af_unix.c revision 921a1650de9eed40dd64d681aba4a4d98856f289
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 118struct hlist_head unix_socket_table[UNIX_HASH_SIZE + 1]; 119EXPORT_SYMBOL_GPL(unix_socket_table); 120DEFINE_SPINLOCK(unix_table_lock); 121EXPORT_SYMBOL_GPL(unix_table_lock); 122static atomic_long_t unix_nr_socks; 123 124#define unix_sockets_unbound (&unix_socket_table[UNIX_HASH_SIZE]) 125 126#define UNIX_ABSTRACT(sk) (unix_sk(sk)->addr->hash != UNIX_HASH_SIZE) 127 128#ifdef CONFIG_SECURITY_NETWORK 129static void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb) 130{ 131 memcpy(UNIXSID(skb), &scm->secid, sizeof(u32)); 132} 133 134static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb) 135{ 136 scm->secid = *UNIXSID(skb); 137} 138#else 139static inline void unix_get_secdata(struct scm_cookie *scm, struct sk_buff *skb) 140{ } 141 142static inline void unix_set_secdata(struct scm_cookie *scm, struct sk_buff *skb) 143{ } 144#endif /* CONFIG_SECURITY_NETWORK */ 145 146/* 147 * SMP locking strategy: 148 * hash table is protected with spinlock unix_table_lock 149 * each socket state is protected by separate spin lock. 150 */ 151 152static inline unsigned int unix_hash_fold(__wsum n) 153{ 154 unsigned int hash = (__force unsigned int)n; 155 156 hash ^= hash>>16; 157 hash ^= hash>>8; 158 return hash&(UNIX_HASH_SIZE-1); 159} 160 161#define unix_peer(sk) (unix_sk(sk)->peer) 162 163static inline int unix_our_peer(struct sock *sk, struct sock *osk) 164{ 165 return unix_peer(osk) == sk; 166} 167 168static inline int unix_may_send(struct sock *sk, struct sock *osk) 169{ 170 return unix_peer(osk) == NULL || unix_our_peer(sk, osk); 171} 172 173static inline int unix_recvq_full(struct sock const *sk) 174{ 175 return skb_queue_len(&sk->sk_receive_queue) > sk->sk_max_ack_backlog; 176} 177 178struct sock *unix_peer_get(struct sock *s) 179{ 180 struct sock *peer; 181 182 unix_state_lock(s); 183 peer = unix_peer(s); 184 if (peer) 185 sock_hold(peer); 186 unix_state_unlock(s); 187 return peer; 188} 189EXPORT_SYMBOL_GPL(unix_peer_get); 190 191static 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 204static int unix_mkname(struct sockaddr_un *sunaddr, int len, unsigned int *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] doesn't 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(sunaddr, len, 0)); 224 return len; 225} 226 227static void __unix_remove_socket(struct sock *sk) 228{ 229 sk_del_node_init(sk); 230} 231 232static void __unix_insert_socket(struct hlist_head *list, struct sock *sk) 233{ 234 WARN_ON(!sk_unhashed(sk)); 235 sk_add_node(sk, list); 236} 237 238static 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 245static 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 252static struct sock *__unix_find_socket_byname(struct net *net, 253 struct sockaddr_un *sunname, 254 int len, int type, unsigned int hash) 255{ 256 struct sock *s; 257 struct hlist_node *node; 258 259 sk_for_each(s, node, &unix_socket_table[hash ^ type]) { 260 struct unix_sock *u = unix_sk(s); 261 262 if (!net_eq(sock_net(s), net)) 263 continue; 264 265 if (u->addr->len == len && 266 !memcmp(u->addr->name, sunname, len)) 267 goto found; 268 } 269 s = NULL; 270found: 271 return s; 272} 273 274static inline struct sock *unix_find_socket_byname(struct net *net, 275 struct sockaddr_un *sunname, 276 int len, int type, 277 unsigned int hash) 278{ 279 struct sock *s; 280 281 spin_lock(&unix_table_lock); 282 s = __unix_find_socket_byname(net, sunname, len, type, hash); 283 if (s) 284 sock_hold(s); 285 spin_unlock(&unix_table_lock); 286 return s; 287} 288 289static struct sock *unix_find_socket_byinode(struct inode *i) 290{ 291 struct sock *s; 292 struct hlist_node *node; 293 294 spin_lock(&unix_table_lock); 295 sk_for_each(s, node, 296 &unix_socket_table[i->i_ino & (UNIX_HASH_SIZE - 1)]) { 297 struct dentry *dentry = unix_sk(s)->path.dentry; 298 299 if (dentry && dentry->d_inode == i) { 300 sock_hold(s); 301 goto found; 302 } 303 } 304 s = NULL; 305found: 306 spin_unlock(&unix_table_lock); 307 return s; 308} 309 310static inline int unix_writable(struct sock *sk) 311{ 312 return (atomic_read(&sk->sk_wmem_alloc) << 2) <= sk->sk_sndbuf; 313} 314 315static void unix_write_space(struct sock *sk) 316{ 317 struct socket_wq *wq; 318 319 rcu_read_lock(); 320 if (unix_writable(sk)) { 321 wq = rcu_dereference(sk->sk_wq); 322 if (wq_has_sleeper(wq)) 323 wake_up_interruptible_sync_poll(&wq->wait, 324 POLLOUT | POLLWRNORM | POLLWRBAND); 325 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT); 326 } 327 rcu_read_unlock(); 328} 329 330/* When dgram socket disconnects (or changes its peer), we clear its receive 331 * queue of packets arrived from previous peer. First, it allows to do 332 * flow control based only on wmem_alloc; second, sk connected to peer 333 * may receive messages only from that peer. */ 334static void unix_dgram_disconnected(struct sock *sk, struct sock *other) 335{ 336 if (!skb_queue_empty(&sk->sk_receive_queue)) { 337 skb_queue_purge(&sk->sk_receive_queue); 338 wake_up_interruptible_all(&unix_sk(sk)->peer_wait); 339 340 /* If one link of bidirectional dgram pipe is disconnected, 341 * we signal error. Messages are lost. Do not make this, 342 * when peer was not connected to us. 343 */ 344 if (!sock_flag(other, SOCK_DEAD) && unix_peer(other) == sk) { 345 other->sk_err = ECONNRESET; 346 other->sk_error_report(other); 347 } 348 } 349} 350 351static void unix_sock_destructor(struct sock *sk) 352{ 353 struct unix_sock *u = unix_sk(sk); 354 355 skb_queue_purge(&sk->sk_receive_queue); 356 357 WARN_ON(atomic_read(&sk->sk_wmem_alloc)); 358 WARN_ON(!sk_unhashed(sk)); 359 WARN_ON(sk->sk_socket); 360 if (!sock_flag(sk, SOCK_DEAD)) { 361 printk(KERN_INFO "Attempt to release alive unix socket: %p\n", sk); 362 return; 363 } 364 365 if (u->addr) 366 unix_release_addr(u->addr); 367 368 atomic_long_dec(&unix_nr_socks); 369 local_bh_disable(); 370 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1); 371 local_bh_enable(); 372#ifdef UNIX_REFCNT_DEBUG 373 printk(KERN_DEBUG "UNIX %p is destroyed, %ld are still alive.\n", sk, 374 atomic_long_read(&unix_nr_socks)); 375#endif 376} 377 378static int unix_release_sock(struct sock *sk, int embrion) 379{ 380 struct unix_sock *u = unix_sk(sk); 381 struct path path; 382 struct sock *skpair; 383 struct sk_buff *skb; 384 int state; 385 386 unix_remove_socket(sk); 387 388 /* Clear state */ 389 unix_state_lock(sk); 390 sock_orphan(sk); 391 sk->sk_shutdown = SHUTDOWN_MASK; 392 path = u->path; 393 u->path.dentry = NULL; 394 u->path.mnt = NULL; 395 state = sk->sk_state; 396 sk->sk_state = TCP_CLOSE; 397 unix_state_unlock(sk); 398 399 wake_up_interruptible_all(&u->peer_wait); 400 401 skpair = unix_peer(sk); 402 403 if (skpair != NULL) { 404 if (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) { 405 unix_state_lock(skpair); 406 /* No more writes */ 407 skpair->sk_shutdown = SHUTDOWN_MASK; 408 if (!skb_queue_empty(&sk->sk_receive_queue) || embrion) 409 skpair->sk_err = ECONNRESET; 410 unix_state_unlock(skpair); 411 skpair->sk_state_change(skpair); 412 sk_wake_async(skpair, SOCK_WAKE_WAITD, POLL_HUP); 413 } 414 sock_put(skpair); /* It may now die */ 415 unix_peer(sk) = NULL; 416 } 417 418 /* Try to flush out this socket. Throw out buffers at least */ 419 420 while ((skb = skb_dequeue(&sk->sk_receive_queue)) != NULL) { 421 if (state == TCP_LISTEN) 422 unix_release_sock(skb->sk, 1); 423 /* passed fds are erased in the kfree_skb hook */ 424 kfree_skb(skb); 425 } 426 427 if (path.dentry) 428 path_put(&path); 429 430 sock_put(sk); 431 432 /* ---- Socket is dead now and most probably destroyed ---- */ 433 434 /* 435 * Fixme: BSD difference: In BSD all sockets connected to use get 436 * ECONNRESET and we die on the spot. In Linux we behave 437 * like files and pipes do and wait for the last 438 * dereference. 439 * 440 * Can't we simply set sock->err? 441 * 442 * What the above comment does talk about? --ANK(980817) 443 */ 444 445 if (unix_tot_inflight) 446 unix_gc(); /* Garbage collect fds */ 447 448 return 0; 449} 450 451static void init_peercred(struct sock *sk) 452{ 453 put_pid(sk->sk_peer_pid); 454 if (sk->sk_peer_cred) 455 put_cred(sk->sk_peer_cred); 456 sk->sk_peer_pid = get_pid(task_tgid(current)); 457 sk->sk_peer_cred = get_current_cred(); 458} 459 460static void copy_peercred(struct sock *sk, struct sock *peersk) 461{ 462 put_pid(sk->sk_peer_pid); 463 if (sk->sk_peer_cred) 464 put_cred(sk->sk_peer_cred); 465 sk->sk_peer_pid = get_pid(peersk->sk_peer_pid); 466 sk->sk_peer_cred = get_cred(peersk->sk_peer_cred); 467} 468 469static int unix_listen(struct socket *sock, int backlog) 470{ 471 int err; 472 struct sock *sk = sock->sk; 473 struct unix_sock *u = unix_sk(sk); 474 struct pid *old_pid = NULL; 475 const struct cred *old_cred = NULL; 476 477 err = -EOPNOTSUPP; 478 if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET) 479 goto out; /* Only stream/seqpacket sockets accept */ 480 err = -EINVAL; 481 if (!u->addr) 482 goto out; /* No listens on an unbound socket */ 483 unix_state_lock(sk); 484 if (sk->sk_state != TCP_CLOSE && sk->sk_state != TCP_LISTEN) 485 goto out_unlock; 486 if (backlog > sk->sk_max_ack_backlog) 487 wake_up_interruptible_all(&u->peer_wait); 488 sk->sk_max_ack_backlog = backlog; 489 sk->sk_state = TCP_LISTEN; 490 /* set credentials so connect can copy them */ 491 init_peercred(sk); 492 err = 0; 493 494out_unlock: 495 unix_state_unlock(sk); 496 put_pid(old_pid); 497 if (old_cred) 498 put_cred(old_cred); 499out: 500 return err; 501} 502 503static int unix_release(struct socket *); 504static int unix_bind(struct socket *, struct sockaddr *, int); 505static int unix_stream_connect(struct socket *, struct sockaddr *, 506 int addr_len, int flags); 507static int unix_socketpair(struct socket *, struct socket *); 508static int unix_accept(struct socket *, struct socket *, int); 509static int unix_getname(struct socket *, struct sockaddr *, int *, int); 510static unsigned int unix_poll(struct file *, struct socket *, poll_table *); 511static unsigned int unix_dgram_poll(struct file *, struct socket *, 512 poll_table *); 513static int unix_ioctl(struct socket *, unsigned int, unsigned long); 514static int unix_shutdown(struct socket *, int); 515static int unix_stream_sendmsg(struct kiocb *, struct socket *, 516 struct msghdr *, size_t); 517static int unix_stream_recvmsg(struct kiocb *, struct socket *, 518 struct msghdr *, size_t, int); 519static int unix_dgram_sendmsg(struct kiocb *, struct socket *, 520 struct msghdr *, size_t); 521static int unix_dgram_recvmsg(struct kiocb *, struct socket *, 522 struct msghdr *, size_t, int); 523static int unix_dgram_connect(struct socket *, struct sockaddr *, 524 int, int); 525static int unix_seqpacket_sendmsg(struct kiocb *, struct socket *, 526 struct msghdr *, size_t); 527static int unix_seqpacket_recvmsg(struct kiocb *, struct socket *, 528 struct msghdr *, size_t, int); 529 530static void unix_set_peek_off(struct sock *sk, int val) 531{ 532 struct unix_sock *u = unix_sk(sk); 533 534 mutex_lock(&u->readlock); 535 sk->sk_peek_off = val; 536 mutex_unlock(&u->readlock); 537} 538 539 540static const struct proto_ops unix_stream_ops = { 541 .family = PF_UNIX, 542 .owner = THIS_MODULE, 543 .release = unix_release, 544 .bind = unix_bind, 545 .connect = unix_stream_connect, 546 .socketpair = unix_socketpair, 547 .accept = unix_accept, 548 .getname = unix_getname, 549 .poll = unix_poll, 550 .ioctl = unix_ioctl, 551 .listen = unix_listen, 552 .shutdown = unix_shutdown, 553 .setsockopt = sock_no_setsockopt, 554 .getsockopt = sock_no_getsockopt, 555 .sendmsg = unix_stream_sendmsg, 556 .recvmsg = unix_stream_recvmsg, 557 .mmap = sock_no_mmap, 558 .sendpage = sock_no_sendpage, 559 .set_peek_off = unix_set_peek_off, 560}; 561 562static const struct proto_ops unix_dgram_ops = { 563 .family = PF_UNIX, 564 .owner = THIS_MODULE, 565 .release = unix_release, 566 .bind = unix_bind, 567 .connect = unix_dgram_connect, 568 .socketpair = unix_socketpair, 569 .accept = sock_no_accept, 570 .getname = unix_getname, 571 .poll = unix_dgram_poll, 572 .ioctl = unix_ioctl, 573 .listen = sock_no_listen, 574 .shutdown = unix_shutdown, 575 .setsockopt = sock_no_setsockopt, 576 .getsockopt = sock_no_getsockopt, 577 .sendmsg = unix_dgram_sendmsg, 578 .recvmsg = unix_dgram_recvmsg, 579 .mmap = sock_no_mmap, 580 .sendpage = sock_no_sendpage, 581 .set_peek_off = unix_set_peek_off, 582}; 583 584static const struct proto_ops unix_seqpacket_ops = { 585 .family = PF_UNIX, 586 .owner = THIS_MODULE, 587 .release = unix_release, 588 .bind = unix_bind, 589 .connect = unix_stream_connect, 590 .socketpair = unix_socketpair, 591 .accept = unix_accept, 592 .getname = unix_getname, 593 .poll = unix_dgram_poll, 594 .ioctl = unix_ioctl, 595 .listen = unix_listen, 596 .shutdown = unix_shutdown, 597 .setsockopt = sock_no_setsockopt, 598 .getsockopt = sock_no_getsockopt, 599 .sendmsg = unix_seqpacket_sendmsg, 600 .recvmsg = unix_seqpacket_recvmsg, 601 .mmap = sock_no_mmap, 602 .sendpage = sock_no_sendpage, 603 .set_peek_off = unix_set_peek_off, 604}; 605 606static struct proto unix_proto = { 607 .name = "UNIX", 608 .owner = THIS_MODULE, 609 .obj_size = sizeof(struct unix_sock), 610}; 611 612/* 613 * AF_UNIX sockets do not interact with hardware, hence they 614 * dont trigger interrupts - so it's safe for them to have 615 * bh-unsafe locking for their sk_receive_queue.lock. Split off 616 * this special lock-class by reinitializing the spinlock key: 617 */ 618static struct lock_class_key af_unix_sk_receive_queue_lock_key; 619 620static struct sock *unix_create1(struct net *net, struct socket *sock) 621{ 622 struct sock *sk = NULL; 623 struct unix_sock *u; 624 625 atomic_long_inc(&unix_nr_socks); 626 if (atomic_long_read(&unix_nr_socks) > 2 * get_max_files()) 627 goto out; 628 629 sk = sk_alloc(net, PF_UNIX, GFP_KERNEL, &unix_proto); 630 if (!sk) 631 goto out; 632 633 sock_init_data(sock, sk); 634 lockdep_set_class(&sk->sk_receive_queue.lock, 635 &af_unix_sk_receive_queue_lock_key); 636 637 sk->sk_write_space = unix_write_space; 638 sk->sk_max_ack_backlog = net->unx.sysctl_max_dgram_qlen; 639 sk->sk_destruct = unix_sock_destructor; 640 u = unix_sk(sk); 641 u->path.dentry = NULL; 642 u->path.mnt = NULL; 643 spin_lock_init(&u->lock); 644 atomic_long_set(&u->inflight, 0); 645 INIT_LIST_HEAD(&u->link); 646 mutex_init(&u->readlock); /* single task reading lock */ 647 init_waitqueue_head(&u->peer_wait); 648 unix_insert_socket(unix_sockets_unbound, sk); 649out: 650 if (sk == NULL) 651 atomic_long_dec(&unix_nr_socks); 652 else { 653 local_bh_disable(); 654 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); 655 local_bh_enable(); 656 } 657 return sk; 658} 659 660static int unix_create(struct net *net, struct socket *sock, int protocol, 661 int kern) 662{ 663 if (protocol && protocol != PF_UNIX) 664 return -EPROTONOSUPPORT; 665 666 sock->state = SS_UNCONNECTED; 667 668 switch (sock->type) { 669 case SOCK_STREAM: 670 sock->ops = &unix_stream_ops; 671 break; 672 /* 673 * Believe it or not BSD has AF_UNIX, SOCK_RAW though 674 * nothing uses it. 675 */ 676 case SOCK_RAW: 677 sock->type = SOCK_DGRAM; 678 case SOCK_DGRAM: 679 sock->ops = &unix_dgram_ops; 680 break; 681 case SOCK_SEQPACKET: 682 sock->ops = &unix_seqpacket_ops; 683 break; 684 default: 685 return -ESOCKTNOSUPPORT; 686 } 687 688 return unix_create1(net, sock) ? 0 : -ENOMEM; 689} 690 691static int unix_release(struct socket *sock) 692{ 693 struct sock *sk = sock->sk; 694 695 if (!sk) 696 return 0; 697 698 sock->sk = NULL; 699 700 return unix_release_sock(sk, 0); 701} 702 703static int unix_autobind(struct socket *sock) 704{ 705 struct sock *sk = sock->sk; 706 struct net *net = sock_net(sk); 707 struct unix_sock *u = unix_sk(sk); 708 static u32 ordernum = 1; 709 struct unix_address *addr; 710 int err; 711 unsigned int retries = 0; 712 713 mutex_lock(&u->readlock); 714 715 err = 0; 716 if (u->addr) 717 goto out; 718 719 err = -ENOMEM; 720 addr = kzalloc(sizeof(*addr) + sizeof(short) + 16, GFP_KERNEL); 721 if (!addr) 722 goto out; 723 724 addr->name->sun_family = AF_UNIX; 725 atomic_set(&addr->refcnt, 1); 726 727retry: 728 addr->len = sprintf(addr->name->sun_path+1, "%05x", ordernum) + 1 + sizeof(short); 729 addr->hash = unix_hash_fold(csum_partial(addr->name, addr->len, 0)); 730 731 spin_lock(&unix_table_lock); 732 ordernum = (ordernum+1)&0xFFFFF; 733 734 if (__unix_find_socket_byname(net, addr->name, addr->len, sock->type, 735 addr->hash)) { 736 spin_unlock(&unix_table_lock); 737 /* 738 * __unix_find_socket_byname() may take long time if many names 739 * are already in use. 740 */ 741 cond_resched(); 742 /* Give up if all names seems to be in use. */ 743 if (retries++ == 0xFFFFF) { 744 err = -ENOSPC; 745 kfree(addr); 746 goto out; 747 } 748 goto retry; 749 } 750 addr->hash ^= sk->sk_type; 751 752 __unix_remove_socket(sk); 753 u->addr = addr; 754 __unix_insert_socket(&unix_socket_table[addr->hash], sk); 755 spin_unlock(&unix_table_lock); 756 err = 0; 757 758out: mutex_unlock(&u->readlock); 759 return err; 760} 761 762static struct sock *unix_find_other(struct net *net, 763 struct sockaddr_un *sunname, int len, 764 int type, unsigned int hash, int *error) 765{ 766 struct sock *u; 767 struct path path; 768 int err = 0; 769 770 if (sunname->sun_path[0]) { 771 struct inode *inode; 772 err = kern_path(sunname->sun_path, LOOKUP_FOLLOW, &path); 773 if (err) 774 goto fail; 775 inode = path.dentry->d_inode; 776 err = inode_permission(inode, MAY_WRITE); 777 if (err) 778 goto put_fail; 779 780 err = -ECONNREFUSED; 781 if (!S_ISSOCK(inode->i_mode)) 782 goto put_fail; 783 u = unix_find_socket_byinode(inode); 784 if (!u) 785 goto put_fail; 786 787 if (u->sk_type == type) 788 touch_atime(&path); 789 790 path_put(&path); 791 792 err = -EPROTOTYPE; 793 if (u->sk_type != type) { 794 sock_put(u); 795 goto fail; 796 } 797 } else { 798 err = -ECONNREFUSED; 799 u = unix_find_socket_byname(net, sunname, len, type, hash); 800 if (u) { 801 struct dentry *dentry; 802 dentry = unix_sk(u)->path.dentry; 803 if (dentry) 804 touch_atime(&unix_sk(u)->path); 805 } else 806 goto fail; 807 } 808 return u; 809 810put_fail: 811 path_put(&path); 812fail: 813 *error = err; 814 return NULL; 815} 816 817 818static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len) 819{ 820 struct sock *sk = sock->sk; 821 struct net *net = sock_net(sk); 822 struct unix_sock *u = unix_sk(sk); 823 struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr; 824 char *sun_path = sunaddr->sun_path; 825 struct dentry *dentry = NULL; 826 struct path path; 827 int err; 828 unsigned int hash; 829 struct unix_address *addr; 830 struct hlist_head *list; 831 832 err = -EINVAL; 833 if (sunaddr->sun_family != AF_UNIX) 834 goto out; 835 836 if (addr_len == sizeof(short)) { 837 err = unix_autobind(sock); 838 goto out; 839 } 840 841 err = unix_mkname(sunaddr, addr_len, &hash); 842 if (err < 0) 843 goto out; 844 addr_len = err; 845 846 mutex_lock(&u->readlock); 847 848 err = -EINVAL; 849 if (u->addr) 850 goto out_up; 851 852 err = -ENOMEM; 853 addr = kmalloc(sizeof(*addr)+addr_len, GFP_KERNEL); 854 if (!addr) 855 goto out_up; 856 857 memcpy(addr->name, sunaddr, addr_len); 858 addr->len = addr_len; 859 addr->hash = hash ^ sk->sk_type; 860 atomic_set(&addr->refcnt, 1); 861 862 if (sun_path[0]) { 863 umode_t mode; 864 err = 0; 865 /* 866 * Get the parent directory, calculate the hash for last 867 * component. 868 */ 869 dentry = kern_path_create(AT_FDCWD, sun_path, &path, 0); 870 err = PTR_ERR(dentry); 871 if (IS_ERR(dentry)) 872 goto out_mknod_parent; 873 874 /* 875 * All right, let's create it. 876 */ 877 mode = S_IFSOCK | 878 (SOCK_INODE(sock)->i_mode & ~current_umask()); 879 err = mnt_want_write(path.mnt); 880 if (err) 881 goto out_mknod_dput; 882 err = security_path_mknod(&path, dentry, mode, 0); 883 if (err) 884 goto out_mknod_drop_write; 885 err = vfs_mknod(path.dentry->d_inode, dentry, mode, 0); 886out_mknod_drop_write: 887 mnt_drop_write(path.mnt); 888 if (err) 889 goto out_mknod_dput; 890 mntget(path.mnt); 891 dget(dentry); 892 done_path_create(&path, dentry); 893 path.dentry = dentry; 894 895 addr->hash = UNIX_HASH_SIZE; 896 } 897 898 spin_lock(&unix_table_lock); 899 900 if (!sun_path[0]) { 901 err = -EADDRINUSE; 902 if (__unix_find_socket_byname(net, sunaddr, addr_len, 903 sk->sk_type, hash)) { 904 unix_release_addr(addr); 905 goto out_unlock; 906 } 907 908 list = &unix_socket_table[addr->hash]; 909 } else { 910 list = &unix_socket_table[dentry->d_inode->i_ino & (UNIX_HASH_SIZE-1)]; 911 u->path = path; 912 } 913 914 err = 0; 915 __unix_remove_socket(sk); 916 u->addr = addr; 917 __unix_insert_socket(list, sk); 918 919out_unlock: 920 spin_unlock(&unix_table_lock); 921out_up: 922 mutex_unlock(&u->readlock); 923out: 924 return err; 925 926out_mknod_dput: 927 done_path_create(&path, dentry); 928out_mknod_parent: 929 if (err == -EEXIST) 930 err = -EADDRINUSE; 931 unix_release_addr(addr); 932 goto out_up; 933} 934 935static void unix_state_double_lock(struct sock *sk1, struct sock *sk2) 936{ 937 if (unlikely(sk1 == sk2) || !sk2) { 938 unix_state_lock(sk1); 939 return; 940 } 941 if (sk1 < sk2) { 942 unix_state_lock(sk1); 943 unix_state_lock_nested(sk2); 944 } else { 945 unix_state_lock(sk2); 946 unix_state_lock_nested(sk1); 947 } 948} 949 950static void unix_state_double_unlock(struct sock *sk1, struct sock *sk2) 951{ 952 if (unlikely(sk1 == sk2) || !sk2) { 953 unix_state_unlock(sk1); 954 return; 955 } 956 unix_state_unlock(sk1); 957 unix_state_unlock(sk2); 958} 959 960static int unix_dgram_connect(struct socket *sock, struct sockaddr *addr, 961 int alen, int flags) 962{ 963 struct sock *sk = sock->sk; 964 struct net *net = sock_net(sk); 965 struct sockaddr_un *sunaddr = (struct sockaddr_un *)addr; 966 struct sock *other; 967 unsigned int hash; 968 int err; 969 970 if (addr->sa_family != AF_UNSPEC) { 971 err = unix_mkname(sunaddr, alen, &hash); 972 if (err < 0) 973 goto out; 974 alen = err; 975 976 if (test_bit(SOCK_PASSCRED, &sock->flags) && 977 !unix_sk(sk)->addr && (err = unix_autobind(sock)) != 0) 978 goto out; 979 980restart: 981 other = unix_find_other(net, sunaddr, alen, sock->type, hash, &err); 982 if (!other) 983 goto out; 984 985 unix_state_double_lock(sk, other); 986 987 /* Apparently VFS overslept socket death. Retry. */ 988 if (sock_flag(other, SOCK_DEAD)) { 989 unix_state_double_unlock(sk, other); 990 sock_put(other); 991 goto restart; 992 } 993 994 err = -EPERM; 995 if (!unix_may_send(sk, other)) 996 goto out_unlock; 997 998 err = security_unix_may_send(sk->sk_socket, other->sk_socket); 999 if (err) 1000 goto out_unlock; 1001 1002 } else { 1003 /* 1004 * 1003.1g breaking connected state with AF_UNSPEC 1005 */ 1006 other = NULL; 1007 unix_state_double_lock(sk, other); 1008 } 1009 1010 /* 1011 * If it was connected, reconnect. 1012 */ 1013 if (unix_peer(sk)) { 1014 struct sock *old_peer = unix_peer(sk); 1015 unix_peer(sk) = other; 1016 unix_state_double_unlock(sk, other); 1017 1018 if (other != old_peer) 1019 unix_dgram_disconnected(sk, old_peer); 1020 sock_put(old_peer); 1021 } else { 1022 unix_peer(sk) = other; 1023 unix_state_double_unlock(sk, other); 1024 } 1025 return 0; 1026 1027out_unlock: 1028 unix_state_double_unlock(sk, other); 1029 sock_put(other); 1030out: 1031 return err; 1032} 1033 1034static long unix_wait_for_peer(struct sock *other, long timeo) 1035{ 1036 struct unix_sock *u = unix_sk(other); 1037 int sched; 1038 DEFINE_WAIT(wait); 1039 1040 prepare_to_wait_exclusive(&u->peer_wait, &wait, TASK_INTERRUPTIBLE); 1041 1042 sched = !sock_flag(other, SOCK_DEAD) && 1043 !(other->sk_shutdown & RCV_SHUTDOWN) && 1044 unix_recvq_full(other); 1045 1046 unix_state_unlock(other); 1047 1048 if (sched) 1049 timeo = schedule_timeout(timeo); 1050 1051 finish_wait(&u->peer_wait, &wait); 1052 return timeo; 1053} 1054 1055static int unix_stream_connect(struct socket *sock, struct sockaddr *uaddr, 1056 int addr_len, int flags) 1057{ 1058 struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr; 1059 struct sock *sk = sock->sk; 1060 struct net *net = sock_net(sk); 1061 struct unix_sock *u = unix_sk(sk), *newu, *otheru; 1062 struct sock *newsk = NULL; 1063 struct sock *other = NULL; 1064 struct sk_buff *skb = NULL; 1065 unsigned int hash; 1066 int st; 1067 int err; 1068 long timeo; 1069 1070 err = unix_mkname(sunaddr, addr_len, &hash); 1071 if (err < 0) 1072 goto out; 1073 addr_len = err; 1074 1075 if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr && 1076 (err = unix_autobind(sock)) != 0) 1077 goto out; 1078 1079 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK); 1080 1081 /* First of all allocate resources. 1082 If we will make it after state is locked, 1083 we will have to recheck all again in any case. 1084 */ 1085 1086 err = -ENOMEM; 1087 1088 /* create new sock for complete connection */ 1089 newsk = unix_create1(sock_net(sk), NULL); 1090 if (newsk == NULL) 1091 goto out; 1092 1093 /* Allocate skb for sending to listening sock */ 1094 skb = sock_wmalloc(newsk, 1, 0, GFP_KERNEL); 1095 if (skb == NULL) 1096 goto out; 1097 1098restart: 1099 /* Find listening sock. */ 1100 other = unix_find_other(net, sunaddr, addr_len, sk->sk_type, hash, &err); 1101 if (!other) 1102 goto out; 1103 1104 /* Latch state of peer */ 1105 unix_state_lock(other); 1106 1107 /* Apparently VFS overslept socket death. Retry. */ 1108 if (sock_flag(other, SOCK_DEAD)) { 1109 unix_state_unlock(other); 1110 sock_put(other); 1111 goto restart; 1112 } 1113 1114 err = -ECONNREFUSED; 1115 if (other->sk_state != TCP_LISTEN) 1116 goto out_unlock; 1117 if (other->sk_shutdown & RCV_SHUTDOWN) 1118 goto out_unlock; 1119 1120 if (unix_recvq_full(other)) { 1121 err = -EAGAIN; 1122 if (!timeo) 1123 goto out_unlock; 1124 1125 timeo = unix_wait_for_peer(other, timeo); 1126 1127 err = sock_intr_errno(timeo); 1128 if (signal_pending(current)) 1129 goto out; 1130 sock_put(other); 1131 goto restart; 1132 } 1133 1134 /* Latch our state. 1135 1136 It is tricky place. We need to grab our state lock and cannot 1137 drop lock on peer. It is dangerous because deadlock is 1138 possible. Connect to self case and simultaneous 1139 attempt to connect are eliminated by checking socket 1140 state. other is TCP_LISTEN, if sk is TCP_LISTEN we 1141 check this before attempt to grab lock. 1142 1143 Well, and we have to recheck the state after socket locked. 1144 */ 1145 st = sk->sk_state; 1146 1147 switch (st) { 1148 case TCP_CLOSE: 1149 /* This is ok... continue with connect */ 1150 break; 1151 case TCP_ESTABLISHED: 1152 /* Socket is already connected */ 1153 err = -EISCONN; 1154 goto out_unlock; 1155 default: 1156 err = -EINVAL; 1157 goto out_unlock; 1158 } 1159 1160 unix_state_lock_nested(sk); 1161 1162 if (sk->sk_state != st) { 1163 unix_state_unlock(sk); 1164 unix_state_unlock(other); 1165 sock_put(other); 1166 goto restart; 1167 } 1168 1169 err = security_unix_stream_connect(sk, other, newsk); 1170 if (err) { 1171 unix_state_unlock(sk); 1172 goto out_unlock; 1173 } 1174 1175 /* The way is open! Fastly set all the necessary fields... */ 1176 1177 sock_hold(sk); 1178 unix_peer(newsk) = sk; 1179 newsk->sk_state = TCP_ESTABLISHED; 1180 newsk->sk_type = sk->sk_type; 1181 init_peercred(newsk); 1182 newu = unix_sk(newsk); 1183 RCU_INIT_POINTER(newsk->sk_wq, &newu->peer_wq); 1184 otheru = unix_sk(other); 1185 1186 /* copy address information from listening to new sock*/ 1187 if (otheru->addr) { 1188 atomic_inc(&otheru->addr->refcnt); 1189 newu->addr = otheru->addr; 1190 } 1191 if (otheru->path.dentry) { 1192 path_get(&otheru->path); 1193 newu->path = otheru->path; 1194 } 1195 1196 /* Set credentials */ 1197 copy_peercred(sk, other); 1198 1199 sock->state = SS_CONNECTED; 1200 sk->sk_state = TCP_ESTABLISHED; 1201 sock_hold(newsk); 1202 1203 smp_mb__after_atomic_inc(); /* sock_hold() does an atomic_inc() */ 1204 unix_peer(sk) = newsk; 1205 1206 unix_state_unlock(sk); 1207 1208 /* take ten and and send info to listening sock */ 1209 spin_lock(&other->sk_receive_queue.lock); 1210 __skb_queue_tail(&other->sk_receive_queue, skb); 1211 spin_unlock(&other->sk_receive_queue.lock); 1212 unix_state_unlock(other); 1213 other->sk_data_ready(other, 0); 1214 sock_put(other); 1215 return 0; 1216 1217out_unlock: 1218 if (other) 1219 unix_state_unlock(other); 1220 1221out: 1222 kfree_skb(skb); 1223 if (newsk) 1224 unix_release_sock(newsk, 0); 1225 if (other) 1226 sock_put(other); 1227 return err; 1228} 1229 1230static int unix_socketpair(struct socket *socka, struct socket *sockb) 1231{ 1232 struct sock *ska = socka->sk, *skb = sockb->sk; 1233 1234 /* Join our sockets back to back */ 1235 sock_hold(ska); 1236 sock_hold(skb); 1237 unix_peer(ska) = skb; 1238 unix_peer(skb) = ska; 1239 init_peercred(ska); 1240 init_peercred(skb); 1241 1242 if (ska->sk_type != SOCK_DGRAM) { 1243 ska->sk_state = TCP_ESTABLISHED; 1244 skb->sk_state = TCP_ESTABLISHED; 1245 socka->state = SS_CONNECTED; 1246 sockb->state = SS_CONNECTED; 1247 } 1248 return 0; 1249} 1250 1251static int unix_accept(struct socket *sock, struct socket *newsock, int flags) 1252{ 1253 struct sock *sk = sock->sk; 1254 struct sock *tsk; 1255 struct sk_buff *skb; 1256 int err; 1257 1258 err = -EOPNOTSUPP; 1259 if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET) 1260 goto out; 1261 1262 err = -EINVAL; 1263 if (sk->sk_state != TCP_LISTEN) 1264 goto out; 1265 1266 /* If socket state is TCP_LISTEN it cannot change (for now...), 1267 * so that no locks are necessary. 1268 */ 1269 1270 skb = skb_recv_datagram(sk, 0, flags&O_NONBLOCK, &err); 1271 if (!skb) { 1272 /* This means receive shutdown. */ 1273 if (err == 0) 1274 err = -EINVAL; 1275 goto out; 1276 } 1277 1278 tsk = skb->sk; 1279 skb_free_datagram(sk, skb); 1280 wake_up_interruptible(&unix_sk(sk)->peer_wait); 1281 1282 /* attach accepted sock to socket */ 1283 unix_state_lock(tsk); 1284 newsock->state = SS_CONNECTED; 1285 sock_graft(tsk, newsock); 1286 unix_state_unlock(tsk); 1287 return 0; 1288 1289out: 1290 return err; 1291} 1292 1293 1294static int unix_getname(struct socket *sock, struct sockaddr *uaddr, int *uaddr_len, int peer) 1295{ 1296 struct sock *sk = sock->sk; 1297 struct unix_sock *u; 1298 DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr, uaddr); 1299 int err = 0; 1300 1301 if (peer) { 1302 sk = unix_peer_get(sk); 1303 1304 err = -ENOTCONN; 1305 if (!sk) 1306 goto out; 1307 err = 0; 1308 } else { 1309 sock_hold(sk); 1310 } 1311 1312 u = unix_sk(sk); 1313 unix_state_lock(sk); 1314 if (!u->addr) { 1315 sunaddr->sun_family = AF_UNIX; 1316 sunaddr->sun_path[0] = 0; 1317 *uaddr_len = sizeof(short); 1318 } else { 1319 struct unix_address *addr = u->addr; 1320 1321 *uaddr_len = addr->len; 1322 memcpy(sunaddr, addr->name, *uaddr_len); 1323 } 1324 unix_state_unlock(sk); 1325 sock_put(sk); 1326out: 1327 return err; 1328} 1329 1330static void unix_detach_fds(struct scm_cookie *scm, struct sk_buff *skb) 1331{ 1332 int i; 1333 1334 scm->fp = UNIXCB(skb).fp; 1335 UNIXCB(skb).fp = NULL; 1336 1337 for (i = scm->fp->count-1; i >= 0; i--) 1338 unix_notinflight(scm->fp->fp[i]); 1339} 1340 1341static void unix_destruct_scm(struct sk_buff *skb) 1342{ 1343 struct scm_cookie scm; 1344 memset(&scm, 0, sizeof(scm)); 1345 scm.pid = UNIXCB(skb).pid; 1346 scm.cred = UNIXCB(skb).cred; 1347 if (UNIXCB(skb).fp) 1348 unix_detach_fds(&scm, skb); 1349 1350 /* Alas, it calls VFS */ 1351 /* So fscking what? fput() had been SMP-safe since the last Summer */ 1352 scm_destroy(&scm); 1353 sock_wfree(skb); 1354} 1355 1356#define MAX_RECURSION_LEVEL 4 1357 1358static int unix_attach_fds(struct scm_cookie *scm, struct sk_buff *skb) 1359{ 1360 int i; 1361 unsigned char max_level = 0; 1362 int unix_sock_count = 0; 1363 1364 for (i = scm->fp->count - 1; i >= 0; i--) { 1365 struct sock *sk = unix_get_socket(scm->fp->fp[i]); 1366 1367 if (sk) { 1368 unix_sock_count++; 1369 max_level = max(max_level, 1370 unix_sk(sk)->recursion_level); 1371 } 1372 } 1373 if (unlikely(max_level > MAX_RECURSION_LEVEL)) 1374 return -ETOOMANYREFS; 1375 1376 /* 1377 * Need to duplicate file references for the sake of garbage 1378 * collection. Otherwise a socket in the fps might become a 1379 * candidate for GC while the skb is not yet queued. 1380 */ 1381 UNIXCB(skb).fp = scm_fp_dup(scm->fp); 1382 if (!UNIXCB(skb).fp) 1383 return -ENOMEM; 1384 1385 if (unix_sock_count) { 1386 for (i = scm->fp->count - 1; i >= 0; i--) 1387 unix_inflight(scm->fp->fp[i]); 1388 } 1389 return max_level; 1390} 1391 1392static int unix_scm_to_skb(struct scm_cookie *scm, struct sk_buff *skb, bool send_fds) 1393{ 1394 int err = 0; 1395 1396 UNIXCB(skb).pid = get_pid(scm->pid); 1397 if (scm->cred) 1398 UNIXCB(skb).cred = get_cred(scm->cred); 1399 UNIXCB(skb).fp = NULL; 1400 if (scm->fp && send_fds) 1401 err = unix_attach_fds(scm, skb); 1402 1403 skb->destructor = unix_destruct_scm; 1404 return err; 1405} 1406 1407/* 1408 * Some apps rely on write() giving SCM_CREDENTIALS 1409 * We include credentials if source or destination socket 1410 * asserted SOCK_PASSCRED. 1411 */ 1412static void maybe_add_creds(struct sk_buff *skb, const struct socket *sock, 1413 const struct sock *other) 1414{ 1415 if (UNIXCB(skb).cred) 1416 return; 1417 if (test_bit(SOCK_PASSCRED, &sock->flags) || 1418 !other->sk_socket || 1419 test_bit(SOCK_PASSCRED, &other->sk_socket->flags)) { 1420 UNIXCB(skb).pid = get_pid(task_tgid(current)); 1421 UNIXCB(skb).cred = get_current_cred(); 1422 } 1423} 1424 1425/* 1426 * Send AF_UNIX data. 1427 */ 1428 1429static int unix_dgram_sendmsg(struct kiocb *kiocb, struct socket *sock, 1430 struct msghdr *msg, size_t len) 1431{ 1432 struct sock_iocb *siocb = kiocb_to_siocb(kiocb); 1433 struct sock *sk = sock->sk; 1434 struct net *net = sock_net(sk); 1435 struct unix_sock *u = unix_sk(sk); 1436 struct sockaddr_un *sunaddr = msg->msg_name; 1437 struct sock *other = NULL; 1438 int namelen = 0; /* fake GCC */ 1439 int err; 1440 unsigned int hash; 1441 struct sk_buff *skb; 1442 long timeo; 1443 struct scm_cookie tmp_scm; 1444 int max_level; 1445 int data_len = 0; 1446 1447 if (NULL == siocb->scm) 1448 siocb->scm = &tmp_scm; 1449 wait_for_unix_gc(); 1450 err = scm_send(sock, msg, siocb->scm); 1451 if (err < 0) 1452 return err; 1453 1454 err = -EOPNOTSUPP; 1455 if (msg->msg_flags&MSG_OOB) 1456 goto out; 1457 1458 if (msg->msg_namelen) { 1459 err = unix_mkname(sunaddr, msg->msg_namelen, &hash); 1460 if (err < 0) 1461 goto out; 1462 namelen = err; 1463 } else { 1464 sunaddr = NULL; 1465 err = -ENOTCONN; 1466 other = unix_peer_get(sk); 1467 if (!other) 1468 goto out; 1469 } 1470 1471 if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr 1472 && (err = unix_autobind(sock)) != 0) 1473 goto out; 1474 1475 err = -EMSGSIZE; 1476 if (len > sk->sk_sndbuf - 32) 1477 goto out; 1478 1479 if (len > SKB_MAX_ALLOC) 1480 data_len = min_t(size_t, 1481 len - SKB_MAX_ALLOC, 1482 MAX_SKB_FRAGS * PAGE_SIZE); 1483 1484 skb = sock_alloc_send_pskb(sk, len - data_len, data_len, 1485 msg->msg_flags & MSG_DONTWAIT, &err); 1486 if (skb == NULL) 1487 goto out; 1488 1489 err = unix_scm_to_skb(siocb->scm, skb, true); 1490 if (err < 0) 1491 goto out_free; 1492 max_level = err + 1; 1493 unix_get_secdata(siocb->scm, skb); 1494 1495 skb_put(skb, len - data_len); 1496 skb->data_len = data_len; 1497 skb->len = len; 1498 err = skb_copy_datagram_from_iovec(skb, 0, msg->msg_iov, 0, len); 1499 if (err) 1500 goto out_free; 1501 1502 timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT); 1503 1504restart: 1505 if (!other) { 1506 err = -ECONNRESET; 1507 if (sunaddr == NULL) 1508 goto out_free; 1509 1510 other = unix_find_other(net, sunaddr, namelen, sk->sk_type, 1511 hash, &err); 1512 if (other == NULL) 1513 goto out_free; 1514 } 1515 1516 if (sk_filter(other, skb) < 0) { 1517 /* Toss the packet but do not return any error to the sender */ 1518 err = len; 1519 goto out_free; 1520 } 1521 1522 unix_state_lock(other); 1523 err = -EPERM; 1524 if (!unix_may_send(sk, other)) 1525 goto out_unlock; 1526 1527 if (sock_flag(other, SOCK_DEAD)) { 1528 /* 1529 * Check with 1003.1g - what should 1530 * datagram error 1531 */ 1532 unix_state_unlock(other); 1533 sock_put(other); 1534 1535 err = 0; 1536 unix_state_lock(sk); 1537 if (unix_peer(sk) == other) { 1538 unix_peer(sk) = NULL; 1539 unix_state_unlock(sk); 1540 1541 unix_dgram_disconnected(sk, other); 1542 sock_put(other); 1543 err = -ECONNREFUSED; 1544 } else { 1545 unix_state_unlock(sk); 1546 } 1547 1548 other = NULL; 1549 if (err) 1550 goto out_free; 1551 goto restart; 1552 } 1553 1554 err = -EPIPE; 1555 if (other->sk_shutdown & RCV_SHUTDOWN) 1556 goto out_unlock; 1557 1558 if (sk->sk_type != SOCK_SEQPACKET) { 1559 err = security_unix_may_send(sk->sk_socket, other->sk_socket); 1560 if (err) 1561 goto out_unlock; 1562 } 1563 1564 if (unix_peer(other) != sk && unix_recvq_full(other)) { 1565 if (!timeo) { 1566 err = -EAGAIN; 1567 goto out_unlock; 1568 } 1569 1570 timeo = unix_wait_for_peer(other, timeo); 1571 1572 err = sock_intr_errno(timeo); 1573 if (signal_pending(current)) 1574 goto out_free; 1575 1576 goto restart; 1577 } 1578 1579 if (sock_flag(other, SOCK_RCVTSTAMP)) 1580 __net_timestamp(skb); 1581 maybe_add_creds(skb, sock, other); 1582 skb_queue_tail(&other->sk_receive_queue, skb); 1583 if (max_level > unix_sk(other)->recursion_level) 1584 unix_sk(other)->recursion_level = max_level; 1585 unix_state_unlock(other); 1586 other->sk_data_ready(other, len); 1587 sock_put(other); 1588 scm_destroy(siocb->scm); 1589 return len; 1590 1591out_unlock: 1592 unix_state_unlock(other); 1593out_free: 1594 kfree_skb(skb); 1595out: 1596 if (other) 1597 sock_put(other); 1598 scm_destroy(siocb->scm); 1599 return err; 1600} 1601 1602 1603static int unix_stream_sendmsg(struct kiocb *kiocb, struct socket *sock, 1604 struct msghdr *msg, size_t len) 1605{ 1606 struct sock_iocb *siocb = kiocb_to_siocb(kiocb); 1607 struct sock *sk = sock->sk; 1608 struct sock *other = NULL; 1609 int err, size; 1610 struct sk_buff *skb; 1611 int sent = 0; 1612 struct scm_cookie tmp_scm; 1613 bool fds_sent = false; 1614 int max_level; 1615 1616 if (NULL == siocb->scm) 1617 siocb->scm = &tmp_scm; 1618 wait_for_unix_gc(); 1619 err = scm_send(sock, msg, siocb->scm); 1620 if (err < 0) 1621 return err; 1622 1623 err = -EOPNOTSUPP; 1624 if (msg->msg_flags&MSG_OOB) 1625 goto out_err; 1626 1627 if (msg->msg_namelen) { 1628 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP; 1629 goto out_err; 1630 } else { 1631 err = -ENOTCONN; 1632 other = unix_peer(sk); 1633 if (!other) 1634 goto out_err; 1635 } 1636 1637 if (sk->sk_shutdown & SEND_SHUTDOWN) 1638 goto pipe_err; 1639 1640 while (sent < len) { 1641 /* 1642 * Optimisation for the fact that under 0.01% of X 1643 * messages typically need breaking up. 1644 */ 1645 1646 size = len-sent; 1647 1648 /* Keep two messages in the pipe so it schedules better */ 1649 if (size > ((sk->sk_sndbuf >> 1) - 64)) 1650 size = (sk->sk_sndbuf >> 1) - 64; 1651 1652 if (size > SKB_MAX_ALLOC) 1653 size = SKB_MAX_ALLOC; 1654 1655 /* 1656 * Grab a buffer 1657 */ 1658 1659 skb = sock_alloc_send_skb(sk, size, msg->msg_flags&MSG_DONTWAIT, 1660 &err); 1661 1662 if (skb == NULL) 1663 goto out_err; 1664 1665 /* 1666 * If you pass two values to the sock_alloc_send_skb 1667 * it tries to grab the large buffer with GFP_NOFS 1668 * (which can fail easily), and if it fails grab the 1669 * fallback size buffer which is under a page and will 1670 * succeed. [Alan] 1671 */ 1672 size = min_t(int, size, skb_tailroom(skb)); 1673 1674 1675 /* Only send the fds in the first buffer */ 1676 err = unix_scm_to_skb(siocb->scm, skb, !fds_sent); 1677 if (err < 0) { 1678 kfree_skb(skb); 1679 goto out_err; 1680 } 1681 max_level = err + 1; 1682 fds_sent = true; 1683 1684 err = memcpy_fromiovec(skb_put(skb, size), msg->msg_iov, size); 1685 if (err) { 1686 kfree_skb(skb); 1687 goto out_err; 1688 } 1689 1690 unix_state_lock(other); 1691 1692 if (sock_flag(other, SOCK_DEAD) || 1693 (other->sk_shutdown & RCV_SHUTDOWN)) 1694 goto pipe_err_free; 1695 1696 maybe_add_creds(skb, sock, other); 1697 skb_queue_tail(&other->sk_receive_queue, skb); 1698 if (max_level > unix_sk(other)->recursion_level) 1699 unix_sk(other)->recursion_level = max_level; 1700 unix_state_unlock(other); 1701 other->sk_data_ready(other, size); 1702 sent += size; 1703 } 1704 1705 scm_destroy(siocb->scm); 1706 siocb->scm = NULL; 1707 1708 return sent; 1709 1710pipe_err_free: 1711 unix_state_unlock(other); 1712 kfree_skb(skb); 1713pipe_err: 1714 if (sent == 0 && !(msg->msg_flags&MSG_NOSIGNAL)) 1715 send_sig(SIGPIPE, current, 0); 1716 err = -EPIPE; 1717out_err: 1718 scm_destroy(siocb->scm); 1719 siocb->scm = NULL; 1720 return sent ? : err; 1721} 1722 1723static int unix_seqpacket_sendmsg(struct kiocb *kiocb, struct socket *sock, 1724 struct msghdr *msg, size_t len) 1725{ 1726 int err; 1727 struct sock *sk = sock->sk; 1728 1729 err = sock_error(sk); 1730 if (err) 1731 return err; 1732 1733 if (sk->sk_state != TCP_ESTABLISHED) 1734 return -ENOTCONN; 1735 1736 if (msg->msg_namelen) 1737 msg->msg_namelen = 0; 1738 1739 return unix_dgram_sendmsg(kiocb, sock, msg, len); 1740} 1741 1742static int unix_seqpacket_recvmsg(struct kiocb *iocb, struct socket *sock, 1743 struct msghdr *msg, size_t size, 1744 int flags) 1745{ 1746 struct sock *sk = sock->sk; 1747 1748 if (sk->sk_state != TCP_ESTABLISHED) 1749 return -ENOTCONN; 1750 1751 return unix_dgram_recvmsg(iocb, sock, msg, size, flags); 1752} 1753 1754static void unix_copy_addr(struct msghdr *msg, struct sock *sk) 1755{ 1756 struct unix_sock *u = unix_sk(sk); 1757 1758 msg->msg_namelen = 0; 1759 if (u->addr) { 1760 msg->msg_namelen = u->addr->len; 1761 memcpy(msg->msg_name, u->addr->name, u->addr->len); 1762 } 1763} 1764 1765static int unix_dgram_recvmsg(struct kiocb *iocb, struct socket *sock, 1766 struct msghdr *msg, size_t size, 1767 int flags) 1768{ 1769 struct sock_iocb *siocb = kiocb_to_siocb(iocb); 1770 struct scm_cookie tmp_scm; 1771 struct sock *sk = sock->sk; 1772 struct unix_sock *u = unix_sk(sk); 1773 int noblock = flags & MSG_DONTWAIT; 1774 struct sk_buff *skb; 1775 int err; 1776 int peeked, skip; 1777 1778 err = -EOPNOTSUPP; 1779 if (flags&MSG_OOB) 1780 goto out; 1781 1782 msg->msg_namelen = 0; 1783 1784 err = mutex_lock_interruptible(&u->readlock); 1785 if (err) { 1786 err = sock_intr_errno(sock_rcvtimeo(sk, noblock)); 1787 goto out; 1788 } 1789 1790 skip = sk_peek_offset(sk, flags); 1791 1792 skb = __skb_recv_datagram(sk, flags, &peeked, &skip, &err); 1793 if (!skb) { 1794 unix_state_lock(sk); 1795 /* Signal EOF on disconnected non-blocking SEQPACKET socket. */ 1796 if (sk->sk_type == SOCK_SEQPACKET && err == -EAGAIN && 1797 (sk->sk_shutdown & RCV_SHUTDOWN)) 1798 err = 0; 1799 unix_state_unlock(sk); 1800 goto out_unlock; 1801 } 1802 1803 wake_up_interruptible_sync_poll(&u->peer_wait, 1804 POLLOUT | POLLWRNORM | POLLWRBAND); 1805 1806 if (msg->msg_name) 1807 unix_copy_addr(msg, skb->sk); 1808 1809 if (size > skb->len - skip) 1810 size = skb->len - skip; 1811 else if (size < skb->len - skip) 1812 msg->msg_flags |= MSG_TRUNC; 1813 1814 err = skb_copy_datagram_iovec(skb, skip, msg->msg_iov, size); 1815 if (err) 1816 goto out_free; 1817 1818 if (sock_flag(sk, SOCK_RCVTSTAMP)) 1819 __sock_recv_timestamp(msg, sk, skb); 1820 1821 if (!siocb->scm) { 1822 siocb->scm = &tmp_scm; 1823 memset(&tmp_scm, 0, sizeof(tmp_scm)); 1824 } 1825 scm_set_cred(siocb->scm, UNIXCB(skb).pid, UNIXCB(skb).cred); 1826 unix_set_secdata(siocb->scm, skb); 1827 1828 if (!(flags & MSG_PEEK)) { 1829 if (UNIXCB(skb).fp) 1830 unix_detach_fds(siocb->scm, skb); 1831 1832 sk_peek_offset_bwd(sk, skb->len); 1833 } else { 1834 /* It is questionable: on PEEK we could: 1835 - do not return fds - good, but too simple 8) 1836 - return fds, and do not return them on read (old strategy, 1837 apparently wrong) 1838 - clone fds (I chose it for now, it is the most universal 1839 solution) 1840 1841 POSIX 1003.1g does not actually define this clearly 1842 at all. POSIX 1003.1g doesn't define a lot of things 1843 clearly however! 1844 1845 */ 1846 1847 sk_peek_offset_fwd(sk, size); 1848 1849 if (UNIXCB(skb).fp) 1850 siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp); 1851 } 1852 err = (flags & MSG_TRUNC) ? skb->len - skip : size; 1853 1854 scm_recv(sock, msg, siocb->scm, flags); 1855 1856out_free: 1857 skb_free_datagram(sk, skb); 1858out_unlock: 1859 mutex_unlock(&u->readlock); 1860out: 1861 return err; 1862} 1863 1864/* 1865 * Sleep until data has arrive. But check for races.. 1866 */ 1867 1868static long unix_stream_data_wait(struct sock *sk, long timeo) 1869{ 1870 DEFINE_WAIT(wait); 1871 1872 unix_state_lock(sk); 1873 1874 for (;;) { 1875 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 1876 1877 if (!skb_queue_empty(&sk->sk_receive_queue) || 1878 sk->sk_err || 1879 (sk->sk_shutdown & RCV_SHUTDOWN) || 1880 signal_pending(current) || 1881 !timeo) 1882 break; 1883 1884 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags); 1885 unix_state_unlock(sk); 1886 timeo = schedule_timeout(timeo); 1887 unix_state_lock(sk); 1888 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags); 1889 } 1890 1891 finish_wait(sk_sleep(sk), &wait); 1892 unix_state_unlock(sk); 1893 return timeo; 1894} 1895 1896 1897 1898static int unix_stream_recvmsg(struct kiocb *iocb, struct socket *sock, 1899 struct msghdr *msg, size_t size, 1900 int flags) 1901{ 1902 struct sock_iocb *siocb = kiocb_to_siocb(iocb); 1903 struct scm_cookie tmp_scm; 1904 struct sock *sk = sock->sk; 1905 struct unix_sock *u = unix_sk(sk); 1906 struct sockaddr_un *sunaddr = msg->msg_name; 1907 int copied = 0; 1908 int check_creds = 0; 1909 int target; 1910 int err = 0; 1911 long timeo; 1912 int skip; 1913 1914 err = -EINVAL; 1915 if (sk->sk_state != TCP_ESTABLISHED) 1916 goto out; 1917 1918 err = -EOPNOTSUPP; 1919 if (flags&MSG_OOB) 1920 goto out; 1921 1922 target = sock_rcvlowat(sk, flags&MSG_WAITALL, size); 1923 timeo = sock_rcvtimeo(sk, flags&MSG_DONTWAIT); 1924 1925 msg->msg_namelen = 0; 1926 1927 /* Lock the socket to prevent queue disordering 1928 * while sleeps in memcpy_tomsg 1929 */ 1930 1931 if (!siocb->scm) { 1932 siocb->scm = &tmp_scm; 1933 memset(&tmp_scm, 0, sizeof(tmp_scm)); 1934 } 1935 1936 err = mutex_lock_interruptible(&u->readlock); 1937 if (err) { 1938 err = sock_intr_errno(timeo); 1939 goto out; 1940 } 1941 1942 skip = sk_peek_offset(sk, flags); 1943 1944 do { 1945 int chunk; 1946 struct sk_buff *skb; 1947 1948 unix_state_lock(sk); 1949 skb = skb_peek(&sk->sk_receive_queue); 1950again: 1951 if (skb == NULL) { 1952 unix_sk(sk)->recursion_level = 0; 1953 if (copied >= target) 1954 goto unlock; 1955 1956 /* 1957 * POSIX 1003.1g mandates this order. 1958 */ 1959 1960 err = sock_error(sk); 1961 if (err) 1962 goto unlock; 1963 if (sk->sk_shutdown & RCV_SHUTDOWN) 1964 goto unlock; 1965 1966 unix_state_unlock(sk); 1967 err = -EAGAIN; 1968 if (!timeo) 1969 break; 1970 mutex_unlock(&u->readlock); 1971 1972 timeo = unix_stream_data_wait(sk, timeo); 1973 1974 if (signal_pending(current) 1975 || mutex_lock_interruptible(&u->readlock)) { 1976 err = sock_intr_errno(timeo); 1977 goto out; 1978 } 1979 1980 continue; 1981 unlock: 1982 unix_state_unlock(sk); 1983 break; 1984 } 1985 1986 if (skip >= skb->len) { 1987 skip -= skb->len; 1988 skb = skb_peek_next(skb, &sk->sk_receive_queue); 1989 goto again; 1990 } 1991 1992 unix_state_unlock(sk); 1993 1994 if (check_creds) { 1995 /* Never glue messages from different writers */ 1996 if ((UNIXCB(skb).pid != siocb->scm->pid) || 1997 (UNIXCB(skb).cred != siocb->scm->cred)) 1998 break; 1999 } else { 2000 /* Copy credentials */ 2001 scm_set_cred(siocb->scm, UNIXCB(skb).pid, UNIXCB(skb).cred); 2002 check_creds = 1; 2003 } 2004 2005 /* Copy address just once */ 2006 if (sunaddr) { 2007 unix_copy_addr(msg, skb->sk); 2008 sunaddr = NULL; 2009 } 2010 2011 chunk = min_t(unsigned int, skb->len - skip, size); 2012 if (memcpy_toiovec(msg->msg_iov, skb->data + skip, chunk)) { 2013 if (copied == 0) 2014 copied = -EFAULT; 2015 break; 2016 } 2017 copied += chunk; 2018 size -= chunk; 2019 2020 /* Mark read part of skb as used */ 2021 if (!(flags & MSG_PEEK)) { 2022 skb_pull(skb, chunk); 2023 2024 sk_peek_offset_bwd(sk, chunk); 2025 2026 if (UNIXCB(skb).fp) 2027 unix_detach_fds(siocb->scm, skb); 2028 2029 if (skb->len) 2030 break; 2031 2032 skb_unlink(skb, &sk->sk_receive_queue); 2033 consume_skb(skb); 2034 2035 if (siocb->scm->fp) 2036 break; 2037 } else { 2038 /* It is questionable, see note in unix_dgram_recvmsg. 2039 */ 2040 if (UNIXCB(skb).fp) 2041 siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp); 2042 2043 sk_peek_offset_fwd(sk, chunk); 2044 2045 break; 2046 } 2047 } while (size); 2048 2049 mutex_unlock(&u->readlock); 2050 scm_recv(sock, msg, siocb->scm, flags); 2051out: 2052 return copied ? : err; 2053} 2054 2055static int unix_shutdown(struct socket *sock, int mode) 2056{ 2057 struct sock *sk = sock->sk; 2058 struct sock *other; 2059 2060 mode = (mode+1)&(RCV_SHUTDOWN|SEND_SHUTDOWN); 2061 2062 if (!mode) 2063 return 0; 2064 2065 unix_state_lock(sk); 2066 sk->sk_shutdown |= mode; 2067 other = unix_peer(sk); 2068 if (other) 2069 sock_hold(other); 2070 unix_state_unlock(sk); 2071 sk->sk_state_change(sk); 2072 2073 if (other && 2074 (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET)) { 2075 2076 int peer_mode = 0; 2077 2078 if (mode&RCV_SHUTDOWN) 2079 peer_mode |= SEND_SHUTDOWN; 2080 if (mode&SEND_SHUTDOWN) 2081 peer_mode |= RCV_SHUTDOWN; 2082 unix_state_lock(other); 2083 other->sk_shutdown |= peer_mode; 2084 unix_state_unlock(other); 2085 other->sk_state_change(other); 2086 if (peer_mode == SHUTDOWN_MASK) 2087 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_HUP); 2088 else if (peer_mode & RCV_SHUTDOWN) 2089 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_IN); 2090 } 2091 if (other) 2092 sock_put(other); 2093 2094 return 0; 2095} 2096 2097long unix_inq_len(struct sock *sk) 2098{ 2099 struct sk_buff *skb; 2100 long amount = 0; 2101 2102 if (sk->sk_state == TCP_LISTEN) 2103 return -EINVAL; 2104 2105 spin_lock(&sk->sk_receive_queue.lock); 2106 if (sk->sk_type == SOCK_STREAM || 2107 sk->sk_type == SOCK_SEQPACKET) { 2108 skb_queue_walk(&sk->sk_receive_queue, skb) 2109 amount += skb->len; 2110 } else { 2111 skb = skb_peek(&sk->sk_receive_queue); 2112 if (skb) 2113 amount = skb->len; 2114 } 2115 spin_unlock(&sk->sk_receive_queue.lock); 2116 2117 return amount; 2118} 2119EXPORT_SYMBOL_GPL(unix_inq_len); 2120 2121long unix_outq_len(struct sock *sk) 2122{ 2123 return sk_wmem_alloc_get(sk); 2124} 2125EXPORT_SYMBOL_GPL(unix_outq_len); 2126 2127static int unix_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 2128{ 2129 struct sock *sk = sock->sk; 2130 long amount = 0; 2131 int err; 2132 2133 switch (cmd) { 2134 case SIOCOUTQ: 2135 amount = unix_outq_len(sk); 2136 err = put_user(amount, (int __user *)arg); 2137 break; 2138 case SIOCINQ: 2139 amount = unix_inq_len(sk); 2140 if (amount < 0) 2141 err = amount; 2142 else 2143 err = put_user(amount, (int __user *)arg); 2144 break; 2145 default: 2146 err = -ENOIOCTLCMD; 2147 break; 2148 } 2149 return err; 2150} 2151 2152static unsigned int unix_poll(struct file *file, struct socket *sock, poll_table *wait) 2153{ 2154 struct sock *sk = sock->sk; 2155 unsigned int mask; 2156 2157 sock_poll_wait(file, sk_sleep(sk), wait); 2158 mask = 0; 2159 2160 /* exceptional events? */ 2161 if (sk->sk_err) 2162 mask |= POLLERR; 2163 if (sk->sk_shutdown == SHUTDOWN_MASK) 2164 mask |= POLLHUP; 2165 if (sk->sk_shutdown & RCV_SHUTDOWN) 2166 mask |= POLLRDHUP | POLLIN | POLLRDNORM; 2167 2168 /* readable? */ 2169 if (!skb_queue_empty(&sk->sk_receive_queue)) 2170 mask |= POLLIN | POLLRDNORM; 2171 2172 /* Connection-based need to check for termination and startup */ 2173 if ((sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) && 2174 sk->sk_state == TCP_CLOSE) 2175 mask |= POLLHUP; 2176 2177 /* 2178 * we set writable also when the other side has shut down the 2179 * connection. This prevents stuck sockets. 2180 */ 2181 if (unix_writable(sk)) 2182 mask |= POLLOUT | POLLWRNORM | POLLWRBAND; 2183 2184 return mask; 2185} 2186 2187static unsigned int unix_dgram_poll(struct file *file, struct socket *sock, 2188 poll_table *wait) 2189{ 2190 struct sock *sk = sock->sk, *other; 2191 unsigned int mask, writable; 2192 2193 sock_poll_wait(file, sk_sleep(sk), wait); 2194 mask = 0; 2195 2196 /* exceptional events? */ 2197 if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue)) 2198 mask |= POLLERR; 2199 if (sk->sk_shutdown & RCV_SHUTDOWN) 2200 mask |= POLLRDHUP | POLLIN | POLLRDNORM; 2201 if (sk->sk_shutdown == SHUTDOWN_MASK) 2202 mask |= POLLHUP; 2203 2204 /* readable? */ 2205 if (!skb_queue_empty(&sk->sk_receive_queue)) 2206 mask |= POLLIN | POLLRDNORM; 2207 2208 /* Connection-based need to check for termination and startup */ 2209 if (sk->sk_type == SOCK_SEQPACKET) { 2210 if (sk->sk_state == TCP_CLOSE) 2211 mask |= POLLHUP; 2212 /* connection hasn't started yet? */ 2213 if (sk->sk_state == TCP_SYN_SENT) 2214 return mask; 2215 } 2216 2217 /* No write status requested, avoid expensive OUT tests. */ 2218 if (!(poll_requested_events(wait) & (POLLWRBAND|POLLWRNORM|POLLOUT))) 2219 return mask; 2220 2221 writable = unix_writable(sk); 2222 other = unix_peer_get(sk); 2223 if (other) { 2224 if (unix_peer(other) != sk) { 2225 sock_poll_wait(file, &unix_sk(other)->peer_wait, wait); 2226 if (unix_recvq_full(other)) 2227 writable = 0; 2228 } 2229 sock_put(other); 2230 } 2231 2232 if (writable) 2233 mask |= POLLOUT | POLLWRNORM | POLLWRBAND; 2234 else 2235 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags); 2236 2237 return mask; 2238} 2239 2240#ifdef CONFIG_PROC_FS 2241static struct sock *first_unix_socket(int *i) 2242{ 2243 for (*i = 0; *i <= UNIX_HASH_SIZE; (*i)++) { 2244 if (!hlist_empty(&unix_socket_table[*i])) 2245 return __sk_head(&unix_socket_table[*i]); 2246 } 2247 return NULL; 2248} 2249 2250static struct sock *next_unix_socket(int *i, struct sock *s) 2251{ 2252 struct sock *next = sk_next(s); 2253 /* More in this chain? */ 2254 if (next) 2255 return next; 2256 /* Look for next non-empty chain. */ 2257 for ((*i)++; *i <= UNIX_HASH_SIZE; (*i)++) { 2258 if (!hlist_empty(&unix_socket_table[*i])) 2259 return __sk_head(&unix_socket_table[*i]); 2260 } 2261 return NULL; 2262} 2263 2264struct unix_iter_state { 2265 struct seq_net_private p; 2266 int i; 2267}; 2268 2269static struct sock *unix_seq_idx(struct seq_file *seq, loff_t pos) 2270{ 2271 struct unix_iter_state *iter = seq->private; 2272 loff_t off = 0; 2273 struct sock *s; 2274 2275 for (s = first_unix_socket(&iter->i); s; s = next_unix_socket(&iter->i, s)) { 2276 if (sock_net(s) != seq_file_net(seq)) 2277 continue; 2278 if (off == pos) 2279 return s; 2280 ++off; 2281 } 2282 return NULL; 2283} 2284 2285static void *unix_seq_start(struct seq_file *seq, loff_t *pos) 2286 __acquires(unix_table_lock) 2287{ 2288 spin_lock(&unix_table_lock); 2289 return *pos ? unix_seq_idx(seq, *pos - 1) : SEQ_START_TOKEN; 2290} 2291 2292static void *unix_seq_next(struct seq_file *seq, void *v, loff_t *pos) 2293{ 2294 struct unix_iter_state *iter = seq->private; 2295 struct sock *sk = v; 2296 ++*pos; 2297 2298 if (v == SEQ_START_TOKEN) 2299 sk = first_unix_socket(&iter->i); 2300 else 2301 sk = next_unix_socket(&iter->i, sk); 2302 while (sk && (sock_net(sk) != seq_file_net(seq))) 2303 sk = next_unix_socket(&iter->i, sk); 2304 return sk; 2305} 2306 2307static void unix_seq_stop(struct seq_file *seq, void *v) 2308 __releases(unix_table_lock) 2309{ 2310 spin_unlock(&unix_table_lock); 2311} 2312 2313static int unix_seq_show(struct seq_file *seq, void *v) 2314{ 2315 2316 if (v == SEQ_START_TOKEN) 2317 seq_puts(seq, "Num RefCount Protocol Flags Type St " 2318 "Inode Path\n"); 2319 else { 2320 struct sock *s = v; 2321 struct unix_sock *u = unix_sk(s); 2322 unix_state_lock(s); 2323 2324 seq_printf(seq, "%pK: %08X %08X %08X %04X %02X %5lu", 2325 s, 2326 atomic_read(&s->sk_refcnt), 2327 0, 2328 s->sk_state == TCP_LISTEN ? __SO_ACCEPTCON : 0, 2329 s->sk_type, 2330 s->sk_socket ? 2331 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTED : SS_UNCONNECTED) : 2332 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTING : SS_DISCONNECTING), 2333 sock_i_ino(s)); 2334 2335 if (u->addr) { 2336 int i, len; 2337 seq_putc(seq, ' '); 2338 2339 i = 0; 2340 len = u->addr->len - sizeof(short); 2341 if (!UNIX_ABSTRACT(s)) 2342 len--; 2343 else { 2344 seq_putc(seq, '@'); 2345 i++; 2346 } 2347 for ( ; i < len; i++) 2348 seq_putc(seq, u->addr->name->sun_path[i]); 2349 } 2350 unix_state_unlock(s); 2351 seq_putc(seq, '\n'); 2352 } 2353 2354 return 0; 2355} 2356 2357static const struct seq_operations unix_seq_ops = { 2358 .start = unix_seq_start, 2359 .next = unix_seq_next, 2360 .stop = unix_seq_stop, 2361 .show = unix_seq_show, 2362}; 2363 2364static int unix_seq_open(struct inode *inode, struct file *file) 2365{ 2366 return seq_open_net(inode, file, &unix_seq_ops, 2367 sizeof(struct unix_iter_state)); 2368} 2369 2370static const struct file_operations unix_seq_fops = { 2371 .owner = THIS_MODULE, 2372 .open = unix_seq_open, 2373 .read = seq_read, 2374 .llseek = seq_lseek, 2375 .release = seq_release_net, 2376}; 2377 2378#endif 2379 2380static const struct net_proto_family unix_family_ops = { 2381 .family = PF_UNIX, 2382 .create = unix_create, 2383 .owner = THIS_MODULE, 2384}; 2385 2386 2387static int __net_init unix_net_init(struct net *net) 2388{ 2389 int error = -ENOMEM; 2390 2391 net->unx.sysctl_max_dgram_qlen = 10; 2392 if (unix_sysctl_register(net)) 2393 goto out; 2394 2395#ifdef CONFIG_PROC_FS 2396 if (!proc_net_fops_create(net, "unix", 0, &unix_seq_fops)) { 2397 unix_sysctl_unregister(net); 2398 goto out; 2399 } 2400#endif 2401 error = 0; 2402out: 2403 return error; 2404} 2405 2406static void __net_exit unix_net_exit(struct net *net) 2407{ 2408 unix_sysctl_unregister(net); 2409 proc_net_remove(net, "unix"); 2410} 2411 2412static struct pernet_operations unix_net_ops = { 2413 .init = unix_net_init, 2414 .exit = unix_net_exit, 2415}; 2416 2417static int __init af_unix_init(void) 2418{ 2419 int rc = -1; 2420 struct sk_buff *dummy_skb; 2421 2422 BUILD_BUG_ON(sizeof(struct unix_skb_parms) > sizeof(dummy_skb->cb)); 2423 2424 rc = proto_register(&unix_proto, 1); 2425 if (rc != 0) { 2426 printk(KERN_CRIT "%s: Cannot create unix_sock SLAB cache!\n", 2427 __func__); 2428 goto out; 2429 } 2430 2431 sock_register(&unix_family_ops); 2432 register_pernet_subsys(&unix_net_ops); 2433out: 2434 return rc; 2435} 2436 2437static void __exit af_unix_exit(void) 2438{ 2439 sock_unregister(PF_UNIX); 2440 proto_unregister(&unix_proto); 2441 unregister_pernet_subsys(&unix_net_ops); 2442} 2443 2444/* Earlier than device_initcall() so that other drivers invoking 2445 request_module() don't end up in a loop when modprobe tries 2446 to use a UNIX socket. But later than subsys_initcall() because 2447 we depend on stuff initialised there */ 2448fs_initcall(af_unix_init); 2449module_exit(af_unix_exit); 2450 2451MODULE_LICENSE("GPL"); 2452MODULE_ALIAS_NETPROTO(PF_UNIX); 2453