sock.c revision 58a5a7b9555ea231b557ebef5cabeaf8e951df0b
1/* 2 * INET An implementation of the TCP/IP protocol suite for the LINUX 3 * operating system. INET is implemented using the BSD Socket 4 * interface as the means of communication with the user level. 5 * 6 * Generic socket support routines. Memory allocators, socket lock/release 7 * handler for protocols to use and generic option handler. 8 * 9 * 10 * Version: $Id: sock.c,v 1.117 2002/02/01 22:01:03 davem Exp $ 11 * 12 * Authors: Ross Biro 13 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 14 * Florian La Roche, <flla@stud.uni-sb.de> 15 * Alan Cox, <A.Cox@swansea.ac.uk> 16 * 17 * Fixes: 18 * Alan Cox : Numerous verify_area() problems 19 * Alan Cox : Connecting on a connecting socket 20 * now returns an error for tcp. 21 * Alan Cox : sock->protocol is set correctly. 22 * and is not sometimes left as 0. 23 * Alan Cox : connect handles icmp errors on a 24 * connect properly. Unfortunately there 25 * is a restart syscall nasty there. I 26 * can't match BSD without hacking the C 27 * library. Ideas urgently sought! 28 * Alan Cox : Disallow bind() to addresses that are 29 * not ours - especially broadcast ones!! 30 * Alan Cox : Socket 1024 _IS_ ok for users. (fencepost) 31 * Alan Cox : sock_wfree/sock_rfree don't destroy sockets, 32 * instead they leave that for the DESTROY timer. 33 * Alan Cox : Clean up error flag in accept 34 * Alan Cox : TCP ack handling is buggy, the DESTROY timer 35 * was buggy. Put a remove_sock() in the handler 36 * for memory when we hit 0. Also altered the timer 37 * code. The ACK stuff can wait and needs major 38 * TCP layer surgery. 39 * Alan Cox : Fixed TCP ack bug, removed remove sock 40 * and fixed timer/inet_bh race. 41 * Alan Cox : Added zapped flag for TCP 42 * Alan Cox : Move kfree_skb into skbuff.c and tidied up surplus code 43 * Alan Cox : for new sk_buff allocations wmalloc/rmalloc now call alloc_skb 44 * Alan Cox : kfree_s calls now are kfree_skbmem so we can track skb resources 45 * Alan Cox : Supports socket option broadcast now as does udp. Packet and raw need fixing. 46 * Alan Cox : Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so... 47 * Rick Sladkey : Relaxed UDP rules for matching packets. 48 * C.E.Hawkins : IFF_PROMISC/SIOCGHWADDR support 49 * Pauline Middelink : identd support 50 * Alan Cox : Fixed connect() taking signals I think. 51 * Alan Cox : SO_LINGER supported 52 * Alan Cox : Error reporting fixes 53 * Anonymous : inet_create tidied up (sk->reuse setting) 54 * Alan Cox : inet sockets don't set sk->type! 55 * Alan Cox : Split socket option code 56 * Alan Cox : Callbacks 57 * Alan Cox : Nagle flag for Charles & Johannes stuff 58 * Alex : Removed restriction on inet fioctl 59 * Alan Cox : Splitting INET from NET core 60 * Alan Cox : Fixed bogus SO_TYPE handling in getsockopt() 61 * Adam Caldwell : Missing return in SO_DONTROUTE/SO_DEBUG code 62 * Alan Cox : Split IP from generic code 63 * Alan Cox : New kfree_skbmem() 64 * Alan Cox : Make SO_DEBUG superuser only. 65 * Alan Cox : Allow anyone to clear SO_DEBUG 66 * (compatibility fix) 67 * Alan Cox : Added optimistic memory grabbing for AF_UNIX throughput. 68 * Alan Cox : Allocator for a socket is settable. 69 * Alan Cox : SO_ERROR includes soft errors. 70 * Alan Cox : Allow NULL arguments on some SO_ opts 71 * Alan Cox : Generic socket allocation to make hooks 72 * easier (suggested by Craig Metz). 73 * Michael Pall : SO_ERROR returns positive errno again 74 * Steve Whitehouse: Added default destructor to free 75 * protocol private data. 76 * Steve Whitehouse: Added various other default routines 77 * common to several socket families. 78 * Chris Evans : Call suser() check last on F_SETOWN 79 * Jay Schulist : Added SO_ATTACH_FILTER and SO_DETACH_FILTER. 80 * Andi Kleen : Add sock_kmalloc()/sock_kfree_s() 81 * Andi Kleen : Fix write_space callback 82 * Chris Evans : Security fixes - signedness again 83 * Arnaldo C. Melo : cleanups, use skb_queue_purge 84 * 85 * To Fix: 86 * 87 * 88 * This program is free software; you can redistribute it and/or 89 * modify it under the terms of the GNU General Public License 90 * as published by the Free Software Foundation; either version 91 * 2 of the License, or (at your option) any later version. 92 */ 93 94#include <linux/capability.h> 95#include <linux/errno.h> 96#include <linux/types.h> 97#include <linux/socket.h> 98#include <linux/in.h> 99#include <linux/kernel.h> 100#include <linux/module.h> 101#include <linux/proc_fs.h> 102#include <linux/seq_file.h> 103#include <linux/sched.h> 104#include <linux/timer.h> 105#include <linux/string.h> 106#include <linux/sockios.h> 107#include <linux/net.h> 108#include <linux/mm.h> 109#include <linux/slab.h> 110#include <linux/interrupt.h> 111#include <linux/poll.h> 112#include <linux/tcp.h> 113#include <linux/init.h> 114 115#include <asm/uaccess.h> 116#include <asm/system.h> 117 118#include <linux/netdevice.h> 119#include <net/protocol.h> 120#include <linux/skbuff.h> 121#include <net/request_sock.h> 122#include <net/sock.h> 123#include <net/xfrm.h> 124#include <linux/ipsec.h> 125 126#include <linux/filter.h> 127 128#ifdef CONFIG_INET 129#include <net/tcp.h> 130#endif 131 132/* 133 * Each address family might have different locking rules, so we have 134 * one slock key per address family: 135 */ 136static struct lock_class_key af_family_keys[AF_MAX]; 137static struct lock_class_key af_family_slock_keys[AF_MAX]; 138 139#ifdef CONFIG_DEBUG_LOCK_ALLOC 140/* 141 * Make lock validator output more readable. (we pre-construct these 142 * strings build-time, so that runtime initialization of socket 143 * locks is fast): 144 */ 145static const char *af_family_key_strings[AF_MAX+1] = { 146 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" , 147 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK", 148 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" , 149 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" , 150 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" , 151 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" , 152 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" , 153 "sk_lock-21" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" , 154 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" , 155 "sk_lock-27" , "sk_lock-28" , "sk_lock-29" , 156 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-AF_MAX" 157}; 158static const char *af_family_slock_key_strings[AF_MAX+1] = { 159 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" , 160 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK", 161 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" , 162 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" , 163 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" , 164 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" , 165 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" , 166 "slock-21" , "slock-AF_SNA" , "slock-AF_IRDA" , 167 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" , 168 "slock-27" , "slock-28" , "slock-29" , 169 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_MAX" 170}; 171#endif 172 173/* 174 * sk_callback_lock locking rules are per-address-family, 175 * so split the lock classes by using a per-AF key: 176 */ 177static struct lock_class_key af_callback_keys[AF_MAX]; 178 179/* Take into consideration the size of the struct sk_buff overhead in the 180 * determination of these values, since that is non-constant across 181 * platforms. This makes socket queueing behavior and performance 182 * not depend upon such differences. 183 */ 184#define _SK_MEM_PACKETS 256 185#define _SK_MEM_OVERHEAD (sizeof(struct sk_buff) + 256) 186#define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 187#define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 188 189/* Run time adjustable parameters. */ 190__u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX; 191__u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX; 192__u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX; 193__u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX; 194 195/* Maximal space eaten by iovec or ancilliary data plus some space */ 196int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512); 197 198static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen) 199{ 200 struct timeval tv; 201 202 if (optlen < sizeof(tv)) 203 return -EINVAL; 204 if (copy_from_user(&tv, optval, sizeof(tv))) 205 return -EFAULT; 206 207 *timeo_p = MAX_SCHEDULE_TIMEOUT; 208 if (tv.tv_sec == 0 && tv.tv_usec == 0) 209 return 0; 210 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1)) 211 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ); 212 return 0; 213} 214 215static void sock_warn_obsolete_bsdism(const char *name) 216{ 217 static int warned; 218 static char warncomm[TASK_COMM_LEN]; 219 if (strcmp(warncomm, current->comm) && warned < 5) { 220 strcpy(warncomm, current->comm); 221 printk(KERN_WARNING "process `%s' is using obsolete " 222 "%s SO_BSDCOMPAT\n", warncomm, name); 223 warned++; 224 } 225} 226 227static void sock_disable_timestamp(struct sock *sk) 228{ 229 if (sock_flag(sk, SOCK_TIMESTAMP)) { 230 sock_reset_flag(sk, SOCK_TIMESTAMP); 231 net_disable_timestamp(); 232 } 233} 234 235 236int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 237{ 238 int err = 0; 239 int skb_len; 240 241 /* Cast skb->rcvbuf to unsigned... It's pointless, but reduces 242 number of warnings when compiling with -W --ANK 243 */ 244 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >= 245 (unsigned)sk->sk_rcvbuf) { 246 err = -ENOMEM; 247 goto out; 248 } 249 250 err = sk_filter(sk, skb); 251 if (err) 252 goto out; 253 254 skb->dev = NULL; 255 skb_set_owner_r(skb, sk); 256 257 /* Cache the SKB length before we tack it onto the receive 258 * queue. Once it is added it no longer belongs to us and 259 * may be freed by other threads of control pulling packets 260 * from the queue. 261 */ 262 skb_len = skb->len; 263 264 skb_queue_tail(&sk->sk_receive_queue, skb); 265 266 if (!sock_flag(sk, SOCK_DEAD)) 267 sk->sk_data_ready(sk, skb_len); 268out: 269 return err; 270} 271EXPORT_SYMBOL(sock_queue_rcv_skb); 272 273int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested) 274{ 275 int rc = NET_RX_SUCCESS; 276 277 if (sk_filter(sk, skb)) 278 goto discard_and_relse; 279 280 skb->dev = NULL; 281 282 if (nested) 283 bh_lock_sock_nested(sk); 284 else 285 bh_lock_sock(sk); 286 if (!sock_owned_by_user(sk)) { 287 /* 288 * trylock + unlock semantics: 289 */ 290 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_); 291 292 rc = sk->sk_backlog_rcv(sk, skb); 293 294 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_); 295 } else 296 sk_add_backlog(sk, skb); 297 bh_unlock_sock(sk); 298out: 299 sock_put(sk); 300 return rc; 301discard_and_relse: 302 kfree_skb(skb); 303 goto out; 304} 305EXPORT_SYMBOL(sk_receive_skb); 306 307struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie) 308{ 309 struct dst_entry *dst = sk->sk_dst_cache; 310 311 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) { 312 sk->sk_dst_cache = NULL; 313 dst_release(dst); 314 return NULL; 315 } 316 317 return dst; 318} 319EXPORT_SYMBOL(__sk_dst_check); 320 321struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie) 322{ 323 struct dst_entry *dst = sk_dst_get(sk); 324 325 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) { 326 sk_dst_reset(sk); 327 dst_release(dst); 328 return NULL; 329 } 330 331 return dst; 332} 333EXPORT_SYMBOL(sk_dst_check); 334 335/* 336 * This is meant for all protocols to use and covers goings on 337 * at the socket level. Everything here is generic. 338 */ 339 340int sock_setsockopt(struct socket *sock, int level, int optname, 341 char __user *optval, int optlen) 342{ 343 struct sock *sk=sock->sk; 344 struct sk_filter *filter; 345 int val; 346 int valbool; 347 struct linger ling; 348 int ret = 0; 349 350 /* 351 * Options without arguments 352 */ 353 354#ifdef SO_DONTLINGER /* Compatibility item... */ 355 if (optname == SO_DONTLINGER) { 356 lock_sock(sk); 357 sock_reset_flag(sk, SOCK_LINGER); 358 release_sock(sk); 359 return 0; 360 } 361#endif 362 363 if(optlen<sizeof(int)) 364 return(-EINVAL); 365 366 if (get_user(val, (int __user *)optval)) 367 return -EFAULT; 368 369 valbool = val?1:0; 370 371 lock_sock(sk); 372 373 switch(optname) 374 { 375 case SO_DEBUG: 376 if(val && !capable(CAP_NET_ADMIN)) 377 { 378 ret = -EACCES; 379 } 380 else if (valbool) 381 sock_set_flag(sk, SOCK_DBG); 382 else 383 sock_reset_flag(sk, SOCK_DBG); 384 break; 385 case SO_REUSEADDR: 386 sk->sk_reuse = valbool; 387 break; 388 case SO_TYPE: 389 case SO_ERROR: 390 ret = -ENOPROTOOPT; 391 break; 392 case SO_DONTROUTE: 393 if (valbool) 394 sock_set_flag(sk, SOCK_LOCALROUTE); 395 else 396 sock_reset_flag(sk, SOCK_LOCALROUTE); 397 break; 398 case SO_BROADCAST: 399 sock_valbool_flag(sk, SOCK_BROADCAST, valbool); 400 break; 401 case SO_SNDBUF: 402 /* Don't error on this BSD doesn't and if you think 403 about it this is right. Otherwise apps have to 404 play 'guess the biggest size' games. RCVBUF/SNDBUF 405 are treated in BSD as hints */ 406 407 if (val > sysctl_wmem_max) 408 val = sysctl_wmem_max; 409set_sndbuf: 410 sk->sk_userlocks |= SOCK_SNDBUF_LOCK; 411 if ((val * 2) < SOCK_MIN_SNDBUF) 412 sk->sk_sndbuf = SOCK_MIN_SNDBUF; 413 else 414 sk->sk_sndbuf = val * 2; 415 416 /* 417 * Wake up sending tasks if we 418 * upped the value. 419 */ 420 sk->sk_write_space(sk); 421 break; 422 423 case SO_SNDBUFFORCE: 424 if (!capable(CAP_NET_ADMIN)) { 425 ret = -EPERM; 426 break; 427 } 428 goto set_sndbuf; 429 430 case SO_RCVBUF: 431 /* Don't error on this BSD doesn't and if you think 432 about it this is right. Otherwise apps have to 433 play 'guess the biggest size' games. RCVBUF/SNDBUF 434 are treated in BSD as hints */ 435 436 if (val > sysctl_rmem_max) 437 val = sysctl_rmem_max; 438set_rcvbuf: 439 sk->sk_userlocks |= SOCK_RCVBUF_LOCK; 440 /* 441 * We double it on the way in to account for 442 * "struct sk_buff" etc. overhead. Applications 443 * assume that the SO_RCVBUF setting they make will 444 * allow that much actual data to be received on that 445 * socket. 446 * 447 * Applications are unaware that "struct sk_buff" and 448 * other overheads allocate from the receive buffer 449 * during socket buffer allocation. 450 * 451 * And after considering the possible alternatives, 452 * returning the value we actually used in getsockopt 453 * is the most desirable behavior. 454 */ 455 if ((val * 2) < SOCK_MIN_RCVBUF) 456 sk->sk_rcvbuf = SOCK_MIN_RCVBUF; 457 else 458 sk->sk_rcvbuf = val * 2; 459 break; 460 461 case SO_RCVBUFFORCE: 462 if (!capable(CAP_NET_ADMIN)) { 463 ret = -EPERM; 464 break; 465 } 466 goto set_rcvbuf; 467 468 case SO_KEEPALIVE: 469#ifdef CONFIG_INET 470 if (sk->sk_protocol == IPPROTO_TCP) 471 tcp_set_keepalive(sk, valbool); 472#endif 473 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool); 474 break; 475 476 case SO_OOBINLINE: 477 sock_valbool_flag(sk, SOCK_URGINLINE, valbool); 478 break; 479 480 case SO_NO_CHECK: 481 sk->sk_no_check = valbool; 482 break; 483 484 case SO_PRIORITY: 485 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN)) 486 sk->sk_priority = val; 487 else 488 ret = -EPERM; 489 break; 490 491 case SO_LINGER: 492 if(optlen<sizeof(ling)) { 493 ret = -EINVAL; /* 1003.1g */ 494 break; 495 } 496 if (copy_from_user(&ling,optval,sizeof(ling))) { 497 ret = -EFAULT; 498 break; 499 } 500 if (!ling.l_onoff) 501 sock_reset_flag(sk, SOCK_LINGER); 502 else { 503#if (BITS_PER_LONG == 32) 504 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ) 505 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT; 506 else 507#endif 508 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ; 509 sock_set_flag(sk, SOCK_LINGER); 510 } 511 break; 512 513 case SO_BSDCOMPAT: 514 sock_warn_obsolete_bsdism("setsockopt"); 515 break; 516 517 case SO_PASSCRED: 518 if (valbool) 519 set_bit(SOCK_PASSCRED, &sock->flags); 520 else 521 clear_bit(SOCK_PASSCRED, &sock->flags); 522 break; 523 524 case SO_TIMESTAMP: 525 if (valbool) { 526 sock_set_flag(sk, SOCK_RCVTSTAMP); 527 sock_enable_timestamp(sk); 528 } else 529 sock_reset_flag(sk, SOCK_RCVTSTAMP); 530 break; 531 532 case SO_RCVLOWAT: 533 if (val < 0) 534 val = INT_MAX; 535 sk->sk_rcvlowat = val ? : 1; 536 break; 537 538 case SO_RCVTIMEO: 539 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen); 540 break; 541 542 case SO_SNDTIMEO: 543 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen); 544 break; 545 546#ifdef CONFIG_NETDEVICES 547 case SO_BINDTODEVICE: 548 { 549 char devname[IFNAMSIZ]; 550 551 /* Sorry... */ 552 if (!capable(CAP_NET_RAW)) { 553 ret = -EPERM; 554 break; 555 } 556 557 /* Bind this socket to a particular device like "eth0", 558 * as specified in the passed interface name. If the 559 * name is "" or the option length is zero the socket 560 * is not bound. 561 */ 562 563 if (!valbool) { 564 sk->sk_bound_dev_if = 0; 565 } else { 566 if (optlen > IFNAMSIZ - 1) 567 optlen = IFNAMSIZ - 1; 568 memset(devname, 0, sizeof(devname)); 569 if (copy_from_user(devname, optval, optlen)) { 570 ret = -EFAULT; 571 break; 572 } 573 574 /* Remove any cached route for this socket. */ 575 sk_dst_reset(sk); 576 577 if (devname[0] == '\0') { 578 sk->sk_bound_dev_if = 0; 579 } else { 580 struct net_device *dev = dev_get_by_name(devname); 581 if (!dev) { 582 ret = -ENODEV; 583 break; 584 } 585 sk->sk_bound_dev_if = dev->ifindex; 586 dev_put(dev); 587 } 588 } 589 break; 590 } 591#endif 592 593 594 case SO_ATTACH_FILTER: 595 ret = -EINVAL; 596 if (optlen == sizeof(struct sock_fprog)) { 597 struct sock_fprog fprog; 598 599 ret = -EFAULT; 600 if (copy_from_user(&fprog, optval, sizeof(fprog))) 601 break; 602 603 ret = sk_attach_filter(&fprog, sk); 604 } 605 break; 606 607 case SO_DETACH_FILTER: 608 rcu_read_lock_bh(); 609 filter = rcu_dereference(sk->sk_filter); 610 if (filter) { 611 rcu_assign_pointer(sk->sk_filter, NULL); 612 sk_filter_release(sk, filter); 613 rcu_read_unlock_bh(); 614 break; 615 } 616 rcu_read_unlock_bh(); 617 ret = -ENONET; 618 break; 619 620 case SO_PASSSEC: 621 if (valbool) 622 set_bit(SOCK_PASSSEC, &sock->flags); 623 else 624 clear_bit(SOCK_PASSSEC, &sock->flags); 625 break; 626 627 /* We implement the SO_SNDLOWAT etc to 628 not be settable (1003.1g 5.3) */ 629 default: 630 ret = -ENOPROTOOPT; 631 break; 632 } 633 release_sock(sk); 634 return ret; 635} 636 637 638int sock_getsockopt(struct socket *sock, int level, int optname, 639 char __user *optval, int __user *optlen) 640{ 641 struct sock *sk = sock->sk; 642 643 union 644 { 645 int val; 646 struct linger ling; 647 struct timeval tm; 648 } v; 649 650 unsigned int lv = sizeof(int); 651 int len; 652 653 if(get_user(len,optlen)) 654 return -EFAULT; 655 if(len < 0) 656 return -EINVAL; 657 658 switch(optname) 659 { 660 case SO_DEBUG: 661 v.val = sock_flag(sk, SOCK_DBG); 662 break; 663 664 case SO_DONTROUTE: 665 v.val = sock_flag(sk, SOCK_LOCALROUTE); 666 break; 667 668 case SO_BROADCAST: 669 v.val = !!sock_flag(sk, SOCK_BROADCAST); 670 break; 671 672 case SO_SNDBUF: 673 v.val = sk->sk_sndbuf; 674 break; 675 676 case SO_RCVBUF: 677 v.val = sk->sk_rcvbuf; 678 break; 679 680 case SO_REUSEADDR: 681 v.val = sk->sk_reuse; 682 break; 683 684 case SO_KEEPALIVE: 685 v.val = !!sock_flag(sk, SOCK_KEEPOPEN); 686 break; 687 688 case SO_TYPE: 689 v.val = sk->sk_type; 690 break; 691 692 case SO_ERROR: 693 v.val = -sock_error(sk); 694 if(v.val==0) 695 v.val = xchg(&sk->sk_err_soft, 0); 696 break; 697 698 case SO_OOBINLINE: 699 v.val = !!sock_flag(sk, SOCK_URGINLINE); 700 break; 701 702 case SO_NO_CHECK: 703 v.val = sk->sk_no_check; 704 break; 705 706 case SO_PRIORITY: 707 v.val = sk->sk_priority; 708 break; 709 710 case SO_LINGER: 711 lv = sizeof(v.ling); 712 v.ling.l_onoff = !!sock_flag(sk, SOCK_LINGER); 713 v.ling.l_linger = sk->sk_lingertime / HZ; 714 break; 715 716 case SO_BSDCOMPAT: 717 sock_warn_obsolete_bsdism("getsockopt"); 718 break; 719 720 case SO_TIMESTAMP: 721 v.val = sock_flag(sk, SOCK_RCVTSTAMP); 722 break; 723 724 case SO_RCVTIMEO: 725 lv=sizeof(struct timeval); 726 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) { 727 v.tm.tv_sec = 0; 728 v.tm.tv_usec = 0; 729 } else { 730 v.tm.tv_sec = sk->sk_rcvtimeo / HZ; 731 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ; 732 } 733 break; 734 735 case SO_SNDTIMEO: 736 lv=sizeof(struct timeval); 737 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) { 738 v.tm.tv_sec = 0; 739 v.tm.tv_usec = 0; 740 } else { 741 v.tm.tv_sec = sk->sk_sndtimeo / HZ; 742 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ; 743 } 744 break; 745 746 case SO_RCVLOWAT: 747 v.val = sk->sk_rcvlowat; 748 break; 749 750 case SO_SNDLOWAT: 751 v.val=1; 752 break; 753 754 case SO_PASSCRED: 755 v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0; 756 break; 757 758 case SO_PEERCRED: 759 if (len > sizeof(sk->sk_peercred)) 760 len = sizeof(sk->sk_peercred); 761 if (copy_to_user(optval, &sk->sk_peercred, len)) 762 return -EFAULT; 763 goto lenout; 764 765 case SO_PEERNAME: 766 { 767 char address[128]; 768 769 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2)) 770 return -ENOTCONN; 771 if (lv < len) 772 return -EINVAL; 773 if (copy_to_user(optval, address, len)) 774 return -EFAULT; 775 goto lenout; 776 } 777 778 /* Dubious BSD thing... Probably nobody even uses it, but 779 * the UNIX standard wants it for whatever reason... -DaveM 780 */ 781 case SO_ACCEPTCONN: 782 v.val = sk->sk_state == TCP_LISTEN; 783 break; 784 785 case SO_PASSSEC: 786 v.val = test_bit(SOCK_PASSSEC, &sock->flags) ? 1 : 0; 787 break; 788 789 case SO_PEERSEC: 790 return security_socket_getpeersec_stream(sock, optval, optlen, len); 791 792 default: 793 return(-ENOPROTOOPT); 794 } 795 if (len > lv) 796 len = lv; 797 if (copy_to_user(optval, &v, len)) 798 return -EFAULT; 799lenout: 800 if (put_user(len, optlen)) 801 return -EFAULT; 802 return 0; 803} 804 805/* 806 * Initialize an sk_lock. 807 * 808 * (We also register the sk_lock with the lock validator.) 809 */ 810static void inline sock_lock_init(struct sock *sk) 811{ 812 spin_lock_init(&sk->sk_lock.slock); 813 sk->sk_lock.owner = NULL; 814 init_waitqueue_head(&sk->sk_lock.wq); 815 /* 816 * Make sure we are not reinitializing a held lock: 817 */ 818 debug_check_no_locks_freed((void *)&sk->sk_lock, sizeof(sk->sk_lock)); 819 820 /* 821 * Mark both the sk_lock and the sk_lock.slock as a 822 * per-address-family lock class: 823 */ 824 lockdep_set_class_and_name(&sk->sk_lock.slock, 825 af_family_slock_keys + sk->sk_family, 826 af_family_slock_key_strings[sk->sk_family]); 827 lockdep_init_map(&sk->sk_lock.dep_map, 828 af_family_key_strings[sk->sk_family], 829 af_family_keys + sk->sk_family, 0); 830} 831 832/** 833 * sk_alloc - All socket objects are allocated here 834 * @family: protocol family 835 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 836 * @prot: struct proto associated with this new sock instance 837 * @zero_it: if we should zero the newly allocated sock 838 */ 839struct sock *sk_alloc(int family, gfp_t priority, 840 struct proto *prot, int zero_it) 841{ 842 struct sock *sk = NULL; 843 kmem_cache_t *slab = prot->slab; 844 845 if (slab != NULL) 846 sk = kmem_cache_alloc(slab, priority); 847 else 848 sk = kmalloc(prot->obj_size, priority); 849 850 if (sk) { 851 if (zero_it) { 852 memset(sk, 0, prot->obj_size); 853 sk->sk_family = family; 854 /* 855 * See comment in struct sock definition to understand 856 * why we need sk_prot_creator -acme 857 */ 858 sk->sk_prot = sk->sk_prot_creator = prot; 859 sock_lock_init(sk); 860 } 861 862 if (security_sk_alloc(sk, family, priority)) 863 goto out_free; 864 865 if (!try_module_get(prot->owner)) 866 goto out_free; 867 } 868 return sk; 869 870out_free: 871 if (slab != NULL) 872 kmem_cache_free(slab, sk); 873 else 874 kfree(sk); 875 return NULL; 876} 877 878void sk_free(struct sock *sk) 879{ 880 struct sk_filter *filter; 881 struct module *owner = sk->sk_prot_creator->owner; 882 883 if (sk->sk_destruct) 884 sk->sk_destruct(sk); 885 886 filter = rcu_dereference(sk->sk_filter); 887 if (filter) { 888 sk_filter_release(sk, filter); 889 rcu_assign_pointer(sk->sk_filter, NULL); 890 } 891 892 sock_disable_timestamp(sk); 893 894 if (atomic_read(&sk->sk_omem_alloc)) 895 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n", 896 __FUNCTION__, atomic_read(&sk->sk_omem_alloc)); 897 898 security_sk_free(sk); 899 if (sk->sk_prot_creator->slab != NULL) 900 kmem_cache_free(sk->sk_prot_creator->slab, sk); 901 else 902 kfree(sk); 903 module_put(owner); 904} 905 906struct sock *sk_clone(const struct sock *sk, const gfp_t priority) 907{ 908 struct sock *newsk = sk_alloc(sk->sk_family, priority, sk->sk_prot, 0); 909 910 if (newsk != NULL) { 911 struct sk_filter *filter; 912 913 sock_copy(newsk, sk); 914 915 /* SANITY */ 916 sk_node_init(&newsk->sk_node); 917 sock_lock_init(newsk); 918 bh_lock_sock(newsk); 919 920 atomic_set(&newsk->sk_rmem_alloc, 0); 921 atomic_set(&newsk->sk_wmem_alloc, 0); 922 atomic_set(&newsk->sk_omem_alloc, 0); 923 skb_queue_head_init(&newsk->sk_receive_queue); 924 skb_queue_head_init(&newsk->sk_write_queue); 925#ifdef CONFIG_NET_DMA 926 skb_queue_head_init(&newsk->sk_async_wait_queue); 927#endif 928 929 rwlock_init(&newsk->sk_dst_lock); 930 rwlock_init(&newsk->sk_callback_lock); 931 lockdep_set_class(&newsk->sk_callback_lock, 932 af_callback_keys + newsk->sk_family); 933 934 newsk->sk_dst_cache = NULL; 935 newsk->sk_wmem_queued = 0; 936 newsk->sk_forward_alloc = 0; 937 newsk->sk_send_head = NULL; 938 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL; 939 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK; 940 941 sock_reset_flag(newsk, SOCK_DONE); 942 skb_queue_head_init(&newsk->sk_error_queue); 943 944 filter = newsk->sk_filter; 945 if (filter != NULL) 946 sk_filter_charge(newsk, filter); 947 948 if (unlikely(xfrm_sk_clone_policy(newsk))) { 949 /* It is still raw copy of parent, so invalidate 950 * destructor and make plain sk_free() */ 951 newsk->sk_destruct = NULL; 952 sk_free(newsk); 953 newsk = NULL; 954 goto out; 955 } 956 957 newsk->sk_err = 0; 958 newsk->sk_priority = 0; 959 atomic_set(&newsk->sk_refcnt, 2); 960 961 /* 962 * Increment the counter in the same struct proto as the master 963 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that 964 * is the same as sk->sk_prot->socks, as this field was copied 965 * with memcpy). 966 * 967 * This _changes_ the previous behaviour, where 968 * tcp_create_openreq_child always was incrementing the 969 * equivalent to tcp_prot->socks (inet_sock_nr), so this have 970 * to be taken into account in all callers. -acme 971 */ 972 sk_refcnt_debug_inc(newsk); 973 newsk->sk_socket = NULL; 974 newsk->sk_sleep = NULL; 975 976 if (newsk->sk_prot->sockets_allocated) 977 atomic_inc(newsk->sk_prot->sockets_allocated); 978 } 979out: 980 return newsk; 981} 982 983EXPORT_SYMBOL_GPL(sk_clone); 984 985void __init sk_init(void) 986{ 987 if (num_physpages <= 4096) { 988 sysctl_wmem_max = 32767; 989 sysctl_rmem_max = 32767; 990 sysctl_wmem_default = 32767; 991 sysctl_rmem_default = 32767; 992 } else if (num_physpages >= 131072) { 993 sysctl_wmem_max = 131071; 994 sysctl_rmem_max = 131071; 995 } 996} 997 998/* 999 * Simple resource managers for sockets. 1000 */ 1001 1002 1003/* 1004 * Write buffer destructor automatically called from kfree_skb. 1005 */ 1006void sock_wfree(struct sk_buff *skb) 1007{ 1008 struct sock *sk = skb->sk; 1009 1010 /* In case it might be waiting for more memory. */ 1011 atomic_sub(skb->truesize, &sk->sk_wmem_alloc); 1012 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) 1013 sk->sk_write_space(sk); 1014 sock_put(sk); 1015} 1016 1017/* 1018 * Read buffer destructor automatically called from kfree_skb. 1019 */ 1020void sock_rfree(struct sk_buff *skb) 1021{ 1022 struct sock *sk = skb->sk; 1023 1024 atomic_sub(skb->truesize, &sk->sk_rmem_alloc); 1025} 1026 1027 1028int sock_i_uid(struct sock *sk) 1029{ 1030 int uid; 1031 1032 read_lock(&sk->sk_callback_lock); 1033 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0; 1034 read_unlock(&sk->sk_callback_lock); 1035 return uid; 1036} 1037 1038unsigned long sock_i_ino(struct sock *sk) 1039{ 1040 unsigned long ino; 1041 1042 read_lock(&sk->sk_callback_lock); 1043 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0; 1044 read_unlock(&sk->sk_callback_lock); 1045 return ino; 1046} 1047 1048/* 1049 * Allocate a skb from the socket's send buffer. 1050 */ 1051struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 1052 gfp_t priority) 1053{ 1054 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) { 1055 struct sk_buff * skb = alloc_skb(size, priority); 1056 if (skb) { 1057 skb_set_owner_w(skb, sk); 1058 return skb; 1059 } 1060 } 1061 return NULL; 1062} 1063 1064/* 1065 * Allocate a skb from the socket's receive buffer. 1066 */ 1067struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force, 1068 gfp_t priority) 1069{ 1070 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) { 1071 struct sk_buff *skb = alloc_skb(size, priority); 1072 if (skb) { 1073 skb_set_owner_r(skb, sk); 1074 return skb; 1075 } 1076 } 1077 return NULL; 1078} 1079 1080/* 1081 * Allocate a memory block from the socket's option memory buffer. 1082 */ 1083void *sock_kmalloc(struct sock *sk, int size, gfp_t priority) 1084{ 1085 if ((unsigned)size <= sysctl_optmem_max && 1086 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) { 1087 void *mem; 1088 /* First do the add, to avoid the race if kmalloc 1089 * might sleep. 1090 */ 1091 atomic_add(size, &sk->sk_omem_alloc); 1092 mem = kmalloc(size, priority); 1093 if (mem) 1094 return mem; 1095 atomic_sub(size, &sk->sk_omem_alloc); 1096 } 1097 return NULL; 1098} 1099 1100/* 1101 * Free an option memory block. 1102 */ 1103void sock_kfree_s(struct sock *sk, void *mem, int size) 1104{ 1105 kfree(mem); 1106 atomic_sub(size, &sk->sk_omem_alloc); 1107} 1108 1109/* It is almost wait_for_tcp_memory minus release_sock/lock_sock. 1110 I think, these locks should be removed for datagram sockets. 1111 */ 1112static long sock_wait_for_wmem(struct sock * sk, long timeo) 1113{ 1114 DEFINE_WAIT(wait); 1115 1116 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags); 1117 for (;;) { 1118 if (!timeo) 1119 break; 1120 if (signal_pending(current)) 1121 break; 1122 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 1123 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE); 1124 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) 1125 break; 1126 if (sk->sk_shutdown & SEND_SHUTDOWN) 1127 break; 1128 if (sk->sk_err) 1129 break; 1130 timeo = schedule_timeout(timeo); 1131 } 1132 finish_wait(sk->sk_sleep, &wait); 1133 return timeo; 1134} 1135 1136 1137/* 1138 * Generic send/receive buffer handlers 1139 */ 1140 1141static struct sk_buff *sock_alloc_send_pskb(struct sock *sk, 1142 unsigned long header_len, 1143 unsigned long data_len, 1144 int noblock, int *errcode) 1145{ 1146 struct sk_buff *skb; 1147 gfp_t gfp_mask; 1148 long timeo; 1149 int err; 1150 1151 gfp_mask = sk->sk_allocation; 1152 if (gfp_mask & __GFP_WAIT) 1153 gfp_mask |= __GFP_REPEAT; 1154 1155 timeo = sock_sndtimeo(sk, noblock); 1156 while (1) { 1157 err = sock_error(sk); 1158 if (err != 0) 1159 goto failure; 1160 1161 err = -EPIPE; 1162 if (sk->sk_shutdown & SEND_SHUTDOWN) 1163 goto failure; 1164 1165 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) { 1166 skb = alloc_skb(header_len, gfp_mask); 1167 if (skb) { 1168 int npages; 1169 int i; 1170 1171 /* No pages, we're done... */ 1172 if (!data_len) 1173 break; 1174 1175 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT; 1176 skb->truesize += data_len; 1177 skb_shinfo(skb)->nr_frags = npages; 1178 for (i = 0; i < npages; i++) { 1179 struct page *page; 1180 skb_frag_t *frag; 1181 1182 page = alloc_pages(sk->sk_allocation, 0); 1183 if (!page) { 1184 err = -ENOBUFS; 1185 skb_shinfo(skb)->nr_frags = i; 1186 kfree_skb(skb); 1187 goto failure; 1188 } 1189 1190 frag = &skb_shinfo(skb)->frags[i]; 1191 frag->page = page; 1192 frag->page_offset = 0; 1193 frag->size = (data_len >= PAGE_SIZE ? 1194 PAGE_SIZE : 1195 data_len); 1196 data_len -= PAGE_SIZE; 1197 } 1198 1199 /* Full success... */ 1200 break; 1201 } 1202 err = -ENOBUFS; 1203 goto failure; 1204 } 1205 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags); 1206 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 1207 err = -EAGAIN; 1208 if (!timeo) 1209 goto failure; 1210 if (signal_pending(current)) 1211 goto interrupted; 1212 timeo = sock_wait_for_wmem(sk, timeo); 1213 } 1214 1215 skb_set_owner_w(skb, sk); 1216 return skb; 1217 1218interrupted: 1219 err = sock_intr_errno(timeo); 1220failure: 1221 *errcode = err; 1222 return NULL; 1223} 1224 1225struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size, 1226 int noblock, int *errcode) 1227{ 1228 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode); 1229} 1230 1231static void __lock_sock(struct sock *sk) 1232{ 1233 DEFINE_WAIT(wait); 1234 1235 for(;;) { 1236 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait, 1237 TASK_UNINTERRUPTIBLE); 1238 spin_unlock_bh(&sk->sk_lock.slock); 1239 schedule(); 1240 spin_lock_bh(&sk->sk_lock.slock); 1241 if(!sock_owned_by_user(sk)) 1242 break; 1243 } 1244 finish_wait(&sk->sk_lock.wq, &wait); 1245} 1246 1247static void __release_sock(struct sock *sk) 1248{ 1249 struct sk_buff *skb = sk->sk_backlog.head; 1250 1251 do { 1252 sk->sk_backlog.head = sk->sk_backlog.tail = NULL; 1253 bh_unlock_sock(sk); 1254 1255 do { 1256 struct sk_buff *next = skb->next; 1257 1258 skb->next = NULL; 1259 sk->sk_backlog_rcv(sk, skb); 1260 1261 /* 1262 * We are in process context here with softirqs 1263 * disabled, use cond_resched_softirq() to preempt. 1264 * This is safe to do because we've taken the backlog 1265 * queue private: 1266 */ 1267 cond_resched_softirq(); 1268 1269 skb = next; 1270 } while (skb != NULL); 1271 1272 bh_lock_sock(sk); 1273 } while((skb = sk->sk_backlog.head) != NULL); 1274} 1275 1276/** 1277 * sk_wait_data - wait for data to arrive at sk_receive_queue 1278 * @sk: sock to wait on 1279 * @timeo: for how long 1280 * 1281 * Now socket state including sk->sk_err is changed only under lock, 1282 * hence we may omit checks after joining wait queue. 1283 * We check receive queue before schedule() only as optimization; 1284 * it is very likely that release_sock() added new data. 1285 */ 1286int sk_wait_data(struct sock *sk, long *timeo) 1287{ 1288 int rc; 1289 DEFINE_WAIT(wait); 1290 1291 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE); 1292 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags); 1293 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue)); 1294 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags); 1295 finish_wait(sk->sk_sleep, &wait); 1296 return rc; 1297} 1298 1299EXPORT_SYMBOL(sk_wait_data); 1300 1301/* 1302 * Set of default routines for initialising struct proto_ops when 1303 * the protocol does not support a particular function. In certain 1304 * cases where it makes no sense for a protocol to have a "do nothing" 1305 * function, some default processing is provided. 1306 */ 1307 1308int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len) 1309{ 1310 return -EOPNOTSUPP; 1311} 1312 1313int sock_no_connect(struct socket *sock, struct sockaddr *saddr, 1314 int len, int flags) 1315{ 1316 return -EOPNOTSUPP; 1317} 1318 1319int sock_no_socketpair(struct socket *sock1, struct socket *sock2) 1320{ 1321 return -EOPNOTSUPP; 1322} 1323 1324int sock_no_accept(struct socket *sock, struct socket *newsock, int flags) 1325{ 1326 return -EOPNOTSUPP; 1327} 1328 1329int sock_no_getname(struct socket *sock, struct sockaddr *saddr, 1330 int *len, int peer) 1331{ 1332 return -EOPNOTSUPP; 1333} 1334 1335unsigned int sock_no_poll(struct file * file, struct socket *sock, poll_table *pt) 1336{ 1337 return 0; 1338} 1339 1340int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 1341{ 1342 return -EOPNOTSUPP; 1343} 1344 1345int sock_no_listen(struct socket *sock, int backlog) 1346{ 1347 return -EOPNOTSUPP; 1348} 1349 1350int sock_no_shutdown(struct socket *sock, int how) 1351{ 1352 return -EOPNOTSUPP; 1353} 1354 1355int sock_no_setsockopt(struct socket *sock, int level, int optname, 1356 char __user *optval, int optlen) 1357{ 1358 return -EOPNOTSUPP; 1359} 1360 1361int sock_no_getsockopt(struct socket *sock, int level, int optname, 1362 char __user *optval, int __user *optlen) 1363{ 1364 return -EOPNOTSUPP; 1365} 1366 1367int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m, 1368 size_t len) 1369{ 1370 return -EOPNOTSUPP; 1371} 1372 1373int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m, 1374 size_t len, int flags) 1375{ 1376 return -EOPNOTSUPP; 1377} 1378 1379int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma) 1380{ 1381 /* Mirror missing mmap method error code */ 1382 return -ENODEV; 1383} 1384 1385ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags) 1386{ 1387 ssize_t res; 1388 struct msghdr msg = {.msg_flags = flags}; 1389 struct kvec iov; 1390 char *kaddr = kmap(page); 1391 iov.iov_base = kaddr + offset; 1392 iov.iov_len = size; 1393 res = kernel_sendmsg(sock, &msg, &iov, 1, size); 1394 kunmap(page); 1395 return res; 1396} 1397 1398/* 1399 * Default Socket Callbacks 1400 */ 1401 1402static void sock_def_wakeup(struct sock *sk) 1403{ 1404 read_lock(&sk->sk_callback_lock); 1405 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) 1406 wake_up_interruptible_all(sk->sk_sleep); 1407 read_unlock(&sk->sk_callback_lock); 1408} 1409 1410static void sock_def_error_report(struct sock *sk) 1411{ 1412 read_lock(&sk->sk_callback_lock); 1413 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) 1414 wake_up_interruptible(sk->sk_sleep); 1415 sk_wake_async(sk,0,POLL_ERR); 1416 read_unlock(&sk->sk_callback_lock); 1417} 1418 1419static void sock_def_readable(struct sock *sk, int len) 1420{ 1421 read_lock(&sk->sk_callback_lock); 1422 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) 1423 wake_up_interruptible(sk->sk_sleep); 1424 sk_wake_async(sk,1,POLL_IN); 1425 read_unlock(&sk->sk_callback_lock); 1426} 1427 1428static void sock_def_write_space(struct sock *sk) 1429{ 1430 read_lock(&sk->sk_callback_lock); 1431 1432 /* Do not wake up a writer until he can make "significant" 1433 * progress. --DaveM 1434 */ 1435 if((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) { 1436 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) 1437 wake_up_interruptible(sk->sk_sleep); 1438 1439 /* Should agree with poll, otherwise some programs break */ 1440 if (sock_writeable(sk)) 1441 sk_wake_async(sk, 2, POLL_OUT); 1442 } 1443 1444 read_unlock(&sk->sk_callback_lock); 1445} 1446 1447static void sock_def_destruct(struct sock *sk) 1448{ 1449 kfree(sk->sk_protinfo); 1450} 1451 1452void sk_send_sigurg(struct sock *sk) 1453{ 1454 if (sk->sk_socket && sk->sk_socket->file) 1455 if (send_sigurg(&sk->sk_socket->file->f_owner)) 1456 sk_wake_async(sk, 3, POLL_PRI); 1457} 1458 1459void sk_reset_timer(struct sock *sk, struct timer_list* timer, 1460 unsigned long expires) 1461{ 1462 if (!mod_timer(timer, expires)) 1463 sock_hold(sk); 1464} 1465 1466EXPORT_SYMBOL(sk_reset_timer); 1467 1468void sk_stop_timer(struct sock *sk, struct timer_list* timer) 1469{ 1470 if (timer_pending(timer) && del_timer(timer)) 1471 __sock_put(sk); 1472} 1473 1474EXPORT_SYMBOL(sk_stop_timer); 1475 1476void sock_init_data(struct socket *sock, struct sock *sk) 1477{ 1478 skb_queue_head_init(&sk->sk_receive_queue); 1479 skb_queue_head_init(&sk->sk_write_queue); 1480 skb_queue_head_init(&sk->sk_error_queue); 1481#ifdef CONFIG_NET_DMA 1482 skb_queue_head_init(&sk->sk_async_wait_queue); 1483#endif 1484 1485 sk->sk_send_head = NULL; 1486 1487 init_timer(&sk->sk_timer); 1488 1489 sk->sk_allocation = GFP_KERNEL; 1490 sk->sk_rcvbuf = sysctl_rmem_default; 1491 sk->sk_sndbuf = sysctl_wmem_default; 1492 sk->sk_state = TCP_CLOSE; 1493 sk->sk_socket = sock; 1494 1495 sock_set_flag(sk, SOCK_ZAPPED); 1496 1497 if(sock) 1498 { 1499 sk->sk_type = sock->type; 1500 sk->sk_sleep = &sock->wait; 1501 sock->sk = sk; 1502 } else 1503 sk->sk_sleep = NULL; 1504 1505 rwlock_init(&sk->sk_dst_lock); 1506 rwlock_init(&sk->sk_callback_lock); 1507 lockdep_set_class(&sk->sk_callback_lock, 1508 af_callback_keys + sk->sk_family); 1509 1510 sk->sk_state_change = sock_def_wakeup; 1511 sk->sk_data_ready = sock_def_readable; 1512 sk->sk_write_space = sock_def_write_space; 1513 sk->sk_error_report = sock_def_error_report; 1514 sk->sk_destruct = sock_def_destruct; 1515 1516 sk->sk_sndmsg_page = NULL; 1517 sk->sk_sndmsg_off = 0; 1518 1519 sk->sk_peercred.pid = 0; 1520 sk->sk_peercred.uid = -1; 1521 sk->sk_peercred.gid = -1; 1522 sk->sk_write_pending = 0; 1523 sk->sk_rcvlowat = 1; 1524 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; 1525 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT; 1526 1527 sk->sk_stamp.tv_sec = -1L; 1528 sk->sk_stamp.tv_usec = -1L; 1529 1530 atomic_set(&sk->sk_refcnt, 1); 1531} 1532 1533void fastcall lock_sock_nested(struct sock *sk, int subclass) 1534{ 1535 might_sleep(); 1536 spin_lock_bh(&sk->sk_lock.slock); 1537 if (sk->sk_lock.owner) 1538 __lock_sock(sk); 1539 sk->sk_lock.owner = (void *)1; 1540 spin_unlock(&sk->sk_lock.slock); 1541 /* 1542 * The sk_lock has mutex_lock() semantics here: 1543 */ 1544 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_); 1545 local_bh_enable(); 1546} 1547 1548EXPORT_SYMBOL(lock_sock_nested); 1549 1550void fastcall release_sock(struct sock *sk) 1551{ 1552 /* 1553 * The sk_lock has mutex_unlock() semantics: 1554 */ 1555 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_); 1556 1557 spin_lock_bh(&sk->sk_lock.slock); 1558 if (sk->sk_backlog.tail) 1559 __release_sock(sk); 1560 sk->sk_lock.owner = NULL; 1561 if (waitqueue_active(&sk->sk_lock.wq)) 1562 wake_up(&sk->sk_lock.wq); 1563 spin_unlock_bh(&sk->sk_lock.slock); 1564} 1565EXPORT_SYMBOL(release_sock); 1566 1567int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp) 1568{ 1569 if (!sock_flag(sk, SOCK_TIMESTAMP)) 1570 sock_enable_timestamp(sk); 1571 if (sk->sk_stamp.tv_sec == -1) 1572 return -ENOENT; 1573 if (sk->sk_stamp.tv_sec == 0) 1574 do_gettimeofday(&sk->sk_stamp); 1575 return copy_to_user(userstamp, &sk->sk_stamp, sizeof(struct timeval)) ? 1576 -EFAULT : 0; 1577} 1578EXPORT_SYMBOL(sock_get_timestamp); 1579 1580void sock_enable_timestamp(struct sock *sk) 1581{ 1582 if (!sock_flag(sk, SOCK_TIMESTAMP)) { 1583 sock_set_flag(sk, SOCK_TIMESTAMP); 1584 net_enable_timestamp(); 1585 } 1586} 1587EXPORT_SYMBOL(sock_enable_timestamp); 1588 1589/* 1590 * Get a socket option on an socket. 1591 * 1592 * FIX: POSIX 1003.1g is very ambiguous here. It states that 1593 * asynchronous errors should be reported by getsockopt. We assume 1594 * this means if you specify SO_ERROR (otherwise whats the point of it). 1595 */ 1596int sock_common_getsockopt(struct socket *sock, int level, int optname, 1597 char __user *optval, int __user *optlen) 1598{ 1599 struct sock *sk = sock->sk; 1600 1601 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen); 1602} 1603 1604EXPORT_SYMBOL(sock_common_getsockopt); 1605 1606#ifdef CONFIG_COMPAT 1607int compat_sock_common_getsockopt(struct socket *sock, int level, int optname, 1608 char __user *optval, int __user *optlen) 1609{ 1610 struct sock *sk = sock->sk; 1611 1612 if (sk->sk_prot->compat_setsockopt != NULL) 1613 return sk->sk_prot->compat_getsockopt(sk, level, optname, 1614 optval, optlen); 1615 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen); 1616} 1617EXPORT_SYMBOL(compat_sock_common_getsockopt); 1618#endif 1619 1620int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock, 1621 struct msghdr *msg, size_t size, int flags) 1622{ 1623 struct sock *sk = sock->sk; 1624 int addr_len = 0; 1625 int err; 1626 1627 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT, 1628 flags & ~MSG_DONTWAIT, &addr_len); 1629 if (err >= 0) 1630 msg->msg_namelen = addr_len; 1631 return err; 1632} 1633 1634EXPORT_SYMBOL(sock_common_recvmsg); 1635 1636/* 1637 * Set socket options on an inet socket. 1638 */ 1639int sock_common_setsockopt(struct socket *sock, int level, int optname, 1640 char __user *optval, int optlen) 1641{ 1642 struct sock *sk = sock->sk; 1643 1644 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen); 1645} 1646 1647EXPORT_SYMBOL(sock_common_setsockopt); 1648 1649#ifdef CONFIG_COMPAT 1650int compat_sock_common_setsockopt(struct socket *sock, int level, int optname, 1651 char __user *optval, int optlen) 1652{ 1653 struct sock *sk = sock->sk; 1654 1655 if (sk->sk_prot->compat_setsockopt != NULL) 1656 return sk->sk_prot->compat_setsockopt(sk, level, optname, 1657 optval, optlen); 1658 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen); 1659} 1660EXPORT_SYMBOL(compat_sock_common_setsockopt); 1661#endif 1662 1663void sk_common_release(struct sock *sk) 1664{ 1665 if (sk->sk_prot->destroy) 1666 sk->sk_prot->destroy(sk); 1667 1668 /* 1669 * Observation: when sock_common_release is called, processes have 1670 * no access to socket. But net still has. 1671 * Step one, detach it from networking: 1672 * 1673 * A. Remove from hash tables. 1674 */ 1675 1676 sk->sk_prot->unhash(sk); 1677 1678 /* 1679 * In this point socket cannot receive new packets, but it is possible 1680 * that some packets are in flight because some CPU runs receiver and 1681 * did hash table lookup before we unhashed socket. They will achieve 1682 * receive queue and will be purged by socket destructor. 1683 * 1684 * Also we still have packets pending on receive queue and probably, 1685 * our own packets waiting in device queues. sock_destroy will drain 1686 * receive queue, but transmitted packets will delay socket destruction 1687 * until the last reference will be released. 1688 */ 1689 1690 sock_orphan(sk); 1691 1692 xfrm_sk_free_policy(sk); 1693 1694 sk_refcnt_debug_release(sk); 1695 sock_put(sk); 1696} 1697 1698EXPORT_SYMBOL(sk_common_release); 1699 1700static DEFINE_RWLOCK(proto_list_lock); 1701static LIST_HEAD(proto_list); 1702 1703int proto_register(struct proto *prot, int alloc_slab) 1704{ 1705 char *request_sock_slab_name = NULL; 1706 char *timewait_sock_slab_name; 1707 int rc = -ENOBUFS; 1708 1709 if (alloc_slab) { 1710 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0, 1711 SLAB_HWCACHE_ALIGN, NULL, NULL); 1712 1713 if (prot->slab == NULL) { 1714 printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n", 1715 prot->name); 1716 goto out; 1717 } 1718 1719 if (prot->rsk_prot != NULL) { 1720 static const char mask[] = "request_sock_%s"; 1721 1722 request_sock_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL); 1723 if (request_sock_slab_name == NULL) 1724 goto out_free_sock_slab; 1725 1726 sprintf(request_sock_slab_name, mask, prot->name); 1727 prot->rsk_prot->slab = kmem_cache_create(request_sock_slab_name, 1728 prot->rsk_prot->obj_size, 0, 1729 SLAB_HWCACHE_ALIGN, NULL, NULL); 1730 1731 if (prot->rsk_prot->slab == NULL) { 1732 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n", 1733 prot->name); 1734 goto out_free_request_sock_slab_name; 1735 } 1736 } 1737 1738 if (prot->twsk_prot != NULL) { 1739 static const char mask[] = "tw_sock_%s"; 1740 1741 timewait_sock_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL); 1742 1743 if (timewait_sock_slab_name == NULL) 1744 goto out_free_request_sock_slab; 1745 1746 sprintf(timewait_sock_slab_name, mask, prot->name); 1747 prot->twsk_prot->twsk_slab = 1748 kmem_cache_create(timewait_sock_slab_name, 1749 prot->twsk_prot->twsk_obj_size, 1750 0, SLAB_HWCACHE_ALIGN, 1751 NULL, NULL); 1752 if (prot->twsk_prot->twsk_slab == NULL) 1753 goto out_free_timewait_sock_slab_name; 1754 } 1755 } 1756 1757 write_lock(&proto_list_lock); 1758 list_add(&prot->node, &proto_list); 1759 write_unlock(&proto_list_lock); 1760 rc = 0; 1761out: 1762 return rc; 1763out_free_timewait_sock_slab_name: 1764 kfree(timewait_sock_slab_name); 1765out_free_request_sock_slab: 1766 if (prot->rsk_prot && prot->rsk_prot->slab) { 1767 kmem_cache_destroy(prot->rsk_prot->slab); 1768 prot->rsk_prot->slab = NULL; 1769 } 1770out_free_request_sock_slab_name: 1771 kfree(request_sock_slab_name); 1772out_free_sock_slab: 1773 kmem_cache_destroy(prot->slab); 1774 prot->slab = NULL; 1775 goto out; 1776} 1777 1778EXPORT_SYMBOL(proto_register); 1779 1780void proto_unregister(struct proto *prot) 1781{ 1782 write_lock(&proto_list_lock); 1783 list_del(&prot->node); 1784 write_unlock(&proto_list_lock); 1785 1786 if (prot->slab != NULL) { 1787 kmem_cache_destroy(prot->slab); 1788 prot->slab = NULL; 1789 } 1790 1791 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) { 1792 const char *name = kmem_cache_name(prot->rsk_prot->slab); 1793 1794 kmem_cache_destroy(prot->rsk_prot->slab); 1795 kfree(name); 1796 prot->rsk_prot->slab = NULL; 1797 } 1798 1799 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) { 1800 const char *name = kmem_cache_name(prot->twsk_prot->twsk_slab); 1801 1802 kmem_cache_destroy(prot->twsk_prot->twsk_slab); 1803 kfree(name); 1804 prot->twsk_prot->twsk_slab = NULL; 1805 } 1806} 1807 1808EXPORT_SYMBOL(proto_unregister); 1809 1810#ifdef CONFIG_PROC_FS 1811static inline struct proto *__proto_head(void) 1812{ 1813 return list_entry(proto_list.next, struct proto, node); 1814} 1815 1816static inline struct proto *proto_head(void) 1817{ 1818 return list_empty(&proto_list) ? NULL : __proto_head(); 1819} 1820 1821static inline struct proto *proto_next(struct proto *proto) 1822{ 1823 return proto->node.next == &proto_list ? NULL : 1824 list_entry(proto->node.next, struct proto, node); 1825} 1826 1827static inline struct proto *proto_get_idx(loff_t pos) 1828{ 1829 struct proto *proto; 1830 loff_t i = 0; 1831 1832 list_for_each_entry(proto, &proto_list, node) 1833 if (i++ == pos) 1834 goto out; 1835 1836 proto = NULL; 1837out: 1838 return proto; 1839} 1840 1841static void *proto_seq_start(struct seq_file *seq, loff_t *pos) 1842{ 1843 read_lock(&proto_list_lock); 1844 return *pos ? proto_get_idx(*pos - 1) : SEQ_START_TOKEN; 1845} 1846 1847static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos) 1848{ 1849 ++*pos; 1850 return v == SEQ_START_TOKEN ? proto_head() : proto_next(v); 1851} 1852 1853static void proto_seq_stop(struct seq_file *seq, void *v) 1854{ 1855 read_unlock(&proto_list_lock); 1856} 1857 1858static char proto_method_implemented(const void *method) 1859{ 1860 return method == NULL ? 'n' : 'y'; 1861} 1862 1863static void proto_seq_printf(struct seq_file *seq, struct proto *proto) 1864{ 1865 seq_printf(seq, "%-9s %4u %6d %6d %-3s %6u %-3s %-10s " 1866 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n", 1867 proto->name, 1868 proto->obj_size, 1869 proto->sockets_allocated != NULL ? atomic_read(proto->sockets_allocated) : -1, 1870 proto->memory_allocated != NULL ? atomic_read(proto->memory_allocated) : -1, 1871 proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI", 1872 proto->max_header, 1873 proto->slab == NULL ? "no" : "yes", 1874 module_name(proto->owner), 1875 proto_method_implemented(proto->close), 1876 proto_method_implemented(proto->connect), 1877 proto_method_implemented(proto->disconnect), 1878 proto_method_implemented(proto->accept), 1879 proto_method_implemented(proto->ioctl), 1880 proto_method_implemented(proto->init), 1881 proto_method_implemented(proto->destroy), 1882 proto_method_implemented(proto->shutdown), 1883 proto_method_implemented(proto->setsockopt), 1884 proto_method_implemented(proto->getsockopt), 1885 proto_method_implemented(proto->sendmsg), 1886 proto_method_implemented(proto->recvmsg), 1887 proto_method_implemented(proto->sendpage), 1888 proto_method_implemented(proto->bind), 1889 proto_method_implemented(proto->backlog_rcv), 1890 proto_method_implemented(proto->hash), 1891 proto_method_implemented(proto->unhash), 1892 proto_method_implemented(proto->get_port), 1893 proto_method_implemented(proto->enter_memory_pressure)); 1894} 1895 1896static int proto_seq_show(struct seq_file *seq, void *v) 1897{ 1898 if (v == SEQ_START_TOKEN) 1899 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s", 1900 "protocol", 1901 "size", 1902 "sockets", 1903 "memory", 1904 "press", 1905 "maxhdr", 1906 "slab", 1907 "module", 1908 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n"); 1909 else 1910 proto_seq_printf(seq, v); 1911 return 0; 1912} 1913 1914static struct seq_operations proto_seq_ops = { 1915 .start = proto_seq_start, 1916 .next = proto_seq_next, 1917 .stop = proto_seq_stop, 1918 .show = proto_seq_show, 1919}; 1920 1921static int proto_seq_open(struct inode *inode, struct file *file) 1922{ 1923 return seq_open(file, &proto_seq_ops); 1924} 1925 1926static struct file_operations proto_seq_fops = { 1927 .owner = THIS_MODULE, 1928 .open = proto_seq_open, 1929 .read = seq_read, 1930 .llseek = seq_lseek, 1931 .release = seq_release, 1932}; 1933 1934static int __init proto_init(void) 1935{ 1936 /* register /proc/net/protocols */ 1937 return proc_net_fops_create("protocols", S_IRUGO, &proto_seq_fops) == NULL ? -ENOBUFS : 0; 1938} 1939 1940subsys_initcall(proto_init); 1941 1942#endif /* PROC_FS */ 1943 1944EXPORT_SYMBOL(sk_alloc); 1945EXPORT_SYMBOL(sk_free); 1946EXPORT_SYMBOL(sk_send_sigurg); 1947EXPORT_SYMBOL(sock_alloc_send_skb); 1948EXPORT_SYMBOL(sock_init_data); 1949EXPORT_SYMBOL(sock_kfree_s); 1950EXPORT_SYMBOL(sock_kmalloc); 1951EXPORT_SYMBOL(sock_no_accept); 1952EXPORT_SYMBOL(sock_no_bind); 1953EXPORT_SYMBOL(sock_no_connect); 1954EXPORT_SYMBOL(sock_no_getname); 1955EXPORT_SYMBOL(sock_no_getsockopt); 1956EXPORT_SYMBOL(sock_no_ioctl); 1957EXPORT_SYMBOL(sock_no_listen); 1958EXPORT_SYMBOL(sock_no_mmap); 1959EXPORT_SYMBOL(sock_no_poll); 1960EXPORT_SYMBOL(sock_no_recvmsg); 1961EXPORT_SYMBOL(sock_no_sendmsg); 1962EXPORT_SYMBOL(sock_no_sendpage); 1963EXPORT_SYMBOL(sock_no_setsockopt); 1964EXPORT_SYMBOL(sock_no_shutdown); 1965EXPORT_SYMBOL(sock_no_socketpair); 1966EXPORT_SYMBOL(sock_rfree); 1967EXPORT_SYMBOL(sock_setsockopt); 1968EXPORT_SYMBOL(sock_wfree); 1969EXPORT_SYMBOL(sock_wmalloc); 1970EXPORT_SYMBOL(sock_i_uid); 1971EXPORT_SYMBOL(sock_i_ino); 1972EXPORT_SYMBOL(sysctl_optmem_max); 1973#ifdef CONFIG_SYSCTL 1974EXPORT_SYMBOL(sysctl_rmem_max); 1975EXPORT_SYMBOL(sysctl_wmem_max); 1976#endif 1977