sock.c revision e71a4783aae059931f63b2d4e7013e36529badef
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#include <linux/highmem.h> 115 116#include <asm/uaccess.h> 117#include <asm/system.h> 118 119#include <linux/netdevice.h> 120#include <net/protocol.h> 121#include <linux/skbuff.h> 122#include <net/request_sock.h> 123#include <net/sock.h> 124#include <net/xfrm.h> 125#include <linux/ipsec.h> 126 127#include <linux/filter.h> 128 129#ifdef CONFIG_INET 130#include <net/tcp.h> 131#endif 132 133/* 134 * Each address family might have different locking rules, so we have 135 * one slock key per address family: 136 */ 137static struct lock_class_key af_family_keys[AF_MAX]; 138static struct lock_class_key af_family_slock_keys[AF_MAX]; 139 140#ifdef CONFIG_DEBUG_LOCK_ALLOC 141/* 142 * Make lock validator output more readable. (we pre-construct these 143 * strings build-time, so that runtime initialization of socket 144 * locks is fast): 145 */ 146static const char *af_family_key_strings[AF_MAX+1] = { 147 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" , 148 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK", 149 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" , 150 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" , 151 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" , 152 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" , 153 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" , 154 "sk_lock-21" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" , 155 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" , 156 "sk_lock-27" , "sk_lock-28" , "sk_lock-29" , 157 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-AF_MAX" 158}; 159static const char *af_family_slock_key_strings[AF_MAX+1] = { 160 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" , 161 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK", 162 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" , 163 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" , 164 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" , 165 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" , 166 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" , 167 "slock-21" , "slock-AF_SNA" , "slock-AF_IRDA" , 168 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" , 169 "slock-27" , "slock-28" , "slock-29" , 170 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_MAX" 171}; 172#endif 173 174/* 175 * sk_callback_lock locking rules are per-address-family, 176 * so split the lock classes by using a per-AF key: 177 */ 178static struct lock_class_key af_callback_keys[AF_MAX]; 179 180/* Take into consideration the size of the struct sk_buff overhead in the 181 * determination of these values, since that is non-constant across 182 * platforms. This makes socket queueing behavior and performance 183 * not depend upon such differences. 184 */ 185#define _SK_MEM_PACKETS 256 186#define _SK_MEM_OVERHEAD (sizeof(struct sk_buff) + 256) 187#define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 188#define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 189 190/* Run time adjustable parameters. */ 191__u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX; 192__u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX; 193__u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX; 194__u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX; 195 196/* Maximal space eaten by iovec or ancilliary data plus some space */ 197int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512); 198 199static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen) 200{ 201 struct timeval tv; 202 203 if (optlen < sizeof(tv)) 204 return -EINVAL; 205 if (copy_from_user(&tv, optval, sizeof(tv))) 206 return -EFAULT; 207 208 *timeo_p = MAX_SCHEDULE_TIMEOUT; 209 if (tv.tv_sec == 0 && tv.tv_usec == 0) 210 return 0; 211 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1)) 212 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ); 213 return 0; 214} 215 216static void sock_warn_obsolete_bsdism(const char *name) 217{ 218 static int warned; 219 static char warncomm[TASK_COMM_LEN]; 220 if (strcmp(warncomm, current->comm) && warned < 5) { 221 strcpy(warncomm, current->comm); 222 printk(KERN_WARNING "process `%s' is using obsolete " 223 "%s SO_BSDCOMPAT\n", warncomm, name); 224 warned++; 225 } 226} 227 228static void sock_disable_timestamp(struct sock *sk) 229{ 230 if (sock_flag(sk, SOCK_TIMESTAMP)) { 231 sock_reset_flag(sk, SOCK_TIMESTAMP); 232 net_disable_timestamp(); 233 } 234} 235 236 237int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 238{ 239 int err = 0; 240 int skb_len; 241 242 /* Cast skb->rcvbuf to unsigned... It's pointless, but reduces 243 number of warnings when compiling with -W --ANK 244 */ 245 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >= 246 (unsigned)sk->sk_rcvbuf) { 247 err = -ENOMEM; 248 goto out; 249 } 250 251 err = sk_filter(sk, skb); 252 if (err) 253 goto out; 254 255 skb->dev = NULL; 256 skb_set_owner_r(skb, sk); 257 258 /* Cache the SKB length before we tack it onto the receive 259 * queue. Once it is added it no longer belongs to us and 260 * may be freed by other threads of control pulling packets 261 * from the queue. 262 */ 263 skb_len = skb->len; 264 265 skb_queue_tail(&sk->sk_receive_queue, skb); 266 267 if (!sock_flag(sk, SOCK_DEAD)) 268 sk->sk_data_ready(sk, skb_len); 269out: 270 return err; 271} 272EXPORT_SYMBOL(sock_queue_rcv_skb); 273 274int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested) 275{ 276 int rc = NET_RX_SUCCESS; 277 278 if (sk_filter(sk, skb)) 279 goto discard_and_relse; 280 281 skb->dev = NULL; 282 283 if (nested) 284 bh_lock_sock_nested(sk); 285 else 286 bh_lock_sock(sk); 287 if (!sock_owned_by_user(sk)) { 288 /* 289 * trylock + unlock semantics: 290 */ 291 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_); 292 293 rc = sk->sk_backlog_rcv(sk, skb); 294 295 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_); 296 } else 297 sk_add_backlog(sk, skb); 298 bh_unlock_sock(sk); 299out: 300 sock_put(sk); 301 return rc; 302discard_and_relse: 303 kfree_skb(skb); 304 goto out; 305} 306EXPORT_SYMBOL(sk_receive_skb); 307 308struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie) 309{ 310 struct dst_entry *dst = sk->sk_dst_cache; 311 312 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) { 313 sk->sk_dst_cache = NULL; 314 dst_release(dst); 315 return NULL; 316 } 317 318 return dst; 319} 320EXPORT_SYMBOL(__sk_dst_check); 321 322struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie) 323{ 324 struct dst_entry *dst = sk_dst_get(sk); 325 326 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) { 327 sk_dst_reset(sk); 328 dst_release(dst); 329 return NULL; 330 } 331 332 return dst; 333} 334EXPORT_SYMBOL(sk_dst_check); 335 336/* 337 * This is meant for all protocols to use and covers goings on 338 * at the socket level. Everything here is generic. 339 */ 340 341int sock_setsockopt(struct socket *sock, int level, int optname, 342 char __user *optval, int optlen) 343{ 344 struct sock *sk=sock->sk; 345 struct sk_filter *filter; 346 int val; 347 int valbool; 348 struct linger ling; 349 int ret = 0; 350 351 /* 352 * Options without arguments 353 */ 354 355#ifdef SO_DONTLINGER /* Compatibility item... */ 356 if (optname == SO_DONTLINGER) { 357 lock_sock(sk); 358 sock_reset_flag(sk, SOCK_LINGER); 359 release_sock(sk); 360 return 0; 361 } 362#endif 363 364 if (optlen < sizeof(int)) 365 return -EINVAL; 366 367 if (get_user(val, (int __user *)optval)) 368 return -EFAULT; 369 370 valbool = val?1:0; 371 372 lock_sock(sk); 373 374 switch(optname) { 375 case SO_DEBUG: 376 if (val && !capable(CAP_NET_ADMIN)) { 377 ret = -EACCES; 378 } 379 else if (valbool) 380 sock_set_flag(sk, SOCK_DBG); 381 else 382 sock_reset_flag(sk, SOCK_DBG); 383 break; 384 case SO_REUSEADDR: 385 sk->sk_reuse = valbool; 386 break; 387 case SO_TYPE: 388 case SO_ERROR: 389 ret = -ENOPROTOOPT; 390 break; 391 case SO_DONTROUTE: 392 if (valbool) 393 sock_set_flag(sk, SOCK_LOCALROUTE); 394 else 395 sock_reset_flag(sk, SOCK_LOCALROUTE); 396 break; 397 case SO_BROADCAST: 398 sock_valbool_flag(sk, SOCK_BROADCAST, valbool); 399 break; 400 case SO_SNDBUF: 401 /* Don't error on this BSD doesn't and if you think 402 about it this is right. Otherwise apps have to 403 play 'guess the biggest size' games. RCVBUF/SNDBUF 404 are treated in BSD as hints */ 405 406 if (val > sysctl_wmem_max) 407 val = sysctl_wmem_max; 408set_sndbuf: 409 sk->sk_userlocks |= SOCK_SNDBUF_LOCK; 410 if ((val * 2) < SOCK_MIN_SNDBUF) 411 sk->sk_sndbuf = SOCK_MIN_SNDBUF; 412 else 413 sk->sk_sndbuf = val * 2; 414 415 /* 416 * Wake up sending tasks if we 417 * upped the value. 418 */ 419 sk->sk_write_space(sk); 420 break; 421 422 case SO_SNDBUFFORCE: 423 if (!capable(CAP_NET_ADMIN)) { 424 ret = -EPERM; 425 break; 426 } 427 goto set_sndbuf; 428 429 case SO_RCVBUF: 430 /* Don't error on this BSD doesn't and if you think 431 about it this is right. Otherwise apps have to 432 play 'guess the biggest size' games. RCVBUF/SNDBUF 433 are treated in BSD as hints */ 434 435 if (val > sysctl_rmem_max) 436 val = sysctl_rmem_max; 437set_rcvbuf: 438 sk->sk_userlocks |= SOCK_RCVBUF_LOCK; 439 /* 440 * We double it on the way in to account for 441 * "struct sk_buff" etc. overhead. Applications 442 * assume that the SO_RCVBUF setting they make will 443 * allow that much actual data to be received on that 444 * socket. 445 * 446 * Applications are unaware that "struct sk_buff" and 447 * other overheads allocate from the receive buffer 448 * during socket buffer allocation. 449 * 450 * And after considering the possible alternatives, 451 * returning the value we actually used in getsockopt 452 * is the most desirable behavior. 453 */ 454 if ((val * 2) < SOCK_MIN_RCVBUF) 455 sk->sk_rcvbuf = SOCK_MIN_RCVBUF; 456 else 457 sk->sk_rcvbuf = val * 2; 458 break; 459 460 case SO_RCVBUFFORCE: 461 if (!capable(CAP_NET_ADMIN)) { 462 ret = -EPERM; 463 break; 464 } 465 goto set_rcvbuf; 466 467 case SO_KEEPALIVE: 468#ifdef CONFIG_INET 469 if (sk->sk_protocol == IPPROTO_TCP) 470 tcp_set_keepalive(sk, valbool); 471#endif 472 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool); 473 break; 474 475 case SO_OOBINLINE: 476 sock_valbool_flag(sk, SOCK_URGINLINE, valbool); 477 break; 478 479 case SO_NO_CHECK: 480 sk->sk_no_check = valbool; 481 break; 482 483 case SO_PRIORITY: 484 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN)) 485 sk->sk_priority = val; 486 else 487 ret = -EPERM; 488 break; 489 490 case SO_LINGER: 491 if (optlen < sizeof(ling)) { 492 ret = -EINVAL; /* 1003.1g */ 493 break; 494 } 495 if (copy_from_user(&ling,optval,sizeof(ling))) { 496 ret = -EFAULT; 497 break; 498 } 499 if (!ling.l_onoff) 500 sock_reset_flag(sk, SOCK_LINGER); 501 else { 502#if (BITS_PER_LONG == 32) 503 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ) 504 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT; 505 else 506#endif 507 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ; 508 sock_set_flag(sk, SOCK_LINGER); 509 } 510 break; 511 512 case SO_BSDCOMPAT: 513 sock_warn_obsolete_bsdism("setsockopt"); 514 break; 515 516 case SO_PASSCRED: 517 if (valbool) 518 set_bit(SOCK_PASSCRED, &sock->flags); 519 else 520 clear_bit(SOCK_PASSCRED, &sock->flags); 521 break; 522 523 case SO_TIMESTAMP: 524 if (valbool) { 525 sock_set_flag(sk, SOCK_RCVTSTAMP); 526 sock_enable_timestamp(sk); 527 } else 528 sock_reset_flag(sk, SOCK_RCVTSTAMP); 529 break; 530 531 case SO_RCVLOWAT: 532 if (val < 0) 533 val = INT_MAX; 534 sk->sk_rcvlowat = val ? : 1; 535 break; 536 537 case SO_RCVTIMEO: 538 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen); 539 break; 540 541 case SO_SNDTIMEO: 542 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen); 543 break; 544 545#ifdef CONFIG_NETDEVICES 546 case SO_BINDTODEVICE: 547 { 548 char devname[IFNAMSIZ]; 549 550 /* Sorry... */ 551 if (!capable(CAP_NET_RAW)) { 552 ret = -EPERM; 553 break; 554 } 555 556 /* Bind this socket to a particular device like "eth0", 557 * as specified in the passed interface name. If the 558 * name is "" or the option length is zero the socket 559 * is not bound. 560 */ 561 562 if (!valbool) { 563 sk->sk_bound_dev_if = 0; 564 } else { 565 if (optlen > IFNAMSIZ - 1) 566 optlen = IFNAMSIZ - 1; 567 memset(devname, 0, sizeof(devname)); 568 if (copy_from_user(devname, optval, optlen)) { 569 ret = -EFAULT; 570 break; 571 } 572 573 /* Remove any cached route for this socket. */ 574 sk_dst_reset(sk); 575 576 if (devname[0] == '\0') { 577 sk->sk_bound_dev_if = 0; 578 } else { 579 struct net_device *dev = dev_get_by_name(devname); 580 if (!dev) { 581 ret = -ENODEV; 582 break; 583 } 584 sk->sk_bound_dev_if = dev->ifindex; 585 dev_put(dev); 586 } 587 } 588 break; 589 } 590#endif 591 592 593 case SO_ATTACH_FILTER: 594 ret = -EINVAL; 595 if (optlen == sizeof(struct sock_fprog)) { 596 struct sock_fprog fprog; 597 598 ret = -EFAULT; 599 if (copy_from_user(&fprog, optval, sizeof(fprog))) 600 break; 601 602 ret = sk_attach_filter(&fprog, sk); 603 } 604 break; 605 606 case SO_DETACH_FILTER: 607 rcu_read_lock_bh(); 608 filter = rcu_dereference(sk->sk_filter); 609 if (filter) { 610 rcu_assign_pointer(sk->sk_filter, NULL); 611 sk_filter_release(sk, filter); 612 rcu_read_unlock_bh(); 613 break; 614 } 615 rcu_read_unlock_bh(); 616 ret = -ENONET; 617 break; 618 619 case SO_PASSSEC: 620 if (valbool) 621 set_bit(SOCK_PASSSEC, &sock->flags); 622 else 623 clear_bit(SOCK_PASSSEC, &sock->flags); 624 break; 625 626 /* We implement the SO_SNDLOWAT etc to 627 not be settable (1003.1g 5.3) */ 628 default: 629 ret = -ENOPROTOOPT; 630 break; 631 } 632 release_sock(sk); 633 return ret; 634} 635 636 637int sock_getsockopt(struct socket *sock, int level, int optname, 638 char __user *optval, int __user *optlen) 639{ 640 struct sock *sk = sock->sk; 641 642 union { 643 int val; 644 struct linger ling; 645 struct timeval tm; 646 } v; 647 648 unsigned int lv = sizeof(int); 649 int len; 650 651 if (get_user(len, optlen)) 652 return -EFAULT; 653 if (len < 0) 654 return -EINVAL; 655 656 switch(optname) { 657 case SO_DEBUG: 658 v.val = sock_flag(sk, SOCK_DBG); 659 break; 660 661 case SO_DONTROUTE: 662 v.val = sock_flag(sk, SOCK_LOCALROUTE); 663 break; 664 665 case SO_BROADCAST: 666 v.val = !!sock_flag(sk, SOCK_BROADCAST); 667 break; 668 669 case SO_SNDBUF: 670 v.val = sk->sk_sndbuf; 671 break; 672 673 case SO_RCVBUF: 674 v.val = sk->sk_rcvbuf; 675 break; 676 677 case SO_REUSEADDR: 678 v.val = sk->sk_reuse; 679 break; 680 681 case SO_KEEPALIVE: 682 v.val = !!sock_flag(sk, SOCK_KEEPOPEN); 683 break; 684 685 case SO_TYPE: 686 v.val = sk->sk_type; 687 break; 688 689 case SO_ERROR: 690 v.val = -sock_error(sk); 691 if (v.val==0) 692 v.val = xchg(&sk->sk_err_soft, 0); 693 break; 694 695 case SO_OOBINLINE: 696 v.val = !!sock_flag(sk, SOCK_URGINLINE); 697 break; 698 699 case SO_NO_CHECK: 700 v.val = sk->sk_no_check; 701 break; 702 703 case SO_PRIORITY: 704 v.val = sk->sk_priority; 705 break; 706 707 case SO_LINGER: 708 lv = sizeof(v.ling); 709 v.ling.l_onoff = !!sock_flag(sk, SOCK_LINGER); 710 v.ling.l_linger = sk->sk_lingertime / HZ; 711 break; 712 713 case SO_BSDCOMPAT: 714 sock_warn_obsolete_bsdism("getsockopt"); 715 break; 716 717 case SO_TIMESTAMP: 718 v.val = sock_flag(sk, SOCK_RCVTSTAMP); 719 break; 720 721 case SO_RCVTIMEO: 722 lv=sizeof(struct timeval); 723 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) { 724 v.tm.tv_sec = 0; 725 v.tm.tv_usec = 0; 726 } else { 727 v.tm.tv_sec = sk->sk_rcvtimeo / HZ; 728 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ; 729 } 730 break; 731 732 case SO_SNDTIMEO: 733 lv=sizeof(struct timeval); 734 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) { 735 v.tm.tv_sec = 0; 736 v.tm.tv_usec = 0; 737 } else { 738 v.tm.tv_sec = sk->sk_sndtimeo / HZ; 739 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ; 740 } 741 break; 742 743 case SO_RCVLOWAT: 744 v.val = sk->sk_rcvlowat; 745 break; 746 747 case SO_SNDLOWAT: 748 v.val=1; 749 break; 750 751 case SO_PASSCRED: 752 v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0; 753 break; 754 755 case SO_PEERCRED: 756 if (len > sizeof(sk->sk_peercred)) 757 len = sizeof(sk->sk_peercred); 758 if (copy_to_user(optval, &sk->sk_peercred, len)) 759 return -EFAULT; 760 goto lenout; 761 762 case SO_PEERNAME: 763 { 764 char address[128]; 765 766 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2)) 767 return -ENOTCONN; 768 if (lv < len) 769 return -EINVAL; 770 if (copy_to_user(optval, address, len)) 771 return -EFAULT; 772 goto lenout; 773 } 774 775 /* Dubious BSD thing... Probably nobody even uses it, but 776 * the UNIX standard wants it for whatever reason... -DaveM 777 */ 778 case SO_ACCEPTCONN: 779 v.val = sk->sk_state == TCP_LISTEN; 780 break; 781 782 case SO_PASSSEC: 783 v.val = test_bit(SOCK_PASSSEC, &sock->flags) ? 1 : 0; 784 break; 785 786 case SO_PEERSEC: 787 return security_socket_getpeersec_stream(sock, optval, optlen, len); 788 789 default: 790 return -ENOPROTOOPT; 791 } 792 793 if (len > lv) 794 len = lv; 795 if (copy_to_user(optval, &v, len)) 796 return -EFAULT; 797lenout: 798 if (put_user(len, optlen)) 799 return -EFAULT; 800 return 0; 801} 802 803/* 804 * Initialize an sk_lock. 805 * 806 * (We also register the sk_lock with the lock validator.) 807 */ 808static inline void sock_lock_init(struct sock *sk) 809{ 810 sock_lock_init_class_and_name(sk, 811 af_family_slock_key_strings[sk->sk_family], 812 af_family_slock_keys + sk->sk_family, 813 af_family_key_strings[sk->sk_family], 814 af_family_keys + sk->sk_family); 815} 816 817/** 818 * sk_alloc - All socket objects are allocated here 819 * @family: protocol family 820 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 821 * @prot: struct proto associated with this new sock instance 822 * @zero_it: if we should zero the newly allocated sock 823 */ 824struct sock *sk_alloc(int family, gfp_t priority, 825 struct proto *prot, int zero_it) 826{ 827 struct sock *sk = NULL; 828 struct kmem_cache *slab = prot->slab; 829 830 if (slab != NULL) 831 sk = kmem_cache_alloc(slab, priority); 832 else 833 sk = kmalloc(prot->obj_size, priority); 834 835 if (sk) { 836 if (zero_it) { 837 memset(sk, 0, prot->obj_size); 838 sk->sk_family = family; 839 /* 840 * See comment in struct sock definition to understand 841 * why we need sk_prot_creator -acme 842 */ 843 sk->sk_prot = sk->sk_prot_creator = prot; 844 sock_lock_init(sk); 845 } 846 847 if (security_sk_alloc(sk, family, priority)) 848 goto out_free; 849 850 if (!try_module_get(prot->owner)) 851 goto out_free; 852 } 853 return sk; 854 855out_free: 856 if (slab != NULL) 857 kmem_cache_free(slab, sk); 858 else 859 kfree(sk); 860 return NULL; 861} 862 863void sk_free(struct sock *sk) 864{ 865 struct sk_filter *filter; 866 struct module *owner = sk->sk_prot_creator->owner; 867 868 if (sk->sk_destruct) 869 sk->sk_destruct(sk); 870 871 filter = rcu_dereference(sk->sk_filter); 872 if (filter) { 873 sk_filter_release(sk, filter); 874 rcu_assign_pointer(sk->sk_filter, NULL); 875 } 876 877 sock_disable_timestamp(sk); 878 879 if (atomic_read(&sk->sk_omem_alloc)) 880 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n", 881 __FUNCTION__, atomic_read(&sk->sk_omem_alloc)); 882 883 security_sk_free(sk); 884 if (sk->sk_prot_creator->slab != NULL) 885 kmem_cache_free(sk->sk_prot_creator->slab, sk); 886 else 887 kfree(sk); 888 module_put(owner); 889} 890 891struct sock *sk_clone(const struct sock *sk, const gfp_t priority) 892{ 893 struct sock *newsk = sk_alloc(sk->sk_family, priority, sk->sk_prot, 0); 894 895 if (newsk != NULL) { 896 struct sk_filter *filter; 897 898 sock_copy(newsk, sk); 899 900 /* SANITY */ 901 sk_node_init(&newsk->sk_node); 902 sock_lock_init(newsk); 903 bh_lock_sock(newsk); 904 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL; 905 906 atomic_set(&newsk->sk_rmem_alloc, 0); 907 atomic_set(&newsk->sk_wmem_alloc, 0); 908 atomic_set(&newsk->sk_omem_alloc, 0); 909 skb_queue_head_init(&newsk->sk_receive_queue); 910 skb_queue_head_init(&newsk->sk_write_queue); 911#ifdef CONFIG_NET_DMA 912 skb_queue_head_init(&newsk->sk_async_wait_queue); 913#endif 914 915 rwlock_init(&newsk->sk_dst_lock); 916 rwlock_init(&newsk->sk_callback_lock); 917 lockdep_set_class(&newsk->sk_callback_lock, 918 af_callback_keys + newsk->sk_family); 919 920 newsk->sk_dst_cache = NULL; 921 newsk->sk_wmem_queued = 0; 922 newsk->sk_forward_alloc = 0; 923 newsk->sk_send_head = NULL; 924 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK; 925 926 sock_reset_flag(newsk, SOCK_DONE); 927 skb_queue_head_init(&newsk->sk_error_queue); 928 929 filter = newsk->sk_filter; 930 if (filter != NULL) 931 sk_filter_charge(newsk, filter); 932 933 if (unlikely(xfrm_sk_clone_policy(newsk))) { 934 /* It is still raw copy of parent, so invalidate 935 * destructor and make plain sk_free() */ 936 newsk->sk_destruct = NULL; 937 sk_free(newsk); 938 newsk = NULL; 939 goto out; 940 } 941 942 newsk->sk_err = 0; 943 newsk->sk_priority = 0; 944 atomic_set(&newsk->sk_refcnt, 2); 945 946 /* 947 * Increment the counter in the same struct proto as the master 948 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that 949 * is the same as sk->sk_prot->socks, as this field was copied 950 * with memcpy). 951 * 952 * This _changes_ the previous behaviour, where 953 * tcp_create_openreq_child always was incrementing the 954 * equivalent to tcp_prot->socks (inet_sock_nr), so this have 955 * to be taken into account in all callers. -acme 956 */ 957 sk_refcnt_debug_inc(newsk); 958 newsk->sk_socket = NULL; 959 newsk->sk_sleep = NULL; 960 961 if (newsk->sk_prot->sockets_allocated) 962 atomic_inc(newsk->sk_prot->sockets_allocated); 963 } 964out: 965 return newsk; 966} 967 968EXPORT_SYMBOL_GPL(sk_clone); 969 970void __init sk_init(void) 971{ 972 if (num_physpages <= 4096) { 973 sysctl_wmem_max = 32767; 974 sysctl_rmem_max = 32767; 975 sysctl_wmem_default = 32767; 976 sysctl_rmem_default = 32767; 977 } else if (num_physpages >= 131072) { 978 sysctl_wmem_max = 131071; 979 sysctl_rmem_max = 131071; 980 } 981} 982 983/* 984 * Simple resource managers for sockets. 985 */ 986 987 988/* 989 * Write buffer destructor automatically called from kfree_skb. 990 */ 991void sock_wfree(struct sk_buff *skb) 992{ 993 struct sock *sk = skb->sk; 994 995 /* In case it might be waiting for more memory. */ 996 atomic_sub(skb->truesize, &sk->sk_wmem_alloc); 997 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) 998 sk->sk_write_space(sk); 999 sock_put(sk); 1000} 1001 1002/* 1003 * Read buffer destructor automatically called from kfree_skb. 1004 */ 1005void sock_rfree(struct sk_buff *skb) 1006{ 1007 struct sock *sk = skb->sk; 1008 1009 atomic_sub(skb->truesize, &sk->sk_rmem_alloc); 1010} 1011 1012 1013int sock_i_uid(struct sock *sk) 1014{ 1015 int uid; 1016 1017 read_lock(&sk->sk_callback_lock); 1018 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0; 1019 read_unlock(&sk->sk_callback_lock); 1020 return uid; 1021} 1022 1023unsigned long sock_i_ino(struct sock *sk) 1024{ 1025 unsigned long ino; 1026 1027 read_lock(&sk->sk_callback_lock); 1028 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0; 1029 read_unlock(&sk->sk_callback_lock); 1030 return ino; 1031} 1032 1033/* 1034 * Allocate a skb from the socket's send buffer. 1035 */ 1036struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 1037 gfp_t priority) 1038{ 1039 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) { 1040 struct sk_buff * skb = alloc_skb(size, priority); 1041 if (skb) { 1042 skb_set_owner_w(skb, sk); 1043 return skb; 1044 } 1045 } 1046 return NULL; 1047} 1048 1049/* 1050 * Allocate a skb from the socket's receive buffer. 1051 */ 1052struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force, 1053 gfp_t priority) 1054{ 1055 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) { 1056 struct sk_buff *skb = alloc_skb(size, priority); 1057 if (skb) { 1058 skb_set_owner_r(skb, sk); 1059 return skb; 1060 } 1061 } 1062 return NULL; 1063} 1064 1065/* 1066 * Allocate a memory block from the socket's option memory buffer. 1067 */ 1068void *sock_kmalloc(struct sock *sk, int size, gfp_t priority) 1069{ 1070 if ((unsigned)size <= sysctl_optmem_max && 1071 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) { 1072 void *mem; 1073 /* First do the add, to avoid the race if kmalloc 1074 * might sleep. 1075 */ 1076 atomic_add(size, &sk->sk_omem_alloc); 1077 mem = kmalloc(size, priority); 1078 if (mem) 1079 return mem; 1080 atomic_sub(size, &sk->sk_omem_alloc); 1081 } 1082 return NULL; 1083} 1084 1085/* 1086 * Free an option memory block. 1087 */ 1088void sock_kfree_s(struct sock *sk, void *mem, int size) 1089{ 1090 kfree(mem); 1091 atomic_sub(size, &sk->sk_omem_alloc); 1092} 1093 1094/* It is almost wait_for_tcp_memory minus release_sock/lock_sock. 1095 I think, these locks should be removed for datagram sockets. 1096 */ 1097static long sock_wait_for_wmem(struct sock * sk, long timeo) 1098{ 1099 DEFINE_WAIT(wait); 1100 1101 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags); 1102 for (;;) { 1103 if (!timeo) 1104 break; 1105 if (signal_pending(current)) 1106 break; 1107 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 1108 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE); 1109 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) 1110 break; 1111 if (sk->sk_shutdown & SEND_SHUTDOWN) 1112 break; 1113 if (sk->sk_err) 1114 break; 1115 timeo = schedule_timeout(timeo); 1116 } 1117 finish_wait(sk->sk_sleep, &wait); 1118 return timeo; 1119} 1120 1121 1122/* 1123 * Generic send/receive buffer handlers 1124 */ 1125 1126static struct sk_buff *sock_alloc_send_pskb(struct sock *sk, 1127 unsigned long header_len, 1128 unsigned long data_len, 1129 int noblock, int *errcode) 1130{ 1131 struct sk_buff *skb; 1132 gfp_t gfp_mask; 1133 long timeo; 1134 int err; 1135 1136 gfp_mask = sk->sk_allocation; 1137 if (gfp_mask & __GFP_WAIT) 1138 gfp_mask |= __GFP_REPEAT; 1139 1140 timeo = sock_sndtimeo(sk, noblock); 1141 while (1) { 1142 err = sock_error(sk); 1143 if (err != 0) 1144 goto failure; 1145 1146 err = -EPIPE; 1147 if (sk->sk_shutdown & SEND_SHUTDOWN) 1148 goto failure; 1149 1150 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) { 1151 skb = alloc_skb(header_len, gfp_mask); 1152 if (skb) { 1153 int npages; 1154 int i; 1155 1156 /* No pages, we're done... */ 1157 if (!data_len) 1158 break; 1159 1160 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT; 1161 skb->truesize += data_len; 1162 skb_shinfo(skb)->nr_frags = npages; 1163 for (i = 0; i < npages; i++) { 1164 struct page *page; 1165 skb_frag_t *frag; 1166 1167 page = alloc_pages(sk->sk_allocation, 0); 1168 if (!page) { 1169 err = -ENOBUFS; 1170 skb_shinfo(skb)->nr_frags = i; 1171 kfree_skb(skb); 1172 goto failure; 1173 } 1174 1175 frag = &skb_shinfo(skb)->frags[i]; 1176 frag->page = page; 1177 frag->page_offset = 0; 1178 frag->size = (data_len >= PAGE_SIZE ? 1179 PAGE_SIZE : 1180 data_len); 1181 data_len -= PAGE_SIZE; 1182 } 1183 1184 /* Full success... */ 1185 break; 1186 } 1187 err = -ENOBUFS; 1188 goto failure; 1189 } 1190 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags); 1191 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 1192 err = -EAGAIN; 1193 if (!timeo) 1194 goto failure; 1195 if (signal_pending(current)) 1196 goto interrupted; 1197 timeo = sock_wait_for_wmem(sk, timeo); 1198 } 1199 1200 skb_set_owner_w(skb, sk); 1201 return skb; 1202 1203interrupted: 1204 err = sock_intr_errno(timeo); 1205failure: 1206 *errcode = err; 1207 return NULL; 1208} 1209 1210struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size, 1211 int noblock, int *errcode) 1212{ 1213 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode); 1214} 1215 1216static void __lock_sock(struct sock *sk) 1217{ 1218 DEFINE_WAIT(wait); 1219 1220 for (;;) { 1221 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait, 1222 TASK_UNINTERRUPTIBLE); 1223 spin_unlock_bh(&sk->sk_lock.slock); 1224 schedule(); 1225 spin_lock_bh(&sk->sk_lock.slock); 1226 if (!sock_owned_by_user(sk)) 1227 break; 1228 } 1229 finish_wait(&sk->sk_lock.wq, &wait); 1230} 1231 1232static void __release_sock(struct sock *sk) 1233{ 1234 struct sk_buff *skb = sk->sk_backlog.head; 1235 1236 do { 1237 sk->sk_backlog.head = sk->sk_backlog.tail = NULL; 1238 bh_unlock_sock(sk); 1239 1240 do { 1241 struct sk_buff *next = skb->next; 1242 1243 skb->next = NULL; 1244 sk->sk_backlog_rcv(sk, skb); 1245 1246 /* 1247 * We are in process context here with softirqs 1248 * disabled, use cond_resched_softirq() to preempt. 1249 * This is safe to do because we've taken the backlog 1250 * queue private: 1251 */ 1252 cond_resched_softirq(); 1253 1254 skb = next; 1255 } while (skb != NULL); 1256 1257 bh_lock_sock(sk); 1258 } while ((skb = sk->sk_backlog.head) != NULL); 1259} 1260 1261/** 1262 * sk_wait_data - wait for data to arrive at sk_receive_queue 1263 * @sk: sock to wait on 1264 * @timeo: for how long 1265 * 1266 * Now socket state including sk->sk_err is changed only under lock, 1267 * hence we may omit checks after joining wait queue. 1268 * We check receive queue before schedule() only as optimization; 1269 * it is very likely that release_sock() added new data. 1270 */ 1271int sk_wait_data(struct sock *sk, long *timeo) 1272{ 1273 int rc; 1274 DEFINE_WAIT(wait); 1275 1276 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE); 1277 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags); 1278 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue)); 1279 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags); 1280 finish_wait(sk->sk_sleep, &wait); 1281 return rc; 1282} 1283 1284EXPORT_SYMBOL(sk_wait_data); 1285 1286/* 1287 * Set of default routines for initialising struct proto_ops when 1288 * the protocol does not support a particular function. In certain 1289 * cases where it makes no sense for a protocol to have a "do nothing" 1290 * function, some default processing is provided. 1291 */ 1292 1293int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len) 1294{ 1295 return -EOPNOTSUPP; 1296} 1297 1298int sock_no_connect(struct socket *sock, struct sockaddr *saddr, 1299 int len, int flags) 1300{ 1301 return -EOPNOTSUPP; 1302} 1303 1304int sock_no_socketpair(struct socket *sock1, struct socket *sock2) 1305{ 1306 return -EOPNOTSUPP; 1307} 1308 1309int sock_no_accept(struct socket *sock, struct socket *newsock, int flags) 1310{ 1311 return -EOPNOTSUPP; 1312} 1313 1314int sock_no_getname(struct socket *sock, struct sockaddr *saddr, 1315 int *len, int peer) 1316{ 1317 return -EOPNOTSUPP; 1318} 1319 1320unsigned int sock_no_poll(struct file * file, struct socket *sock, poll_table *pt) 1321{ 1322 return 0; 1323} 1324 1325int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 1326{ 1327 return -EOPNOTSUPP; 1328} 1329 1330int sock_no_listen(struct socket *sock, int backlog) 1331{ 1332 return -EOPNOTSUPP; 1333} 1334 1335int sock_no_shutdown(struct socket *sock, int how) 1336{ 1337 return -EOPNOTSUPP; 1338} 1339 1340int sock_no_setsockopt(struct socket *sock, int level, int optname, 1341 char __user *optval, int optlen) 1342{ 1343 return -EOPNOTSUPP; 1344} 1345 1346int sock_no_getsockopt(struct socket *sock, int level, int optname, 1347 char __user *optval, int __user *optlen) 1348{ 1349 return -EOPNOTSUPP; 1350} 1351 1352int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m, 1353 size_t len) 1354{ 1355 return -EOPNOTSUPP; 1356} 1357 1358int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m, 1359 size_t len, int flags) 1360{ 1361 return -EOPNOTSUPP; 1362} 1363 1364int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma) 1365{ 1366 /* Mirror missing mmap method error code */ 1367 return -ENODEV; 1368} 1369 1370ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags) 1371{ 1372 ssize_t res; 1373 struct msghdr msg = {.msg_flags = flags}; 1374 struct kvec iov; 1375 char *kaddr = kmap(page); 1376 iov.iov_base = kaddr + offset; 1377 iov.iov_len = size; 1378 res = kernel_sendmsg(sock, &msg, &iov, 1, size); 1379 kunmap(page); 1380 return res; 1381} 1382 1383/* 1384 * Default Socket Callbacks 1385 */ 1386 1387static void sock_def_wakeup(struct sock *sk) 1388{ 1389 read_lock(&sk->sk_callback_lock); 1390 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) 1391 wake_up_interruptible_all(sk->sk_sleep); 1392 read_unlock(&sk->sk_callback_lock); 1393} 1394 1395static void sock_def_error_report(struct sock *sk) 1396{ 1397 read_lock(&sk->sk_callback_lock); 1398 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) 1399 wake_up_interruptible(sk->sk_sleep); 1400 sk_wake_async(sk,0,POLL_ERR); 1401 read_unlock(&sk->sk_callback_lock); 1402} 1403 1404static void sock_def_readable(struct sock *sk, int len) 1405{ 1406 read_lock(&sk->sk_callback_lock); 1407 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) 1408 wake_up_interruptible(sk->sk_sleep); 1409 sk_wake_async(sk,1,POLL_IN); 1410 read_unlock(&sk->sk_callback_lock); 1411} 1412 1413static void sock_def_write_space(struct sock *sk) 1414{ 1415 read_lock(&sk->sk_callback_lock); 1416 1417 /* Do not wake up a writer until he can make "significant" 1418 * progress. --DaveM 1419 */ 1420 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) { 1421 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) 1422 wake_up_interruptible(sk->sk_sleep); 1423 1424 /* Should agree with poll, otherwise some programs break */ 1425 if (sock_writeable(sk)) 1426 sk_wake_async(sk, 2, POLL_OUT); 1427 } 1428 1429 read_unlock(&sk->sk_callback_lock); 1430} 1431 1432static void sock_def_destruct(struct sock *sk) 1433{ 1434 kfree(sk->sk_protinfo); 1435} 1436 1437void sk_send_sigurg(struct sock *sk) 1438{ 1439 if (sk->sk_socket && sk->sk_socket->file) 1440 if (send_sigurg(&sk->sk_socket->file->f_owner)) 1441 sk_wake_async(sk, 3, POLL_PRI); 1442} 1443 1444void sk_reset_timer(struct sock *sk, struct timer_list* timer, 1445 unsigned long expires) 1446{ 1447 if (!mod_timer(timer, expires)) 1448 sock_hold(sk); 1449} 1450 1451EXPORT_SYMBOL(sk_reset_timer); 1452 1453void sk_stop_timer(struct sock *sk, struct timer_list* timer) 1454{ 1455 if (timer_pending(timer) && del_timer(timer)) 1456 __sock_put(sk); 1457} 1458 1459EXPORT_SYMBOL(sk_stop_timer); 1460 1461void sock_init_data(struct socket *sock, struct sock *sk) 1462{ 1463 skb_queue_head_init(&sk->sk_receive_queue); 1464 skb_queue_head_init(&sk->sk_write_queue); 1465 skb_queue_head_init(&sk->sk_error_queue); 1466#ifdef CONFIG_NET_DMA 1467 skb_queue_head_init(&sk->sk_async_wait_queue); 1468#endif 1469 1470 sk->sk_send_head = NULL; 1471 1472 init_timer(&sk->sk_timer); 1473 1474 sk->sk_allocation = GFP_KERNEL; 1475 sk->sk_rcvbuf = sysctl_rmem_default; 1476 sk->sk_sndbuf = sysctl_wmem_default; 1477 sk->sk_state = TCP_CLOSE; 1478 sk->sk_socket = sock; 1479 1480 sock_set_flag(sk, SOCK_ZAPPED); 1481 1482 if (sock) { 1483 sk->sk_type = sock->type; 1484 sk->sk_sleep = &sock->wait; 1485 sock->sk = sk; 1486 } else 1487 sk->sk_sleep = NULL; 1488 1489 rwlock_init(&sk->sk_dst_lock); 1490 rwlock_init(&sk->sk_callback_lock); 1491 lockdep_set_class(&sk->sk_callback_lock, 1492 af_callback_keys + sk->sk_family); 1493 1494 sk->sk_state_change = sock_def_wakeup; 1495 sk->sk_data_ready = sock_def_readable; 1496 sk->sk_write_space = sock_def_write_space; 1497 sk->sk_error_report = sock_def_error_report; 1498 sk->sk_destruct = sock_def_destruct; 1499 1500 sk->sk_sndmsg_page = NULL; 1501 sk->sk_sndmsg_off = 0; 1502 1503 sk->sk_peercred.pid = 0; 1504 sk->sk_peercred.uid = -1; 1505 sk->sk_peercred.gid = -1; 1506 sk->sk_write_pending = 0; 1507 sk->sk_rcvlowat = 1; 1508 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; 1509 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT; 1510 1511 sk->sk_stamp = ktime_set(-1L, -1L); 1512 1513 atomic_set(&sk->sk_refcnt, 1); 1514} 1515 1516void fastcall lock_sock_nested(struct sock *sk, int subclass) 1517{ 1518 might_sleep(); 1519 spin_lock_bh(&sk->sk_lock.slock); 1520 if (sk->sk_lock.owner) 1521 __lock_sock(sk); 1522 sk->sk_lock.owner = (void *)1; 1523 spin_unlock(&sk->sk_lock.slock); 1524 /* 1525 * The sk_lock has mutex_lock() semantics here: 1526 */ 1527 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_); 1528 local_bh_enable(); 1529} 1530 1531EXPORT_SYMBOL(lock_sock_nested); 1532 1533void fastcall release_sock(struct sock *sk) 1534{ 1535 /* 1536 * The sk_lock has mutex_unlock() semantics: 1537 */ 1538 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_); 1539 1540 spin_lock_bh(&sk->sk_lock.slock); 1541 if (sk->sk_backlog.tail) 1542 __release_sock(sk); 1543 sk->sk_lock.owner = NULL; 1544 if (waitqueue_active(&sk->sk_lock.wq)) 1545 wake_up(&sk->sk_lock.wq); 1546 spin_unlock_bh(&sk->sk_lock.slock); 1547} 1548EXPORT_SYMBOL(release_sock); 1549 1550int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp) 1551{ 1552 struct timeval tv; 1553 if (!sock_flag(sk, SOCK_TIMESTAMP)) 1554 sock_enable_timestamp(sk); 1555 tv = ktime_to_timeval(sk->sk_stamp); 1556 if (tv.tv_sec == -1) 1557 return -ENOENT; 1558 if (tv.tv_sec == 0) { 1559 sk->sk_stamp = ktime_get_real(); 1560 tv = ktime_to_timeval(sk->sk_stamp); 1561 } 1562 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0; 1563} 1564EXPORT_SYMBOL(sock_get_timestamp); 1565 1566int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp) 1567{ 1568 struct timespec ts; 1569 if (!sock_flag(sk, SOCK_TIMESTAMP)) 1570 sock_enable_timestamp(sk); 1571 ts = ktime_to_timespec(sk->sk_stamp); 1572 if (ts.tv_sec == -1) 1573 return -ENOENT; 1574 if (ts.tv_sec == 0) { 1575 sk->sk_stamp = ktime_get_real(); 1576 ts = ktime_to_timespec(sk->sk_stamp); 1577 } 1578 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0; 1579} 1580EXPORT_SYMBOL(sock_get_timestampns); 1581 1582void sock_enable_timestamp(struct sock *sk) 1583{ 1584 if (!sock_flag(sk, SOCK_TIMESTAMP)) { 1585 sock_set_flag(sk, SOCK_TIMESTAMP); 1586 net_enable_timestamp(); 1587 } 1588} 1589EXPORT_SYMBOL(sock_enable_timestamp); 1590 1591/* 1592 * Get a socket option on an socket. 1593 * 1594 * FIX: POSIX 1003.1g is very ambiguous here. It states that 1595 * asynchronous errors should be reported by getsockopt. We assume 1596 * this means if you specify SO_ERROR (otherwise whats the point of it). 1597 */ 1598int sock_common_getsockopt(struct socket *sock, int level, int optname, 1599 char __user *optval, int __user *optlen) 1600{ 1601 struct sock *sk = sock->sk; 1602 1603 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen); 1604} 1605 1606EXPORT_SYMBOL(sock_common_getsockopt); 1607 1608#ifdef CONFIG_COMPAT 1609int compat_sock_common_getsockopt(struct socket *sock, int level, int optname, 1610 char __user *optval, int __user *optlen) 1611{ 1612 struct sock *sk = sock->sk; 1613 1614 if (sk->sk_prot->compat_getsockopt != NULL) 1615 return sk->sk_prot->compat_getsockopt(sk, level, optname, 1616 optval, optlen); 1617 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen); 1618} 1619EXPORT_SYMBOL(compat_sock_common_getsockopt); 1620#endif 1621 1622int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock, 1623 struct msghdr *msg, size_t size, int flags) 1624{ 1625 struct sock *sk = sock->sk; 1626 int addr_len = 0; 1627 int err; 1628 1629 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT, 1630 flags & ~MSG_DONTWAIT, &addr_len); 1631 if (err >= 0) 1632 msg->msg_namelen = addr_len; 1633 return err; 1634} 1635 1636EXPORT_SYMBOL(sock_common_recvmsg); 1637 1638/* 1639 * Set socket options on an inet socket. 1640 */ 1641int sock_common_setsockopt(struct socket *sock, int level, int optname, 1642 char __user *optval, int optlen) 1643{ 1644 struct sock *sk = sock->sk; 1645 1646 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen); 1647} 1648 1649EXPORT_SYMBOL(sock_common_setsockopt); 1650 1651#ifdef CONFIG_COMPAT 1652int compat_sock_common_setsockopt(struct socket *sock, int level, int optname, 1653 char __user *optval, int optlen) 1654{ 1655 struct sock *sk = sock->sk; 1656 1657 if (sk->sk_prot->compat_setsockopt != NULL) 1658 return sk->sk_prot->compat_setsockopt(sk, level, optname, 1659 optval, optlen); 1660 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen); 1661} 1662EXPORT_SYMBOL(compat_sock_common_setsockopt); 1663#endif 1664 1665void sk_common_release(struct sock *sk) 1666{ 1667 if (sk->sk_prot->destroy) 1668 sk->sk_prot->destroy(sk); 1669 1670 /* 1671 * Observation: when sock_common_release is called, processes have 1672 * no access to socket. But net still has. 1673 * Step one, detach it from networking: 1674 * 1675 * A. Remove from hash tables. 1676 */ 1677 1678 sk->sk_prot->unhash(sk); 1679 1680 /* 1681 * In this point socket cannot receive new packets, but it is possible 1682 * that some packets are in flight because some CPU runs receiver and 1683 * did hash table lookup before we unhashed socket. They will achieve 1684 * receive queue and will be purged by socket destructor. 1685 * 1686 * Also we still have packets pending on receive queue and probably, 1687 * our own packets waiting in device queues. sock_destroy will drain 1688 * receive queue, but transmitted packets will delay socket destruction 1689 * until the last reference will be released. 1690 */ 1691 1692 sock_orphan(sk); 1693 1694 xfrm_sk_free_policy(sk); 1695 1696 sk_refcnt_debug_release(sk); 1697 sock_put(sk); 1698} 1699 1700EXPORT_SYMBOL(sk_common_release); 1701 1702static DEFINE_RWLOCK(proto_list_lock); 1703static LIST_HEAD(proto_list); 1704 1705int proto_register(struct proto *prot, int alloc_slab) 1706{ 1707 char *request_sock_slab_name = NULL; 1708 char *timewait_sock_slab_name; 1709 int rc = -ENOBUFS; 1710 1711 if (alloc_slab) { 1712 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0, 1713 SLAB_HWCACHE_ALIGN, NULL, NULL); 1714 1715 if (prot->slab == NULL) { 1716 printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n", 1717 prot->name); 1718 goto out; 1719 } 1720 1721 if (prot->rsk_prot != NULL) { 1722 static const char mask[] = "request_sock_%s"; 1723 1724 request_sock_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL); 1725 if (request_sock_slab_name == NULL) 1726 goto out_free_sock_slab; 1727 1728 sprintf(request_sock_slab_name, mask, prot->name); 1729 prot->rsk_prot->slab = kmem_cache_create(request_sock_slab_name, 1730 prot->rsk_prot->obj_size, 0, 1731 SLAB_HWCACHE_ALIGN, NULL, NULL); 1732 1733 if (prot->rsk_prot->slab == NULL) { 1734 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n", 1735 prot->name); 1736 goto out_free_request_sock_slab_name; 1737 } 1738 } 1739 1740 if (prot->twsk_prot != NULL) { 1741 static const char mask[] = "tw_sock_%s"; 1742 1743 timewait_sock_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL); 1744 1745 if (timewait_sock_slab_name == NULL) 1746 goto out_free_request_sock_slab; 1747 1748 sprintf(timewait_sock_slab_name, mask, prot->name); 1749 prot->twsk_prot->twsk_slab = 1750 kmem_cache_create(timewait_sock_slab_name, 1751 prot->twsk_prot->twsk_obj_size, 1752 0, SLAB_HWCACHE_ALIGN, 1753 NULL, NULL); 1754 if (prot->twsk_prot->twsk_slab == NULL) 1755 goto out_free_timewait_sock_slab_name; 1756 } 1757 } 1758 1759 write_lock(&proto_list_lock); 1760 list_add(&prot->node, &proto_list); 1761 write_unlock(&proto_list_lock); 1762 rc = 0; 1763out: 1764 return rc; 1765out_free_timewait_sock_slab_name: 1766 kfree(timewait_sock_slab_name); 1767out_free_request_sock_slab: 1768 if (prot->rsk_prot && prot->rsk_prot->slab) { 1769 kmem_cache_destroy(prot->rsk_prot->slab); 1770 prot->rsk_prot->slab = NULL; 1771 } 1772out_free_request_sock_slab_name: 1773 kfree(request_sock_slab_name); 1774out_free_sock_slab: 1775 kmem_cache_destroy(prot->slab); 1776 prot->slab = NULL; 1777 goto out; 1778} 1779 1780EXPORT_SYMBOL(proto_register); 1781 1782void proto_unregister(struct proto *prot) 1783{ 1784 write_lock(&proto_list_lock); 1785 list_del(&prot->node); 1786 write_unlock(&proto_list_lock); 1787 1788 if (prot->slab != NULL) { 1789 kmem_cache_destroy(prot->slab); 1790 prot->slab = NULL; 1791 } 1792 1793 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) { 1794 const char *name = kmem_cache_name(prot->rsk_prot->slab); 1795 1796 kmem_cache_destroy(prot->rsk_prot->slab); 1797 kfree(name); 1798 prot->rsk_prot->slab = NULL; 1799 } 1800 1801 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) { 1802 const char *name = kmem_cache_name(prot->twsk_prot->twsk_slab); 1803 1804 kmem_cache_destroy(prot->twsk_prot->twsk_slab); 1805 kfree(name); 1806 prot->twsk_prot->twsk_slab = NULL; 1807 } 1808} 1809 1810EXPORT_SYMBOL(proto_unregister); 1811 1812#ifdef CONFIG_PROC_FS 1813static inline struct proto *__proto_head(void) 1814{ 1815 return list_entry(proto_list.next, struct proto, node); 1816} 1817 1818static inline struct proto *proto_head(void) 1819{ 1820 return list_empty(&proto_list) ? NULL : __proto_head(); 1821} 1822 1823static inline struct proto *proto_next(struct proto *proto) 1824{ 1825 return proto->node.next == &proto_list ? NULL : 1826 list_entry(proto->node.next, struct proto, node); 1827} 1828 1829static inline struct proto *proto_get_idx(loff_t pos) 1830{ 1831 struct proto *proto; 1832 loff_t i = 0; 1833 1834 list_for_each_entry(proto, &proto_list, node) 1835 if (i++ == pos) 1836 goto out; 1837 1838 proto = NULL; 1839out: 1840 return proto; 1841} 1842 1843static void *proto_seq_start(struct seq_file *seq, loff_t *pos) 1844{ 1845 read_lock(&proto_list_lock); 1846 return *pos ? proto_get_idx(*pos - 1) : SEQ_START_TOKEN; 1847} 1848 1849static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos) 1850{ 1851 ++*pos; 1852 return v == SEQ_START_TOKEN ? proto_head() : proto_next(v); 1853} 1854 1855static void proto_seq_stop(struct seq_file *seq, void *v) 1856{ 1857 read_unlock(&proto_list_lock); 1858} 1859 1860static char proto_method_implemented(const void *method) 1861{ 1862 return method == NULL ? 'n' : 'y'; 1863} 1864 1865static void proto_seq_printf(struct seq_file *seq, struct proto *proto) 1866{ 1867 seq_printf(seq, "%-9s %4u %6d %6d %-3s %6u %-3s %-10s " 1868 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n", 1869 proto->name, 1870 proto->obj_size, 1871 proto->sockets_allocated != NULL ? atomic_read(proto->sockets_allocated) : -1, 1872 proto->memory_allocated != NULL ? atomic_read(proto->memory_allocated) : -1, 1873 proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI", 1874 proto->max_header, 1875 proto->slab == NULL ? "no" : "yes", 1876 module_name(proto->owner), 1877 proto_method_implemented(proto->close), 1878 proto_method_implemented(proto->connect), 1879 proto_method_implemented(proto->disconnect), 1880 proto_method_implemented(proto->accept), 1881 proto_method_implemented(proto->ioctl), 1882 proto_method_implemented(proto->init), 1883 proto_method_implemented(proto->destroy), 1884 proto_method_implemented(proto->shutdown), 1885 proto_method_implemented(proto->setsockopt), 1886 proto_method_implemented(proto->getsockopt), 1887 proto_method_implemented(proto->sendmsg), 1888 proto_method_implemented(proto->recvmsg), 1889 proto_method_implemented(proto->sendpage), 1890 proto_method_implemented(proto->bind), 1891 proto_method_implemented(proto->backlog_rcv), 1892 proto_method_implemented(proto->hash), 1893 proto_method_implemented(proto->unhash), 1894 proto_method_implemented(proto->get_port), 1895 proto_method_implemented(proto->enter_memory_pressure)); 1896} 1897 1898static int proto_seq_show(struct seq_file *seq, void *v) 1899{ 1900 if (v == SEQ_START_TOKEN) 1901 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s", 1902 "protocol", 1903 "size", 1904 "sockets", 1905 "memory", 1906 "press", 1907 "maxhdr", 1908 "slab", 1909 "module", 1910 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n"); 1911 else 1912 proto_seq_printf(seq, v); 1913 return 0; 1914} 1915 1916static struct seq_operations proto_seq_ops = { 1917 .start = proto_seq_start, 1918 .next = proto_seq_next, 1919 .stop = proto_seq_stop, 1920 .show = proto_seq_show, 1921}; 1922 1923static int proto_seq_open(struct inode *inode, struct file *file) 1924{ 1925 return seq_open(file, &proto_seq_ops); 1926} 1927 1928static const struct file_operations proto_seq_fops = { 1929 .owner = THIS_MODULE, 1930 .open = proto_seq_open, 1931 .read = seq_read, 1932 .llseek = seq_lseek, 1933 .release = seq_release, 1934}; 1935 1936static int __init proto_init(void) 1937{ 1938 /* register /proc/net/protocols */ 1939 return proc_net_fops_create("protocols", S_IRUGO, &proto_seq_fops) == NULL ? -ENOBUFS : 0; 1940} 1941 1942subsys_initcall(proto_init); 1943 1944#endif /* PROC_FS */ 1945 1946EXPORT_SYMBOL(sk_alloc); 1947EXPORT_SYMBOL(sk_free); 1948EXPORT_SYMBOL(sk_send_sigurg); 1949EXPORT_SYMBOL(sock_alloc_send_skb); 1950EXPORT_SYMBOL(sock_init_data); 1951EXPORT_SYMBOL(sock_kfree_s); 1952EXPORT_SYMBOL(sock_kmalloc); 1953EXPORT_SYMBOL(sock_no_accept); 1954EXPORT_SYMBOL(sock_no_bind); 1955EXPORT_SYMBOL(sock_no_connect); 1956EXPORT_SYMBOL(sock_no_getname); 1957EXPORT_SYMBOL(sock_no_getsockopt); 1958EXPORT_SYMBOL(sock_no_ioctl); 1959EXPORT_SYMBOL(sock_no_listen); 1960EXPORT_SYMBOL(sock_no_mmap); 1961EXPORT_SYMBOL(sock_no_poll); 1962EXPORT_SYMBOL(sock_no_recvmsg); 1963EXPORT_SYMBOL(sock_no_sendmsg); 1964EXPORT_SYMBOL(sock_no_sendpage); 1965EXPORT_SYMBOL(sock_no_setsockopt); 1966EXPORT_SYMBOL(sock_no_shutdown); 1967EXPORT_SYMBOL(sock_no_socketpair); 1968EXPORT_SYMBOL(sock_rfree); 1969EXPORT_SYMBOL(sock_setsockopt); 1970EXPORT_SYMBOL(sock_wfree); 1971EXPORT_SYMBOL(sock_wmalloc); 1972EXPORT_SYMBOL(sock_i_uid); 1973EXPORT_SYMBOL(sock_i_ino); 1974EXPORT_SYMBOL(sysctl_optmem_max); 1975#ifdef CONFIG_SYSCTL 1976EXPORT_SYMBOL(sysctl_rmem_max); 1977EXPORT_SYMBOL(sysctl_wmem_max); 1978#endif 1979