct file *file, struct socket *sock, poll_table *wait)
2039 {
2040         struct sock *sk = sock->sk;
2041         unsigned int mask;
2042
2043         sock_poll_wait(file, sk->sk_sleep, wait);
2044         mask = 0;
2045
2046         /* exceptional events? */
2047         if (sk->sk_err)
2048                 mask |= POLLERR;
2049         if (sk->sk_shutdown == SHUTDOWN_MASK)
2050                 mask |= POLLHUP;
2051         if (sk->sk_shutdown & RCV_SHUTDOWN)
2052                 mask |= POLLRDHUP;
2053
2054         /* readable? */
2055         if (!skb_queue_empty(&sk->sk_receive_queue) ||
2056             (sk->sk_shutdown & RCV_SHUTDOWN))
2057                 mask |= POLLIN | POLLRDNORM;
2058
2059         /* Connection-based need to check for termination and startup */
2060         if ((sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) &&
2061             sk->sk_state == TCP_CLOSE)
2062                 mask |= POLLHUP;
2063
2064         /*
2065          * we set writable also when the other side has shut down the
2066          * connection. This prevents stuck sockets.
2067          */
2068         if (unix_writable(sk))
2069                 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2070
2071         return mask;
2072 }
2073
2074 static unsigned int unix_dgram_poll(struct file *file, struct socket *sock,
2075                                     poll_table *wait)
2076 {
2077         struct sock *sk = sock->sk, *other;
2078         unsigned int mask, writable;
2079
2080         sock_poll_wait(file, sk->sk_sleep, wait);
2081         mask = 0;
2082
2083         /* exceptional events? */
2084         if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue))
2085                 mask |= POLLERR;
2086         if (sk->sk_shutdown & RCV_SHUTDOWN)
2087                 mask |= POLLRDHUP;
2088         if (sk->sk_shutdown == SHUTDOWN_MASK)
2089                 mask |= POLLHUP;
2090
2091         /* readable? */
2092         if (!skb_queue_empty(&sk->sk_receive_queue) ||
2093             (sk->sk_shutdown & RCV_SHUTDOWN))
2094                 mask |= POLLIN | POLLRDNORM;
2095
2096         /* Connection-based need to check for termination and startup */
2097         if (sk->sk_type == SOCK_SEQPACKET) {
2098                 if (sk->sk_state == TCP_CLOSE)
2099                         mask |= POLLHUP;
2100                 /* connection hasn't started yet? */
2101                 if (sk->sk_state == TCP_SYN_SENT)
2102                         return mask;
2103         }
2104
2105         /* writable? */
2106         writable = unix_writable(sk);
2107         if (writable) {
2108                 other = unix_peer_get(sk);
2109                 if (other) {
2110                         if (unix_peer(other) != sk) {
2111                                 sock_poll_wait(file, &unix_sk(other)->peer_wait,
2112                                           wait);
2113                                 if (unix_recvq_full(other))
2114                                         writable = 0;
2115                         }
2116
2117                         sock_put(other);
2118                 }
2119         }
2120
2121         if (writable)
2122                 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
2123         else
2124                 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
2125
2126         return mask;
2127 }
2128
2129 #ifdef CONFIG_PROC_FS
2130 static struct sock *first_unix_socket(int *i)
2131 {
2132         for (*i = 0; *i <= UNIX_HASH_SIZE; (*i)++) {
2133                 if (!hlist_empty(&unix_socket_table[*i]))
2134                         return __sk_head(&unix_socket_table[*i]);
2135         }
2136         return NULL;
2137 }
2138
2139 static struct sock *next_unix_socket(int *i, struct sock *s)
2140 {
2141         struct sock *next = sk_next(s);
2142         /* More in this chain? */
2143         if (next)
2144                 return next;
2145         /* Look for next non-empty chain. */
2146         for ((*i)++; *i <= UNIX_HASH_SIZE; (*i)++) {
2147                 if (!hlist_empty(&unix_socket_table[*i]))
2148                         return __sk_head(&unix_socket_table[*i]);
2149         }
2150         return NULL;
2151 }
2152
2153 struct unix_iter_state {
2154         struct seq_net_private p;
2155         int i;
2156 };
2157
2158 static struct sock *unix_seq_idx(struct seq_file *seq, loff_t pos)
2159 {
2160         struct unix_iter_state *iter = seq->private;
2161         loff_t off = 0;
2162         struct sock *s;
2163
2164         for (s = first_unix_socket(&iter->i); s; s = next_unix_socket(&iter->i, s)) {
2165                 if (sock_net(s) != seq_file_net(seq))
2166                         continue;
2167                 if (off == pos)
2168                         return s;
2169                 ++off;
2170         }
2171         return NULL;
2172 }
2173
2174 static void *unix_seq_start(struct seq_file *seq, loff_t *pos)
2175         __acquires(unix_table_lock)
2176 {
2177         spin_lock(&unix_table_lock);
2178         return *pos ? unix_seq_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2179 }
2180
2181 static void *unix_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2182 {
2183         struct unix_iter_state *iter = seq->private;
2184         struct sock *sk = v;
2185         ++*pos;
2186
2187         if (v == SEQ_START_TOKEN)
2188                 sk = first_unix_socket(&iter->i);
2189         else
2190                 sk = next_unix_socket(&iter->i, sk);
2191         while (sk && (sock_net(sk) != seq_file_net(seq)))
2192                 sk = next_unix_socket(&iter->i, sk);
2193         return sk;
2194 }
2195
2196 static void unix_seq_stop(struct seq_file *seq, void *v)
2197         __releases(unix_table_lock)
2198 {
2199         spin_unlock(&unix_table_lock);
2200 }
2201
2202 static int unix_seq_show(struct seq_file *seq, void *v)
2203 {
2204
2205         if (v == SEQ_START_TOKEN)
2206                 seq_puts(seq, "Num       RefCount Protocol Flags    Type St "
2207                          "Inode Path\n");
2208         else {
2209                 struct sock *s = v;
2210                 struct unix_sock *u = unix_sk(s);
2211                 unix_state_lock(s);
2212
2213                 seq_printf(seq, "%p: %08X %08X %08X %04X %02X %5lu",
2214                         s,
2215                         atomic_read(&s->sk_refcnt),
2216                         0,
2217                         s->sk_state == TCP_LISTEN ? __SO_ACCEPTCON : 0,
2218                         s->sk_type,
2219                         s->sk_socket ?
2220                         (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTED : SS_UNCONNECTED) :
2221                         (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTING : SS_DISCONNECTING),
2222                         sock_i_ino(s));
2223
2224                 if (u->addr) {
2225                         int i, len;
2226                         seq_putc(seq, ' ');
2227
2228                         i = 0;
2229                         len = u->addr->len - sizeof(short);
2230                         if (!UNIX_ABSTRACT(s))
2231                                 len--;
2232                         else {
2233                                 seq_putc(seq, '@');
2234                                 i++;
2235                         }
2236                         for ( ; i < len; i++)
2237                                 seq_putc(seq, u->addr->name->sun_path[i]);
2238                 }
2239                 unix_state_unlock(s);
2240                 seq_putc(seq, '\n');
2241         }
2242
2243         return 0;
2244 }
2245
2246 static const struct seq_operations unix_seq_ops = {
2247         .start  = unix_seq_start,
2248         .next   = unix_seq_next,
2249         .stop   = unix_seq_stop,
2250         .show   = unix_seq_show,
2251 };
2252
2253 static int unix_seq_open(struct inode *inode, struct file *file)
2254 {
2255         return seq_open_net(inode, file, &unix_seq_ops,
2256                             sizeof(struct unix_iter_state));
2257 }
2258
2259 static const struct file_operations unix_seq_fops = {
2260         .owner          = THIS_MODULE,
2261         .open           = unix_seq_open,
2262         .read           = seq_read,
2263         .llseek         = seq_lseek,
2264         .release        = seq_release_net,
2265 };
2266
2267 #endif
2268
2269 static struct net_proto_family unix_family_ops = {
2270         .family = PF_UNIX,
2271         .create = unix_create,
2272         .owner  = THIS_MODULE,
2273 };
2274
2275
2276 static int unix_net_init(struct net *net)
2277 {
2278         int error = -ENOMEM;
2279
2280         net->unx.sysctl_max_dgram_qlen = 10;
2281         if (unix_sysctl_register(net))
2282                 goto out;
2283
2284 #ifdef CONFIG_PROC_FS
2285         if (!proc_net_fops_create(net, "unix", 0, &unix_seq_fops)) {
2286                 unix_sysctl_unregister(net);
2287                 goto out;
2288         }
2289 #endif
2290         error = 0;
2291 out:
2292         return error;
2293 }
2294
2295 static void unix_net_exit(struct net *net)
2296 {
2297         unix_sysctl_unregister(net);
2298         proc_net_remove(net, "unix");
2299 }
2300
2301 static struct pernet_operations unix_net_ops = {
2302         .init = unix_net_init,
2303         .exit = unix_net_exit,
2304 };
2305
2306 static int __init af_unix_init(void)
2307 {
2308         int rc = -1;
2309         struct sk_buff *dummy_skb;
2310
2311         BUILD_BUG_ON(sizeof(struct unix_skb_parms) > sizeof(dummy_skb->cb));
2312
2313         rc = proto_register(&unix_proto, 1);
2314         if (rc != 0) {
2315                 printk(KERN_CRIT "%s: Cannot create unix_sock SLAB cache!\n",
2316                        __func__);
2317                 goto out;
2318         }
2319
2320         sock_register(&unix_family_ops);
2321         register_pernet_subsys(&unix_net_ops);
2322 out:
2323         return rc;
2324 }
2325
2326 static void __exit af_unix_exit(void)
2327 {
2328         sock_unregister(PF_UNIX);
2329         proto_unregister(&unix_proto);
2330         unregister_pernet_subsys(&unix_net_ops);
2331 }
2332
2333 /* Earlier than device_initcall() so that other drivers invoking
2334    request_module() don't end up in a loop when modprobe tries
2335    to use a UNIX socket. But later than subsys_initcall() because
2336    we depend on stuff initialised there */
2337 fs_initcall(af_unix_init);
2338 module_exit(af_unix_exit);
2339
2340 MODULE_LICENSE("GPL");
2341 MODULE_ALIAS_NETPROTO(PF_UNIX);