af_unix.c revision fc0d753641f7b919c7273d9bd21ae6ab45e757f3
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 unix_hash_fold(__wsum n) 153{ 154 unsigned hash = (__force unsigned)n; 155 hash ^= hash>>16; 156 hash ^= hash>>8; 157 return hash&(UNIX_HASH_SIZE-1); 158} 159 160#define unix_peer(sk) (unix_sk(sk)->peer) 161 162static inline int unix_our_peer(struct sock *sk, struct sock *osk) 163{ 164 return unix_peer(osk) == sk; 165} 166 167static inline int unix_may_send(struct sock *sk, struct sock *osk) 168{ 169 return unix_peer(osk) == NULL || unix_our_peer(sk, osk); 170} 171 172static inline int unix_recvq_full(struct sock const *sk) 173{ 174 return skb_queue_len(&sk->sk_receive_queue) > sk->sk_max_ack_backlog; 175} 176 177struct sock *unix_peer_get(struct sock *s) 178{ 179 struct sock *peer; 180 181 unix_state_lock(s); 182 peer = unix_peer(s); 183 if (peer) 184 sock_hold(peer); 185 unix_state_unlock(s); 186 return peer; 187} 188EXPORT_SYMBOL_GPL(unix_peer_get); 189 190static inline void unix_release_addr(struct unix_address *addr) 191{ 192 if (atomic_dec_and_test(&addr->refcnt)) 193 kfree(addr); 194} 195 196/* 197 * Check unix socket name: 198 * - should be not zero length. 199 * - if started by not zero, should be NULL terminated (FS object) 200 * - if started by zero, it is abstract name. 201 */ 202 203static int unix_mkname(struct sockaddr_un *sunaddr, int len, unsigned *hashp) 204{ 205 if (len <= sizeof(short) || len > sizeof(*sunaddr)) 206 return -EINVAL; 207 if (!sunaddr || sunaddr->sun_family != AF_UNIX) 208 return -EINVAL; 209 if (sunaddr->sun_path[0]) { 210 /* 211 * This may look like an off by one error but it is a bit more 212 * subtle. 108 is the longest valid AF_UNIX path for a binding. 213 * sun_path[108] doesn't as such exist. However in kernel space 214 * we are guaranteed that it is a valid memory location in our 215 * kernel address buffer. 216 */ 217 ((char *)sunaddr)[len] = 0; 218 len = strlen(sunaddr->sun_path)+1+sizeof(short); 219 return len; 220 } 221 222 *hashp = unix_hash_fold(csum_partial(sunaddr, len, 0)); 223 return len; 224} 225 226static void __unix_remove_socket(struct sock *sk) 227{ 228 sk_del_node_init(sk); 229} 230 231static void __unix_insert_socket(struct hlist_head *list, struct sock *sk) 232{ 233 WARN_ON(!sk_unhashed(sk)); 234 sk_add_node(sk, list); 235} 236 237static inline void unix_remove_socket(struct sock *sk) 238{ 239 spin_lock(&unix_table_lock); 240 __unix_remove_socket(sk); 241 spin_unlock(&unix_table_lock); 242} 243 244static inline void unix_insert_socket(struct hlist_head *list, struct sock *sk) 245{ 246 spin_lock(&unix_table_lock); 247 __unix_insert_socket(list, sk); 248 spin_unlock(&unix_table_lock); 249} 250 251static struct sock *__unix_find_socket_byname(struct net *net, 252 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 (!net_eq(sock_net(s), net)) 262 continue; 263 264 if (u->addr->len == len && 265 !memcmp(u->addr->name, sunname, len)) 266 goto found; 267 } 268 s = NULL; 269found: 270 return s; 271} 272 273static inline struct sock *unix_find_socket_byname(struct net *net, 274 struct sockaddr_un *sunname, 275 int len, int type, 276 unsigned hash) 277{ 278 struct sock *s; 279 280 spin_lock(&unix_table_lock); 281 s = __unix_find_socket_byname(net, sunname, len, type, hash); 282 if (s) 283 sock_hold(s); 284 spin_unlock(&unix_table_lock); 285 return s; 286} 287 288static struct sock *unix_find_socket_byinode(struct inode *i) 289{ 290 struct sock *s; 291 struct hlist_node *node; 292 293 spin_lock(&unix_table_lock); 294 sk_for_each(s, node, 295 &unix_socket_table[i->i_ino & (UNIX_HASH_SIZE - 1)]) { 296 struct dentry *dentry = unix_sk(s)->dentry; 297 298 if (dentry && dentry->d_inode == i) { 299 sock_hold(s); 300 goto found; 301 } 302 } 303 s = NULL; 304found: 305 spin_unlock(&unix_table_lock); 306 return s; 307} 308 309static inline int unix_writable(struct sock *sk) 310{ 311 return (atomic_read(&sk->sk_wmem_alloc) << 2) <= sk->sk_sndbuf; 312} 313 314static void unix_write_space(struct sock *sk) 315{ 316 struct socket_wq *wq; 317 318 rcu_read_lock(); 319 if (unix_writable(sk)) { 320 wq = rcu_dereference(sk->sk_wq); 321 if (wq_has_sleeper(wq)) 322 wake_up_interruptible_sync_poll(&wq->wait, 323 POLLOUT | POLLWRNORM | POLLWRBAND); 324 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT); 325 } 326 rcu_read_unlock(); 327} 328 329/* When dgram socket disconnects (or changes its peer), we clear its receive 330 * queue of packets arrived from previous peer. First, it allows to do 331 * flow control based only on wmem_alloc; second, sk connected to peer 332 * may receive messages only from that peer. */ 333static void unix_dgram_disconnected(struct sock *sk, struct sock *other) 334{ 335 if (!skb_queue_empty(&sk->sk_receive_queue)) { 336 skb_queue_purge(&sk->sk_receive_queue); 337 wake_up_interruptible_all(&unix_sk(sk)->peer_wait); 338 339 /* If one link of bidirectional dgram pipe is disconnected, 340 * we signal error. Messages are lost. Do not make this, 341 * when peer was not connected to us. 342 */ 343 if (!sock_flag(other, SOCK_DEAD) && unix_peer(other) == sk) { 344 other->sk_err = ECONNRESET; 345 other->sk_error_report(other); 346 } 347 } 348} 349 350static void unix_sock_destructor(struct sock *sk) 351{ 352 struct unix_sock *u = unix_sk(sk); 353 354 skb_queue_purge(&sk->sk_receive_queue); 355 356 WARN_ON(atomic_read(&sk->sk_wmem_alloc)); 357 WARN_ON(!sk_unhashed(sk)); 358 WARN_ON(sk->sk_socket); 359 if (!sock_flag(sk, SOCK_DEAD)) { 360 printk(KERN_INFO "Attempt to release alive unix socket: %p\n", sk); 361 return; 362 } 363 364 if (u->addr) 365 unix_release_addr(u->addr); 366 367 atomic_long_dec(&unix_nr_socks); 368 local_bh_disable(); 369 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1); 370 local_bh_enable(); 371#ifdef UNIX_REFCNT_DEBUG 372 printk(KERN_DEBUG "UNIX %p is destroyed, %ld are still alive.\n", sk, 373 atomic_long_read(&unix_nr_socks)); 374#endif 375} 376 377static int unix_release_sock(struct sock *sk, int embrion) 378{ 379 struct unix_sock *u = unix_sk(sk); 380 struct dentry *dentry; 381 struct vfsmount *mnt; 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 dentry = u->dentry; 393 u->dentry = NULL; 394 mnt = u->mnt; 395 u->mnt = NULL; 396 state = sk->sk_state; 397 sk->sk_state = TCP_CLOSE; 398 unix_state_unlock(sk); 399 400 wake_up_interruptible_all(&u->peer_wait); 401 402 skpair = unix_peer(sk); 403 404 if (skpair != NULL) { 405 if (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) { 406 unix_state_lock(skpair); 407 /* No more writes */ 408 skpair->sk_shutdown = SHUTDOWN_MASK; 409 if (!skb_queue_empty(&sk->sk_receive_queue) || embrion) 410 skpair->sk_err = ECONNRESET; 411 unix_state_unlock(skpair); 412 skpair->sk_state_change(skpair); 413 sk_wake_async(skpair, SOCK_WAKE_WAITD, POLL_HUP); 414 } 415 sock_put(skpair); /* It may now die */ 416 unix_peer(sk) = NULL; 417 } 418 419 /* Try to flush out this socket. Throw out buffers at least */ 420 421 while ((skb = skb_dequeue(&sk->sk_receive_queue)) != NULL) { 422 if (state == TCP_LISTEN) 423 unix_release_sock(skb->sk, 1); 424 /* passed fds are erased in the kfree_skb hook */ 425 kfree_skb(skb); 426 } 427 428 if (dentry) { 429 dput(dentry); 430 mntput(mnt); 431 } 432 433 sock_put(sk); 434 435 /* ---- Socket is dead now and most probably destroyed ---- */ 436 437 /* 438 * Fixme: BSD difference: In BSD all sockets connected to use get 439 * ECONNRESET and we die on the spot. In Linux we behave 440 * like files and pipes do and wait for the last 441 * dereference. 442 * 443 * Can't we simply set sock->err? 444 * 445 * What the above comment does talk about? --ANK(980817) 446 */ 447 448 if (unix_tot_inflight) 449 unix_gc(); /* Garbage collect fds */ 450 451 return 0; 452} 453 454static void init_peercred(struct sock *sk) 455{ 456 put_pid(sk->sk_peer_pid); 457 if (sk->sk_peer_cred) 458 put_cred(sk->sk_peer_cred); 459 sk->sk_peer_pid = get_pid(task_tgid(current)); 460 sk->sk_peer_cred = get_current_cred(); 461} 462 463static void copy_peercred(struct sock *sk, struct sock *peersk) 464{ 465 put_pid(sk->sk_peer_pid); 466 if (sk->sk_peer_cred) 467 put_cred(sk->sk_peer_cred); 468 sk->sk_peer_pid = get_pid(peersk->sk_peer_pid); 469 sk->sk_peer_cred = get_cred(peersk->sk_peer_cred); 470} 471 472static int unix_listen(struct socket *sock, int backlog) 473{ 474 int err; 475 struct sock *sk = sock->sk; 476 struct unix_sock *u = unix_sk(sk); 477 struct pid *old_pid = NULL; 478 const struct cred *old_cred = NULL; 479 480 err = -EOPNOTSUPP; 481 if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET) 482 goto out; /* Only stream/seqpacket sockets accept */ 483 err = -EINVAL; 484 if (!u->addr) 485 goto out; /* No listens on an unbound socket */ 486 unix_state_lock(sk); 487 if (sk->sk_state != TCP_CLOSE && sk->sk_state != TCP_LISTEN) 488 goto out_unlock; 489 if (backlog > sk->sk_max_ack_backlog) 490 wake_up_interruptible_all(&u->peer_wait); 491 sk->sk_max_ack_backlog = backlog; 492 sk->sk_state = TCP_LISTEN; 493 /* set credentials so connect can copy them */ 494 init_peercred(sk); 495 err = 0; 496 497out_unlock: 498 unix_state_unlock(sk); 499 put_pid(old_pid); 500 if (old_cred) 501 put_cred(old_cred); 502out: 503 return err; 504} 505 506static int unix_release(struct socket *); 507static int unix_bind(struct socket *, struct sockaddr *, int); 508static int unix_stream_connect(struct socket *, struct sockaddr *, 509 int addr_len, int flags); 510static int unix_socketpair(struct socket *, struct socket *); 511static int unix_accept(struct socket *, struct socket *, int); 512static int unix_getname(struct socket *, struct sockaddr *, int *, int); 513static unsigned int unix_poll(struct file *, struct socket *, poll_table *); 514static unsigned int unix_dgram_poll(struct file *, struct socket *, 515 poll_table *); 516static int unix_ioctl(struct socket *, unsigned int, unsigned long); 517static int unix_shutdown(struct socket *, int); 518static int unix_stream_sendmsg(struct kiocb *, struct socket *, 519 struct msghdr *, size_t); 520static int unix_stream_recvmsg(struct kiocb *, struct socket *, 521 struct msghdr *, size_t, int); 522static int unix_dgram_sendmsg(struct kiocb *, struct socket *, 523 struct msghdr *, size_t); 524static int unix_dgram_recvmsg(struct kiocb *, struct socket *, 525 struct msghdr *, size_t, int); 526static int unix_dgram_connect(struct socket *, struct sockaddr *, 527 int, int); 528static int unix_seqpacket_sendmsg(struct kiocb *, struct socket *, 529 struct msghdr *, size_t); 530static int unix_seqpacket_recvmsg(struct kiocb *, struct socket *, 531 struct msghdr *, size_t, int); 532 533static void unix_set_peek_off(struct sock *sk, int val) 534{ 535 struct unix_sock *u = unix_sk(sk); 536 537 mutex_lock(&u->readlock); 538 sk->sk_peek_off = val; 539 mutex_unlock(&u->readlock); 540} 541 542 543static const struct proto_ops unix_stream_ops = { 544 .family = PF_UNIX, 545 .owner = THIS_MODULE, 546 .release = unix_release, 547 .bind = unix_bind, 548 .connect = unix_stream_connect, 549 .socketpair = unix_socketpair, 550 .accept = unix_accept, 551 .getname = unix_getname, 552 .poll = unix_poll, 553 .ioctl = unix_ioctl, 554 .listen = unix_listen, 555 .shutdown = unix_shutdown, 556 .setsockopt = sock_no_setsockopt, 557 .getsockopt = sock_no_getsockopt, 558 .sendmsg = unix_stream_sendmsg, 559 .recvmsg = unix_stream_recvmsg, 560 .mmap = sock_no_mmap, 561 .sendpage = sock_no_sendpage, 562 .set_peek_off = unix_set_peek_off, 563}; 564 565static const struct proto_ops unix_dgram_ops = { 566 .family = PF_UNIX, 567 .owner = THIS_MODULE, 568 .release = unix_release, 569 .bind = unix_bind, 570 .connect = unix_dgram_connect, 571 .socketpair = unix_socketpair, 572 .accept = sock_no_accept, 573 .getname = unix_getname, 574 .poll = unix_dgram_poll, 575 .ioctl = unix_ioctl, 576 .listen = sock_no_listen, 577 .shutdown = unix_shutdown, 578 .setsockopt = sock_no_setsockopt, 579 .getsockopt = sock_no_getsockopt, 580 .sendmsg = unix_dgram_sendmsg, 581 .recvmsg = unix_dgram_recvmsg, 582 .mmap = sock_no_mmap, 583 .sendpage = sock_no_sendpage, 584 .set_peek_off = unix_set_peek_off, 585}; 586 587static const struct proto_ops unix_seqpacket_ops = { 588 .family = PF_UNIX, 589 .owner = THIS_MODULE, 590 .release = unix_release, 591 .bind = unix_bind, 592 .connect = unix_stream_connect, 593 .socketpair = unix_socketpair, 594 .accept = unix_accept, 595 .getname = unix_getname, 596 .poll = unix_dgram_poll, 597 .ioctl = unix_ioctl, 598 .listen = unix_listen, 599 .shutdown = unix_shutdown, 600 .setsockopt = sock_no_setsockopt, 601 .getsockopt = sock_no_getsockopt, 602 .sendmsg = unix_seqpacket_sendmsg, 603 .recvmsg = unix_seqpacket_recvmsg, 604 .mmap = sock_no_mmap, 605 .sendpage = sock_no_sendpage, 606 .set_peek_off = unix_set_peek_off, 607}; 608 609static struct proto unix_proto = { 610 .name = "UNIX", 611 .owner = THIS_MODULE, 612 .obj_size = sizeof(struct unix_sock), 613}; 614 615/* 616 * AF_UNIX sockets do not interact with hardware, hence they 617 * dont trigger interrupts - so it's safe for them to have 618 * bh-unsafe locking for their sk_receive_queue.lock. Split off 619 * this special lock-class by reinitializing the spinlock key: 620 */ 621static struct lock_class_key af_unix_sk_receive_queue_lock_key; 622 623static struct sock *unix_create1(struct net *net, struct socket *sock) 624{ 625 struct sock *sk = NULL; 626 struct unix_sock *u; 627 628 atomic_long_inc(&unix_nr_socks); 629 if (atomic_long_read(&unix_nr_socks) > 2 * get_max_files()) 630 goto out; 631 632 sk = sk_alloc(net, PF_UNIX, GFP_KERNEL, &unix_proto); 633 if (!sk) 634 goto out; 635 636 sock_init_data(sock, sk); 637 lockdep_set_class(&sk->sk_receive_queue.lock, 638 &af_unix_sk_receive_queue_lock_key); 639 640 sk->sk_write_space = unix_write_space; 641 sk->sk_max_ack_backlog = net->unx.sysctl_max_dgram_qlen; 642 sk->sk_destruct = unix_sock_destructor; 643 u = unix_sk(sk); 644 u->dentry = NULL; 645 u->mnt = NULL; 646 spin_lock_init(&u->lock); 647 atomic_long_set(&u->inflight, 0); 648 INIT_LIST_HEAD(&u->link); 649 mutex_init(&u->readlock); /* single task reading lock */ 650 init_waitqueue_head(&u->peer_wait); 651 unix_insert_socket(unix_sockets_unbound, sk); 652out: 653 if (sk == NULL) 654 atomic_long_dec(&unix_nr_socks); 655 else { 656 local_bh_disable(); 657 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); 658 local_bh_enable(); 659 } 660 return sk; 661} 662 663static int unix_create(struct net *net, struct socket *sock, int protocol, 664 int kern) 665{ 666 if (protocol && protocol != PF_UNIX) 667 return -EPROTONOSUPPORT; 668 669 sock->state = SS_UNCONNECTED; 670 671 switch (sock->type) { 672 case SOCK_STREAM: 673 sock->ops = &unix_stream_ops; 674 break; 675 /* 676 * Believe it or not BSD has AF_UNIX, SOCK_RAW though 677 * nothing uses it. 678 */ 679 case SOCK_RAW: 680 sock->type = SOCK_DGRAM; 681 case SOCK_DGRAM: 682 sock->ops = &unix_dgram_ops; 683 break; 684 case SOCK_SEQPACKET: 685 sock->ops = &unix_seqpacket_ops; 686 break; 687 default: 688 return -ESOCKTNOSUPPORT; 689 } 690 691 return unix_create1(net, sock) ? 0 : -ENOMEM; 692} 693 694static int unix_release(struct socket *sock) 695{ 696 struct sock *sk = sock->sk; 697 698 if (!sk) 699 return 0; 700 701 sock->sk = NULL; 702 703 return unix_release_sock(sk, 0); 704} 705 706static int unix_autobind(struct socket *sock) 707{ 708 struct sock *sk = sock->sk; 709 struct net *net = sock_net(sk); 710 struct unix_sock *u = unix_sk(sk); 711 static u32 ordernum = 1; 712 struct unix_address *addr; 713 int err; 714 unsigned int retries = 0; 715 716 mutex_lock(&u->readlock); 717 718 err = 0; 719 if (u->addr) 720 goto out; 721 722 err = -ENOMEM; 723 addr = kzalloc(sizeof(*addr) + sizeof(short) + 16, GFP_KERNEL); 724 if (!addr) 725 goto out; 726 727 addr->name->sun_family = AF_UNIX; 728 atomic_set(&addr->refcnt, 1); 729 730retry: 731 addr->len = sprintf(addr->name->sun_path+1, "%05x", ordernum) + 1 + sizeof(short); 732 addr->hash = unix_hash_fold(csum_partial(addr->name, addr->len, 0)); 733 734 spin_lock(&unix_table_lock); 735 ordernum = (ordernum+1)&0xFFFFF; 736 737 if (__unix_find_socket_byname(net, addr->name, addr->len, sock->type, 738 addr->hash)) { 739 spin_unlock(&unix_table_lock); 740 /* 741 * __unix_find_socket_byname() may take long time if many names 742 * are already in use. 743 */ 744 cond_resched(); 745 /* Give up if all names seems to be in use. */ 746 if (retries++ == 0xFFFFF) { 747 err = -ENOSPC; 748 kfree(addr); 749 goto out; 750 } 751 goto retry; 752 } 753 addr->hash ^= sk->sk_type; 754 755 __unix_remove_socket(sk); 756 u->addr = addr; 757 __unix_insert_socket(&unix_socket_table[addr->hash], sk); 758 spin_unlock(&unix_table_lock); 759 err = 0; 760 761out: mutex_unlock(&u->readlock); 762 return err; 763} 764 765static struct sock *unix_find_other(struct net *net, 766 struct sockaddr_un *sunname, int len, 767 int type, unsigned hash, int *error) 768{ 769 struct sock *u; 770 struct path path; 771 int err = 0; 772 773 if (sunname->sun_path[0]) { 774 struct inode *inode; 775 err = kern_path(sunname->sun_path, LOOKUP_FOLLOW, &path); 776 if (err) 777 goto fail; 778 inode = path.dentry->d_inode; 779 err = inode_permission(inode, MAY_WRITE); 780 if (err) 781 goto put_fail; 782 783 err = -ECONNREFUSED; 784 if (!S_ISSOCK(inode->i_mode)) 785 goto put_fail; 786 u = unix_find_socket_byinode(inode); 787 if (!u) 788 goto put_fail; 789 790 if (u->sk_type == type) 791 touch_atime(path.mnt, path.dentry); 792 793 path_put(&path); 794 795 err = -EPROTOTYPE; 796 if (u->sk_type != type) { 797 sock_put(u); 798 goto fail; 799 } 800 } else { 801 err = -ECONNREFUSED; 802 u = unix_find_socket_byname(net, sunname, len, type, hash); 803 if (u) { 804 struct dentry *dentry; 805 dentry = unix_sk(u)->dentry; 806 if (dentry) 807 touch_atime(unix_sk(u)->mnt, dentry); 808 } else 809 goto fail; 810 } 811 return u; 812 813put_fail: 814 path_put(&path); 815fail: 816 *error = err; 817 return NULL; 818} 819 820 821static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len) 822{ 823 struct sock *sk = sock->sk; 824 struct net *net = sock_net(sk); 825 struct unix_sock *u = unix_sk(sk); 826 struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr; 827 char *sun_path = sunaddr->sun_path; 828 struct dentry *dentry = NULL; 829 struct path path; 830 int err; 831 unsigned hash; 832 struct unix_address *addr; 833 struct hlist_head *list; 834 835 err = -EINVAL; 836 if (sunaddr->sun_family != AF_UNIX) 837 goto out; 838 839 if (addr_len == sizeof(short)) { 840 err = unix_autobind(sock); 841 goto out; 842 } 843 844 err = unix_mkname(sunaddr, addr_len, &hash); 845 if (err < 0) 846 goto out; 847 addr_len = err; 848 849 mutex_lock(&u->readlock); 850 851 err = -EINVAL; 852 if (u->addr) 853 goto out_up; 854 855 err = -ENOMEM; 856 addr = kmalloc(sizeof(*addr)+addr_len, GFP_KERNEL); 857 if (!addr) 858 goto out_up; 859 860 memcpy(addr->name, sunaddr, addr_len); 861 addr->len = addr_len; 862 addr->hash = hash ^ sk->sk_type; 863 atomic_set(&addr->refcnt, 1); 864 865 if (sun_path[0]) { 866 umode_t mode; 867 err = 0; 868 /* 869 * Get the parent directory, calculate the hash for last 870 * component. 871 */ 872 dentry = kern_path_create(AT_FDCWD, sun_path, &path, 0); 873 err = PTR_ERR(dentry); 874 if (IS_ERR(dentry)) 875 goto out_mknod_parent; 876 877 /* 878 * All right, let's create it. 879 */ 880 mode = S_IFSOCK | 881 (SOCK_INODE(sock)->i_mode & ~current_umask()); 882 err = mnt_want_write(path.mnt); 883 if (err) 884 goto out_mknod_dput; 885 err = security_path_mknod(&path, dentry, mode, 0); 886 if (err) 887 goto out_mknod_drop_write; 888 err = vfs_mknod(path.dentry->d_inode, dentry, mode, 0); 889out_mknod_drop_write: 890 mnt_drop_write(path.mnt); 891 if (err) 892 goto out_mknod_dput; 893 mutex_unlock(&path.dentry->d_inode->i_mutex); 894 dput(path.dentry); 895 path.dentry = dentry; 896 897 addr->hash = UNIX_HASH_SIZE; 898 } 899 900 spin_lock(&unix_table_lock); 901 902 if (!sun_path[0]) { 903 err = -EADDRINUSE; 904 if (__unix_find_socket_byname(net, sunaddr, addr_len, 905 sk->sk_type, hash)) { 906 unix_release_addr(addr); 907 goto out_unlock; 908 } 909 910 list = &unix_socket_table[addr->hash]; 911 } else { 912 list = &unix_socket_table[dentry->d_inode->i_ino & (UNIX_HASH_SIZE-1)]; 913 u->dentry = path.dentry; 914 u->mnt = path.mnt; 915 } 916 917 err = 0; 918 __unix_remove_socket(sk); 919 u->addr = addr; 920 __unix_insert_socket(list, sk); 921 922out_unlock: 923 spin_unlock(&unix_table_lock); 924out_up: 925 mutex_unlock(&u->readlock); 926out: 927 return err; 928 929out_mknod_dput: 930 dput(dentry); 931 mutex_unlock(&path.dentry->d_inode->i_mutex); 932 path_put(&path); 933out_mknod_parent: 934 if (err == -EEXIST) 935 err = -EADDRINUSE; 936 unix_release_addr(addr); 937 goto out_up; 938} 939 940static void unix_state_double_lock(struct sock *sk1, struct sock *sk2) 941{ 942 if (unlikely(sk1 == sk2) || !sk2) { 943 unix_state_lock(sk1); 944 return; 945 } 946 if (sk1 < sk2) { 947 unix_state_lock(sk1); 948 unix_state_lock_nested(sk2); 949 } else { 950 unix_state_lock(sk2); 951 unix_state_lock_nested(sk1); 952 } 953} 954 955static void unix_state_double_unlock(struct sock *sk1, struct sock *sk2) 956{ 957 if (unlikely(sk1 == sk2) || !sk2) { 958 unix_state_unlock(sk1); 959 return; 960 } 961 unix_state_unlock(sk1); 962 unix_state_unlock(sk2); 963} 964 965static int unix_dgram_connect(struct socket *sock, struct sockaddr *addr, 966 int alen, int flags) 967{ 968 struct sock *sk = sock->sk; 969 struct net *net = sock_net(sk); 970 struct sockaddr_un *sunaddr = (struct sockaddr_un *)addr; 971 struct sock *other; 972 unsigned hash; 973 int err; 974 975 if (addr->sa_family != AF_UNSPEC) { 976 err = unix_mkname(sunaddr, alen, &hash); 977 if (err < 0) 978 goto out; 979 alen = err; 980 981 if (test_bit(SOCK_PASSCRED, &sock->flags) && 982 !unix_sk(sk)->addr && (err = unix_autobind(sock)) != 0) 983 goto out; 984 985restart: 986 other = unix_find_other(net, sunaddr, alen, sock->type, hash, &err); 987 if (!other) 988 goto out; 989 990 unix_state_double_lock(sk, other); 991 992 /* Apparently VFS overslept socket death. Retry. */ 993 if (sock_flag(other, SOCK_DEAD)) { 994 unix_state_double_unlock(sk, other); 995 sock_put(other); 996 goto restart; 997 } 998 999 err = -EPERM; 1000 if (!unix_may_send(sk, other)) 1001 goto out_unlock; 1002 1003 err = security_unix_may_send(sk->sk_socket, other->sk_socket); 1004 if (err) 1005 goto out_unlock; 1006 1007 } else { 1008 /* 1009 * 1003.1g breaking connected state with AF_UNSPEC 1010 */ 1011 other = NULL; 1012 unix_state_double_lock(sk, other); 1013 } 1014 1015 /* 1016 * If it was connected, reconnect. 1017 */ 1018 if (unix_peer(sk)) { 1019 struct sock *old_peer = unix_peer(sk); 1020 unix_peer(sk) = other; 1021 unix_state_double_unlock(sk, other); 1022 1023 if (other != old_peer) 1024 unix_dgram_disconnected(sk, old_peer); 1025 sock_put(old_peer); 1026 } else { 1027 unix_peer(sk) = other; 1028 unix_state_double_unlock(sk, other); 1029 } 1030 return 0; 1031 1032out_unlock: 1033 unix_state_double_unlock(sk, other); 1034 sock_put(other); 1035out: 1036 return err; 1037} 1038 1039static long unix_wait_for_peer(struct sock *other, long timeo) 1040{ 1041 struct unix_sock *u = unix_sk(other); 1042 int sched; 1043 DEFINE_WAIT(wait); 1044 1045 prepare_to_wait_exclusive(&u->peer_wait, &wait, TASK_INTERRUPTIBLE); 1046 1047 sched = !sock_flag(other, SOCK_DEAD) && 1048 !(other->sk_shutdown & RCV_SHUTDOWN) && 1049 unix_recvq_full(other); 1050 1051 unix_state_unlock(other); 1052 1053 if (sched) 1054 timeo = schedule_timeout(timeo); 1055 1056 finish_wait(&u->peer_wait, &wait); 1057 return timeo; 1058} 1059 1060static int unix_stream_connect(struct socket *sock, struct sockaddr *uaddr, 1061 int addr_len, int flags) 1062{ 1063 struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr; 1064 struct sock *sk = sock->sk; 1065 struct net *net = sock_net(sk); 1066 struct unix_sock *u = unix_sk(sk), *newu, *otheru; 1067 struct sock *newsk = NULL; 1068 struct sock *other = NULL; 1069 struct sk_buff *skb = NULL; 1070 unsigned hash; 1071 int st; 1072 int err; 1073 long timeo; 1074 1075 err = unix_mkname(sunaddr, addr_len, &hash); 1076 if (err < 0) 1077 goto out; 1078 addr_len = err; 1079 1080 if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr && 1081 (err = unix_autobind(sock)) != 0) 1082 goto out; 1083 1084 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK); 1085 1086 /* First of all allocate resources. 1087 If we will make it after state is locked, 1088 we will have to recheck all again in any case. 1089 */ 1090 1091 err = -ENOMEM; 1092 1093 /* create new sock for complete connection */ 1094 newsk = unix_create1(sock_net(sk), NULL); 1095 if (newsk == NULL) 1096 goto out; 1097 1098 /* Allocate skb for sending to listening sock */ 1099 skb = sock_wmalloc(newsk, 1, 0, GFP_KERNEL); 1100 if (skb == NULL) 1101 goto out; 1102 1103restart: 1104 /* Find listening sock. */ 1105 other = unix_find_other(net, sunaddr, addr_len, sk->sk_type, hash, &err); 1106 if (!other) 1107 goto out; 1108 1109 /* Latch state of peer */ 1110 unix_state_lock(other); 1111 1112 /* Apparently VFS overslept socket death. Retry. */ 1113 if (sock_flag(other, SOCK_DEAD)) { 1114 unix_state_unlock(other); 1115 sock_put(other); 1116 goto restart; 1117 } 1118 1119 err = -ECONNREFUSED; 1120 if (other->sk_state != TCP_LISTEN) 1121 goto out_unlock; 1122 if (other->sk_shutdown & RCV_SHUTDOWN) 1123 goto out_unlock; 1124 1125 if (unix_recvq_full(other)) { 1126 err = -EAGAIN; 1127 if (!timeo) 1128 goto out_unlock; 1129 1130 timeo = unix_wait_for_peer(other, timeo); 1131 1132 err = sock_intr_errno(timeo); 1133 if (signal_pending(current)) 1134 goto out; 1135 sock_put(other); 1136 goto restart; 1137 } 1138 1139 /* Latch our state. 1140 1141 It is tricky place. We need to grab our state lock and cannot 1142 drop lock on peer. It is dangerous because deadlock is 1143 possible. Connect to self case and simultaneous 1144 attempt to connect are eliminated by checking socket 1145 state. other is TCP_LISTEN, if sk is TCP_LISTEN we 1146 check this before attempt to grab lock. 1147 1148 Well, and we have to recheck the state after socket locked. 1149 */ 1150 st = sk->sk_state; 1151 1152 switch (st) { 1153 case TCP_CLOSE: 1154 /* This is ok... continue with connect */ 1155 break; 1156 case TCP_ESTABLISHED: 1157 /* Socket is already connected */ 1158 err = -EISCONN; 1159 goto out_unlock; 1160 default: 1161 err = -EINVAL; 1162 goto out_unlock; 1163 } 1164 1165 unix_state_lock_nested(sk); 1166 1167 if (sk->sk_state != st) { 1168 unix_state_unlock(sk); 1169 unix_state_unlock(other); 1170 sock_put(other); 1171 goto restart; 1172 } 1173 1174 err = security_unix_stream_connect(sk, other, newsk); 1175 if (err) { 1176 unix_state_unlock(sk); 1177 goto out_unlock; 1178 } 1179 1180 /* The way is open! Fastly set all the necessary fields... */ 1181 1182 sock_hold(sk); 1183 unix_peer(newsk) = sk; 1184 newsk->sk_state = TCP_ESTABLISHED; 1185 newsk->sk_type = sk->sk_type; 1186 init_peercred(newsk); 1187 newu = unix_sk(newsk); 1188 RCU_INIT_POINTER(newsk->sk_wq, &newu->peer_wq); 1189 otheru = unix_sk(other); 1190 1191 /* copy address information from listening to new sock*/ 1192 if (otheru->addr) { 1193 atomic_inc(&otheru->addr->refcnt); 1194 newu->addr = otheru->addr; 1195 } 1196 if (otheru->dentry) { 1197 newu->dentry = dget(otheru->dentry); 1198 newu->mnt = mntget(otheru->mnt); 1199 } 1200 1201 /* Set credentials */ 1202 copy_peercred(sk, other); 1203 1204 sock->state = SS_CONNECTED; 1205 sk->sk_state = TCP_ESTABLISHED; 1206 sock_hold(newsk); 1207 1208 smp_mb__after_atomic_inc(); /* sock_hold() does an atomic_inc() */ 1209 unix_peer(sk) = newsk; 1210 1211 unix_state_unlock(sk); 1212 1213 /* take ten and and send info to listening sock */ 1214 spin_lock(&other->sk_receive_queue.lock); 1215 __skb_queue_tail(&other->sk_receive_queue, skb); 1216 spin_unlock(&other->sk_receive_queue.lock); 1217 unix_state_unlock(other); 1218 other->sk_data_ready(other, 0); 1219 sock_put(other); 1220 return 0; 1221 1222out_unlock: 1223 if (other) 1224 unix_state_unlock(other); 1225 1226out: 1227 kfree_skb(skb); 1228 if (newsk) 1229 unix_release_sock(newsk, 0); 1230 if (other) 1231 sock_put(other); 1232 return err; 1233} 1234 1235static int unix_socketpair(struct socket *socka, struct socket *sockb) 1236{ 1237 struct sock *ska = socka->sk, *skb = sockb->sk; 1238 1239 /* Join our sockets back to back */ 1240 sock_hold(ska); 1241 sock_hold(skb); 1242 unix_peer(ska) = skb; 1243 unix_peer(skb) = ska; 1244 init_peercred(ska); 1245 init_peercred(skb); 1246 1247 if (ska->sk_type != SOCK_DGRAM) { 1248 ska->sk_state = TCP_ESTABLISHED; 1249 skb->sk_state = TCP_ESTABLISHED; 1250 socka->state = SS_CONNECTED; 1251 sockb->state = SS_CONNECTED; 1252 } 1253 return 0; 1254} 1255 1256static int unix_accept(struct socket *sock, struct socket *newsock, int flags) 1257{ 1258 struct sock *sk = sock->sk; 1259 struct sock *tsk; 1260 struct sk_buff *skb; 1261 int err; 1262 1263 err = -EOPNOTSUPP; 1264 if (sock->type != SOCK_STREAM && sock->type != SOCK_SEQPACKET) 1265 goto out; 1266 1267 err = -EINVAL; 1268 if (sk->sk_state != TCP_LISTEN) 1269 goto out; 1270 1271 /* If socket state is TCP_LISTEN it cannot change (for now...), 1272 * so that no locks are necessary. 1273 */ 1274 1275 skb = skb_recv_datagram(sk, 0, flags&O_NONBLOCK, &err); 1276 if (!skb) { 1277 /* This means receive shutdown. */ 1278 if (err == 0) 1279 err = -EINVAL; 1280 goto out; 1281 } 1282 1283 tsk = skb->sk; 1284 skb_free_datagram(sk, skb); 1285 wake_up_interruptible(&unix_sk(sk)->peer_wait); 1286 1287 /* attach accepted sock to socket */ 1288 unix_state_lock(tsk); 1289 newsock->state = SS_CONNECTED; 1290 sock_graft(tsk, newsock); 1291 unix_state_unlock(tsk); 1292 return 0; 1293 1294out: 1295 return err; 1296} 1297 1298 1299static int unix_getname(struct socket *sock, struct sockaddr *uaddr, int *uaddr_len, int peer) 1300{ 1301 struct sock *sk = sock->sk; 1302 struct unix_sock *u; 1303 DECLARE_SOCKADDR(struct sockaddr_un *, sunaddr, uaddr); 1304 int err = 0; 1305 1306 if (peer) { 1307 sk = unix_peer_get(sk); 1308 1309 err = -ENOTCONN; 1310 if (!sk) 1311 goto out; 1312 err = 0; 1313 } else { 1314 sock_hold(sk); 1315 } 1316 1317 u = unix_sk(sk); 1318 unix_state_lock(sk); 1319 if (!u->addr) { 1320 sunaddr->sun_family = AF_UNIX; 1321 sunaddr->sun_path[0] = 0; 1322 *uaddr_len = sizeof(short); 1323 } else { 1324 struct unix_address *addr = u->addr; 1325 1326 *uaddr_len = addr->len; 1327 memcpy(sunaddr, addr->name, *uaddr_len); 1328 } 1329 unix_state_unlock(sk); 1330 sock_put(sk); 1331out: 1332 return err; 1333} 1334 1335static void unix_detach_fds(struct scm_cookie *scm, struct sk_buff *skb) 1336{ 1337 int i; 1338 1339 scm->fp = UNIXCB(skb).fp; 1340 UNIXCB(skb).fp = NULL; 1341 1342 for (i = scm->fp->count-1; i >= 0; i--) 1343 unix_notinflight(scm->fp->fp[i]); 1344} 1345 1346static void unix_destruct_scm(struct sk_buff *skb) 1347{ 1348 struct scm_cookie scm; 1349 memset(&scm, 0, sizeof(scm)); 1350 scm.pid = UNIXCB(skb).pid; 1351 scm.cred = UNIXCB(skb).cred; 1352 if (UNIXCB(skb).fp) 1353 unix_detach_fds(&scm, skb); 1354 1355 /* Alas, it calls VFS */ 1356 /* So fscking what? fput() had been SMP-safe since the last Summer */ 1357 scm_destroy(&scm); 1358 sock_wfree(skb); 1359} 1360 1361#define MAX_RECURSION_LEVEL 4 1362 1363static int unix_attach_fds(struct scm_cookie *scm, struct sk_buff *skb) 1364{ 1365 int i; 1366 unsigned char max_level = 0; 1367 int unix_sock_count = 0; 1368 1369 for (i = scm->fp->count - 1; i >= 0; i--) { 1370 struct sock *sk = unix_get_socket(scm->fp->fp[i]); 1371 1372 if (sk) { 1373 unix_sock_count++; 1374 max_level = max(max_level, 1375 unix_sk(sk)->recursion_level); 1376 } 1377 } 1378 if (unlikely(max_level > MAX_RECURSION_LEVEL)) 1379 return -ETOOMANYREFS; 1380 1381 /* 1382 * Need to duplicate file references for the sake of garbage 1383 * collection. Otherwise a socket in the fps might become a 1384 * candidate for GC while the skb is not yet queued. 1385 */ 1386 UNIXCB(skb).fp = scm_fp_dup(scm->fp); 1387 if (!UNIXCB(skb).fp) 1388 return -ENOMEM; 1389 1390 if (unix_sock_count) { 1391 for (i = scm->fp->count - 1; i >= 0; i--) 1392 unix_inflight(scm->fp->fp[i]); 1393 } 1394 return max_level; 1395} 1396 1397static int unix_scm_to_skb(struct scm_cookie *scm, struct sk_buff *skb, bool send_fds) 1398{ 1399 int err = 0; 1400 1401 UNIXCB(skb).pid = get_pid(scm->pid); 1402 if (scm->cred) 1403 UNIXCB(skb).cred = get_cred(scm->cred); 1404 UNIXCB(skb).fp = NULL; 1405 if (scm->fp && send_fds) 1406 err = unix_attach_fds(scm, skb); 1407 1408 skb->destructor = unix_destruct_scm; 1409 return err; 1410} 1411 1412/* 1413 * Some apps rely on write() giving SCM_CREDENTIALS 1414 * We include credentials if source or destination socket 1415 * asserted SOCK_PASSCRED. 1416 */ 1417static void maybe_add_creds(struct sk_buff *skb, const struct socket *sock, 1418 const struct sock *other) 1419{ 1420 if (UNIXCB(skb).cred) 1421 return; 1422 if (test_bit(SOCK_PASSCRED, &sock->flags) || 1423 !other->sk_socket || 1424 test_bit(SOCK_PASSCRED, &other->sk_socket->flags)) { 1425 UNIXCB(skb).pid = get_pid(task_tgid(current)); 1426 UNIXCB(skb).cred = get_current_cred(); 1427 } 1428} 1429 1430/* 1431 * Send AF_UNIX data. 1432 */ 1433 1434static int unix_dgram_sendmsg(struct kiocb *kiocb, struct socket *sock, 1435 struct msghdr *msg, size_t len) 1436{ 1437 struct sock_iocb *siocb = kiocb_to_siocb(kiocb); 1438 struct sock *sk = sock->sk; 1439 struct net *net = sock_net(sk); 1440 struct unix_sock *u = unix_sk(sk); 1441 struct sockaddr_un *sunaddr = msg->msg_name; 1442 struct sock *other = NULL; 1443 int namelen = 0; /* fake GCC */ 1444 int err; 1445 unsigned hash; 1446 struct sk_buff *skb; 1447 long timeo; 1448 struct scm_cookie tmp_scm; 1449 int max_level; 1450 1451 if (NULL == siocb->scm) 1452 siocb->scm = &tmp_scm; 1453 wait_for_unix_gc(); 1454 err = scm_send(sock, msg, siocb->scm); 1455 if (err < 0) 1456 return err; 1457 1458 err = -EOPNOTSUPP; 1459 if (msg->msg_flags&MSG_OOB) 1460 goto out; 1461 1462 if (msg->msg_namelen) { 1463 err = unix_mkname(sunaddr, msg->msg_namelen, &hash); 1464 if (err < 0) 1465 goto out; 1466 namelen = err; 1467 } else { 1468 sunaddr = NULL; 1469 err = -ENOTCONN; 1470 other = unix_peer_get(sk); 1471 if (!other) 1472 goto out; 1473 } 1474 1475 if (test_bit(SOCK_PASSCRED, &sock->flags) && !u->addr 1476 && (err = unix_autobind(sock)) != 0) 1477 goto out; 1478 1479 err = -EMSGSIZE; 1480 if (len > sk->sk_sndbuf - 32) 1481 goto out; 1482 1483 skb = sock_alloc_send_skb(sk, len, msg->msg_flags&MSG_DONTWAIT, &err); 1484 if (skb == NULL) 1485 goto out; 1486 1487 err = unix_scm_to_skb(siocb->scm, skb, true); 1488 if (err < 0) 1489 goto out_free; 1490 max_level = err + 1; 1491 unix_get_secdata(siocb->scm, skb); 1492 1493 skb_reset_transport_header(skb); 1494 err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len); 1495 if (err) 1496 goto out_free; 1497 1498 timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT); 1499 1500restart: 1501 if (!other) { 1502 err = -ECONNRESET; 1503 if (sunaddr == NULL) 1504 goto out_free; 1505 1506 other = unix_find_other(net, sunaddr, namelen, sk->sk_type, 1507 hash, &err); 1508 if (other == NULL) 1509 goto out_free; 1510 } 1511 1512 if (sk_filter(other, skb) < 0) { 1513 /* Toss the packet but do not return any error to the sender */ 1514 err = len; 1515 goto out_free; 1516 } 1517 1518 unix_state_lock(other); 1519 err = -EPERM; 1520 if (!unix_may_send(sk, other)) 1521 goto out_unlock; 1522 1523 if (sock_flag(other, SOCK_DEAD)) { 1524 /* 1525 * Check with 1003.1g - what should 1526 * datagram error 1527 */ 1528 unix_state_unlock(other); 1529 sock_put(other); 1530 1531 err = 0; 1532 unix_state_lock(sk); 1533 if (unix_peer(sk) == other) { 1534 unix_peer(sk) = NULL; 1535 unix_state_unlock(sk); 1536 1537 unix_dgram_disconnected(sk, other); 1538 sock_put(other); 1539 err = -ECONNREFUSED; 1540 } else { 1541 unix_state_unlock(sk); 1542 } 1543 1544 other = NULL; 1545 if (err) 1546 goto out_free; 1547 goto restart; 1548 } 1549 1550 err = -EPIPE; 1551 if (other->sk_shutdown & RCV_SHUTDOWN) 1552 goto out_unlock; 1553 1554 if (sk->sk_type != SOCK_SEQPACKET) { 1555 err = security_unix_may_send(sk->sk_socket, other->sk_socket); 1556 if (err) 1557 goto out_unlock; 1558 } 1559 1560 if (unix_peer(other) != sk && unix_recvq_full(other)) { 1561 if (!timeo) { 1562 err = -EAGAIN; 1563 goto out_unlock; 1564 } 1565 1566 timeo = unix_wait_for_peer(other, timeo); 1567 1568 err = sock_intr_errno(timeo); 1569 if (signal_pending(current)) 1570 goto out_free; 1571 1572 goto restart; 1573 } 1574 1575 if (sock_flag(other, SOCK_RCVTSTAMP)) 1576 __net_timestamp(skb); 1577 maybe_add_creds(skb, sock, other); 1578 skb_queue_tail(&other->sk_receive_queue, skb); 1579 if (max_level > unix_sk(other)->recursion_level) 1580 unix_sk(other)->recursion_level = max_level; 1581 unix_state_unlock(other); 1582 other->sk_data_ready(other, len); 1583 sock_put(other); 1584 scm_destroy(siocb->scm); 1585 return len; 1586 1587out_unlock: 1588 unix_state_unlock(other); 1589out_free: 1590 kfree_skb(skb); 1591out: 1592 if (other) 1593 sock_put(other); 1594 scm_destroy(siocb->scm); 1595 return err; 1596} 1597 1598 1599static int unix_stream_sendmsg(struct kiocb *kiocb, struct socket *sock, 1600 struct msghdr *msg, size_t len) 1601{ 1602 struct sock_iocb *siocb = kiocb_to_siocb(kiocb); 1603 struct sock *sk = sock->sk; 1604 struct sock *other = NULL; 1605 int err, size; 1606 struct sk_buff *skb; 1607 int sent = 0; 1608 struct scm_cookie tmp_scm; 1609 bool fds_sent = false; 1610 int max_level; 1611 1612 if (NULL == siocb->scm) 1613 siocb->scm = &tmp_scm; 1614 wait_for_unix_gc(); 1615 err = scm_send(sock, msg, siocb->scm); 1616 if (err < 0) 1617 return err; 1618 1619 err = -EOPNOTSUPP; 1620 if (msg->msg_flags&MSG_OOB) 1621 goto out_err; 1622 1623 if (msg->msg_namelen) { 1624 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP; 1625 goto out_err; 1626 } else { 1627 err = -ENOTCONN; 1628 other = unix_peer(sk); 1629 if (!other) 1630 goto out_err; 1631 } 1632 1633 if (sk->sk_shutdown & SEND_SHUTDOWN) 1634 goto pipe_err; 1635 1636 while (sent < len) { 1637 /* 1638 * Optimisation for the fact that under 0.01% of X 1639 * messages typically need breaking up. 1640 */ 1641 1642 size = len-sent; 1643 1644 /* Keep two messages in the pipe so it schedules better */ 1645 if (size > ((sk->sk_sndbuf >> 1) - 64)) 1646 size = (sk->sk_sndbuf >> 1) - 64; 1647 1648 if (size > SKB_MAX_ALLOC) 1649 size = SKB_MAX_ALLOC; 1650 1651 /* 1652 * Grab a buffer 1653 */ 1654 1655 skb = sock_alloc_send_skb(sk, size, msg->msg_flags&MSG_DONTWAIT, 1656 &err); 1657 1658 if (skb == NULL) 1659 goto out_err; 1660 1661 /* 1662 * If you pass two values to the sock_alloc_send_skb 1663 * it tries to grab the large buffer with GFP_NOFS 1664 * (which can fail easily), and if it fails grab the 1665 * fallback size buffer which is under a page and will 1666 * succeed. [Alan] 1667 */ 1668 size = min_t(int, size, skb_tailroom(skb)); 1669 1670 1671 /* Only send the fds in the first buffer */ 1672 err = unix_scm_to_skb(siocb->scm, skb, !fds_sent); 1673 if (err < 0) { 1674 kfree_skb(skb); 1675 goto out_err; 1676 } 1677 max_level = err + 1; 1678 fds_sent = true; 1679 1680 err = memcpy_fromiovec(skb_put(skb, size), msg->msg_iov, size); 1681 if (err) { 1682 kfree_skb(skb); 1683 goto out_err; 1684 } 1685 1686 unix_state_lock(other); 1687 1688 if (sock_flag(other, SOCK_DEAD) || 1689 (other->sk_shutdown & RCV_SHUTDOWN)) 1690 goto pipe_err_free; 1691 1692 maybe_add_creds(skb, sock, other); 1693 skb_queue_tail(&other->sk_receive_queue, skb); 1694 if (max_level > unix_sk(other)->recursion_level) 1695 unix_sk(other)->recursion_level = max_level; 1696 unix_state_unlock(other); 1697 other->sk_data_ready(other, size); 1698 sent += size; 1699 } 1700 1701 scm_destroy(siocb->scm); 1702 siocb->scm = NULL; 1703 1704 return sent; 1705 1706pipe_err_free: 1707 unix_state_unlock(other); 1708 kfree_skb(skb); 1709pipe_err: 1710 if (sent == 0 && !(msg->msg_flags&MSG_NOSIGNAL)) 1711 send_sig(SIGPIPE, current, 0); 1712 err = -EPIPE; 1713out_err: 1714 scm_destroy(siocb->scm); 1715 siocb->scm = NULL; 1716 return sent ? : err; 1717} 1718 1719static int unix_seqpacket_sendmsg(struct kiocb *kiocb, struct socket *sock, 1720 struct msghdr *msg, size_t len) 1721{ 1722 int err; 1723 struct sock *sk = sock->sk; 1724 1725 err = sock_error(sk); 1726 if (err) 1727 return err; 1728 1729 if (sk->sk_state != TCP_ESTABLISHED) 1730 return -ENOTCONN; 1731 1732 if (msg->msg_namelen) 1733 msg->msg_namelen = 0; 1734 1735 return unix_dgram_sendmsg(kiocb, sock, msg, len); 1736} 1737 1738static int unix_seqpacket_recvmsg(struct kiocb *iocb, struct socket *sock, 1739 struct msghdr *msg, size_t size, 1740 int flags) 1741{ 1742 struct sock *sk = sock->sk; 1743 1744 if (sk->sk_state != TCP_ESTABLISHED) 1745 return -ENOTCONN; 1746 1747 return unix_dgram_recvmsg(iocb, sock, msg, size, flags); 1748} 1749 1750static void unix_copy_addr(struct msghdr *msg, struct sock *sk) 1751{ 1752 struct unix_sock *u = unix_sk(sk); 1753 1754 msg->msg_namelen = 0; 1755 if (u->addr) { 1756 msg->msg_namelen = u->addr->len; 1757 memcpy(msg->msg_name, u->addr->name, u->addr->len); 1758 } 1759} 1760 1761static int unix_dgram_recvmsg(struct kiocb *iocb, struct socket *sock, 1762 struct msghdr *msg, size_t size, 1763 int flags) 1764{ 1765 struct sock_iocb *siocb = kiocb_to_siocb(iocb); 1766 struct scm_cookie tmp_scm; 1767 struct sock *sk = sock->sk; 1768 struct unix_sock *u = unix_sk(sk); 1769 int noblock = flags & MSG_DONTWAIT; 1770 struct sk_buff *skb; 1771 int err; 1772 int peeked, skip; 1773 1774 err = -EOPNOTSUPP; 1775 if (flags&MSG_OOB) 1776 goto out; 1777 1778 msg->msg_namelen = 0; 1779 1780 err = mutex_lock_interruptible(&u->readlock); 1781 if (err) { 1782 err = sock_intr_errno(sock_rcvtimeo(sk, noblock)); 1783 goto out; 1784 } 1785 1786 skip = sk_peek_offset(sk, flags); 1787 1788 skb = __skb_recv_datagram(sk, flags, &peeked, &skip, &err); 1789 if (!skb) { 1790 unix_state_lock(sk); 1791 /* Signal EOF on disconnected non-blocking SEQPACKET socket. */ 1792 if (sk->sk_type == SOCK_SEQPACKET && err == -EAGAIN && 1793 (sk->sk_shutdown & RCV_SHUTDOWN)) 1794 err = 0; 1795 unix_state_unlock(sk); 1796 goto out_unlock; 1797 } 1798 1799 wake_up_interruptible_sync_poll(&u->peer_wait, 1800 POLLOUT | POLLWRNORM | POLLWRBAND); 1801 1802 if (msg->msg_name) 1803 unix_copy_addr(msg, skb->sk); 1804 1805 if (size > skb->len - skip) 1806 size = skb->len - skip; 1807 else if (size < skb->len - skip) 1808 msg->msg_flags |= MSG_TRUNC; 1809 1810 err = skb_copy_datagram_iovec(skb, skip, msg->msg_iov, size); 1811 if (err) 1812 goto out_free; 1813 1814 if (sock_flag(sk, SOCK_RCVTSTAMP)) 1815 __sock_recv_timestamp(msg, sk, skb); 1816 1817 if (!siocb->scm) { 1818 siocb->scm = &tmp_scm; 1819 memset(&tmp_scm, 0, sizeof(tmp_scm)); 1820 } 1821 scm_set_cred(siocb->scm, UNIXCB(skb).pid, UNIXCB(skb).cred); 1822 unix_set_secdata(siocb->scm, skb); 1823 1824 if (!(flags & MSG_PEEK)) { 1825 if (UNIXCB(skb).fp) 1826 unix_detach_fds(siocb->scm, skb); 1827 1828 sk_peek_offset_bwd(sk, skb->len); 1829 } else { 1830 /* It is questionable: on PEEK we could: 1831 - do not return fds - good, but too simple 8) 1832 - return fds, and do not return them on read (old strategy, 1833 apparently wrong) 1834 - clone fds (I chose it for now, it is the most universal 1835 solution) 1836 1837 POSIX 1003.1g does not actually define this clearly 1838 at all. POSIX 1003.1g doesn't define a lot of things 1839 clearly however! 1840 1841 */ 1842 1843 sk_peek_offset_fwd(sk, size); 1844 1845 if (UNIXCB(skb).fp) 1846 siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp); 1847 } 1848 err = size; 1849 1850 scm_recv(sock, msg, siocb->scm, flags); 1851 1852out_free: 1853 skb_free_datagram(sk, skb); 1854out_unlock: 1855 mutex_unlock(&u->readlock); 1856out: 1857 return err; 1858} 1859 1860/* 1861 * Sleep until data has arrive. But check for races.. 1862 */ 1863 1864static long unix_stream_data_wait(struct sock *sk, long timeo) 1865{ 1866 DEFINE_WAIT(wait); 1867 1868 unix_state_lock(sk); 1869 1870 for (;;) { 1871 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 1872 1873 if (!skb_queue_empty(&sk->sk_receive_queue) || 1874 sk->sk_err || 1875 (sk->sk_shutdown & RCV_SHUTDOWN) || 1876 signal_pending(current) || 1877 !timeo) 1878 break; 1879 1880 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags); 1881 unix_state_unlock(sk); 1882 timeo = schedule_timeout(timeo); 1883 unix_state_lock(sk); 1884 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags); 1885 } 1886 1887 finish_wait(sk_sleep(sk), &wait); 1888 unix_state_unlock(sk); 1889 return timeo; 1890} 1891 1892 1893 1894static int unix_stream_recvmsg(struct kiocb *iocb, struct socket *sock, 1895 struct msghdr *msg, size_t size, 1896 int flags) 1897{ 1898 struct sock_iocb *siocb = kiocb_to_siocb(iocb); 1899 struct scm_cookie tmp_scm; 1900 struct sock *sk = sock->sk; 1901 struct unix_sock *u = unix_sk(sk); 1902 struct sockaddr_un *sunaddr = msg->msg_name; 1903 int copied = 0; 1904 int check_creds = 0; 1905 int target; 1906 int err = 0; 1907 long timeo; 1908 int skip; 1909 1910 err = -EINVAL; 1911 if (sk->sk_state != TCP_ESTABLISHED) 1912 goto out; 1913 1914 err = -EOPNOTSUPP; 1915 if (flags&MSG_OOB) 1916 goto out; 1917 1918 target = sock_rcvlowat(sk, flags&MSG_WAITALL, size); 1919 timeo = sock_rcvtimeo(sk, flags&MSG_DONTWAIT); 1920 1921 msg->msg_namelen = 0; 1922 1923 /* Lock the socket to prevent queue disordering 1924 * while sleeps in memcpy_tomsg 1925 */ 1926 1927 if (!siocb->scm) { 1928 siocb->scm = &tmp_scm; 1929 memset(&tmp_scm, 0, sizeof(tmp_scm)); 1930 } 1931 1932 err = mutex_lock_interruptible(&u->readlock); 1933 if (err) { 1934 err = sock_intr_errno(timeo); 1935 goto out; 1936 } 1937 1938 skip = sk_peek_offset(sk, flags); 1939 1940 do { 1941 int chunk; 1942 struct sk_buff *skb; 1943 1944 unix_state_lock(sk); 1945 skb = skb_peek(&sk->sk_receive_queue); 1946again: 1947 if (skb == NULL) { 1948 unix_sk(sk)->recursion_level = 0; 1949 if (copied >= target) 1950 goto unlock; 1951 1952 /* 1953 * POSIX 1003.1g mandates this order. 1954 */ 1955 1956 err = sock_error(sk); 1957 if (err) 1958 goto unlock; 1959 if (sk->sk_shutdown & RCV_SHUTDOWN) 1960 goto unlock; 1961 1962 unix_state_unlock(sk); 1963 err = -EAGAIN; 1964 if (!timeo) 1965 break; 1966 mutex_unlock(&u->readlock); 1967 1968 timeo = unix_stream_data_wait(sk, timeo); 1969 1970 if (signal_pending(current) 1971 || mutex_lock_interruptible(&u->readlock)) { 1972 err = sock_intr_errno(timeo); 1973 goto out; 1974 } 1975 1976 continue; 1977 unlock: 1978 unix_state_unlock(sk); 1979 break; 1980 } 1981 1982 if (skip >= skb->len) { 1983 skip -= skb->len; 1984 skb = skb_peek_next(skb, &sk->sk_receive_queue); 1985 goto again; 1986 } 1987 1988 unix_state_unlock(sk); 1989 1990 if (check_creds) { 1991 /* Never glue messages from different writers */ 1992 if ((UNIXCB(skb).pid != siocb->scm->pid) || 1993 (UNIXCB(skb).cred != siocb->scm->cred)) 1994 break; 1995 } else { 1996 /* Copy credentials */ 1997 scm_set_cred(siocb->scm, UNIXCB(skb).pid, UNIXCB(skb).cred); 1998 check_creds = 1; 1999 } 2000 2001 /* Copy address just once */ 2002 if (sunaddr) { 2003 unix_copy_addr(msg, skb->sk); 2004 sunaddr = NULL; 2005 } 2006 2007 chunk = min_t(unsigned int, skb->len - skip, size); 2008 if (memcpy_toiovec(msg->msg_iov, skb->data + skip, chunk)) { 2009 if (copied == 0) 2010 copied = -EFAULT; 2011 break; 2012 } 2013 copied += chunk; 2014 size -= chunk; 2015 2016 /* Mark read part of skb as used */ 2017 if (!(flags & MSG_PEEK)) { 2018 skb_pull(skb, chunk); 2019 2020 sk_peek_offset_bwd(sk, chunk); 2021 2022 if (UNIXCB(skb).fp) 2023 unix_detach_fds(siocb->scm, skb); 2024 2025 if (skb->len) 2026 break; 2027 2028 skb_unlink(skb, &sk->sk_receive_queue); 2029 consume_skb(skb); 2030 2031 if (siocb->scm->fp) 2032 break; 2033 } else { 2034 /* It is questionable, see note in unix_dgram_recvmsg. 2035 */ 2036 if (UNIXCB(skb).fp) 2037 siocb->scm->fp = scm_fp_dup(UNIXCB(skb).fp); 2038 2039 sk_peek_offset_fwd(sk, chunk); 2040 2041 break; 2042 } 2043 } while (size); 2044 2045 mutex_unlock(&u->readlock); 2046 scm_recv(sock, msg, siocb->scm, flags); 2047out: 2048 return copied ? : err; 2049} 2050 2051static int unix_shutdown(struct socket *sock, int mode) 2052{ 2053 struct sock *sk = sock->sk; 2054 struct sock *other; 2055 2056 mode = (mode+1)&(RCV_SHUTDOWN|SEND_SHUTDOWN); 2057 2058 if (!mode) 2059 return 0; 2060 2061 unix_state_lock(sk); 2062 sk->sk_shutdown |= mode; 2063 other = unix_peer(sk); 2064 if (other) 2065 sock_hold(other); 2066 unix_state_unlock(sk); 2067 sk->sk_state_change(sk); 2068 2069 if (other && 2070 (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET)) { 2071 2072 int peer_mode = 0; 2073 2074 if (mode&RCV_SHUTDOWN) 2075 peer_mode |= SEND_SHUTDOWN; 2076 if (mode&SEND_SHUTDOWN) 2077 peer_mode |= RCV_SHUTDOWN; 2078 unix_state_lock(other); 2079 other->sk_shutdown |= peer_mode; 2080 unix_state_unlock(other); 2081 other->sk_state_change(other); 2082 if (peer_mode == SHUTDOWN_MASK) 2083 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_HUP); 2084 else if (peer_mode & RCV_SHUTDOWN) 2085 sk_wake_async(other, SOCK_WAKE_WAITD, POLL_IN); 2086 } 2087 if (other) 2088 sock_put(other); 2089 2090 return 0; 2091} 2092 2093long unix_inq_len(struct sock *sk) 2094{ 2095 struct sk_buff *skb; 2096 long amount = 0; 2097 2098 if (sk->sk_state == TCP_LISTEN) 2099 return -EINVAL; 2100 2101 spin_lock(&sk->sk_receive_queue.lock); 2102 if (sk->sk_type == SOCK_STREAM || 2103 sk->sk_type == SOCK_SEQPACKET) { 2104 skb_queue_walk(&sk->sk_receive_queue, skb) 2105 amount += skb->len; 2106 } else { 2107 skb = skb_peek(&sk->sk_receive_queue); 2108 if (skb) 2109 amount = skb->len; 2110 } 2111 spin_unlock(&sk->sk_receive_queue.lock); 2112 2113 return amount; 2114} 2115EXPORT_SYMBOL_GPL(unix_inq_len); 2116 2117long unix_outq_len(struct sock *sk) 2118{ 2119 return sk_wmem_alloc_get(sk); 2120} 2121EXPORT_SYMBOL_GPL(unix_outq_len); 2122 2123static int unix_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 2124{ 2125 struct sock *sk = sock->sk; 2126 long amount = 0; 2127 int err; 2128 2129 switch (cmd) { 2130 case SIOCOUTQ: 2131 amount = unix_outq_len(sk); 2132 err = put_user(amount, (int __user *)arg); 2133 break; 2134 case SIOCINQ: 2135 amount = unix_inq_len(sk); 2136 if (amount < 0) 2137 err = amount; 2138 else 2139 err = put_user(amount, (int __user *)arg); 2140 break; 2141 default: 2142 err = -ENOIOCTLCMD; 2143 break; 2144 } 2145 return err; 2146} 2147 2148static unsigned int unix_poll(struct file *file, struct socket *sock, poll_table *wait) 2149{ 2150 struct sock *sk = sock->sk; 2151 unsigned int mask; 2152 2153 sock_poll_wait(file, sk_sleep(sk), wait); 2154 mask = 0; 2155 2156 /* exceptional events? */ 2157 if (sk->sk_err) 2158 mask |= POLLERR; 2159 if (sk->sk_shutdown == SHUTDOWN_MASK) 2160 mask |= POLLHUP; 2161 if (sk->sk_shutdown & RCV_SHUTDOWN) 2162 mask |= POLLRDHUP | POLLIN | POLLRDNORM; 2163 2164 /* readable? */ 2165 if (!skb_queue_empty(&sk->sk_receive_queue)) 2166 mask |= POLLIN | POLLRDNORM; 2167 2168 /* Connection-based need to check for termination and startup */ 2169 if ((sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) && 2170 sk->sk_state == TCP_CLOSE) 2171 mask |= POLLHUP; 2172 2173 /* 2174 * we set writable also when the other side has shut down the 2175 * connection. This prevents stuck sockets. 2176 */ 2177 if (unix_writable(sk)) 2178 mask |= POLLOUT | POLLWRNORM | POLLWRBAND; 2179 2180 return mask; 2181} 2182 2183static unsigned int unix_dgram_poll(struct file *file, struct socket *sock, 2184 poll_table *wait) 2185{ 2186 struct sock *sk = sock->sk, *other; 2187 unsigned int mask, writable; 2188 2189 sock_poll_wait(file, sk_sleep(sk), wait); 2190 mask = 0; 2191 2192 /* exceptional events? */ 2193 if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue)) 2194 mask |= POLLERR; 2195 if (sk->sk_shutdown & RCV_SHUTDOWN) 2196 mask |= POLLRDHUP | POLLIN | POLLRDNORM; 2197 if (sk->sk_shutdown == SHUTDOWN_MASK) 2198 mask |= POLLHUP; 2199 2200 /* readable? */ 2201 if (!skb_queue_empty(&sk->sk_receive_queue)) 2202 mask |= POLLIN | POLLRDNORM; 2203 2204 /* Connection-based need to check for termination and startup */ 2205 if (sk->sk_type == SOCK_SEQPACKET) { 2206 if (sk->sk_state == TCP_CLOSE) 2207 mask |= POLLHUP; 2208 /* connection hasn't started yet? */ 2209 if (sk->sk_state == TCP_SYN_SENT) 2210 return mask; 2211 } 2212 2213 /* No write status requested, avoid expensive OUT tests. */ 2214 if (wait && !(wait->key & (POLLWRBAND | POLLWRNORM | POLLOUT))) 2215 return mask; 2216 2217 writable = unix_writable(sk); 2218 other = unix_peer_get(sk); 2219 if (other) { 2220 if (unix_peer(other) != sk) { 2221 sock_poll_wait(file, &unix_sk(other)->peer_wait, wait); 2222 if (unix_recvq_full(other)) 2223 writable = 0; 2224 } 2225 sock_put(other); 2226 } 2227 2228 if (writable) 2229 mask |= POLLOUT | POLLWRNORM | POLLWRBAND; 2230 else 2231 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags); 2232 2233 return mask; 2234} 2235 2236#ifdef CONFIG_PROC_FS 2237static struct sock *first_unix_socket(int *i) 2238{ 2239 for (*i = 0; *i <= UNIX_HASH_SIZE; (*i)++) { 2240 if (!hlist_empty(&unix_socket_table[*i])) 2241 return __sk_head(&unix_socket_table[*i]); 2242 } 2243 return NULL; 2244} 2245 2246static struct sock *next_unix_socket(int *i, struct sock *s) 2247{ 2248 struct sock *next = sk_next(s); 2249 /* More in this chain? */ 2250 if (next) 2251 return next; 2252 /* Look for next non-empty chain. */ 2253 for ((*i)++; *i <= UNIX_HASH_SIZE; (*i)++) { 2254 if (!hlist_empty(&unix_socket_table[*i])) 2255 return __sk_head(&unix_socket_table[*i]); 2256 } 2257 return NULL; 2258} 2259 2260struct unix_iter_state { 2261 struct seq_net_private p; 2262 int i; 2263}; 2264 2265static struct sock *unix_seq_idx(struct seq_file *seq, loff_t pos) 2266{ 2267 struct unix_iter_state *iter = seq->private; 2268 loff_t off = 0; 2269 struct sock *s; 2270 2271 for (s = first_unix_socket(&iter->i); s; s = next_unix_socket(&iter->i, s)) { 2272 if (sock_net(s) != seq_file_net(seq)) 2273 continue; 2274 if (off == pos) 2275 return s; 2276 ++off; 2277 } 2278 return NULL; 2279} 2280 2281static void *unix_seq_start(struct seq_file *seq, loff_t *pos) 2282 __acquires(unix_table_lock) 2283{ 2284 spin_lock(&unix_table_lock); 2285 return *pos ? unix_seq_idx(seq, *pos - 1) : SEQ_START_TOKEN; 2286} 2287 2288static void *unix_seq_next(struct seq_file *seq, void *v, loff_t *pos) 2289{ 2290 struct unix_iter_state *iter = seq->private; 2291 struct sock *sk = v; 2292 ++*pos; 2293 2294 if (v == SEQ_START_TOKEN) 2295 sk = first_unix_socket(&iter->i); 2296 else 2297 sk = next_unix_socket(&iter->i, sk); 2298 while (sk && (sock_net(sk) != seq_file_net(seq))) 2299 sk = next_unix_socket(&iter->i, sk); 2300 return sk; 2301} 2302 2303static void unix_seq_stop(struct seq_file *seq, void *v) 2304 __releases(unix_table_lock) 2305{ 2306 spin_unlock(&unix_table_lock); 2307} 2308 2309static int unix_seq_show(struct seq_file *seq, void *v) 2310{ 2311 2312 if (v == SEQ_START_TOKEN) 2313 seq_puts(seq, "Num RefCount Protocol Flags Type St " 2314 "Inode Path\n"); 2315 else { 2316 struct sock *s = v; 2317 struct unix_sock *u = unix_sk(s); 2318 unix_state_lock(s); 2319 2320 seq_printf(seq, "%pK: %08X %08X %08X %04X %02X %5lu", 2321 s, 2322 atomic_read(&s->sk_refcnt), 2323 0, 2324 s->sk_state == TCP_LISTEN ? __SO_ACCEPTCON : 0, 2325 s->sk_type, 2326 s->sk_socket ? 2327 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTED : SS_UNCONNECTED) : 2328 (s->sk_state == TCP_ESTABLISHED ? SS_CONNECTING : SS_DISCONNECTING), 2329 sock_i_ino(s)); 2330 2331 if (u->addr) { 2332 int i, len; 2333 seq_putc(seq, ' '); 2334 2335 i = 0; 2336 len = u->addr->len - sizeof(short); 2337 if (!UNIX_ABSTRACT(s)) 2338 len--; 2339 else { 2340 seq_putc(seq, '@'); 2341 i++; 2342 } 2343 for ( ; i < len; i++) 2344 seq_putc(seq, u->addr->name->sun_path[i]); 2345 } 2346 unix_state_unlock(s); 2347 seq_putc(seq, '\n'); 2348 } 2349 2350 return 0; 2351} 2352 2353static const struct seq_operations unix_seq_ops = { 2354 .start = unix_seq_start, 2355 .next = unix_seq_next, 2356 .stop = unix_seq_stop, 2357 .show = unix_seq_show, 2358}; 2359 2360static int unix_seq_open(struct inode *inode, struct file *file) 2361{ 2362 return seq_open_net(inode, file, &unix_seq_ops, 2363 sizeof(struct unix_iter_state)); 2364} 2365 2366static const struct file_operations unix_seq_fops = { 2367 .owner = THIS_MODULE, 2368 .open = unix_seq_open, 2369 .read = seq_read, 2370 .llseek = seq_lseek, 2371 .release = seq_release_net, 2372}; 2373 2374#endif 2375 2376static const struct net_proto_family unix_family_ops = { 2377 .family = PF_UNIX, 2378 .create = unix_create, 2379 .owner = THIS_MODULE, 2380}; 2381 2382 2383static int __net_init unix_net_init(struct net *net) 2384{ 2385 int error = -ENOMEM; 2386 2387 net->unx.sysctl_max_dgram_qlen = 10; 2388 if (unix_sysctl_register(net)) 2389 goto out; 2390 2391#ifdef CONFIG_PROC_FS 2392 if (!proc_net_fops_create(net, "unix", 0, &unix_seq_fops)) { 2393 unix_sysctl_unregister(net); 2394 goto out; 2395 } 2396#endif 2397 error = 0; 2398out: 2399 return error; 2400} 2401 2402static void __net_exit unix_net_exit(struct net *net) 2403{ 2404 unix_sysctl_unregister(net); 2405 proc_net_remove(net, "unix"); 2406} 2407 2408static struct pernet_operations unix_net_ops = { 2409 .init = unix_net_init, 2410 .exit = unix_net_exit, 2411}; 2412 2413static int __init af_unix_init(void) 2414{ 2415 int rc = -1; 2416 struct sk_buff *dummy_skb; 2417 2418 BUILD_BUG_ON(sizeof(struct unix_skb_parms) > sizeof(dummy_skb->cb)); 2419 2420 rc = proto_register(&unix_proto, 1); 2421 if (rc != 0) { 2422 printk(KERN_CRIT "%s: Cannot create unix_sock SLAB cache!\n", 2423 __func__); 2424 goto out; 2425 } 2426 2427 sock_register(&unix_family_ops); 2428 register_pernet_subsys(&unix_net_ops); 2429out: 2430 return rc; 2431} 2432 2433static void __exit af_unix_exit(void) 2434{ 2435 sock_unregister(PF_UNIX); 2436 proto_unregister(&unix_proto); 2437 unregister_pernet_subsys(&unix_net_ops); 2438} 2439 2440/* Earlier than device_initcall() so that other drivers invoking 2441 request_module() don't end up in a loop when modprobe tries 2442 to use a UNIX socket. But later than subsys_initcall() because 2443 we depend on stuff initialised there */ 2444fs_initcall(af_unix_init); 2445module_exit(af_unix_exit); 2446 2447MODULE_LICENSE("GPL"); 2448MODULE_ALIAS_NETPROTO(PF_UNIX); 2449