sock.c revision 51e4e7faba786d33e5e33f8776c5027a1c8d6fb7
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#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 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#include <linux/user_namespace.h> 116#include <linux/static_key.h> 117#include <linux/memcontrol.h> 118#include <linux/prefetch.h> 119 120#include <asm/uaccess.h> 121 122#include <linux/netdevice.h> 123#include <net/protocol.h> 124#include <linux/skbuff.h> 125#include <net/net_namespace.h> 126#include <net/request_sock.h> 127#include <net/sock.h> 128#include <linux/net_tstamp.h> 129#include <net/xfrm.h> 130#include <linux/ipsec.h> 131#include <net/cls_cgroup.h> 132#include <net/netprio_cgroup.h> 133 134#include <linux/filter.h> 135 136#include <trace/events/sock.h> 137 138#ifdef CONFIG_INET 139#include <net/tcp.h> 140#endif 141 142static DEFINE_MUTEX(proto_list_mutex); 143static LIST_HEAD(proto_list); 144 145#ifdef CONFIG_MEMCG_KMEM 146int mem_cgroup_sockets_init(struct mem_cgroup *memcg, struct cgroup_subsys *ss) 147{ 148 struct proto *proto; 149 int ret = 0; 150 151 mutex_lock(&proto_list_mutex); 152 list_for_each_entry(proto, &proto_list, node) { 153 if (proto->init_cgroup) { 154 ret = proto->init_cgroup(memcg, ss); 155 if (ret) 156 goto out; 157 } 158 } 159 160 mutex_unlock(&proto_list_mutex); 161 return ret; 162out: 163 list_for_each_entry_continue_reverse(proto, &proto_list, node) 164 if (proto->destroy_cgroup) 165 proto->destroy_cgroup(memcg); 166 mutex_unlock(&proto_list_mutex); 167 return ret; 168} 169 170void mem_cgroup_sockets_destroy(struct mem_cgroup *memcg) 171{ 172 struct proto *proto; 173 174 mutex_lock(&proto_list_mutex); 175 list_for_each_entry_reverse(proto, &proto_list, node) 176 if (proto->destroy_cgroup) 177 proto->destroy_cgroup(memcg); 178 mutex_unlock(&proto_list_mutex); 179} 180#endif 181 182/* 183 * Each address family might have different locking rules, so we have 184 * one slock key per address family: 185 */ 186static struct lock_class_key af_family_keys[AF_MAX]; 187static struct lock_class_key af_family_slock_keys[AF_MAX]; 188 189struct static_key memcg_socket_limit_enabled; 190EXPORT_SYMBOL(memcg_socket_limit_enabled); 191 192/* 193 * Make lock validator output more readable. (we pre-construct these 194 * strings build-time, so that runtime initialization of socket 195 * locks is fast): 196 */ 197static const char *const af_family_key_strings[AF_MAX+1] = { 198 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" , 199 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK", 200 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" , 201 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" , 202 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" , 203 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" , 204 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" , 205 "sk_lock-AF_RDS" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" , 206 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" , 207 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" , 208 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" , 209 "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN" , "sk_lock-AF_PHONET" , 210 "sk_lock-AF_IEEE802154", "sk_lock-AF_CAIF" , "sk_lock-AF_ALG" , 211 "sk_lock-AF_NFC" , "sk_lock-AF_MAX" 212}; 213static const char *const af_family_slock_key_strings[AF_MAX+1] = { 214 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" , 215 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK", 216 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" , 217 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" , 218 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" , 219 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" , 220 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" , 221 "slock-AF_RDS" , "slock-AF_SNA" , "slock-AF_IRDA" , 222 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" , 223 "slock-27" , "slock-28" , "slock-AF_CAN" , 224 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" , 225 "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" , 226 "slock-AF_IEEE802154", "slock-AF_CAIF" , "slock-AF_ALG" , 227 "slock-AF_NFC" , "slock-AF_MAX" 228}; 229static const char *const af_family_clock_key_strings[AF_MAX+1] = { 230 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" , 231 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK", 232 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" , 233 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" , 234 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" , 235 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" , 236 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" , 237 "clock-AF_RDS" , "clock-AF_SNA" , "clock-AF_IRDA" , 238 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" , 239 "clock-27" , "clock-28" , "clock-AF_CAN" , 240 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" , 241 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" , 242 "clock-AF_IEEE802154", "clock-AF_CAIF" , "clock-AF_ALG" , 243 "clock-AF_NFC" , "clock-AF_MAX" 244}; 245 246/* 247 * sk_callback_lock locking rules are per-address-family, 248 * so split the lock classes by using a per-AF key: 249 */ 250static struct lock_class_key af_callback_keys[AF_MAX]; 251 252/* Take into consideration the size of the struct sk_buff overhead in the 253 * determination of these values, since that is non-constant across 254 * platforms. This makes socket queueing behavior and performance 255 * not depend upon such differences. 256 */ 257#define _SK_MEM_PACKETS 256 258#define _SK_MEM_OVERHEAD SKB_TRUESIZE(256) 259#define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 260#define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 261 262/* Run time adjustable parameters. */ 263__u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX; 264EXPORT_SYMBOL(sysctl_wmem_max); 265__u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX; 266EXPORT_SYMBOL(sysctl_rmem_max); 267__u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX; 268__u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX; 269 270/* Maximal space eaten by iovec or ancillary data plus some space */ 271int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512); 272EXPORT_SYMBOL(sysctl_optmem_max); 273 274struct static_key memalloc_socks = STATIC_KEY_INIT_FALSE; 275EXPORT_SYMBOL_GPL(memalloc_socks); 276 277/** 278 * sk_set_memalloc - sets %SOCK_MEMALLOC 279 * @sk: socket to set it on 280 * 281 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves. 282 * It's the responsibility of the admin to adjust min_free_kbytes 283 * to meet the requirements 284 */ 285void sk_set_memalloc(struct sock *sk) 286{ 287 sock_set_flag(sk, SOCK_MEMALLOC); 288 sk->sk_allocation |= __GFP_MEMALLOC; 289 static_key_slow_inc(&memalloc_socks); 290} 291EXPORT_SYMBOL_GPL(sk_set_memalloc); 292 293void sk_clear_memalloc(struct sock *sk) 294{ 295 sock_reset_flag(sk, SOCK_MEMALLOC); 296 sk->sk_allocation &= ~__GFP_MEMALLOC; 297 static_key_slow_dec(&memalloc_socks); 298 299 /* 300 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward 301 * progress of swapping. However, if SOCK_MEMALLOC is cleared while 302 * it has rmem allocations there is a risk that the user of the 303 * socket cannot make forward progress due to exceeding the rmem 304 * limits. By rights, sk_clear_memalloc() should only be called 305 * on sockets being torn down but warn and reset the accounting if 306 * that assumption breaks. 307 */ 308 if (WARN_ON(sk->sk_forward_alloc)) 309 sk_mem_reclaim(sk); 310} 311EXPORT_SYMBOL_GPL(sk_clear_memalloc); 312 313int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb) 314{ 315 int ret; 316 unsigned long pflags = current->flags; 317 318 /* these should have been dropped before queueing */ 319 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC)); 320 321 current->flags |= PF_MEMALLOC; 322 ret = sk->sk_backlog_rcv(sk, skb); 323 tsk_restore_flags(current, pflags, PF_MEMALLOC); 324 325 return ret; 326} 327EXPORT_SYMBOL(__sk_backlog_rcv); 328 329#if defined(CONFIG_CGROUPS) 330#if !defined(CONFIG_NET_CLS_CGROUP) 331int net_cls_subsys_id = -1; 332EXPORT_SYMBOL_GPL(net_cls_subsys_id); 333#endif 334#if !defined(CONFIG_NETPRIO_CGROUP) 335int net_prio_subsys_id = -1; 336EXPORT_SYMBOL_GPL(net_prio_subsys_id); 337#endif 338#endif 339 340static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen) 341{ 342 struct timeval tv; 343 344 if (optlen < sizeof(tv)) 345 return -EINVAL; 346 if (copy_from_user(&tv, optval, sizeof(tv))) 347 return -EFAULT; 348 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC) 349 return -EDOM; 350 351 if (tv.tv_sec < 0) { 352 static int warned __read_mostly; 353 354 *timeo_p = 0; 355 if (warned < 10 && net_ratelimit()) { 356 warned++; 357 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n", 358 __func__, current->comm, task_pid_nr(current)); 359 } 360 return 0; 361 } 362 *timeo_p = MAX_SCHEDULE_TIMEOUT; 363 if (tv.tv_sec == 0 && tv.tv_usec == 0) 364 return 0; 365 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1)) 366 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ); 367 return 0; 368} 369 370static void sock_warn_obsolete_bsdism(const char *name) 371{ 372 static int warned; 373 static char warncomm[TASK_COMM_LEN]; 374 if (strcmp(warncomm, current->comm) && warned < 5) { 375 strcpy(warncomm, current->comm); 376 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n", 377 warncomm, name); 378 warned++; 379 } 380} 381 382#define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE)) 383 384static void sock_disable_timestamp(struct sock *sk, unsigned long flags) 385{ 386 if (sk->sk_flags & flags) { 387 sk->sk_flags &= ~flags; 388 if (!(sk->sk_flags & SK_FLAGS_TIMESTAMP)) 389 net_disable_timestamp(); 390 } 391} 392 393 394int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 395{ 396 int err; 397 int skb_len; 398 unsigned long flags; 399 struct sk_buff_head *list = &sk->sk_receive_queue; 400 401 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) { 402 atomic_inc(&sk->sk_drops); 403 trace_sock_rcvqueue_full(sk, skb); 404 return -ENOMEM; 405 } 406 407 err = sk_filter(sk, skb); 408 if (err) 409 return err; 410 411 if (!sk_rmem_schedule(sk, skb, skb->truesize)) { 412 atomic_inc(&sk->sk_drops); 413 return -ENOBUFS; 414 } 415 416 skb->dev = NULL; 417 skb_set_owner_r(skb, sk); 418 419 /* Cache the SKB length before we tack it onto the receive 420 * queue. Once it is added it no longer belongs to us and 421 * may be freed by other threads of control pulling packets 422 * from the queue. 423 */ 424 skb_len = skb->len; 425 426 /* we escape from rcu protected region, make sure we dont leak 427 * a norefcounted dst 428 */ 429 skb_dst_force(skb); 430 431 spin_lock_irqsave(&list->lock, flags); 432 skb->dropcount = atomic_read(&sk->sk_drops); 433 __skb_queue_tail(list, skb); 434 spin_unlock_irqrestore(&list->lock, flags); 435 436 if (!sock_flag(sk, SOCK_DEAD)) 437 sk->sk_data_ready(sk, skb_len); 438 return 0; 439} 440EXPORT_SYMBOL(sock_queue_rcv_skb); 441 442int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested) 443{ 444 int rc = NET_RX_SUCCESS; 445 446 if (sk_filter(sk, skb)) 447 goto discard_and_relse; 448 449 skb->dev = NULL; 450 451 if (sk_rcvqueues_full(sk, skb, sk->sk_rcvbuf)) { 452 atomic_inc(&sk->sk_drops); 453 goto discard_and_relse; 454 } 455 if (nested) 456 bh_lock_sock_nested(sk); 457 else 458 bh_lock_sock(sk); 459 if (!sock_owned_by_user(sk)) { 460 /* 461 * trylock + unlock semantics: 462 */ 463 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_); 464 465 rc = sk_backlog_rcv(sk, skb); 466 467 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_); 468 } else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) { 469 bh_unlock_sock(sk); 470 atomic_inc(&sk->sk_drops); 471 goto discard_and_relse; 472 } 473 474 bh_unlock_sock(sk); 475out: 476 sock_put(sk); 477 return rc; 478discard_and_relse: 479 kfree_skb(skb); 480 goto out; 481} 482EXPORT_SYMBOL(sk_receive_skb); 483 484void sk_reset_txq(struct sock *sk) 485{ 486 sk_tx_queue_clear(sk); 487} 488EXPORT_SYMBOL(sk_reset_txq); 489 490struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie) 491{ 492 struct dst_entry *dst = __sk_dst_get(sk); 493 494 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) { 495 sk_tx_queue_clear(sk); 496 RCU_INIT_POINTER(sk->sk_dst_cache, NULL); 497 dst_release(dst); 498 return NULL; 499 } 500 501 return dst; 502} 503EXPORT_SYMBOL(__sk_dst_check); 504 505struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie) 506{ 507 struct dst_entry *dst = sk_dst_get(sk); 508 509 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) { 510 sk_dst_reset(sk); 511 dst_release(dst); 512 return NULL; 513 } 514 515 return dst; 516} 517EXPORT_SYMBOL(sk_dst_check); 518 519static int sock_bindtodevice(struct sock *sk, char __user *optval, int optlen) 520{ 521 int ret = -ENOPROTOOPT; 522#ifdef CONFIG_NETDEVICES 523 struct net *net = sock_net(sk); 524 char devname[IFNAMSIZ]; 525 int index; 526 527 /* Sorry... */ 528 ret = -EPERM; 529 if (!capable(CAP_NET_RAW)) 530 goto out; 531 532 ret = -EINVAL; 533 if (optlen < 0) 534 goto out; 535 536 /* Bind this socket to a particular device like "eth0", 537 * as specified in the passed interface name. If the 538 * name is "" or the option length is zero the socket 539 * is not bound. 540 */ 541 if (optlen > IFNAMSIZ - 1) 542 optlen = IFNAMSIZ - 1; 543 memset(devname, 0, sizeof(devname)); 544 545 ret = -EFAULT; 546 if (copy_from_user(devname, optval, optlen)) 547 goto out; 548 549 index = 0; 550 if (devname[0] != '\0') { 551 struct net_device *dev; 552 553 rcu_read_lock(); 554 dev = dev_get_by_name_rcu(net, devname); 555 if (dev) 556 index = dev->ifindex; 557 rcu_read_unlock(); 558 ret = -ENODEV; 559 if (!dev) 560 goto out; 561 } 562 563 lock_sock(sk); 564 sk->sk_bound_dev_if = index; 565 sk_dst_reset(sk); 566 release_sock(sk); 567 568 ret = 0; 569 570out: 571#endif 572 573 return ret; 574} 575 576static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool) 577{ 578 if (valbool) 579 sock_set_flag(sk, bit); 580 else 581 sock_reset_flag(sk, bit); 582} 583 584/* 585 * This is meant for all protocols to use and covers goings on 586 * at the socket level. Everything here is generic. 587 */ 588 589int sock_setsockopt(struct socket *sock, int level, int optname, 590 char __user *optval, unsigned int optlen) 591{ 592 struct sock *sk = sock->sk; 593 int val; 594 int valbool; 595 struct linger ling; 596 int ret = 0; 597 598 /* 599 * Options without arguments 600 */ 601 602 if (optname == SO_BINDTODEVICE) 603 return sock_bindtodevice(sk, optval, optlen); 604 605 if (optlen < sizeof(int)) 606 return -EINVAL; 607 608 if (get_user(val, (int __user *)optval)) 609 return -EFAULT; 610 611 valbool = val ? 1 : 0; 612 613 lock_sock(sk); 614 615 switch (optname) { 616 case SO_DEBUG: 617 if (val && !capable(CAP_NET_ADMIN)) 618 ret = -EACCES; 619 else 620 sock_valbool_flag(sk, SOCK_DBG, valbool); 621 break; 622 case SO_REUSEADDR: 623 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE); 624 break; 625 case SO_TYPE: 626 case SO_PROTOCOL: 627 case SO_DOMAIN: 628 case SO_ERROR: 629 ret = -ENOPROTOOPT; 630 break; 631 case SO_DONTROUTE: 632 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool); 633 break; 634 case SO_BROADCAST: 635 sock_valbool_flag(sk, SOCK_BROADCAST, valbool); 636 break; 637 case SO_SNDBUF: 638 /* Don't error on this BSD doesn't and if you think 639 * about it this is right. Otherwise apps have to 640 * play 'guess the biggest size' games. RCVBUF/SNDBUF 641 * are treated in BSD as hints 642 */ 643 val = min_t(u32, val, sysctl_wmem_max); 644set_sndbuf: 645 sk->sk_userlocks |= SOCK_SNDBUF_LOCK; 646 sk->sk_sndbuf = max_t(u32, val * 2, SOCK_MIN_SNDBUF); 647 /* Wake up sending tasks if we upped the value. */ 648 sk->sk_write_space(sk); 649 break; 650 651 case SO_SNDBUFFORCE: 652 if (!capable(CAP_NET_ADMIN)) { 653 ret = -EPERM; 654 break; 655 } 656 goto set_sndbuf; 657 658 case SO_RCVBUF: 659 /* Don't error on this BSD doesn't and if you think 660 * about it this is right. Otherwise apps have to 661 * play 'guess the biggest size' games. RCVBUF/SNDBUF 662 * are treated in BSD as hints 663 */ 664 val = min_t(u32, val, sysctl_rmem_max); 665set_rcvbuf: 666 sk->sk_userlocks |= SOCK_RCVBUF_LOCK; 667 /* 668 * We double it on the way in to account for 669 * "struct sk_buff" etc. overhead. Applications 670 * assume that the SO_RCVBUF setting they make will 671 * allow that much actual data to be received on that 672 * socket. 673 * 674 * Applications are unaware that "struct sk_buff" and 675 * other overheads allocate from the receive buffer 676 * during socket buffer allocation. 677 * 678 * And after considering the possible alternatives, 679 * returning the value we actually used in getsockopt 680 * is the most desirable behavior. 681 */ 682 sk->sk_rcvbuf = max_t(u32, val * 2, SOCK_MIN_RCVBUF); 683 break; 684 685 case SO_RCVBUFFORCE: 686 if (!capable(CAP_NET_ADMIN)) { 687 ret = -EPERM; 688 break; 689 } 690 goto set_rcvbuf; 691 692 case SO_KEEPALIVE: 693#ifdef CONFIG_INET 694 if (sk->sk_protocol == IPPROTO_TCP) 695 tcp_set_keepalive(sk, valbool); 696#endif 697 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool); 698 break; 699 700 case SO_OOBINLINE: 701 sock_valbool_flag(sk, SOCK_URGINLINE, valbool); 702 break; 703 704 case SO_NO_CHECK: 705 sk->sk_no_check = valbool; 706 break; 707 708 case SO_PRIORITY: 709 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN)) 710 sk->sk_priority = val; 711 else 712 ret = -EPERM; 713 break; 714 715 case SO_LINGER: 716 if (optlen < sizeof(ling)) { 717 ret = -EINVAL; /* 1003.1g */ 718 break; 719 } 720 if (copy_from_user(&ling, optval, sizeof(ling))) { 721 ret = -EFAULT; 722 break; 723 } 724 if (!ling.l_onoff) 725 sock_reset_flag(sk, SOCK_LINGER); 726 else { 727#if (BITS_PER_LONG == 32) 728 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ) 729 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT; 730 else 731#endif 732 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ; 733 sock_set_flag(sk, SOCK_LINGER); 734 } 735 break; 736 737 case SO_BSDCOMPAT: 738 sock_warn_obsolete_bsdism("setsockopt"); 739 break; 740 741 case SO_PASSCRED: 742 if (valbool) 743 set_bit(SOCK_PASSCRED, &sock->flags); 744 else 745 clear_bit(SOCK_PASSCRED, &sock->flags); 746 break; 747 748 case SO_TIMESTAMP: 749 case SO_TIMESTAMPNS: 750 if (valbool) { 751 if (optname == SO_TIMESTAMP) 752 sock_reset_flag(sk, SOCK_RCVTSTAMPNS); 753 else 754 sock_set_flag(sk, SOCK_RCVTSTAMPNS); 755 sock_set_flag(sk, SOCK_RCVTSTAMP); 756 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 757 } else { 758 sock_reset_flag(sk, SOCK_RCVTSTAMP); 759 sock_reset_flag(sk, SOCK_RCVTSTAMPNS); 760 } 761 break; 762 763 case SO_TIMESTAMPING: 764 if (val & ~SOF_TIMESTAMPING_MASK) { 765 ret = -EINVAL; 766 break; 767 } 768 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE, 769 val & SOF_TIMESTAMPING_TX_HARDWARE); 770 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE, 771 val & SOF_TIMESTAMPING_TX_SOFTWARE); 772 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE, 773 val & SOF_TIMESTAMPING_RX_HARDWARE); 774 if (val & SOF_TIMESTAMPING_RX_SOFTWARE) 775 sock_enable_timestamp(sk, 776 SOCK_TIMESTAMPING_RX_SOFTWARE); 777 else 778 sock_disable_timestamp(sk, 779 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE)); 780 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SOFTWARE, 781 val & SOF_TIMESTAMPING_SOFTWARE); 782 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE, 783 val & SOF_TIMESTAMPING_SYS_HARDWARE); 784 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE, 785 val & SOF_TIMESTAMPING_RAW_HARDWARE); 786 break; 787 788 case SO_RCVLOWAT: 789 if (val < 0) 790 val = INT_MAX; 791 sk->sk_rcvlowat = val ? : 1; 792 break; 793 794 case SO_RCVTIMEO: 795 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen); 796 break; 797 798 case SO_SNDTIMEO: 799 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen); 800 break; 801 802 case SO_ATTACH_FILTER: 803 ret = -EINVAL; 804 if (optlen == sizeof(struct sock_fprog)) { 805 struct sock_fprog fprog; 806 807 ret = -EFAULT; 808 if (copy_from_user(&fprog, optval, sizeof(fprog))) 809 break; 810 811 ret = sk_attach_filter(&fprog, sk); 812 } 813 break; 814 815 case SO_DETACH_FILTER: 816 ret = sk_detach_filter(sk); 817 break; 818 819 case SO_PASSSEC: 820 if (valbool) 821 set_bit(SOCK_PASSSEC, &sock->flags); 822 else 823 clear_bit(SOCK_PASSSEC, &sock->flags); 824 break; 825 case SO_MARK: 826 if (!capable(CAP_NET_ADMIN)) 827 ret = -EPERM; 828 else 829 sk->sk_mark = val; 830 break; 831 832 /* We implement the SO_SNDLOWAT etc to 833 not be settable (1003.1g 5.3) */ 834 case SO_RXQ_OVFL: 835 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool); 836 break; 837 838 case SO_WIFI_STATUS: 839 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool); 840 break; 841 842 case SO_PEEK_OFF: 843 if (sock->ops->set_peek_off) 844 sock->ops->set_peek_off(sk, val); 845 else 846 ret = -EOPNOTSUPP; 847 break; 848 849 case SO_NOFCS: 850 sock_valbool_flag(sk, SOCK_NOFCS, valbool); 851 break; 852 853 default: 854 ret = -ENOPROTOOPT; 855 break; 856 } 857 release_sock(sk); 858 return ret; 859} 860EXPORT_SYMBOL(sock_setsockopt); 861 862 863void cred_to_ucred(struct pid *pid, const struct cred *cred, 864 struct ucred *ucred) 865{ 866 ucred->pid = pid_vnr(pid); 867 ucred->uid = ucred->gid = -1; 868 if (cred) { 869 struct user_namespace *current_ns = current_user_ns(); 870 871 ucred->uid = from_kuid(current_ns, cred->euid); 872 ucred->gid = from_kgid(current_ns, cred->egid); 873 } 874} 875EXPORT_SYMBOL_GPL(cred_to_ucred); 876 877int sock_getsockopt(struct socket *sock, int level, int optname, 878 char __user *optval, int __user *optlen) 879{ 880 struct sock *sk = sock->sk; 881 882 union { 883 int val; 884 struct linger ling; 885 struct timeval tm; 886 } v; 887 888 int lv = sizeof(int); 889 int len; 890 891 if (get_user(len, optlen)) 892 return -EFAULT; 893 if (len < 0) 894 return -EINVAL; 895 896 memset(&v, 0, sizeof(v)); 897 898 switch (optname) { 899 case SO_DEBUG: 900 v.val = sock_flag(sk, SOCK_DBG); 901 break; 902 903 case SO_DONTROUTE: 904 v.val = sock_flag(sk, SOCK_LOCALROUTE); 905 break; 906 907 case SO_BROADCAST: 908 v.val = sock_flag(sk, SOCK_BROADCAST); 909 break; 910 911 case SO_SNDBUF: 912 v.val = sk->sk_sndbuf; 913 break; 914 915 case SO_RCVBUF: 916 v.val = sk->sk_rcvbuf; 917 break; 918 919 case SO_REUSEADDR: 920 v.val = sk->sk_reuse; 921 break; 922 923 case SO_KEEPALIVE: 924 v.val = sock_flag(sk, SOCK_KEEPOPEN); 925 break; 926 927 case SO_TYPE: 928 v.val = sk->sk_type; 929 break; 930 931 case SO_PROTOCOL: 932 v.val = sk->sk_protocol; 933 break; 934 935 case SO_DOMAIN: 936 v.val = sk->sk_family; 937 break; 938 939 case SO_ERROR: 940 v.val = -sock_error(sk); 941 if (v.val == 0) 942 v.val = xchg(&sk->sk_err_soft, 0); 943 break; 944 945 case SO_OOBINLINE: 946 v.val = sock_flag(sk, SOCK_URGINLINE); 947 break; 948 949 case SO_NO_CHECK: 950 v.val = sk->sk_no_check; 951 break; 952 953 case SO_PRIORITY: 954 v.val = sk->sk_priority; 955 break; 956 957 case SO_LINGER: 958 lv = sizeof(v.ling); 959 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER); 960 v.ling.l_linger = sk->sk_lingertime / HZ; 961 break; 962 963 case SO_BSDCOMPAT: 964 sock_warn_obsolete_bsdism("getsockopt"); 965 break; 966 967 case SO_TIMESTAMP: 968 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && 969 !sock_flag(sk, SOCK_RCVTSTAMPNS); 970 break; 971 972 case SO_TIMESTAMPNS: 973 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS); 974 break; 975 976 case SO_TIMESTAMPING: 977 v.val = 0; 978 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE)) 979 v.val |= SOF_TIMESTAMPING_TX_HARDWARE; 980 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE)) 981 v.val |= SOF_TIMESTAMPING_TX_SOFTWARE; 982 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE)) 983 v.val |= SOF_TIMESTAMPING_RX_HARDWARE; 984 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE)) 985 v.val |= SOF_TIMESTAMPING_RX_SOFTWARE; 986 if (sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE)) 987 v.val |= SOF_TIMESTAMPING_SOFTWARE; 988 if (sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE)) 989 v.val |= SOF_TIMESTAMPING_SYS_HARDWARE; 990 if (sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE)) 991 v.val |= SOF_TIMESTAMPING_RAW_HARDWARE; 992 break; 993 994 case SO_RCVTIMEO: 995 lv = sizeof(struct timeval); 996 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) { 997 v.tm.tv_sec = 0; 998 v.tm.tv_usec = 0; 999 } else { 1000 v.tm.tv_sec = sk->sk_rcvtimeo / HZ; 1001 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ; 1002 } 1003 break; 1004 1005 case SO_SNDTIMEO: 1006 lv = sizeof(struct timeval); 1007 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) { 1008 v.tm.tv_sec = 0; 1009 v.tm.tv_usec = 0; 1010 } else { 1011 v.tm.tv_sec = sk->sk_sndtimeo / HZ; 1012 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ; 1013 } 1014 break; 1015 1016 case SO_RCVLOWAT: 1017 v.val = sk->sk_rcvlowat; 1018 break; 1019 1020 case SO_SNDLOWAT: 1021 v.val = 1; 1022 break; 1023 1024 case SO_PASSCRED: 1025 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags); 1026 break; 1027 1028 case SO_PEERCRED: 1029 { 1030 struct ucred peercred; 1031 if (len > sizeof(peercred)) 1032 len = sizeof(peercred); 1033 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred); 1034 if (copy_to_user(optval, &peercred, len)) 1035 return -EFAULT; 1036 goto lenout; 1037 } 1038 1039 case SO_PEERNAME: 1040 { 1041 char address[128]; 1042 1043 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2)) 1044 return -ENOTCONN; 1045 if (lv < len) 1046 return -EINVAL; 1047 if (copy_to_user(optval, address, len)) 1048 return -EFAULT; 1049 goto lenout; 1050 } 1051 1052 /* Dubious BSD thing... Probably nobody even uses it, but 1053 * the UNIX standard wants it for whatever reason... -DaveM 1054 */ 1055 case SO_ACCEPTCONN: 1056 v.val = sk->sk_state == TCP_LISTEN; 1057 break; 1058 1059 case SO_PASSSEC: 1060 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags); 1061 break; 1062 1063 case SO_PEERSEC: 1064 return security_socket_getpeersec_stream(sock, optval, optlen, len); 1065 1066 case SO_MARK: 1067 v.val = sk->sk_mark; 1068 break; 1069 1070 case SO_RXQ_OVFL: 1071 v.val = sock_flag(sk, SOCK_RXQ_OVFL); 1072 break; 1073 1074 case SO_WIFI_STATUS: 1075 v.val = sock_flag(sk, SOCK_WIFI_STATUS); 1076 break; 1077 1078 case SO_PEEK_OFF: 1079 if (!sock->ops->set_peek_off) 1080 return -EOPNOTSUPP; 1081 1082 v.val = sk->sk_peek_off; 1083 break; 1084 case SO_NOFCS: 1085 v.val = sock_flag(sk, SOCK_NOFCS); 1086 break; 1087 default: 1088 return -ENOPROTOOPT; 1089 } 1090 1091 if (len > lv) 1092 len = lv; 1093 if (copy_to_user(optval, &v, len)) 1094 return -EFAULT; 1095lenout: 1096 if (put_user(len, optlen)) 1097 return -EFAULT; 1098 return 0; 1099} 1100 1101/* 1102 * Initialize an sk_lock. 1103 * 1104 * (We also register the sk_lock with the lock validator.) 1105 */ 1106static inline void sock_lock_init(struct sock *sk) 1107{ 1108 sock_lock_init_class_and_name(sk, 1109 af_family_slock_key_strings[sk->sk_family], 1110 af_family_slock_keys + sk->sk_family, 1111 af_family_key_strings[sk->sk_family], 1112 af_family_keys + sk->sk_family); 1113} 1114 1115/* 1116 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet, 1117 * even temporarly, because of RCU lookups. sk_node should also be left as is. 1118 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end 1119 */ 1120static void sock_copy(struct sock *nsk, const struct sock *osk) 1121{ 1122#ifdef CONFIG_SECURITY_NETWORK 1123 void *sptr = nsk->sk_security; 1124#endif 1125 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin)); 1126 1127 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end, 1128 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end)); 1129 1130#ifdef CONFIG_SECURITY_NETWORK 1131 nsk->sk_security = sptr; 1132 security_sk_clone(osk, nsk); 1133#endif 1134} 1135 1136/* 1137 * caches using SLAB_DESTROY_BY_RCU should let .next pointer from nulls nodes 1138 * un-modified. Special care is taken when initializing object to zero. 1139 */ 1140static inline void sk_prot_clear_nulls(struct sock *sk, int size) 1141{ 1142 if (offsetof(struct sock, sk_node.next) != 0) 1143 memset(sk, 0, offsetof(struct sock, sk_node.next)); 1144 memset(&sk->sk_node.pprev, 0, 1145 size - offsetof(struct sock, sk_node.pprev)); 1146} 1147 1148void sk_prot_clear_portaddr_nulls(struct sock *sk, int size) 1149{ 1150 unsigned long nulls1, nulls2; 1151 1152 nulls1 = offsetof(struct sock, __sk_common.skc_node.next); 1153 nulls2 = offsetof(struct sock, __sk_common.skc_portaddr_node.next); 1154 if (nulls1 > nulls2) 1155 swap(nulls1, nulls2); 1156 1157 if (nulls1 != 0) 1158 memset((char *)sk, 0, nulls1); 1159 memset((char *)sk + nulls1 + sizeof(void *), 0, 1160 nulls2 - nulls1 - sizeof(void *)); 1161 memset((char *)sk + nulls2 + sizeof(void *), 0, 1162 size - nulls2 - sizeof(void *)); 1163} 1164EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls); 1165 1166static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority, 1167 int family) 1168{ 1169 struct sock *sk; 1170 struct kmem_cache *slab; 1171 1172 slab = prot->slab; 1173 if (slab != NULL) { 1174 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO); 1175 if (!sk) 1176 return sk; 1177 if (priority & __GFP_ZERO) { 1178 if (prot->clear_sk) 1179 prot->clear_sk(sk, prot->obj_size); 1180 else 1181 sk_prot_clear_nulls(sk, prot->obj_size); 1182 } 1183 } else 1184 sk = kmalloc(prot->obj_size, priority); 1185 1186 if (sk != NULL) { 1187 kmemcheck_annotate_bitfield(sk, flags); 1188 1189 if (security_sk_alloc(sk, family, priority)) 1190 goto out_free; 1191 1192 if (!try_module_get(prot->owner)) 1193 goto out_free_sec; 1194 sk_tx_queue_clear(sk); 1195 } 1196 1197 return sk; 1198 1199out_free_sec: 1200 security_sk_free(sk); 1201out_free: 1202 if (slab != NULL) 1203 kmem_cache_free(slab, sk); 1204 else 1205 kfree(sk); 1206 return NULL; 1207} 1208 1209static void sk_prot_free(struct proto *prot, struct sock *sk) 1210{ 1211 struct kmem_cache *slab; 1212 struct module *owner; 1213 1214 owner = prot->owner; 1215 slab = prot->slab; 1216 1217 security_sk_free(sk); 1218 if (slab != NULL) 1219 kmem_cache_free(slab, sk); 1220 else 1221 kfree(sk); 1222 module_put(owner); 1223} 1224 1225#ifdef CONFIG_CGROUPS 1226#if IS_ENABLED(CONFIG_NET_CLS_CGROUP) 1227void sock_update_classid(struct sock *sk) 1228{ 1229 u32 classid; 1230 1231 rcu_read_lock(); /* doing current task, which cannot vanish. */ 1232 classid = task_cls_classid(current); 1233 rcu_read_unlock(); 1234 if (classid && classid != sk->sk_classid) 1235 sk->sk_classid = classid; 1236} 1237EXPORT_SYMBOL(sock_update_classid); 1238#endif 1239 1240#if IS_ENABLED(CONFIG_NETPRIO_CGROUP) 1241void sock_update_netprioidx(struct sock *sk, struct task_struct *task) 1242{ 1243 if (in_interrupt()) 1244 return; 1245 1246 sk->sk_cgrp_prioidx = task_netprioidx(task); 1247} 1248EXPORT_SYMBOL_GPL(sock_update_netprioidx); 1249#endif 1250#endif 1251 1252/** 1253 * sk_alloc - All socket objects are allocated here 1254 * @net: the applicable net namespace 1255 * @family: protocol family 1256 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 1257 * @prot: struct proto associated with this new sock instance 1258 */ 1259struct sock *sk_alloc(struct net *net, int family, gfp_t priority, 1260 struct proto *prot) 1261{ 1262 struct sock *sk; 1263 1264 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family); 1265 if (sk) { 1266 sk->sk_family = family; 1267 /* 1268 * See comment in struct sock definition to understand 1269 * why we need sk_prot_creator -acme 1270 */ 1271 sk->sk_prot = sk->sk_prot_creator = prot; 1272 sock_lock_init(sk); 1273 sock_net_set(sk, get_net(net)); 1274 atomic_set(&sk->sk_wmem_alloc, 1); 1275 1276 sock_update_classid(sk); 1277 sock_update_netprioidx(sk, current); 1278 } 1279 1280 return sk; 1281} 1282EXPORT_SYMBOL(sk_alloc); 1283 1284static void __sk_free(struct sock *sk) 1285{ 1286 struct sk_filter *filter; 1287 1288 if (sk->sk_destruct) 1289 sk->sk_destruct(sk); 1290 1291 filter = rcu_dereference_check(sk->sk_filter, 1292 atomic_read(&sk->sk_wmem_alloc) == 0); 1293 if (filter) { 1294 sk_filter_uncharge(sk, filter); 1295 RCU_INIT_POINTER(sk->sk_filter, NULL); 1296 } 1297 1298 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP); 1299 1300 if (atomic_read(&sk->sk_omem_alloc)) 1301 pr_debug("%s: optmem leakage (%d bytes) detected\n", 1302 __func__, atomic_read(&sk->sk_omem_alloc)); 1303 1304 if (sk->sk_peer_cred) 1305 put_cred(sk->sk_peer_cred); 1306 put_pid(sk->sk_peer_pid); 1307 put_net(sock_net(sk)); 1308 sk_prot_free(sk->sk_prot_creator, sk); 1309} 1310 1311void sk_free(struct sock *sk) 1312{ 1313 /* 1314 * We subtract one from sk_wmem_alloc and can know if 1315 * some packets are still in some tx queue. 1316 * If not null, sock_wfree() will call __sk_free(sk) later 1317 */ 1318 if (atomic_dec_and_test(&sk->sk_wmem_alloc)) 1319 __sk_free(sk); 1320} 1321EXPORT_SYMBOL(sk_free); 1322 1323/* 1324 * Last sock_put should drop reference to sk->sk_net. It has already 1325 * been dropped in sk_change_net. Taking reference to stopping namespace 1326 * is not an option. 1327 * Take reference to a socket to remove it from hash _alive_ and after that 1328 * destroy it in the context of init_net. 1329 */ 1330void sk_release_kernel(struct sock *sk) 1331{ 1332 if (sk == NULL || sk->sk_socket == NULL) 1333 return; 1334 1335 sock_hold(sk); 1336 sock_release(sk->sk_socket); 1337 release_net(sock_net(sk)); 1338 sock_net_set(sk, get_net(&init_net)); 1339 sock_put(sk); 1340} 1341EXPORT_SYMBOL(sk_release_kernel); 1342 1343static void sk_update_clone(const struct sock *sk, struct sock *newsk) 1344{ 1345 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) 1346 sock_update_memcg(newsk); 1347} 1348 1349/** 1350 * sk_clone_lock - clone a socket, and lock its clone 1351 * @sk: the socket to clone 1352 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 1353 * 1354 * Caller must unlock socket even in error path (bh_unlock_sock(newsk)) 1355 */ 1356struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority) 1357{ 1358 struct sock *newsk; 1359 1360 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family); 1361 if (newsk != NULL) { 1362 struct sk_filter *filter; 1363 1364 sock_copy(newsk, sk); 1365 1366 /* SANITY */ 1367 get_net(sock_net(newsk)); 1368 sk_node_init(&newsk->sk_node); 1369 sock_lock_init(newsk); 1370 bh_lock_sock(newsk); 1371 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL; 1372 newsk->sk_backlog.len = 0; 1373 1374 atomic_set(&newsk->sk_rmem_alloc, 0); 1375 /* 1376 * sk_wmem_alloc set to one (see sk_free() and sock_wfree()) 1377 */ 1378 atomic_set(&newsk->sk_wmem_alloc, 1); 1379 atomic_set(&newsk->sk_omem_alloc, 0); 1380 skb_queue_head_init(&newsk->sk_receive_queue); 1381 skb_queue_head_init(&newsk->sk_write_queue); 1382#ifdef CONFIG_NET_DMA 1383 skb_queue_head_init(&newsk->sk_async_wait_queue); 1384#endif 1385 1386 spin_lock_init(&newsk->sk_dst_lock); 1387 rwlock_init(&newsk->sk_callback_lock); 1388 lockdep_set_class_and_name(&newsk->sk_callback_lock, 1389 af_callback_keys + newsk->sk_family, 1390 af_family_clock_key_strings[newsk->sk_family]); 1391 1392 newsk->sk_dst_cache = NULL; 1393 newsk->sk_wmem_queued = 0; 1394 newsk->sk_forward_alloc = 0; 1395 newsk->sk_send_head = NULL; 1396 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK; 1397 1398 sock_reset_flag(newsk, SOCK_DONE); 1399 skb_queue_head_init(&newsk->sk_error_queue); 1400 1401 filter = rcu_dereference_protected(newsk->sk_filter, 1); 1402 if (filter != NULL) 1403 sk_filter_charge(newsk, filter); 1404 1405 if (unlikely(xfrm_sk_clone_policy(newsk))) { 1406 /* It is still raw copy of parent, so invalidate 1407 * destructor and make plain sk_free() */ 1408 newsk->sk_destruct = NULL; 1409 bh_unlock_sock(newsk); 1410 sk_free(newsk); 1411 newsk = NULL; 1412 goto out; 1413 } 1414 1415 newsk->sk_err = 0; 1416 newsk->sk_priority = 0; 1417 /* 1418 * Before updating sk_refcnt, we must commit prior changes to memory 1419 * (Documentation/RCU/rculist_nulls.txt for details) 1420 */ 1421 smp_wmb(); 1422 atomic_set(&newsk->sk_refcnt, 2); 1423 1424 /* 1425 * Increment the counter in the same struct proto as the master 1426 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that 1427 * is the same as sk->sk_prot->socks, as this field was copied 1428 * with memcpy). 1429 * 1430 * This _changes_ the previous behaviour, where 1431 * tcp_create_openreq_child always was incrementing the 1432 * equivalent to tcp_prot->socks (inet_sock_nr), so this have 1433 * to be taken into account in all callers. -acme 1434 */ 1435 sk_refcnt_debug_inc(newsk); 1436 sk_set_socket(newsk, NULL); 1437 newsk->sk_wq = NULL; 1438 1439 sk_update_clone(sk, newsk); 1440 1441 if (newsk->sk_prot->sockets_allocated) 1442 sk_sockets_allocated_inc(newsk); 1443 1444 if (newsk->sk_flags & SK_FLAGS_TIMESTAMP) 1445 net_enable_timestamp(); 1446 } 1447out: 1448 return newsk; 1449} 1450EXPORT_SYMBOL_GPL(sk_clone_lock); 1451 1452void sk_setup_caps(struct sock *sk, struct dst_entry *dst) 1453{ 1454 __sk_dst_set(sk, dst); 1455 sk->sk_route_caps = dst->dev->features; 1456 if (sk->sk_route_caps & NETIF_F_GSO) 1457 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE; 1458 sk->sk_route_caps &= ~sk->sk_route_nocaps; 1459 if (sk_can_gso(sk)) { 1460 if (dst->header_len) { 1461 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 1462 } else { 1463 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM; 1464 sk->sk_gso_max_size = dst->dev->gso_max_size; 1465 sk->sk_gso_max_segs = dst->dev->gso_max_segs; 1466 } 1467 } 1468} 1469EXPORT_SYMBOL_GPL(sk_setup_caps); 1470 1471void __init sk_init(void) 1472{ 1473 if (totalram_pages <= 4096) { 1474 sysctl_wmem_max = 32767; 1475 sysctl_rmem_max = 32767; 1476 sysctl_wmem_default = 32767; 1477 sysctl_rmem_default = 32767; 1478 } else if (totalram_pages >= 131072) { 1479 sysctl_wmem_max = 131071; 1480 sysctl_rmem_max = 131071; 1481 } 1482} 1483 1484/* 1485 * Simple resource managers for sockets. 1486 */ 1487 1488 1489/* 1490 * Write buffer destructor automatically called from kfree_skb. 1491 */ 1492void sock_wfree(struct sk_buff *skb) 1493{ 1494 struct sock *sk = skb->sk; 1495 unsigned int len = skb->truesize; 1496 1497 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) { 1498 /* 1499 * Keep a reference on sk_wmem_alloc, this will be released 1500 * after sk_write_space() call 1501 */ 1502 atomic_sub(len - 1, &sk->sk_wmem_alloc); 1503 sk->sk_write_space(sk); 1504 len = 1; 1505 } 1506 /* 1507 * if sk_wmem_alloc reaches 0, we must finish what sk_free() 1508 * could not do because of in-flight packets 1509 */ 1510 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc)) 1511 __sk_free(sk); 1512} 1513EXPORT_SYMBOL(sock_wfree); 1514 1515/* 1516 * Read buffer destructor automatically called from kfree_skb. 1517 */ 1518void sock_rfree(struct sk_buff *skb) 1519{ 1520 struct sock *sk = skb->sk; 1521 unsigned int len = skb->truesize; 1522 1523 atomic_sub(len, &sk->sk_rmem_alloc); 1524 sk_mem_uncharge(sk, len); 1525} 1526EXPORT_SYMBOL(sock_rfree); 1527 1528void sock_edemux(struct sk_buff *skb) 1529{ 1530 sock_put(skb->sk); 1531} 1532EXPORT_SYMBOL(sock_edemux); 1533 1534int sock_i_uid(struct sock *sk) 1535{ 1536 int uid; 1537 1538 read_lock_bh(&sk->sk_callback_lock); 1539 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0; 1540 read_unlock_bh(&sk->sk_callback_lock); 1541 return uid; 1542} 1543EXPORT_SYMBOL(sock_i_uid); 1544 1545unsigned long sock_i_ino(struct sock *sk) 1546{ 1547 unsigned long ino; 1548 1549 read_lock_bh(&sk->sk_callback_lock); 1550 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0; 1551 read_unlock_bh(&sk->sk_callback_lock); 1552 return ino; 1553} 1554EXPORT_SYMBOL(sock_i_ino); 1555 1556/* 1557 * Allocate a skb from the socket's send buffer. 1558 */ 1559struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 1560 gfp_t priority) 1561{ 1562 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) { 1563 struct sk_buff *skb = alloc_skb(size, priority); 1564 if (skb) { 1565 skb_set_owner_w(skb, sk); 1566 return skb; 1567 } 1568 } 1569 return NULL; 1570} 1571EXPORT_SYMBOL(sock_wmalloc); 1572 1573/* 1574 * Allocate a skb from the socket's receive buffer. 1575 */ 1576struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force, 1577 gfp_t priority) 1578{ 1579 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) { 1580 struct sk_buff *skb = alloc_skb(size, priority); 1581 if (skb) { 1582 skb_set_owner_r(skb, sk); 1583 return skb; 1584 } 1585 } 1586 return NULL; 1587} 1588 1589/* 1590 * Allocate a memory block from the socket's option memory buffer. 1591 */ 1592void *sock_kmalloc(struct sock *sk, int size, gfp_t priority) 1593{ 1594 if ((unsigned int)size <= sysctl_optmem_max && 1595 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) { 1596 void *mem; 1597 /* First do the add, to avoid the race if kmalloc 1598 * might sleep. 1599 */ 1600 atomic_add(size, &sk->sk_omem_alloc); 1601 mem = kmalloc(size, priority); 1602 if (mem) 1603 return mem; 1604 atomic_sub(size, &sk->sk_omem_alloc); 1605 } 1606 return NULL; 1607} 1608EXPORT_SYMBOL(sock_kmalloc); 1609 1610/* 1611 * Free an option memory block. 1612 */ 1613void sock_kfree_s(struct sock *sk, void *mem, int size) 1614{ 1615 kfree(mem); 1616 atomic_sub(size, &sk->sk_omem_alloc); 1617} 1618EXPORT_SYMBOL(sock_kfree_s); 1619 1620/* It is almost wait_for_tcp_memory minus release_sock/lock_sock. 1621 I think, these locks should be removed for datagram sockets. 1622 */ 1623static long sock_wait_for_wmem(struct sock *sk, long timeo) 1624{ 1625 DEFINE_WAIT(wait); 1626 1627 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags); 1628 for (;;) { 1629 if (!timeo) 1630 break; 1631 if (signal_pending(current)) 1632 break; 1633 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 1634 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 1635 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) 1636 break; 1637 if (sk->sk_shutdown & SEND_SHUTDOWN) 1638 break; 1639 if (sk->sk_err) 1640 break; 1641 timeo = schedule_timeout(timeo); 1642 } 1643 finish_wait(sk_sleep(sk), &wait); 1644 return timeo; 1645} 1646 1647 1648/* 1649 * Generic send/receive buffer handlers 1650 */ 1651 1652struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 1653 unsigned long data_len, int noblock, 1654 int *errcode) 1655{ 1656 struct sk_buff *skb; 1657 gfp_t gfp_mask; 1658 long timeo; 1659 int err; 1660 int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT; 1661 1662 err = -EMSGSIZE; 1663 if (npages > MAX_SKB_FRAGS) 1664 goto failure; 1665 1666 gfp_mask = sk->sk_allocation; 1667 if (gfp_mask & __GFP_WAIT) 1668 gfp_mask |= __GFP_REPEAT; 1669 1670 timeo = sock_sndtimeo(sk, noblock); 1671 while (1) { 1672 err = sock_error(sk); 1673 if (err != 0) 1674 goto failure; 1675 1676 err = -EPIPE; 1677 if (sk->sk_shutdown & SEND_SHUTDOWN) 1678 goto failure; 1679 1680 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) { 1681 skb = alloc_skb(header_len, gfp_mask); 1682 if (skb) { 1683 int i; 1684 1685 /* No pages, we're done... */ 1686 if (!data_len) 1687 break; 1688 1689 skb->truesize += data_len; 1690 skb_shinfo(skb)->nr_frags = npages; 1691 for (i = 0; i < npages; i++) { 1692 struct page *page; 1693 1694 page = alloc_pages(sk->sk_allocation, 0); 1695 if (!page) { 1696 err = -ENOBUFS; 1697 skb_shinfo(skb)->nr_frags = i; 1698 kfree_skb(skb); 1699 goto failure; 1700 } 1701 1702 __skb_fill_page_desc(skb, i, 1703 page, 0, 1704 (data_len >= PAGE_SIZE ? 1705 PAGE_SIZE : 1706 data_len)); 1707 data_len -= PAGE_SIZE; 1708 } 1709 1710 /* Full success... */ 1711 break; 1712 } 1713 err = -ENOBUFS; 1714 goto failure; 1715 } 1716 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags); 1717 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 1718 err = -EAGAIN; 1719 if (!timeo) 1720 goto failure; 1721 if (signal_pending(current)) 1722 goto interrupted; 1723 timeo = sock_wait_for_wmem(sk, timeo); 1724 } 1725 1726 skb_set_owner_w(skb, sk); 1727 return skb; 1728 1729interrupted: 1730 err = sock_intr_errno(timeo); 1731failure: 1732 *errcode = err; 1733 return NULL; 1734} 1735EXPORT_SYMBOL(sock_alloc_send_pskb); 1736 1737struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size, 1738 int noblock, int *errcode) 1739{ 1740 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode); 1741} 1742EXPORT_SYMBOL(sock_alloc_send_skb); 1743 1744static void __lock_sock(struct sock *sk) 1745 __releases(&sk->sk_lock.slock) 1746 __acquires(&sk->sk_lock.slock) 1747{ 1748 DEFINE_WAIT(wait); 1749 1750 for (;;) { 1751 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait, 1752 TASK_UNINTERRUPTIBLE); 1753 spin_unlock_bh(&sk->sk_lock.slock); 1754 schedule(); 1755 spin_lock_bh(&sk->sk_lock.slock); 1756 if (!sock_owned_by_user(sk)) 1757 break; 1758 } 1759 finish_wait(&sk->sk_lock.wq, &wait); 1760} 1761 1762static void __release_sock(struct sock *sk) 1763 __releases(&sk->sk_lock.slock) 1764 __acquires(&sk->sk_lock.slock) 1765{ 1766 struct sk_buff *skb = sk->sk_backlog.head; 1767 1768 do { 1769 sk->sk_backlog.head = sk->sk_backlog.tail = NULL; 1770 bh_unlock_sock(sk); 1771 1772 do { 1773 struct sk_buff *next = skb->next; 1774 1775 prefetch(next); 1776 WARN_ON_ONCE(skb_dst_is_noref(skb)); 1777 skb->next = NULL; 1778 sk_backlog_rcv(sk, skb); 1779 1780 /* 1781 * We are in process context here with softirqs 1782 * disabled, use cond_resched_softirq() to preempt. 1783 * This is safe to do because we've taken the backlog 1784 * queue private: 1785 */ 1786 cond_resched_softirq(); 1787 1788 skb = next; 1789 } while (skb != NULL); 1790 1791 bh_lock_sock(sk); 1792 } while ((skb = sk->sk_backlog.head) != NULL); 1793 1794 /* 1795 * Doing the zeroing here guarantee we can not loop forever 1796 * while a wild producer attempts to flood us. 1797 */ 1798 sk->sk_backlog.len = 0; 1799} 1800 1801/** 1802 * sk_wait_data - wait for data to arrive at sk_receive_queue 1803 * @sk: sock to wait on 1804 * @timeo: for how long 1805 * 1806 * Now socket state including sk->sk_err is changed only under lock, 1807 * hence we may omit checks after joining wait queue. 1808 * We check receive queue before schedule() only as optimization; 1809 * it is very likely that release_sock() added new data. 1810 */ 1811int sk_wait_data(struct sock *sk, long *timeo) 1812{ 1813 int rc; 1814 DEFINE_WAIT(wait); 1815 1816 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 1817 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags); 1818 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue)); 1819 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags); 1820 finish_wait(sk_sleep(sk), &wait); 1821 return rc; 1822} 1823EXPORT_SYMBOL(sk_wait_data); 1824 1825/** 1826 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated 1827 * @sk: socket 1828 * @size: memory size to allocate 1829 * @kind: allocation type 1830 * 1831 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means 1832 * rmem allocation. This function assumes that protocols which have 1833 * memory_pressure use sk_wmem_queued as write buffer accounting. 1834 */ 1835int __sk_mem_schedule(struct sock *sk, int size, int kind) 1836{ 1837 struct proto *prot = sk->sk_prot; 1838 int amt = sk_mem_pages(size); 1839 long allocated; 1840 int parent_status = UNDER_LIMIT; 1841 1842 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM; 1843 1844 allocated = sk_memory_allocated_add(sk, amt, &parent_status); 1845 1846 /* Under limit. */ 1847 if (parent_status == UNDER_LIMIT && 1848 allocated <= sk_prot_mem_limits(sk, 0)) { 1849 sk_leave_memory_pressure(sk); 1850 return 1; 1851 } 1852 1853 /* Under pressure. (we or our parents) */ 1854 if ((parent_status > SOFT_LIMIT) || 1855 allocated > sk_prot_mem_limits(sk, 1)) 1856 sk_enter_memory_pressure(sk); 1857 1858 /* Over hard limit (we or our parents) */ 1859 if ((parent_status == OVER_LIMIT) || 1860 (allocated > sk_prot_mem_limits(sk, 2))) 1861 goto suppress_allocation; 1862 1863 /* guarantee minimum buffer size under pressure */ 1864 if (kind == SK_MEM_RECV) { 1865 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0]) 1866 return 1; 1867 1868 } else { /* SK_MEM_SEND */ 1869 if (sk->sk_type == SOCK_STREAM) { 1870 if (sk->sk_wmem_queued < prot->sysctl_wmem[0]) 1871 return 1; 1872 } else if (atomic_read(&sk->sk_wmem_alloc) < 1873 prot->sysctl_wmem[0]) 1874 return 1; 1875 } 1876 1877 if (sk_has_memory_pressure(sk)) { 1878 int alloc; 1879 1880 if (!sk_under_memory_pressure(sk)) 1881 return 1; 1882 alloc = sk_sockets_allocated_read_positive(sk); 1883 if (sk_prot_mem_limits(sk, 2) > alloc * 1884 sk_mem_pages(sk->sk_wmem_queued + 1885 atomic_read(&sk->sk_rmem_alloc) + 1886 sk->sk_forward_alloc)) 1887 return 1; 1888 } 1889 1890suppress_allocation: 1891 1892 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) { 1893 sk_stream_moderate_sndbuf(sk); 1894 1895 /* Fail only if socket is _under_ its sndbuf. 1896 * In this case we cannot block, so that we have to fail. 1897 */ 1898 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf) 1899 return 1; 1900 } 1901 1902 trace_sock_exceed_buf_limit(sk, prot, allocated); 1903 1904 /* Alas. Undo changes. */ 1905 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM; 1906 1907 sk_memory_allocated_sub(sk, amt); 1908 1909 return 0; 1910} 1911EXPORT_SYMBOL(__sk_mem_schedule); 1912 1913/** 1914 * __sk_reclaim - reclaim memory_allocated 1915 * @sk: socket 1916 */ 1917void __sk_mem_reclaim(struct sock *sk) 1918{ 1919 sk_memory_allocated_sub(sk, 1920 sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT); 1921 sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1; 1922 1923 if (sk_under_memory_pressure(sk) && 1924 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0))) 1925 sk_leave_memory_pressure(sk); 1926} 1927EXPORT_SYMBOL(__sk_mem_reclaim); 1928 1929 1930/* 1931 * Set of default routines for initialising struct proto_ops when 1932 * the protocol does not support a particular function. In certain 1933 * cases where it makes no sense for a protocol to have a "do nothing" 1934 * function, some default processing is provided. 1935 */ 1936 1937int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len) 1938{ 1939 return -EOPNOTSUPP; 1940} 1941EXPORT_SYMBOL(sock_no_bind); 1942 1943int sock_no_connect(struct socket *sock, struct sockaddr *saddr, 1944 int len, int flags) 1945{ 1946 return -EOPNOTSUPP; 1947} 1948EXPORT_SYMBOL(sock_no_connect); 1949 1950int sock_no_socketpair(struct socket *sock1, struct socket *sock2) 1951{ 1952 return -EOPNOTSUPP; 1953} 1954EXPORT_SYMBOL(sock_no_socketpair); 1955 1956int sock_no_accept(struct socket *sock, struct socket *newsock, int flags) 1957{ 1958 return -EOPNOTSUPP; 1959} 1960EXPORT_SYMBOL(sock_no_accept); 1961 1962int sock_no_getname(struct socket *sock, struct sockaddr *saddr, 1963 int *len, int peer) 1964{ 1965 return -EOPNOTSUPP; 1966} 1967EXPORT_SYMBOL(sock_no_getname); 1968 1969unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt) 1970{ 1971 return 0; 1972} 1973EXPORT_SYMBOL(sock_no_poll); 1974 1975int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 1976{ 1977 return -EOPNOTSUPP; 1978} 1979EXPORT_SYMBOL(sock_no_ioctl); 1980 1981int sock_no_listen(struct socket *sock, int backlog) 1982{ 1983 return -EOPNOTSUPP; 1984} 1985EXPORT_SYMBOL(sock_no_listen); 1986 1987int sock_no_shutdown(struct socket *sock, int how) 1988{ 1989 return -EOPNOTSUPP; 1990} 1991EXPORT_SYMBOL(sock_no_shutdown); 1992 1993int sock_no_setsockopt(struct socket *sock, int level, int optname, 1994 char __user *optval, unsigned int optlen) 1995{ 1996 return -EOPNOTSUPP; 1997} 1998EXPORT_SYMBOL(sock_no_setsockopt); 1999 2000int sock_no_getsockopt(struct socket *sock, int level, int optname, 2001 char __user *optval, int __user *optlen) 2002{ 2003 return -EOPNOTSUPP; 2004} 2005EXPORT_SYMBOL(sock_no_getsockopt); 2006 2007int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m, 2008 size_t len) 2009{ 2010 return -EOPNOTSUPP; 2011} 2012EXPORT_SYMBOL(sock_no_sendmsg); 2013 2014int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m, 2015 size_t len, int flags) 2016{ 2017 return -EOPNOTSUPP; 2018} 2019EXPORT_SYMBOL(sock_no_recvmsg); 2020 2021int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma) 2022{ 2023 /* Mirror missing mmap method error code */ 2024 return -ENODEV; 2025} 2026EXPORT_SYMBOL(sock_no_mmap); 2027 2028ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags) 2029{ 2030 ssize_t res; 2031 struct msghdr msg = {.msg_flags = flags}; 2032 struct kvec iov; 2033 char *kaddr = kmap(page); 2034 iov.iov_base = kaddr + offset; 2035 iov.iov_len = size; 2036 res = kernel_sendmsg(sock, &msg, &iov, 1, size); 2037 kunmap(page); 2038 return res; 2039} 2040EXPORT_SYMBOL(sock_no_sendpage); 2041 2042/* 2043 * Default Socket Callbacks 2044 */ 2045 2046static void sock_def_wakeup(struct sock *sk) 2047{ 2048 struct socket_wq *wq; 2049 2050 rcu_read_lock(); 2051 wq = rcu_dereference(sk->sk_wq); 2052 if (wq_has_sleeper(wq)) 2053 wake_up_interruptible_all(&wq->wait); 2054 rcu_read_unlock(); 2055} 2056 2057static void sock_def_error_report(struct sock *sk) 2058{ 2059 struct socket_wq *wq; 2060 2061 rcu_read_lock(); 2062 wq = rcu_dereference(sk->sk_wq); 2063 if (wq_has_sleeper(wq)) 2064 wake_up_interruptible_poll(&wq->wait, POLLERR); 2065 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR); 2066 rcu_read_unlock(); 2067} 2068 2069static void sock_def_readable(struct sock *sk, int len) 2070{ 2071 struct socket_wq *wq; 2072 2073 rcu_read_lock(); 2074 wq = rcu_dereference(sk->sk_wq); 2075 if (wq_has_sleeper(wq)) 2076 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI | 2077 POLLRDNORM | POLLRDBAND); 2078 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN); 2079 rcu_read_unlock(); 2080} 2081 2082static void sock_def_write_space(struct sock *sk) 2083{ 2084 struct socket_wq *wq; 2085 2086 rcu_read_lock(); 2087 2088 /* Do not wake up a writer until he can make "significant" 2089 * progress. --DaveM 2090 */ 2091 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) { 2092 wq = rcu_dereference(sk->sk_wq); 2093 if (wq_has_sleeper(wq)) 2094 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT | 2095 POLLWRNORM | POLLWRBAND); 2096 2097 /* Should agree with poll, otherwise some programs break */ 2098 if (sock_writeable(sk)) 2099 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT); 2100 } 2101 2102 rcu_read_unlock(); 2103} 2104 2105static void sock_def_destruct(struct sock *sk) 2106{ 2107 kfree(sk->sk_protinfo); 2108} 2109 2110void sk_send_sigurg(struct sock *sk) 2111{ 2112 if (sk->sk_socket && sk->sk_socket->file) 2113 if (send_sigurg(&sk->sk_socket->file->f_owner)) 2114 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI); 2115} 2116EXPORT_SYMBOL(sk_send_sigurg); 2117 2118void sk_reset_timer(struct sock *sk, struct timer_list* timer, 2119 unsigned long expires) 2120{ 2121 if (!mod_timer(timer, expires)) 2122 sock_hold(sk); 2123} 2124EXPORT_SYMBOL(sk_reset_timer); 2125 2126void sk_stop_timer(struct sock *sk, struct timer_list* timer) 2127{ 2128 if (timer_pending(timer) && del_timer(timer)) 2129 __sock_put(sk); 2130} 2131EXPORT_SYMBOL(sk_stop_timer); 2132 2133void sock_init_data(struct socket *sock, struct sock *sk) 2134{ 2135 skb_queue_head_init(&sk->sk_receive_queue); 2136 skb_queue_head_init(&sk->sk_write_queue); 2137 skb_queue_head_init(&sk->sk_error_queue); 2138#ifdef CONFIG_NET_DMA 2139 skb_queue_head_init(&sk->sk_async_wait_queue); 2140#endif 2141 2142 sk->sk_send_head = NULL; 2143 2144 init_timer(&sk->sk_timer); 2145 2146 sk->sk_allocation = GFP_KERNEL; 2147 sk->sk_rcvbuf = sysctl_rmem_default; 2148 sk->sk_sndbuf = sysctl_wmem_default; 2149 sk->sk_state = TCP_CLOSE; 2150 sk_set_socket(sk, sock); 2151 2152 sock_set_flag(sk, SOCK_ZAPPED); 2153 2154 if (sock) { 2155 sk->sk_type = sock->type; 2156 sk->sk_wq = sock->wq; 2157 sock->sk = sk; 2158 } else 2159 sk->sk_wq = NULL; 2160 2161 spin_lock_init(&sk->sk_dst_lock); 2162 rwlock_init(&sk->sk_callback_lock); 2163 lockdep_set_class_and_name(&sk->sk_callback_lock, 2164 af_callback_keys + sk->sk_family, 2165 af_family_clock_key_strings[sk->sk_family]); 2166 2167 sk->sk_state_change = sock_def_wakeup; 2168 sk->sk_data_ready = sock_def_readable; 2169 sk->sk_write_space = sock_def_write_space; 2170 sk->sk_error_report = sock_def_error_report; 2171 sk->sk_destruct = sock_def_destruct; 2172 2173 sk->sk_sndmsg_page = NULL; 2174 sk->sk_sndmsg_off = 0; 2175 sk->sk_peek_off = -1; 2176 2177 sk->sk_peer_pid = NULL; 2178 sk->sk_peer_cred = NULL; 2179 sk->sk_write_pending = 0; 2180 sk->sk_rcvlowat = 1; 2181 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; 2182 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT; 2183 2184 sk->sk_stamp = ktime_set(-1L, 0); 2185 2186 /* 2187 * Before updating sk_refcnt, we must commit prior changes to memory 2188 * (Documentation/RCU/rculist_nulls.txt for details) 2189 */ 2190 smp_wmb(); 2191 atomic_set(&sk->sk_refcnt, 1); 2192 atomic_set(&sk->sk_drops, 0); 2193} 2194EXPORT_SYMBOL(sock_init_data); 2195 2196void lock_sock_nested(struct sock *sk, int subclass) 2197{ 2198 might_sleep(); 2199 spin_lock_bh(&sk->sk_lock.slock); 2200 if (sk->sk_lock.owned) 2201 __lock_sock(sk); 2202 sk->sk_lock.owned = 1; 2203 spin_unlock(&sk->sk_lock.slock); 2204 /* 2205 * The sk_lock has mutex_lock() semantics here: 2206 */ 2207 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_); 2208 local_bh_enable(); 2209} 2210EXPORT_SYMBOL(lock_sock_nested); 2211 2212void release_sock(struct sock *sk) 2213{ 2214 /* 2215 * The sk_lock has mutex_unlock() semantics: 2216 */ 2217 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_); 2218 2219 spin_lock_bh(&sk->sk_lock.slock); 2220 if (sk->sk_backlog.tail) 2221 __release_sock(sk); 2222 2223 if (sk->sk_prot->release_cb) 2224 sk->sk_prot->release_cb(sk); 2225 2226 sk->sk_lock.owned = 0; 2227 if (waitqueue_active(&sk->sk_lock.wq)) 2228 wake_up(&sk->sk_lock.wq); 2229 spin_unlock_bh(&sk->sk_lock.slock); 2230} 2231EXPORT_SYMBOL(release_sock); 2232 2233/** 2234 * lock_sock_fast - fast version of lock_sock 2235 * @sk: socket 2236 * 2237 * This version should be used for very small section, where process wont block 2238 * return false if fast path is taken 2239 * sk_lock.slock locked, owned = 0, BH disabled 2240 * return true if slow path is taken 2241 * sk_lock.slock unlocked, owned = 1, BH enabled 2242 */ 2243bool lock_sock_fast(struct sock *sk) 2244{ 2245 might_sleep(); 2246 spin_lock_bh(&sk->sk_lock.slock); 2247 2248 if (!sk->sk_lock.owned) 2249 /* 2250 * Note : We must disable BH 2251 */ 2252 return false; 2253 2254 __lock_sock(sk); 2255 sk->sk_lock.owned = 1; 2256 spin_unlock(&sk->sk_lock.slock); 2257 /* 2258 * The sk_lock has mutex_lock() semantics here: 2259 */ 2260 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_); 2261 local_bh_enable(); 2262 return true; 2263} 2264EXPORT_SYMBOL(lock_sock_fast); 2265 2266int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp) 2267{ 2268 struct timeval tv; 2269 if (!sock_flag(sk, SOCK_TIMESTAMP)) 2270 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 2271 tv = ktime_to_timeval(sk->sk_stamp); 2272 if (tv.tv_sec == -1) 2273 return -ENOENT; 2274 if (tv.tv_sec == 0) { 2275 sk->sk_stamp = ktime_get_real(); 2276 tv = ktime_to_timeval(sk->sk_stamp); 2277 } 2278 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0; 2279} 2280EXPORT_SYMBOL(sock_get_timestamp); 2281 2282int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp) 2283{ 2284 struct timespec ts; 2285 if (!sock_flag(sk, SOCK_TIMESTAMP)) 2286 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 2287 ts = ktime_to_timespec(sk->sk_stamp); 2288 if (ts.tv_sec == -1) 2289 return -ENOENT; 2290 if (ts.tv_sec == 0) { 2291 sk->sk_stamp = ktime_get_real(); 2292 ts = ktime_to_timespec(sk->sk_stamp); 2293 } 2294 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0; 2295} 2296EXPORT_SYMBOL(sock_get_timestampns); 2297 2298void sock_enable_timestamp(struct sock *sk, int flag) 2299{ 2300 if (!sock_flag(sk, flag)) { 2301 unsigned long previous_flags = sk->sk_flags; 2302 2303 sock_set_flag(sk, flag); 2304 /* 2305 * we just set one of the two flags which require net 2306 * time stamping, but time stamping might have been on 2307 * already because of the other one 2308 */ 2309 if (!(previous_flags & SK_FLAGS_TIMESTAMP)) 2310 net_enable_timestamp(); 2311 } 2312} 2313 2314/* 2315 * Get a socket option on an socket. 2316 * 2317 * FIX: POSIX 1003.1g is very ambiguous here. It states that 2318 * asynchronous errors should be reported by getsockopt. We assume 2319 * this means if you specify SO_ERROR (otherwise whats the point of it). 2320 */ 2321int sock_common_getsockopt(struct socket *sock, int level, int optname, 2322 char __user *optval, int __user *optlen) 2323{ 2324 struct sock *sk = sock->sk; 2325 2326 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen); 2327} 2328EXPORT_SYMBOL(sock_common_getsockopt); 2329 2330#ifdef CONFIG_COMPAT 2331int compat_sock_common_getsockopt(struct socket *sock, int level, int optname, 2332 char __user *optval, int __user *optlen) 2333{ 2334 struct sock *sk = sock->sk; 2335 2336 if (sk->sk_prot->compat_getsockopt != NULL) 2337 return sk->sk_prot->compat_getsockopt(sk, level, optname, 2338 optval, optlen); 2339 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen); 2340} 2341EXPORT_SYMBOL(compat_sock_common_getsockopt); 2342#endif 2343 2344int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock, 2345 struct msghdr *msg, size_t size, int flags) 2346{ 2347 struct sock *sk = sock->sk; 2348 int addr_len = 0; 2349 int err; 2350 2351 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT, 2352 flags & ~MSG_DONTWAIT, &addr_len); 2353 if (err >= 0) 2354 msg->msg_namelen = addr_len; 2355 return err; 2356} 2357EXPORT_SYMBOL(sock_common_recvmsg); 2358 2359/* 2360 * Set socket options on an inet socket. 2361 */ 2362int sock_common_setsockopt(struct socket *sock, int level, int optname, 2363 char __user *optval, unsigned int optlen) 2364{ 2365 struct sock *sk = sock->sk; 2366 2367 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen); 2368} 2369EXPORT_SYMBOL(sock_common_setsockopt); 2370 2371#ifdef CONFIG_COMPAT 2372int compat_sock_common_setsockopt(struct socket *sock, int level, int optname, 2373 char __user *optval, unsigned int optlen) 2374{ 2375 struct sock *sk = sock->sk; 2376 2377 if (sk->sk_prot->compat_setsockopt != NULL) 2378 return sk->sk_prot->compat_setsockopt(sk, level, optname, 2379 optval, optlen); 2380 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen); 2381} 2382EXPORT_SYMBOL(compat_sock_common_setsockopt); 2383#endif 2384 2385void sk_common_release(struct sock *sk) 2386{ 2387 if (sk->sk_prot->destroy) 2388 sk->sk_prot->destroy(sk); 2389 2390 /* 2391 * Observation: when sock_common_release is called, processes have 2392 * no access to socket. But net still has. 2393 * Step one, detach it from networking: 2394 * 2395 * A. Remove from hash tables. 2396 */ 2397 2398 sk->sk_prot->unhash(sk); 2399 2400 /* 2401 * In this point socket cannot receive new packets, but it is possible 2402 * that some packets are in flight because some CPU runs receiver and 2403 * did hash table lookup before we unhashed socket. They will achieve 2404 * receive queue and will be purged by socket destructor. 2405 * 2406 * Also we still have packets pending on receive queue and probably, 2407 * our own packets waiting in device queues. sock_destroy will drain 2408 * receive queue, but transmitted packets will delay socket destruction 2409 * until the last reference will be released. 2410 */ 2411 2412 sock_orphan(sk); 2413 2414 xfrm_sk_free_policy(sk); 2415 2416 sk_refcnt_debug_release(sk); 2417 sock_put(sk); 2418} 2419EXPORT_SYMBOL(sk_common_release); 2420 2421#ifdef CONFIG_PROC_FS 2422#define PROTO_INUSE_NR 64 /* should be enough for the first time */ 2423struct prot_inuse { 2424 int val[PROTO_INUSE_NR]; 2425}; 2426 2427static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR); 2428 2429#ifdef CONFIG_NET_NS 2430void sock_prot_inuse_add(struct net *net, struct proto *prot, int val) 2431{ 2432 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val); 2433} 2434EXPORT_SYMBOL_GPL(sock_prot_inuse_add); 2435 2436int sock_prot_inuse_get(struct net *net, struct proto *prot) 2437{ 2438 int cpu, idx = prot->inuse_idx; 2439 int res = 0; 2440 2441 for_each_possible_cpu(cpu) 2442 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx]; 2443 2444 return res >= 0 ? res : 0; 2445} 2446EXPORT_SYMBOL_GPL(sock_prot_inuse_get); 2447 2448static int __net_init sock_inuse_init_net(struct net *net) 2449{ 2450 net->core.inuse = alloc_percpu(struct prot_inuse); 2451 return net->core.inuse ? 0 : -ENOMEM; 2452} 2453 2454static void __net_exit sock_inuse_exit_net(struct net *net) 2455{ 2456 free_percpu(net->core.inuse); 2457} 2458 2459static struct pernet_operations net_inuse_ops = { 2460 .init = sock_inuse_init_net, 2461 .exit = sock_inuse_exit_net, 2462}; 2463 2464static __init int net_inuse_init(void) 2465{ 2466 if (register_pernet_subsys(&net_inuse_ops)) 2467 panic("Cannot initialize net inuse counters"); 2468 2469 return 0; 2470} 2471 2472core_initcall(net_inuse_init); 2473#else 2474static DEFINE_PER_CPU(struct prot_inuse, prot_inuse); 2475 2476void sock_prot_inuse_add(struct net *net, struct proto *prot, int val) 2477{ 2478 __this_cpu_add(prot_inuse.val[prot->inuse_idx], val); 2479} 2480EXPORT_SYMBOL_GPL(sock_prot_inuse_add); 2481 2482int sock_prot_inuse_get(struct net *net, struct proto *prot) 2483{ 2484 int cpu, idx = prot->inuse_idx; 2485 int res = 0; 2486 2487 for_each_possible_cpu(cpu) 2488 res += per_cpu(prot_inuse, cpu).val[idx]; 2489 2490 return res >= 0 ? res : 0; 2491} 2492EXPORT_SYMBOL_GPL(sock_prot_inuse_get); 2493#endif 2494 2495static void assign_proto_idx(struct proto *prot) 2496{ 2497 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR); 2498 2499 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) { 2500 pr_err("PROTO_INUSE_NR exhausted\n"); 2501 return; 2502 } 2503 2504 set_bit(prot->inuse_idx, proto_inuse_idx); 2505} 2506 2507static void release_proto_idx(struct proto *prot) 2508{ 2509 if (prot->inuse_idx != PROTO_INUSE_NR - 1) 2510 clear_bit(prot->inuse_idx, proto_inuse_idx); 2511} 2512#else 2513static inline void assign_proto_idx(struct proto *prot) 2514{ 2515} 2516 2517static inline void release_proto_idx(struct proto *prot) 2518{ 2519} 2520#endif 2521 2522int proto_register(struct proto *prot, int alloc_slab) 2523{ 2524 if (alloc_slab) { 2525 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0, 2526 SLAB_HWCACHE_ALIGN | prot->slab_flags, 2527 NULL); 2528 2529 if (prot->slab == NULL) { 2530 pr_crit("%s: Can't create sock SLAB cache!\n", 2531 prot->name); 2532 goto out; 2533 } 2534 2535 if (prot->rsk_prot != NULL) { 2536 prot->rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", prot->name); 2537 if (prot->rsk_prot->slab_name == NULL) 2538 goto out_free_sock_slab; 2539 2540 prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name, 2541 prot->rsk_prot->obj_size, 0, 2542 SLAB_HWCACHE_ALIGN, NULL); 2543 2544 if (prot->rsk_prot->slab == NULL) { 2545 pr_crit("%s: Can't create request sock SLAB cache!\n", 2546 prot->name); 2547 goto out_free_request_sock_slab_name; 2548 } 2549 } 2550 2551 if (prot->twsk_prot != NULL) { 2552 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name); 2553 2554 if (prot->twsk_prot->twsk_slab_name == NULL) 2555 goto out_free_request_sock_slab; 2556 2557 prot->twsk_prot->twsk_slab = 2558 kmem_cache_create(prot->twsk_prot->twsk_slab_name, 2559 prot->twsk_prot->twsk_obj_size, 2560 0, 2561 SLAB_HWCACHE_ALIGN | 2562 prot->slab_flags, 2563 NULL); 2564 if (prot->twsk_prot->twsk_slab == NULL) 2565 goto out_free_timewait_sock_slab_name; 2566 } 2567 } 2568 2569 mutex_lock(&proto_list_mutex); 2570 list_add(&prot->node, &proto_list); 2571 assign_proto_idx(prot); 2572 mutex_unlock(&proto_list_mutex); 2573 return 0; 2574 2575out_free_timewait_sock_slab_name: 2576 kfree(prot->twsk_prot->twsk_slab_name); 2577out_free_request_sock_slab: 2578 if (prot->rsk_prot && prot->rsk_prot->slab) { 2579 kmem_cache_destroy(prot->rsk_prot->slab); 2580 prot->rsk_prot->slab = NULL; 2581 } 2582out_free_request_sock_slab_name: 2583 if (prot->rsk_prot) 2584 kfree(prot->rsk_prot->slab_name); 2585out_free_sock_slab: 2586 kmem_cache_destroy(prot->slab); 2587 prot->slab = NULL; 2588out: 2589 return -ENOBUFS; 2590} 2591EXPORT_SYMBOL(proto_register); 2592 2593void proto_unregister(struct proto *prot) 2594{ 2595 mutex_lock(&proto_list_mutex); 2596 release_proto_idx(prot); 2597 list_del(&prot->node); 2598 mutex_unlock(&proto_list_mutex); 2599 2600 if (prot->slab != NULL) { 2601 kmem_cache_destroy(prot->slab); 2602 prot->slab = NULL; 2603 } 2604 2605 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) { 2606 kmem_cache_destroy(prot->rsk_prot->slab); 2607 kfree(prot->rsk_prot->slab_name); 2608 prot->rsk_prot->slab = NULL; 2609 } 2610 2611 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) { 2612 kmem_cache_destroy(prot->twsk_prot->twsk_slab); 2613 kfree(prot->twsk_prot->twsk_slab_name); 2614 prot->twsk_prot->twsk_slab = NULL; 2615 } 2616} 2617EXPORT_SYMBOL(proto_unregister); 2618 2619#ifdef CONFIG_PROC_FS 2620static void *proto_seq_start(struct seq_file *seq, loff_t *pos) 2621 __acquires(proto_list_mutex) 2622{ 2623 mutex_lock(&proto_list_mutex); 2624 return seq_list_start_head(&proto_list, *pos); 2625} 2626 2627static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos) 2628{ 2629 return seq_list_next(v, &proto_list, pos); 2630} 2631 2632static void proto_seq_stop(struct seq_file *seq, void *v) 2633 __releases(proto_list_mutex) 2634{ 2635 mutex_unlock(&proto_list_mutex); 2636} 2637 2638static char proto_method_implemented(const void *method) 2639{ 2640 return method == NULL ? 'n' : 'y'; 2641} 2642static long sock_prot_memory_allocated(struct proto *proto) 2643{ 2644 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L; 2645} 2646 2647static char *sock_prot_memory_pressure(struct proto *proto) 2648{ 2649 return proto->memory_pressure != NULL ? 2650 proto_memory_pressure(proto) ? "yes" : "no" : "NI"; 2651} 2652 2653static void proto_seq_printf(struct seq_file *seq, struct proto *proto) 2654{ 2655 2656 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s " 2657 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n", 2658 proto->name, 2659 proto->obj_size, 2660 sock_prot_inuse_get(seq_file_net(seq), proto), 2661 sock_prot_memory_allocated(proto), 2662 sock_prot_memory_pressure(proto), 2663 proto->max_header, 2664 proto->slab == NULL ? "no" : "yes", 2665 module_name(proto->owner), 2666 proto_method_implemented(proto->close), 2667 proto_method_implemented(proto->connect), 2668 proto_method_implemented(proto->disconnect), 2669 proto_method_implemented(proto->accept), 2670 proto_method_implemented(proto->ioctl), 2671 proto_method_implemented(proto->init), 2672 proto_method_implemented(proto->destroy), 2673 proto_method_implemented(proto->shutdown), 2674 proto_method_implemented(proto->setsockopt), 2675 proto_method_implemented(proto->getsockopt), 2676 proto_method_implemented(proto->sendmsg), 2677 proto_method_implemented(proto->recvmsg), 2678 proto_method_implemented(proto->sendpage), 2679 proto_method_implemented(proto->bind), 2680 proto_method_implemented(proto->backlog_rcv), 2681 proto_method_implemented(proto->hash), 2682 proto_method_implemented(proto->unhash), 2683 proto_method_implemented(proto->get_port), 2684 proto_method_implemented(proto->enter_memory_pressure)); 2685} 2686 2687static int proto_seq_show(struct seq_file *seq, void *v) 2688{ 2689 if (v == &proto_list) 2690 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s", 2691 "protocol", 2692 "size", 2693 "sockets", 2694 "memory", 2695 "press", 2696 "maxhdr", 2697 "slab", 2698 "module", 2699 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n"); 2700 else 2701 proto_seq_printf(seq, list_entry(v, struct proto, node)); 2702 return 0; 2703} 2704 2705static const struct seq_operations proto_seq_ops = { 2706 .start = proto_seq_start, 2707 .next = proto_seq_next, 2708 .stop = proto_seq_stop, 2709 .show = proto_seq_show, 2710}; 2711 2712static int proto_seq_open(struct inode *inode, struct file *file) 2713{ 2714 return seq_open_net(inode, file, &proto_seq_ops, 2715 sizeof(struct seq_net_private)); 2716} 2717 2718static const struct file_operations proto_seq_fops = { 2719 .owner = THIS_MODULE, 2720 .open = proto_seq_open, 2721 .read = seq_read, 2722 .llseek = seq_lseek, 2723 .release = seq_release_net, 2724}; 2725 2726static __net_init int proto_init_net(struct net *net) 2727{ 2728 if (!proc_net_fops_create(net, "protocols", S_IRUGO, &proto_seq_fops)) 2729 return -ENOMEM; 2730 2731 return 0; 2732} 2733 2734static __net_exit void proto_exit_net(struct net *net) 2735{ 2736 proc_net_remove(net, "protocols"); 2737} 2738 2739 2740static __net_initdata struct pernet_operations proto_net_ops = { 2741 .init = proto_init_net, 2742 .exit = proto_exit_net, 2743}; 2744 2745static int __init proto_init(void) 2746{ 2747 return register_pernet_subsys(&proto_net_ops); 2748} 2749 2750subsys_initcall(proto_init); 2751 2752#endif /* PROC_FS */ 2753