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 struct hlist_head unix_socket_table[2 * UNIX_HASH_SIZE]; 119 EXPORT_SYMBOL_GPL(unix_socket_table); 120 DEFINE_SPINLOCK(unix_table_lock); 121 EXPORT_SYMBOL_GPL(unix_table_lock); 122 static atomic_long_t unix_nr_socks; 123 124 125 static struct hlist_head *unix_sockets_unbound(void *addr) 126 { 127 unsigned long hash = (unsigned long)addr; 128 129 hash ^= hash >> 16; 130 hash ^= hash >> 8; 131 hash %= UNIX_HASH_SIZE; 132 return &unix_socket_table[UNIX_HASH_SIZE + hash]; 133 } 134 135 #define UNIX_ABSTRACT(sk) (unix_sk(sk)->addr->hash < UNIX_HASH_SIZE) 136 137 #ifdef CONFIG_SECURITY_NETWORK 138 static void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb) 139 { 140 memcpy(UNIXSID(skb), &scm->secid, sizeof(u32)); 141 } 142 143 static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb) 144 { 145 scm->secid = *UNIXSID(skb); 146 } 147 #else 148 static inline void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb) 149 { } 150 151 static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb) 152 { } 153 #endif /* CONFIG_SECURITY_NETWORK */ 154 155 /* 156 * SMP locking strategy: 157 * hash table is protected with spinlock unix_table_lock 158 * each socket state is protected by separate spin lock. 159 */ 160 161 static inline unsigned int unix_hash_fold(__wsum n) 162 { 163 unsigned int hash = (__force unsigned int)n; 164 165 hash ^= hash>>16; 166 hash ^= hash>>8; 167 return hash&(UNIX_HASH_SIZE-1); 168 } 169 170 #define unix_peer(sk) (unix_sk(sk)->peer) 171 172 static inline int unix_our_peer(struct sock *sk, struct sock *osk) 173 { 174 return unix_peer(osk) == sk; 175 } 176 177 static inline int unix_may_send(struct sock *sk, struct sock *osk) 178 { 179 return unix_peer(osk) == NULL || unix_our_peer(sk, osk); 180 } 181 182 static inline int unix_recvq_full(struct sock const *sk) 183 { 184 return skb_queue_len(&sk->sk_receive_queue) > sk->sk_max_ack_backlog; 185 } 186 187 struct sock *unix_peer_get(struct sock *s) 188 { 189 struct sock *peer; 190 191 unix_state_lock(s); 192 peer = unix_peer(s); 193 if (peer) 194 sock_hold(peer); 195 unix_state_unlock(s); 196 return peer; 197 } 198 EXPORT_SYMBOL_GPL(unix_peer_get); 199 200 static inline void unix_release_addr(struct unix_address *addr) 201 { 202 if (atomic_dec_and_test(&addr->refcnt)) 203 kfree(addr); 204 } 205 206 /* 207 * Check unix socket name: 208 * - should be not zero length. 209 * - if started by not zero, should be NULL terminated (FS object) 210 * - if started by zero, it is abstract name. 211 */ 212 213 static int unix_mkname(struct sockaddr_un *sunaddr, int len, unsigned int *hashp) 214 { 215 if (len <= sizeof(short) || len > sizeof(*sunaddr)) 216 return -EINVAL; 217 if (!sunaddr || sunaddr->sun_family != AF_UNIX) 218 return -EINVAL; 219 if (sunaddr->sun_path[0]) { 220 /* 221 * This may look like an off by one error but it is a bit more 222 * subtle. 108 is the longest valid AF_UNIX path for a binding. 223 * sun_path[108] doesn't as such exist. However in kernel space 224 * we are guaranteed that it is a valid memory location in our 225 * kernel address buffer. 226 */ 227 ((char *)sunaddr)[len] = 0; 228 len = strlen(sunaddr->sun_path)+1+sizeof(short); 229 return len; 230 } 231 232 *hashp = unix_hash_fold(csum_partial(sunaddr, len, 0)); 233 return len; 234 } 235 236 static void __unix_remove_socket(struct sock *sk) 237 { 238 sk_del_node_init(sk); 239 } 240 241 static void __unix_insert_socket(struct hlist_head *list, struct sock *sk) 242 { 243 WARN_ON(!sk_unhashed(sk)); 244 sk_add_node(sk, list); 245 } 246 247 static inline void unix_remove_socket(struct sock *sk) 248 { 249 spin_lock(&unix_table_lock); 250 __unix_remove_socket(sk); 251 spin_unlock(&unix_table_lock); 252 } 253 254 static inline void unix_insert_socket(struct hlist_head *list, struct sock *sk) 255 { 256 spin_lock(&unix_table_lock); 257 __unix_insert_socket(list, sk); 258 spin_unlock(&unix_table_lock); 259 } 260 261 static struct sock *__unix_find_socket_byname(struct net *net, 262 struct sockaddr_un *sunname, 263 int len, int type, unsigned int hash) 264 { 265 struct sock *s; 266 267 sk_for_each(s, &unix_socket_table[hash ^ type]) { 268 struct unix_sock *u = unix_sk(s); 269 270 if (!net_eq(sock_net(s), net)) 271 continue; 272 273 if (u->addr->len == len && 274 !memcmp(u->addr->name, sunname, len)) 275 goto found; 276 } 277 s = NULL; 278 found: 279 return s; 280 } 281 282 static inline struct sock *unix_find_socket_byname(struct net *net, 283 struct sockaddr_un *sunname, 284 int len, int type, 285 unsigned int hash) 286 { 287 struct sock *s; 288 289 spin_lock(&unix_table_lock); 290 s = __unix_find_socket_byname(net, sunname, len, type, hash); 291 if (s) 292 sock_hold(s); 293 spin_unlock(&unix_table_lock); 294 return s; 295 } 296 297 static struct sock *unix_find_socket_byinode(struct inode *i) 298 { 299 struct sock *s; 300 301 spin_lock(&unix_table_lock); 302 sk_for_each(s, 303 &unix_socket_table[i->i_ino & (UNIX_HASH_SIZE - 1)]) { 304 struct dentry *dentry = unix_sk(s)->path.dentry; 305 306 if (dentry && dentry->d_inode == i) { 307 sock_hold(s); 308 goto found; 309 } 310 } 311 s = NULL; 312 found: 313 spin_unlock(&unix_table_lock); 314 return s; 315 } 316 317 static inline int unix_writable(struct sock *sk) 318 { 319 return (atomic_read(&sk->sk_wmem_alloc) << 2) <= sk->sk_sndbuf; 320 } 321 322 static void unix_write_space(struct sock *sk) 323 { 324 struct socket_wq *wq; 325 326 rcu_read_lock(); 327 if (unix_writable(sk)) { 328 wq = rcu_dereference(sk->sk_wq); 329 if (wq_has_sleeper(wq)) 330 wake_up_interruptible_sync_poll(&wq->wait, 331 POLLOUT | POLLWRNORM | POLLWRBAND); 332 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT); 333 } 334 rcu_read_unlock(); 335 } 336 337 /* When dgram socket disconnects (or changes its peer), we clear its receive 338 * queue of packets arrived from previous peer. First, it allows to do 339 * flow control based only on wmem_alloc; second, sk connected to peer 340 * may receive messages only from that peer. */ 341 static void unix_dgram_disconnected(struct sock *sk, struct sock *other) 342 { 343 if (!skb_queue_empty(&sk->sk_receive_queue)) { 344 skb_queue_purge(&sk->sk_receive_queue); 345 wake_up_interruptible_all(&unix_sk(sk)->peer_wait); 346 347 /* If one link of bidirectional dgram pipe is disconnected, 348 * we signal error. Messages are lost. Do not make this, 349 * when peer was not connected to us. 350 */ 351 if (!sock_flag(other, SOCK_DEAD) && unix_peer(other) == sk) { 352 other->sk_err = ECONNRESET; 353 other->sk_error_report(other); 354 } 355 } 356 } 357 358 static void unix_sock_destructor(struct sock *sk) 359 { 360 struct unix_sock *u = unix_sk(sk); 361 362 skb_queue_purge(&sk->sk_receive_queue); 363 364 WARN_ON(atomic_read(&sk->sk_wmem_alloc)); 365 WARN_ON(!sk_unhashed(sk)); 366 WARN_ON(sk->sk_socket); 367 if (!sock_flag(sk, SOCK_DEAD)) { 368 printk(KERN_INFO "Attempt to release alive unix socket: %p\n", sk); 369 return; 370 } 371 372 if (u->addr) 373 unix_release_addr(u->addr); 374 375 atomic_long_dec(&unix_nr_socks); 376 local_bh_disable(); 377 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1); 378 local_bh_enable(); 379 #ifdef UNIX_REFCNT_DEBUG 380 printk(KERN_DEBUG "UNIX %p is destroyed, %ld are still alive.\n", sk, 381 atomic_long_read(&unix_nr_socks)); 382 #endif 383 } 384 385 static void unix_release_sock(struct sock *sk, int embrion) 386 { 387 struct unix_sock *u = unix_sk(sk); 388 struct path path; 389 struct sock *skpair; 390 struct sk_buff *skb; 391 int state; 392 393 unix_remove_socket(sk); 394 395 /* Clear state */ 396 unix_state_lock(sk); 397 sock_orphan(sk); 398 sk->sk_shutdown = SHUTDOWN_MASK; 399 path = u->path; 400 u->path.dentry = NULL; 401 u->path.mnt = NULL; 402 state = sk->sk_state; 403 sk->sk_state = TCP_CLOSE; 404 unix_state_unlock(sk); 405 406 wake_up_interruptible_all(&u->peer_wait); 407 408 skpair = unix_peer(sk); 409 410 if (skpair != NULL) { 411 if (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) { 412 unix_state_lock(skpair); 413 /* No more writes */ 414 skpair->sk_shutdown = SHUTDOWN_MASK; 415 if (!skb_queue_empty(&sk->sk_receive_queue) || embrion) 416 skpair->sk_err = ECONNRESET; 417 unix_state_unlock(skpair); 418 skpair->sk_state_change(skpair); 419 sk_wake_async(skpair, SOCK_WAKE_WAITD, POLL_HUP); 420 } 421 sock_put(skpair); /* It may now die */ 422 unix_peer(sk) = NULL; 423 } 424 425 /* Try to flush out this socket. Throw out buffers at least */ 426 427 while ((skb = skb_dequeue(&sk->sk_receive_queue)) != NULL) { 428 if (state == TCP_LISTEN) 429 unix_release_sock(skb->sk, 1); 430 /* passed fds are erased in the kfree_skb hook */ 431 kfree_skb(skb); 432 } 433 434 if (path.dentry) 435 path_put(&path); 436 437 sock_put(sk); 438 439 /* ---- Socket is dead now and most probably destroyed ---- */ 440 441 /* 442 * Fixme: BSD difference: In BSD all sockets connected to us get 443 * ECONNRESET and we die on the spot. In Linux we behave 444 * like files and pipes do and wait for the last 445 * dereference. 446 * 447 * Can't we simply set sock->err? 448 * 449 * What the above comment does talk about? --ANK(980817) 450 */ 451 452 if (unix_tot_inflight) 453 unix_gc(); /* Garbage collect fds */ 454 } 455 456 static void init_peercred(struct sock *sk) 457 { 458 put_pid(sk->sk_peer_pid); 459 if (sk->sk_peer_cred) 460 put_cred(sk->sk_peer_cred); 461 sk->sk_peer_pid = get_pid(task_tgid(current)); 462 sk->sk_peer_cred = get_current_cred(); 463 } 464 465 static void copy_peercred(struct sock *sk, struct sock *peersk) 466 { 467 put_pid(sk->sk_peer_pid); 468 if (sk->sk_peer_cred) 469 put_cred(sk->sk_peer_cred); 470 sk->sk_peer_pid = get_pid(peersk->sk_peer_pid); 471 sk->sk_peer_cred = get_cred(peersk->sk_peer_cred); 472 } 473 474 static int unix_listen(struct socket *sock, int backlog) 475 { 476 int err; 477 struct sock *sk = sock->sk; 478 struct unix_sock *u = unix_sk(sk); 479 struct pid *old_pid = NULL; 480 481 err = -EOPNOTSUPP; 482 if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET) 483 goto out; /* Only stream/seqpacket sockets accept */ 484 err = -EINVAL; 485 if (!u->addr) 486 goto out; /* No listens on an unbound socket */ 487 unix_state_lock(sk); 488 if (sk->sk_state != TCP_CLOSE && sk->sk_state != TCP_LISTEN) 489 goto out_unlock; 490 if (backlog > sk->sk_max_ack_backlog) 491 wake_up_interruptible_all(&u->peer_wait); 492 sk->sk_max_ack_backlog = backlog; 493 sk->sk_state = TCP_LISTEN; 494 /* set credentials so connect can copy them */ 495 init_peercred(sk); 496 err = 0; 497 498 out_unlock: 499 unix_state_unlock(sk); 500 put_pid(old_pid); 501 out: 502 return err; 503 } 504 505 static int unix_release(struct socket *); 506 static int unix_bind(struct socket *, struct sockaddr *, int); 507 static int unix_stream_connect(struct socket *, struct sockaddr *, 508 int addr_len, int flags); 509 static int unix_socketpair(struct socket *, struct socket *); 510 static int unix_accept(struct socket *, struct socket *, int); 511 static int unix_getname(struct socket *, struct sockaddr *, int *, int); 512 static unsigned int unix_poll(struct file *, struct socket *, poll_table *); 513 static unsigned int unix_dgram_poll(struct file *, struct socket *, 514 poll_table *); 515 static int unix_ioctl(struct socket *, unsigned int, unsigned long); 516 static int unix_shutdown(struct socket *, int); 517 static int unix_stream_sendmsg(struct kiocb *, struct socket *, 518 struct msghdr *, size_t); 519 static int unix_stream_recvmsg(struct kiocb *, struct socket *, 520 struct msghdr *, size_t, int); 521 static int unix_dgram_sendmsg(struct kiocb *, struct socket *, 522 struct msghdr *, size_t); 523 static int unix_dgram_recvmsg(struct kiocb *, struct socket *, 524 struct msghdr *, size_t, int); 525 static int unix_dgram_connect(struct socket *, struct sockaddr *, 526 int, int); 527 static int unix_seqpacket_sendmsg(struct kiocb *, struct socket *, 528 struct msghdr *, size_t); 529 static int unix_seqpacket_recvmsg(struct kiocb *, struct socket *, 530 struct msghdr *, size_t, int); 531 532 static void unix_set_peek_off(struct sock *sk, int val) 533 { 534 struct unix_sock *u = unix_sk(sk); 535 536 mutex_lock(&u->readlock); 537 sk->sk_peek_off = val; 538 mutex_unlock(&u->readlock); 539 } 540 541 542 static const struct proto_ops unix_stream_ops = { 543 .family = PF_UNIX, 544 .owner = THIS_MODULE, 545 .release = unix_release, 546 .bind = unix_bind, 547 .connect = unix_stream_connect, 548 .socketpair = unix_socketpair, 549 .accept = unix_accept, 550 .getname = unix_getname, 551 .poll = unix_poll, 552 .ioctl = unix_ioctl, 553 .listen = unix_listen, 554 .shutdown = unix_shutdown, 555 .setsockopt = sock_no_setsockopt, 556 .getsockopt = sock_no_getsockopt, 557 .sendmsg = unix_stream_sendmsg, 558 .recvmsg = unix_stream_recvmsg, 559 .mmap = sock_no_mmap, 560 .sendpage = sock_no_sendpage, 561 .set_peek_off = unix_set_peek_off, 562 }; 563 564 static const struct proto_ops unix_dgram_ops = { 565 .family = PF_UNIX, 566 .owner = THIS_MODULE, 567 .release = unix_release, 568 .bind = unix_bind, 569 .connect = unix_dgram_connect, 570 .socketpair = unix_socketpair, 571 .accept = sock_no_accept, 572 .getname = unix_getname, 573 .poll = unix_dgram_poll, 574 .ioctl = unix_ioctl, 575 .listen = sock_no_listen, 576 .shutdown = unix_shutdown, 577 .setsockopt = sock_no_setsockopt, 578 .getsockopt = sock_no_getsockopt, 579 .sendmsg = unix_dgram_sendmsg, 580 .recvmsg = unix_dgram_recvmsg, 581 .mmap = sock_no_mmap, 582 .sendpage = sock_no_sendpage, 583 .set_peek_off = unix_set_peek_off, 584 }; 585 586 static const struct proto_ops unix_seqpacket_ops = { 587 .family = PF_UNIX, 588 .owner = THIS_MODULE, 589 .release = unix_release, 590 .bind = unix_bind, 591 .connect = unix_stream_connect, 592 .socketpair = unix_socketpair, 593 .accept = unix_accept, 594 .getname = unix_getname, 595 .poll = unix_dgram_poll, 596 .ioctl = unix_ioctl, 597 .listen = unix_listen, 598 .shutdown = unix_shutdown, 599 .setsockopt = sock_no_setsockopt, 600 .getsockopt = sock_no_getsockopt, 601 .sendmsg = unix_seqpacket_sendmsg, 602 .recvmsg = unix_seqpacket_recvmsg, 603 .mmap = sock_no_mmap, 604 .sendpage = sock_no_sendpage, 605 .set_peek_off = unix_set_peek_off, 606 }; 607 608 static struct proto unix_proto = { 609 .name = "UNIX", 610 .owner = THIS_MODULE, 611 .obj_size = sizeof(struct unix_sock), 612 }; 613 614 /* 615 * AF_UNIX sockets do not interact with hardware, hence they 616 * dont trigger interrupts - so it's safe for them to have 617 * bh-unsafe locking for their sk_receive_queue.lock. Split off 618 * this special lock-class by reinitializing the spinlock key: 619 */ 620 static struct lock_class_key af_unix_sk_receive_queue_lock_key; 621 622 static struct sock *unix_create1(struct net *net, struct socket *sock) 623 { 624 struct sock *sk = NULL; 625 struct unix_sock *u; 626 627 atomic_long_inc(&unix_nr_socks); 628 if (atomic_long_read(&unix_nr_socks) > 2 * get_max_files()) 629 goto out; 630 631 sk = sk_alloc(net, PF_UNIX, GFP_KERNEL, &unix_proto); 632 if (!sk) 633 goto out; 634 635 sock_init_data(sock, sk); 636 lockdep_set_class(&sk->sk_receive_queue.lock, 637 &af_unix_sk_receive_queue_lock_key); 638 639 sk->sk_write_space = unix_write_space; 640 sk->sk_max_ack_backlog = net->unx.sysctl_max_dgram_qlen; 641 sk->sk_destruct = unix_sock_destructor; 642 u = unix_sk(sk); 643 u->path.dentry = NULL; 644 u->path.mnt = NULL; 645 spin_lock_init(&u->lock); 646 atomic_long_set(&u->inflight, 0); 647 INIT_LIST_HEAD(&u->link); 648 mutex_init(&u->readlock); /* single task reading lock */ 649 init_waitqueue_head(&u->peer_wait); 650 unix_insert_socket(unix_sockets_unbound(sk), sk); 651 out: 652 if (sk == NULL) 653 atomic_long_dec(&unix_nr_socks); 654 else { 655 local_bh_disable(); 656 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); 657 local_bh_enable(); 658 } 659 return sk; 660 } 661 662 static int unix_create(struct net *net, struct socket *sock, int protocol, 663 int kern) 664 { 665 if (protocol && protocol != PF_UNIX) 666 return -EPROTONOSUPPORT; 667 668 sock->state = SS_UNCONNECTED; 669 670 switch (sock->type) { 671 case SOCK_STREAM: 672 sock->ops = &unix_stream_ops; 673 break; 674 /* 675 * Believe it or not BSD has AF_UNIX, SOCK_RAW though 676 * nothing uses it. 677 */ 678 case SOCK_RAW: 679 sock->type = SOCK_DGRAM; 680 case SOCK_DGRAM: 681 sock->ops = &unix_dgram_ops; 682 break; 683 case SOCK_SEQPACKET: 684 sock->ops = &unix_seqpacket_ops; 685 break; 686 default: 687 return -ESOCKTNOSUPPORT; 688 } 689 690 return unix_create1(net, sock) ? 0 : -ENOMEM; 691 } 692 693 static int unix_release(struct socket *sock) 694 { 695 struct sock *sk = sock->sk; 696 697 if (!sk) 698 return 0; 699 700 unix_release_sock(sk, 0); 701 sock->sk = NULL; 702 703 return 0; 704 } 705 706 static int unix_autobind(struct socket *sock) 707 { 708 struct sock *sk = sock->sk; 709 struct net *net = sock_net(sk); 710 struct unix_sock *u = unix_sk(sk); 711 static u32 ordernum = 1; 712 struct unix_address *addr; 713 int err; 714 unsigned int retries = 0; 715 716 mutex_lock(&u->readlock); 717 718 err = 0; 719 if (u->addr) 720 goto out; 721 722 err = -ENOMEM; 723 addr = kzalloc(sizeof(*addr) + sizeof(short) + 16, GFP_KERNEL); 724 if (!addr) 725 goto out; 726 727 addr->name->sun_family = AF_UNIX; 728 atomic_set(&addr->refcnt, 1); 729 730 retry: 731 addr->len = sprintf(addr->name->sun_path+1, "%05x", ordernum) + 1 + sizeof(short); 732 addr->hash = unix_hash_fold(csum_partial(addr->name, addr->len, 0)); 733 734 spin_lock(&unix_table_lock); 735 ordernum = (ordernum+1)&0xFFFFF; 736 737 if (__unix_find_socket_byname(net, addr->name, addr->len, sock->type, 738 addr->hash)) { 739 spin_unlock(&unix_table_lock); 740 /* 741 * __unix_find_socket_byname() may take long time if many names 742 * are already in use. 743 */ 744 cond_resched(); 745 /* Give up if all names seems to be in use. */ 746 if (retries++ == 0xFFFFF) { 747 err = -ENOSPC; 748 kfree(addr); 749 goto out; 750 } 751 goto retry; 752 } 753 addr->hash ^= sk->sk_type; 754 755 __unix_remove_socket(sk); 756 u->addr = addr; 757 __unix_insert_socket(&unix_socket_table[addr->hash], sk); 758 spin_unlock(&unix_table_lock); 759 err = 0; 760 761 out: mutex_unlock(&u->readlock); 762 return err; 763 } 764 765 static struct sock *unix_find_other(struct net *net, 766 struct sockaddr_un *sunname, int len, 767 int type, unsigned int hash, int *error) 768 { 769 struct sock *u; 770 struct path path; 771 int err = 0; 772 773 if (sunname->sun_path[0]) { 774 struct inode *inode; 775 err = kern_path(sunname->sun_path, LOOKUP_FOLLOW, &path); 776 if (err) 777 goto fail; 778 inode = path.dentry->d_inode; 779 err = inode_permission(inode, MAY_WRITE); 780 if (err) 781 goto put_fail; 782 783 err = -ECONNREFUSED; 784 if (!S_ISSOCK(inode->i_mode)) 785 goto put_fail; 786 u = unix_find_socket_byinode(inode); 787 if (!u) 788 goto put_fail; 789 790 if (u->sk_type == type) 791 touch_atime(&path); 792 793 path_put(&path); 794 795 err = -EPROTOTYPE; 796 if (u->sk_type != type) { 797 sock_put(u); 798 goto fail; 799 } 800 } else { 801 err = -ECONNREFUSED; 802 u = unix_find_socket_byname(net, sunname, len, type, hash); 803 if (u) { 804 struct dentry *dentry; 805 dentry = unix_sk(u)->path.dentry; 806 if (dentry) 807 touch_atime(&unix_sk(u)->path); 808 } else 809 goto fail; 810 } 811 return u; 812 813 put_fail: 814 path_put(&path); 815 fail: 816 *error = err; 817 return NULL; 818 } 819 820 static int unix_mknod(const char *sun_path, umode_t mode, struct path *res) 821 { 822 struct dentry *dentry; 823 struct path path; 824 int err = 0; 825 /* 826 * Get the parent directory, calculate the hash for last 827 * component. 828 */ 829 dentry = kern_path_create(AT_FDCWD, sun_path, &path, 0); 830 err = PTR_ERR(dentry); 831 if (IS_ERR(dentry)) 832 return err; 833 834 /* 835 * All right, let's create it. 836 */ 837 err = security_path_mknod(&path, dentry, mode, 0); 838 if (!err) { 839 err = vfs_mknod(path.dentry->d_inode, dentry, mode, 0); 840 if (!err) { 841 res->mnt = mntget(path.mnt); 842 res->dentry = dget(dentry); 843 } 844 } 845 done_path_create(&path, dentry); 846 return err; 847 } 848 849 static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len) 850 { 851 struct sock *sk = sock->sk; 852 struct net *net = sock_net(sk); 853 struct unix_sock *u = unix_sk(sk); 854 struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr; 855 char *sun_path = sunaddr->sun_path; 856 int err; 857 unsigned int hash; 858 struct unix_address *addr; 859 struct hlist_head *list; 860 861 err = -EINVAL; 862 if (sunaddr->sun_family != AF_UNIX) 863 goto out; 864 865 if (addr_len == sizeof(short)) { 866 err = unix_autobind(sock); 867 goto out; 868 } 869 870 err = unix_mkname(sunaddr, addr_len, &hash); 871 if (err < 0) 872 goto out; 873 addr_len = err; 874 875 mutex_lock(&u->readlock); 876 877 err = -EINVAL; 878 if (u->addr) 879 goto out_up; 880 881 err = -ENOMEM; 882 addr = kmalloc(sizeof(*addr)+addr_len, GFP_KERNEL); 883 if (!addr) 884 goto out_up; 885 886 memcpy(addr->name, sunaddr, addr_len); 887 addr->len = addr_len; 888 addr->hash = hash ^ sk->sk_type; 889 atomic_set(&addr->refcnt, 1); 890 891 if (sun_path[0]) { 892 struct path path; 893 umode_t mode = S_IFSOCK | 894 (SOCK_INODE(sock)->i_mode & ~current_umask()); 895 err = unix_mknod(sun_path, mode, &path); 896 if (err) { 897 if (err == -EEXIST) 898 err = -EADDRINUSE; 899 unix_release_addr(addr); 900 goto out_up; 901 } 902 addr->hash = UNIX_HASH_SIZE; 903 hash = path.dentry->d_inode->i_ino & (UNIX_HASH_SIZE-1); 904 spin_lock(&unix_table_lock); 905 u->path = path; 906 list = &unix_socket_table[hash]; 907 } else { 908 spin_lock(&unix_table_lock); 909 err = -EADDRINUSE; 910 if (__unix_find_socket_byname(net, sunaddr, addr_len, 911 sk->sk_type, hash)) { 912 unix_release_addr(addr); 913 goto out_unlock; 914 } 915 916 list = &unix_socket_table[addr->hash]; 917 } 918 919 err = 0; 920 __unix_remove_socket(sk); 921 u->addr = addr; 922 __unix_insert_socket(list, sk); 923 924 out_unlock: 925 spin_unlock(&unix_table_lock); 926 out_up: 927 mutex_unlock(&u->readlock); 928 out: 929 return err; 930 } 931 932 static void unix_state_double_lock(struct sock *sk1, struct sock *sk2) 933 { 934 if (unlikely(sk1 == sk2) || !sk2) { 935 unix_state_lock(sk1); 936 return; 937 } 938 if (sk1 < sk2) { 939 unix_state_lock(sk1); 940 unix_state_lock_nested(sk2); 941 } else { 942 unix_state_lock(sk2); 943 unix_state_lock_nested(sk1); 944 } 945 } 946 947 static void unix_state_double_unlock(struct sock *sk1, struct sock *sk2) 948 { 949 if (unlikely(sk1 == sk2) || !sk2) { 950 unix_state_unlock(sk1); 951 return; 952 } 953 unix_state_unlock(sk1); 954 unix_state_unlock(sk2); 955 } 956 957 static int unix_dgram_connect(struct socket *sock, struct sockaddr *addr, 958 int alen, int flags) 959 { 960 struct sock *sk = sock->sk; 961 struct net *net = sock_net(sk); 962 struct sockaddr_un *sunaddr = (struct sockaddr_un *)addr; 963 struct sock *other; 964 unsigned int hash; 965 int err; 966 967 if (addr->sa_family != AF_UNSPEC) { 968 err = unix_mkname(sunaddr, alen, &hash); 969 if (err < 0) 970 goto out; 971 alen = err; 972 973 if (test_bit(SOCK_PASSCRED, &sock->flags) && 974 !unix_sk(sk)->addr && (err = unix_autobind(sock)) != 0) 975 goto out; 976 977 restart: 978 other = unix_find_other(net, sunaddr, alen, sock->type, hash, &err); 979 if (!other) 980 goto out; 981 982 unix_state_double_lock(sk, other); 983 984 /* Apparently VFS overslept socket death. Retry. */ 985 if (sock_flag(other, SOCK_DEAD)) { 986 unix_state_double_unlock(sk, other); 987 sock_put(other); 988 goto restart; 989 } 990 991 err = -EPERM; 992 if (!unix_may_send(sk, other)) 993 goto out_unlock; 994 995 err = security_unix_may_send(sk->sk_socket, other->sk_socket); 996 if (err) 997 goto out_unlock; 998 999 } else { 1000 /* 1001 * 1003.1g breaking connected state with AF_UNSPEC 1002 */ 1003 other = NULL; 1004 unix_state_double_lock(sk, other); 1005 } 1006 1007 /* 1008 * If it was connected, reconnect. 1009 */ 1010 if (unix_peer(sk)) { 1011 struct sock *old_peer = unix_peer(sk); 1012 unix_peer(sk) = other; 1013 unix_state_double_unlock(sk, other); 1014 1015 if (other != old_peer) 1016 unix_dgram_disconnected(sk, old_peer); 1017 sock_put(old_peer); 1018 } else { 1019 unix_peer(sk) = other; 1020 unix_state_double_unlock(sk, other); 1021 } 1022 return 0; 1023 1024 out_unlock: 1025 unix_state_double_unlock(sk, other); 1026 sock_put(other); 1027 out: 1028 return err; 1029 } 1030 1031 static long unix_wait_for_peer(struct sock *other, long timeo) 1032 { 1033 struct unix_sock *u = unix_sk(other); 1034 int sched; 1035 DEFINE_WAIT(wait); 1036 1037 prepare_to_wait_exclusive(&u->peer_wait, &wait, TASK_INTERRUPTIBLE); 1038 1039 sched = !sock_flag(other, SOCK_DEAD) && 1040 !(other->sk_shutdown & RCV_SHUTDOWN) && 1041 unix_recvq_full(other); 1042 1043 unix_state_unlock(other); 1044 1045 if (sched) 1046 timeo = schedule_timeout(timeo); 1047 1048 finish_wait(&u->peer_wait, &wait); 1049 return timeo; 1050 } 1051 1052 static int unix_stream_connect(struct socket *sock, struct sockaddr *uaddr, 1053 int addr_len, int flags) 1054 { 1055 struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr; 1056 struct sock *sk = sock->sk; 1057 struct net *net = sock_net(sk); 1058 struct unix_sock *u = unix_sk(sk), *newu, *otheru; 1059 struct sock *newsk = NULL; 1060 struct sock *other = NULL; 1061 struct sk_buff *skb = NULL; 1062 unsigned int hash; 1063 int st; 1064 int err; 1065 long timeo; 1066 1067 err = unix_mkname(sunaddr, addr_len, &hash); 1068 if (err < 0) 1069 goto out; 1070 addr_len = err; 1071 1072 if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr && 1073 (err = unix_autobind(sock)) != 0) 1074 goto out; 1075 1076 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK); 1077 1078 /* First of all allocate resources. 1079 If we will make it after state is locked, 1080 we will have to recheck all again in any case. 1081 */ 1082 1083 err = -ENOMEM; 1084 1085 /* create new sock for complete connection */ 1086 newsk = unix_create1(sock_net(sk), NULL); 1087 if (newsk == NULL) 1088 goto out; 1089 1090 /* Allocate skb for sending to listening sock */ 1091 skb = sock_wmalloc(newsk, 1, 0, GFP_KERNEL); 1092 if (skb == NULL) 1093 goto out; 1094 1095 restart: 1096 /* Find listening sock. */ 1097 other = unix_find_other(net, sunaddr, addr_len, sk->sk_type, hash, &err); 1098 if (!other) 1099 goto out; 1100 1101 /* Latch state of peer */ 1102 unix_state_lock(other); 1103 1104 /* Apparently VFS overslept socket death. Retry. */ 1105 if (sock_flag(other, SOCK_DEAD)) { 1106 unix_state_unlock(other); 1107 sock_put(other); 1108 goto restart; 1109 } 1110 1111 err = -ECONNREFUSED; 1112 if (other->sk_state != TCP_LISTEN) 1113 goto out_unlock; 1114 if (other->sk_shutdown & RCV_SHUTDOWN) 1115 goto out_unlock; 1116 1117 if (unix_recvq_full(other)) { 1118 err = -EAGAIN; 1119 if (!timeo) 1120 goto out_unlock; 1121 1122 timeo = unix_wait_for_peer(other, timeo); 1123 1124 err = sock_intr_errno(timeo); 1125 if (signal_pending(current)) 1126 goto out; 1127 sock_put(other); 1128 goto restart; 1129 } 1130 1131 /* Latch our state. 1132 1133 It is tricky place. We need to grab our state lock and cannot 1134 drop lock on peer. It is dangerous because deadlock is 1135 possible. Connect to self case and simultaneous 1136 attempt to connect are eliminated by checking socket 1137 state. other is TCP_LISTEN, if sk is TCP_LISTEN we 1138 check this before attempt to grab lock. 1139 1140 Well, and we have to recheck the state after socket locked. 1141 */ 1142 st = sk->sk_state; 1143 1144 switch (st) { 1145 case TCP_CLOSE: 1146 /* This is ok... continue with connect */ 1147 break; 1148 case TCP_ESTABLISHED: 1149 /* Socket is already connected */ 1150 err = -EISCONN; 1151 goto out_unlock; 1152 default: 1153 err = -EINVAL; 1154 goto out_unlock; 1155 } 1156 1157 unix_state_lock_nested(sk); 1158 1159 if (sk->sk_state != st) { 1160 unix_state_unlock(sk); 1161 unix_state_unlock(other); 1162 sock_put(other); 1163 goto restart; 1164 } 1165 1166 err = security_unix_stream_connect(sk, other, newsk); 1167 if (err) { 1168 unix_state_unlock(sk); 1169 goto out_unlock; 1170 } 1171 1172 /* The way is open! Fastly set all the necessary fields... */ 1173 1174 sock_hold(sk); 1175 unix_peer(newsk) = sk; 1176 newsk->sk_state = TCP_ESTABLISHED; 1177 newsk->sk_type = sk->sk_type; 1178 init_peercred(newsk); 1179 newu = unix_sk(newsk); 1180 RCU_INIT_POINTER(newsk->sk_wq, &newu->peer_wq); 1181 otheru = unix_sk(other); 1182 1183 /* copy address information from listening to new sock*/ 1184 if (otheru->addr) { 1185 atomic_inc(&otheru->addr->refcnt); 1186 newu->addr = otheru->addr; 1187 } 1188 if (otheru->path.dentry) { 1189 path_get(&otheru->path); 1190 newu->path = otheru->path; 1191 } 1192 1193 /* Set credentials */ 1194 copy_peercred(sk, other); 1195 1196 sock->state = SS_CONNECTED; 1197 sk->sk_state = TCP_ESTABLISHED; 1198 sock_hold(newsk); 1199 1200 smp_mb__after_atomic_inc(); /* sock_hold() does an atomic_inc() */ 1201 unix_peer(sk) = newsk; 1202 1203 unix_state_unlock(sk); 1204 1205 /* take ten and and send info to listening sock */ 1206 spin_lock(&other->sk_receive_queue.lock); 1207 __skb_queue_tail(&other->sk_receive_queue, skb); 1208 spin_unlock(&other->sk_receive_queue.lock); 1209 unix_state_unlock(other); 1210 other->sk_data_ready(other, 0); 1211 sock_put(other); 1212 return 0; 1213 1214 out_unlock: 1215 if (other) 1216 unix_state_unlock(other); 1217 1218 out: 1219 kfree_skb(skb); 1220 if (newsk) 1221 unix_release_sock(newsk, 0); 1222 if (other) 1223 sock_put(other); 1224 return err; 1225 } 1226 1227 static int unix_socketpair(struct socket *socka, struct socket *sockb) 1228 { 1229 struct sock *ska = socka->sk, *skb = sockb->sk; 1230 1231 /* Join our sockets back to back */ 1232 sock_hold(ska); 1233 sock_hold(skb); 1234 unix_peer(ska) = skb; 1235 unix_peer(skb) = ska; 1236 init_peercred(ska); 1237 init_peercred(skb); 1238 1239 if (ska->sk_type != SOCK_DGRAM) { 1240 ska->sk_state = TCP_ESTABLISHED; 1241 skb->sk_state = TCP_ESTABLISHED; 1242 socka->state = SS_CONNECTED; 1243 sockb->state = SS_CONNECTED; 1244 } 1245 return 0; 1246 } 1247 1248 static int unix_accept(struct socket *sock, struct socket *newsock, int flags) 1249 { 1250 struct sock *sk = sock->sk; 1251 struct sock *tsk; 1252 struct sk_buff *skb; 1253 int err; 1254 1255 err = -EOPNOTSUPP; 1256 if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET) 1257 goto out; 1258 1259 err = -EINVAL; 1260 if (sk->sk_state != TCP_LISTEN) 1261 goto out; 1262 1263 /* If socket state is TCP_LISTEN it cannot change (for now...), 1264 * so that no locks are necessary. 1265 */ 1266 1267 skb = skb_recv_datagram(sk, 0, flags&O_NONBLOCK, &err); 1268 if (!skb) { 1269 /* This means receive shutdown. */ 1270 if (err == 0) 1271 err = -EINVAL; 1272 goto out; 1273 } 1274 1275 tsk = skb->sk; 1276 skb_free_datagram(sk, skb); 1277 wake_up_interruptible(&unix_sk(sk)->peer_wait); 1278 1279 /* attach accepted sock to socket */ 1280 unix_state_lock(tsk); 1281 newsock->state = SS_CONNECTED; 1282 sock_graft(tsk, newsock); 1283 unix_state_unlock(tsk); 1284 return 0; 1285 1286 out: 1287 return err; 1288 } 1289 1290 1291 static int unix_getname(struct socket *sock, struct sockaddr *uaddr, int *uaddr_len, int peer) 1292 { 1293 struct sock *sk = sock->sk; 1294 struct unix_sock *u; 1295 DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr, uaddr); 1296 int err = 0; 1297 1298 if (peer) { 1299 sk = unix_peer_get(sk); 1300 1301 err = -ENOTCONN; 1302 if (!sk) 1303 goto out; 1304 err = 0; 1305 } else { 1306 sock_hold(sk); 1307 } 1308 1309 u = unix_sk(sk); 1310 unix_state_lock(sk); 1311 if (!u->addr) { 1312 sunaddr->sun_family = AF_UNIX; 1313 sunaddr->sun_path[0] = 0; 1314 *uaddr_len = sizeof(short); 1315 } else { 1316 struct unix_address *addr = u->addr; 1317 1318 *uaddr_len = addr->len; 1319 memcpy(sunaddr, addr->name, *uaddr_len); 1320 } 1321 unix_state_unlock(sk); 1322 sock_put(sk); 1323 out: 1324 return err; 1325 } 1326 1327 static void unix_detach_fds(struct scm_cookie *scm, struct sk_buff *skb) 1328 { 1329 int i; 1330 1331 scm->fp = UNIXCB(skb).fp; 1332 UNIXCB(skb).fp = NULL; 1333 1334 for (i = scm->fp->count-1; i >= 0; i--) 1335 unix_notinflight(scm->fp->fp[i]); 1336 } 1337 1338 static void unix_destruct_scm(struct sk_buff *skb) 1339 { 1340 struct scm_cookie scm; 1341 memset(&scm, 0, sizeof(scm)); 1342 scm.pid = UNIXCB(skb).pid; 1343 scm.cred = UNIXCB(skb).cred; 1344 if (UNIXCB(skb).fp) 1345 unix_detach_fds(&scm, skb); 1346 1347 /* Alas, it calls VFS */ 1348 /* So fscking what? fput() had been SMP-safe since the last Summer */ 1349 scm_destroy(&scm); 1350 sock_wfree(skb); 1351 } 1352 1353 #define MAX_RECURSION_LEVEL 4 1354 1355 static int unix_attach_fds(struct scm_cookie *scm, struct sk_buff *skb) 1356 { 1357 int i; 1358 unsigned char max_level = 0; 1359 int unix_sock_count = 0; 1360 1361 for (i = scm->fp->count - 1; i >= 0; i--) { 1362 struct sock *sk = unix_get_socket(scm->fp->fp[i]); 1363 1364 if (sk) { 1365 unix_sock_count++; 1366 max_level = max(max_level, 1367 unix_sk(sk)->recursion_level); 1368 } 1369 } 1370 if (unlikely(max_level > MAX_RECURSION_LEVEL)) 1371 return -ETOOMANYREFS; 1372 1373 /* 1374 * Need to duplicate file references for the sake of garbage 1375 * collection. Otherwise a socket in the fps might become a 1376 * candidate for GC while the skb is not yet queued. 1377 */ 1378 UNIXCB(skb).fp = scm_fp_dup(scm->fp); 1379 if (!UNIXCB(skb).fp) 1380 return -ENOMEM; 1381 1382 if (unix_sock_count) { 1383 for (i = scm->fp->count - 1; i >= 0; i--) 1384 unix_inflight(scm->fp->fp[i]); 1385 } 1386 return max_level; 1387 } 1388 1389 static int unix_scm_to_skb(struct scm_cookie *scm, struct sk_buff *skb, bool send_fds) 1390 { 1391 int err = 0; 1392 1393 UNIXCB(skb).pid = get_pid(scm->pid); 1394 if (scm->cred) 1395 UNIXCB(skb).cred = get_cred(scm->cred); 1396 UNIXCB(skb).fp = NULL; 1397 if (scm->fp && send_fds) 1398 err = unix_attach_fds(scm, skb); 1399 1400 skb->destructor = unix_destruct_scm; 1401 return err; 1402 } 1403 1404 /* 1405 * Some apps rely on write() giving SCM_CREDENTIALS 1406 * We include credentials if source or destination socket 1407 * asserted SOCK_PASSCRED. 1408 */ 1409 static void maybe_add_creds(struct sk_buff *skb, const struct socket *sock, 1410 const struct sock *other) 1411 { 1412 if (UNIXCB(skb).cred) 1413 return; 1414 if (test_bit(SOCK_PASSCRED, &sock->flags) || 1415 !other->sk_socket || 1416 test_bit(SOCK_PASSCRED, &other->sk_socket->flags)) { 1417 UNIXCB(skb).pid = get_pid(task_tgid(current)); 1418 UNIXCB(skb).cred = get_current_cred(); 1419 } 1420 } 1421 1422 /* 1423 * Send AF_UNIX data. 1424 */ 1425 1426 static int unix_dgram_sendmsg(struct kiocb *kiocb, struct socket *sock, 1427 struct msghdr *msg, size_t len) 1428 { 1429 struct sock_iocb *siocb = kiocb_to_siocb(kiocb); 1430 struct sock *sk = sock->sk; 1431 struct net *net = sock_net(sk); 1432 struct unix_sock *u = unix_sk(sk); 1433 struct sockaddr_un *sunaddr = msg->msg_name; 1434 struct sock *other = NULL; 1435 int namelen = 0; /* fake GCC */ 1436 int err; 1437 unsigned int hash; 1438 struct sk_buff *skb; 1439 long timeo; 1440 struct scm_cookie tmp_scm; 1441 int max_level; 1442 int data_len = 0; 1443 1444 if (NULL == siocb->scm) 1445 siocb->scm = &tmp_scm; 1446 wait_for_unix_gc(); 1447 err = scm_send(sock, msg, siocb->scm, false); 1448 if (err < 0) 1449 return err; 1450 1451 err = -EOPNOTSUPP; 1452 if (msg->msg_flags&MSG_OOB) 1453 goto out; 1454 1455 if (msg->msg_namelen) { 1456 err = unix_mkname(sunaddr, msg->msg_namelen, &hash); 1457 if (err < 0) 1458 goto out; 1459 namelen = err; 1460 } else { 1461 sunaddr = NULL; 1462 err = -ENOTCONN; 1463 other = unix_peer_get(sk); 1464 if (!other) 1465 goto out; 1466 } 1467 1468 if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr 1469 && (err = unix_autobind(sock)) != 0) 1470 goto out; 1471 1472 err = -EMSGSIZE; 1473 if (len > sk->sk_sndbuf - 32) 1474 goto out; 1475 1476 if (len > SKB_MAX_ALLOC) 1477 data_len = min_t(size_t, 1478 len - SKB_MAX_ALLOC, 1479 MAX_SKB_FRAGS * PAGE_SIZE); 1480 1481 skb = sock_alloc_send_pskb(sk, len - data_len, data_len, 1482 msg->msg_flags & MSG_DONTWAIT, &err); 1483 if (skb == NULL) 1484 goto out; 1485 1486 err = unix_scm_to_skb(siocb->scm, skb, true); 1487 if (err < 0) 1488 goto out_free; 1489 max_level = err + 1; 1490 unix_get_secdata(siocb->scm, skb); 1491 1492 skb_put(skb, len - data_len); 1493 skb->data_len = data_len; 1494 skb->len = len; 1495 err = skb_copy_datagram_from_iovec(skb, 0, msg->msg_iov, 0, len); 1496 if (err) 1497 goto out_free; 1498 1499 timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT); 1500 1501 restart: 1502 if (!other) { 1503 err = -ECONNRESET; 1504 if (sunaddr == NULL) 1505 goto out_free; 1506 1507 other = unix_find_other(net, sunaddr, namelen, sk->sk_type, 1508 hash, &err); 1509 if (other == NULL) 1510 goto out_free; 1511 } 1512 1513 if (sk_filter(other, skb) < 0) { 1514 /* Toss the packet but do not return any error to the sender */ 1515 err = len; 1516 goto out_free; 1517 } 1518 1519 unix_state_lock(other); 1520 err = -EPERM; 1521 if (!unix_may_send(sk, other)) 1522 goto out_unlock; 1523 1524 if (sock_flag(other, SOCK_DEAD)) { 1525 /* 1526 * Check with 1003.1g - what should 1527 * datagram error 1528 */ 1529 unix_state_unlock(other); 1530 sock_put(other); 1531 1532 err = 0; 1533 unix_state_lock(sk); 1534 if (unix_peer(sk) == other) { 1535 unix_peer(sk) = NULL; 1536 unix_state_unlock(sk); 1537 1538 unix_dgram_disconnected(sk, other); 1539 sock_put(other); 1540 err = -ECONNREFUSED; 1541 } else { 1542 unix_state_unlock(sk); 1543 } 1544 1545 other = NULL; 1546 if (err) 1547 goto out_free; 1548 goto restart; 1549 } 1550 1551 err = -EPIPE; 1552 if (other->sk_shutdown & RCV_SHUTDOWN) 1553 goto out_unlock; 1554 1555 if (sk->sk_type != SOCK_SEQPACKET) { 1556 err = security_unix_may_send(sk->sk_socket, other->sk_socket); 1557 if (err) 1558 goto out_unlock; 1559 } 1560 1561 if (unix_peer(other) != sk && unix_recvq_full(other)) { 1562 if (!timeo) { 1563 err = -EAGAIN; 1564 goto out_unlock; 1565 } 1566 1567 timeo = unix_wait_for_peer(other, timeo); 1568 1569 err = sock_intr_errno(timeo); 1570 if (signal_pending(current)) 1571 goto out_free; 1572 1573 goto restart; 1574 } 1575 1576 if (sock_flag(other, SOCK_RCVTSTAMP)) 1577 __net_timestamp(skb); 1578 maybe_add_creds(skb, sock, other); 1579 skb_queue_tail(&other->sk_receive_queue, skb); 1580 if (max_level > unix_sk(other)->recursion_level) 1581 unix_sk(other)->recursion_level = max_level; 1582 unix_state_unlock(other); 1583 other->sk_data_ready(other, len); 1584 sock_put(other); 1585 scm_destroy(siocb->scm); 1586 return len; 1587 1588 out_unlock: 1589 unix_state_unlock(other); 1590 out_free: 1591 kfree_skb(skb); 1592 out: 1593 if (other) 1594 sock_put(other); 1595 scm_destroy(siocb->scm); 1596 return err; 1597 } 1598 1599 1600 static int unix_stream_sendmsg(struct kiocb *kiocb, struct socket *sock, 1601 struct msghdr *msg, size_t len) 1602 { 1603 struct sock_iocb *siocb = kiocb_to_siocb(kiocb); 1604 struct sock *sk = sock->sk; 1605 struct sock *other = NULL; 1606 int err, size; 1607 struct sk_buff *skb; 1608 int sent = 0; 1609 struct scm_cookie tmp_scm; 1610 bool fds_sent = false; 1611 int max_level; 1612 1613 if (NULL == siocb->scm) 1614 siocb->scm = &tmp_scm; 1615 wait_for_unix_gc(); 1616 err = scm_send(sock, msg, siocb->scm, false); 1617 if (err < 0) 1618 return err; 1619 1620 err = -EOPNOTSUPP; 1621 if (msg->msg_flags&MSG_OOB) 1622 goto out_err; 1623 1624 if (msg->msg_namelen) { 1625 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP; 1626 goto out_err; 1627 } else { 1628 err = -ENOTCONN; 1629 other = unix_peer(sk); 1630 if (!other) 1631 goto out_err; 1632 } 1633 1634 if (sk->sk_shutdown & SEND_SHUTDOWN) 1635 goto pipe_err; 1636 1637 while (sent < len) { 1638 /* 1639 * Optimisation for the fact that under 0.01% of X 1640 * messages typically need breaking up. 1641 */ 1642 1643 size = len-sent; 1644 1645 /* Keep two messages in the pipe so it schedules better */ 1646 if (size > ((sk->sk_sndbuf >> 1) - 64)) 1647 size = (sk->sk_sndbuf >> 1) - 64; 1648 1649 if (size > SKB_MAX_ALLOC) 1650 size = SKB_MAX_ALLOC; 1651 1652 /* 1653 * Grab a buffer 1654 */ 1655 1656 skb = sock_alloc_send_skb(sk, size, msg->msg_flags&MSG_DONTWAIT, 1657 &err); 1658 1659 if (skb == NULL) 1660 goto out_err; 1661 1662 /* 1663 * If you pass two values to the sock_alloc_send_skb 1664 * it tries to grab the large buffer with GFP_NOFS 1665 * (which can fail easily), and if it fails grab the 1666 * fallback size buffer which is under a page and will 1667 * succeed. [Alan] 1668 */ 1669 size = min_t(int, size, skb_tailroom(skb)); 1670 1671 1672 /* Only send the fds in the first buffer */ 1673 err = unix_scm_to_skb(siocb->scm, skb, !fds_sent); 1674 if (err < 0) { 1675 kfree_skb(skb); 1676 goto out_err; 1677 } 1678 max_level = err + 1; 1679 fds_sent = true; 1680 1681 err = memcpy_fromiovec(skb_put(skb, size), msg->msg_iov, size); 1682 if (err) { 1683 kfree_skb(skb); 1684 goto out_err; 1685 } 1686 1687 unix_state_lock(other); 1688 1689 if (sock_flag(other, SOCK_DEAD) || 1690 (other->sk_shutdown & RCV_SHUTDOWN)) 1691 goto pipe_err_free; 1692 1693 maybe_add_creds(skb, sock, other); 1694 skb_queue_tail(&other->sk_receive_queue, skb); 1695 if (max_level > unix_sk(other)->recursion_level) 1696 unix_sk(other)->recursion_level = max_level; 1697 unix_state_unlock(other); 1698 other->sk_data_ready(other, size); 1699 sent += size; 1700 } 1701 1702 scm_destroy(siocb->scm); 1703 siocb->scm = NULL; 1704 1705 return sent; 1706 1707 pipe_err_free: 1708 unix_state_unlock(other); 1709 kfree_skb(skb); 1710 pipe_err: 1711 if (sent == 0 && !(msg->msg_flags&MSG_NOSIGNAL)) 1712 send_sig(SIGPIPE, current, 0); 1713 err = -EPIPE; 1714 out_err: 1715 scm_destroy(siocb->scm); 1716 siocb->scm = NULL; 1717 return sent ? : err; 1718 } 1719 1720 static int unix_seqpacket_sendmsg(struct kiocb *kiocb, struct socket *sock, 1721 struct msghdr *msg, size_t len) 1722 { 1723 int err; 1724 struct sock *sk = sock->sk; 1725 1726 err = sock_error(sk); 1727 if (err) 1728 return err; 1729 1730 if (sk->sk_state != TCP_ESTABLISHED) 1731 return -ENOTCONN; 1732 1733 if (msg->msg_namelen) 1734 msg->msg_namelen = 0; 1735 1736 return unix_dgram_sendmsg(kiocb, sock, msg, len); 1737 } 1738 1739 static int unix_seqpacket_recvmsg(struct kiocb *iocb, struct socket *sock, 1740 struct msghdr *msg, size_t size, 1741 int flags) 1742 { 1743 struct sock *sk = sock->sk; 1744 1745 if (sk->sk_state != TCP_ESTABLISHED) 1746 return -ENOTCONN; 1747 1748 return unix_dgram_recvmsg(iocb, sock, msg, size, flags); 1749 } 1750 1751 static void unix_copy_addr(struct msghdr *msg, struct sock *sk) 1752 { 1753 struct unix_sock *u = unix_sk(sk); 1754 1755 msg->msg_namelen = 0; 1756 if (u->addr) { 1757 msg->msg_namelen = u->addr->len; 1758 memcpy(msg->msg_name, u->addr->name, u->addr->len); 1759 } 1760 } 1761 1762 static int unix_dgram_recvmsg(struct kiocb *iocb, struct socket *sock, 1763 struct msghdr *msg, size_t size, 1764 int flags) 1765 { 1766 struct sock_iocb *siocb = kiocb_to_siocb(iocb); 1767 struct scm_cookie tmp_scm; 1768 struct sock *sk = sock->sk; 1769 struct unix_sock *u = unix_sk(sk); 1770 int noblock = flags & MSG_DONTWAIT; 1771 struct sk_buff *skb; 1772 int err; 1773 int peeked, skip; 1774 1775 err = -EOPNOTSUPP; 1776 if (flags&MSG_OOB) 1777 goto out; 1778 1779 msg->msg_namelen = 0; 1780 1781 err = mutex_lock_interruptible(&u->readlock); 1782 if (err) { 1783 err = sock_intr_errno(sock_rcvtimeo(sk, noblock)); 1784 goto out; 1785 } 1786 1787 skip = sk_peek_offset(sk, flags); 1788 1789 skb = __skb_recv_datagram(sk, flags, &peeked, &skip, &err); 1790 if (!skb) { 1791 unix_state_lock(sk); 1792 /* Signal EOF on disconnected non-blocking SEQPACKET socket. */ 1793 if (sk->sk_type == SOCK_SEQPACKET && err == -EAGAIN && 1794 (sk->sk_shutdown & RCV_SHUTDOWN)) 1795 err = 0; 1796 unix_state_unlock(sk); 1797 goto out_unlock; 1798 } 1799 1800 wake_up_interruptible_sync_poll(&u->peer_wait, 1801 POLLOUT | POLLWRNORM | POLLWRBAND); 1802 1803 if (msg->msg_name) 1804 unix_copy_addr(msg, skb->sk); 1805 1806 if (size > skb->len - skip) 1807 size = skb->len - skip; 1808 else if (size < skb->len - skip) 1809 msg->msg_flags |= MSG_TRUNC; 1810 1811 err = skb_copy_datagram_iovec(skb, skip, msg->msg_iov, size); 1812 if (err) 1813 goto out_free; 1814 1815 if (sock_flag(sk, SOCK_RCVTSTAMP)) 1816 __sock_recv_timestamp(msg, sk, skb); 1817 1818 if (!siocb->scm) { 1819 siocb->scm = &tmp_scm; 1820 memset(&tmp_scm, 0, sizeof(tmp_scm)); 1821 } 1822 scm_set_cred(siocb->scm, UNIXCB(skb).pid, UNIXCB(skb).cred); 1823 unix_set_secdata(siocb->scm, skb); 1824 1825 if (!(flags & MSG_PEEK)) { 1826 if (UNIXCB(skb).fp) 1827 unix_detach_fds(siocb->scm, skb); 1828 1829 sk_peek_offset_bwd(sk, skb->len); 1830 } else { 1831 /* It is questionable: on PEEK we could: 1832 - do not return fds - good, but too simple 8) 1833 - return fds, and do not return them on read (old strategy, 1834 apparently wrong) 1835 - clone fds (I chose it for now, it is the most universal 1836 solution) 1837 1838 POSIX 1003.1g does not actually define this clearly 1839 at all. POSIX 1003.1g doesn't define a lot of things 1840 clearly however! 1841 1842 */ 1843 1844 sk_peek_offset_fwd(sk, size); 1845 1846 if (UNIXCB(skb).fp) 1847 siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp); 1848 } 1849 err = (flags & MSG_TRUNC) ? skb->len - skip : size; 1850 1851 scm_recv(sock, msg, siocb->scm, flags); 1852 1853 out_free: 1854 skb_free_datagram(sk, skb); 1855 out_unlock: 1856 mutex_unlock(&u->readlock); 1857 out: 1858 return err; 1859 } 1860 1861 /* 1862 * Sleep until data has arrive. But check for races.. 1863 */ 1864 1865 static long unix_stream_data_wait(struct sock *sk, long timeo) 1866 { 1867 DEFINE_WAIT(wait); 1868 1869 unix_state_lock(sk); 1870 1871 for (;;) { 1872 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 1873 1874 if (!skb_queue_empty(&sk->sk_receive_queue) || 1875 sk->sk_err || 1876 (sk->sk_shutdown & RCV_SHUTDOWN) || 1877 signal_pending(current) || 1878 !timeo) 1879 break; 1880 1881 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags); 1882 unix_state_unlock(sk); 1883 timeo = schedule_timeout(timeo); 1884 unix_state_lock(sk); 1885 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags); 1886 } 1887 1888 finish_wait(sk_sleep(sk), &wait); 1889 unix_state_unlock(sk); 1890 return timeo; 1891 } 1892 1893 1894 1895 static int unix_stream_recvmsg(struct kiocb *iocb, struct socket *sock, 1896 struct msghdr *msg, size_t size, 1897 int flags) 1898 { 1899 struct sock_iocb *siocb = kiocb_to_siocb(iocb); 1900 struct scm_cookie tmp_scm; 1901 struct sock *sk = sock->sk; 1902 struct unix_sock *u = unix_sk(sk); 1903 struct sockaddr_un *sunaddr = msg->msg_name; 1904 int copied = 0; 1905 int check_creds = 0; 1906 int target; 1907 int err = 0; 1908 long timeo; 1909 int skip; 1910 1911 err = -EINVAL; 1912 if (sk->sk_state != TCP_ESTABLISHED) 1913 goto out; 1914 1915 err = -EOPNOTSUPP; 1916 if (flags&MSG_OOB) 1917 goto out; 1918 1919 target = sock_rcvlowat(sk, flags&MSG_WAITALL, size); 1920 timeo = sock_rcvtimeo(sk, flags&MSG_DONTWAIT); 1921 1922 msg->msg_namelen = 0; 1923 1924 /* Lock the socket to prevent queue disordering 1925 * while sleeps in memcpy_tomsg 1926 */ 1927 1928 if (!siocb->scm) { 1929 siocb->scm = &tmp_scm; 1930 memset(&tmp_scm, 0, sizeof(tmp_scm)); 1931 } 1932 1933 err = mutex_lock_interruptible(&u->readlock); 1934 if (err) { 1935 err = sock_intr_errno(timeo); 1936 goto out; 1937 } 1938 1939 skip = sk_peek_offset(sk, flags); 1940 1941 do { 1942 int chunk; 1943 struct sk_buff *skb; 1944 1945 unix_state_lock(sk); 1946 skb = skb_peek(&sk->sk_receive_queue); 1947 again: 1948 if (skb == NULL) { 1949 unix_sk(sk)->recursion_level = 0; 1950 if (copied >= target) 1951 goto unlock; 1952 1953 /* 1954 * POSIX 1003.1g mandates this order. 1955 */ 1956 1957 err = sock_error(sk); 1958 if (err) 1959 goto unlock; 1960 if (sk->sk_shutdown & RCV_SHUTDOWN) 1961 goto unlock; 1962 1963 unix_state_unlock(sk); 1964 err = -EAGAIN; 1965 if (!timeo) 1966 break; 1967 mutex_unlock(&u->readlock); 1968 1969 timeo = unix_stream_data_wait(sk, timeo); 1970 1971 if (signal_pending(current) 1972 || mutex_lock_interruptible(&u->readlock)) { 1973 err = sock_intr_errno(timeo); 1974 goto out; 1975 } 1976 1977 continue; 1978 unlock: 1979 unix_state_unlock(sk); 1980 break; 1981 } 1982 1983 if (skip >= skb->len) { 1984 skip -= skb->len; 1985 skb = skb_peek_next(skb, &sk->sk_receive_queue); 1986 goto again; 1987 } 1988 1989 unix_state_unlock(sk); 1990 1991 if (check_creds) { 1992 /* Never glue messages from different writers */ 1993 if ((UNIXCB(skb).pid != siocb->scm->pid) || 1994 (UNIXCB(skb).cred != siocb->scm->cred)) 1995 break; 1996 } else if (test_bit(SOCK_PASSCRED, &sock->flags)) { 1997 /* Copy credentials */ 1998 scm_set_cred(siocb->scm, UNIXCB(skb).pid, UNIXCB(skb).cred); 1999 check_creds = 1; 2000 } 2001 2002 /* Copy address just once */ 2003 if (sunaddr) { 2004 unix_copy_addr(msg, skb->sk); 2005 sunaddr = NULL; 2006 } 2007 2008 chunk = min_t(unsigned int, skb->len - skip, size); 2009 if (memcpy_toiovec(msg->msg_iov, skb->data + skip, chunk)) { 2010 if (copied == 0) 2011 copied = -EFAULT; 2012 break; 2013 } 2014 copied += chunk; 2015 size -= chunk; 2016 2017 /* Mark read part of skb as used */ 2018 if (!(flags & MSG_PEEK)) { 2019 skb_pull(skb, chunk); 2020 2021 sk_peek_offset_bwd(sk, chunk); 2022 2023 if (UNIXCB(skb).fp) 2024 unix_detach_fds(siocb->scm, skb); 2025 2026 if (skb->len) 2027 break; 2028 2029 skb_unlink(skb, &sk->sk_receive_queue); 2030 consume_skb(skb); 2031 2032 if (siocb->scm->fp) 2033 break; 2034 } else { 2035 /* It is questionable, see note in unix_dgram_recvmsg. 2036 */ 2037 if (UNIXCB(skb).fp) 2038 siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp); 2039 2040 sk_peek_offset_fwd(sk, chunk); 2041 2042 break; 2043 } 2044 } while (size); 2045 2046 mutex_unlock(&u->readlock); 2047 scm_recv(sock, msg, siocb->scm, flags); 2048 out: 2049 return copied ? : err; 2050 } 2051 2052 static int unix_shutdown(struct socket *sock, int mode) 2053 { 2054 struct sock *sk = sock->sk; 2055 struct sock *other; 2056 2057 if (mode < SHUT_RD || mode > SHUT_RDWR) 2058 return -EINVAL; 2059 /* This maps: 2060 * SHUT_RD (0) -> RCV_SHUTDOWN (1) 2061 * SHUT_WR (1) -> SEND_SHUTDOWN (2) 2062 * SHUT_RDWR (2) -> SHUTDOWN_MASK (3) 2063 */ 2064 ++mode; 2065 2066 unix_state_lock(sk); 2067 sk->sk_shutdown |= mode; 2068 other = unix_peer(sk); 2069 if (other) 2070 sock_hold(other); 2071 unix_state_unlock(sk); 2072 sk->sk_state_change(sk); 2073 2074 if (other && 2075 (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET)) { 2076 2077 int peer_mode = 0; 2078 2079 if (mode&RCV_SHUTDOWN) 2080 peer_mode |= SEND_SHUTDOWN; 2081 if (mode&SEND_SHUTDOWN) 2082 peer_mode |= RCV_SHUTDOWN; 2083 unix_state_lock(other); 2084 other->sk_shutdown |= peer_mode; 2085 unix_state_unlock(other); 2086 other->sk_state_change(other); 2087 if (peer_mode == SHUTDOWN_MASK) 2088 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_HUP); 2089 else if (peer_mode & RCV_SHUTDOWN) 2090 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_IN); 2091 } 2092 if (other) 2093 sock_put(other); 2094 2095 return 0; 2096 } 2097 2098 long unix_inq_len(struct sock *sk) 2099 { 2100 struct sk_buff *skb; 2101 long amount = 0; 2102 2103 if (sk->sk_state == TCP_LISTEN) 2104 return -EINVAL; 2105 2106 spin_lock(&sk->sk_receive_queue.lock); 2107 if (sk->sk_type == SOCK_STREAM || 2108 sk->sk_type == SOCK_SEQPACKET) { 2109 skb_queue_walk(&sk->sk_receive_queue, skb) 2110 amount += skb->len; 2111 } else { 2112 skb = skb_peek(&sk->sk_receive_queue); 2113 if (skb) 2114 amount = skb->len; 2115 } 2116 spin_unlock(&sk->sk_receive_queue.lock); 2117 2118 return amount; 2119 } 2120 EXPORT_SYMBOL_GPL(unix_inq_len); 2121 2122 long unix_outq_len(struct sock *sk) 2123 { 2124 return sk_wmem_alloc_get(sk); 2125 } 2126 EXPORT_SYMBOL_GPL(unix_outq_len); 2127 2128 static int unix_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 2129 { 2130 struct sock *sk = sock->sk; 2131 long amount = 0; 2132 int err; 2133 2134 switch (cmd) { 2135 case SIOCOUTQ: 2136 amount = unix_outq_len(sk); 2137 err = put_user(amount, (int __user *)arg); 2138 break; 2139 case SIOCINQ: 2140 amount = unix_inq_len(sk); 2141 if (amount < 0) 2142 err = amount; 2143 else 2144 err = put_user(amount, (int __user *)arg); 2145 break; 2146 default: 2147 err = -ENOIOCTLCMD; 2148 break; 2149 } 2150 return err; 2151 } 2152 2153 static unsigned int unix_poll(struct file *file, struct socket *sock, poll_table *wait) 2154 { 2155 struct sock *sk = sock->sk; 2156 unsigned int mask; 2157 2158 sock_poll_wait(file, sk_sleep(sk), wait); 2159 mask = 0; 2160 2161 /* exceptional events? */ 2162 if (sk->sk_err) 2163 mask |= POLLERR; 2164 if (sk->sk_shutdown == SHUTDOWN_MASK) 2165 mask |= POLLHUP; 2166 if (sk->sk_shutdown & RCV_SHUTDOWN) 2167 mask |= POLLRDHUP | POLLIN | POLLRDNORM; 2168 2169 /* readable? */ 2170 if (!skb_queue_empty(&sk->sk_receive_queue)) 2171 mask |= POLLIN | POLLRDNORM; 2172 2173 /* Connection-based need to check for termination and startup */ 2174 if ((sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) && 2175 sk->sk_state == TCP_CLOSE) 2176 mask |= POLLHUP; 2177 2178 /* 2179 * we set writable also when the other side has shut down the 2180 * connection. This prevents stuck sockets. 2181 */ 2182 if (unix_writable(sk)) 2183 mask |= POLLOUT | POLLWRNORM | POLLWRBAND; 2184 2185 return mask; 2186 } 2187 2188 static unsigned int unix_dgram_poll(struct file *file, struct socket *sock, 2189 poll_table *wait) 2190 { 2191 struct sock *sk = sock->sk, *other; 2192 unsigned int mask, writable; 2193 2194 sock_poll_wait(file, sk_sleep(sk), wait); 2195 mask = 0; 2196 2197 /* exceptional events? */ 2198 if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue)) 2199 mask |= POLLERR; 2200 if (sk->sk_shutdown & RCV_SHUTDOWN) 2201 mask |= POLLRDHUP | POLLIN | POLLRDNORM; 2202 if (sk->sk_shutdown == SHUTDOWN_MASK) 2203 mask |= POLLHUP; 2204 2205 /* readable? */ 2206 if (!skb_queue_empty(&sk->sk_receive_queue)) 2207 mask |= POLLIN | POLLRDNORM; 2208 2209 /* Connection-based need to check for termination and startup */ 2210 if (sk->sk_type == SOCK_SEQPACKET) { 2211 if (sk->sk_state == TCP_CLOSE) 2212 mask |= POLLHUP; 2213 /* connection hasn't started yet? */ 2214 if (sk->sk_state == TCP_SYN_SENT) 2215 return mask; 2216 } 2217 2218 /* No write status requested, avoid expensive OUT tests. */ 2219 if (!(poll_requested_events(wait) & (POLLWRBAND|POLLWRNORM|POLLOUT))) 2220 return mask; 2221 2222 writable = unix_writable(sk); 2223 other = unix_peer_get(sk); 2224 if (other) { 2225 if (unix_peer(other) != sk) { 2226 sock_poll_wait(file, &unix_sk(other)->peer_wait, wait); 2227 if (unix_recvq_full(other)) 2228 writable = 0; 2229 } 2230 sock_put(other); 2231 } 2232 2233 if (writable) 2234 mask |= POLLOUT | POLLWRNORM | POLLWRBAND; 2235 else 2236 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags); 2237 2238 return mask; 2239 } 2240 2241 #ifdef CONFIG_PROC_FS 2242 2243 #define BUCKET_SPACE (BITS_PER_LONG - (UNIX_HASH_BITS + 1) - 1) 2244 2245 #define get_bucket(x) ((x) >> BUCKET_SPACE) 2246 #define get_offset(x) ((x) & ((1L << BUCKET_SPACE) - 1)) 2247 #define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o)) 2248 2249 static struct sock *unix_from_bucket(struct seq_file *seq, loff_t *pos) 2250 { 2251 unsigned long offset = get_offset(*pos); 2252 unsigned long bucket = get_bucket(*pos); 2253 struct sock *sk; 2254 unsigned long count = 0; 2255 2256 for (sk = sk_head(&unix_socket_table[bucket]); sk; sk = sk_next(sk)) { 2257 if (sock_net(sk) != seq_file_net(seq)) 2258 continue; 2259 if (++count == offset) 2260 break; 2261 } 2262 2263 return sk; 2264 } 2265 2266 static struct sock *unix_next_socket(struct seq_file *seq, 2267 struct sock *sk, 2268 loff_t *pos) 2269 { 2270 unsigned long bucket; 2271 2272 while (sk > (struct sock *)SEQ_START_TOKEN) { 2273 sk = sk_next(sk); 2274 if (!sk) 2275 goto next_bucket; 2276 if (sock_net(sk) == seq_file_net(seq)) 2277 return sk; 2278 } 2279 2280 do { 2281 sk = unix_from_bucket(seq, pos); 2282 if (sk) 2283 return sk; 2284 2285 next_bucket: 2286 bucket = get_bucket(*pos) + 1; 2287 *pos = set_bucket_offset(bucket, 1); 2288 } while (bucket < ARRAY_SIZE(unix_socket_table)); 2289 2290 return NULL; 2291 } 2292 2293 static void *unix_seq_start(struct seq_file *seq, loff_t *pos) 2294 __acquires(unix_table_lock) 2295 { 2296 spin_lock(&unix_table_lock); 2297 2298 if (!*pos) 2299 return SEQ_START_TOKEN; 2300 2301 if (get_bucket(*pos) >= ARRAY_SIZE(unix_socket_table)) 2302 return NULL; 2303 2304 return unix_next_socket(seq, NULL, pos); 2305 } 2306 2307 static void *unix_seq_next(struct seq_file *seq, void *v, loff_t *pos) 2308 { 2309 ++*pos; 2310 return unix_next_socket(seq, v, pos); 2311 } 2312 2313 static void unix_seq_stop(struct seq_file *seq, void *v) 2314 __releases(unix_table_lock) 2315 { 2316 spin_unlock(&unix_table_lock); 2317 } 2318 2319 static int unix_seq_show(struct seq_file *seq, void *v) 2320 { 2321 2322 if (v == SEQ_START_TOKEN) 2323 seq_puts(seq, "Num RefCount Protocol Flags Type St " 2324 "Inode Path\n"); 2325 else { 2326 struct sock *s = v; 2327 struct unix_sock *u = unix_sk(s); 2328 unix_state_lock(s); 2329 2330 seq_printf(seq, "%pK: %08X %08X %08X %04X %02X %5lu", 2331 s, 2332 atomic_read(&s->sk_refcnt), 2333 0, 2334 s->sk_state == TCP_LISTEN ? __SO_ACCEPTCON : 0, 2335 s->sk_type, 2336 s->sk_socket ? 2337 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTED : SS_UNCONNECTED) : 2338 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTING : SS_DISCONNECTING), 2339 sock_i_ino(s)); 2340 2341 if (u->addr) { 2342 int i, len; 2343 seq_putc(seq, ' '); 2344 2345 i = 0; 2346 len = u->addr->len - sizeof(short); 2347 if (!UNIX_ABSTRACT(s)) 2348 len--; 2349 else { 2350 seq_putc(seq, '@'); 2351 i++; 2352 } 2353 for ( ; i < len; i++) 2354 seq_putc(seq, u->addr->name->sun_path[i]); 2355 } 2356 unix_state_unlock(s); 2357 seq_putc(seq, '\n'); 2358 } 2359 2360 return 0; 2361 } 2362 2363 static const struct seq_operations unix_seq_ops = { 2364 .start = unix_seq_start, 2365 .next = unix_seq_next, 2366 .stop = unix_seq_stop, 2367 .show = unix_seq_show, 2368 }; 2369 2370 static int unix_seq_open(struct inode *inode, struct file *file) 2371 { 2372 return seq_open_net(inode, file, &unix_seq_ops, 2373 sizeof(struct seq_net_private)); 2374 } 2375 2376 static const struct file_operations unix_seq_fops = { 2377 .owner = THIS_MODULE, 2378 .open = unix_seq_open, 2379 .read = seq_read, 2380 .llseek = seq_lseek, 2381 .release = seq_release_net, 2382 }; 2383 2384 #endif 2385 2386 static const struct net_proto_family unix_family_ops = { 2387 .family = PF_UNIX, 2388 .create = unix_create, 2389 .owner = THIS_MODULE, 2390 }; 2391 2392 2393 static int __net_init unix_net_init(struct net *net) 2394 { 2395 int error = -ENOMEM; 2396 2397 net->unx.sysctl_max_dgram_qlen = 10; 2398 if (unix_sysctl_register(net)) 2399 goto out; 2400 2401 #ifdef CONFIG_PROC_FS 2402 if (!proc_create("unix", 0, net->proc_net, &unix_seq_fops)) { 2403 unix_sysctl_unregister(net); 2404 goto out; 2405 } 2406 #endif 2407 error = 0; 2408 out: 2409 return error; 2410 } 2411 2412 static void __net_exit unix_net_exit(struct net *net) 2413 { 2414 unix_sysctl_unregister(net); 2415 remove_proc_entry("unix", net->proc_net); 2416 } 2417 2418 static struct pernet_operations unix_net_ops = { 2419 .init = unix_net_init, 2420 .exit = unix_net_exit, 2421 }; 2422 2423 static int __init af_unix_init(void) 2424 { 2425 int rc = -1; 2426 2427 BUILD_BUG_ON(sizeof(struct unix_skb_parms) > FIELD_SIZEOF(struct sk_buff, cb)); 2428 2429 rc = proto_register(&unix_proto, 1); 2430 if (rc != 0) { 2431 printk(KERN_CRIT "%s: Cannot create unix_sock SLAB cache!\n", 2432 __func__); 2433 goto out; 2434 } 2435 2436 sock_register(&unix_family_ops); 2437 register_pernet_subsys(&unix_net_ops); 2438 out: 2439 return rc; 2440 } 2441 2442 static void __exit af_unix_exit(void) 2443 { 2444 sock_unregister(PF_UNIX); 2445 proto_unregister(&unix_proto); 2446 unregister_pernet_subsys(&unix_net_ops); 2447 } 2448 2449 /* Earlier than device_initcall() so that other drivers invoking 2450 request_module() don't end up in a loop when modprobe tries 2451 to use a UNIX socket. But later than subsys_initcall() because 2452 we depend on stuff initialised there */ 2453 fs_initcall(af_unix_init); 2454 module_exit(af_unix_exit); 2455 2456 MODULE_LICENSE("GPL"); 2457 MODULE_ALIAS_NETPROTO(PF_UNIX); 2458