1 /* 2 * NET4: Implementation of BSD Unix domain sockets. 3 * 4 * Authors: Alan Cox, <alan.cox@linux.org> 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 * Version: $Id: af_unix.c,v 1.133 2002/02/08 03:57:19 davem Exp $ 12 * 13 * Fixes: 14 * Linus Torvalds : Assorted bug cures. 15 * Niibe Yutaka : async I/O support. 16 * Carsten Paeth : PF_UNIX check, address fixes. 17 * Alan Cox : Limit size of allocated blocks. 18 * Alan Cox : Fixed the stupid socketpair bug. 19 * Alan Cox : BSD compatibility fine tuning. 20 * Alan Cox : Fixed a bug in connect when interrupted. 21 * Alan Cox : Sorted out a proper draft version of 22 * file descriptor passing hacked up from 23 * Mike Shaver's work. 24 * Marty Leisner : Fixes to fd passing 25 * Nick Nevin : recvmsg bugfix. 26 * Alan Cox : Started proper garbage collector 27 * Heiko EiBfeldt : Missing verify_area check 28 * Alan Cox : Started POSIXisms 29 * Andreas Schwab : Replace inode by dentry for proper 30 * reference counting 31 * Kirk Petersen : Made this a module 32 * Christoph Rohland : Elegant non-blocking accept/connect algorithm. 33 * Lots of bug fixes. 34 * Alexey Kuznetosv : Repaired (I hope) bugs introduces 35 * by above two patches. 36 * Andrea Arcangeli : If possible we block in connect(2) 37 * if the max backlog of the listen socket 38 * is been reached. This won't break 39 * old apps and it will avoid huge amount 40 * of socks hashed (this for unix_gc() 41 * performances reasons). 42 * Security fix that limits the max 43 * number of socks to 2*max_files and 44 * the number of skb queueable in the 45 * dgram receiver. 46 * Artur Skawina : Hash function optimizations 47 * Alexey Kuznetsov : Full scale SMP. Lot of bugs are introduced 8) 48 * Malcolm Beattie : Set peercred for socketpair 49 * Michal Ostrowski : Module initialization cleanup. 50 * Arnaldo C. Melo : Remove MOD_{INC,DEC}_USE_COUNT, 51 * the core infrastructure is doing that 52 * for all net proto families now (2.5.69+) 53 * 54 * 55 * Known differences from reference BSD that was tested: 56 * 57 * [TO FIX] 58 * ECONNREFUSED is not returned from one end of a connected() socket to the 59 * other the moment one end closes. 60 * fstat() doesn't return st_dev=0, and give the blksize as high water mark 61 * and a fake inode identifier (nor the BSD first socket fstat twice bug). 62 * [NOT TO FIX] 63 * accept() returns a path name even if the connecting socket has closed 64 * in the meantime (BSD loses the path and gives up). 65 * accept() returns 0 length path for an unbound connector. BSD returns 16 66 * and a null first byte in the path (but not for gethost/peername - BSD bug ??) 67 * socketpair(...SOCK_RAW..) doesn't panic the kernel. 68 * BSD af_unix apparently has connect forgetting to block properly. 69 * (need to check this with the POSIX spec in detail) 70 * 71 * Differences from 2.0.0-11-... (ANK) 72 * Bug fixes and improvements. 73 * - client shutdown killed server socket. 74 * - removed all useless cli/sti pairs. 75 * 76 * Semantic changes/extensions. 77 * - generic control message passing. 78 * - SCM_CREDENTIALS control message. 79 * - "Abstract" (not FS based) socket bindings. 80 * Abstract names are sequences of bytes (not zero terminated) 81 * started by 0, so that this name space does not intersect 82 * with BSD names. 83 */ 84 85 #include <linux/module.h> 86 #include <linux/kernel.h> 87 #include <linux/signal.h> 88 #include <linux/sched.h> 89 #include <linux/errno.h> 90 #include <linux/string.h> 91 #include <linux/stat.h> 92 #include <linux/dcache.h> 93 #include <linux/namei.h> 94 #include <linux/socket.h> 95 #include <linux/un.h> 96 #include <linux/fcntl.h> 97 #include <linux/termios.h> 98 #include <linux/sockios.h> 99 #include <linux/net.h> 100 #include <linux/in.h> 101 #include <linux/fs.h> 102 #include <linux/slab.h> 103 #include <asm/uaccess.h> 104 #include <linux/skbuff.h> 105 #include <linux/netdevice.h> 106 #include <net/sock.h> 107 #include <net/tcp_states.h> 108 #include <net/af_unix.h> 109 #include <linux/proc_fs.h> 110 #include <linux/seq_file.h> 111 #include <net/scm.h> 112 #include <linux/init.h> 113 #include <linux/poll.h> 114 #include <linux/smp_lock.h> 115 #include <linux/rtnetlink.h> 116 #include <linux/mount.h> 117 #include <net/checksum.h> 118 #include <linux/security.h> 119 120 int sysctl_unix_max_dgram_qlen = 10; 121 122 struct hlist_head unix_socket_table[UNIX_HASH_SIZE + 1]; 123 DEFINE_SPINLOCK(unix_table_lock); 124 static atomic_t unix_nr_socks = ATOMIC_INIT(0); 125 126 #define unix_sockets_unbound (&unix_socket_table[UNIX_HASH_SIZE]) 127 128 #define UNIX_ABSTRACT(sk) (unix_sk(sk)->addr->hash != UNIX_HASH_SIZE) 129 130 #ifdef CONFIG_SECURITY_NETWORK 131 static void unix_get_peersec_dgram(struct sk_buff *skb) 132 { 133 int err; 134 135 err = security_socket_getpeersec_dgram(skb, UNIXSECDATA(skb), 136 UNIXSECLEN(skb)); 137 if (err) 138 *(UNIXSECDATA(skb)) = NULL; 139 } 140 141 static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb) 142 { 143 scm->secdata = *UNIXSECDATA(skb); 144 scm->seclen = *UNIXSECLEN(skb); 145 } 146 #else 147 static inline void unix_get_peersec_dgram(struct sk_buff *skb) 148 { } 149 150 static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb) 151 { } 152 #endif /* CONFIG_SECURITY_NETWORK */ 153 154 /* 155 * SMP locking strategy: 156 * hash table is protected with spinlock unix_table_lock 157 * each socket state is protected by separate rwlock. 158 */ 159 160 static inline unsigned unix_hash_fold(unsigned hash) 161 { 162 hash ^= hash>>16; 163 hash ^= hash>>8; 164 return hash&(UNIX_HASH_SIZE-1); 165 } 166 167 #define unix_peer(sk) (unix_sk(sk)->peer) 168 169 static inline int unix_our_peer(struct sock *sk, struct sock *osk) 170 { 171 return unix_peer(osk) == sk; 172 } 173 174 static inline int unix_may_send(struct sock *sk, struct sock *osk) 175 { 176 return (unix_peer(osk) == NULL || unix_our_peer(sk, osk)); 177 } 178 179 static struct sock *unix_peer_get(struct sock *s) 180 { 181 struct sock *peer; 182 183 unix_state_rlock(s); 184 peer = unix_peer(s); 185 if (peer) 186 sock_hold(peer); 187 unix_state_runlock(s); 188 return peer; 189 } 190 191 static inline void unix_release_addr(struct unix_address *addr) 192 { 193 if (atomic_dec_and_test(&addr->refcnt)) 194 kfree(addr); 195 } 196 197 /* 198 * Check unix socket name: 199 * - should be not zero length. 200 * - if started by not zero, should be NULL terminated (FS object) 201 * - if started by zero, it is abstract name. 202 */ 203 204 static int unix_mkname(struct sockaddr_un * sunaddr, int len, unsigned *hashp) 205 { 206 if (len <= sizeof(short) || len > sizeof(*sunaddr)) 207 return -EINVAL; 208 if (!sunaddr || sunaddr->sun_family != AF_UNIX) 209 return -EINVAL; 210 if (sunaddr->sun_path[0]) { 211 /* 212 * This may look like an off by one error but it is a bit more 213 * subtle. 108 is the longest valid AF_UNIX path for a binding. 214 * sun_path[108] doesnt as such exist. However in kernel space 215 * we are guaranteed that it is a valid memory location in our 216 * kernel address buffer. 217 */ 218 ((char *)sunaddr)[len]=0; 219 len = strlen(sunaddr->sun_path)+1+sizeof(short); 220 return len; 221 } 222 223 *hashp = unix_hash_fold(csum_partial((char*)sunaddr, len, 0)); 224 return len; 225 } 226 227 static void __unix_remove_socket(struct sock *sk) 228 { 229 sk_del_node_init(sk); 230 } 231 232 static void __unix_insert_socket(struct hlist_head *list, struct sock *sk) 233 { 234 BUG_TRAP(sk_unhashed(sk)); 235 sk_add_node(sk, list); 236 } 237 238 static inline void unix_remove_socket(struct sock *sk) 239 { 240 spin_lock(&unix_table_lock); 241 __unix_remove_socket(sk); 242 spin_unlock(&unix_table_lock); 243 } 244 245 static inline void unix_insert_socket(struct hlist_head *list, struct sock *sk) 246 { 247 spin_lock(&unix_table_lock); 248 __unix_insert_socket(list, sk); 249 spin_unlock(&unix_table_lock); 250 } 251 252 static struct sock *__unix_find_socket_byname(struct sockaddr_un *sunname, 253 int len, int type, unsigned hash) 254 { 255 struct sock *s; 256 struct hlist_node *node; 257 258 sk_for_each(s, node, &unix_socket_table[hash ^ type]) { 259 struct unix_sock *u = unix_sk(s); 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 sockaddr_un *sunname, 271 int len, int type, 272 unsigned hash) 273 { 274 struct sock *s; 275 276 spin_lock(&unix_table_lock); 277 s = __unix_find_socket_byname(sunname, len, type, hash); 278 if (s) 279 sock_hold(s); 280 spin_unlock(&unix_table_lock); 281 return s; 282 } 283 284 static struct sock *unix_find_socket_byinode(struct inode *i) 285 { 286 struct sock *s; 287 struct hlist_node *node; 288 289 spin_lock(&unix_table_lock); 290 sk_for_each(s, node, 291 &unix_socket_table[i->i_ino & (UNIX_HASH_SIZE - 1)]) { 292 struct dentry *dentry = unix_sk(s)->dentry; 293 294 if(dentry && dentry->d_inode == i) 295 { 296 sock_hold(s); 297 goto found; 298 } 299 } 300 s = NULL; 301 found: 302 spin_unlock(&unix_table_lock); 303 return s; 304 } 305 306 static inline int unix_writable(struct sock *sk) 307 { 308 return (atomic_read(&sk->sk_wmem_alloc) << 2) <= sk->sk_sndbuf; 309 } 310 311 static void unix_write_space(struct sock *sk) 312 { 313 read_lock(&sk->sk_callback_lock); 314 if (unix_writable(sk)) { 315 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) 316 wake_up_interruptible(sk->sk_sleep); 317 sk_wake_async(sk, 2, POLL_OUT); 318 } 319 read_unlock(&sk->sk_callback_lock); 320 } 321 322 /* When dgram socket disconnects (or changes its peer), we clear its receive 323 * queue of packets arrived from previous peer. First, it allows to do 324 * flow control based only on wmem_alloc; second, sk connected to peer 325 * may receive messages only from that peer. */ 326 static void unix_dgram_disconnected(struct sock *sk, struct sock *other) 327 { 328 if (!skb_queue_empty(&sk->sk_receive_queue)) { 329 skb_queue_purge(&sk->sk_receive_queue); 330 wake_up_interruptible_all(&unix_sk(sk)->peer_wait); 331 332 /* If one link of bidirectional dgram pipe is disconnected, 333 * we signal error. Messages are lost. Do not make this, 334 * when peer was not connected to us. 335 */ 336 if (!sock_flag(other, SOCK_DEAD) && unix_peer(other) == sk) { 337 other->sk_err = ECONNRESET; 338 other->sk_error_report(other); 339 } 340 } 341 } 342 343 static void unix_sock_destructor(struct sock *sk) 344 { 345 struct unix_sock *u = unix_sk(sk); 346 347 skb_queue_purge(&sk->sk_receive_queue); 348 349 BUG_TRAP(!atomic_read(&sk->sk_wmem_alloc)); 350 BUG_TRAP(sk_unhashed(sk)); 351 BUG_TRAP(!sk->sk_socket); 352 if (!sock_flag(sk, SOCK_DEAD)) { 353 printk("Attempt to release alive unix socket: %p\n", sk); 354 return; 355 } 356 357 if (u->addr) 358 unix_release_addr(u->addr); 359 360 atomic_dec(&unix_nr_socks); 361 #ifdef UNIX_REFCNT_DEBUG 362 printk(KERN_DEBUG "UNIX %p is destroyed, %d are still alive.\n", sk, atomic_read(&unix_nr_socks)); 363 #endif 364 } 365 366 static int unix_release_sock (struct sock *sk, int embrion) 367 { 368 struct unix_sock *u = unix_sk(sk); 369 struct dentry *dentry; 370 struct vfsmount *mnt; 371 struct sock *skpair; 372 struct sk_buff *skb; 373 int state; 374 375 unix_remove_socket(sk); 376 377 /* Clear state */ 378 unix_state_wlock(sk); 379 sock_orphan(sk); 380 sk->sk_shutdown = SHUTDOWN_MASK; 381 dentry = u->dentry; 382 u->dentry = NULL; 383 mnt = u->mnt; 384 u->mnt = NULL; 385 state = sk->sk_state; 386 sk->sk_state = TCP_CLOSE; 387 unix_state_wunlock(sk); 388 389 wake_up_interruptible_all(&u->peer_wait); 390 391 skpair=unix_peer(sk); 392 393 if (skpair!=NULL) { 394 if (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) { 395 unix_state_wlock(skpair); 396 /* No more writes */ 397 skpair->sk_shutdown = SHUTDOWN_MASK; 398 if (!skb_queue_empty(&sk->sk_receive_queue) || embrion) 399 skpair->sk_err = ECONNRESET; 400 unix_state_wunlock(skpair); 401 skpair->sk_state_change(skpair); 402 read_lock(&skpair->sk_callback_lock); 403 sk_wake_async(skpair,1,POLL_HUP); 404 read_unlock(&skpair->sk_callback_lock); 405 } 406 sock_put(skpair); /* It may now die */ 407 unix_peer(sk) = NULL; 408 } 409 410 /* Try to flush out this socket. Throw out buffers at least */ 411 412 while ((skb = skb_dequeue(&sk->sk_receive_queue)) != NULL) { 413 if (state==TCP_LISTEN) 414 unix_release_sock(skb->sk, 1); 415 /* passed fds are erased in the kfree_skb hook */ 416 kfree_skb(skb); 417 } 418 419 if (dentry) { 420 dput(dentry); 421 mntput(mnt); 422 } 423 424 sock_put(sk); 425 426 /* ---- Socket is dead now and most probably destroyed ---- */ 427 428 /* 429 * Fixme: BSD difference: In BSD all sockets connected to use get 430 * ECONNRESET and we die on the spot. In Linux we behave 431 * like files and pipes do and wait for the last 432 * dereference. 433 * 434 * Can't we simply set sock->err? 435 * 436 * What the above comment does talk about? --ANK(980817) 437 */ 438 439 if (atomic_read(&unix_tot_inflight)) 440 unix_gc(); /* Garbage collect fds */ 441 442 return 0; 443 } 444 445 static int unix_listen(struct socket *sock, int backlog) 446 { 447 int err; 448 struct sock *sk = sock->sk; 449 struct unix_sock *u = unix_sk(sk); 450 451 err = -EOPNOTSUPP; 452 if (sock->type!=SOCK_STREAM && sock->type!=SOCK_SEQPACKET) 453 goto out; /* Only stream/seqpacket sockets accept */ 454 err = -EINVAL; 455 if (!u->addr) 456 goto out; /* No listens on an unbound socket */ 457 unix_state_wlock(sk); 458 if (sk->sk_state != TCP_CLOSE && sk->sk_state != TCP_LISTEN) 459 goto out_unlock; 460 if (backlog > sk->sk_max_ack_backlog) 461 wake_up_interruptible_all(&u->peer_wait); 462 sk->sk_max_ack_backlog = backlog; 463 sk->sk_state = TCP_LISTEN; 464 /* set credentials so connect can copy them */ 465 sk->sk_peercred.pid = current->tgid; 466 sk->sk_peercred.uid = current->euid; 467 sk->sk_peercred.gid = current->egid; 468 err = 0; 469 470 out_unlock: 471 unix_state_wunlock(sk); 472 out: 473 return err; 474 } 475 476 static int unix_release(struct socket *); 477 static int unix_bind(struct socket *, struct sockaddr *, int); 478 static int unix_stream_connect(struct socket *, struct sockaddr *, 479 int addr_len, int flags); 480 static int unix_socketpair(struct socket *, struct socket *); 481 static int unix_accept(struct socket *, struct socket *, int); 482 static int unix_getname(struct socket *, struct sockaddr *, int *, int); 483 static unsigned int unix_poll(struct file *, struct socket *, poll_table *); 484 static int unix_ioctl(struct socket *, unsigned int, unsigned long); 485 static int unix_shutdown(struct socket *, int); 486 static int unix_stream_sendmsg(struct kiocb *, struct socket *, 487 struct msghdr *, size_t); 488 static int unix_stream_recvmsg(struct kiocb *, struct socket *, 489 struct msghdr *, size_t, int); 490 static int unix_dgram_sendmsg(struct kiocb *, struct socket *, 491 struct msghdr *, size_t); 492 static int unix_dgram_recvmsg(struct kiocb *, struct socket *, 493 struct msghdr *, size_t, int); 494 static int unix_dgram_connect(struct socket *, struct sockaddr *, 495 int, int); 496 static int unix_seqpacket_sendmsg(struct kiocb *, struct socket *, 497 struct msghdr *, size_t); 498 499 static const struct proto_ops unix_stream_ops = { 500 .family = PF_UNIX, 501 .owner = THIS_MODULE, 502 .release = unix_release, 503 .bind = unix_bind, 504 .connect = unix_stream_connect, 505 .socketpair = unix_socketpair, 506 .accept = unix_accept, 507 .getname = unix_getname, 508 .poll = unix_poll, 509 .ioctl = unix_ioctl, 510 .listen = unix_listen, 511 .shutdown = unix_shutdown, 512 .setsockopt = sock_no_setsockopt, 513 .getsockopt = sock_no_getsockopt, 514 .sendmsg = unix_stream_sendmsg, 515 .recvmsg = unix_stream_recvmsg, 516 .mmap = sock_no_mmap, 517 .sendpage = sock_no_sendpage, 518 }; 519 520 static const struct proto_ops unix_dgram_ops = { 521 .family = PF_UNIX, 522 .owner = THIS_MODULE, 523 .release = unix_release, 524 .bind = unix_bind, 525 .connect = unix_dgram_connect, 526 .socketpair = unix_socketpair, 527 .accept = sock_no_accept, 528 .getname = unix_getname, 529 .poll = datagram_poll, 530 .ioctl = unix_ioctl, 531 .listen = sock_no_listen, 532 .shutdown = unix_shutdown, 533 .setsockopt = sock_no_setsockopt, 534 .getsockopt = sock_no_getsockopt, 535 .sendmsg = unix_dgram_sendmsg, 536 .recvmsg = unix_dgram_recvmsg, 537 .mmap = sock_no_mmap, 538 .sendpage = sock_no_sendpage, 539 }; 540 541 static const struct proto_ops unix_seqpacket_ops = { 542 .family = PF_UNIX, 543 .owner = THIS_MODULE, 544 .release = unix_release, 545 .bind = unix_bind, 546 .connect = unix_stream_connect, 547 .socketpair = unix_socketpair, 548 .accept = unix_accept, 549 .getname = unix_getname, 550 .poll = datagram_poll, 551 .ioctl = unix_ioctl, 552 .listen = unix_listen, 553 .shutdown = unix_shutdown, 554 .setsockopt = sock_no_setsockopt, 555 .getsockopt = sock_no_getsockopt, 556 .sendmsg = unix_seqpacket_sendmsg, 557 .recvmsg = unix_dgram_recvmsg, 558 .mmap = sock_no_mmap, 559 .sendpage = sock_no_sendpage, 560 }; 561 562 static struct proto unix_proto = { 563 .name = "UNIX", 564 .owner = THIS_MODULE, 565 .obj_size = sizeof(struct unix_sock), 566 }; 567 568 /* 569 * AF_UNIX sockets do not interact with hardware, hence they 570 * dont trigger interrupts - so it's safe for them to have 571 * bh-unsafe locking for their sk_receive_queue.lock. Split off 572 * this special lock-class by reinitializing the spinlock key: 573 */ 574 static struct lock_class_key af_unix_sk_receive_queue_lock_key; 575 576 static struct sock * unix_create1(struct socket *sock) 577 { 578 struct sock *sk = NULL; 579 struct unix_sock *u; 580 581 if (atomic_read(&unix_nr_socks) >= 2*get_max_files()) 582 goto out; 583 584 sk = sk_alloc(PF_UNIX, GFP_KERNEL, &unix_proto, 1); 585 if (!sk) 586 goto out; 587 588 atomic_inc(&unix_nr_socks); 589 590 sock_init_data(sock,sk); 591 lockdep_set_class(&sk->sk_receive_queue.lock, 592 &af_unix_sk_receive_queue_lock_key); 593 594 sk->sk_write_space = unix_write_space; 595 sk->sk_max_ack_backlog = sysctl_unix_max_dgram_qlen; 596 sk->sk_destruct = unix_sock_destructor; 597 u = unix_sk(sk); 598 u->dentry = NULL; 599 u->mnt = NULL; 600 spin_lock_init(&u->lock); 601 atomic_set(&u->inflight, sock ? 0 : -1); 602 mutex_init(&u->readlock); /* single task reading lock */ 603 init_waitqueue_head(&u->peer_wait); 604 unix_insert_socket(unix_sockets_unbound, sk); 605 out: 606 return sk; 607 } 608 609 static int unix_create(struct socket *sock, int protocol) 610 { 611 if (protocol && protocol != PF_UNIX) 612 return -EPROTONOSUPPORT; 613 614 sock->state = SS_UNCONNECTED; 615 616 switch (sock->type) { 617 case SOCK_STREAM: 618 sock->ops = &unix_stream_ops; 619 break; 620 /* 621 * Believe it or not BSD has AF_UNIX, SOCK_RAW though 622 * nothing uses it. 623 */ 624 case SOCK_RAW: 625 sock->type=SOCK_DGRAM; 626 case SOCK_DGRAM: 627 sock->ops = &unix_dgram_ops; 628 break; 629 case SOCK_SEQPACKET: 630 sock->ops = &unix_seqpacket_ops; 631 break; 632 default: 633 return -ESOCKTNOSUPPORT; 634 } 635 636 return unix_create1(sock) ? 0 : -ENOMEM; 637 } 638 639 static int unix_release(struct socket *sock) 640 { 641 struct sock *sk = sock->sk; 642 643 if (!sk) 644 return 0; 645 646 sock->sk = NULL; 647 648 return unix_release_sock (sk, 0); 649 } 650 651 static int unix_autobind(struct socket *sock) 652 { 653 struct sock *sk = sock->sk; 654 struct unix_sock *u = unix_sk(sk); 655 static u32 ordernum = 1; 656 struct unix_address * addr; 657 int err; 658 659 mutex_lock(&u->readlock); 660 661 err = 0; 662 if (u->addr) 663 goto out; 664 665 err = -ENOMEM; 666 addr = kmalloc(sizeof(*addr) + sizeof(short) + 16, GFP_KERNEL); 667 if (!addr) 668 goto out; 669 670 memset(addr, 0, sizeof(*addr) + sizeof(short) + 16); 671 addr->name->sun_family = AF_UNIX; 672 atomic_set(&addr->refcnt, 1); 673 674 retry: 675 addr->len = sprintf(addr->name->sun_path+1, "%05x", ordernum) + 1 + sizeof(short); 676 addr->hash = unix_hash_fold(csum_partial((void*)addr->name, addr->len, 0)); 677 678 spin_lock(&unix_table_lock); 679 ordernum = (ordernum+1)&0xFFFFF; 680 681 if (__unix_find_socket_byname(addr->name, addr->len, sock->type, 682 addr->hash)) { 683 spin_unlock(&unix_table_lock); 684 /* Sanity yield. It is unusual case, but yet... */ 685 if (!(ordernum&0xFF)) 686 yield(); 687 goto retry; 688 } 689 addr->hash ^= sk->sk_type; 690 691 __unix_remove_socket(sk); 692 u->addr = addr; 693 __unix_insert_socket(&unix_socket_table[addr->hash], sk); 694 spin_unlock(&unix_table_lock); 695 err = 0; 696 697 out: mutex_unlock(&u->readlock); 698 return err; 699 } 700 701 static struct sock *unix_find_other(struct sockaddr_un *sunname, int len, 702 int type, unsigned hash, int *error) 703 { 704 struct sock *u; 705 struct nameidata nd; 706 int err = 0; 707 708 if (sunname->sun_path[0]) { 709 err = path_lookup(sunname->sun_path, LOOKUP_FOLLOW, &nd); 710 if (err) 711 goto fail; 712 err = vfs_permission(&nd, MAY_WRITE); 713 if (err) 714 goto put_fail; 715 716 err = -ECONNREFUSED; 717 if (!S_ISSOCK(nd.dentry->d_inode->i_mode)) 718 goto put_fail; 719 u=unix_find_socket_byinode(nd.dentry->d_inode); 720 if (!u) 721 goto put_fail; 722 723 if (u->sk_type == type) 724 touch_atime(nd.mnt, nd.dentry); 725 726 path_release(&nd); 727 728 err=-EPROTOTYPE; 729 if (u->sk_type != type) { 730 sock_put(u); 731 goto fail; 732 } 733 } else { 734 err = -ECONNREFUSED; 735 u=unix_find_socket_byname(sunname, len, type, hash); 736 if (u) { 737 struct dentry *dentry; 738 dentry = unix_sk(u)->dentry; 739 if (dentry) 740 touch_atime(unix_sk(u)->mnt, dentry); 741 } else 742 goto fail; 743 } 744 return u; 745 746 put_fail: 747 path_release(&nd); 748 fail: 749 *error=err; 750 return NULL; 751 } 752 753 754 static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len) 755 { 756 struct sock *sk = sock->sk; 757 struct unix_sock *u = unix_sk(sk); 758 struct sockaddr_un *sunaddr=(struct sockaddr_un *)uaddr; 759 struct dentry * dentry = NULL; 760 struct nameidata nd; 761 int err; 762 unsigned hash; 763 struct unix_address *addr; 764 struct hlist_head *list; 765 766 err = -EINVAL; 767 if (sunaddr->sun_family != AF_UNIX) 768 goto out; 769 770 if (addr_len==sizeof(short)) { 771 err = unix_autobind(sock); 772 goto out; 773 } 774 775 err = unix_mkname(sunaddr, addr_len, &hash); 776 if (err < 0) 777 goto out; 778 addr_len = err; 779 780 mutex_lock(&u->readlock); 781 782 err = -EINVAL; 783 if (u->addr) 784 goto out_up; 785 786 err = -ENOMEM; 787 addr = kmalloc(sizeof(*addr)+addr_len, GFP_KERNEL); 788 if (!addr) 789 goto out_up; 790 791 memcpy(addr->name, sunaddr, addr_len); 792 addr->len = addr_len; 793 addr->hash = hash ^ sk->sk_type; 794 atomic_set(&addr->refcnt, 1); 795 796 if (sunaddr->sun_path[0]) { 797 unsigned int mode; 798 err = 0; 799 /* 800 * Get the parent directory, calculate the hash for last 801 * component. 802 */ 803 err = path_lookup(sunaddr->sun_path, LOOKUP_PARENT, &nd); 804 if (err) 805 goto out_mknod_parent; 806 807 dentry = lookup_create(&nd, 0); 808 err = PTR_ERR(dentry); 809 if (IS_ERR(dentry)) 810 goto out_mknod_unlock; 811 812 /* 813 * All right, let's create it. 814 */ 815 mode = S_IFSOCK | 816 (SOCK_INODE(sock)->i_mode & ~current->fs->umask); 817 err = vfs_mknod(nd.dentry->d_inode, dentry, mode, 0); 818 if (err) 819 goto out_mknod_dput; 820 mutex_unlock(&nd.dentry->d_inode->i_mutex); 821 dput(nd.dentry); 822 nd.dentry = dentry; 823 824 addr->hash = UNIX_HASH_SIZE; 825 } 826 827 spin_lock(&unix_table_lock); 828 829 if (!sunaddr->sun_path[0]) { 830 err = -EADDRINUSE; 831 if (__unix_find_socket_byname(sunaddr, addr_len, 832 sk->sk_type, hash)) { 833 unix_release_addr(addr); 834 goto out_unlock; 835 } 836 837 list = &unix_socket_table[addr->hash]; 838 } else { 839 list = &unix_socket_table[dentry->d_inode->i_ino & (UNIX_HASH_SIZE-1)]; 840 u->dentry = nd.dentry; 841 u->mnt = nd.mnt; 842 } 843 844 err = 0; 845 __unix_remove_socket(sk); 846 u->addr = addr; 847 __unix_insert_socket(list, sk); 848 849 out_unlock: 850 spin_unlock(&unix_table_lock); 851 out_up: 852 mutex_unlock(&u->readlock); 853 out: 854 return err; 855 856 out_mknod_dput: 857 dput(dentry); 858 out_mknod_unlock: 859 mutex_unlock(&nd.dentry->d_inode->i_mutex); 860 path_release(&nd); 861 out_mknod_parent: 862 if (err==-EEXIST) 863 err=-EADDRINUSE; 864 unix_release_addr(addr); 865 goto out_up; 866 } 867 868 static int unix_dgram_connect(struct socket *sock, struct sockaddr *addr, 869 int alen, int flags) 870 { 871 struct sock *sk = sock->sk; 872 struct sockaddr_un *sunaddr=(struct sockaddr_un*)addr; 873 struct sock *other; 874 unsigned hash; 875 int err; 876 877 if (addr->sa_family != AF_UNSPEC) { 878 err = unix_mkname(sunaddr, alen, &hash); 879 if (err < 0) 880 goto out; 881 alen = err; 882 883 if (test_bit(SOCK_PASSCRED, &sock->flags) && 884 !unix_sk(sk)->addr && (err = unix_autobind(sock)) != 0) 885 goto out; 886 887 other=unix_find_other(sunaddr, alen, sock->type, hash, &err); 888 if (!other) 889 goto out; 890 891 unix_state_wlock(sk); 892 893 err = -EPERM; 894 if (!unix_may_send(sk, other)) 895 goto out_unlock; 896 897 err = security_unix_may_send(sk->sk_socket, other->sk_socket); 898 if (err) 899 goto out_unlock; 900 901 } else { 902 /* 903 * 1003.1g breaking connected state with AF_UNSPEC 904 */ 905 other = NULL; 906 unix_state_wlock(sk); 907 } 908 909 /* 910 * If it was connected, reconnect. 911 */ 912 if (unix_peer(sk)) { 913 struct sock *old_peer = unix_peer(sk); 914 unix_peer(sk)=other; 915 unix_state_wunlock(sk); 916 917 if (other != old_peer) 918 unix_dgram_disconnected(sk, old_peer); 919 sock_put(old_peer); 920 } else { 921 unix_peer(sk)=other; 922 unix_state_wunlock(sk); 923 } 924 return 0; 925 926 out_unlock: 927 unix_state_wunlock(sk); 928 sock_put(other); 929 out: 930 return err; 931 } 932 933 static long unix_wait_for_peer(struct sock *other, long timeo) 934 { 935 struct unix_sock *u = unix_sk(other); 936 int sched; 937 DEFINE_WAIT(wait); 938 939 prepare_to_wait_exclusive(&u->peer_wait, &wait, TASK_INTERRUPTIBLE); 940 941 sched = !sock_flag(other, SOCK_DEAD) && 942 !(other->sk_shutdown & RCV_SHUTDOWN) && 943 (skb_queue_len(&other->sk_receive_queue) > 944 other->sk_max_ack_backlog); 945 946 unix_state_runlock(other); 947 948 if (sched) 949 timeo = schedule_timeout(timeo); 950 951 finish_wait(&u->peer_wait, &wait); 952 return timeo; 953 } 954 955 static int unix_stream_connect(struct socket *sock, struct sockaddr *uaddr, 956 int addr_len, int flags) 957 { 958 struct sockaddr_un *sunaddr=(struct sockaddr_un *)uaddr; 959 struct sock *sk = sock->sk; 960 struct unix_sock *u = unix_sk(sk), *newu, *otheru; 961 struct sock *newsk = NULL; 962 struct sock *other = NULL; 963 struct sk_buff *skb = NULL; 964 unsigned hash; 965 int st; 966 int err; 967 long timeo; 968 969 err = unix_mkname(sunaddr, addr_len, &hash); 970 if (err < 0) 971 goto out; 972 addr_len = err; 973 974 if (test_bit(SOCK_PASSCRED, &sock->flags) 975 && !u->addr && (err = unix_autobind(sock)) != 0) 976 goto out; 977 978 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK); 979 980 /* First of all allocate resources. 981 If we will make it after state is locked, 982 we will have to recheck all again in any case. 983 */ 984 985 err = -ENOMEM; 986 987 /* create new sock for complete connection */ 988 newsk = unix_create1(NULL); 989 if (newsk == NULL) 990 goto out; 991 992 /* Allocate skb for sending to listening sock */ 993 skb = sock_wmalloc(newsk, 1, 0, GFP_KERNEL); 994 if (skb == NULL) 995 goto out; 996 997 restart: 998 /* Find listening sock. */ 999 other = unix_find_other(sunaddr, addr_len, sk->sk_type, hash, &err); 1000 if (!other) 1001 goto out; 1002 1003 /* Latch state of peer */ 1004 unix_state_rlock(other); 1005 1006 /* Apparently VFS overslept socket death. Retry. */ 1007 if (sock_flag(other, SOCK_DEAD)) { 1008 unix_state_runlock(other); 1009 sock_put(other); 1010 goto restart; 1011 } 1012 1013 err = -ECONNREFUSED; 1014 if (other->sk_state != TCP_LISTEN) 1015 goto out_unlock; 1016 1017 if (skb_queue_len(&other->sk_receive_queue) > 1018 other->sk_max_ack_backlog) { 1019 err = -EAGAIN; 1020 if (!timeo) 1021 goto out_unlock; 1022 1023 timeo = unix_wait_for_peer(other, timeo); 1024 1025 err = sock_intr_errno(timeo); 1026 if (signal_pending(current)) 1027 goto out; 1028 sock_put(other); 1029 goto restart; 1030 } 1031 1032 /* Latch our state. 1033 1034 It is tricky place. We need to grab write lock and cannot 1035 drop lock on peer. It is dangerous because deadlock is 1036 possible. Connect to self case and simultaneous 1037 attempt to connect are eliminated by checking socket 1038 state. other is TCP_LISTEN, if sk is TCP_LISTEN we 1039 check this before attempt to grab lock. 1040 1041 Well, and we have to recheck the state after socket locked. 1042 */ 1043 st = sk->sk_state; 1044 1045 switch (st) { 1046 case TCP_CLOSE: 1047 /* This is ok... continue with connect */ 1048 break; 1049 case TCP_ESTABLISHED: 1050 /* Socket is already connected */ 1051 err = -EISCONN; 1052 goto out_unlock; 1053 default: 1054 err = -EINVAL; 1055 goto out_unlock; 1056 } 1057 1058 unix_state_wlock_nested(sk); 1059 1060 if (sk->sk_state != st) { 1061 unix_state_wunlock(sk); 1062 unix_state_runlock(other); 1063 sock_put(other); 1064 goto restart; 1065 } 1066 1067 err = security_unix_stream_connect(sock, other->sk_socket, newsk); 1068 if (err) { 1069 unix_state_wunlock(sk); 1070 goto out_unlock; 1071 } 1072 1073 /* The way is open! Fastly set all the necessary fields... */ 1074 1075 sock_hold(sk); 1076 unix_peer(newsk) = sk; 1077 newsk->sk_state = TCP_ESTABLISHED; 1078 newsk->sk_type = sk->sk_type; 1079 newsk->sk_peercred.pid = current->tgid; 1080 newsk->sk_peercred.uid = current->euid; 1081 newsk->sk_peercred.gid = current->egid; 1082 newu = unix_sk(newsk); 1083 newsk->sk_sleep = &newu->peer_wait; 1084 otheru = unix_sk(other); 1085 1086 /* copy address information from listening to new sock*/ 1087 if (otheru->addr) { 1088 atomic_inc(&otheru->addr->refcnt); 1089 newu->addr = otheru->addr; 1090 } 1091 if (otheru->dentry) { 1092 newu->dentry = dget(otheru->dentry); 1093 newu->mnt = mntget(otheru->mnt); 1094 } 1095 1096 /* Set credentials */ 1097 sk->sk_peercred = other->sk_peercred; 1098 1099 sock->state = SS_CONNECTED; 1100 sk->sk_state = TCP_ESTABLISHED; 1101 sock_hold(newsk); 1102 1103 smp_mb__after_atomic_inc(); /* sock_hold() does an atomic_inc() */ 1104 unix_peer(sk) = newsk; 1105 1106 unix_state_wunlock(sk); 1107 1108 /* take ten and and send info to listening sock */ 1109 spin_lock(&other->sk_receive_queue.lock); 1110 __skb_queue_tail(&other->sk_receive_queue, skb); 1111 /* Undo artificially decreased inflight after embrion 1112 * is installed to listening socket. */ 1113 atomic_inc(&newu->inflight); 1114 spin_unlock(&other->sk_receive_queue.lock); 1115 unix_state_runlock(other); 1116 other->sk_data_ready(other, 0); 1117 sock_put(other); 1118 return 0; 1119 1120 out_unlock: 1121 if (other) 1122 unix_state_runlock(other); 1123 1124 out: 1125 if (skb) 1126 kfree_skb(skb); 1127 if (newsk) 1128 unix_release_sock(newsk, 0); 1129 if (other) 1130 sock_put(other); 1131 return err; 1132 } 1133 1134 static int unix_socketpair(struct socket *socka, struct socket *sockb) 1135 { 1136 struct sock *ska=socka->sk, *skb = sockb->sk; 1137 1138 /* Join our sockets back to back */ 1139 sock_hold(ska); 1140 sock_hold(skb); 1141 unix_peer(ska)=skb; 1142 unix_peer(skb)=ska; 1143 ska->sk_peercred.pid = skb->sk_peercred.pid = current->tgid; 1144 ska->sk_peercred.uid = skb->sk_peercred.uid = current->euid; 1145 ska->sk_peercred.gid = skb->sk_peercred.gid = current->egid; 1146 1147 if (ska->sk_type != SOCK_DGRAM) { 1148 ska->sk_state = TCP_ESTABLISHED; 1149 skb->sk_state = TCP_ESTABLISHED; 1150 socka->state = SS_CONNECTED; 1151 sockb->state = SS_CONNECTED; 1152 } 1153 return 0; 1154 } 1155 1156 static int unix_accept(struct socket *sock, struct socket *newsock, int flags) 1157 { 1158 struct sock *sk = sock->sk; 1159 struct sock *tsk; 1160 struct sk_buff *skb; 1161 int err; 1162 1163 err = -EOPNOTSUPP; 1164 if (sock->type!=SOCK_STREAM && sock->type!=SOCK_SEQPACKET) 1165 goto out; 1166 1167 err = -EINVAL; 1168 if (sk->sk_state != TCP_LISTEN) 1169 goto out; 1170 1171 /* If socket state is TCP_LISTEN it cannot change (for now...), 1172 * so that no locks are necessary. 1173 */ 1174 1175 skb = skb_recv_datagram(sk, 0, flags&O_NONBLOCK, &err); 1176 if (!skb) { 1177 /* This means receive shutdown. */ 1178 if (err == 0) 1179 err = -EINVAL; 1180 goto out; 1181 } 1182 1183 tsk = skb->sk; 1184 skb_free_datagram(sk, skb); 1185 wake_up_interruptible(&unix_sk(sk)->peer_wait); 1186 1187 /* attach accepted sock to socket */ 1188 unix_state_wlock(tsk); 1189 newsock->state = SS_CONNECTED; 1190 sock_graft(tsk, newsock); 1191 unix_state_wunlock(tsk); 1192 return 0; 1193 1194 out: 1195 return err; 1196 } 1197 1198 1199 static int unix_getname(struct socket *sock, struct sockaddr *uaddr, int *uaddr_len, int peer) 1200 { 1201 struct sock *sk = sock->sk; 1202 struct unix_sock *u; 1203 struct sockaddr_un *sunaddr=(struct sockaddr_un *)uaddr; 1204 int err = 0; 1205 1206 if (peer) { 1207 sk = unix_peer_get(sk); 1208 1209 err = -ENOTCONN; 1210 if (!sk) 1211 goto out; 1212 err = 0; 1213 } else { 1214 sock_hold(sk); 1215 } 1216 1217 u = unix_sk(sk); 1218 unix_state_rlock(sk); 1219 if (!u->addr) { 1220 sunaddr->sun_family = AF_UNIX; 1221 sunaddr->sun_path[0] = 0; 1222 *uaddr_len = sizeof(short); 1223 } else { 1224 struct unix_address *addr = u->addr; 1225 1226 *uaddr_len = addr->len; 1227 memcpy(sunaddr, addr->name, *uaddr_len); 1228 } 1229 unix_state_runlock(sk); 1230 sock_put(sk); 1231 out: 1232 return err; 1233 } 1234 1235 static void unix_detach_fds(struct scm_cookie *scm, struct sk_buff *skb) 1236 { 1237 int i; 1238 1239 scm->fp = UNIXCB(skb).fp; 1240 skb->destructor = sock_wfree; 1241 UNIXCB(skb).fp = NULL; 1242 1243 for (i=scm->fp->count-1; i>=0; i--) 1244 unix_notinflight(scm->fp->fp[i]); 1245 } 1246 1247 static void unix_destruct_fds(struct sk_buff *skb) 1248 { 1249 struct scm_cookie scm; 1250 memset(&scm, 0, sizeof(scm)); 1251 unix_detach_fds(&scm, skb); 1252 1253 /* Alas, it calls VFS */ 1254 /* So fscking what? fput() had been SMP-safe since the last Summer */ 1255 scm_destroy(&scm); 1256 sock_wfree(skb); 1257 } 1258 1259 static void unix_attach_fds(struct scm_cookie *scm, struct sk_buff *skb) 1260 { 1261 int i; 1262 for (i=scm->fp->count-1; i>=0; i--) 1263 unix_inflight(scm->fp->fp[i]); 1264 UNIXCB(skb).fp = scm->fp; 1265 skb->destructor = unix_destruct_fds; 1266 scm->fp = NULL; 1267 } 1268 1269 /* 1270 * Send AF_UNIX data. 1271 */ 1272 1273 static int unix_dgram_sendmsg(struct kiocb *kiocb, struct socket *sock, 1274 struct msghdr *msg, size_t len) 1275 { 1276 struct sock_iocb *siocb = kiocb_to_siocb(kiocb); 1277 struct sock *sk = sock->sk; 1278 struct unix_sock *u = unix_sk(sk); 1279 struct sockaddr_un *sunaddr=msg->msg_name; 1280 struct sock *other = NULL; 1281 int namelen = 0; /* fake GCC */ 1282 int err; 1283 unsigned hash; 1284 struct sk_buff *skb; 1285 long timeo; 1286 struct scm_cookie tmp_scm; 1287 1288 if (NULL == siocb->scm) 1289 siocb->scm = &tmp_scm; 1290 err = scm_send(sock, msg, siocb->scm); 1291 if (err < 0) 1292 return err; 1293 1294 err = -EOPNOTSUPP; 1295 if (msg->msg_flags&MSG_OOB) 1296 goto out; 1297 1298 if (msg->msg_namelen) { 1299 err = unix_mkname(sunaddr, msg->msg_namelen, &hash); 1300 if (err < 0) 1301 goto out; 1302 namelen = err; 1303 } else { 1304 sunaddr = NULL; 1305 err = -ENOTCONN; 1306 other = unix_peer_get(sk); 1307 if (!other) 1308 goto out; 1309 } 1310 1311 if (test_bit(SOCK_PASSCRED, &sock->flags) 1312 && !u->addr && (err = unix_autobind(sock)) != 0) 1313 goto out; 1314 1315 err = -EMSGSIZE; 1316 if (len > sk->sk_sndbuf - 32) 1317 goto out; 1318 1319 skb = sock_alloc_send_skb(sk, len, msg->msg_flags&MSG_DONTWAIT, &err); 1320 if (skb==NULL) 1321 goto out; 1322 1323 memcpy(UNIXCREDS(skb), &siocb->scm->creds, sizeof(struct ucred)); 1324 if (siocb->scm->fp) 1325 unix_attach_fds(siocb->scm, skb); 1326 1327 unix_get_peersec_dgram(skb); 1328 1329 skb->h.raw = skb->data; 1330 err = memcpy_fromiovec(skb_put(skb,len), msg->msg_iov, len); 1331 if (err) 1332 goto out_free; 1333 1334 timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT); 1335 1336 restart: 1337 if (!other) { 1338 err = -ECONNRESET; 1339 if (sunaddr == NULL) 1340 goto out_free; 1341 1342 other = unix_find_other(sunaddr, namelen, sk->sk_type, 1343 hash, &err); 1344 if (other==NULL) 1345 goto out_free; 1346 } 1347 1348 unix_state_rlock(other); 1349 err = -EPERM; 1350 if (!unix_may_send(sk, other)) 1351 goto out_unlock; 1352 1353 if (sock_flag(other, SOCK_DEAD)) { 1354 /* 1355 * Check with 1003.1g - what should 1356 * datagram error 1357 */ 1358 unix_state_runlock(other); 1359 sock_put(other); 1360 1361 err = 0; 1362 unix_state_wlock(sk); 1363 if (unix_peer(sk) == other) { 1364 unix_peer(sk)=NULL; 1365 unix_state_wunlock(sk); 1366 1367 unix_dgram_disconnected(sk, other); 1368 sock_put(other); 1369 err = -ECONNREFUSED; 1370 } else { 1371 unix_state_wunlock(sk); 1372 } 1373 1374 other = NULL; 1375 if (err) 1376 goto out_free; 1377 goto restart; 1378 } 1379 1380 err = -EPIPE; 1381 if (other->sk_shutdown & RCV_SHUTDOWN) 1382 goto out_unlock; 1383 1384 if (sk->sk_type != SOCK_SEQPACKET) { 1385 err = security_unix_may_send(sk->sk_socket, other->sk_socket); 1386 if (err) 1387 goto out_unlock; 1388 } 1389 1390 if (unix_peer(other) != sk && 1391 (skb_queue_len(&other->sk_receive_queue) > 1392 other->sk_max_ack_backlog)) { 1393 if (!timeo) { 1394 err = -EAGAIN; 1395 goto out_unlock; 1396 } 1397 1398 timeo = unix_wait_for_peer(other, timeo); 1399 1400 err = sock_intr_errno(timeo); 1401 if (signal_pending(current)) 1402 goto out_free; 1403 1404 goto restart; 1405 } 1406 1407 skb_queue_tail(&other->sk_receive_queue, skb); 1408 unix_state_runlock(other); 1409 other->sk_data_ready(other, len); 1410 sock_put(other); 1411 scm_destroy(siocb->scm); 1412 return len; 1413 1414 out_unlock: 1415 unix_state_runlock(other); 1416 out_free: 1417 kfree_skb(skb); 1418 out: 1419 if (other) 1420 sock_put(other); 1421 scm_destroy(siocb->scm); 1422 return err; 1423 } 1424 1425 1426 static int unix_stream_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 sock *other = NULL; 1432 struct sockaddr_un *sunaddr=msg->msg_name; 1433 int err,size; 1434 struct sk_buff *skb; 1435 int sent=0; 1436 struct scm_cookie tmp_scm; 1437 1438 if (NULL == siocb->scm) 1439 siocb->scm = &tmp_scm; 1440 err = scm_send(sock, msg, siocb->scm); 1441 if (err < 0) 1442 return err; 1443 1444 err = -EOPNOTSUPP; 1445 if (msg->msg_flags&MSG_OOB) 1446 goto out_err; 1447 1448 if (msg->msg_namelen) { 1449 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP; 1450 goto out_err; 1451 } else { 1452 sunaddr = NULL; 1453 err = -ENOTCONN; 1454 other = unix_peer(sk); 1455 if (!other) 1456 goto out_err; 1457 } 1458 1459 if (sk->sk_shutdown & SEND_SHUTDOWN) 1460 goto pipe_err; 1461 1462 while(sent < len) 1463 { 1464 /* 1465 * Optimisation for the fact that under 0.01% of X 1466 * messages typically need breaking up. 1467 */ 1468 1469 size = len-sent; 1470 1471 /* Keep two messages in the pipe so it schedules better */ 1472 if (size > ((sk->sk_sndbuf >> 1) - 64)) 1473 size = (sk->sk_sndbuf >> 1) - 64; 1474 1475 if (size > SKB_MAX_ALLOC) 1476 size = SKB_MAX_ALLOC; 1477 1478 /* 1479 * Grab a buffer 1480 */ 1481 1482 skb=sock_alloc_send_skb(sk,size,msg->msg_flags&MSG_DONTWAIT, &err); 1483 1484 if (skb==NULL) 1485 goto out_err; 1486 1487 /* 1488 * If you pass two values to the sock_alloc_send_skb 1489 * it tries to grab the large buffer with GFP_NOFS 1490 * (which can fail easily), and if it fails grab the 1491 * fallback size buffer which is under a page and will 1492 * succeed. [Alan] 1493 */ 1494 size = min_t(int, size, skb_tailroom(skb)); 1495 1496 memcpy(UNIXCREDS(skb), &siocb->scm->creds, sizeof(struct ucred)); 1497 if (siocb->scm->fp) 1498 unix_attach_fds(siocb->scm, skb); 1499 1500 if ((err = memcpy_fromiovec(skb_put(skb,size), msg->msg_iov, size)) != 0) { 1501 kfree_skb(skb); 1502 goto out_err; 1503 } 1504 1505 unix_state_rlock(other); 1506 1507 if (sock_flag(other, SOCK_DEAD) || 1508 (other->sk_shutdown & RCV_SHUTDOWN)) 1509 goto pipe_err_free; 1510 1511 skb_queue_tail(&other->sk_receive_queue, skb); 1512 unix_state_runlock(other); 1513 other->sk_data_ready(other, size); 1514 sent+=size; 1515 } 1516 1517 scm_destroy(siocb->scm); 1518 siocb->scm = NULL; 1519 1520 return sent; 1521 1522 pipe_err_free: 1523 unix_state_runlock(other); 1524 kfree_skb(skb); 1525 pipe_err: 1526 if (sent==0 && !(msg->msg_flags&MSG_NOSIGNAL)) 1527 send_sig(SIGPIPE,current,0); 1528 err = -EPIPE; 1529 out_err: 1530 scm_destroy(siocb->scm); 1531 siocb->scm = NULL; 1532 return sent ? : err; 1533 } 1534 1535 static int unix_seqpacket_sendmsg(struct kiocb *kiocb, struct socket *sock, 1536 struct msghdr *msg, size_t len) 1537 { 1538 int err; 1539 struct sock *sk = sock->sk; 1540 1541 err = sock_error(sk); 1542 if (err) 1543 return err; 1544 1545 if (sk->sk_state != TCP_ESTABLISHED) 1546 return -ENOTCONN; 1547 1548 if (msg->msg_namelen) 1549 msg->msg_namelen = 0; 1550 1551 return unix_dgram_sendmsg(kiocb, sock, msg, len); 1552 } 1553 1554 static void unix_copy_addr(struct msghdr *msg, struct sock *sk) 1555 { 1556 struct unix_sock *u = unix_sk(sk); 1557 1558 msg->msg_namelen = 0; 1559 if (u->addr) { 1560 msg->msg_namelen = u->addr->len; 1561 memcpy(msg->msg_name, u->addr->name, u->addr->len); 1562 } 1563 } 1564 1565 static int unix_dgram_recvmsg(struct kiocb *iocb, struct socket *sock, 1566 struct msghdr *msg, size_t size, 1567 int flags) 1568 { 1569 struct sock_iocb *siocb = kiocb_to_siocb(iocb); 1570 struct scm_cookie tmp_scm; 1571 struct sock *sk = sock->sk; 1572 struct unix_sock *u = unix_sk(sk); 1573 int noblock = flags & MSG_DONTWAIT; 1574 struct sk_buff *skb; 1575 int err; 1576 1577 err = -EOPNOTSUPP; 1578 if (flags&MSG_OOB) 1579 goto out; 1580 1581 msg->msg_namelen = 0; 1582 1583 mutex_lock(&u->readlock); 1584 1585 skb = skb_recv_datagram(sk, flags, noblock, &err); 1586 if (!skb) 1587 goto out_unlock; 1588 1589 wake_up_interruptible(&u->peer_wait); 1590 1591 if (msg->msg_name) 1592 unix_copy_addr(msg, skb->sk); 1593 1594 if (size > skb->len) 1595 size = skb->len; 1596 else if (size < skb->len) 1597 msg->msg_flags |= MSG_TRUNC; 1598 1599 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, size); 1600 if (err) 1601 goto out_free; 1602 1603 if (!siocb->scm) { 1604 siocb->scm = &tmp_scm; 1605 memset(&tmp_scm, 0, sizeof(tmp_scm)); 1606 } 1607 siocb->scm->creds = *UNIXCREDS(skb); 1608 unix_set_secdata(siocb->scm, skb); 1609 1610 if (!(flags & MSG_PEEK)) 1611 { 1612 if (UNIXCB(skb).fp) 1613 unix_detach_fds(siocb->scm, skb); 1614 } 1615 else 1616 { 1617 /* It is questionable: on PEEK we could: 1618 - do not return fds - good, but too simple 8) 1619 - return fds, and do not return them on read (old strategy, 1620 apparently wrong) 1621 - clone fds (I chose it for now, it is the most universal 1622 solution) 1623 1624 POSIX 1003.1g does not actually define this clearly 1625 at all. POSIX 1003.1g doesn't define a lot of things 1626 clearly however! 1627 1628 */ 1629 if (UNIXCB(skb).fp) 1630 siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp); 1631 } 1632 err = size; 1633 1634 scm_recv(sock, msg, siocb->scm, flags); 1635 1636 out_free: 1637 skb_free_datagram(sk,skb); 1638 out_unlock: 1639 mutex_unlock(&u->readlock); 1640 out: 1641 return err; 1642 } 1643 1644 /* 1645 * Sleep until data has arrive. But check for races.. 1646 */ 1647 1648 static long unix_stream_data_wait(struct sock * sk, long timeo) 1649 { 1650 DEFINE_WAIT(wait); 1651 1652 unix_state_rlock(sk); 1653 1654 for (;;) { 1655 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE); 1656 1657 if (!skb_queue_empty(&sk->sk_receive_queue) || 1658 sk->sk_err || 1659 (sk->sk_shutdown & RCV_SHUTDOWN) || 1660 signal_pending(current) || 1661 !timeo) 1662 break; 1663 1664 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags); 1665 unix_state_runlock(sk); 1666 timeo = schedule_timeout(timeo); 1667 unix_state_rlock(sk); 1668 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags); 1669 } 1670 1671 finish_wait(sk->sk_sleep, &wait); 1672 unix_state_runlock(sk); 1673 return timeo; 1674 } 1675 1676 1677 1678 static int unix_stream_recvmsg(struct kiocb *iocb, struct socket *sock, 1679 struct msghdr *msg, size_t size, 1680 int flags) 1681 { 1682 struct sock_iocb *siocb = kiocb_to_siocb(iocb); 1683 struct scm_cookie tmp_scm; 1684 struct sock *sk = sock->sk; 1685 struct unix_sock *u = unix_sk(sk); 1686 struct sockaddr_un *sunaddr=msg->msg_name; 1687 int copied = 0; 1688 int check_creds = 0; 1689 int target; 1690 int err = 0; 1691 long timeo; 1692 1693 err = -EINVAL; 1694 if (sk->sk_state != TCP_ESTABLISHED) 1695 goto out; 1696 1697 err = -EOPNOTSUPP; 1698 if (flags&MSG_OOB) 1699 goto out; 1700 1701 target = sock_rcvlowat(sk, flags&MSG_WAITALL, size); 1702 timeo = sock_rcvtimeo(sk, flags&MSG_DONTWAIT); 1703 1704 msg->msg_namelen = 0; 1705 1706 /* Lock the socket to prevent queue disordering 1707 * while sleeps in memcpy_tomsg 1708 */ 1709 1710 if (!siocb->scm) { 1711 siocb->scm = &tmp_scm; 1712 memset(&tmp_scm, 0, sizeof(tmp_scm)); 1713 } 1714 1715 mutex_lock(&u->readlock); 1716 1717 do 1718 { 1719 int chunk; 1720 struct sk_buff *skb; 1721 1722 skb = skb_dequeue(&sk->sk_receive_queue); 1723 if (skb==NULL) 1724 { 1725 if (copied >= target) 1726 break; 1727 1728 /* 1729 * POSIX 1003.1g mandates this order. 1730 */ 1731 1732 if ((err = sock_error(sk)) != 0) 1733 break; 1734 if (sk->sk_shutdown & RCV_SHUTDOWN) 1735 break; 1736 err = -EAGAIN; 1737 if (!timeo) 1738 break; 1739 mutex_unlock(&u->readlock); 1740 1741 timeo = unix_stream_data_wait(sk, timeo); 1742 1743 if (signal_pending(current)) { 1744 err = sock_intr_errno(timeo); 1745 goto out; 1746 } 1747 mutex_lock(&u->readlock); 1748 continue; 1749 } 1750 1751 if (check_creds) { 1752 /* Never glue messages from different writers */ 1753 if (memcmp(UNIXCREDS(skb), &siocb->scm->creds, sizeof(siocb->scm->creds)) != 0) { 1754 skb_queue_head(&sk->sk_receive_queue, skb); 1755 break; 1756 } 1757 } else { 1758 /* Copy credentials */ 1759 siocb->scm->creds = *UNIXCREDS(skb); 1760 check_creds = 1; 1761 } 1762 1763 /* Copy address just once */ 1764 if (sunaddr) 1765 { 1766 unix_copy_addr(msg, skb->sk); 1767 sunaddr = NULL; 1768 } 1769 1770 chunk = min_t(unsigned int, skb->len, size); 1771 if (memcpy_toiovec(msg->msg_iov, skb->data, chunk)) { 1772 skb_queue_head(&sk->sk_receive_queue, skb); 1773 if (copied == 0) 1774 copied = -EFAULT; 1775 break; 1776 } 1777 copied += chunk; 1778 size -= chunk; 1779 1780 /* Mark read part of skb as used */ 1781 if (!(flags & MSG_PEEK)) 1782 { 1783 skb_pull(skb, chunk); 1784 1785 if (UNIXCB(skb).fp) 1786 unix_detach_fds(siocb->scm, skb); 1787 1788 /* put the skb back if we didn't use it up.. */ 1789 if (skb->len) 1790 { 1791 skb_queue_head(&sk->sk_receive_queue, skb); 1792 break; 1793 } 1794 1795 kfree_skb(skb); 1796 1797 if (siocb->scm->fp) 1798 break; 1799 } 1800 else 1801 { 1802 /* It is questionable, see note in unix_dgram_recvmsg. 1803 */ 1804 if (UNIXCB(skb).fp) 1805 siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp); 1806 1807 /* put message back and return */ 1808 skb_queue_head(&sk->sk_receive_queue, skb); 1809 break; 1810 } 1811 } while (size); 1812 1813 mutex_unlock(&u->readlock); 1814 scm_recv(sock, msg, siocb->scm, flags); 1815 out: 1816 return copied ? : err; 1817 } 1818 1819 static int unix_shutdown(struct socket *sock, int mode) 1820 { 1821 struct sock *sk = sock->sk; 1822 struct sock *other; 1823 1824 mode = (mode+1)&(RCV_SHUTDOWN|SEND_SHUTDOWN); 1825 1826 if (mode) { 1827 unix_state_wlock(sk); 1828 sk->sk_shutdown |= mode; 1829 other=unix_peer(sk); 1830 if (other) 1831 sock_hold(other); 1832 unix_state_wunlock(sk); 1833 sk->sk_state_change(sk); 1834 1835 if (other && 1836 (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET)) { 1837 1838 int peer_mode = 0; 1839 1840 if (mode&RCV_SHUTDOWN) 1841 peer_mode |= SEND_SHUTDOWN; 1842 if (mode&SEND_SHUTDOWN) 1843 peer_mode |= RCV_SHUTDOWN; 1844 unix_state_wlock(other); 1845 other->sk_shutdown |= peer_mode; 1846 unix_state_wunlock(other); 1847 other->sk_state_change(other); 1848 read_lock(&other->sk_callback_lock); 1849 if (peer_mode == SHUTDOWN_MASK) 1850 sk_wake_async(other,1,POLL_HUP); 1851 else if (peer_mode & RCV_SHUTDOWN) 1852 sk_wake_async(other,1,POLL_IN); 1853 read_unlock(&other->sk_callback_lock); 1854 } 1855 if (other) 1856 sock_put(other); 1857 } 1858 return 0; 1859 } 1860 1861 static int unix_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 1862 { 1863 struct sock *sk = sock->sk; 1864 long amount=0; 1865 int err; 1866 1867 switch(cmd) 1868 { 1869 case SIOCOUTQ: 1870 amount = atomic_read(&sk->sk_wmem_alloc); 1871 err = put_user(amount, (int __user *)arg); 1872 break; 1873 case SIOCINQ: 1874 { 1875 struct sk_buff *skb; 1876 1877 if (sk->sk_state == TCP_LISTEN) { 1878 err = -EINVAL; 1879 break; 1880 } 1881 1882 spin_lock(&sk->sk_receive_queue.lock); 1883 if (sk->sk_type == SOCK_STREAM || 1884 sk->sk_type == SOCK_SEQPACKET) { 1885 skb_queue_walk(&sk->sk_receive_queue, skb) 1886 amount += skb->len; 1887 } else { 1888 skb = skb_peek(&sk->sk_receive_queue); 1889 if (skb) 1890 amount=skb->len; 1891 } 1892 spin_unlock(&sk->sk_receive_queue.lock); 1893 err = put_user(amount, (int __user *)arg); 1894 break; 1895 } 1896 1897 default: 1898 err = -ENOIOCTLCMD; 1899 break; 1900 } 1901 return err; 1902 } 1903 1904 static unsigned int unix_poll(struct file * file, struct socket *sock, poll_table *wait) 1905 { 1906 struct sock *sk = sock->sk; 1907 unsigned int mask; 1908 1909 poll_wait(file, sk->sk_sleep, wait); 1910 mask = 0; 1911 1912 /* exceptional events? */ 1913 if (sk->sk_err) 1914 mask |= POLLERR; 1915 if (sk->sk_shutdown == SHUTDOWN_MASK) 1916 mask |= POLLHUP; 1917 if (sk->sk_shutdown & RCV_SHUTDOWN) 1918 mask |= POLLRDHUP; 1919 1920 /* readable? */ 1921 if (!skb_queue_empty(&sk->sk_receive_queue) || 1922 (sk->sk_shutdown & RCV_SHUTDOWN)) 1923 mask |= POLLIN | POLLRDNORM; 1924 1925 /* Connection-based need to check for termination and startup */ 1926 if ((sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) && sk->sk_state == TCP_CLOSE) 1927 mask |= POLLHUP; 1928 1929 /* 1930 * we set writable also when the other side has shut down the 1931 * connection. This prevents stuck sockets. 1932 */ 1933 if (unix_writable(sk)) 1934 mask |= POLLOUT | POLLWRNORM | POLLWRBAND; 1935 1936 return mask; 1937 } 1938 1939 1940 #ifdef CONFIG_PROC_FS 1941 static struct sock *unix_seq_idx(int *iter, loff_t pos) 1942 { 1943 loff_t off = 0; 1944 struct sock *s; 1945 1946 for (s = first_unix_socket(iter); s; s = next_unix_socket(iter, s)) { 1947 if (off == pos) 1948 return s; 1949 ++off; 1950 } 1951 return NULL; 1952 } 1953 1954 1955 static void *unix_seq_start(struct seq_file *seq, loff_t *pos) 1956 { 1957 spin_lock(&unix_table_lock); 1958 return *pos ? unix_seq_idx(seq->private, *pos - 1) : ((void *) 1); 1959 } 1960 1961 static void *unix_seq_next(struct seq_file *seq, void *v, loff_t *pos) 1962 { 1963 ++*pos; 1964 1965 if (v == (void *)1) 1966 return first_unix_socket(seq->private); 1967 return next_unix_socket(seq->private, v); 1968 } 1969 1970 static void unix_seq_stop(struct seq_file *seq, void *v) 1971 { 1972 spin_unlock(&unix_table_lock); 1973 } 1974 1975 static int unix_seq_show(struct seq_file *seq, void *v) 1976 { 1977 1978 if (v == (void *)1) 1979 seq_puts(seq, "Num RefCount Protocol Flags Type St " 1980 "Inode Path\n"); 1981 else { 1982 struct sock *s = v; 1983 struct unix_sock *u = unix_sk(s); 1984 unix_state_rlock(s); 1985 1986 seq_printf(seq, "%p: %08X %08X %08X %04X %02X %5lu", 1987 s, 1988 atomic_read(&s->sk_refcnt), 1989 0, 1990 s->sk_state == TCP_LISTEN ? __SO_ACCEPTCON : 0, 1991 s->sk_type, 1992 s->sk_socket ? 1993 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTED : SS_UNCONNECTED) : 1994 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTING : SS_DISCONNECTING), 1995 sock_i_ino(s)); 1996 1997 if (u->addr) { 1998 int i, len; 1999 seq_putc(seq, ' '); 2000 2001 i = 0; 2002 len = u->addr->len - sizeof(short); 2003 if (!UNIX_ABSTRACT(s)) 2004 len--; 2005 else { 2006 seq_putc(seq, '@'); 2007 i++; 2008 } 2009 for ( ; i < len; i++) 2010 seq_putc(seq, u->addr->name->sun_path[i]); 2011 } 2012 unix_state_runlock(s); 2013 seq_putc(seq, '\n'); 2014 } 2015 2016 return 0; 2017 } 2018 2019 static struct seq_operations unix_seq_ops = { 2020 .start = unix_seq_start, 2021 .next = unix_seq_next, 2022 .stop = unix_seq_stop, 2023 .show = unix_seq_show, 2024 }; 2025 2026 2027 static int unix_seq_open(struct inode *inode, struct file *file) 2028 { 2029 struct seq_file *seq; 2030 int rc = -ENOMEM; 2031 int *iter = kmalloc(sizeof(int), GFP_KERNEL); 2032 2033 if (!iter) 2034 goto out; 2035 2036 rc = seq_open(file, &unix_seq_ops); 2037 if (rc) 2038 goto out_kfree; 2039 2040 seq = file->private_data; 2041 seq->private = iter; 2042 *iter = 0; 2043 out: 2044 return rc; 2045 out_kfree: 2046 kfree(iter); 2047 goto out; 2048 } 2049 2050 static struct file_operations unix_seq_fops = { 2051 .owner = THIS_MODULE, 2052 .open = unix_seq_open, 2053 .read = seq_read, 2054 .llseek = seq_lseek, 2055 .release = seq_release_private, 2056 }; 2057 2058 #endif 2059 2060 static struct net_proto_family unix_family_ops = { 2061 .family = PF_UNIX, 2062 .create = unix_create, 2063 .owner = THIS_MODULE, 2064 }; 2065 2066 static int __init af_unix_init(void) 2067 { 2068 int rc = -1; 2069 struct sk_buff *dummy_skb; 2070 2071 if (sizeof(struct unix_skb_parms) > sizeof(dummy_skb->cb)) { 2072 printk(KERN_CRIT "%s: panic\n", __FUNCTION__); 2073 goto out; 2074 } 2075 2076 rc = proto_register(&unix_proto, 1); 2077 if (rc != 0) { 2078 printk(KERN_CRIT "%s: Cannot create unix_sock SLAB cache!\n", 2079 __FUNCTION__); 2080 goto out; 2081 } 2082 2083 sock_register(&unix_family_ops); 2084 #ifdef CONFIG_PROC_FS 2085 proc_net_fops_create("unix", 0, &unix_seq_fops); 2086 #endif 2087 unix_sysctl_register(); 2088 out: 2089 return rc; 2090 } 2091 2092 static void __exit af_unix_exit(void) 2093 { 2094 sock_unregister(PF_UNIX); 2095 unix_sysctl_unregister(); 2096 proc_net_remove("unix"); 2097 proto_unregister(&unix_proto); 2098 } 2099 2100 module_init(af_unix_init); 2101 module_exit(af_unix_exit); 2102 2103 MODULE_LICENSE("GPL"); 2104 MODULE_ALIAS_NETPROTO(PF_UNIX); 2105