sock.c revision 0fd7bac6b6157eed6cf0cb86a1e88ba29e57c033
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 * Authors: Ross Biro 11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 12 * Florian La Roche, <flla@stud.uni-sb.de> 13 * Alan Cox, <A.Cox@swansea.ac.uk> 14 * 15 * Fixes: 16 * Alan Cox : Numerous verify_area() problems 17 * Alan Cox : Connecting on a connecting socket 18 * now returns an error for tcp. 19 * Alan Cox : sock->protocol is set correctly. 20 * and is not sometimes left as 0. 21 * Alan Cox : connect handles icmp errors on a 22 * connect properly. Unfortunately there 23 * is a restart syscall nasty there. I 24 * can't match BSD without hacking the C 25 * library. Ideas urgently sought! 26 * Alan Cox : Disallow bind() to addresses that are 27 * not ours - especially broadcast ones!! 28 * Alan Cox : Socket 1024 _IS_ ok for users. (fencepost) 29 * Alan Cox : sock_wfree/sock_rfree don't destroy sockets, 30 * instead they leave that for the DESTROY timer. 31 * Alan Cox : Clean up error flag in accept 32 * Alan Cox : TCP ack handling is buggy, the DESTROY timer 33 * was buggy. Put a remove_sock() in the handler 34 * for memory when we hit 0. Also altered the timer 35 * code. The ACK stuff can wait and needs major 36 * TCP layer surgery. 37 * Alan Cox : Fixed TCP ack bug, removed remove sock 38 * and fixed timer/inet_bh race. 39 * Alan Cox : Added zapped flag for TCP 40 * Alan Cox : Move kfree_skb into skbuff.c and tidied up surplus code 41 * Alan Cox : for new sk_buff allocations wmalloc/rmalloc now call alloc_skb 42 * Alan Cox : kfree_s calls now are kfree_skbmem so we can track skb resources 43 * Alan Cox : Supports socket option broadcast now as does udp. Packet and raw need fixing. 44 * Alan Cox : Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so... 45 * Rick Sladkey : Relaxed UDP rules for matching packets. 46 * C.E.Hawkins : IFF_PROMISC/SIOCGHWADDR support 47 * Pauline Middelink : identd support 48 * Alan Cox : Fixed connect() taking signals I think. 49 * Alan Cox : SO_LINGER supported 50 * Alan Cox : Error reporting fixes 51 * Anonymous : inet_create tidied up (sk->reuse setting) 52 * Alan Cox : inet sockets don't set sk->type! 53 * Alan Cox : Split socket option code 54 * Alan Cox : Callbacks 55 * Alan Cox : Nagle flag for Charles & Johannes stuff 56 * Alex : Removed restriction on inet fioctl 57 * Alan Cox : Splitting INET from NET core 58 * Alan Cox : Fixed bogus SO_TYPE handling in getsockopt() 59 * Adam Caldwell : Missing return in SO_DONTROUTE/SO_DEBUG code 60 * Alan Cox : Split IP from generic code 61 * Alan Cox : New kfree_skbmem() 62 * Alan Cox : Make SO_DEBUG superuser only. 63 * Alan Cox : Allow anyone to clear SO_DEBUG 64 * (compatibility fix) 65 * Alan Cox : Added optimistic memory grabbing for AF_UNIX throughput. 66 * Alan Cox : Allocator for a socket is settable. 67 * Alan Cox : SO_ERROR includes soft errors. 68 * Alan Cox : Allow NULL arguments on some SO_ opts 69 * Alan Cox : Generic socket allocation to make hooks 70 * easier (suggested by Craig Metz). 71 * Michael Pall : SO_ERROR returns positive errno again 72 * Steve Whitehouse: Added default destructor to free 73 * protocol private data. 74 * Steve Whitehouse: Added various other default routines 75 * common to several socket families. 76 * Chris Evans : Call suser() check last on F_SETOWN 77 * Jay Schulist : Added SO_ATTACH_FILTER and SO_DETACH_FILTER. 78 * Andi Kleen : Add sock_kmalloc()/sock_kfree_s() 79 * Andi Kleen : Fix write_space callback 80 * Chris Evans : Security fixes - signedness again 81 * Arnaldo C. Melo : cleanups, use skb_queue_purge 82 * 83 * To Fix: 84 * 85 * 86 * This program is free software; you can redistribute it and/or 87 * modify it under the terms of the GNU General Public License 88 * as published by the Free Software Foundation; either version 89 * 2 of the License, or (at your option) any later version. 90 */ 91 92#include <linux/capability.h> 93#include <linux/errno.h> 94#include <linux/types.h> 95#include <linux/socket.h> 96#include <linux/in.h> 97#include <linux/kernel.h> 98#include <linux/module.h> 99#include <linux/proc_fs.h> 100#include <linux/seq_file.h> 101#include <linux/sched.h> 102#include <linux/timer.h> 103#include <linux/string.h> 104#include <linux/sockios.h> 105#include <linux/net.h> 106#include <linux/mm.h> 107#include <linux/slab.h> 108#include <linux/interrupt.h> 109#include <linux/poll.h> 110#include <linux/tcp.h> 111#include <linux/init.h> 112#include <linux/highmem.h> 113#include <linux/user_namespace.h> 114 115#include <asm/uaccess.h> 116#include <asm/system.h> 117 118#include <linux/netdevice.h> 119#include <net/protocol.h> 120#include <linux/skbuff.h> 121#include <net/net_namespace.h> 122#include <net/request_sock.h> 123#include <net/sock.h> 124#include <linux/net_tstamp.h> 125#include <net/xfrm.h> 126#include <linux/ipsec.h> 127#include <net/cls_cgroup.h> 128 129#include <linux/filter.h> 130 131#include <trace/events/sock.h> 132 133#ifdef CONFIG_INET 134#include <net/tcp.h> 135#endif 136 137/* 138 * Each address family might have different locking rules, so we have 139 * one slock key per address family: 140 */ 141static struct lock_class_key af_family_keys[AF_MAX]; 142static struct lock_class_key af_family_slock_keys[AF_MAX]; 143 144/* 145 * Make lock validator output more readable. (we pre-construct these 146 * strings build-time, so that runtime initialization of socket 147 * locks is fast): 148 */ 149static const char *const af_family_key_strings[AF_MAX+1] = { 150 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" , 151 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK", 152 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" , 153 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" , 154 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" , 155 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" , 156 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" , 157 "sk_lock-AF_RDS" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" , 158 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" , 159 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" , 160 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" , 161 "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN" , "sk_lock-AF_PHONET" , 162 "sk_lock-AF_IEEE802154", "sk_lock-AF_CAIF" , "sk_lock-AF_ALG" , 163 "sk_lock-AF_NFC" , "sk_lock-AF_MAX" 164}; 165static const char *const af_family_slock_key_strings[AF_MAX+1] = { 166 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" , 167 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK", 168 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" , 169 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" , 170 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" , 171 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" , 172 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" , 173 "slock-AF_RDS" , "slock-AF_SNA" , "slock-AF_IRDA" , 174 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" , 175 "slock-27" , "slock-28" , "slock-AF_CAN" , 176 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" , 177 "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" , 178 "slock-AF_IEEE802154", "slock-AF_CAIF" , "slock-AF_ALG" , 179 "slock-AF_NFC" , "slock-AF_MAX" 180}; 181static const char *const af_family_clock_key_strings[AF_MAX+1] = { 182 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" , 183 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK", 184 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" , 185 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" , 186 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" , 187 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" , 188 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" , 189 "clock-AF_RDS" , "clock-AF_SNA" , "clock-AF_IRDA" , 190 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" , 191 "clock-27" , "clock-28" , "clock-AF_CAN" , 192 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" , 193 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" , 194 "clock-AF_IEEE802154", "clock-AF_CAIF" , "clock-AF_ALG" , 195 "clock-AF_NFC" , "clock-AF_MAX" 196}; 197 198/* 199 * sk_callback_lock locking rules are per-address-family, 200 * so split the lock classes by using a per-AF key: 201 */ 202static struct lock_class_key af_callback_keys[AF_MAX]; 203 204/* Take into consideration the size of the struct sk_buff overhead in the 205 * determination of these values, since that is non-constant across 206 * platforms. This makes socket queueing behavior and performance 207 * not depend upon such differences. 208 */ 209#define _SK_MEM_PACKETS 256 210#define _SK_MEM_OVERHEAD SKB_TRUESIZE(256) 211#define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 212#define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 213 214/* Run time adjustable parameters. */ 215__u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX; 216__u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX; 217__u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX; 218__u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX; 219 220/* Maximal space eaten by iovec or ancillary data plus some space */ 221int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512); 222EXPORT_SYMBOL(sysctl_optmem_max); 223 224#if defined(CONFIG_CGROUPS) && !defined(CONFIG_NET_CLS_CGROUP) 225int net_cls_subsys_id = -1; 226EXPORT_SYMBOL_GPL(net_cls_subsys_id); 227#endif 228 229static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen) 230{ 231 struct timeval tv; 232 233 if (optlen < sizeof(tv)) 234 return -EINVAL; 235 if (copy_from_user(&tv, optval, sizeof(tv))) 236 return -EFAULT; 237 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC) 238 return -EDOM; 239 240 if (tv.tv_sec < 0) { 241 static int warned __read_mostly; 242 243 *timeo_p = 0; 244 if (warned < 10 && net_ratelimit()) { 245 warned++; 246 printk(KERN_INFO "sock_set_timeout: `%s' (pid %d) " 247 "tries to set negative timeout\n", 248 current->comm, task_pid_nr(current)); 249 } 250 return 0; 251 } 252 *timeo_p = MAX_SCHEDULE_TIMEOUT; 253 if (tv.tv_sec == 0 && tv.tv_usec == 0) 254 return 0; 255 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1)) 256 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ); 257 return 0; 258} 259 260static void sock_warn_obsolete_bsdism(const char *name) 261{ 262 static int warned; 263 static char warncomm[TASK_COMM_LEN]; 264 if (strcmp(warncomm, current->comm) && warned < 5) { 265 strcpy(warncomm, current->comm); 266 printk(KERN_WARNING "process `%s' is using obsolete " 267 "%s SO_BSDCOMPAT\n", warncomm, name); 268 warned++; 269 } 270} 271 272static void sock_disable_timestamp(struct sock *sk, int flag) 273{ 274 if (sock_flag(sk, flag)) { 275 sock_reset_flag(sk, flag); 276 if (!sock_flag(sk, SOCK_TIMESTAMP) && 277 !sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE)) { 278 net_disable_timestamp(); 279 } 280 } 281} 282 283 284int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 285{ 286 int err; 287 int skb_len; 288 unsigned long flags; 289 struct sk_buff_head *list = &sk->sk_receive_queue; 290 291 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) { 292 atomic_inc(&sk->sk_drops); 293 trace_sock_rcvqueue_full(sk, skb); 294 return -ENOMEM; 295 } 296 297 err = sk_filter(sk, skb); 298 if (err) 299 return err; 300 301 if (!sk_rmem_schedule(sk, skb->truesize)) { 302 atomic_inc(&sk->sk_drops); 303 return -ENOBUFS; 304 } 305 306 skb->dev = NULL; 307 skb_set_owner_r(skb, sk); 308 309 /* Cache the SKB length before we tack it onto the receive 310 * queue. Once it is added it no longer belongs to us and 311 * may be freed by other threads of control pulling packets 312 * from the queue. 313 */ 314 skb_len = skb->len; 315 316 /* we escape from rcu protected region, make sure we dont leak 317 * a norefcounted dst 318 */ 319 skb_dst_force(skb); 320 321 spin_lock_irqsave(&list->lock, flags); 322 skb->dropcount = atomic_read(&sk->sk_drops); 323 __skb_queue_tail(list, skb); 324 spin_unlock_irqrestore(&list->lock, flags); 325 326 if (!sock_flag(sk, SOCK_DEAD)) 327 sk->sk_data_ready(sk, skb_len); 328 return 0; 329} 330EXPORT_SYMBOL(sock_queue_rcv_skb); 331 332int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested) 333{ 334 int rc = NET_RX_SUCCESS; 335 336 if (sk_filter(sk, skb)) 337 goto discard_and_relse; 338 339 skb->dev = NULL; 340 341 if (sk_rcvqueues_full(sk, skb)) { 342 atomic_inc(&sk->sk_drops); 343 goto discard_and_relse; 344 } 345 if (nested) 346 bh_lock_sock_nested(sk); 347 else 348 bh_lock_sock(sk); 349 if (!sock_owned_by_user(sk)) { 350 /* 351 * trylock + unlock semantics: 352 */ 353 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_); 354 355 rc = sk_backlog_rcv(sk, skb); 356 357 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_); 358 } else if (sk_add_backlog(sk, skb)) { 359 bh_unlock_sock(sk); 360 atomic_inc(&sk->sk_drops); 361 goto discard_and_relse; 362 } 363 364 bh_unlock_sock(sk); 365out: 366 sock_put(sk); 367 return rc; 368discard_and_relse: 369 kfree_skb(skb); 370 goto out; 371} 372EXPORT_SYMBOL(sk_receive_skb); 373 374void sk_reset_txq(struct sock *sk) 375{ 376 sk_tx_queue_clear(sk); 377} 378EXPORT_SYMBOL(sk_reset_txq); 379 380struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie) 381{ 382 struct dst_entry *dst = __sk_dst_get(sk); 383 384 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) { 385 sk_tx_queue_clear(sk); 386 RCU_INIT_POINTER(sk->sk_dst_cache, NULL); 387 dst_release(dst); 388 return NULL; 389 } 390 391 return dst; 392} 393EXPORT_SYMBOL(__sk_dst_check); 394 395struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie) 396{ 397 struct dst_entry *dst = sk_dst_get(sk); 398 399 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) { 400 sk_dst_reset(sk); 401 dst_release(dst); 402 return NULL; 403 } 404 405 return dst; 406} 407EXPORT_SYMBOL(sk_dst_check); 408 409static int sock_bindtodevice(struct sock *sk, char __user *optval, int optlen) 410{ 411 int ret = -ENOPROTOOPT; 412#ifdef CONFIG_NETDEVICES 413 struct net *net = sock_net(sk); 414 char devname[IFNAMSIZ]; 415 int index; 416 417 /* Sorry... */ 418 ret = -EPERM; 419 if (!capable(CAP_NET_RAW)) 420 goto out; 421 422 ret = -EINVAL; 423 if (optlen < 0) 424 goto out; 425 426 /* Bind this socket to a particular device like "eth0", 427 * as specified in the passed interface name. If the 428 * name is "" or the option length is zero the socket 429 * is not bound. 430 */ 431 if (optlen > IFNAMSIZ - 1) 432 optlen = IFNAMSIZ - 1; 433 memset(devname, 0, sizeof(devname)); 434 435 ret = -EFAULT; 436 if (copy_from_user(devname, optval, optlen)) 437 goto out; 438 439 index = 0; 440 if (devname[0] != '\0') { 441 struct net_device *dev; 442 443 rcu_read_lock(); 444 dev = dev_get_by_name_rcu(net, devname); 445 if (dev) 446 index = dev->ifindex; 447 rcu_read_unlock(); 448 ret = -ENODEV; 449 if (!dev) 450 goto out; 451 } 452 453 lock_sock(sk); 454 sk->sk_bound_dev_if = index; 455 sk_dst_reset(sk); 456 release_sock(sk); 457 458 ret = 0; 459 460out: 461#endif 462 463 return ret; 464} 465 466static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool) 467{ 468 if (valbool) 469 sock_set_flag(sk, bit); 470 else 471 sock_reset_flag(sk, bit); 472} 473 474/* 475 * This is meant for all protocols to use and covers goings on 476 * at the socket level. Everything here is generic. 477 */ 478 479int sock_setsockopt(struct socket *sock, int level, int optname, 480 char __user *optval, unsigned int optlen) 481{ 482 struct sock *sk = sock->sk; 483 int val; 484 int valbool; 485 struct linger ling; 486 int ret = 0; 487 488 /* 489 * Options without arguments 490 */ 491 492 if (optname == SO_BINDTODEVICE) 493 return sock_bindtodevice(sk, optval, optlen); 494 495 if (optlen < sizeof(int)) 496 return -EINVAL; 497 498 if (get_user(val, (int __user *)optval)) 499 return -EFAULT; 500 501 valbool = val ? 1 : 0; 502 503 lock_sock(sk); 504 505 switch (optname) { 506 case SO_DEBUG: 507 if (val && !capable(CAP_NET_ADMIN)) 508 ret = -EACCES; 509 else 510 sock_valbool_flag(sk, SOCK_DBG, valbool); 511 break; 512 case SO_REUSEADDR: 513 sk->sk_reuse = valbool; 514 break; 515 case SO_TYPE: 516 case SO_PROTOCOL: 517 case SO_DOMAIN: 518 case SO_ERROR: 519 ret = -ENOPROTOOPT; 520 break; 521 case SO_DONTROUTE: 522 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool); 523 break; 524 case SO_BROADCAST: 525 sock_valbool_flag(sk, SOCK_BROADCAST, valbool); 526 break; 527 case SO_SNDBUF: 528 /* Don't error on this BSD doesn't and if you think 529 about it this is right. Otherwise apps have to 530 play 'guess the biggest size' games. RCVBUF/SNDBUF 531 are treated in BSD as hints */ 532 533 if (val > sysctl_wmem_max) 534 val = sysctl_wmem_max; 535set_sndbuf: 536 sk->sk_userlocks |= SOCK_SNDBUF_LOCK; 537 if ((val * 2) < SOCK_MIN_SNDBUF) 538 sk->sk_sndbuf = SOCK_MIN_SNDBUF; 539 else 540 sk->sk_sndbuf = val * 2; 541 542 /* 543 * Wake up sending tasks if we 544 * upped the value. 545 */ 546 sk->sk_write_space(sk); 547 break; 548 549 case SO_SNDBUFFORCE: 550 if (!capable(CAP_NET_ADMIN)) { 551 ret = -EPERM; 552 break; 553 } 554 goto set_sndbuf; 555 556 case SO_RCVBUF: 557 /* Don't error on this BSD doesn't and if you think 558 about it this is right. Otherwise apps have to 559 play 'guess the biggest size' games. RCVBUF/SNDBUF 560 are treated in BSD as hints */ 561 562 if (val > sysctl_rmem_max) 563 val = sysctl_rmem_max; 564set_rcvbuf: 565 sk->sk_userlocks |= SOCK_RCVBUF_LOCK; 566 /* 567 * We double it on the way in to account for 568 * "struct sk_buff" etc. overhead. Applications 569 * assume that the SO_RCVBUF setting they make will 570 * allow that much actual data to be received on that 571 * socket. 572 * 573 * Applications are unaware that "struct sk_buff" and 574 * other overheads allocate from the receive buffer 575 * during socket buffer allocation. 576 * 577 * And after considering the possible alternatives, 578 * returning the value we actually used in getsockopt 579 * is the most desirable behavior. 580 */ 581 if ((val * 2) < SOCK_MIN_RCVBUF) 582 sk->sk_rcvbuf = SOCK_MIN_RCVBUF; 583 else 584 sk->sk_rcvbuf = val * 2; 585 break; 586 587 case SO_RCVBUFFORCE: 588 if (!capable(CAP_NET_ADMIN)) { 589 ret = -EPERM; 590 break; 591 } 592 goto set_rcvbuf; 593 594 case SO_KEEPALIVE: 595#ifdef CONFIG_INET 596 if (sk->sk_protocol == IPPROTO_TCP) 597 tcp_set_keepalive(sk, valbool); 598#endif 599 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool); 600 break; 601 602 case SO_OOBINLINE: 603 sock_valbool_flag(sk, SOCK_URGINLINE, valbool); 604 break; 605 606 case SO_NO_CHECK: 607 sk->sk_no_check = valbool; 608 break; 609 610 case SO_PRIORITY: 611 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN)) 612 sk->sk_priority = val; 613 else 614 ret = -EPERM; 615 break; 616 617 case SO_LINGER: 618 if (optlen < sizeof(ling)) { 619 ret = -EINVAL; /* 1003.1g */ 620 break; 621 } 622 if (copy_from_user(&ling, optval, sizeof(ling))) { 623 ret = -EFAULT; 624 break; 625 } 626 if (!ling.l_onoff) 627 sock_reset_flag(sk, SOCK_LINGER); 628 else { 629#if (BITS_PER_LONG == 32) 630 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ) 631 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT; 632 else 633#endif 634 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ; 635 sock_set_flag(sk, SOCK_LINGER); 636 } 637 break; 638 639 case SO_BSDCOMPAT: 640 sock_warn_obsolete_bsdism("setsockopt"); 641 break; 642 643 case SO_PASSCRED: 644 if (valbool) 645 set_bit(SOCK_PASSCRED, &sock->flags); 646 else 647 clear_bit(SOCK_PASSCRED, &sock->flags); 648 break; 649 650 case SO_TIMESTAMP: 651 case SO_TIMESTAMPNS: 652 if (valbool) { 653 if (optname == SO_TIMESTAMP) 654 sock_reset_flag(sk, SOCK_RCVTSTAMPNS); 655 else 656 sock_set_flag(sk, SOCK_RCVTSTAMPNS); 657 sock_set_flag(sk, SOCK_RCVTSTAMP); 658 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 659 } else { 660 sock_reset_flag(sk, SOCK_RCVTSTAMP); 661 sock_reset_flag(sk, SOCK_RCVTSTAMPNS); 662 } 663 break; 664 665 case SO_TIMESTAMPING: 666 if (val & ~SOF_TIMESTAMPING_MASK) { 667 ret = -EINVAL; 668 break; 669 } 670 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE, 671 val & SOF_TIMESTAMPING_TX_HARDWARE); 672 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE, 673 val & SOF_TIMESTAMPING_TX_SOFTWARE); 674 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE, 675 val & SOF_TIMESTAMPING_RX_HARDWARE); 676 if (val & SOF_TIMESTAMPING_RX_SOFTWARE) 677 sock_enable_timestamp(sk, 678 SOCK_TIMESTAMPING_RX_SOFTWARE); 679 else 680 sock_disable_timestamp(sk, 681 SOCK_TIMESTAMPING_RX_SOFTWARE); 682 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SOFTWARE, 683 val & SOF_TIMESTAMPING_SOFTWARE); 684 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE, 685 val & SOF_TIMESTAMPING_SYS_HARDWARE); 686 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE, 687 val & SOF_TIMESTAMPING_RAW_HARDWARE); 688 break; 689 690 case SO_RCVLOWAT: 691 if (val < 0) 692 val = INT_MAX; 693 sk->sk_rcvlowat = val ? : 1; 694 break; 695 696 case SO_RCVTIMEO: 697 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen); 698 break; 699 700 case SO_SNDTIMEO: 701 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen); 702 break; 703 704 case SO_ATTACH_FILTER: 705 ret = -EINVAL; 706 if (optlen == sizeof(struct sock_fprog)) { 707 struct sock_fprog fprog; 708 709 ret = -EFAULT; 710 if (copy_from_user(&fprog, optval, sizeof(fprog))) 711 break; 712 713 ret = sk_attach_filter(&fprog, sk); 714 } 715 break; 716 717 case SO_DETACH_FILTER: 718 ret = sk_detach_filter(sk); 719 break; 720 721 case SO_PASSSEC: 722 if (valbool) 723 set_bit(SOCK_PASSSEC, &sock->flags); 724 else 725 clear_bit(SOCK_PASSSEC, &sock->flags); 726 break; 727 case SO_MARK: 728 if (!capable(CAP_NET_ADMIN)) 729 ret = -EPERM; 730 else 731 sk->sk_mark = val; 732 break; 733 734 /* We implement the SO_SNDLOWAT etc to 735 not be settable (1003.1g 5.3) */ 736 case SO_RXQ_OVFL: 737 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool); 738 break; 739 default: 740 ret = -ENOPROTOOPT; 741 break; 742 } 743 release_sock(sk); 744 return ret; 745} 746EXPORT_SYMBOL(sock_setsockopt); 747 748 749void cred_to_ucred(struct pid *pid, const struct cred *cred, 750 struct ucred *ucred) 751{ 752 ucred->pid = pid_vnr(pid); 753 ucred->uid = ucred->gid = -1; 754 if (cred) { 755 struct user_namespace *current_ns = current_user_ns(); 756 757 ucred->uid = user_ns_map_uid(current_ns, cred, cred->euid); 758 ucred->gid = user_ns_map_gid(current_ns, cred, cred->egid); 759 } 760} 761EXPORT_SYMBOL_GPL(cred_to_ucred); 762 763int sock_getsockopt(struct socket *sock, int level, int optname, 764 char __user *optval, int __user *optlen) 765{ 766 struct sock *sk = sock->sk; 767 768 union { 769 int val; 770 struct linger ling; 771 struct timeval tm; 772 } v; 773 774 int lv = sizeof(int); 775 int len; 776 777 if (get_user(len, optlen)) 778 return -EFAULT; 779 if (len < 0) 780 return -EINVAL; 781 782 memset(&v, 0, sizeof(v)); 783 784 switch (optname) { 785 case SO_DEBUG: 786 v.val = sock_flag(sk, SOCK_DBG); 787 break; 788 789 case SO_DONTROUTE: 790 v.val = sock_flag(sk, SOCK_LOCALROUTE); 791 break; 792 793 case SO_BROADCAST: 794 v.val = !!sock_flag(sk, SOCK_BROADCAST); 795 break; 796 797 case SO_SNDBUF: 798 v.val = sk->sk_sndbuf; 799 break; 800 801 case SO_RCVBUF: 802 v.val = sk->sk_rcvbuf; 803 break; 804 805 case SO_REUSEADDR: 806 v.val = sk->sk_reuse; 807 break; 808 809 case SO_KEEPALIVE: 810 v.val = !!sock_flag(sk, SOCK_KEEPOPEN); 811 break; 812 813 case SO_TYPE: 814 v.val = sk->sk_type; 815 break; 816 817 case SO_PROTOCOL: 818 v.val = sk->sk_protocol; 819 break; 820 821 case SO_DOMAIN: 822 v.val = sk->sk_family; 823 break; 824 825 case SO_ERROR: 826 v.val = -sock_error(sk); 827 if (v.val == 0) 828 v.val = xchg(&sk->sk_err_soft, 0); 829 break; 830 831 case SO_OOBINLINE: 832 v.val = !!sock_flag(sk, SOCK_URGINLINE); 833 break; 834 835 case SO_NO_CHECK: 836 v.val = sk->sk_no_check; 837 break; 838 839 case SO_PRIORITY: 840 v.val = sk->sk_priority; 841 break; 842 843 case SO_LINGER: 844 lv = sizeof(v.ling); 845 v.ling.l_onoff = !!sock_flag(sk, SOCK_LINGER); 846 v.ling.l_linger = sk->sk_lingertime / HZ; 847 break; 848 849 case SO_BSDCOMPAT: 850 sock_warn_obsolete_bsdism("getsockopt"); 851 break; 852 853 case SO_TIMESTAMP: 854 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && 855 !sock_flag(sk, SOCK_RCVTSTAMPNS); 856 break; 857 858 case SO_TIMESTAMPNS: 859 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS); 860 break; 861 862 case SO_TIMESTAMPING: 863 v.val = 0; 864 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE)) 865 v.val |= SOF_TIMESTAMPING_TX_HARDWARE; 866 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE)) 867 v.val |= SOF_TIMESTAMPING_TX_SOFTWARE; 868 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE)) 869 v.val |= SOF_TIMESTAMPING_RX_HARDWARE; 870 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE)) 871 v.val |= SOF_TIMESTAMPING_RX_SOFTWARE; 872 if (sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE)) 873 v.val |= SOF_TIMESTAMPING_SOFTWARE; 874 if (sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE)) 875 v.val |= SOF_TIMESTAMPING_SYS_HARDWARE; 876 if (sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE)) 877 v.val |= SOF_TIMESTAMPING_RAW_HARDWARE; 878 break; 879 880 case SO_RCVTIMEO: 881 lv = sizeof(struct timeval); 882 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) { 883 v.tm.tv_sec = 0; 884 v.tm.tv_usec = 0; 885 } else { 886 v.tm.tv_sec = sk->sk_rcvtimeo / HZ; 887 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ; 888 } 889 break; 890 891 case SO_SNDTIMEO: 892 lv = sizeof(struct timeval); 893 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) { 894 v.tm.tv_sec = 0; 895 v.tm.tv_usec = 0; 896 } else { 897 v.tm.tv_sec = sk->sk_sndtimeo / HZ; 898 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ; 899 } 900 break; 901 902 case SO_RCVLOWAT: 903 v.val = sk->sk_rcvlowat; 904 break; 905 906 case SO_SNDLOWAT: 907 v.val = 1; 908 break; 909 910 case SO_PASSCRED: 911 v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0; 912 break; 913 914 case SO_PEERCRED: 915 { 916 struct ucred peercred; 917 if (len > sizeof(peercred)) 918 len = sizeof(peercred); 919 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred); 920 if (copy_to_user(optval, &peercred, len)) 921 return -EFAULT; 922 goto lenout; 923 } 924 925 case SO_PEERNAME: 926 { 927 char address[128]; 928 929 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2)) 930 return -ENOTCONN; 931 if (lv < len) 932 return -EINVAL; 933 if (copy_to_user(optval, address, len)) 934 return -EFAULT; 935 goto lenout; 936 } 937 938 /* Dubious BSD thing... Probably nobody even uses it, but 939 * the UNIX standard wants it for whatever reason... -DaveM 940 */ 941 case SO_ACCEPTCONN: 942 v.val = sk->sk_state == TCP_LISTEN; 943 break; 944 945 case SO_PASSSEC: 946 v.val = test_bit(SOCK_PASSSEC, &sock->flags) ? 1 : 0; 947 break; 948 949 case SO_PEERSEC: 950 return security_socket_getpeersec_stream(sock, optval, optlen, len); 951 952 case SO_MARK: 953 v.val = sk->sk_mark; 954 break; 955 956 case SO_RXQ_OVFL: 957 v.val = !!sock_flag(sk, SOCK_RXQ_OVFL); 958 break; 959 960 default: 961 return -ENOPROTOOPT; 962 } 963 964 if (len > lv) 965 len = lv; 966 if (copy_to_user(optval, &v, len)) 967 return -EFAULT; 968lenout: 969 if (put_user(len, optlen)) 970 return -EFAULT; 971 return 0; 972} 973 974/* 975 * Initialize an sk_lock. 976 * 977 * (We also register the sk_lock with the lock validator.) 978 */ 979static inline void sock_lock_init(struct sock *sk) 980{ 981 sock_lock_init_class_and_name(sk, 982 af_family_slock_key_strings[sk->sk_family], 983 af_family_slock_keys + sk->sk_family, 984 af_family_key_strings[sk->sk_family], 985 af_family_keys + sk->sk_family); 986} 987 988/* 989 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet, 990 * even temporarly, because of RCU lookups. sk_node should also be left as is. 991 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end 992 */ 993static void sock_copy(struct sock *nsk, const struct sock *osk) 994{ 995#ifdef CONFIG_SECURITY_NETWORK 996 void *sptr = nsk->sk_security; 997#endif 998 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin)); 999 1000 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end, 1001 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end)); 1002 1003#ifdef CONFIG_SECURITY_NETWORK 1004 nsk->sk_security = sptr; 1005 security_sk_clone(osk, nsk); 1006#endif 1007} 1008 1009/* 1010 * caches using SLAB_DESTROY_BY_RCU should let .next pointer from nulls nodes 1011 * un-modified. Special care is taken when initializing object to zero. 1012 */ 1013static inline void sk_prot_clear_nulls(struct sock *sk, int size) 1014{ 1015 if (offsetof(struct sock, sk_node.next) != 0) 1016 memset(sk, 0, offsetof(struct sock, sk_node.next)); 1017 memset(&sk->sk_node.pprev, 0, 1018 size - offsetof(struct sock, sk_node.pprev)); 1019} 1020 1021void sk_prot_clear_portaddr_nulls(struct sock *sk, int size) 1022{ 1023 unsigned long nulls1, nulls2; 1024 1025 nulls1 = offsetof(struct sock, __sk_common.skc_node.next); 1026 nulls2 = offsetof(struct sock, __sk_common.skc_portaddr_node.next); 1027 if (nulls1 > nulls2) 1028 swap(nulls1, nulls2); 1029 1030 if (nulls1 != 0) 1031 memset((char *)sk, 0, nulls1); 1032 memset((char *)sk + nulls1 + sizeof(void *), 0, 1033 nulls2 - nulls1 - sizeof(void *)); 1034 memset((char *)sk + nulls2 + sizeof(void *), 0, 1035 size - nulls2 - sizeof(void *)); 1036} 1037EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls); 1038 1039static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority, 1040 int family) 1041{ 1042 struct sock *sk; 1043 struct kmem_cache *slab; 1044 1045 slab = prot->slab; 1046 if (slab != NULL) { 1047 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO); 1048 if (!sk) 1049 return sk; 1050 if (priority & __GFP_ZERO) { 1051 if (prot->clear_sk) 1052 prot->clear_sk(sk, prot->obj_size); 1053 else 1054 sk_prot_clear_nulls(sk, prot->obj_size); 1055 } 1056 } else 1057 sk = kmalloc(prot->obj_size, priority); 1058 1059 if (sk != NULL) { 1060 kmemcheck_annotate_bitfield(sk, flags); 1061 1062 if (security_sk_alloc(sk, family, priority)) 1063 goto out_free; 1064 1065 if (!try_module_get(prot->owner)) 1066 goto out_free_sec; 1067 sk_tx_queue_clear(sk); 1068 } 1069 1070 return sk; 1071 1072out_free_sec: 1073 security_sk_free(sk); 1074out_free: 1075 if (slab != NULL) 1076 kmem_cache_free(slab, sk); 1077 else 1078 kfree(sk); 1079 return NULL; 1080} 1081 1082static void sk_prot_free(struct proto *prot, struct sock *sk) 1083{ 1084 struct kmem_cache *slab; 1085 struct module *owner; 1086 1087 owner = prot->owner; 1088 slab = prot->slab; 1089 1090 security_sk_free(sk); 1091 if (slab != NULL) 1092 kmem_cache_free(slab, sk); 1093 else 1094 kfree(sk); 1095 module_put(owner); 1096} 1097 1098#ifdef CONFIG_CGROUPS 1099void sock_update_classid(struct sock *sk) 1100{ 1101 u32 classid; 1102 1103 rcu_read_lock(); /* doing current task, which cannot vanish. */ 1104 classid = task_cls_classid(current); 1105 rcu_read_unlock(); 1106 if (classid && classid != sk->sk_classid) 1107 sk->sk_classid = classid; 1108} 1109EXPORT_SYMBOL(sock_update_classid); 1110#endif 1111 1112/** 1113 * sk_alloc - All socket objects are allocated here 1114 * @net: the applicable net namespace 1115 * @family: protocol family 1116 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 1117 * @prot: struct proto associated with this new sock instance 1118 */ 1119struct sock *sk_alloc(struct net *net, int family, gfp_t priority, 1120 struct proto *prot) 1121{ 1122 struct sock *sk; 1123 1124 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family); 1125 if (sk) { 1126 sk->sk_family = family; 1127 /* 1128 * See comment in struct sock definition to understand 1129 * why we need sk_prot_creator -acme 1130 */ 1131 sk->sk_prot = sk->sk_prot_creator = prot; 1132 sock_lock_init(sk); 1133 sock_net_set(sk, get_net(net)); 1134 atomic_set(&sk->sk_wmem_alloc, 1); 1135 1136 sock_update_classid(sk); 1137 } 1138 1139 return sk; 1140} 1141EXPORT_SYMBOL(sk_alloc); 1142 1143static void __sk_free(struct sock *sk) 1144{ 1145 struct sk_filter *filter; 1146 1147 if (sk->sk_destruct) 1148 sk->sk_destruct(sk); 1149 1150 filter = rcu_dereference_check(sk->sk_filter, 1151 atomic_read(&sk->sk_wmem_alloc) == 0); 1152 if (filter) { 1153 sk_filter_uncharge(sk, filter); 1154 RCU_INIT_POINTER(sk->sk_filter, NULL); 1155 } 1156 1157 sock_disable_timestamp(sk, SOCK_TIMESTAMP); 1158 sock_disable_timestamp(sk, SOCK_TIMESTAMPING_RX_SOFTWARE); 1159 1160 if (atomic_read(&sk->sk_omem_alloc)) 1161 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n", 1162 __func__, atomic_read(&sk->sk_omem_alloc)); 1163 1164 if (sk->sk_peer_cred) 1165 put_cred(sk->sk_peer_cred); 1166 put_pid(sk->sk_peer_pid); 1167 put_net(sock_net(sk)); 1168 sk_prot_free(sk->sk_prot_creator, sk); 1169} 1170 1171void sk_free(struct sock *sk) 1172{ 1173 /* 1174 * We subtract one from sk_wmem_alloc and can know if 1175 * some packets are still in some tx queue. 1176 * If not null, sock_wfree() will call __sk_free(sk) later 1177 */ 1178 if (atomic_dec_and_test(&sk->sk_wmem_alloc)) 1179 __sk_free(sk); 1180} 1181EXPORT_SYMBOL(sk_free); 1182 1183/* 1184 * Last sock_put should drop reference to sk->sk_net. It has already 1185 * been dropped in sk_change_net. Taking reference to stopping namespace 1186 * is not an option. 1187 * Take reference to a socket to remove it from hash _alive_ and after that 1188 * destroy it in the context of init_net. 1189 */ 1190void sk_release_kernel(struct sock *sk) 1191{ 1192 if (sk == NULL || sk->sk_socket == NULL) 1193 return; 1194 1195 sock_hold(sk); 1196 sock_release(sk->sk_socket); 1197 release_net(sock_net(sk)); 1198 sock_net_set(sk, get_net(&init_net)); 1199 sock_put(sk); 1200} 1201EXPORT_SYMBOL(sk_release_kernel); 1202 1203struct sock *sk_clone(const struct sock *sk, const gfp_t priority) 1204{ 1205 struct sock *newsk; 1206 1207 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family); 1208 if (newsk != NULL) { 1209 struct sk_filter *filter; 1210 1211 sock_copy(newsk, sk); 1212 1213 /* SANITY */ 1214 get_net(sock_net(newsk)); 1215 sk_node_init(&newsk->sk_node); 1216 sock_lock_init(newsk); 1217 bh_lock_sock(newsk); 1218 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL; 1219 newsk->sk_backlog.len = 0; 1220 1221 atomic_set(&newsk->sk_rmem_alloc, 0); 1222 /* 1223 * sk_wmem_alloc set to one (see sk_free() and sock_wfree()) 1224 */ 1225 atomic_set(&newsk->sk_wmem_alloc, 1); 1226 atomic_set(&newsk->sk_omem_alloc, 0); 1227 skb_queue_head_init(&newsk->sk_receive_queue); 1228 skb_queue_head_init(&newsk->sk_write_queue); 1229#ifdef CONFIG_NET_DMA 1230 skb_queue_head_init(&newsk->sk_async_wait_queue); 1231#endif 1232 1233 spin_lock_init(&newsk->sk_dst_lock); 1234 rwlock_init(&newsk->sk_callback_lock); 1235 lockdep_set_class_and_name(&newsk->sk_callback_lock, 1236 af_callback_keys + newsk->sk_family, 1237 af_family_clock_key_strings[newsk->sk_family]); 1238 1239 newsk->sk_dst_cache = NULL; 1240 newsk->sk_wmem_queued = 0; 1241 newsk->sk_forward_alloc = 0; 1242 newsk->sk_send_head = NULL; 1243 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK; 1244 1245 sock_reset_flag(newsk, SOCK_DONE); 1246 skb_queue_head_init(&newsk->sk_error_queue); 1247 1248 filter = rcu_dereference_protected(newsk->sk_filter, 1); 1249 if (filter != NULL) 1250 sk_filter_charge(newsk, filter); 1251 1252 if (unlikely(xfrm_sk_clone_policy(newsk))) { 1253 /* It is still raw copy of parent, so invalidate 1254 * destructor and make plain sk_free() */ 1255 newsk->sk_destruct = NULL; 1256 bh_unlock_sock(newsk); 1257 sk_free(newsk); 1258 newsk = NULL; 1259 goto out; 1260 } 1261 1262 newsk->sk_err = 0; 1263 newsk->sk_priority = 0; 1264 /* 1265 * Before updating sk_refcnt, we must commit prior changes to memory 1266 * (Documentation/RCU/rculist_nulls.txt for details) 1267 */ 1268 smp_wmb(); 1269 atomic_set(&newsk->sk_refcnt, 2); 1270 1271 /* 1272 * Increment the counter in the same struct proto as the master 1273 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that 1274 * is the same as sk->sk_prot->socks, as this field was copied 1275 * with memcpy). 1276 * 1277 * This _changes_ the previous behaviour, where 1278 * tcp_create_openreq_child always was incrementing the 1279 * equivalent to tcp_prot->socks (inet_sock_nr), so this have 1280 * to be taken into account in all callers. -acme 1281 */ 1282 sk_refcnt_debug_inc(newsk); 1283 sk_set_socket(newsk, NULL); 1284 newsk->sk_wq = NULL; 1285 1286 if (newsk->sk_prot->sockets_allocated) 1287 percpu_counter_inc(newsk->sk_prot->sockets_allocated); 1288 1289 if (sock_flag(newsk, SOCK_TIMESTAMP) || 1290 sock_flag(newsk, SOCK_TIMESTAMPING_RX_SOFTWARE)) 1291 net_enable_timestamp(); 1292 } 1293out: 1294 return newsk; 1295} 1296EXPORT_SYMBOL_GPL(sk_clone); 1297 1298void sk_setup_caps(struct sock *sk, struct dst_entry *dst) 1299{ 1300 __sk_dst_set(sk, dst); 1301 sk->sk_route_caps = dst->dev->features; 1302 if (sk->sk_route_caps & NETIF_F_GSO) 1303 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE; 1304 sk->sk_route_caps &= ~sk->sk_route_nocaps; 1305 if (sk_can_gso(sk)) { 1306 if (dst->header_len) { 1307 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 1308 } else { 1309 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM; 1310 sk->sk_gso_max_size = dst->dev->gso_max_size; 1311 } 1312 } 1313} 1314EXPORT_SYMBOL_GPL(sk_setup_caps); 1315 1316void __init sk_init(void) 1317{ 1318 if (totalram_pages <= 4096) { 1319 sysctl_wmem_max = 32767; 1320 sysctl_rmem_max = 32767; 1321 sysctl_wmem_default = 32767; 1322 sysctl_rmem_default = 32767; 1323 } else if (totalram_pages >= 131072) { 1324 sysctl_wmem_max = 131071; 1325 sysctl_rmem_max = 131071; 1326 } 1327} 1328 1329/* 1330 * Simple resource managers for sockets. 1331 */ 1332 1333 1334/* 1335 * Write buffer destructor automatically called from kfree_skb. 1336 */ 1337void sock_wfree(struct sk_buff *skb) 1338{ 1339 struct sock *sk = skb->sk; 1340 unsigned int len = skb->truesize; 1341 1342 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) { 1343 /* 1344 * Keep a reference on sk_wmem_alloc, this will be released 1345 * after sk_write_space() call 1346 */ 1347 atomic_sub(len - 1, &sk->sk_wmem_alloc); 1348 sk->sk_write_space(sk); 1349 len = 1; 1350 } 1351 /* 1352 * if sk_wmem_alloc reaches 0, we must finish what sk_free() 1353 * could not do because of in-flight packets 1354 */ 1355 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc)) 1356 __sk_free(sk); 1357} 1358EXPORT_SYMBOL(sock_wfree); 1359 1360/* 1361 * Read buffer destructor automatically called from kfree_skb. 1362 */ 1363void sock_rfree(struct sk_buff *skb) 1364{ 1365 struct sock *sk = skb->sk; 1366 unsigned int len = skb->truesize; 1367 1368 atomic_sub(len, &sk->sk_rmem_alloc); 1369 sk_mem_uncharge(sk, len); 1370} 1371EXPORT_SYMBOL(sock_rfree); 1372 1373 1374int sock_i_uid(struct sock *sk) 1375{ 1376 int uid; 1377 1378 read_lock_bh(&sk->sk_callback_lock); 1379 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0; 1380 read_unlock_bh(&sk->sk_callback_lock); 1381 return uid; 1382} 1383EXPORT_SYMBOL(sock_i_uid); 1384 1385unsigned long sock_i_ino(struct sock *sk) 1386{ 1387 unsigned long ino; 1388 1389 read_lock_bh(&sk->sk_callback_lock); 1390 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0; 1391 read_unlock_bh(&sk->sk_callback_lock); 1392 return ino; 1393} 1394EXPORT_SYMBOL(sock_i_ino); 1395 1396/* 1397 * Allocate a skb from the socket's send buffer. 1398 */ 1399struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 1400 gfp_t priority) 1401{ 1402 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) { 1403 struct sk_buff *skb = alloc_skb(size, priority); 1404 if (skb) { 1405 skb_set_owner_w(skb, sk); 1406 return skb; 1407 } 1408 } 1409 return NULL; 1410} 1411EXPORT_SYMBOL(sock_wmalloc); 1412 1413/* 1414 * Allocate a skb from the socket's receive buffer. 1415 */ 1416struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force, 1417 gfp_t priority) 1418{ 1419 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) { 1420 struct sk_buff *skb = alloc_skb(size, priority); 1421 if (skb) { 1422 skb_set_owner_r(skb, sk); 1423 return skb; 1424 } 1425 } 1426 return NULL; 1427} 1428 1429/* 1430 * Allocate a memory block from the socket's option memory buffer. 1431 */ 1432void *sock_kmalloc(struct sock *sk, int size, gfp_t priority) 1433{ 1434 if ((unsigned)size <= sysctl_optmem_max && 1435 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) { 1436 void *mem; 1437 /* First do the add, to avoid the race if kmalloc 1438 * might sleep. 1439 */ 1440 atomic_add(size, &sk->sk_omem_alloc); 1441 mem = kmalloc(size, priority); 1442 if (mem) 1443 return mem; 1444 atomic_sub(size, &sk->sk_omem_alloc); 1445 } 1446 return NULL; 1447} 1448EXPORT_SYMBOL(sock_kmalloc); 1449 1450/* 1451 * Free an option memory block. 1452 */ 1453void sock_kfree_s(struct sock *sk, void *mem, int size) 1454{ 1455 kfree(mem); 1456 atomic_sub(size, &sk->sk_omem_alloc); 1457} 1458EXPORT_SYMBOL(sock_kfree_s); 1459 1460/* It is almost wait_for_tcp_memory minus release_sock/lock_sock. 1461 I think, these locks should be removed for datagram sockets. 1462 */ 1463static long sock_wait_for_wmem(struct sock *sk, long timeo) 1464{ 1465 DEFINE_WAIT(wait); 1466 1467 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags); 1468 for (;;) { 1469 if (!timeo) 1470 break; 1471 if (signal_pending(current)) 1472 break; 1473 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 1474 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 1475 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) 1476 break; 1477 if (sk->sk_shutdown & SEND_SHUTDOWN) 1478 break; 1479 if (sk->sk_err) 1480 break; 1481 timeo = schedule_timeout(timeo); 1482 } 1483 finish_wait(sk_sleep(sk), &wait); 1484 return timeo; 1485} 1486 1487 1488/* 1489 * Generic send/receive buffer handlers 1490 */ 1491 1492struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 1493 unsigned long data_len, int noblock, 1494 int *errcode) 1495{ 1496 struct sk_buff *skb; 1497 gfp_t gfp_mask; 1498 long timeo; 1499 int err; 1500 1501 gfp_mask = sk->sk_allocation; 1502 if (gfp_mask & __GFP_WAIT) 1503 gfp_mask |= __GFP_REPEAT; 1504 1505 timeo = sock_sndtimeo(sk, noblock); 1506 while (1) { 1507 err = sock_error(sk); 1508 if (err != 0) 1509 goto failure; 1510 1511 err = -EPIPE; 1512 if (sk->sk_shutdown & SEND_SHUTDOWN) 1513 goto failure; 1514 1515 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) { 1516 skb = alloc_skb(header_len, gfp_mask); 1517 if (skb) { 1518 int npages; 1519 int i; 1520 1521 /* No pages, we're done... */ 1522 if (!data_len) 1523 break; 1524 1525 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT; 1526 skb->truesize += data_len; 1527 skb_shinfo(skb)->nr_frags = npages; 1528 for (i = 0; i < npages; i++) { 1529 struct page *page; 1530 1531 page = alloc_pages(sk->sk_allocation, 0); 1532 if (!page) { 1533 err = -ENOBUFS; 1534 skb_shinfo(skb)->nr_frags = i; 1535 kfree_skb(skb); 1536 goto failure; 1537 } 1538 1539 __skb_fill_page_desc(skb, i, 1540 page, 0, 1541 (data_len >= PAGE_SIZE ? 1542 PAGE_SIZE : 1543 data_len)); 1544 data_len -= PAGE_SIZE; 1545 } 1546 1547 /* Full success... */ 1548 break; 1549 } 1550 err = -ENOBUFS; 1551 goto failure; 1552 } 1553 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags); 1554 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 1555 err = -EAGAIN; 1556 if (!timeo) 1557 goto failure; 1558 if (signal_pending(current)) 1559 goto interrupted; 1560 timeo = sock_wait_for_wmem(sk, timeo); 1561 } 1562 1563 skb_set_owner_w(skb, sk); 1564 return skb; 1565 1566interrupted: 1567 err = sock_intr_errno(timeo); 1568failure: 1569 *errcode = err; 1570 return NULL; 1571} 1572EXPORT_SYMBOL(sock_alloc_send_pskb); 1573 1574struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size, 1575 int noblock, int *errcode) 1576{ 1577 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode); 1578} 1579EXPORT_SYMBOL(sock_alloc_send_skb); 1580 1581static void __lock_sock(struct sock *sk) 1582 __releases(&sk->sk_lock.slock) 1583 __acquires(&sk->sk_lock.slock) 1584{ 1585 DEFINE_WAIT(wait); 1586 1587 for (;;) { 1588 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait, 1589 TASK_UNINTERRUPTIBLE); 1590 spin_unlock_bh(&sk->sk_lock.slock); 1591 schedule(); 1592 spin_lock_bh(&sk->sk_lock.slock); 1593 if (!sock_owned_by_user(sk)) 1594 break; 1595 } 1596 finish_wait(&sk->sk_lock.wq, &wait); 1597} 1598 1599static void __release_sock(struct sock *sk) 1600 __releases(&sk->sk_lock.slock) 1601 __acquires(&sk->sk_lock.slock) 1602{ 1603 struct sk_buff *skb = sk->sk_backlog.head; 1604 1605 do { 1606 sk->sk_backlog.head = sk->sk_backlog.tail = NULL; 1607 bh_unlock_sock(sk); 1608 1609 do { 1610 struct sk_buff *next = skb->next; 1611 1612 WARN_ON_ONCE(skb_dst_is_noref(skb)); 1613 skb->next = NULL; 1614 sk_backlog_rcv(sk, skb); 1615 1616 /* 1617 * We are in process context here with softirqs 1618 * disabled, use cond_resched_softirq() to preempt. 1619 * This is safe to do because we've taken the backlog 1620 * queue private: 1621 */ 1622 cond_resched_softirq(); 1623 1624 skb = next; 1625 } while (skb != NULL); 1626 1627 bh_lock_sock(sk); 1628 } while ((skb = sk->sk_backlog.head) != NULL); 1629 1630 /* 1631 * Doing the zeroing here guarantee we can not loop forever 1632 * while a wild producer attempts to flood us. 1633 */ 1634 sk->sk_backlog.len = 0; 1635} 1636 1637/** 1638 * sk_wait_data - wait for data to arrive at sk_receive_queue 1639 * @sk: sock to wait on 1640 * @timeo: for how long 1641 * 1642 * Now socket state including sk->sk_err is changed only under lock, 1643 * hence we may omit checks after joining wait queue. 1644 * We check receive queue before schedule() only as optimization; 1645 * it is very likely that release_sock() added new data. 1646 */ 1647int sk_wait_data(struct sock *sk, long *timeo) 1648{ 1649 int rc; 1650 DEFINE_WAIT(wait); 1651 1652 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 1653 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags); 1654 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue)); 1655 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags); 1656 finish_wait(sk_sleep(sk), &wait); 1657 return rc; 1658} 1659EXPORT_SYMBOL(sk_wait_data); 1660 1661/** 1662 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated 1663 * @sk: socket 1664 * @size: memory size to allocate 1665 * @kind: allocation type 1666 * 1667 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means 1668 * rmem allocation. This function assumes that protocols which have 1669 * memory_pressure use sk_wmem_queued as write buffer accounting. 1670 */ 1671int __sk_mem_schedule(struct sock *sk, int size, int kind) 1672{ 1673 struct proto *prot = sk->sk_prot; 1674 int amt = sk_mem_pages(size); 1675 long allocated; 1676 1677 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM; 1678 allocated = atomic_long_add_return(amt, prot->memory_allocated); 1679 1680 /* Under limit. */ 1681 if (allocated <= prot->sysctl_mem[0]) { 1682 if (prot->memory_pressure && *prot->memory_pressure) 1683 *prot->memory_pressure = 0; 1684 return 1; 1685 } 1686 1687 /* Under pressure. */ 1688 if (allocated > prot->sysctl_mem[1]) 1689 if (prot->enter_memory_pressure) 1690 prot->enter_memory_pressure(sk); 1691 1692 /* Over hard limit. */ 1693 if (allocated > prot->sysctl_mem[2]) 1694 goto suppress_allocation; 1695 1696 /* guarantee minimum buffer size under pressure */ 1697 if (kind == SK_MEM_RECV) { 1698 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0]) 1699 return 1; 1700 } else { /* SK_MEM_SEND */ 1701 if (sk->sk_type == SOCK_STREAM) { 1702 if (sk->sk_wmem_queued < prot->sysctl_wmem[0]) 1703 return 1; 1704 } else if (atomic_read(&sk->sk_wmem_alloc) < 1705 prot->sysctl_wmem[0]) 1706 return 1; 1707 } 1708 1709 if (prot->memory_pressure) { 1710 int alloc; 1711 1712 if (!*prot->memory_pressure) 1713 return 1; 1714 alloc = percpu_counter_read_positive(prot->sockets_allocated); 1715 if (prot->sysctl_mem[2] > alloc * 1716 sk_mem_pages(sk->sk_wmem_queued + 1717 atomic_read(&sk->sk_rmem_alloc) + 1718 sk->sk_forward_alloc)) 1719 return 1; 1720 } 1721 1722suppress_allocation: 1723 1724 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) { 1725 sk_stream_moderate_sndbuf(sk); 1726 1727 /* Fail only if socket is _under_ its sndbuf. 1728 * In this case we cannot block, so that we have to fail. 1729 */ 1730 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf) 1731 return 1; 1732 } 1733 1734 trace_sock_exceed_buf_limit(sk, prot, allocated); 1735 1736 /* Alas. Undo changes. */ 1737 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM; 1738 atomic_long_sub(amt, prot->memory_allocated); 1739 return 0; 1740} 1741EXPORT_SYMBOL(__sk_mem_schedule); 1742 1743/** 1744 * __sk_reclaim - reclaim memory_allocated 1745 * @sk: socket 1746 */ 1747void __sk_mem_reclaim(struct sock *sk) 1748{ 1749 struct proto *prot = sk->sk_prot; 1750 1751 atomic_long_sub(sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT, 1752 prot->memory_allocated); 1753 sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1; 1754 1755 if (prot->memory_pressure && *prot->memory_pressure && 1756 (atomic_long_read(prot->memory_allocated) < prot->sysctl_mem[0])) 1757 *prot->memory_pressure = 0; 1758} 1759EXPORT_SYMBOL(__sk_mem_reclaim); 1760 1761 1762/* 1763 * Set of default routines for initialising struct proto_ops when 1764 * the protocol does not support a particular function. In certain 1765 * cases where it makes no sense for a protocol to have a "do nothing" 1766 * function, some default processing is provided. 1767 */ 1768 1769int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len) 1770{ 1771 return -EOPNOTSUPP; 1772} 1773EXPORT_SYMBOL(sock_no_bind); 1774 1775int sock_no_connect(struct socket *sock, struct sockaddr *saddr, 1776 int len, int flags) 1777{ 1778 return -EOPNOTSUPP; 1779} 1780EXPORT_SYMBOL(sock_no_connect); 1781 1782int sock_no_socketpair(struct socket *sock1, struct socket *sock2) 1783{ 1784 return -EOPNOTSUPP; 1785} 1786EXPORT_SYMBOL(sock_no_socketpair); 1787 1788int sock_no_accept(struct socket *sock, struct socket *newsock, int flags) 1789{ 1790 return -EOPNOTSUPP; 1791} 1792EXPORT_SYMBOL(sock_no_accept); 1793 1794int sock_no_getname(struct socket *sock, struct sockaddr *saddr, 1795 int *len, int peer) 1796{ 1797 return -EOPNOTSUPP; 1798} 1799EXPORT_SYMBOL(sock_no_getname); 1800 1801unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt) 1802{ 1803 return 0; 1804} 1805EXPORT_SYMBOL(sock_no_poll); 1806 1807int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 1808{ 1809 return -EOPNOTSUPP; 1810} 1811EXPORT_SYMBOL(sock_no_ioctl); 1812 1813int sock_no_listen(struct socket *sock, int backlog) 1814{ 1815 return -EOPNOTSUPP; 1816} 1817EXPORT_SYMBOL(sock_no_listen); 1818 1819int sock_no_shutdown(struct socket *sock, int how) 1820{ 1821 return -EOPNOTSUPP; 1822} 1823EXPORT_SYMBOL(sock_no_shutdown); 1824 1825int sock_no_setsockopt(struct socket *sock, int level, int optname, 1826 char __user *optval, unsigned int optlen) 1827{ 1828 return -EOPNOTSUPP; 1829} 1830EXPORT_SYMBOL(sock_no_setsockopt); 1831 1832int sock_no_getsockopt(struct socket *sock, int level, int optname, 1833 char __user *optval, int __user *optlen) 1834{ 1835 return -EOPNOTSUPP; 1836} 1837EXPORT_SYMBOL(sock_no_getsockopt); 1838 1839int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m, 1840 size_t len) 1841{ 1842 return -EOPNOTSUPP; 1843} 1844EXPORT_SYMBOL(sock_no_sendmsg); 1845 1846int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m, 1847 size_t len, int flags) 1848{ 1849 return -EOPNOTSUPP; 1850} 1851EXPORT_SYMBOL(sock_no_recvmsg); 1852 1853int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma) 1854{ 1855 /* Mirror missing mmap method error code */ 1856 return -ENODEV; 1857} 1858EXPORT_SYMBOL(sock_no_mmap); 1859 1860ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags) 1861{ 1862 ssize_t res; 1863 struct msghdr msg = {.msg_flags = flags}; 1864 struct kvec iov; 1865 char *kaddr = kmap(page); 1866 iov.iov_base = kaddr + offset; 1867 iov.iov_len = size; 1868 res = kernel_sendmsg(sock, &msg, &iov, 1, size); 1869 kunmap(page); 1870 return res; 1871} 1872EXPORT_SYMBOL(sock_no_sendpage); 1873 1874/* 1875 * Default Socket Callbacks 1876 */ 1877 1878static void sock_def_wakeup(struct sock *sk) 1879{ 1880 struct socket_wq *wq; 1881 1882 rcu_read_lock(); 1883 wq = rcu_dereference(sk->sk_wq); 1884 if (wq_has_sleeper(wq)) 1885 wake_up_interruptible_all(&wq->wait); 1886 rcu_read_unlock(); 1887} 1888 1889static void sock_def_error_report(struct sock *sk) 1890{ 1891 struct socket_wq *wq; 1892 1893 rcu_read_lock(); 1894 wq = rcu_dereference(sk->sk_wq); 1895 if (wq_has_sleeper(wq)) 1896 wake_up_interruptible_poll(&wq->wait, POLLERR); 1897 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR); 1898 rcu_read_unlock(); 1899} 1900 1901static void sock_def_readable(struct sock *sk, int len) 1902{ 1903 struct socket_wq *wq; 1904 1905 rcu_read_lock(); 1906 wq = rcu_dereference(sk->sk_wq); 1907 if (wq_has_sleeper(wq)) 1908 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI | 1909 POLLRDNORM | POLLRDBAND); 1910 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN); 1911 rcu_read_unlock(); 1912} 1913 1914static void sock_def_write_space(struct sock *sk) 1915{ 1916 struct socket_wq *wq; 1917 1918 rcu_read_lock(); 1919 1920 /* Do not wake up a writer until he can make "significant" 1921 * progress. --DaveM 1922 */ 1923 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) { 1924 wq = rcu_dereference(sk->sk_wq); 1925 if (wq_has_sleeper(wq)) 1926 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT | 1927 POLLWRNORM | POLLWRBAND); 1928 1929 /* Should agree with poll, otherwise some programs break */ 1930 if (sock_writeable(sk)) 1931 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT); 1932 } 1933 1934 rcu_read_unlock(); 1935} 1936 1937static void sock_def_destruct(struct sock *sk) 1938{ 1939 kfree(sk->sk_protinfo); 1940} 1941 1942void sk_send_sigurg(struct sock *sk) 1943{ 1944 if (sk->sk_socket && sk->sk_socket->file) 1945 if (send_sigurg(&sk->sk_socket->file->f_owner)) 1946 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI); 1947} 1948EXPORT_SYMBOL(sk_send_sigurg); 1949 1950void sk_reset_timer(struct sock *sk, struct timer_list* timer, 1951 unsigned long expires) 1952{ 1953 if (!mod_timer(timer, expires)) 1954 sock_hold(sk); 1955} 1956EXPORT_SYMBOL(sk_reset_timer); 1957 1958void sk_stop_timer(struct sock *sk, struct timer_list* timer) 1959{ 1960 if (timer_pending(timer) && del_timer(timer)) 1961 __sock_put(sk); 1962} 1963EXPORT_SYMBOL(sk_stop_timer); 1964 1965void sock_init_data(struct socket *sock, struct sock *sk) 1966{ 1967 skb_queue_head_init(&sk->sk_receive_queue); 1968 skb_queue_head_init(&sk->sk_write_queue); 1969 skb_queue_head_init(&sk->sk_error_queue); 1970#ifdef CONFIG_NET_DMA 1971 skb_queue_head_init(&sk->sk_async_wait_queue); 1972#endif 1973 1974 sk->sk_send_head = NULL; 1975 1976 init_timer(&sk->sk_timer); 1977 1978 sk->sk_allocation = GFP_KERNEL; 1979 sk->sk_rcvbuf = sysctl_rmem_default; 1980 sk->sk_sndbuf = sysctl_wmem_default; 1981 sk->sk_state = TCP_CLOSE; 1982 sk_set_socket(sk, sock); 1983 1984 sock_set_flag(sk, SOCK_ZAPPED); 1985 1986 if (sock) { 1987 sk->sk_type = sock->type; 1988 sk->sk_wq = sock->wq; 1989 sock->sk = sk; 1990 } else 1991 sk->sk_wq = NULL; 1992 1993 spin_lock_init(&sk->sk_dst_lock); 1994 rwlock_init(&sk->sk_callback_lock); 1995 lockdep_set_class_and_name(&sk->sk_callback_lock, 1996 af_callback_keys + sk->sk_family, 1997 af_family_clock_key_strings[sk->sk_family]); 1998 1999 sk->sk_state_change = sock_def_wakeup; 2000 sk->sk_data_ready = sock_def_readable; 2001 sk->sk_write_space = sock_def_write_space; 2002 sk->sk_error_report = sock_def_error_report; 2003 sk->sk_destruct = sock_def_destruct; 2004 2005 sk->sk_sndmsg_page = NULL; 2006 sk->sk_sndmsg_off = 0; 2007 2008 sk->sk_peer_pid = NULL; 2009 sk->sk_peer_cred = NULL; 2010 sk->sk_write_pending = 0; 2011 sk->sk_rcvlowat = 1; 2012 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; 2013 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT; 2014 2015 sk->sk_stamp = ktime_set(-1L, 0); 2016 2017 /* 2018 * Before updating sk_refcnt, we must commit prior changes to memory 2019 * (Documentation/RCU/rculist_nulls.txt for details) 2020 */ 2021 smp_wmb(); 2022 atomic_set(&sk->sk_refcnt, 1); 2023 atomic_set(&sk->sk_drops, 0); 2024} 2025EXPORT_SYMBOL(sock_init_data); 2026 2027void lock_sock_nested(struct sock *sk, int subclass) 2028{ 2029 might_sleep(); 2030 spin_lock_bh(&sk->sk_lock.slock); 2031 if (sk->sk_lock.owned) 2032 __lock_sock(sk); 2033 sk->sk_lock.owned = 1; 2034 spin_unlock(&sk->sk_lock.slock); 2035 /* 2036 * The sk_lock has mutex_lock() semantics here: 2037 */ 2038 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_); 2039 local_bh_enable(); 2040} 2041EXPORT_SYMBOL(lock_sock_nested); 2042 2043void release_sock(struct sock *sk) 2044{ 2045 /* 2046 * The sk_lock has mutex_unlock() semantics: 2047 */ 2048 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_); 2049 2050 spin_lock_bh(&sk->sk_lock.slock); 2051 if (sk->sk_backlog.tail) 2052 __release_sock(sk); 2053 sk->sk_lock.owned = 0; 2054 if (waitqueue_active(&sk->sk_lock.wq)) 2055 wake_up(&sk->sk_lock.wq); 2056 spin_unlock_bh(&sk->sk_lock.slock); 2057} 2058EXPORT_SYMBOL(release_sock); 2059 2060/** 2061 * lock_sock_fast - fast version of lock_sock 2062 * @sk: socket 2063 * 2064 * This version should be used for very small section, where process wont block 2065 * return false if fast path is taken 2066 * sk_lock.slock locked, owned = 0, BH disabled 2067 * return true if slow path is taken 2068 * sk_lock.slock unlocked, owned = 1, BH enabled 2069 */ 2070bool lock_sock_fast(struct sock *sk) 2071{ 2072 might_sleep(); 2073 spin_lock_bh(&sk->sk_lock.slock); 2074 2075 if (!sk->sk_lock.owned) 2076 /* 2077 * Note : We must disable BH 2078 */ 2079 return false; 2080 2081 __lock_sock(sk); 2082 sk->sk_lock.owned = 1; 2083 spin_unlock(&sk->sk_lock.slock); 2084 /* 2085 * The sk_lock has mutex_lock() semantics here: 2086 */ 2087 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_); 2088 local_bh_enable(); 2089 return true; 2090} 2091EXPORT_SYMBOL(lock_sock_fast); 2092 2093int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp) 2094{ 2095 struct timeval tv; 2096 if (!sock_flag(sk, SOCK_TIMESTAMP)) 2097 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 2098 tv = ktime_to_timeval(sk->sk_stamp); 2099 if (tv.tv_sec == -1) 2100 return -ENOENT; 2101 if (tv.tv_sec == 0) { 2102 sk->sk_stamp = ktime_get_real(); 2103 tv = ktime_to_timeval(sk->sk_stamp); 2104 } 2105 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0; 2106} 2107EXPORT_SYMBOL(sock_get_timestamp); 2108 2109int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp) 2110{ 2111 struct timespec ts; 2112 if (!sock_flag(sk, SOCK_TIMESTAMP)) 2113 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 2114 ts = ktime_to_timespec(sk->sk_stamp); 2115 if (ts.tv_sec == -1) 2116 return -ENOENT; 2117 if (ts.tv_sec == 0) { 2118 sk->sk_stamp = ktime_get_real(); 2119 ts = ktime_to_timespec(sk->sk_stamp); 2120 } 2121 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0; 2122} 2123EXPORT_SYMBOL(sock_get_timestampns); 2124 2125void sock_enable_timestamp(struct sock *sk, int flag) 2126{ 2127 if (!sock_flag(sk, flag)) { 2128 sock_set_flag(sk, flag); 2129 /* 2130 * we just set one of the two flags which require net 2131 * time stamping, but time stamping might have been on 2132 * already because of the other one 2133 */ 2134 if (!sock_flag(sk, 2135 flag == SOCK_TIMESTAMP ? 2136 SOCK_TIMESTAMPING_RX_SOFTWARE : 2137 SOCK_TIMESTAMP)) 2138 net_enable_timestamp(); 2139 } 2140} 2141 2142/* 2143 * Get a socket option on an socket. 2144 * 2145 * FIX: POSIX 1003.1g is very ambiguous here. It states that 2146 * asynchronous errors should be reported by getsockopt. We assume 2147 * this means if you specify SO_ERROR (otherwise whats the point of it). 2148 */ 2149int sock_common_getsockopt(struct socket *sock, int level, int optname, 2150 char __user *optval, int __user *optlen) 2151{ 2152 struct sock *sk = sock->sk; 2153 2154 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen); 2155} 2156EXPORT_SYMBOL(sock_common_getsockopt); 2157 2158#ifdef CONFIG_COMPAT 2159int compat_sock_common_getsockopt(struct socket *sock, int level, int optname, 2160 char __user *optval, int __user *optlen) 2161{ 2162 struct sock *sk = sock->sk; 2163 2164 if (sk->sk_prot->compat_getsockopt != NULL) 2165 return sk->sk_prot->compat_getsockopt(sk, level, optname, 2166 optval, optlen); 2167 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen); 2168} 2169EXPORT_SYMBOL(compat_sock_common_getsockopt); 2170#endif 2171 2172int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock, 2173 struct msghdr *msg, size_t size, int flags) 2174{ 2175 struct sock *sk = sock->sk; 2176 int addr_len = 0; 2177 int err; 2178 2179 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT, 2180 flags & ~MSG_DONTWAIT, &addr_len); 2181 if (err >= 0) 2182 msg->msg_namelen = addr_len; 2183 return err; 2184} 2185EXPORT_SYMBOL(sock_common_recvmsg); 2186 2187/* 2188 * Set socket options on an inet socket. 2189 */ 2190int sock_common_setsockopt(struct socket *sock, int level, int optname, 2191 char __user *optval, unsigned int optlen) 2192{ 2193 struct sock *sk = sock->sk; 2194 2195 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen); 2196} 2197EXPORT_SYMBOL(sock_common_setsockopt); 2198 2199#ifdef CONFIG_COMPAT 2200int compat_sock_common_setsockopt(struct socket *sock, int level, int optname, 2201 char __user *optval, unsigned int optlen) 2202{ 2203 struct sock *sk = sock->sk; 2204 2205 if (sk->sk_prot->compat_setsockopt != NULL) 2206 return sk->sk_prot->compat_setsockopt(sk, level, optname, 2207 optval, optlen); 2208 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen); 2209} 2210EXPORT_SYMBOL(compat_sock_common_setsockopt); 2211#endif 2212 2213void sk_common_release(struct sock *sk) 2214{ 2215 if (sk->sk_prot->destroy) 2216 sk->sk_prot->destroy(sk); 2217 2218 /* 2219 * Observation: when sock_common_release is called, processes have 2220 * no access to socket. But net still has. 2221 * Step one, detach it from networking: 2222 * 2223 * A. Remove from hash tables. 2224 */ 2225 2226 sk->sk_prot->unhash(sk); 2227 2228 /* 2229 * In this point socket cannot receive new packets, but it is possible 2230 * that some packets are in flight because some CPU runs receiver and 2231 * did hash table lookup before we unhashed socket. They will achieve 2232 * receive queue and will be purged by socket destructor. 2233 * 2234 * Also we still have packets pending on receive queue and probably, 2235 * our own packets waiting in device queues. sock_destroy will drain 2236 * receive queue, but transmitted packets will delay socket destruction 2237 * until the last reference will be released. 2238 */ 2239 2240 sock_orphan(sk); 2241 2242 xfrm_sk_free_policy(sk); 2243 2244 sk_refcnt_debug_release(sk); 2245 sock_put(sk); 2246} 2247EXPORT_SYMBOL(sk_common_release); 2248 2249static DEFINE_RWLOCK(proto_list_lock); 2250static LIST_HEAD(proto_list); 2251 2252#ifdef CONFIG_PROC_FS 2253#define PROTO_INUSE_NR 64 /* should be enough for the first time */ 2254struct prot_inuse { 2255 int val[PROTO_INUSE_NR]; 2256}; 2257 2258static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR); 2259 2260#ifdef CONFIG_NET_NS 2261void sock_prot_inuse_add(struct net *net, struct proto *prot, int val) 2262{ 2263 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val); 2264} 2265EXPORT_SYMBOL_GPL(sock_prot_inuse_add); 2266 2267int sock_prot_inuse_get(struct net *net, struct proto *prot) 2268{ 2269 int cpu, idx = prot->inuse_idx; 2270 int res = 0; 2271 2272 for_each_possible_cpu(cpu) 2273 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx]; 2274 2275 return res >= 0 ? res : 0; 2276} 2277EXPORT_SYMBOL_GPL(sock_prot_inuse_get); 2278 2279static int __net_init sock_inuse_init_net(struct net *net) 2280{ 2281 net->core.inuse = alloc_percpu(struct prot_inuse); 2282 return net->core.inuse ? 0 : -ENOMEM; 2283} 2284 2285static void __net_exit sock_inuse_exit_net(struct net *net) 2286{ 2287 free_percpu(net->core.inuse); 2288} 2289 2290static struct pernet_operations net_inuse_ops = { 2291 .init = sock_inuse_init_net, 2292 .exit = sock_inuse_exit_net, 2293}; 2294 2295static __init int net_inuse_init(void) 2296{ 2297 if (register_pernet_subsys(&net_inuse_ops)) 2298 panic("Cannot initialize net inuse counters"); 2299 2300 return 0; 2301} 2302 2303core_initcall(net_inuse_init); 2304#else 2305static DEFINE_PER_CPU(struct prot_inuse, prot_inuse); 2306 2307void sock_prot_inuse_add(struct net *net, struct proto *prot, int val) 2308{ 2309 __this_cpu_add(prot_inuse.val[prot->inuse_idx], val); 2310} 2311EXPORT_SYMBOL_GPL(sock_prot_inuse_add); 2312 2313int sock_prot_inuse_get(struct net *net, struct proto *prot) 2314{ 2315 int cpu, idx = prot->inuse_idx; 2316 int res = 0; 2317 2318 for_each_possible_cpu(cpu) 2319 res += per_cpu(prot_inuse, cpu).val[idx]; 2320 2321 return res >= 0 ? res : 0; 2322} 2323EXPORT_SYMBOL_GPL(sock_prot_inuse_get); 2324#endif 2325 2326static void assign_proto_idx(struct proto *prot) 2327{ 2328 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR); 2329 2330 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) { 2331 printk(KERN_ERR "PROTO_INUSE_NR exhausted\n"); 2332 return; 2333 } 2334 2335 set_bit(prot->inuse_idx, proto_inuse_idx); 2336} 2337 2338static void release_proto_idx(struct proto *prot) 2339{ 2340 if (prot->inuse_idx != PROTO_INUSE_NR - 1) 2341 clear_bit(prot->inuse_idx, proto_inuse_idx); 2342} 2343#else 2344static inline void assign_proto_idx(struct proto *prot) 2345{ 2346} 2347 2348static inline void release_proto_idx(struct proto *prot) 2349{ 2350} 2351#endif 2352 2353int proto_register(struct proto *prot, int alloc_slab) 2354{ 2355 if (alloc_slab) { 2356 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0, 2357 SLAB_HWCACHE_ALIGN | prot->slab_flags, 2358 NULL); 2359 2360 if (prot->slab == NULL) { 2361 printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n", 2362 prot->name); 2363 goto out; 2364 } 2365 2366 if (prot->rsk_prot != NULL) { 2367 prot->rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", prot->name); 2368 if (prot->rsk_prot->slab_name == NULL) 2369 goto out_free_sock_slab; 2370 2371 prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name, 2372 prot->rsk_prot->obj_size, 0, 2373 SLAB_HWCACHE_ALIGN, NULL); 2374 2375 if (prot->rsk_prot->slab == NULL) { 2376 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n", 2377 prot->name); 2378 goto out_free_request_sock_slab_name; 2379 } 2380 } 2381 2382 if (prot->twsk_prot != NULL) { 2383 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name); 2384 2385 if (prot->twsk_prot->twsk_slab_name == NULL) 2386 goto out_free_request_sock_slab; 2387 2388 prot->twsk_prot->twsk_slab = 2389 kmem_cache_create(prot->twsk_prot->twsk_slab_name, 2390 prot->twsk_prot->twsk_obj_size, 2391 0, 2392 SLAB_HWCACHE_ALIGN | 2393 prot->slab_flags, 2394 NULL); 2395 if (prot->twsk_prot->twsk_slab == NULL) 2396 goto out_free_timewait_sock_slab_name; 2397 } 2398 } 2399 2400 write_lock(&proto_list_lock); 2401 list_add(&prot->node, &proto_list); 2402 assign_proto_idx(prot); 2403 write_unlock(&proto_list_lock); 2404 return 0; 2405 2406out_free_timewait_sock_slab_name: 2407 kfree(prot->twsk_prot->twsk_slab_name); 2408out_free_request_sock_slab: 2409 if (prot->rsk_prot && prot->rsk_prot->slab) { 2410 kmem_cache_destroy(prot->rsk_prot->slab); 2411 prot->rsk_prot->slab = NULL; 2412 } 2413out_free_request_sock_slab_name: 2414 if (prot->rsk_prot) 2415 kfree(prot->rsk_prot->slab_name); 2416out_free_sock_slab: 2417 kmem_cache_destroy(prot->slab); 2418 prot->slab = NULL; 2419out: 2420 return -ENOBUFS; 2421} 2422EXPORT_SYMBOL(proto_register); 2423 2424void proto_unregister(struct proto *prot) 2425{ 2426 write_lock(&proto_list_lock); 2427 release_proto_idx(prot); 2428 list_del(&prot->node); 2429 write_unlock(&proto_list_lock); 2430 2431 if (prot->slab != NULL) { 2432 kmem_cache_destroy(prot->slab); 2433 prot->slab = NULL; 2434 } 2435 2436 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) { 2437 kmem_cache_destroy(prot->rsk_prot->slab); 2438 kfree(prot->rsk_prot->slab_name); 2439 prot->rsk_prot->slab = NULL; 2440 } 2441 2442 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) { 2443 kmem_cache_destroy(prot->twsk_prot->twsk_slab); 2444 kfree(prot->twsk_prot->twsk_slab_name); 2445 prot->twsk_prot->twsk_slab = NULL; 2446 } 2447} 2448EXPORT_SYMBOL(proto_unregister); 2449 2450#ifdef CONFIG_PROC_FS 2451static void *proto_seq_start(struct seq_file *seq, loff_t *pos) 2452 __acquires(proto_list_lock) 2453{ 2454 read_lock(&proto_list_lock); 2455 return seq_list_start_head(&proto_list, *pos); 2456} 2457 2458static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos) 2459{ 2460 return seq_list_next(v, &proto_list, pos); 2461} 2462 2463static void proto_seq_stop(struct seq_file *seq, void *v) 2464 __releases(proto_list_lock) 2465{ 2466 read_unlock(&proto_list_lock); 2467} 2468 2469static char proto_method_implemented(const void *method) 2470{ 2471 return method == NULL ? 'n' : 'y'; 2472} 2473 2474static void proto_seq_printf(struct seq_file *seq, struct proto *proto) 2475{ 2476 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s " 2477 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n", 2478 proto->name, 2479 proto->obj_size, 2480 sock_prot_inuse_get(seq_file_net(seq), proto), 2481 proto->memory_allocated != NULL ? atomic_long_read(proto->memory_allocated) : -1L, 2482 proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI", 2483 proto->max_header, 2484 proto->slab == NULL ? "no" : "yes", 2485 module_name(proto->owner), 2486 proto_method_implemented(proto->close), 2487 proto_method_implemented(proto->connect), 2488 proto_method_implemented(proto->disconnect), 2489 proto_method_implemented(proto->accept), 2490 proto_method_implemented(proto->ioctl), 2491 proto_method_implemented(proto->init), 2492 proto_method_implemented(proto->destroy), 2493 proto_method_implemented(proto->shutdown), 2494 proto_method_implemented(proto->setsockopt), 2495 proto_method_implemented(proto->getsockopt), 2496 proto_method_implemented(proto->sendmsg), 2497 proto_method_implemented(proto->recvmsg), 2498 proto_method_implemented(proto->sendpage), 2499 proto_method_implemented(proto->bind), 2500 proto_method_implemented(proto->backlog_rcv), 2501 proto_method_implemented(proto->hash), 2502 proto_method_implemented(proto->unhash), 2503 proto_method_implemented(proto->get_port), 2504 proto_method_implemented(proto->enter_memory_pressure)); 2505} 2506 2507static int proto_seq_show(struct seq_file *seq, void *v) 2508{ 2509 if (v == &proto_list) 2510 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s", 2511 "protocol", 2512 "size", 2513 "sockets", 2514 "memory", 2515 "press", 2516 "maxhdr", 2517 "slab", 2518 "module", 2519 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n"); 2520 else 2521 proto_seq_printf(seq, list_entry(v, struct proto, node)); 2522 return 0; 2523} 2524 2525static const struct seq_operations proto_seq_ops = { 2526 .start = proto_seq_start, 2527 .next = proto_seq_next, 2528 .stop = proto_seq_stop, 2529 .show = proto_seq_show, 2530}; 2531 2532static int proto_seq_open(struct inode *inode, struct file *file) 2533{ 2534 return seq_open_net(inode, file, &proto_seq_ops, 2535 sizeof(struct seq_net_private)); 2536} 2537 2538static const struct file_operations proto_seq_fops = { 2539 .owner = THIS_MODULE, 2540 .open = proto_seq_open, 2541 .read = seq_read, 2542 .llseek = seq_lseek, 2543 .release = seq_release_net, 2544}; 2545 2546static __net_init int proto_init_net(struct net *net) 2547{ 2548 if (!proc_net_fops_create(net, "protocols", S_IRUGO, &proto_seq_fops)) 2549 return -ENOMEM; 2550 2551 return 0; 2552} 2553 2554static __net_exit void proto_exit_net(struct net *net) 2555{ 2556 proc_net_remove(net, "protocols"); 2557} 2558 2559 2560static __net_initdata struct pernet_operations proto_net_ops = { 2561 .init = proto_init_net, 2562 .exit = proto_exit_net, 2563}; 2564 2565static int __init proto_init(void) 2566{ 2567 return register_pernet_subsys(&proto_net_ops); 2568} 2569 2570subsys_initcall(proto_init); 2571 2572#endif /* PROC_FS */ 2573