sock.c revision 9958089a43ae8a9af07402461c0b2b7548c7341e
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 case SO_TIMESTAMPNS: 525 if (valbool) { 526 if (optname == SO_TIMESTAMP) 527 sock_reset_flag(sk, SOCK_RCVTSTAMPNS); 528 else 529 sock_set_flag(sk, SOCK_RCVTSTAMPNS); 530 sock_set_flag(sk, SOCK_RCVTSTAMP); 531 sock_enable_timestamp(sk); 532 } else { 533 sock_reset_flag(sk, SOCK_RCVTSTAMP); 534 sock_reset_flag(sk, SOCK_RCVTSTAMPNS); 535 } 536 break; 537 538 case SO_RCVLOWAT: 539 if (val < 0) 540 val = INT_MAX; 541 sk->sk_rcvlowat = val ? : 1; 542 break; 543 544 case SO_RCVTIMEO: 545 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen); 546 break; 547 548 case SO_SNDTIMEO: 549 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen); 550 break; 551 552#ifdef CONFIG_NETDEVICES 553 case SO_BINDTODEVICE: 554 { 555 char devname[IFNAMSIZ]; 556 557 /* Sorry... */ 558 if (!capable(CAP_NET_RAW)) { 559 ret = -EPERM; 560 break; 561 } 562 563 /* Bind this socket to a particular device like "eth0", 564 * as specified in the passed interface name. If the 565 * name is "" or the option length is zero the socket 566 * is not bound. 567 */ 568 569 if (!valbool) { 570 sk->sk_bound_dev_if = 0; 571 } else { 572 if (optlen > IFNAMSIZ - 1) 573 optlen = IFNAMSIZ - 1; 574 memset(devname, 0, sizeof(devname)); 575 if (copy_from_user(devname, optval, optlen)) { 576 ret = -EFAULT; 577 break; 578 } 579 580 /* Remove any cached route for this socket. */ 581 sk_dst_reset(sk); 582 583 if (devname[0] == '\0') { 584 sk->sk_bound_dev_if = 0; 585 } else { 586 struct net_device *dev = dev_get_by_name(devname); 587 if (!dev) { 588 ret = -ENODEV; 589 break; 590 } 591 sk->sk_bound_dev_if = dev->ifindex; 592 dev_put(dev); 593 } 594 } 595 break; 596 } 597#endif 598 599 600 case SO_ATTACH_FILTER: 601 ret = -EINVAL; 602 if (optlen == sizeof(struct sock_fprog)) { 603 struct sock_fprog fprog; 604 605 ret = -EFAULT; 606 if (copy_from_user(&fprog, optval, sizeof(fprog))) 607 break; 608 609 ret = sk_attach_filter(&fprog, sk); 610 } 611 break; 612 613 case SO_DETACH_FILTER: 614 rcu_read_lock_bh(); 615 filter = rcu_dereference(sk->sk_filter); 616 if (filter) { 617 rcu_assign_pointer(sk->sk_filter, NULL); 618 sk_filter_release(sk, filter); 619 rcu_read_unlock_bh(); 620 break; 621 } 622 rcu_read_unlock_bh(); 623 ret = -ENONET; 624 break; 625 626 case SO_PASSSEC: 627 if (valbool) 628 set_bit(SOCK_PASSSEC, &sock->flags); 629 else 630 clear_bit(SOCK_PASSSEC, &sock->flags); 631 break; 632 633 /* We implement the SO_SNDLOWAT etc to 634 not be settable (1003.1g 5.3) */ 635 default: 636 ret = -ENOPROTOOPT; 637 break; 638 } 639 release_sock(sk); 640 return ret; 641} 642 643 644int sock_getsockopt(struct socket *sock, int level, int optname, 645 char __user *optval, int __user *optlen) 646{ 647 struct sock *sk = sock->sk; 648 649 union { 650 int val; 651 struct linger ling; 652 struct timeval tm; 653 } v; 654 655 unsigned int lv = sizeof(int); 656 int len; 657 658 if (get_user(len, optlen)) 659 return -EFAULT; 660 if (len < 0) 661 return -EINVAL; 662 663 switch(optname) { 664 case SO_DEBUG: 665 v.val = sock_flag(sk, SOCK_DBG); 666 break; 667 668 case SO_DONTROUTE: 669 v.val = sock_flag(sk, SOCK_LOCALROUTE); 670 break; 671 672 case SO_BROADCAST: 673 v.val = !!sock_flag(sk, SOCK_BROADCAST); 674 break; 675 676 case SO_SNDBUF: 677 v.val = sk->sk_sndbuf; 678 break; 679 680 case SO_RCVBUF: 681 v.val = sk->sk_rcvbuf; 682 break; 683 684 case SO_REUSEADDR: 685 v.val = sk->sk_reuse; 686 break; 687 688 case SO_KEEPALIVE: 689 v.val = !!sock_flag(sk, SOCK_KEEPOPEN); 690 break; 691 692 case SO_TYPE: 693 v.val = sk->sk_type; 694 break; 695 696 case SO_ERROR: 697 v.val = -sock_error(sk); 698 if (v.val==0) 699 v.val = xchg(&sk->sk_err_soft, 0); 700 break; 701 702 case SO_OOBINLINE: 703 v.val = !!sock_flag(sk, SOCK_URGINLINE); 704 break; 705 706 case SO_NO_CHECK: 707 v.val = sk->sk_no_check; 708 break; 709 710 case SO_PRIORITY: 711 v.val = sk->sk_priority; 712 break; 713 714 case SO_LINGER: 715 lv = sizeof(v.ling); 716 v.ling.l_onoff = !!sock_flag(sk, SOCK_LINGER); 717 v.ling.l_linger = sk->sk_lingertime / HZ; 718 break; 719 720 case SO_BSDCOMPAT: 721 sock_warn_obsolete_bsdism("getsockopt"); 722 break; 723 724 case SO_TIMESTAMP: 725 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && 726 !sock_flag(sk, SOCK_RCVTSTAMPNS); 727 break; 728 729 case SO_TIMESTAMPNS: 730 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS); 731 break; 732 733 case SO_RCVTIMEO: 734 lv=sizeof(struct timeval); 735 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) { 736 v.tm.tv_sec = 0; 737 v.tm.tv_usec = 0; 738 } else { 739 v.tm.tv_sec = sk->sk_rcvtimeo / HZ; 740 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ; 741 } 742 break; 743 744 case SO_SNDTIMEO: 745 lv=sizeof(struct timeval); 746 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) { 747 v.tm.tv_sec = 0; 748 v.tm.tv_usec = 0; 749 } else { 750 v.tm.tv_sec = sk->sk_sndtimeo / HZ; 751 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ; 752 } 753 break; 754 755 case SO_RCVLOWAT: 756 v.val = sk->sk_rcvlowat; 757 break; 758 759 case SO_SNDLOWAT: 760 v.val=1; 761 break; 762 763 case SO_PASSCRED: 764 v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0; 765 break; 766 767 case SO_PEERCRED: 768 if (len > sizeof(sk->sk_peercred)) 769 len = sizeof(sk->sk_peercred); 770 if (copy_to_user(optval, &sk->sk_peercred, len)) 771 return -EFAULT; 772 goto lenout; 773 774 case SO_PEERNAME: 775 { 776 char address[128]; 777 778 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2)) 779 return -ENOTCONN; 780 if (lv < len) 781 return -EINVAL; 782 if (copy_to_user(optval, address, len)) 783 return -EFAULT; 784 goto lenout; 785 } 786 787 /* Dubious BSD thing... Probably nobody even uses it, but 788 * the UNIX standard wants it for whatever reason... -DaveM 789 */ 790 case SO_ACCEPTCONN: 791 v.val = sk->sk_state == TCP_LISTEN; 792 break; 793 794 case SO_PASSSEC: 795 v.val = test_bit(SOCK_PASSSEC, &sock->flags) ? 1 : 0; 796 break; 797 798 case SO_PEERSEC: 799 return security_socket_getpeersec_stream(sock, optval, optlen, len); 800 801 default: 802 return -ENOPROTOOPT; 803 } 804 805 if (len > lv) 806 len = lv; 807 if (copy_to_user(optval, &v, len)) 808 return -EFAULT; 809lenout: 810 if (put_user(len, optlen)) 811 return -EFAULT; 812 return 0; 813} 814 815/* 816 * Initialize an sk_lock. 817 * 818 * (We also register the sk_lock with the lock validator.) 819 */ 820static inline void sock_lock_init(struct sock *sk) 821{ 822 sock_lock_init_class_and_name(sk, 823 af_family_slock_key_strings[sk->sk_family], 824 af_family_slock_keys + sk->sk_family, 825 af_family_key_strings[sk->sk_family], 826 af_family_keys + sk->sk_family); 827} 828 829/** 830 * sk_alloc - All socket objects are allocated here 831 * @family: protocol family 832 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 833 * @prot: struct proto associated with this new sock instance 834 * @zero_it: if we should zero the newly allocated sock 835 */ 836struct sock *sk_alloc(int family, gfp_t priority, 837 struct proto *prot, int zero_it) 838{ 839 struct sock *sk = NULL; 840 struct kmem_cache *slab = prot->slab; 841 842 if (slab != NULL) 843 sk = kmem_cache_alloc(slab, priority); 844 else 845 sk = kmalloc(prot->obj_size, priority); 846 847 if (sk) { 848 if (zero_it) { 849 memset(sk, 0, prot->obj_size); 850 sk->sk_family = family; 851 /* 852 * See comment in struct sock definition to understand 853 * why we need sk_prot_creator -acme 854 */ 855 sk->sk_prot = sk->sk_prot_creator = prot; 856 sock_lock_init(sk); 857 } 858 859 if (security_sk_alloc(sk, family, priority)) 860 goto out_free; 861 862 if (!try_module_get(prot->owner)) 863 goto out_free; 864 } 865 return sk; 866 867out_free: 868 if (slab != NULL) 869 kmem_cache_free(slab, sk); 870 else 871 kfree(sk); 872 return NULL; 873} 874 875void sk_free(struct sock *sk) 876{ 877 struct sk_filter *filter; 878 struct module *owner = sk->sk_prot_creator->owner; 879 880 if (sk->sk_destruct) 881 sk->sk_destruct(sk); 882 883 filter = rcu_dereference(sk->sk_filter); 884 if (filter) { 885 sk_filter_release(sk, filter); 886 rcu_assign_pointer(sk->sk_filter, NULL); 887 } 888 889 sock_disable_timestamp(sk); 890 891 if (atomic_read(&sk->sk_omem_alloc)) 892 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n", 893 __FUNCTION__, atomic_read(&sk->sk_omem_alloc)); 894 895 security_sk_free(sk); 896 if (sk->sk_prot_creator->slab != NULL) 897 kmem_cache_free(sk->sk_prot_creator->slab, sk); 898 else 899 kfree(sk); 900 module_put(owner); 901} 902 903struct sock *sk_clone(const struct sock *sk, const gfp_t priority) 904{ 905 struct sock *newsk = sk_alloc(sk->sk_family, priority, sk->sk_prot, 0); 906 907 if (newsk != NULL) { 908 struct sk_filter *filter; 909 910 sock_copy(newsk, sk); 911 912 /* SANITY */ 913 sk_node_init(&newsk->sk_node); 914 sock_lock_init(newsk); 915 bh_lock_sock(newsk); 916 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL; 917 918 atomic_set(&newsk->sk_rmem_alloc, 0); 919 atomic_set(&newsk->sk_wmem_alloc, 0); 920 atomic_set(&newsk->sk_omem_alloc, 0); 921 skb_queue_head_init(&newsk->sk_receive_queue); 922 skb_queue_head_init(&newsk->sk_write_queue); 923#ifdef CONFIG_NET_DMA 924 skb_queue_head_init(&newsk->sk_async_wait_queue); 925#endif 926 927 rwlock_init(&newsk->sk_dst_lock); 928 rwlock_init(&newsk->sk_callback_lock); 929 lockdep_set_class(&newsk->sk_callback_lock, 930 af_callback_keys + newsk->sk_family); 931 932 newsk->sk_dst_cache = NULL; 933 newsk->sk_wmem_queued = 0; 934 newsk->sk_forward_alloc = 0; 935 newsk->sk_send_head = NULL; 936 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK; 937 938 sock_reset_flag(newsk, SOCK_DONE); 939 skb_queue_head_init(&newsk->sk_error_queue); 940 941 filter = newsk->sk_filter; 942 if (filter != NULL) 943 sk_filter_charge(newsk, filter); 944 945 if (unlikely(xfrm_sk_clone_policy(newsk))) { 946 /* It is still raw copy of parent, so invalidate 947 * destructor and make plain sk_free() */ 948 newsk->sk_destruct = NULL; 949 sk_free(newsk); 950 newsk = NULL; 951 goto out; 952 } 953 954 newsk->sk_err = 0; 955 newsk->sk_priority = 0; 956 atomic_set(&newsk->sk_refcnt, 2); 957 958 /* 959 * Increment the counter in the same struct proto as the master 960 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that 961 * is the same as sk->sk_prot->socks, as this field was copied 962 * with memcpy). 963 * 964 * This _changes_ the previous behaviour, where 965 * tcp_create_openreq_child always was incrementing the 966 * equivalent to tcp_prot->socks (inet_sock_nr), so this have 967 * to be taken into account in all callers. -acme 968 */ 969 sk_refcnt_debug_inc(newsk); 970 newsk->sk_socket = NULL; 971 newsk->sk_sleep = NULL; 972 973 if (newsk->sk_prot->sockets_allocated) 974 atomic_inc(newsk->sk_prot->sockets_allocated); 975 } 976out: 977 return newsk; 978} 979 980EXPORT_SYMBOL_GPL(sk_clone); 981 982void sk_setup_caps(struct sock *sk, struct dst_entry *dst) 983{ 984 __sk_dst_set(sk, dst); 985 sk->sk_route_caps = dst->dev->features; 986 if (sk->sk_route_caps & NETIF_F_GSO) 987 sk->sk_route_caps |= NETIF_F_GSO_MASK; 988 if (sk_can_gso(sk)) { 989 if (dst->header_len) 990 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 991 else 992 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM; 993 } 994} 995EXPORT_SYMBOL_GPL(sk_setup_caps); 996 997void __init sk_init(void) 998{ 999 if (num_physpages <= 4096) { 1000 sysctl_wmem_max = 32767; 1001 sysctl_rmem_max = 32767; 1002 sysctl_wmem_default = 32767; 1003 sysctl_rmem_default = 32767; 1004 } else if (num_physpages >= 131072) { 1005 sysctl_wmem_max = 131071; 1006 sysctl_rmem_max = 131071; 1007 } 1008} 1009 1010/* 1011 * Simple resource managers for sockets. 1012 */ 1013 1014 1015/* 1016 * Write buffer destructor automatically called from kfree_skb. 1017 */ 1018void sock_wfree(struct sk_buff *skb) 1019{ 1020 struct sock *sk = skb->sk; 1021 1022 /* In case it might be waiting for more memory. */ 1023 atomic_sub(skb->truesize, &sk->sk_wmem_alloc); 1024 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) 1025 sk->sk_write_space(sk); 1026 sock_put(sk); 1027} 1028 1029/* 1030 * Read buffer destructor automatically called from kfree_skb. 1031 */ 1032void sock_rfree(struct sk_buff *skb) 1033{ 1034 struct sock *sk = skb->sk; 1035 1036 atomic_sub(skb->truesize, &sk->sk_rmem_alloc); 1037} 1038 1039 1040int sock_i_uid(struct sock *sk) 1041{ 1042 int uid; 1043 1044 read_lock(&sk->sk_callback_lock); 1045 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0; 1046 read_unlock(&sk->sk_callback_lock); 1047 return uid; 1048} 1049 1050unsigned long sock_i_ino(struct sock *sk) 1051{ 1052 unsigned long ino; 1053 1054 read_lock(&sk->sk_callback_lock); 1055 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0; 1056 read_unlock(&sk->sk_callback_lock); 1057 return ino; 1058} 1059 1060/* 1061 * Allocate a skb from the socket's send buffer. 1062 */ 1063struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 1064 gfp_t priority) 1065{ 1066 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) { 1067 struct sk_buff * skb = alloc_skb(size, priority); 1068 if (skb) { 1069 skb_set_owner_w(skb, sk); 1070 return skb; 1071 } 1072 } 1073 return NULL; 1074} 1075 1076/* 1077 * Allocate a skb from the socket's receive buffer. 1078 */ 1079struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force, 1080 gfp_t priority) 1081{ 1082 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) { 1083 struct sk_buff *skb = alloc_skb(size, priority); 1084 if (skb) { 1085 skb_set_owner_r(skb, sk); 1086 return skb; 1087 } 1088 } 1089 return NULL; 1090} 1091 1092/* 1093 * Allocate a memory block from the socket's option memory buffer. 1094 */ 1095void *sock_kmalloc(struct sock *sk, int size, gfp_t priority) 1096{ 1097 if ((unsigned)size <= sysctl_optmem_max && 1098 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) { 1099 void *mem; 1100 /* First do the add, to avoid the race if kmalloc 1101 * might sleep. 1102 */ 1103 atomic_add(size, &sk->sk_omem_alloc); 1104 mem = kmalloc(size, priority); 1105 if (mem) 1106 return mem; 1107 atomic_sub(size, &sk->sk_omem_alloc); 1108 } 1109 return NULL; 1110} 1111 1112/* 1113 * Free an option memory block. 1114 */ 1115void sock_kfree_s(struct sock *sk, void *mem, int size) 1116{ 1117 kfree(mem); 1118 atomic_sub(size, &sk->sk_omem_alloc); 1119} 1120 1121/* It is almost wait_for_tcp_memory minus release_sock/lock_sock. 1122 I think, these locks should be removed for datagram sockets. 1123 */ 1124static long sock_wait_for_wmem(struct sock * sk, long timeo) 1125{ 1126 DEFINE_WAIT(wait); 1127 1128 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags); 1129 for (;;) { 1130 if (!timeo) 1131 break; 1132 if (signal_pending(current)) 1133 break; 1134 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 1135 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE); 1136 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) 1137 break; 1138 if (sk->sk_shutdown & SEND_SHUTDOWN) 1139 break; 1140 if (sk->sk_err) 1141 break; 1142 timeo = schedule_timeout(timeo); 1143 } 1144 finish_wait(sk->sk_sleep, &wait); 1145 return timeo; 1146} 1147 1148 1149/* 1150 * Generic send/receive buffer handlers 1151 */ 1152 1153static struct sk_buff *sock_alloc_send_pskb(struct sock *sk, 1154 unsigned long header_len, 1155 unsigned long data_len, 1156 int noblock, int *errcode) 1157{ 1158 struct sk_buff *skb; 1159 gfp_t gfp_mask; 1160 long timeo; 1161 int err; 1162 1163 gfp_mask = sk->sk_allocation; 1164 if (gfp_mask & __GFP_WAIT) 1165 gfp_mask |= __GFP_REPEAT; 1166 1167 timeo = sock_sndtimeo(sk, noblock); 1168 while (1) { 1169 err = sock_error(sk); 1170 if (err != 0) 1171 goto failure; 1172 1173 err = -EPIPE; 1174 if (sk->sk_shutdown & SEND_SHUTDOWN) 1175 goto failure; 1176 1177 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) { 1178 skb = alloc_skb(header_len, gfp_mask); 1179 if (skb) { 1180 int npages; 1181 int i; 1182 1183 /* No pages, we're done... */ 1184 if (!data_len) 1185 break; 1186 1187 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT; 1188 skb->truesize += data_len; 1189 skb_shinfo(skb)->nr_frags = npages; 1190 for (i = 0; i < npages; i++) { 1191 struct page *page; 1192 skb_frag_t *frag; 1193 1194 page = alloc_pages(sk->sk_allocation, 0); 1195 if (!page) { 1196 err = -ENOBUFS; 1197 skb_shinfo(skb)->nr_frags = i; 1198 kfree_skb(skb); 1199 goto failure; 1200 } 1201 1202 frag = &skb_shinfo(skb)->frags[i]; 1203 frag->page = page; 1204 frag->page_offset = 0; 1205 frag->size = (data_len >= PAGE_SIZE ? 1206 PAGE_SIZE : 1207 data_len); 1208 data_len -= PAGE_SIZE; 1209 } 1210 1211 /* Full success... */ 1212 break; 1213 } 1214 err = -ENOBUFS; 1215 goto failure; 1216 } 1217 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags); 1218 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 1219 err = -EAGAIN; 1220 if (!timeo) 1221 goto failure; 1222 if (signal_pending(current)) 1223 goto interrupted; 1224 timeo = sock_wait_for_wmem(sk, timeo); 1225 } 1226 1227 skb_set_owner_w(skb, sk); 1228 return skb; 1229 1230interrupted: 1231 err = sock_intr_errno(timeo); 1232failure: 1233 *errcode = err; 1234 return NULL; 1235} 1236 1237struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size, 1238 int noblock, int *errcode) 1239{ 1240 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode); 1241} 1242 1243static void __lock_sock(struct sock *sk) 1244{ 1245 DEFINE_WAIT(wait); 1246 1247 for (;;) { 1248 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait, 1249 TASK_UNINTERRUPTIBLE); 1250 spin_unlock_bh(&sk->sk_lock.slock); 1251 schedule(); 1252 spin_lock_bh(&sk->sk_lock.slock); 1253 if (!sock_owned_by_user(sk)) 1254 break; 1255 } 1256 finish_wait(&sk->sk_lock.wq, &wait); 1257} 1258 1259static void __release_sock(struct sock *sk) 1260{ 1261 struct sk_buff *skb = sk->sk_backlog.head; 1262 1263 do { 1264 sk->sk_backlog.head = sk->sk_backlog.tail = NULL; 1265 bh_unlock_sock(sk); 1266 1267 do { 1268 struct sk_buff *next = skb->next; 1269 1270 skb->next = NULL; 1271 sk->sk_backlog_rcv(sk, skb); 1272 1273 /* 1274 * We are in process context here with softirqs 1275 * disabled, use cond_resched_softirq() to preempt. 1276 * This is safe to do because we've taken the backlog 1277 * queue private: 1278 */ 1279 cond_resched_softirq(); 1280 1281 skb = next; 1282 } while (skb != NULL); 1283 1284 bh_lock_sock(sk); 1285 } while ((skb = sk->sk_backlog.head) != NULL); 1286} 1287 1288/** 1289 * sk_wait_data - wait for data to arrive at sk_receive_queue 1290 * @sk: sock to wait on 1291 * @timeo: for how long 1292 * 1293 * Now socket state including sk->sk_err is changed only under lock, 1294 * hence we may omit checks after joining wait queue. 1295 * We check receive queue before schedule() only as optimization; 1296 * it is very likely that release_sock() added new data. 1297 */ 1298int sk_wait_data(struct sock *sk, long *timeo) 1299{ 1300 int rc; 1301 DEFINE_WAIT(wait); 1302 1303 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE); 1304 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags); 1305 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue)); 1306 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags); 1307 finish_wait(sk->sk_sleep, &wait); 1308 return rc; 1309} 1310 1311EXPORT_SYMBOL(sk_wait_data); 1312 1313/* 1314 * Set of default routines for initialising struct proto_ops when 1315 * the protocol does not support a particular function. In certain 1316 * cases where it makes no sense for a protocol to have a "do nothing" 1317 * function, some default processing is provided. 1318 */ 1319 1320int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len) 1321{ 1322 return -EOPNOTSUPP; 1323} 1324 1325int sock_no_connect(struct socket *sock, struct sockaddr *saddr, 1326 int len, int flags) 1327{ 1328 return -EOPNOTSUPP; 1329} 1330 1331int sock_no_socketpair(struct socket *sock1, struct socket *sock2) 1332{ 1333 return -EOPNOTSUPP; 1334} 1335 1336int sock_no_accept(struct socket *sock, struct socket *newsock, int flags) 1337{ 1338 return -EOPNOTSUPP; 1339} 1340 1341int sock_no_getname(struct socket *sock, struct sockaddr *saddr, 1342 int *len, int peer) 1343{ 1344 return -EOPNOTSUPP; 1345} 1346 1347unsigned int sock_no_poll(struct file * file, struct socket *sock, poll_table *pt) 1348{ 1349 return 0; 1350} 1351 1352int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 1353{ 1354 return -EOPNOTSUPP; 1355} 1356 1357int sock_no_listen(struct socket *sock, int backlog) 1358{ 1359 return -EOPNOTSUPP; 1360} 1361 1362int sock_no_shutdown(struct socket *sock, int how) 1363{ 1364 return -EOPNOTSUPP; 1365} 1366 1367int sock_no_setsockopt(struct socket *sock, int level, int optname, 1368 char __user *optval, int optlen) 1369{ 1370 return -EOPNOTSUPP; 1371} 1372 1373int sock_no_getsockopt(struct socket *sock, int level, int optname, 1374 char __user *optval, int __user *optlen) 1375{ 1376 return -EOPNOTSUPP; 1377} 1378 1379int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m, 1380 size_t len) 1381{ 1382 return -EOPNOTSUPP; 1383} 1384 1385int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m, 1386 size_t len, int flags) 1387{ 1388 return -EOPNOTSUPP; 1389} 1390 1391int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma) 1392{ 1393 /* Mirror missing mmap method error code */ 1394 return -ENODEV; 1395} 1396 1397ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags) 1398{ 1399 ssize_t res; 1400 struct msghdr msg = {.msg_flags = flags}; 1401 struct kvec iov; 1402 char *kaddr = kmap(page); 1403 iov.iov_base = kaddr + offset; 1404 iov.iov_len = size; 1405 res = kernel_sendmsg(sock, &msg, &iov, 1, size); 1406 kunmap(page); 1407 return res; 1408} 1409 1410/* 1411 * Default Socket Callbacks 1412 */ 1413 1414static void sock_def_wakeup(struct sock *sk) 1415{ 1416 read_lock(&sk->sk_callback_lock); 1417 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) 1418 wake_up_interruptible_all(sk->sk_sleep); 1419 read_unlock(&sk->sk_callback_lock); 1420} 1421 1422static void sock_def_error_report(struct sock *sk) 1423{ 1424 read_lock(&sk->sk_callback_lock); 1425 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) 1426 wake_up_interruptible(sk->sk_sleep); 1427 sk_wake_async(sk,0,POLL_ERR); 1428 read_unlock(&sk->sk_callback_lock); 1429} 1430 1431static void sock_def_readable(struct sock *sk, int len) 1432{ 1433 read_lock(&sk->sk_callback_lock); 1434 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) 1435 wake_up_interruptible(sk->sk_sleep); 1436 sk_wake_async(sk,1,POLL_IN); 1437 read_unlock(&sk->sk_callback_lock); 1438} 1439 1440static void sock_def_write_space(struct sock *sk) 1441{ 1442 read_lock(&sk->sk_callback_lock); 1443 1444 /* Do not wake up a writer until he can make "significant" 1445 * progress. --DaveM 1446 */ 1447 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) { 1448 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) 1449 wake_up_interruptible(sk->sk_sleep); 1450 1451 /* Should agree with poll, otherwise some programs break */ 1452 if (sock_writeable(sk)) 1453 sk_wake_async(sk, 2, POLL_OUT); 1454 } 1455 1456 read_unlock(&sk->sk_callback_lock); 1457} 1458 1459static void sock_def_destruct(struct sock *sk) 1460{ 1461 kfree(sk->sk_protinfo); 1462} 1463 1464void sk_send_sigurg(struct sock *sk) 1465{ 1466 if (sk->sk_socket && sk->sk_socket->file) 1467 if (send_sigurg(&sk->sk_socket->file->f_owner)) 1468 sk_wake_async(sk, 3, POLL_PRI); 1469} 1470 1471void sk_reset_timer(struct sock *sk, struct timer_list* timer, 1472 unsigned long expires) 1473{ 1474 if (!mod_timer(timer, expires)) 1475 sock_hold(sk); 1476} 1477 1478EXPORT_SYMBOL(sk_reset_timer); 1479 1480void sk_stop_timer(struct sock *sk, struct timer_list* timer) 1481{ 1482 if (timer_pending(timer) && del_timer(timer)) 1483 __sock_put(sk); 1484} 1485 1486EXPORT_SYMBOL(sk_stop_timer); 1487 1488void sock_init_data(struct socket *sock, struct sock *sk) 1489{ 1490 skb_queue_head_init(&sk->sk_receive_queue); 1491 skb_queue_head_init(&sk->sk_write_queue); 1492 skb_queue_head_init(&sk->sk_error_queue); 1493#ifdef CONFIG_NET_DMA 1494 skb_queue_head_init(&sk->sk_async_wait_queue); 1495#endif 1496 1497 sk->sk_send_head = NULL; 1498 1499 init_timer(&sk->sk_timer); 1500 1501 sk->sk_allocation = GFP_KERNEL; 1502 sk->sk_rcvbuf = sysctl_rmem_default; 1503 sk->sk_sndbuf = sysctl_wmem_default; 1504 sk->sk_state = TCP_CLOSE; 1505 sk->sk_socket = sock; 1506 1507 sock_set_flag(sk, SOCK_ZAPPED); 1508 1509 if (sock) { 1510 sk->sk_type = sock->type; 1511 sk->sk_sleep = &sock->wait; 1512 sock->sk = sk; 1513 } else 1514 sk->sk_sleep = NULL; 1515 1516 rwlock_init(&sk->sk_dst_lock); 1517 rwlock_init(&sk->sk_callback_lock); 1518 lockdep_set_class(&sk->sk_callback_lock, 1519 af_callback_keys + sk->sk_family); 1520 1521 sk->sk_state_change = sock_def_wakeup; 1522 sk->sk_data_ready = sock_def_readable; 1523 sk->sk_write_space = sock_def_write_space; 1524 sk->sk_error_report = sock_def_error_report; 1525 sk->sk_destruct = sock_def_destruct; 1526 1527 sk->sk_sndmsg_page = NULL; 1528 sk->sk_sndmsg_off = 0; 1529 1530 sk->sk_peercred.pid = 0; 1531 sk->sk_peercred.uid = -1; 1532 sk->sk_peercred.gid = -1; 1533 sk->sk_write_pending = 0; 1534 sk->sk_rcvlowat = 1; 1535 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; 1536 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT; 1537 1538 sk->sk_stamp = ktime_set(-1L, -1L); 1539 1540 atomic_set(&sk->sk_refcnt, 1); 1541} 1542 1543void fastcall lock_sock_nested(struct sock *sk, int subclass) 1544{ 1545 might_sleep(); 1546 spin_lock_bh(&sk->sk_lock.slock); 1547 if (sk->sk_lock.owner) 1548 __lock_sock(sk); 1549 sk->sk_lock.owner = (void *)1; 1550 spin_unlock(&sk->sk_lock.slock); 1551 /* 1552 * The sk_lock has mutex_lock() semantics here: 1553 */ 1554 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_); 1555 local_bh_enable(); 1556} 1557 1558EXPORT_SYMBOL(lock_sock_nested); 1559 1560void fastcall release_sock(struct sock *sk) 1561{ 1562 /* 1563 * The sk_lock has mutex_unlock() semantics: 1564 */ 1565 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_); 1566 1567 spin_lock_bh(&sk->sk_lock.slock); 1568 if (sk->sk_backlog.tail) 1569 __release_sock(sk); 1570 sk->sk_lock.owner = NULL; 1571 if (waitqueue_active(&sk->sk_lock.wq)) 1572 wake_up(&sk->sk_lock.wq); 1573 spin_unlock_bh(&sk->sk_lock.slock); 1574} 1575EXPORT_SYMBOL(release_sock); 1576 1577int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp) 1578{ 1579 struct timeval tv; 1580 if (!sock_flag(sk, SOCK_TIMESTAMP)) 1581 sock_enable_timestamp(sk); 1582 tv = ktime_to_timeval(sk->sk_stamp); 1583 if (tv.tv_sec == -1) 1584 return -ENOENT; 1585 if (tv.tv_sec == 0) { 1586 sk->sk_stamp = ktime_get_real(); 1587 tv = ktime_to_timeval(sk->sk_stamp); 1588 } 1589 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0; 1590} 1591EXPORT_SYMBOL(sock_get_timestamp); 1592 1593int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp) 1594{ 1595 struct timespec ts; 1596 if (!sock_flag(sk, SOCK_TIMESTAMP)) 1597 sock_enable_timestamp(sk); 1598 ts = ktime_to_timespec(sk->sk_stamp); 1599 if (ts.tv_sec == -1) 1600 return -ENOENT; 1601 if (ts.tv_sec == 0) { 1602 sk->sk_stamp = ktime_get_real(); 1603 ts = ktime_to_timespec(sk->sk_stamp); 1604 } 1605 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0; 1606} 1607EXPORT_SYMBOL(sock_get_timestampns); 1608 1609void sock_enable_timestamp(struct sock *sk) 1610{ 1611 if (!sock_flag(sk, SOCK_TIMESTAMP)) { 1612 sock_set_flag(sk, SOCK_TIMESTAMP); 1613 net_enable_timestamp(); 1614 } 1615} 1616EXPORT_SYMBOL(sock_enable_timestamp); 1617 1618/* 1619 * Get a socket option on an socket. 1620 * 1621 * FIX: POSIX 1003.1g is very ambiguous here. It states that 1622 * asynchronous errors should be reported by getsockopt. We assume 1623 * this means if you specify SO_ERROR (otherwise whats the point of it). 1624 */ 1625int sock_common_getsockopt(struct socket *sock, int level, int optname, 1626 char __user *optval, int __user *optlen) 1627{ 1628 struct sock *sk = sock->sk; 1629 1630 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen); 1631} 1632 1633EXPORT_SYMBOL(sock_common_getsockopt); 1634 1635#ifdef CONFIG_COMPAT 1636int compat_sock_common_getsockopt(struct socket *sock, int level, int optname, 1637 char __user *optval, int __user *optlen) 1638{ 1639 struct sock *sk = sock->sk; 1640 1641 if (sk->sk_prot->compat_getsockopt != NULL) 1642 return sk->sk_prot->compat_getsockopt(sk, level, optname, 1643 optval, optlen); 1644 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen); 1645} 1646EXPORT_SYMBOL(compat_sock_common_getsockopt); 1647#endif 1648 1649int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock, 1650 struct msghdr *msg, size_t size, int flags) 1651{ 1652 struct sock *sk = sock->sk; 1653 int addr_len = 0; 1654 int err; 1655 1656 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT, 1657 flags & ~MSG_DONTWAIT, &addr_len); 1658 if (err >= 0) 1659 msg->msg_namelen = addr_len; 1660 return err; 1661} 1662 1663EXPORT_SYMBOL(sock_common_recvmsg); 1664 1665/* 1666 * Set socket options on an inet socket. 1667 */ 1668int sock_common_setsockopt(struct socket *sock, int level, int optname, 1669 char __user *optval, int optlen) 1670{ 1671 struct sock *sk = sock->sk; 1672 1673 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen); 1674} 1675 1676EXPORT_SYMBOL(sock_common_setsockopt); 1677 1678#ifdef CONFIG_COMPAT 1679int compat_sock_common_setsockopt(struct socket *sock, int level, int optname, 1680 char __user *optval, int optlen) 1681{ 1682 struct sock *sk = sock->sk; 1683 1684 if (sk->sk_prot->compat_setsockopt != NULL) 1685 return sk->sk_prot->compat_setsockopt(sk, level, optname, 1686 optval, optlen); 1687 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen); 1688} 1689EXPORT_SYMBOL(compat_sock_common_setsockopt); 1690#endif 1691 1692void sk_common_release(struct sock *sk) 1693{ 1694 if (sk->sk_prot->destroy) 1695 sk->sk_prot->destroy(sk); 1696 1697 /* 1698 * Observation: when sock_common_release is called, processes have 1699 * no access to socket. But net still has. 1700 * Step one, detach it from networking: 1701 * 1702 * A. Remove from hash tables. 1703 */ 1704 1705 sk->sk_prot->unhash(sk); 1706 1707 /* 1708 * In this point socket cannot receive new packets, but it is possible 1709 * that some packets are in flight because some CPU runs receiver and 1710 * did hash table lookup before we unhashed socket. They will achieve 1711 * receive queue and will be purged by socket destructor. 1712 * 1713 * Also we still have packets pending on receive queue and probably, 1714 * our own packets waiting in device queues. sock_destroy will drain 1715 * receive queue, but transmitted packets will delay socket destruction 1716 * until the last reference will be released. 1717 */ 1718 1719 sock_orphan(sk); 1720 1721 xfrm_sk_free_policy(sk); 1722 1723 sk_refcnt_debug_release(sk); 1724 sock_put(sk); 1725} 1726 1727EXPORT_SYMBOL(sk_common_release); 1728 1729static DEFINE_RWLOCK(proto_list_lock); 1730static LIST_HEAD(proto_list); 1731 1732int proto_register(struct proto *prot, int alloc_slab) 1733{ 1734 char *request_sock_slab_name = NULL; 1735 char *timewait_sock_slab_name; 1736 int rc = -ENOBUFS; 1737 1738 if (alloc_slab) { 1739 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0, 1740 SLAB_HWCACHE_ALIGN, NULL, NULL); 1741 1742 if (prot->slab == NULL) { 1743 printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n", 1744 prot->name); 1745 goto out; 1746 } 1747 1748 if (prot->rsk_prot != NULL) { 1749 static const char mask[] = "request_sock_%s"; 1750 1751 request_sock_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL); 1752 if (request_sock_slab_name == NULL) 1753 goto out_free_sock_slab; 1754 1755 sprintf(request_sock_slab_name, mask, prot->name); 1756 prot->rsk_prot->slab = kmem_cache_create(request_sock_slab_name, 1757 prot->rsk_prot->obj_size, 0, 1758 SLAB_HWCACHE_ALIGN, NULL, NULL); 1759 1760 if (prot->rsk_prot->slab == NULL) { 1761 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n", 1762 prot->name); 1763 goto out_free_request_sock_slab_name; 1764 } 1765 } 1766 1767 if (prot->twsk_prot != NULL) { 1768 static const char mask[] = "tw_sock_%s"; 1769 1770 timewait_sock_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL); 1771 1772 if (timewait_sock_slab_name == NULL) 1773 goto out_free_request_sock_slab; 1774 1775 sprintf(timewait_sock_slab_name, mask, prot->name); 1776 prot->twsk_prot->twsk_slab = 1777 kmem_cache_create(timewait_sock_slab_name, 1778 prot->twsk_prot->twsk_obj_size, 1779 0, SLAB_HWCACHE_ALIGN, 1780 NULL, NULL); 1781 if (prot->twsk_prot->twsk_slab == NULL) 1782 goto out_free_timewait_sock_slab_name; 1783 } 1784 } 1785 1786 write_lock(&proto_list_lock); 1787 list_add(&prot->node, &proto_list); 1788 write_unlock(&proto_list_lock); 1789 rc = 0; 1790out: 1791 return rc; 1792out_free_timewait_sock_slab_name: 1793 kfree(timewait_sock_slab_name); 1794out_free_request_sock_slab: 1795 if (prot->rsk_prot && prot->rsk_prot->slab) { 1796 kmem_cache_destroy(prot->rsk_prot->slab); 1797 prot->rsk_prot->slab = NULL; 1798 } 1799out_free_request_sock_slab_name: 1800 kfree(request_sock_slab_name); 1801out_free_sock_slab: 1802 kmem_cache_destroy(prot->slab); 1803 prot->slab = NULL; 1804 goto out; 1805} 1806 1807EXPORT_SYMBOL(proto_register); 1808 1809void proto_unregister(struct proto *prot) 1810{ 1811 write_lock(&proto_list_lock); 1812 list_del(&prot->node); 1813 write_unlock(&proto_list_lock); 1814 1815 if (prot->slab != NULL) { 1816 kmem_cache_destroy(prot->slab); 1817 prot->slab = NULL; 1818 } 1819 1820 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) { 1821 const char *name = kmem_cache_name(prot->rsk_prot->slab); 1822 1823 kmem_cache_destroy(prot->rsk_prot->slab); 1824 kfree(name); 1825 prot->rsk_prot->slab = NULL; 1826 } 1827 1828 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) { 1829 const char *name = kmem_cache_name(prot->twsk_prot->twsk_slab); 1830 1831 kmem_cache_destroy(prot->twsk_prot->twsk_slab); 1832 kfree(name); 1833 prot->twsk_prot->twsk_slab = NULL; 1834 } 1835} 1836 1837EXPORT_SYMBOL(proto_unregister); 1838 1839#ifdef CONFIG_PROC_FS 1840static inline struct proto *__proto_head(void) 1841{ 1842 return list_entry(proto_list.next, struct proto, node); 1843} 1844 1845static inline struct proto *proto_head(void) 1846{ 1847 return list_empty(&proto_list) ? NULL : __proto_head(); 1848} 1849 1850static inline struct proto *proto_next(struct proto *proto) 1851{ 1852 return proto->node.next == &proto_list ? NULL : 1853 list_entry(proto->node.next, struct proto, node); 1854} 1855 1856static inline struct proto *proto_get_idx(loff_t pos) 1857{ 1858 struct proto *proto; 1859 loff_t i = 0; 1860 1861 list_for_each_entry(proto, &proto_list, node) 1862 if (i++ == pos) 1863 goto out; 1864 1865 proto = NULL; 1866out: 1867 return proto; 1868} 1869 1870static void *proto_seq_start(struct seq_file *seq, loff_t *pos) 1871{ 1872 read_lock(&proto_list_lock); 1873 return *pos ? proto_get_idx(*pos - 1) : SEQ_START_TOKEN; 1874} 1875 1876static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos) 1877{ 1878 ++*pos; 1879 return v == SEQ_START_TOKEN ? proto_head() : proto_next(v); 1880} 1881 1882static void proto_seq_stop(struct seq_file *seq, void *v) 1883{ 1884 read_unlock(&proto_list_lock); 1885} 1886 1887static char proto_method_implemented(const void *method) 1888{ 1889 return method == NULL ? 'n' : 'y'; 1890} 1891 1892static void proto_seq_printf(struct seq_file *seq, struct proto *proto) 1893{ 1894 seq_printf(seq, "%-9s %4u %6d %6d %-3s %6u %-3s %-10s " 1895 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n", 1896 proto->name, 1897 proto->obj_size, 1898 proto->sockets_allocated != NULL ? atomic_read(proto->sockets_allocated) : -1, 1899 proto->memory_allocated != NULL ? atomic_read(proto->memory_allocated) : -1, 1900 proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI", 1901 proto->max_header, 1902 proto->slab == NULL ? "no" : "yes", 1903 module_name(proto->owner), 1904 proto_method_implemented(proto->close), 1905 proto_method_implemented(proto->connect), 1906 proto_method_implemented(proto->disconnect), 1907 proto_method_implemented(proto->accept), 1908 proto_method_implemented(proto->ioctl), 1909 proto_method_implemented(proto->init), 1910 proto_method_implemented(proto->destroy), 1911 proto_method_implemented(proto->shutdown), 1912 proto_method_implemented(proto->setsockopt), 1913 proto_method_implemented(proto->getsockopt), 1914 proto_method_implemented(proto->sendmsg), 1915 proto_method_implemented(proto->recvmsg), 1916 proto_method_implemented(proto->sendpage), 1917 proto_method_implemented(proto->bind), 1918 proto_method_implemented(proto->backlog_rcv), 1919 proto_method_implemented(proto->hash), 1920 proto_method_implemented(proto->unhash), 1921 proto_method_implemented(proto->get_port), 1922 proto_method_implemented(proto->enter_memory_pressure)); 1923} 1924 1925static int proto_seq_show(struct seq_file *seq, void *v) 1926{ 1927 if (v == SEQ_START_TOKEN) 1928 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s", 1929 "protocol", 1930 "size", 1931 "sockets", 1932 "memory", 1933 "press", 1934 "maxhdr", 1935 "slab", 1936 "module", 1937 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n"); 1938 else 1939 proto_seq_printf(seq, v); 1940 return 0; 1941} 1942 1943static const struct seq_operations proto_seq_ops = { 1944 .start = proto_seq_start, 1945 .next = proto_seq_next, 1946 .stop = proto_seq_stop, 1947 .show = proto_seq_show, 1948}; 1949 1950static int proto_seq_open(struct inode *inode, struct file *file) 1951{ 1952 return seq_open(file, &proto_seq_ops); 1953} 1954 1955static const struct file_operations proto_seq_fops = { 1956 .owner = THIS_MODULE, 1957 .open = proto_seq_open, 1958 .read = seq_read, 1959 .llseek = seq_lseek, 1960 .release = seq_release, 1961}; 1962 1963static int __init proto_init(void) 1964{ 1965 /* register /proc/net/protocols */ 1966 return proc_net_fops_create("protocols", S_IRUGO, &proto_seq_fops) == NULL ? -ENOBUFS : 0; 1967} 1968 1969subsys_initcall(proto_init); 1970 1971#endif /* PROC_FS */ 1972 1973EXPORT_SYMBOL(sk_alloc); 1974EXPORT_SYMBOL(sk_free); 1975EXPORT_SYMBOL(sk_send_sigurg); 1976EXPORT_SYMBOL(sock_alloc_send_skb); 1977EXPORT_SYMBOL(sock_init_data); 1978EXPORT_SYMBOL(sock_kfree_s); 1979EXPORT_SYMBOL(sock_kmalloc); 1980EXPORT_SYMBOL(sock_no_accept); 1981EXPORT_SYMBOL(sock_no_bind); 1982EXPORT_SYMBOL(sock_no_connect); 1983EXPORT_SYMBOL(sock_no_getname); 1984EXPORT_SYMBOL(sock_no_getsockopt); 1985EXPORT_SYMBOL(sock_no_ioctl); 1986EXPORT_SYMBOL(sock_no_listen); 1987EXPORT_SYMBOL(sock_no_mmap); 1988EXPORT_SYMBOL(sock_no_poll); 1989EXPORT_SYMBOL(sock_no_recvmsg); 1990EXPORT_SYMBOL(sock_no_sendmsg); 1991EXPORT_SYMBOL(sock_no_sendpage); 1992EXPORT_SYMBOL(sock_no_setsockopt); 1993EXPORT_SYMBOL(sock_no_shutdown); 1994EXPORT_SYMBOL(sock_no_socketpair); 1995EXPORT_SYMBOL(sock_rfree); 1996EXPORT_SYMBOL(sock_setsockopt); 1997EXPORT_SYMBOL(sock_wfree); 1998EXPORT_SYMBOL(sock_wmalloc); 1999EXPORT_SYMBOL(sock_i_uid); 2000EXPORT_SYMBOL(sock_i_ino); 2001EXPORT_SYMBOL(sysctl_optmem_max); 2002#ifdef CONFIG_SYSCTL 2003EXPORT_SYMBOL(sysctl_rmem_max); 2004EXPORT_SYMBOL(sysctl_wmem_max); 2005#endif 2006