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 1.1.1.1 2007/10/11 23:35:15 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-IUCV" , 158 "sk_lock-AF_RXRPC" , "sk_lock-AF_MAX" 159}; 160static const char *af_family_slock_key_strings[AF_MAX+1] = { 161 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" , 162 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK", 163 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" , 164 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" , 165 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" , 166 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" , 167 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" , 168 "slock-21" , "slock-AF_SNA" , "slock-AF_IRDA" , 169 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" , 170 "slock-27" , "slock-28" , "slock-29" , 171 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" , 172 "slock-AF_RXRPC" , "slock-AF_MAX" 173}; 174#endif 175 176/* 177 * sk_callback_lock locking rules are per-address-family, 178 * so split the lock classes by using a per-AF key: 179 */ 180static struct lock_class_key af_callback_keys[AF_MAX]; 181 182/* Take into consideration the size of the struct sk_buff overhead in the 183 * determination of these values, since that is non-constant across 184 * platforms. This makes socket queueing behavior and performance 185 * not depend upon such differences. 186 */ 187#define _SK_MEM_PACKETS 256 188#define _SK_MEM_OVERHEAD (sizeof(struct sk_buff) + 256) 189#define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 190#define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 191 192/* Run time adjustable parameters. */ 193__u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX; 194__u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX; 195__u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX; 196__u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX; 197 198/* Maximal space eaten by iovec or ancilliary data plus some space */ 199int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512); 200 201static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen) 202{ 203 struct timeval tv; 204 205 if (optlen < sizeof(tv)) 206 return -EINVAL; 207 if (copy_from_user(&tv, optval, sizeof(tv))) 208 return -EFAULT; 209 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC) 210 return -EDOM; 211 212 if (tv.tv_sec < 0) { 213 static int warned = 0; 214 *timeo_p = 0; 215 if (warned < 10 && net_ratelimit()) 216 warned++; 217 printk(KERN_INFO "sock_set_timeout: `%s' (pid %d) " 218 "tries to set negative timeout\n", 219 current->comm, current->pid); 220 return 0; 221 } 222 *timeo_p = MAX_SCHEDULE_TIMEOUT; 223 if (tv.tv_sec == 0 && tv.tv_usec == 0) 224 return 0; 225 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1)) 226 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ); 227 return 0; 228} 229 230static void sock_warn_obsolete_bsdism(const char *name) 231{ 232 static int warned; 233 static char warncomm[TASK_COMM_LEN]; 234 if (strcmp(warncomm, current->comm) && warned < 5) { 235 strcpy(warncomm, current->comm); 236 printk(KERN_WARNING "process `%s' is using obsolete " 237 "%s SO_BSDCOMPAT\n", warncomm, name); 238 warned++; 239 } 240} 241 242static void sock_disable_timestamp(struct sock *sk) 243{ 244 if (sock_flag(sk, SOCK_TIMESTAMP)) { 245 sock_reset_flag(sk, SOCK_TIMESTAMP); 246 net_disable_timestamp(); 247 } 248} 249 250 251int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 252{ 253 int err = 0; 254 int skb_len; 255 256 /* Cast skb->rcvbuf to unsigned... It's pointless, but reduces 257 number of warnings when compiling with -W --ANK 258 */ 259 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >= 260 (unsigned)sk->sk_rcvbuf) { 261 err = -ENOMEM; 262 goto out; 263 } 264 265 err = sk_filter(sk, skb); 266 if (err) 267 goto out; 268 269 skb->dev = NULL; 270 skb_set_owner_r(skb, sk); 271 272 /* Cache the SKB length before we tack it onto the receive 273 * queue. Once it is added it no longer belongs to us and 274 * may be freed by other threads of control pulling packets 275 * from the queue. 276 */ 277 skb_len = skb->len; 278 279 skb_queue_tail(&sk->sk_receive_queue, skb); 280 281 if (!sock_flag(sk, SOCK_DEAD)) 282 sk->sk_data_ready(sk, skb_len); 283out: 284 return err; 285} 286EXPORT_SYMBOL(sock_queue_rcv_skb); 287 288int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested) 289{ 290 int rc = NET_RX_SUCCESS; 291 292 if (sk_filter(sk, skb)) 293 goto discard_and_relse; 294 295 skb->dev = NULL; 296 297 if (nested) 298 bh_lock_sock_nested(sk); 299 else 300 bh_lock_sock(sk); 301 if (!sock_owned_by_user(sk)) { 302 /* 303 * trylock + unlock semantics: 304 */ 305 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_); 306 307 rc = sk->sk_backlog_rcv(sk, skb); 308 309 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_); 310 } else 311 sk_add_backlog(sk, skb); 312 bh_unlock_sock(sk); 313out: 314 sock_put(sk); 315 return rc; 316discard_and_relse: 317 kfree_skb(skb); 318 goto out; 319} 320EXPORT_SYMBOL(sk_receive_skb); 321 322struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie) 323{ 324 struct dst_entry *dst = sk->sk_dst_cache; 325 326 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) { 327 sk->sk_dst_cache = NULL; 328 dst_release(dst); 329 return NULL; 330 } 331 332 return dst; 333} 334EXPORT_SYMBOL(__sk_dst_check); 335 336struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie) 337{ 338 struct dst_entry *dst = sk_dst_get(sk); 339 340 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) { 341 sk_dst_reset(sk); 342 dst_release(dst); 343 return NULL; 344 } 345 346 return dst; 347} 348EXPORT_SYMBOL(sk_dst_check); 349 350/* 351 * This is meant for all protocols to use and covers goings on 352 * at the socket level. Everything here is generic. 353 */ 354 355int sock_setsockopt(struct socket *sock, int level, int optname, 356 char __user *optval, int optlen) 357{ 358 struct sock *sk=sock->sk; 359 struct sk_filter *filter; 360 int val; 361 int valbool; 362 struct linger ling; 363 int ret = 0; 364 365 /* 366 * Options without arguments 367 */ 368 369#ifdef SO_DONTLINGER /* Compatibility item... */ 370 if (optname == SO_DONTLINGER) { 371 lock_sock(sk); 372 sock_reset_flag(sk, SOCK_LINGER); 373 release_sock(sk); 374 return 0; 375 } 376#endif 377 378 if (optlen < sizeof(int)) 379 return -EINVAL; 380 381 if (get_user(val, (int __user *)optval)) 382 return -EFAULT; 383 384 valbool = val?1:0; 385 386 lock_sock(sk); 387 388 switch(optname) { 389 case SO_DEBUG: 390 if (val && !capable(CAP_NET_ADMIN)) { 391 ret = -EACCES; 392 } 393 else if (valbool) 394 sock_set_flag(sk, SOCK_DBG); 395 else 396 sock_reset_flag(sk, SOCK_DBG); 397 break; 398 case SO_REUSEADDR: 399 sk->sk_reuse = valbool; 400 break; 401 case SO_TYPE: 402 case SO_ERROR: 403 ret = -ENOPROTOOPT; 404 break; 405 case SO_DONTROUTE: 406 if (valbool) 407 sock_set_flag(sk, SOCK_LOCALROUTE); 408 else 409 sock_reset_flag(sk, SOCK_LOCALROUTE); 410 break; 411 case SO_BROADCAST: 412 sock_valbool_flag(sk, SOCK_BROADCAST, valbool); 413 break; 414 case SO_SNDBUF: 415 /* Don't error on this BSD doesn't and if you think 416 about it this is right. Otherwise apps have to 417 play 'guess the biggest size' games. RCVBUF/SNDBUF 418 are treated in BSD as hints */ 419 420 if (val > sysctl_wmem_max) 421 val = sysctl_wmem_max; 422set_sndbuf: 423 sk->sk_userlocks |= SOCK_SNDBUF_LOCK; 424 if ((val * 2) < SOCK_MIN_SNDBUF) 425 sk->sk_sndbuf = SOCK_MIN_SNDBUF; 426 else 427 sk->sk_sndbuf = val * 2; 428 429 /* 430 * Wake up sending tasks if we 431 * upped the value. 432 */ 433 sk->sk_write_space(sk); 434 break; 435 436 case SO_SNDBUFFORCE: 437 if (!capable(CAP_NET_ADMIN)) { 438 ret = -EPERM; 439 break; 440 } 441 goto set_sndbuf; 442 443 case SO_RCVBUF: 444 /* Don't error on this BSD doesn't and if you think 445 about it this is right. Otherwise apps have to 446 play 'guess the biggest size' games. RCVBUF/SNDBUF 447 are treated in BSD as hints */ 448 449 if (val > sysctl_rmem_max) 450 val = sysctl_rmem_max; 451set_rcvbuf: 452 sk->sk_userlocks |= SOCK_RCVBUF_LOCK; 453 /* 454 * We double it on the way in to account for 455 * "struct sk_buff" etc. overhead. Applications 456 * assume that the SO_RCVBUF setting they make will 457 * allow that much actual data to be received on that 458 * socket. 459 * 460 * Applications are unaware that "struct sk_buff" and 461 * other overheads allocate from the receive buffer 462 * during socket buffer allocation. 463 * 464 * And after considering the possible alternatives, 465 * returning the value we actually used in getsockopt 466 * is the most desirable behavior. 467 */ 468 if ((val * 2) < SOCK_MIN_RCVBUF) 469 sk->sk_rcvbuf = SOCK_MIN_RCVBUF; 470 else 471 sk->sk_rcvbuf = val * 2; 472 break; 473 474 case SO_RCVBUFFORCE: 475 if (!capable(CAP_NET_ADMIN)) { 476 ret = -EPERM; 477 break; 478 } 479 goto set_rcvbuf; 480 481 case SO_KEEPALIVE: 482#ifdef CONFIG_INET 483 if (sk->sk_protocol == IPPROTO_TCP) 484 tcp_set_keepalive(sk, valbool); 485#endif 486 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool); 487 break; 488 489 case SO_OOBINLINE: 490 sock_valbool_flag(sk, SOCK_URGINLINE, valbool); 491 break; 492 493 case SO_NO_CHECK: 494 sk->sk_no_check = valbool; 495 break; 496 497 case SO_PRIORITY: 498 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN)) 499 sk->sk_priority = val; 500 else 501 ret = -EPERM; 502 break; 503 504 case SO_LINGER: 505 if (optlen < sizeof(ling)) { 506 ret = -EINVAL; /* 1003.1g */ 507 break; 508 } 509 if (copy_from_user(&ling,optval,sizeof(ling))) { 510 ret = -EFAULT; 511 break; 512 } 513 if (!ling.l_onoff) 514 sock_reset_flag(sk, SOCK_LINGER); 515 else { 516#if (BITS_PER_LONG == 32) 517 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ) 518 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT; 519 else 520#endif 521 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ; 522 sock_set_flag(sk, SOCK_LINGER); 523 } 524 break; 525 526 case SO_BSDCOMPAT: 527 sock_warn_obsolete_bsdism("setsockopt"); 528 break; 529 530 case SO_PASSCRED: 531 if (valbool) 532 set_bit(SOCK_PASSCRED, &sock->flags); 533 else 534 clear_bit(SOCK_PASSCRED, &sock->flags); 535 break; 536 537 case SO_TIMESTAMP: 538 case SO_TIMESTAMPNS: 539 if (valbool) { 540 if (optname == SO_TIMESTAMP) 541 sock_reset_flag(sk, SOCK_RCVTSTAMPNS); 542 else 543 sock_set_flag(sk, SOCK_RCVTSTAMPNS); 544 sock_set_flag(sk, SOCK_RCVTSTAMP); 545 sock_enable_timestamp(sk); 546 } else { 547 sock_reset_flag(sk, SOCK_RCVTSTAMP); 548 sock_reset_flag(sk, SOCK_RCVTSTAMPNS); 549 } 550 break; 551 552 case SO_RCVLOWAT: 553 if (val < 0) 554 val = INT_MAX; 555 sk->sk_rcvlowat = val ? : 1; 556 break; 557 558 case SO_RCVTIMEO: 559 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen); 560 break; 561 562 case SO_SNDTIMEO: 563 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen); 564 break; 565 566#ifdef CONFIG_NETDEVICES 567 case SO_BINDTODEVICE: 568 { 569 char devname[IFNAMSIZ]; 570 571 /* Sorry... */ 572 if (!capable(CAP_NET_RAW)) { 573 ret = -EPERM; 574 break; 575 } 576 577 /* Bind this socket to a particular device like "eth0", 578 * as specified in the passed interface name. If the 579 * name is "" or the option length is zero the socket 580 * is not bound. 581 */ 582 583 if (!valbool) { 584 sk->sk_bound_dev_if = 0; 585 } else { 586 if (optlen > IFNAMSIZ - 1) 587 optlen = IFNAMSIZ - 1; 588 memset(devname, 0, sizeof(devname)); 589 if (copy_from_user(devname, optval, optlen)) { 590 ret = -EFAULT; 591 break; 592 } 593 594 /* Remove any cached route for this socket. */ 595 sk_dst_reset(sk); 596 597 if (devname[0] == '\0') { 598 sk->sk_bound_dev_if = 0; 599 } else { 600 struct net_device *dev = dev_get_by_name(devname); 601 if (!dev) { 602 ret = -ENODEV; 603 break; 604 } 605 sk->sk_bound_dev_if = dev->ifindex; 606 dev_put(dev); 607 } 608 } 609 break; 610 } 611#endif 612 613 614 case SO_ATTACH_FILTER: 615 ret = -EINVAL; 616 if (optlen == sizeof(struct sock_fprog)) { 617 struct sock_fprog fprog; 618 619 ret = -EFAULT; 620 if (copy_from_user(&fprog, optval, sizeof(fprog))) 621 break; 622 623 ret = sk_attach_filter(&fprog, sk); 624 } 625 break; 626 627 case SO_DETACH_FILTER: 628 rcu_read_lock_bh(); 629 filter = rcu_dereference(sk->sk_filter); 630 if (filter) { 631 rcu_assign_pointer(sk->sk_filter, NULL); 632 sk_filter_release(sk, filter); 633 rcu_read_unlock_bh(); 634 break; 635 } 636 rcu_read_unlock_bh(); 637 ret = -ENONET; 638 break; 639 640 case SO_PASSSEC: 641 if (valbool) 642 set_bit(SOCK_PASSSEC, &sock->flags); 643 else 644 clear_bit(SOCK_PASSSEC, &sock->flags); 645 break; 646 647 /* We implement the SO_SNDLOWAT etc to 648 not be settable (1003.1g 5.3) */ 649 default: 650 ret = -ENOPROTOOPT; 651 break; 652 } 653 release_sock(sk); 654 return ret; 655} 656 657 658int sock_getsockopt(struct socket *sock, int level, int optname, 659 char __user *optval, int __user *optlen) 660{ 661 struct sock *sk = sock->sk; 662 663 union { 664 int val; 665 struct linger ling; 666 struct timeval tm; 667 } v; 668 669 unsigned int lv = sizeof(int); 670 int len; 671 672 if (get_user(len, optlen)) 673 return -EFAULT; 674 if (len < 0) 675 return -EINVAL; 676 677 switch(optname) { 678 case SO_DEBUG: 679 v.val = sock_flag(sk, SOCK_DBG); 680 break; 681 682 case SO_DONTROUTE: 683 v.val = sock_flag(sk, SOCK_LOCALROUTE); 684 break; 685 686 case SO_BROADCAST: 687 v.val = !!sock_flag(sk, SOCK_BROADCAST); 688 break; 689 690 case SO_SNDBUF: 691 v.val = sk->sk_sndbuf; 692 break; 693 694 case SO_RCVBUF: 695 v.val = sk->sk_rcvbuf; 696 break; 697 698 case SO_REUSEADDR: 699 v.val = sk->sk_reuse; 700 break; 701 702 case SO_KEEPALIVE: 703 v.val = !!sock_flag(sk, SOCK_KEEPOPEN); 704 break; 705 706 case SO_TYPE: 707 v.val = sk->sk_type; 708 break; 709 710 case SO_ERROR: 711 v.val = -sock_error(sk); 712 if (v.val==0) 713 v.val = xchg(&sk->sk_err_soft, 0); 714 break; 715 716 case SO_OOBINLINE: 717 v.val = !!sock_flag(sk, SOCK_URGINLINE); 718 break; 719 720 case SO_NO_CHECK: 721 v.val = sk->sk_no_check; 722 break; 723 724 case SO_PRIORITY: 725 v.val = sk->sk_priority; 726 break; 727 728 case SO_LINGER: 729 lv = sizeof(v.ling); 730 v.ling.l_onoff = !!sock_flag(sk, SOCK_LINGER); 731 v.ling.l_linger = sk->sk_lingertime / HZ; 732 break; 733 734 case SO_BSDCOMPAT: 735 sock_warn_obsolete_bsdism("getsockopt"); 736 break; 737 738 case SO_TIMESTAMP: 739 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && 740 !sock_flag(sk, SOCK_RCVTSTAMPNS); 741 break; 742 743 case SO_TIMESTAMPNS: 744 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS); 745 break; 746 747 case SO_RCVTIMEO: 748 lv=sizeof(struct timeval); 749 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) { 750 v.tm.tv_sec = 0; 751 v.tm.tv_usec = 0; 752 } else { 753 v.tm.tv_sec = sk->sk_rcvtimeo / HZ; 754 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ; 755 } 756 break; 757 758 case SO_SNDTIMEO: 759 lv=sizeof(struct timeval); 760 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) { 761 v.tm.tv_sec = 0; 762 v.tm.tv_usec = 0; 763 } else { 764 v.tm.tv_sec = sk->sk_sndtimeo / HZ; 765 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ; 766 } 767 break; 768 769 case SO_RCVLOWAT: 770 v.val = sk->sk_rcvlowat; 771 break; 772 773 case SO_SNDLOWAT: 774 v.val=1; 775 break; 776 777 case SO_PASSCRED: 778 v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0; 779 break; 780 781 case SO_PEERCRED: 782 if (len > sizeof(sk->sk_peercred)) 783 len = sizeof(sk->sk_peercred); 784 if (copy_to_user(optval, &sk->sk_peercred, len)) 785 return -EFAULT; 786 goto lenout; 787 788 case SO_PEERNAME: 789 { 790 char address[128]; 791 792 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2)) 793 return -ENOTCONN; 794 if (lv < len) 795 return -EINVAL; 796 if (copy_to_user(optval, address, len)) 797 return -EFAULT; 798 goto lenout; 799 } 800 801 /* Dubious BSD thing... Probably nobody even uses it, but 802 * the UNIX standard wants it for whatever reason... -DaveM 803 */ 804 case SO_ACCEPTCONN: 805 v.val = sk->sk_state == TCP_LISTEN; 806 break; 807 808 case SO_PASSSEC: 809 v.val = test_bit(SOCK_PASSSEC, &sock->flags) ? 1 : 0; 810 break; 811 812 case SO_PEERSEC: 813 return security_socket_getpeersec_stream(sock, optval, optlen, len); 814 815 default: 816 return -ENOPROTOOPT; 817 } 818 819 if (len > lv) 820 len = lv; 821 if (copy_to_user(optval, &v, len)) 822 return -EFAULT; 823lenout: 824 if (put_user(len, optlen)) 825 return -EFAULT; 826 return 0; 827} 828 829/* 830 * Initialize an sk_lock. 831 * 832 * (We also register the sk_lock with the lock validator.) 833 */ 834static inline void sock_lock_init(struct sock *sk) 835{ 836 sock_lock_init_class_and_name(sk, 837 af_family_slock_key_strings[sk->sk_family], 838 af_family_slock_keys + sk->sk_family, 839 af_family_key_strings[sk->sk_family], 840 af_family_keys + sk->sk_family); 841} 842 843/** 844 * sk_alloc - All socket objects are allocated here 845 * @family: protocol family 846 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 847 * @prot: struct proto associated with this new sock instance 848 * @zero_it: if we should zero the newly allocated sock 849 */ 850struct sock *sk_alloc(int family, gfp_t priority, 851 struct proto *prot, int zero_it) 852{ 853 struct sock *sk = NULL; 854 struct kmem_cache *slab = prot->slab; 855 856 if (slab != NULL) 857 sk = kmem_cache_alloc(slab, priority); 858 else 859 sk = kmalloc(prot->obj_size, priority); 860 861 if (sk) { 862 if (zero_it) { 863 memset(sk, 0, prot->obj_size); 864 sk->sk_family = family; 865 /* 866 * See comment in struct sock definition to understand 867 * why we need sk_prot_creator -acme 868 */ 869 sk->sk_prot = sk->sk_prot_creator = prot; 870 sock_lock_init(sk); 871 } 872 873 if (security_sk_alloc(sk, family, priority)) 874 goto out_free; 875 876 if (!try_module_get(prot->owner)) 877 goto out_free; 878 } 879 return sk; 880 881out_free: 882 if (slab != NULL) 883 kmem_cache_free(slab, sk); 884 else 885 kfree(sk); 886 return NULL; 887} 888 889void sk_free(struct sock *sk) 890{ 891 struct sk_filter *filter; 892 struct module *owner = sk->sk_prot_creator->owner; 893 894 if (sk->sk_destruct) 895 sk->sk_destruct(sk); 896 897 filter = rcu_dereference(sk->sk_filter); 898 if (filter) { 899 sk_filter_release(sk, filter); 900 rcu_assign_pointer(sk->sk_filter, NULL); 901 } 902 903 sock_disable_timestamp(sk); 904 905 if (atomic_read(&sk->sk_omem_alloc)) 906 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n", 907 __FUNCTION__, atomic_read(&sk->sk_omem_alloc)); 908 909 security_sk_free(sk); 910 if (sk->sk_prot_creator->slab != NULL) 911 kmem_cache_free(sk->sk_prot_creator->slab, sk); 912 else 913 kfree(sk); 914 module_put(owner); 915} 916 917struct sock *sk_clone(const struct sock *sk, const gfp_t priority) 918{ 919 struct sock *newsk = sk_alloc(sk->sk_family, priority, sk->sk_prot, 0); 920 921 if (newsk != NULL) { 922 struct sk_filter *filter; 923 924 sock_copy(newsk, sk); 925 926 /* SANITY */ 927 sk_node_init(&newsk->sk_node); 928 sock_lock_init(newsk); 929 bh_lock_sock(newsk); 930 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL; 931 932 atomic_set(&newsk->sk_rmem_alloc, 0); 933 atomic_set(&newsk->sk_wmem_alloc, 0); 934 atomic_set(&newsk->sk_omem_alloc, 0); 935 skb_queue_head_init(&newsk->sk_receive_queue); 936 skb_queue_head_init(&newsk->sk_write_queue); 937#ifdef CONFIG_NET_DMA 938 skb_queue_head_init(&newsk->sk_async_wait_queue); 939#endif 940 941 rwlock_init(&newsk->sk_dst_lock); 942 rwlock_init(&newsk->sk_callback_lock); 943 lockdep_set_class(&newsk->sk_callback_lock, 944 af_callback_keys + newsk->sk_family); 945 946 newsk->sk_dst_cache = NULL; 947 newsk->sk_wmem_queued = 0; 948 newsk->sk_forward_alloc = 0; 949 newsk->sk_send_head = NULL; 950 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK; 951 952 sock_reset_flag(newsk, SOCK_DONE); 953 skb_queue_head_init(&newsk->sk_error_queue); 954 955 filter = newsk->sk_filter; 956 if (filter != NULL) 957 sk_filter_charge(newsk, filter); 958 959 if (unlikely(xfrm_sk_clone_policy(newsk))) { 960 /* It is still raw copy of parent, so invalidate 961 * destructor and make plain sk_free() */ 962 newsk->sk_destruct = NULL; 963 sk_free(newsk); 964 newsk = NULL; 965 goto out; 966 } 967 968 newsk->sk_err = 0; 969 newsk->sk_priority = 0; 970 atomic_set(&newsk->sk_refcnt, 2); 971 972 /* 973 * Increment the counter in the same struct proto as the master 974 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that 975 * is the same as sk->sk_prot->socks, as this field was copied 976 * with memcpy). 977 * 978 * This _changes_ the previous behaviour, where 979 * tcp_create_openreq_child always was incrementing the 980 * equivalent to tcp_prot->socks (inet_sock_nr), so this have 981 * to be taken into account in all callers. -acme 982 */ 983 sk_refcnt_debug_inc(newsk); 984 newsk->sk_socket = NULL; 985 newsk->sk_sleep = NULL; 986 987 if (newsk->sk_prot->sockets_allocated) 988 atomic_inc(newsk->sk_prot->sockets_allocated); 989 } 990out: 991 return newsk; 992} 993 994EXPORT_SYMBOL_GPL(sk_clone); 995 996void sk_setup_caps(struct sock *sk, struct dst_entry *dst) 997{ 998 __sk_dst_set(sk, dst); 999 sk->sk_route_caps = dst->dev->features; 1000 if (sk->sk_route_caps & NETIF_F_GSO) 1001 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE; 1002 if (sk_can_gso(sk)) { 1003 if (dst->header_len) 1004 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 1005 else 1006 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM; 1007 } 1008} 1009EXPORT_SYMBOL_GPL(sk_setup_caps); 1010 1011void __init sk_init(void) 1012{ 1013 if (num_physpages <= 4096) { 1014 sysctl_wmem_max = 32767; 1015 sysctl_rmem_max = 32767; 1016 sysctl_wmem_default = 32767; 1017 sysctl_rmem_default = 32767; 1018 } else if (num_physpages >= 131072) { 1019 sysctl_wmem_max = 131071; 1020 sysctl_rmem_max = 131071; 1021 } 1022} 1023 1024/* 1025 * Simple resource managers for sockets. 1026 */ 1027 1028 1029/* 1030 * Write buffer destructor automatically called from kfree_skb. 1031 */ 1032void sock_wfree(struct sk_buff *skb) 1033{ 1034 struct sock *sk = skb->sk; 1035 1036 /* In case it might be waiting for more memory. */ 1037 atomic_sub(skb->truesize, &sk->sk_wmem_alloc); 1038 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) 1039 sk->sk_write_space(sk); 1040 sock_put(sk); 1041} 1042 1043/* 1044 * Read buffer destructor automatically called from kfree_skb. 1045 */ 1046void sock_rfree(struct sk_buff *skb) 1047{ 1048 struct sock *sk = skb->sk; 1049 1050 atomic_sub(skb->truesize, &sk->sk_rmem_alloc); 1051} 1052 1053 1054int sock_i_uid(struct sock *sk) 1055{ 1056 int uid; 1057 1058 read_lock(&sk->sk_callback_lock); 1059 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0; 1060 read_unlock(&sk->sk_callback_lock); 1061 return uid; 1062} 1063 1064unsigned long sock_i_ino(struct sock *sk) 1065{ 1066 unsigned long ino; 1067 1068 read_lock(&sk->sk_callback_lock); 1069 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0; 1070 read_unlock(&sk->sk_callback_lock); 1071 return ino; 1072} 1073 1074/* 1075 * Allocate a skb from the socket's send buffer. 1076 */ 1077struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 1078 gfp_t priority) 1079{ 1080 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) { 1081 struct sk_buff * skb = alloc_skb(size, priority); 1082 if (skb) { 1083 skb_set_owner_w(skb, sk); 1084 return skb; 1085 } 1086 } 1087 return NULL; 1088} 1089 1090/* 1091 * Allocate a skb from the socket's receive buffer. 1092 */ 1093struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force, 1094 gfp_t priority) 1095{ 1096 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) { 1097 struct sk_buff *skb = alloc_skb(size, priority); 1098 if (skb) { 1099 skb_set_owner_r(skb, sk); 1100 return skb; 1101 } 1102 } 1103 return NULL; 1104} 1105 1106/* 1107 * Allocate a memory block from the socket's option memory buffer. 1108 */ 1109void *sock_kmalloc(struct sock *sk, int size, gfp_t priority) 1110{ 1111 if ((unsigned)size <= sysctl_optmem_max && 1112 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) { 1113 void *mem; 1114 /* First do the add, to avoid the race if kmalloc 1115 * might sleep. 1116 */ 1117 atomic_add(size, &sk->sk_omem_alloc); 1118 mem = kmalloc(size, priority); 1119 if (mem) 1120 return mem; 1121 atomic_sub(size, &sk->sk_omem_alloc); 1122 } 1123 return NULL; 1124} 1125 1126/* 1127 * Free an option memory block. 1128 */ 1129void sock_kfree_s(struct sock *sk, void *mem, int size) 1130{ 1131 kfree(mem); 1132 atomic_sub(size, &sk->sk_omem_alloc); 1133} 1134 1135/* It is almost wait_for_tcp_memory minus release_sock/lock_sock. 1136 I think, these locks should be removed for datagram sockets. 1137 */ 1138static long sock_wait_for_wmem(struct sock * sk, long timeo) 1139{ 1140 DEFINE_WAIT(wait); 1141 1142 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags); 1143 for (;;) { 1144 if (!timeo) 1145 break; 1146 if (signal_pending(current)) 1147 break; 1148 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 1149 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE); 1150 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) 1151 break; 1152 if (sk->sk_shutdown & SEND_SHUTDOWN) 1153 break; 1154 if (sk->sk_err) 1155 break; 1156 timeo = schedule_timeout(timeo); 1157 } 1158 finish_wait(sk->sk_sleep, &wait); 1159 return timeo; 1160} 1161 1162 1163/* 1164 * Generic send/receive buffer handlers 1165 */ 1166 1167static struct sk_buff *sock_alloc_send_pskb(struct sock *sk, 1168 unsigned long header_len, 1169 unsigned long data_len, 1170 int noblock, int *errcode) 1171{ 1172 struct sk_buff *skb; 1173 gfp_t gfp_mask; 1174 long timeo; 1175 int err; 1176 1177 gfp_mask = sk->sk_allocation; 1178 if (gfp_mask & __GFP_WAIT) 1179 gfp_mask |= __GFP_REPEAT; 1180 1181 timeo = sock_sndtimeo(sk, noblock); 1182 while (1) { 1183 err = sock_error(sk); 1184 if (err != 0) 1185 goto failure; 1186 1187 err = -EPIPE; 1188 if (sk->sk_shutdown & SEND_SHUTDOWN) 1189 goto failure; 1190 1191 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) { 1192 skb = alloc_skb(header_len, gfp_mask); 1193 if (skb) { 1194 int npages; 1195 int i; 1196 1197 /* No pages, we're done... */ 1198 if (!data_len) 1199 break; 1200 1201 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT; 1202 skb->truesize += data_len; 1203 skb_shinfo(skb)->nr_frags = npages; 1204 for (i = 0; i < npages; i++) { 1205 struct page *page; 1206 skb_frag_t *frag; 1207 1208 page = alloc_pages(sk->sk_allocation, 0); 1209 if (!page) { 1210 err = -ENOBUFS; 1211 skb_shinfo(skb)->nr_frags = i; 1212 kfree_skb(skb); 1213 goto failure; 1214 } 1215 1216 frag = &skb_shinfo(skb)->frags[i]; 1217 frag->page = page; 1218 frag->page_offset = 0; 1219 frag->size = (data_len >= PAGE_SIZE ? 1220 PAGE_SIZE : 1221 data_len); 1222 data_len -= PAGE_SIZE; 1223 } 1224 1225 /* Full success... */ 1226 break; 1227 } 1228 err = -ENOBUFS; 1229 goto failure; 1230 } 1231 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags); 1232 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 1233 err = -EAGAIN; 1234 if (!timeo) 1235 goto failure; 1236 if (signal_pending(current)) 1237 goto interrupted; 1238 timeo = sock_wait_for_wmem(sk, timeo); 1239 } 1240 1241 skb_set_owner_w(skb, sk); 1242 return skb; 1243 1244interrupted: 1245 err = sock_intr_errno(timeo); 1246failure: 1247 *errcode = err; 1248 return NULL; 1249} 1250 1251struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size, 1252 int noblock, int *errcode) 1253{ 1254 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode); 1255} 1256 1257static void __lock_sock(struct sock *sk) 1258{ 1259 DEFINE_WAIT(wait); 1260 1261 for (;;) { 1262 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait, 1263 TASK_UNINTERRUPTIBLE); 1264 spin_unlock_bh(&sk->sk_lock.slock); 1265 schedule(); 1266 spin_lock_bh(&sk->sk_lock.slock); 1267 if (!sock_owned_by_user(sk)) 1268 break; 1269 } 1270 finish_wait(&sk->sk_lock.wq, &wait); 1271} 1272 1273static void __release_sock(struct sock *sk) 1274{ 1275 struct sk_buff *skb = sk->sk_backlog.head; 1276 1277 do { 1278 sk->sk_backlog.head = sk->sk_backlog.tail = NULL; 1279 bh_unlock_sock(sk); 1280 1281 do { 1282 struct sk_buff *next = skb->next; 1283 1284 skb->next = NULL; 1285 sk->sk_backlog_rcv(sk, skb); 1286 1287 /* 1288 * We are in process context here with softirqs 1289 * disabled, use cond_resched_softirq() to preempt. 1290 * This is safe to do because we've taken the backlog 1291 * queue private: 1292 */ 1293 cond_resched_softirq(); 1294 1295 skb = next; 1296 } while (skb != NULL); 1297 1298 bh_lock_sock(sk); 1299 } while ((skb = sk->sk_backlog.head) != NULL); 1300} 1301 1302/** 1303 * sk_wait_data - wait for data to arrive at sk_receive_queue 1304 * @sk: sock to wait on 1305 * @timeo: for how long 1306 * 1307 * Now socket state including sk->sk_err is changed only under lock, 1308 * hence we may omit checks after joining wait queue. 1309 * We check receive queue before schedule() only as optimization; 1310 * it is very likely that release_sock() added new data. 1311 */ 1312int sk_wait_data(struct sock *sk, long *timeo) 1313{ 1314 int rc; 1315 DEFINE_WAIT(wait); 1316 1317 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE); 1318 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags); 1319 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue)); 1320 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags); 1321 finish_wait(sk->sk_sleep, &wait); 1322 return rc; 1323} 1324 1325EXPORT_SYMBOL(sk_wait_data); 1326 1327/* 1328 * Set of default routines for initialising struct proto_ops when 1329 * the protocol does not support a particular function. In certain 1330 * cases where it makes no sense for a protocol to have a "do nothing" 1331 * function, some default processing is provided. 1332 */ 1333 1334int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len) 1335{ 1336 return -EOPNOTSUPP; 1337} 1338 1339int sock_no_connect(struct socket *sock, struct sockaddr *saddr, 1340 int len, int flags) 1341{ 1342 return -EOPNOTSUPP; 1343} 1344 1345int sock_no_socketpair(struct socket *sock1, struct socket *sock2) 1346{ 1347 return -EOPNOTSUPP; 1348} 1349 1350int sock_no_accept(struct socket *sock, struct socket *newsock, int flags) 1351{ 1352 return -EOPNOTSUPP; 1353} 1354 1355int sock_no_getname(struct socket *sock, struct sockaddr *saddr, 1356 int *len, int peer) 1357{ 1358 return -EOPNOTSUPP; 1359} 1360 1361unsigned int sock_no_poll(struct file * file, struct socket *sock, poll_table *pt) 1362{ 1363 return 0; 1364} 1365 1366int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 1367{ 1368 return -EOPNOTSUPP; 1369} 1370 1371int sock_no_listen(struct socket *sock, int backlog) 1372{ 1373 return -EOPNOTSUPP; 1374} 1375 1376int sock_no_shutdown(struct socket *sock, int how) 1377{ 1378 return -EOPNOTSUPP; 1379} 1380 1381int sock_no_setsockopt(struct socket *sock, int level, int optname, 1382 char __user *optval, int optlen) 1383{ 1384 return -EOPNOTSUPP; 1385} 1386 1387int sock_no_getsockopt(struct socket *sock, int level, int optname, 1388 char __user *optval, int __user *optlen) 1389{ 1390 return -EOPNOTSUPP; 1391} 1392 1393int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m, 1394 size_t len) 1395{ 1396 return -EOPNOTSUPP; 1397} 1398 1399int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m, 1400 size_t len, int flags) 1401{ 1402 return -EOPNOTSUPP; 1403} 1404 1405int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma) 1406{ 1407 /* Mirror missing mmap method error code */ 1408 return -ENODEV; 1409} 1410 1411ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags) 1412{ 1413 ssize_t res; 1414 struct msghdr msg = {.msg_flags = flags}; 1415 struct kvec iov; 1416 char *kaddr = kmap(page); 1417 iov.iov_base = kaddr + offset; 1418 iov.iov_len = size; 1419 res = kernel_sendmsg(sock, &msg, &iov, 1, size); 1420 kunmap(page); 1421 return res; 1422} 1423 1424/* 1425 * Default Socket Callbacks 1426 */ 1427 1428static void sock_def_wakeup(struct sock *sk) 1429{ 1430 read_lock(&sk->sk_callback_lock); 1431 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) 1432 wake_up_interruptible_all(sk->sk_sleep); 1433 read_unlock(&sk->sk_callback_lock); 1434} 1435 1436static void sock_def_error_report(struct sock *sk) 1437{ 1438 read_lock(&sk->sk_callback_lock); 1439 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) 1440 wake_up_interruptible(sk->sk_sleep); 1441 sk_wake_async(sk,0,POLL_ERR); 1442 read_unlock(&sk->sk_callback_lock); 1443} 1444 1445static void sock_def_readable(struct sock *sk, int len) 1446{ 1447 read_lock(&sk->sk_callback_lock); 1448 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) 1449 wake_up_interruptible(sk->sk_sleep); 1450 sk_wake_async(sk,1,POLL_IN); 1451 read_unlock(&sk->sk_callback_lock); 1452} 1453 1454static void sock_def_write_space(struct sock *sk) 1455{ 1456 read_lock(&sk->sk_callback_lock); 1457 1458 /* Do not wake up a writer until he can make "significant" 1459 * progress. --DaveM 1460 */ 1461 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) { 1462 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) 1463 wake_up_interruptible(sk->sk_sleep); 1464 1465 /* Should agree with poll, otherwise some programs break */ 1466 if (sock_writeable(sk)) 1467 sk_wake_async(sk, 2, POLL_OUT); 1468 } 1469 1470 read_unlock(&sk->sk_callback_lock); 1471} 1472 1473static void sock_def_destruct(struct sock *sk) 1474{ 1475 kfree(sk->sk_protinfo); 1476} 1477 1478void sk_send_sigurg(struct sock *sk) 1479{ 1480 if (sk->sk_socket && sk->sk_socket->file) 1481 if (send_sigurg(&sk->sk_socket->file->f_owner)) 1482 sk_wake_async(sk, 3, POLL_PRI); 1483} 1484 1485void sk_reset_timer(struct sock *sk, struct timer_list* timer, 1486 unsigned long expires) 1487{ 1488 if (!mod_timer(timer, expires)) 1489 sock_hold(sk); 1490} 1491 1492EXPORT_SYMBOL(sk_reset_timer); 1493 1494void sk_stop_timer(struct sock *sk, struct timer_list* timer) 1495{ 1496 if (timer_pending(timer) && del_timer(timer)) 1497 __sock_put(sk); 1498} 1499 1500EXPORT_SYMBOL(sk_stop_timer); 1501 1502void sock_init_data(struct socket *sock, struct sock *sk) 1503{ 1504 skb_queue_head_init(&sk->sk_receive_queue); 1505 skb_queue_head_init(&sk->sk_write_queue); 1506 skb_queue_head_init(&sk->sk_error_queue); 1507#ifdef CONFIG_NET_DMA 1508 skb_queue_head_init(&sk->sk_async_wait_queue); 1509#endif 1510 1511 sk->sk_send_head = NULL; 1512 1513 init_timer(&sk->sk_timer); 1514 1515 sk->sk_allocation = GFP_KERNEL; 1516 sk->sk_rcvbuf = sysctl_rmem_default; 1517 sk->sk_sndbuf = sysctl_wmem_default; 1518 sk->sk_state = TCP_CLOSE; 1519 sk->sk_socket = sock; 1520 1521 sock_set_flag(sk, SOCK_ZAPPED); 1522 1523 if (sock) { 1524 sk->sk_type = sock->type; 1525 sk->sk_sleep = &sock->wait; 1526 sock->sk = sk; 1527 } else 1528 sk->sk_sleep = NULL; 1529 1530 rwlock_init(&sk->sk_dst_lock); 1531 rwlock_init(&sk->sk_callback_lock); 1532 lockdep_set_class(&sk->sk_callback_lock, 1533 af_callback_keys + sk->sk_family); 1534 1535 sk->sk_state_change = sock_def_wakeup; 1536 sk->sk_data_ready = sock_def_readable; 1537 sk->sk_write_space = sock_def_write_space; 1538 sk->sk_error_report = sock_def_error_report; 1539 sk->sk_destruct = sock_def_destruct; 1540 1541 sk->sk_sndmsg_page = NULL; 1542 sk->sk_sndmsg_off = 0; 1543 1544 sk->sk_peercred.pid = 0; 1545 sk->sk_peercred.uid = -1; 1546 sk->sk_peercred.gid = -1; 1547 sk->sk_write_pending = 0; 1548 sk->sk_rcvlowat = 1; 1549 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; 1550 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT; 1551 1552 sk->sk_stamp = ktime_set(-1L, -1L); 1553 1554 atomic_set(&sk->sk_refcnt, 1); 1555} 1556 1557void fastcall lock_sock_nested(struct sock *sk, int subclass) 1558{ 1559 might_sleep(); 1560 spin_lock_bh(&sk->sk_lock.slock); 1561 if (sk->sk_lock.owner) 1562 __lock_sock(sk); 1563 sk->sk_lock.owner = (void *)1; 1564 spin_unlock(&sk->sk_lock.slock); 1565 /* 1566 * The sk_lock has mutex_lock() semantics here: 1567 */ 1568 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_); 1569 local_bh_enable(); 1570} 1571 1572EXPORT_SYMBOL(lock_sock_nested); 1573 1574void fastcall release_sock(struct sock *sk) 1575{ 1576 /* 1577 * The sk_lock has mutex_unlock() semantics: 1578 */ 1579 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_); 1580 1581 spin_lock_bh(&sk->sk_lock.slock); 1582 if (sk->sk_backlog.tail) 1583 __release_sock(sk); 1584 sk->sk_lock.owner = NULL; 1585 if (waitqueue_active(&sk->sk_lock.wq)) 1586 wake_up(&sk->sk_lock.wq); 1587 spin_unlock_bh(&sk->sk_lock.slock); 1588} 1589EXPORT_SYMBOL(release_sock); 1590 1591int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp) 1592{ 1593 struct timeval tv; 1594 if (!sock_flag(sk, SOCK_TIMESTAMP)) 1595 sock_enable_timestamp(sk); 1596 tv = ktime_to_timeval(sk->sk_stamp); 1597 if (tv.tv_sec == -1) 1598 return -ENOENT; 1599 if (tv.tv_sec == 0) { 1600 sk->sk_stamp = ktime_get_real(); 1601 tv = ktime_to_timeval(sk->sk_stamp); 1602 } 1603 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0; 1604} 1605EXPORT_SYMBOL(sock_get_timestamp); 1606 1607int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp) 1608{ 1609 struct timespec ts; 1610 if (!sock_flag(sk, SOCK_TIMESTAMP)) 1611 sock_enable_timestamp(sk); 1612 ts = ktime_to_timespec(sk->sk_stamp); 1613 if (ts.tv_sec == -1) 1614 return -ENOENT; 1615 if (ts.tv_sec == 0) { 1616 sk->sk_stamp = ktime_get_real(); 1617 ts = ktime_to_timespec(sk->sk_stamp); 1618 } 1619 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0; 1620} 1621EXPORT_SYMBOL(sock_get_timestampns); 1622 1623void sock_enable_timestamp(struct sock *sk) 1624{ 1625 if (!sock_flag(sk, SOCK_TIMESTAMP)) { 1626 sock_set_flag(sk, SOCK_TIMESTAMP); 1627 net_enable_timestamp(); 1628 } 1629} 1630EXPORT_SYMBOL(sock_enable_timestamp); 1631 1632/* 1633 * Get a socket option on an socket. 1634 * 1635 * FIX: POSIX 1003.1g is very ambiguous here. It states that 1636 * asynchronous errors should be reported by getsockopt. We assume 1637 * this means if you specify SO_ERROR (otherwise whats the point of it). 1638 */ 1639int sock_common_getsockopt(struct socket *sock, int level, int optname, 1640 char __user *optval, int __user *optlen) 1641{ 1642 struct sock *sk = sock->sk; 1643 1644 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen); 1645} 1646 1647EXPORT_SYMBOL(sock_common_getsockopt); 1648 1649#ifdef CONFIG_COMPAT 1650int compat_sock_common_getsockopt(struct socket *sock, int level, int optname, 1651 char __user *optval, int __user *optlen) 1652{ 1653 struct sock *sk = sock->sk; 1654 1655 if (sk->sk_prot->compat_getsockopt != NULL) 1656 return sk->sk_prot->compat_getsockopt(sk, level, optname, 1657 optval, optlen); 1658 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen); 1659} 1660EXPORT_SYMBOL(compat_sock_common_getsockopt); 1661#endif 1662 1663int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock, 1664 struct msghdr *msg, size_t size, int flags) 1665{ 1666 struct sock *sk = sock->sk; 1667 int addr_len = 0; 1668 int err; 1669 1670 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT, 1671 flags & ~MSG_DONTWAIT, &addr_len); 1672 if (err >= 0) 1673 msg->msg_namelen = addr_len; 1674 return err; 1675} 1676 1677EXPORT_SYMBOL(sock_common_recvmsg); 1678 1679/* 1680 * Set socket options on an inet socket. 1681 */ 1682int sock_common_setsockopt(struct socket *sock, int level, int optname, 1683 char __user *optval, int optlen) 1684{ 1685 struct sock *sk = sock->sk; 1686 1687 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen); 1688} 1689 1690EXPORT_SYMBOL(sock_common_setsockopt); 1691 1692#ifdef CONFIG_COMPAT 1693int compat_sock_common_setsockopt(struct socket *sock, int level, int optname, 1694 char __user *optval, int optlen) 1695{ 1696 struct sock *sk = sock->sk; 1697 1698 if (sk->sk_prot->compat_setsockopt != NULL) 1699 return sk->sk_prot->compat_setsockopt(sk, level, optname, 1700 optval, optlen); 1701 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen); 1702} 1703EXPORT_SYMBOL(compat_sock_common_setsockopt); 1704#endif 1705 1706void sk_common_release(struct sock *sk) 1707{ 1708 if (sk->sk_prot->destroy) 1709 sk->sk_prot->destroy(sk); 1710 1711 /* 1712 * Observation: when sock_common_release is called, processes have 1713 * no access to socket. But net still has. 1714 * Step one, detach it from networking: 1715 * 1716 * A. Remove from hash tables. 1717 */ 1718 1719 sk->sk_prot->unhash(sk); 1720 1721 /* 1722 * In this point socket cannot receive new packets, but it is possible 1723 * that some packets are in flight because some CPU runs receiver and 1724 * did hash table lookup before we unhashed socket. They will achieve 1725 * receive queue and will be purged by socket destructor. 1726 * 1727 * Also we still have packets pending on receive queue and probably, 1728 * our own packets waiting in device queues. sock_destroy will drain 1729 * receive queue, but transmitted packets will delay socket destruction 1730 * until the last reference will be released. 1731 */ 1732 1733 sock_orphan(sk); 1734 1735 xfrm_sk_free_policy(sk); 1736 1737 sk_refcnt_debug_release(sk); 1738 sock_put(sk); 1739} 1740 1741EXPORT_SYMBOL(sk_common_release); 1742 1743static DEFINE_RWLOCK(proto_list_lock); 1744static LIST_HEAD(proto_list); 1745 1746int proto_register(struct proto *prot, int alloc_slab) 1747{ 1748 char *request_sock_slab_name = NULL; 1749 char *timewait_sock_slab_name; 1750 int rc = -ENOBUFS; 1751 1752 if (alloc_slab) { 1753 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0, 1754 SLAB_HWCACHE_ALIGN, NULL, NULL); 1755 1756 if (prot->slab == NULL) { 1757 printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n", 1758 prot->name); 1759 goto out; 1760 } 1761 1762 if (prot->rsk_prot != NULL) { 1763 static const char mask[] = "request_sock_%s"; 1764 1765 request_sock_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL); 1766 if (request_sock_slab_name == NULL) 1767 goto out_free_sock_slab; 1768 1769 sprintf(request_sock_slab_name, mask, prot->name); 1770 prot->rsk_prot->slab = kmem_cache_create(request_sock_slab_name, 1771 prot->rsk_prot->obj_size, 0, 1772 SLAB_HWCACHE_ALIGN, NULL, NULL); 1773 1774 if (prot->rsk_prot->slab == NULL) { 1775 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n", 1776 prot->name); 1777 goto out_free_request_sock_slab_name; 1778 } 1779 } 1780 1781 if (prot->twsk_prot != NULL) { 1782 static const char mask[] = "tw_sock_%s"; 1783 1784 timewait_sock_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL); 1785 1786 if (timewait_sock_slab_name == NULL) 1787 goto out_free_request_sock_slab; 1788 1789 sprintf(timewait_sock_slab_name, mask, prot->name); 1790 prot->twsk_prot->twsk_slab = 1791 kmem_cache_create(timewait_sock_slab_name, 1792 prot->twsk_prot->twsk_obj_size, 1793 0, SLAB_HWCACHE_ALIGN, 1794 NULL, NULL); 1795 if (prot->twsk_prot->twsk_slab == NULL) 1796 goto out_free_timewait_sock_slab_name; 1797 } 1798 } 1799 1800 write_lock(&proto_list_lock); 1801 list_add(&prot->node, &proto_list); 1802 write_unlock(&proto_list_lock); 1803 rc = 0; 1804out: 1805 return rc; 1806out_free_timewait_sock_slab_name: 1807 kfree(timewait_sock_slab_name); 1808out_free_request_sock_slab: 1809 if (prot->rsk_prot && prot->rsk_prot->slab) { 1810 kmem_cache_destroy(prot->rsk_prot->slab); 1811 prot->rsk_prot->slab = NULL; 1812 } 1813out_free_request_sock_slab_name: 1814 kfree(request_sock_slab_name); 1815out_free_sock_slab: 1816 kmem_cache_destroy(prot->slab); 1817 prot->slab = NULL; 1818 goto out; 1819} 1820 1821EXPORT_SYMBOL(proto_register); 1822 1823void proto_unregister(struct proto *prot) 1824{ 1825 write_lock(&proto_list_lock); 1826 list_del(&prot->node); 1827 write_unlock(&proto_list_lock); 1828 1829 if (prot->slab != NULL) { 1830 kmem_cache_destroy(prot->slab); 1831 prot->slab = NULL; 1832 } 1833 1834 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) { 1835 const char *name = kmem_cache_name(prot->rsk_prot->slab); 1836 1837 kmem_cache_destroy(prot->rsk_prot->slab); 1838 kfree(name); 1839 prot->rsk_prot->slab = NULL; 1840 } 1841 1842 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) { 1843 const char *name = kmem_cache_name(prot->twsk_prot->twsk_slab); 1844 1845 kmem_cache_destroy(prot->twsk_prot->twsk_slab); 1846 kfree(name); 1847 prot->twsk_prot->twsk_slab = NULL; 1848 } 1849} 1850 1851EXPORT_SYMBOL(proto_unregister); 1852 1853#ifdef CONFIG_PROC_FS 1854static inline struct proto *__proto_head(void) 1855{ 1856 return list_entry(proto_list.next, struct proto, node); 1857} 1858 1859static inline struct proto *proto_head(void) 1860{ 1861 return list_empty(&proto_list) ? NULL : __proto_head(); 1862} 1863 1864static inline struct proto *proto_next(struct proto *proto) 1865{ 1866 return proto->node.next == &proto_list ? NULL : 1867 list_entry(proto->node.next, struct proto, node); 1868} 1869 1870static inline struct proto *proto_get_idx(loff_t pos) 1871{ 1872 struct proto *proto; 1873 loff_t i = 0; 1874 1875 list_for_each_entry(proto, &proto_list, node) 1876 if (i++ == pos) 1877 goto out; 1878 1879 proto = NULL; 1880out: 1881 return proto; 1882} 1883 1884static void *proto_seq_start(struct seq_file *seq, loff_t *pos) 1885{ 1886 read_lock(&proto_list_lock); 1887 return *pos ? proto_get_idx(*pos - 1) : SEQ_START_TOKEN; 1888} 1889 1890static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos) 1891{ 1892 ++*pos; 1893 return v == SEQ_START_TOKEN ? proto_head() : proto_next(v); 1894} 1895 1896static void proto_seq_stop(struct seq_file *seq, void *v) 1897{ 1898 read_unlock(&proto_list_lock); 1899} 1900 1901static char proto_method_implemented(const void *method) 1902{ 1903 return method == NULL ? 'n' : 'y'; 1904} 1905 1906static void proto_seq_printf(struct seq_file *seq, struct proto *proto) 1907{ 1908 seq_printf(seq, "%-9s %4u %6d %6d %-3s %6u %-3s %-10s " 1909 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n", 1910 proto->name, 1911 proto->obj_size, 1912 proto->sockets_allocated != NULL ? atomic_read(proto->sockets_allocated) : -1, 1913 proto->memory_allocated != NULL ? atomic_read(proto->memory_allocated) : -1, 1914 proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI", 1915 proto->max_header, 1916 proto->slab == NULL ? "no" : "yes", 1917 module_name(proto->owner), 1918 proto_method_implemented(proto->close), 1919 proto_method_implemented(proto->connect), 1920 proto_method_implemented(proto->disconnect), 1921 proto_method_implemented(proto->accept), 1922 proto_method_implemented(proto->ioctl), 1923 proto_method_implemented(proto->init), 1924 proto_method_implemented(proto->destroy), 1925 proto_method_implemented(proto->shutdown), 1926 proto_method_implemented(proto->setsockopt), 1927 proto_method_implemented(proto->getsockopt), 1928 proto_method_implemented(proto->sendmsg), 1929 proto_method_implemented(proto->recvmsg), 1930 proto_method_implemented(proto->sendpage), 1931 proto_method_implemented(proto->bind), 1932 proto_method_implemented(proto->backlog_rcv), 1933 proto_method_implemented(proto->hash), 1934 proto_method_implemented(proto->unhash), 1935 proto_method_implemented(proto->get_port), 1936 proto_method_implemented(proto->enter_memory_pressure)); 1937} 1938 1939static int proto_seq_show(struct seq_file *seq, void *v) 1940{ 1941 if (v == SEQ_START_TOKEN) 1942 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s", 1943 "protocol", 1944 "size", 1945 "sockets", 1946 "memory", 1947 "press", 1948 "maxhdr", 1949 "slab", 1950 "module", 1951 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n"); 1952 else 1953 proto_seq_printf(seq, v); 1954 return 0; 1955} 1956 1957static const struct seq_operations proto_seq_ops = { 1958 .start = proto_seq_start, 1959 .next = proto_seq_next, 1960 .stop = proto_seq_stop, 1961 .show = proto_seq_show, 1962}; 1963 1964static int proto_seq_open(struct inode *inode, struct file *file) 1965{ 1966 return seq_open(file, &proto_seq_ops); 1967} 1968 1969static const struct file_operations proto_seq_fops = { 1970 .owner = THIS_MODULE, 1971 .open = proto_seq_open, 1972 .read = seq_read, 1973 .llseek = seq_lseek, 1974 .release = seq_release, 1975}; 1976 1977static int __init proto_init(void) 1978{ 1979 /* register /proc/net/protocols */ 1980 return proc_net_fops_create("protocols", S_IRUGO, &proto_seq_fops) == NULL ? -ENOBUFS : 0; 1981} 1982 1983subsys_initcall(proto_init); 1984 1985#endif /* PROC_FS */ 1986 1987EXPORT_SYMBOL(sk_alloc); 1988EXPORT_SYMBOL(sk_free); 1989EXPORT_SYMBOL(sk_send_sigurg); 1990EXPORT_SYMBOL(sock_alloc_send_skb); 1991EXPORT_SYMBOL(sock_init_data); 1992EXPORT_SYMBOL(sock_kfree_s); 1993EXPORT_SYMBOL(sock_kmalloc); 1994EXPORT_SYMBOL(sock_no_accept); 1995EXPORT_SYMBOL(sock_no_bind); 1996EXPORT_SYMBOL(sock_no_connect); 1997EXPORT_SYMBOL(sock_no_getname); 1998EXPORT_SYMBOL(sock_no_getsockopt); 1999EXPORT_SYMBOL(sock_no_ioctl); 2000EXPORT_SYMBOL(sock_no_listen); 2001EXPORT_SYMBOL(sock_no_mmap); 2002EXPORT_SYMBOL(sock_no_poll); 2003EXPORT_SYMBOL(sock_no_recvmsg); 2004EXPORT_SYMBOL(sock_no_sendmsg); 2005EXPORT_SYMBOL(sock_no_sendpage); 2006EXPORT_SYMBOL(sock_no_setsockopt); 2007EXPORT_SYMBOL(sock_no_shutdown); 2008EXPORT_SYMBOL(sock_no_socketpair); 2009EXPORT_SYMBOL(sock_rfree); 2010EXPORT_SYMBOL(sock_setsockopt); 2011EXPORT_SYMBOL(sock_wfree); 2012EXPORT_SYMBOL(sock_wmalloc); 2013EXPORT_SYMBOL(sock_i_uid); 2014EXPORT_SYMBOL(sock_i_ino); 2015EXPORT_SYMBOL(sysctl_optmem_max); 2016#ifdef CONFIG_SYSCTL 2017EXPORT_SYMBOL(sysctl_rmem_max); 2018EXPORT_SYMBOL(sysctl_wmem_max); 2019#endif 2020