sock.c revision f7b86bfe8d9f10e5a9fcacf52dddf29d0d58a33b
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 329static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen) 330{ 331 struct timeval tv; 332 333 if (optlen < sizeof(tv)) 334 return -EINVAL; 335 if (copy_from_user(&tv, optval, sizeof(tv))) 336 return -EFAULT; 337 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC) 338 return -EDOM; 339 340 if (tv.tv_sec < 0) { 341 static int warned __read_mostly; 342 343 *timeo_p = 0; 344 if (warned < 10 && net_ratelimit()) { 345 warned++; 346 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n", 347 __func__, current->comm, task_pid_nr(current)); 348 } 349 return 0; 350 } 351 *timeo_p = MAX_SCHEDULE_TIMEOUT; 352 if (tv.tv_sec == 0 && tv.tv_usec == 0) 353 return 0; 354 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1)) 355 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ); 356 return 0; 357} 358 359static void sock_warn_obsolete_bsdism(const char *name) 360{ 361 static int warned; 362 static char warncomm[TASK_COMM_LEN]; 363 if (strcmp(warncomm, current->comm) && warned < 5) { 364 strcpy(warncomm, current->comm); 365 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n", 366 warncomm, name); 367 warned++; 368 } 369} 370 371#define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE)) 372 373static void sock_disable_timestamp(struct sock *sk, unsigned long flags) 374{ 375 if (sk->sk_flags & flags) { 376 sk->sk_flags &= ~flags; 377 if (!(sk->sk_flags & SK_FLAGS_TIMESTAMP)) 378 net_disable_timestamp(); 379 } 380} 381 382 383int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 384{ 385 int err; 386 int skb_len; 387 unsigned long flags; 388 struct sk_buff_head *list = &sk->sk_receive_queue; 389 390 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) { 391 atomic_inc(&sk->sk_drops); 392 trace_sock_rcvqueue_full(sk, skb); 393 return -ENOMEM; 394 } 395 396 err = sk_filter(sk, skb); 397 if (err) 398 return err; 399 400 if (!sk_rmem_schedule(sk, skb, skb->truesize)) { 401 atomic_inc(&sk->sk_drops); 402 return -ENOBUFS; 403 } 404 405 skb->dev = NULL; 406 skb_set_owner_r(skb, sk); 407 408 /* Cache the SKB length before we tack it onto the receive 409 * queue. Once it is added it no longer belongs to us and 410 * may be freed by other threads of control pulling packets 411 * from the queue. 412 */ 413 skb_len = skb->len; 414 415 /* we escape from rcu protected region, make sure we dont leak 416 * a norefcounted dst 417 */ 418 skb_dst_force(skb); 419 420 spin_lock_irqsave(&list->lock, flags); 421 skb->dropcount = atomic_read(&sk->sk_drops); 422 __skb_queue_tail(list, skb); 423 spin_unlock_irqrestore(&list->lock, flags); 424 425 if (!sock_flag(sk, SOCK_DEAD)) 426 sk->sk_data_ready(sk, skb_len); 427 return 0; 428} 429EXPORT_SYMBOL(sock_queue_rcv_skb); 430 431int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested) 432{ 433 int rc = NET_RX_SUCCESS; 434 435 if (sk_filter(sk, skb)) 436 goto discard_and_relse; 437 438 skb->dev = NULL; 439 440 if (sk_rcvqueues_full(sk, skb, sk->sk_rcvbuf)) { 441 atomic_inc(&sk->sk_drops); 442 goto discard_and_relse; 443 } 444 if (nested) 445 bh_lock_sock_nested(sk); 446 else 447 bh_lock_sock(sk); 448 if (!sock_owned_by_user(sk)) { 449 /* 450 * trylock + unlock semantics: 451 */ 452 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_); 453 454 rc = sk_backlog_rcv(sk, skb); 455 456 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_); 457 } else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) { 458 bh_unlock_sock(sk); 459 atomic_inc(&sk->sk_drops); 460 goto discard_and_relse; 461 } 462 463 bh_unlock_sock(sk); 464out: 465 sock_put(sk); 466 return rc; 467discard_and_relse: 468 kfree_skb(skb); 469 goto out; 470} 471EXPORT_SYMBOL(sk_receive_skb); 472 473void sk_reset_txq(struct sock *sk) 474{ 475 sk_tx_queue_clear(sk); 476} 477EXPORT_SYMBOL(sk_reset_txq); 478 479struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie) 480{ 481 struct dst_entry *dst = __sk_dst_get(sk); 482 483 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) { 484 sk_tx_queue_clear(sk); 485 RCU_INIT_POINTER(sk->sk_dst_cache, NULL); 486 dst_release(dst); 487 return NULL; 488 } 489 490 return dst; 491} 492EXPORT_SYMBOL(__sk_dst_check); 493 494struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie) 495{ 496 struct dst_entry *dst = sk_dst_get(sk); 497 498 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) { 499 sk_dst_reset(sk); 500 dst_release(dst); 501 return NULL; 502 } 503 504 return dst; 505} 506EXPORT_SYMBOL(sk_dst_check); 507 508static int sock_bindtodevice(struct sock *sk, char __user *optval, int optlen) 509{ 510 int ret = -ENOPROTOOPT; 511#ifdef CONFIG_NETDEVICES 512 struct net *net = sock_net(sk); 513 char devname[IFNAMSIZ]; 514 int index; 515 516 /* Sorry... */ 517 ret = -EPERM; 518 if (!capable(CAP_NET_RAW)) 519 goto out; 520 521 ret = -EINVAL; 522 if (optlen < 0) 523 goto out; 524 525 /* Bind this socket to a particular device like "eth0", 526 * as specified in the passed interface name. If the 527 * name is "" or the option length is zero the socket 528 * is not bound. 529 */ 530 if (optlen > IFNAMSIZ - 1) 531 optlen = IFNAMSIZ - 1; 532 memset(devname, 0, sizeof(devname)); 533 534 ret = -EFAULT; 535 if (copy_from_user(devname, optval, optlen)) 536 goto out; 537 538 index = 0; 539 if (devname[0] != '\0') { 540 struct net_device *dev; 541 542 rcu_read_lock(); 543 dev = dev_get_by_name_rcu(net, devname); 544 if (dev) 545 index = dev->ifindex; 546 rcu_read_unlock(); 547 ret = -ENODEV; 548 if (!dev) 549 goto out; 550 } 551 552 lock_sock(sk); 553 sk->sk_bound_dev_if = index; 554 sk_dst_reset(sk); 555 release_sock(sk); 556 557 ret = 0; 558 559out: 560#endif 561 562 return ret; 563} 564 565static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool) 566{ 567 if (valbool) 568 sock_set_flag(sk, bit); 569 else 570 sock_reset_flag(sk, bit); 571} 572 573/* 574 * This is meant for all protocols to use and covers goings on 575 * at the socket level. Everything here is generic. 576 */ 577 578int sock_setsockopt(struct socket *sock, int level, int optname, 579 char __user *optval, unsigned int optlen) 580{ 581 struct sock *sk = sock->sk; 582 int val; 583 int valbool; 584 struct linger ling; 585 int ret = 0; 586 587 /* 588 * Options without arguments 589 */ 590 591 if (optname == SO_BINDTODEVICE) 592 return sock_bindtodevice(sk, optval, optlen); 593 594 if (optlen < sizeof(int)) 595 return -EINVAL; 596 597 if (get_user(val, (int __user *)optval)) 598 return -EFAULT; 599 600 valbool = val ? 1 : 0; 601 602 lock_sock(sk); 603 604 switch (optname) { 605 case SO_DEBUG: 606 if (val && !capable(CAP_NET_ADMIN)) 607 ret = -EACCES; 608 else 609 sock_valbool_flag(sk, SOCK_DBG, valbool); 610 break; 611 case SO_REUSEADDR: 612 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE); 613 break; 614 case SO_TYPE: 615 case SO_PROTOCOL: 616 case SO_DOMAIN: 617 case SO_ERROR: 618 ret = -ENOPROTOOPT; 619 break; 620 case SO_DONTROUTE: 621 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool); 622 break; 623 case SO_BROADCAST: 624 sock_valbool_flag(sk, SOCK_BROADCAST, valbool); 625 break; 626 case SO_SNDBUF: 627 /* Don't error on this BSD doesn't and if you think 628 * about it this is right. Otherwise apps have to 629 * play 'guess the biggest size' games. RCVBUF/SNDBUF 630 * are treated in BSD as hints 631 */ 632 val = min_t(u32, val, sysctl_wmem_max); 633set_sndbuf: 634 sk->sk_userlocks |= SOCK_SNDBUF_LOCK; 635 sk->sk_sndbuf = max_t(u32, val * 2, SOCK_MIN_SNDBUF); 636 /* Wake up sending tasks if we upped the value. */ 637 sk->sk_write_space(sk); 638 break; 639 640 case SO_SNDBUFFORCE: 641 if (!capable(CAP_NET_ADMIN)) { 642 ret = -EPERM; 643 break; 644 } 645 goto set_sndbuf; 646 647 case SO_RCVBUF: 648 /* Don't error on this BSD doesn't and if you think 649 * about it this is right. Otherwise apps have to 650 * play 'guess the biggest size' games. RCVBUF/SNDBUF 651 * are treated in BSD as hints 652 */ 653 val = min_t(u32, val, sysctl_rmem_max); 654set_rcvbuf: 655 sk->sk_userlocks |= SOCK_RCVBUF_LOCK; 656 /* 657 * We double it on the way in to account for 658 * "struct sk_buff" etc. overhead. Applications 659 * assume that the SO_RCVBUF setting they make will 660 * allow that much actual data to be received on that 661 * socket. 662 * 663 * Applications are unaware that "struct sk_buff" and 664 * other overheads allocate from the receive buffer 665 * during socket buffer allocation. 666 * 667 * And after considering the possible alternatives, 668 * returning the value we actually used in getsockopt 669 * is the most desirable behavior. 670 */ 671 sk->sk_rcvbuf = max_t(u32, val * 2, SOCK_MIN_RCVBUF); 672 break; 673 674 case SO_RCVBUFFORCE: 675 if (!capable(CAP_NET_ADMIN)) { 676 ret = -EPERM; 677 break; 678 } 679 goto set_rcvbuf; 680 681 case SO_KEEPALIVE: 682#ifdef CONFIG_INET 683 if (sk->sk_protocol == IPPROTO_TCP && 684 sk->sk_type == SOCK_STREAM) 685 tcp_set_keepalive(sk, valbool); 686#endif 687 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool); 688 break; 689 690 case SO_OOBINLINE: 691 sock_valbool_flag(sk, SOCK_URGINLINE, valbool); 692 break; 693 694 case SO_NO_CHECK: 695 sk->sk_no_check = valbool; 696 break; 697 698 case SO_PRIORITY: 699 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN)) 700 sk->sk_priority = val; 701 else 702 ret = -EPERM; 703 break; 704 705 case SO_LINGER: 706 if (optlen < sizeof(ling)) { 707 ret = -EINVAL; /* 1003.1g */ 708 break; 709 } 710 if (copy_from_user(&ling, optval, sizeof(ling))) { 711 ret = -EFAULT; 712 break; 713 } 714 if (!ling.l_onoff) 715 sock_reset_flag(sk, SOCK_LINGER); 716 else { 717#if (BITS_PER_LONG == 32) 718 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ) 719 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT; 720 else 721#endif 722 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ; 723 sock_set_flag(sk, SOCK_LINGER); 724 } 725 break; 726 727 case SO_BSDCOMPAT: 728 sock_warn_obsolete_bsdism("setsockopt"); 729 break; 730 731 case SO_PASSCRED: 732 if (valbool) 733 set_bit(SOCK_PASSCRED, &sock->flags); 734 else 735 clear_bit(SOCK_PASSCRED, &sock->flags); 736 break; 737 738 case SO_TIMESTAMP: 739 case SO_TIMESTAMPNS: 740 if (valbool) { 741 if (optname == SO_TIMESTAMP) 742 sock_reset_flag(sk, SOCK_RCVTSTAMPNS); 743 else 744 sock_set_flag(sk, SOCK_RCVTSTAMPNS); 745 sock_set_flag(sk, SOCK_RCVTSTAMP); 746 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 747 } else { 748 sock_reset_flag(sk, SOCK_RCVTSTAMP); 749 sock_reset_flag(sk, SOCK_RCVTSTAMPNS); 750 } 751 break; 752 753 case SO_TIMESTAMPING: 754 if (val & ~SOF_TIMESTAMPING_MASK) { 755 ret = -EINVAL; 756 break; 757 } 758 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE, 759 val & SOF_TIMESTAMPING_TX_HARDWARE); 760 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE, 761 val & SOF_TIMESTAMPING_TX_SOFTWARE); 762 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE, 763 val & SOF_TIMESTAMPING_RX_HARDWARE); 764 if (val & SOF_TIMESTAMPING_RX_SOFTWARE) 765 sock_enable_timestamp(sk, 766 SOCK_TIMESTAMPING_RX_SOFTWARE); 767 else 768 sock_disable_timestamp(sk, 769 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE)); 770 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SOFTWARE, 771 val & SOF_TIMESTAMPING_SOFTWARE); 772 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE, 773 val & SOF_TIMESTAMPING_SYS_HARDWARE); 774 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE, 775 val & SOF_TIMESTAMPING_RAW_HARDWARE); 776 break; 777 778 case SO_RCVLOWAT: 779 if (val < 0) 780 val = INT_MAX; 781 sk->sk_rcvlowat = val ? : 1; 782 break; 783 784 case SO_RCVTIMEO: 785 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen); 786 break; 787 788 case SO_SNDTIMEO: 789 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen); 790 break; 791 792 case SO_ATTACH_FILTER: 793 ret = -EINVAL; 794 if (optlen == sizeof(struct sock_fprog)) { 795 struct sock_fprog fprog; 796 797 ret = -EFAULT; 798 if (copy_from_user(&fprog, optval, sizeof(fprog))) 799 break; 800 801 ret = sk_attach_filter(&fprog, sk); 802 } 803 break; 804 805 case SO_DETACH_FILTER: 806 ret = sk_detach_filter(sk); 807 break; 808 809 case SO_PASSSEC: 810 if (valbool) 811 set_bit(SOCK_PASSSEC, &sock->flags); 812 else 813 clear_bit(SOCK_PASSSEC, &sock->flags); 814 break; 815 case SO_MARK: 816 if (!capable(CAP_NET_ADMIN)) 817 ret = -EPERM; 818 else 819 sk->sk_mark = val; 820 break; 821 822 /* We implement the SO_SNDLOWAT etc to 823 not be settable (1003.1g 5.3) */ 824 case SO_RXQ_OVFL: 825 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool); 826 break; 827 828 case SO_WIFI_STATUS: 829 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool); 830 break; 831 832 case SO_PEEK_OFF: 833 if (sock->ops->set_peek_off) 834 sock->ops->set_peek_off(sk, val); 835 else 836 ret = -EOPNOTSUPP; 837 break; 838 839 case SO_NOFCS: 840 sock_valbool_flag(sk, SOCK_NOFCS, valbool); 841 break; 842 843 default: 844 ret = -ENOPROTOOPT; 845 break; 846 } 847 release_sock(sk); 848 return ret; 849} 850EXPORT_SYMBOL(sock_setsockopt); 851 852 853void cred_to_ucred(struct pid *pid, const struct cred *cred, 854 struct ucred *ucred) 855{ 856 ucred->pid = pid_vnr(pid); 857 ucred->uid = ucred->gid = -1; 858 if (cred) { 859 struct user_namespace *current_ns = current_user_ns(); 860 861 ucred->uid = from_kuid_munged(current_ns, cred->euid); 862 ucred->gid = from_kgid_munged(current_ns, cred->egid); 863 } 864} 865EXPORT_SYMBOL_GPL(cred_to_ucred); 866 867int sock_getsockopt(struct socket *sock, int level, int optname, 868 char __user *optval, int __user *optlen) 869{ 870 struct sock *sk = sock->sk; 871 872 union { 873 int val; 874 struct linger ling; 875 struct timeval tm; 876 } v; 877 878 int lv = sizeof(int); 879 int len; 880 881 if (get_user(len, optlen)) 882 return -EFAULT; 883 if (len < 0) 884 return -EINVAL; 885 886 memset(&v, 0, sizeof(v)); 887 888 switch (optname) { 889 case SO_DEBUG: 890 v.val = sock_flag(sk, SOCK_DBG); 891 break; 892 893 case SO_DONTROUTE: 894 v.val = sock_flag(sk, SOCK_LOCALROUTE); 895 break; 896 897 case SO_BROADCAST: 898 v.val = sock_flag(sk, SOCK_BROADCAST); 899 break; 900 901 case SO_SNDBUF: 902 v.val = sk->sk_sndbuf; 903 break; 904 905 case SO_RCVBUF: 906 v.val = sk->sk_rcvbuf; 907 break; 908 909 case SO_REUSEADDR: 910 v.val = sk->sk_reuse; 911 break; 912 913 case SO_KEEPALIVE: 914 v.val = sock_flag(sk, SOCK_KEEPOPEN); 915 break; 916 917 case SO_TYPE: 918 v.val = sk->sk_type; 919 break; 920 921 case SO_PROTOCOL: 922 v.val = sk->sk_protocol; 923 break; 924 925 case SO_DOMAIN: 926 v.val = sk->sk_family; 927 break; 928 929 case SO_ERROR: 930 v.val = -sock_error(sk); 931 if (v.val == 0) 932 v.val = xchg(&sk->sk_err_soft, 0); 933 break; 934 935 case SO_OOBINLINE: 936 v.val = sock_flag(sk, SOCK_URGINLINE); 937 break; 938 939 case SO_NO_CHECK: 940 v.val = sk->sk_no_check; 941 break; 942 943 case SO_PRIORITY: 944 v.val = sk->sk_priority; 945 break; 946 947 case SO_LINGER: 948 lv = sizeof(v.ling); 949 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER); 950 v.ling.l_linger = sk->sk_lingertime / HZ; 951 break; 952 953 case SO_BSDCOMPAT: 954 sock_warn_obsolete_bsdism("getsockopt"); 955 break; 956 957 case SO_TIMESTAMP: 958 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && 959 !sock_flag(sk, SOCK_RCVTSTAMPNS); 960 break; 961 962 case SO_TIMESTAMPNS: 963 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS); 964 break; 965 966 case SO_TIMESTAMPING: 967 v.val = 0; 968 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE)) 969 v.val |= SOF_TIMESTAMPING_TX_HARDWARE; 970 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE)) 971 v.val |= SOF_TIMESTAMPING_TX_SOFTWARE; 972 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE)) 973 v.val |= SOF_TIMESTAMPING_RX_HARDWARE; 974 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE)) 975 v.val |= SOF_TIMESTAMPING_RX_SOFTWARE; 976 if (sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE)) 977 v.val |= SOF_TIMESTAMPING_SOFTWARE; 978 if (sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE)) 979 v.val |= SOF_TIMESTAMPING_SYS_HARDWARE; 980 if (sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE)) 981 v.val |= SOF_TIMESTAMPING_RAW_HARDWARE; 982 break; 983 984 case SO_RCVTIMEO: 985 lv = sizeof(struct timeval); 986 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) { 987 v.tm.tv_sec = 0; 988 v.tm.tv_usec = 0; 989 } else { 990 v.tm.tv_sec = sk->sk_rcvtimeo / HZ; 991 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ; 992 } 993 break; 994 995 case SO_SNDTIMEO: 996 lv = sizeof(struct timeval); 997 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) { 998 v.tm.tv_sec = 0; 999 v.tm.tv_usec = 0; 1000 } else { 1001 v.tm.tv_sec = sk->sk_sndtimeo / HZ; 1002 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ; 1003 } 1004 break; 1005 1006 case SO_RCVLOWAT: 1007 v.val = sk->sk_rcvlowat; 1008 break; 1009 1010 case SO_SNDLOWAT: 1011 v.val = 1; 1012 break; 1013 1014 case SO_PASSCRED: 1015 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags); 1016 break; 1017 1018 case SO_PEERCRED: 1019 { 1020 struct ucred peercred; 1021 if (len > sizeof(peercred)) 1022 len = sizeof(peercred); 1023 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred); 1024 if (copy_to_user(optval, &peercred, len)) 1025 return -EFAULT; 1026 goto lenout; 1027 } 1028 1029 case SO_PEERNAME: 1030 { 1031 char address[128]; 1032 1033 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2)) 1034 return -ENOTCONN; 1035 if (lv < len) 1036 return -EINVAL; 1037 if (copy_to_user(optval, address, len)) 1038 return -EFAULT; 1039 goto lenout; 1040 } 1041 1042 /* Dubious BSD thing... Probably nobody even uses it, but 1043 * the UNIX standard wants it for whatever reason... -DaveM 1044 */ 1045 case SO_ACCEPTCONN: 1046 v.val = sk->sk_state == TCP_LISTEN; 1047 break; 1048 1049 case SO_PASSSEC: 1050 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags); 1051 break; 1052 1053 case SO_PEERSEC: 1054 return security_socket_getpeersec_stream(sock, optval, optlen, len); 1055 1056 case SO_MARK: 1057 v.val = sk->sk_mark; 1058 break; 1059 1060 case SO_RXQ_OVFL: 1061 v.val = sock_flag(sk, SOCK_RXQ_OVFL); 1062 break; 1063 1064 case SO_WIFI_STATUS: 1065 v.val = sock_flag(sk, SOCK_WIFI_STATUS); 1066 break; 1067 1068 case SO_PEEK_OFF: 1069 if (!sock->ops->set_peek_off) 1070 return -EOPNOTSUPP; 1071 1072 v.val = sk->sk_peek_off; 1073 break; 1074 case SO_NOFCS: 1075 v.val = sock_flag(sk, SOCK_NOFCS); 1076 break; 1077 case SO_BINDTODEVICE: 1078 v.val = sk->sk_bound_dev_if; 1079 break; 1080 default: 1081 return -ENOPROTOOPT; 1082 } 1083 1084 if (len > lv) 1085 len = lv; 1086 if (copy_to_user(optval, &v, len)) 1087 return -EFAULT; 1088lenout: 1089 if (put_user(len, optlen)) 1090 return -EFAULT; 1091 return 0; 1092} 1093 1094/* 1095 * Initialize an sk_lock. 1096 * 1097 * (We also register the sk_lock with the lock validator.) 1098 */ 1099static inline void sock_lock_init(struct sock *sk) 1100{ 1101 sock_lock_init_class_and_name(sk, 1102 af_family_slock_key_strings[sk->sk_family], 1103 af_family_slock_keys + sk->sk_family, 1104 af_family_key_strings[sk->sk_family], 1105 af_family_keys + sk->sk_family); 1106} 1107 1108/* 1109 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet, 1110 * even temporarly, because of RCU lookups. sk_node should also be left as is. 1111 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end 1112 */ 1113static void sock_copy(struct sock *nsk, const struct sock *osk) 1114{ 1115#ifdef CONFIG_SECURITY_NETWORK 1116 void *sptr = nsk->sk_security; 1117#endif 1118 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin)); 1119 1120 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end, 1121 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end)); 1122 1123#ifdef CONFIG_SECURITY_NETWORK 1124 nsk->sk_security = sptr; 1125 security_sk_clone(osk, nsk); 1126#endif 1127} 1128 1129/* 1130 * caches using SLAB_DESTROY_BY_RCU should let .next pointer from nulls nodes 1131 * un-modified. Special care is taken when initializing object to zero. 1132 */ 1133static inline void sk_prot_clear_nulls(struct sock *sk, int size) 1134{ 1135 if (offsetof(struct sock, sk_node.next) != 0) 1136 memset(sk, 0, offsetof(struct sock, sk_node.next)); 1137 memset(&sk->sk_node.pprev, 0, 1138 size - offsetof(struct sock, sk_node.pprev)); 1139} 1140 1141void sk_prot_clear_portaddr_nulls(struct sock *sk, int size) 1142{ 1143 unsigned long nulls1, nulls2; 1144 1145 nulls1 = offsetof(struct sock, __sk_common.skc_node.next); 1146 nulls2 = offsetof(struct sock, __sk_common.skc_portaddr_node.next); 1147 if (nulls1 > nulls2) 1148 swap(nulls1, nulls2); 1149 1150 if (nulls1 != 0) 1151 memset((char *)sk, 0, nulls1); 1152 memset((char *)sk + nulls1 + sizeof(void *), 0, 1153 nulls2 - nulls1 - sizeof(void *)); 1154 memset((char *)sk + nulls2 + sizeof(void *), 0, 1155 size - nulls2 - sizeof(void *)); 1156} 1157EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls); 1158 1159static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority, 1160 int family) 1161{ 1162 struct sock *sk; 1163 struct kmem_cache *slab; 1164 1165 slab = prot->slab; 1166 if (slab != NULL) { 1167 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO); 1168 if (!sk) 1169 return sk; 1170 if (priority & __GFP_ZERO) { 1171 if (prot->clear_sk) 1172 prot->clear_sk(sk, prot->obj_size); 1173 else 1174 sk_prot_clear_nulls(sk, prot->obj_size); 1175 } 1176 } else 1177 sk = kmalloc(prot->obj_size, priority); 1178 1179 if (sk != NULL) { 1180 kmemcheck_annotate_bitfield(sk, flags); 1181 1182 if (security_sk_alloc(sk, family, priority)) 1183 goto out_free; 1184 1185 if (!try_module_get(prot->owner)) 1186 goto out_free_sec; 1187 sk_tx_queue_clear(sk); 1188 } 1189 1190 return sk; 1191 1192out_free_sec: 1193 security_sk_free(sk); 1194out_free: 1195 if (slab != NULL) 1196 kmem_cache_free(slab, sk); 1197 else 1198 kfree(sk); 1199 return NULL; 1200} 1201 1202static void sk_prot_free(struct proto *prot, struct sock *sk) 1203{ 1204 struct kmem_cache *slab; 1205 struct module *owner; 1206 1207 owner = prot->owner; 1208 slab = prot->slab; 1209 1210 security_sk_free(sk); 1211 if (slab != NULL) 1212 kmem_cache_free(slab, sk); 1213 else 1214 kfree(sk); 1215 module_put(owner); 1216} 1217 1218#ifdef CONFIG_CGROUPS 1219#if IS_ENABLED(CONFIG_NET_CLS_CGROUP) 1220void sock_update_classid(struct sock *sk) 1221{ 1222 u32 classid; 1223 1224 rcu_read_lock(); /* doing current task, which cannot vanish. */ 1225 classid = task_cls_classid(current); 1226 rcu_read_unlock(); 1227 if (classid != sk->sk_classid) 1228 sk->sk_classid = classid; 1229} 1230EXPORT_SYMBOL(sock_update_classid); 1231#endif 1232 1233#if IS_ENABLED(CONFIG_NETPRIO_CGROUP) 1234void sock_update_netprioidx(struct sock *sk, struct task_struct *task) 1235{ 1236 if (in_interrupt()) 1237 return; 1238 1239 sk->sk_cgrp_prioidx = task_netprioidx(task); 1240} 1241EXPORT_SYMBOL_GPL(sock_update_netprioidx); 1242#endif 1243#endif 1244 1245/** 1246 * sk_alloc - All socket objects are allocated here 1247 * @net: the applicable net namespace 1248 * @family: protocol family 1249 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 1250 * @prot: struct proto associated with this new sock instance 1251 */ 1252struct sock *sk_alloc(struct net *net, int family, gfp_t priority, 1253 struct proto *prot) 1254{ 1255 struct sock *sk; 1256 1257 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family); 1258 if (sk) { 1259 sk->sk_family = family; 1260 /* 1261 * See comment in struct sock definition to understand 1262 * why we need sk_prot_creator -acme 1263 */ 1264 sk->sk_prot = sk->sk_prot_creator = prot; 1265 sock_lock_init(sk); 1266 sock_net_set(sk, get_net(net)); 1267 atomic_set(&sk->sk_wmem_alloc, 1); 1268 1269 sock_update_classid(sk); 1270 sock_update_netprioidx(sk, current); 1271 } 1272 1273 return sk; 1274} 1275EXPORT_SYMBOL(sk_alloc); 1276 1277static void __sk_free(struct sock *sk) 1278{ 1279 struct sk_filter *filter; 1280 1281 if (sk->sk_destruct) 1282 sk->sk_destruct(sk); 1283 1284 filter = rcu_dereference_check(sk->sk_filter, 1285 atomic_read(&sk->sk_wmem_alloc) == 0); 1286 if (filter) { 1287 sk_filter_uncharge(sk, filter); 1288 RCU_INIT_POINTER(sk->sk_filter, NULL); 1289 } 1290 1291 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP); 1292 1293 if (atomic_read(&sk->sk_omem_alloc)) 1294 pr_debug("%s: optmem leakage (%d bytes) detected\n", 1295 __func__, atomic_read(&sk->sk_omem_alloc)); 1296 1297 if (sk->sk_peer_cred) 1298 put_cred(sk->sk_peer_cred); 1299 put_pid(sk->sk_peer_pid); 1300 put_net(sock_net(sk)); 1301 sk_prot_free(sk->sk_prot_creator, sk); 1302} 1303 1304void sk_free(struct sock *sk) 1305{ 1306 /* 1307 * We subtract one from sk_wmem_alloc and can know if 1308 * some packets are still in some tx queue. 1309 * If not null, sock_wfree() will call __sk_free(sk) later 1310 */ 1311 if (atomic_dec_and_test(&sk->sk_wmem_alloc)) 1312 __sk_free(sk); 1313} 1314EXPORT_SYMBOL(sk_free); 1315 1316/* 1317 * Last sock_put should drop reference to sk->sk_net. It has already 1318 * been dropped in sk_change_net. Taking reference to stopping namespace 1319 * is not an option. 1320 * Take reference to a socket to remove it from hash _alive_ and after that 1321 * destroy it in the context of init_net. 1322 */ 1323void sk_release_kernel(struct sock *sk) 1324{ 1325 if (sk == NULL || sk->sk_socket == NULL) 1326 return; 1327 1328 sock_hold(sk); 1329 sock_release(sk->sk_socket); 1330 release_net(sock_net(sk)); 1331 sock_net_set(sk, get_net(&init_net)); 1332 sock_put(sk); 1333} 1334EXPORT_SYMBOL(sk_release_kernel); 1335 1336static void sk_update_clone(const struct sock *sk, struct sock *newsk) 1337{ 1338 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) 1339 sock_update_memcg(newsk); 1340} 1341 1342/** 1343 * sk_clone_lock - clone a socket, and lock its clone 1344 * @sk: the socket to clone 1345 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 1346 * 1347 * Caller must unlock socket even in error path (bh_unlock_sock(newsk)) 1348 */ 1349struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority) 1350{ 1351 struct sock *newsk; 1352 1353 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family); 1354 if (newsk != NULL) { 1355 struct sk_filter *filter; 1356 1357 sock_copy(newsk, sk); 1358 1359 /* SANITY */ 1360 get_net(sock_net(newsk)); 1361 sk_node_init(&newsk->sk_node); 1362 sock_lock_init(newsk); 1363 bh_lock_sock(newsk); 1364 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL; 1365 newsk->sk_backlog.len = 0; 1366 1367 atomic_set(&newsk->sk_rmem_alloc, 0); 1368 /* 1369 * sk_wmem_alloc set to one (see sk_free() and sock_wfree()) 1370 */ 1371 atomic_set(&newsk->sk_wmem_alloc, 1); 1372 atomic_set(&newsk->sk_omem_alloc, 0); 1373 skb_queue_head_init(&newsk->sk_receive_queue); 1374 skb_queue_head_init(&newsk->sk_write_queue); 1375#ifdef CONFIG_NET_DMA 1376 skb_queue_head_init(&newsk->sk_async_wait_queue); 1377#endif 1378 1379 spin_lock_init(&newsk->sk_dst_lock); 1380 rwlock_init(&newsk->sk_callback_lock); 1381 lockdep_set_class_and_name(&newsk->sk_callback_lock, 1382 af_callback_keys + newsk->sk_family, 1383 af_family_clock_key_strings[newsk->sk_family]); 1384 1385 newsk->sk_dst_cache = NULL; 1386 newsk->sk_wmem_queued = 0; 1387 newsk->sk_forward_alloc = 0; 1388 newsk->sk_send_head = NULL; 1389 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK; 1390 1391 sock_reset_flag(newsk, SOCK_DONE); 1392 skb_queue_head_init(&newsk->sk_error_queue); 1393 1394 filter = rcu_dereference_protected(newsk->sk_filter, 1); 1395 if (filter != NULL) 1396 sk_filter_charge(newsk, filter); 1397 1398 if (unlikely(xfrm_sk_clone_policy(newsk))) { 1399 /* It is still raw copy of parent, so invalidate 1400 * destructor and make plain sk_free() */ 1401 newsk->sk_destruct = NULL; 1402 bh_unlock_sock(newsk); 1403 sk_free(newsk); 1404 newsk = NULL; 1405 goto out; 1406 } 1407 1408 newsk->sk_err = 0; 1409 newsk->sk_priority = 0; 1410 /* 1411 * Before updating sk_refcnt, we must commit prior changes to memory 1412 * (Documentation/RCU/rculist_nulls.txt for details) 1413 */ 1414 smp_wmb(); 1415 atomic_set(&newsk->sk_refcnt, 2); 1416 1417 /* 1418 * Increment the counter in the same struct proto as the master 1419 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that 1420 * is the same as sk->sk_prot->socks, as this field was copied 1421 * with memcpy). 1422 * 1423 * This _changes_ the previous behaviour, where 1424 * tcp_create_openreq_child always was incrementing the 1425 * equivalent to tcp_prot->socks (inet_sock_nr), so this have 1426 * to be taken into account in all callers. -acme 1427 */ 1428 sk_refcnt_debug_inc(newsk); 1429 sk_set_socket(newsk, NULL); 1430 newsk->sk_wq = NULL; 1431 1432 sk_update_clone(sk, newsk); 1433 1434 if (newsk->sk_prot->sockets_allocated) 1435 sk_sockets_allocated_inc(newsk); 1436 1437 if (newsk->sk_flags & SK_FLAGS_TIMESTAMP) 1438 net_enable_timestamp(); 1439 } 1440out: 1441 return newsk; 1442} 1443EXPORT_SYMBOL_GPL(sk_clone_lock); 1444 1445void sk_setup_caps(struct sock *sk, struct dst_entry *dst) 1446{ 1447 __sk_dst_set(sk, dst); 1448 sk->sk_route_caps = dst->dev->features; 1449 if (sk->sk_route_caps & NETIF_F_GSO) 1450 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE; 1451 sk->sk_route_caps &= ~sk->sk_route_nocaps; 1452 if (sk_can_gso(sk)) { 1453 if (dst->header_len) { 1454 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 1455 } else { 1456 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM; 1457 sk->sk_gso_max_size = dst->dev->gso_max_size; 1458 sk->sk_gso_max_segs = dst->dev->gso_max_segs; 1459 } 1460 } 1461} 1462EXPORT_SYMBOL_GPL(sk_setup_caps); 1463 1464/* 1465 * Simple resource managers for sockets. 1466 */ 1467 1468 1469/* 1470 * Write buffer destructor automatically called from kfree_skb. 1471 */ 1472void sock_wfree(struct sk_buff *skb) 1473{ 1474 struct sock *sk = skb->sk; 1475 unsigned int len = skb->truesize; 1476 1477 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) { 1478 /* 1479 * Keep a reference on sk_wmem_alloc, this will be released 1480 * after sk_write_space() call 1481 */ 1482 atomic_sub(len - 1, &sk->sk_wmem_alloc); 1483 sk->sk_write_space(sk); 1484 len = 1; 1485 } 1486 /* 1487 * if sk_wmem_alloc reaches 0, we must finish what sk_free() 1488 * could not do because of in-flight packets 1489 */ 1490 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc)) 1491 __sk_free(sk); 1492} 1493EXPORT_SYMBOL(sock_wfree); 1494 1495/* 1496 * Read buffer destructor automatically called from kfree_skb. 1497 */ 1498void sock_rfree(struct sk_buff *skb) 1499{ 1500 struct sock *sk = skb->sk; 1501 unsigned int len = skb->truesize; 1502 1503 atomic_sub(len, &sk->sk_rmem_alloc); 1504 sk_mem_uncharge(sk, len); 1505} 1506EXPORT_SYMBOL(sock_rfree); 1507 1508void sock_edemux(struct sk_buff *skb) 1509{ 1510 struct sock *sk = skb->sk; 1511 1512#ifdef CONFIG_INET 1513 if (sk->sk_state == TCP_TIME_WAIT) 1514 inet_twsk_put(inet_twsk(sk)); 1515 else 1516#endif 1517 sock_put(sk); 1518} 1519EXPORT_SYMBOL(sock_edemux); 1520 1521kuid_t sock_i_uid(struct sock *sk) 1522{ 1523 kuid_t uid; 1524 1525 read_lock_bh(&sk->sk_callback_lock); 1526 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID; 1527 read_unlock_bh(&sk->sk_callback_lock); 1528 return uid; 1529} 1530EXPORT_SYMBOL(sock_i_uid); 1531 1532unsigned long sock_i_ino(struct sock *sk) 1533{ 1534 unsigned long ino; 1535 1536 read_lock_bh(&sk->sk_callback_lock); 1537 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0; 1538 read_unlock_bh(&sk->sk_callback_lock); 1539 return ino; 1540} 1541EXPORT_SYMBOL(sock_i_ino); 1542 1543/* 1544 * Allocate a skb from the socket's send buffer. 1545 */ 1546struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 1547 gfp_t priority) 1548{ 1549 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) { 1550 struct sk_buff *skb = alloc_skb(size, priority); 1551 if (skb) { 1552 skb_set_owner_w(skb, sk); 1553 return skb; 1554 } 1555 } 1556 return NULL; 1557} 1558EXPORT_SYMBOL(sock_wmalloc); 1559 1560/* 1561 * Allocate a skb from the socket's receive buffer. 1562 */ 1563struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force, 1564 gfp_t priority) 1565{ 1566 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) { 1567 struct sk_buff *skb = alloc_skb(size, priority); 1568 if (skb) { 1569 skb_set_owner_r(skb, sk); 1570 return skb; 1571 } 1572 } 1573 return NULL; 1574} 1575 1576/* 1577 * Allocate a memory block from the socket's option memory buffer. 1578 */ 1579void *sock_kmalloc(struct sock *sk, int size, gfp_t priority) 1580{ 1581 if ((unsigned int)size <= sysctl_optmem_max && 1582 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) { 1583 void *mem; 1584 /* First do the add, to avoid the race if kmalloc 1585 * might sleep. 1586 */ 1587 atomic_add(size, &sk->sk_omem_alloc); 1588 mem = kmalloc(size, priority); 1589 if (mem) 1590 return mem; 1591 atomic_sub(size, &sk->sk_omem_alloc); 1592 } 1593 return NULL; 1594} 1595EXPORT_SYMBOL(sock_kmalloc); 1596 1597/* 1598 * Free an option memory block. 1599 */ 1600void sock_kfree_s(struct sock *sk, void *mem, int size) 1601{ 1602 kfree(mem); 1603 atomic_sub(size, &sk->sk_omem_alloc); 1604} 1605EXPORT_SYMBOL(sock_kfree_s); 1606 1607/* It is almost wait_for_tcp_memory minus release_sock/lock_sock. 1608 I think, these locks should be removed for datagram sockets. 1609 */ 1610static long sock_wait_for_wmem(struct sock *sk, long timeo) 1611{ 1612 DEFINE_WAIT(wait); 1613 1614 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags); 1615 for (;;) { 1616 if (!timeo) 1617 break; 1618 if (signal_pending(current)) 1619 break; 1620 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 1621 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 1622 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) 1623 break; 1624 if (sk->sk_shutdown & SEND_SHUTDOWN) 1625 break; 1626 if (sk->sk_err) 1627 break; 1628 timeo = schedule_timeout(timeo); 1629 } 1630 finish_wait(sk_sleep(sk), &wait); 1631 return timeo; 1632} 1633 1634 1635/* 1636 * Generic send/receive buffer handlers 1637 */ 1638 1639struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 1640 unsigned long data_len, int noblock, 1641 int *errcode) 1642{ 1643 struct sk_buff *skb; 1644 gfp_t gfp_mask; 1645 long timeo; 1646 int err; 1647 int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT; 1648 1649 err = -EMSGSIZE; 1650 if (npages > MAX_SKB_FRAGS) 1651 goto failure; 1652 1653 gfp_mask = sk->sk_allocation; 1654 if (gfp_mask & __GFP_WAIT) 1655 gfp_mask |= __GFP_REPEAT; 1656 1657 timeo = sock_sndtimeo(sk, noblock); 1658 while (1) { 1659 err = sock_error(sk); 1660 if (err != 0) 1661 goto failure; 1662 1663 err = -EPIPE; 1664 if (sk->sk_shutdown & SEND_SHUTDOWN) 1665 goto failure; 1666 1667 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) { 1668 skb = alloc_skb(header_len, gfp_mask); 1669 if (skb) { 1670 int i; 1671 1672 /* No pages, we're done... */ 1673 if (!data_len) 1674 break; 1675 1676 skb->truesize += data_len; 1677 skb_shinfo(skb)->nr_frags = npages; 1678 for (i = 0; i < npages; i++) { 1679 struct page *page; 1680 1681 page = alloc_pages(sk->sk_allocation, 0); 1682 if (!page) { 1683 err = -ENOBUFS; 1684 skb_shinfo(skb)->nr_frags = i; 1685 kfree_skb(skb); 1686 goto failure; 1687 } 1688 1689 __skb_fill_page_desc(skb, i, 1690 page, 0, 1691 (data_len >= PAGE_SIZE ? 1692 PAGE_SIZE : 1693 data_len)); 1694 data_len -= PAGE_SIZE; 1695 } 1696 1697 /* Full success... */ 1698 break; 1699 } 1700 err = -ENOBUFS; 1701 goto failure; 1702 } 1703 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags); 1704 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 1705 err = -EAGAIN; 1706 if (!timeo) 1707 goto failure; 1708 if (signal_pending(current)) 1709 goto interrupted; 1710 timeo = sock_wait_for_wmem(sk, timeo); 1711 } 1712 1713 skb_set_owner_w(skb, sk); 1714 return skb; 1715 1716interrupted: 1717 err = sock_intr_errno(timeo); 1718failure: 1719 *errcode = err; 1720 return NULL; 1721} 1722EXPORT_SYMBOL(sock_alloc_send_pskb); 1723 1724struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size, 1725 int noblock, int *errcode) 1726{ 1727 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode); 1728} 1729EXPORT_SYMBOL(sock_alloc_send_skb); 1730 1731/* On 32bit arches, an skb frag is limited to 2^15 */ 1732#define SKB_FRAG_PAGE_ORDER get_order(32768) 1733 1734bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag) 1735{ 1736 int order; 1737 1738 if (pfrag->page) { 1739 if (atomic_read(&pfrag->page->_count) == 1) { 1740 pfrag->offset = 0; 1741 return true; 1742 } 1743 if (pfrag->offset < pfrag->size) 1744 return true; 1745 put_page(pfrag->page); 1746 } 1747 1748 /* We restrict high order allocations to users that can afford to wait */ 1749 order = (sk->sk_allocation & __GFP_WAIT) ? SKB_FRAG_PAGE_ORDER : 0; 1750 1751 do { 1752 gfp_t gfp = sk->sk_allocation; 1753 1754 if (order) 1755 gfp |= __GFP_COMP | __GFP_NOWARN; 1756 pfrag->page = alloc_pages(gfp, order); 1757 if (likely(pfrag->page)) { 1758 pfrag->offset = 0; 1759 pfrag->size = PAGE_SIZE << order; 1760 return true; 1761 } 1762 } while (--order >= 0); 1763 1764 sk_enter_memory_pressure(sk); 1765 sk_stream_moderate_sndbuf(sk); 1766 return false; 1767} 1768EXPORT_SYMBOL(sk_page_frag_refill); 1769 1770static void __lock_sock(struct sock *sk) 1771 __releases(&sk->sk_lock.slock) 1772 __acquires(&sk->sk_lock.slock) 1773{ 1774 DEFINE_WAIT(wait); 1775 1776 for (;;) { 1777 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait, 1778 TASK_UNINTERRUPTIBLE); 1779 spin_unlock_bh(&sk->sk_lock.slock); 1780 schedule(); 1781 spin_lock_bh(&sk->sk_lock.slock); 1782 if (!sock_owned_by_user(sk)) 1783 break; 1784 } 1785 finish_wait(&sk->sk_lock.wq, &wait); 1786} 1787 1788static void __release_sock(struct sock *sk) 1789 __releases(&sk->sk_lock.slock) 1790 __acquires(&sk->sk_lock.slock) 1791{ 1792 struct sk_buff *skb = sk->sk_backlog.head; 1793 1794 do { 1795 sk->sk_backlog.head = sk->sk_backlog.tail = NULL; 1796 bh_unlock_sock(sk); 1797 1798 do { 1799 struct sk_buff *next = skb->next; 1800 1801 prefetch(next); 1802 WARN_ON_ONCE(skb_dst_is_noref(skb)); 1803 skb->next = NULL; 1804 sk_backlog_rcv(sk, skb); 1805 1806 /* 1807 * We are in process context here with softirqs 1808 * disabled, use cond_resched_softirq() to preempt. 1809 * This is safe to do because we've taken the backlog 1810 * queue private: 1811 */ 1812 cond_resched_softirq(); 1813 1814 skb = next; 1815 } while (skb != NULL); 1816 1817 bh_lock_sock(sk); 1818 } while ((skb = sk->sk_backlog.head) != NULL); 1819 1820 /* 1821 * Doing the zeroing here guarantee we can not loop forever 1822 * while a wild producer attempts to flood us. 1823 */ 1824 sk->sk_backlog.len = 0; 1825} 1826 1827/** 1828 * sk_wait_data - wait for data to arrive at sk_receive_queue 1829 * @sk: sock to wait on 1830 * @timeo: for how long 1831 * 1832 * Now socket state including sk->sk_err is changed only under lock, 1833 * hence we may omit checks after joining wait queue. 1834 * We check receive queue before schedule() only as optimization; 1835 * it is very likely that release_sock() added new data. 1836 */ 1837int sk_wait_data(struct sock *sk, long *timeo) 1838{ 1839 int rc; 1840 DEFINE_WAIT(wait); 1841 1842 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 1843 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags); 1844 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue)); 1845 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags); 1846 finish_wait(sk_sleep(sk), &wait); 1847 return rc; 1848} 1849EXPORT_SYMBOL(sk_wait_data); 1850 1851/** 1852 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated 1853 * @sk: socket 1854 * @size: memory size to allocate 1855 * @kind: allocation type 1856 * 1857 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means 1858 * rmem allocation. This function assumes that protocols which have 1859 * memory_pressure use sk_wmem_queued as write buffer accounting. 1860 */ 1861int __sk_mem_schedule(struct sock *sk, int size, int kind) 1862{ 1863 struct proto *prot = sk->sk_prot; 1864 int amt = sk_mem_pages(size); 1865 long allocated; 1866 int parent_status = UNDER_LIMIT; 1867 1868 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM; 1869 1870 allocated = sk_memory_allocated_add(sk, amt, &parent_status); 1871 1872 /* Under limit. */ 1873 if (parent_status == UNDER_LIMIT && 1874 allocated <= sk_prot_mem_limits(sk, 0)) { 1875 sk_leave_memory_pressure(sk); 1876 return 1; 1877 } 1878 1879 /* Under pressure. (we or our parents) */ 1880 if ((parent_status > SOFT_LIMIT) || 1881 allocated > sk_prot_mem_limits(sk, 1)) 1882 sk_enter_memory_pressure(sk); 1883 1884 /* Over hard limit (we or our parents) */ 1885 if ((parent_status == OVER_LIMIT) || 1886 (allocated > sk_prot_mem_limits(sk, 2))) 1887 goto suppress_allocation; 1888 1889 /* guarantee minimum buffer size under pressure */ 1890 if (kind == SK_MEM_RECV) { 1891 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0]) 1892 return 1; 1893 1894 } else { /* SK_MEM_SEND */ 1895 if (sk->sk_type == SOCK_STREAM) { 1896 if (sk->sk_wmem_queued < prot->sysctl_wmem[0]) 1897 return 1; 1898 } else if (atomic_read(&sk->sk_wmem_alloc) < 1899 prot->sysctl_wmem[0]) 1900 return 1; 1901 } 1902 1903 if (sk_has_memory_pressure(sk)) { 1904 int alloc; 1905 1906 if (!sk_under_memory_pressure(sk)) 1907 return 1; 1908 alloc = sk_sockets_allocated_read_positive(sk); 1909 if (sk_prot_mem_limits(sk, 2) > alloc * 1910 sk_mem_pages(sk->sk_wmem_queued + 1911 atomic_read(&sk->sk_rmem_alloc) + 1912 sk->sk_forward_alloc)) 1913 return 1; 1914 } 1915 1916suppress_allocation: 1917 1918 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) { 1919 sk_stream_moderate_sndbuf(sk); 1920 1921 /* Fail only if socket is _under_ its sndbuf. 1922 * In this case we cannot block, so that we have to fail. 1923 */ 1924 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf) 1925 return 1; 1926 } 1927 1928 trace_sock_exceed_buf_limit(sk, prot, allocated); 1929 1930 /* Alas. Undo changes. */ 1931 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM; 1932 1933 sk_memory_allocated_sub(sk, amt); 1934 1935 return 0; 1936} 1937EXPORT_SYMBOL(__sk_mem_schedule); 1938 1939/** 1940 * __sk_reclaim - reclaim memory_allocated 1941 * @sk: socket 1942 */ 1943void __sk_mem_reclaim(struct sock *sk) 1944{ 1945 sk_memory_allocated_sub(sk, 1946 sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT); 1947 sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1; 1948 1949 if (sk_under_memory_pressure(sk) && 1950 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0))) 1951 sk_leave_memory_pressure(sk); 1952} 1953EXPORT_SYMBOL(__sk_mem_reclaim); 1954 1955 1956/* 1957 * Set of default routines for initialising struct proto_ops when 1958 * the protocol does not support a particular function. In certain 1959 * cases where it makes no sense for a protocol to have a "do nothing" 1960 * function, some default processing is provided. 1961 */ 1962 1963int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len) 1964{ 1965 return -EOPNOTSUPP; 1966} 1967EXPORT_SYMBOL(sock_no_bind); 1968 1969int sock_no_connect(struct socket *sock, struct sockaddr *saddr, 1970 int len, int flags) 1971{ 1972 return -EOPNOTSUPP; 1973} 1974EXPORT_SYMBOL(sock_no_connect); 1975 1976int sock_no_socketpair(struct socket *sock1, struct socket *sock2) 1977{ 1978 return -EOPNOTSUPP; 1979} 1980EXPORT_SYMBOL(sock_no_socketpair); 1981 1982int sock_no_accept(struct socket *sock, struct socket *newsock, int flags) 1983{ 1984 return -EOPNOTSUPP; 1985} 1986EXPORT_SYMBOL(sock_no_accept); 1987 1988int sock_no_getname(struct socket *sock, struct sockaddr *saddr, 1989 int *len, int peer) 1990{ 1991 return -EOPNOTSUPP; 1992} 1993EXPORT_SYMBOL(sock_no_getname); 1994 1995unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt) 1996{ 1997 return 0; 1998} 1999EXPORT_SYMBOL(sock_no_poll); 2000 2001int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 2002{ 2003 return -EOPNOTSUPP; 2004} 2005EXPORT_SYMBOL(sock_no_ioctl); 2006 2007int sock_no_listen(struct socket *sock, int backlog) 2008{ 2009 return -EOPNOTSUPP; 2010} 2011EXPORT_SYMBOL(sock_no_listen); 2012 2013int sock_no_shutdown(struct socket *sock, int how) 2014{ 2015 return -EOPNOTSUPP; 2016} 2017EXPORT_SYMBOL(sock_no_shutdown); 2018 2019int sock_no_setsockopt(struct socket *sock, int level, int optname, 2020 char __user *optval, unsigned int optlen) 2021{ 2022 return -EOPNOTSUPP; 2023} 2024EXPORT_SYMBOL(sock_no_setsockopt); 2025 2026int sock_no_getsockopt(struct socket *sock, int level, int optname, 2027 char __user *optval, int __user *optlen) 2028{ 2029 return -EOPNOTSUPP; 2030} 2031EXPORT_SYMBOL(sock_no_getsockopt); 2032 2033int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m, 2034 size_t len) 2035{ 2036 return -EOPNOTSUPP; 2037} 2038EXPORT_SYMBOL(sock_no_sendmsg); 2039 2040int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m, 2041 size_t len, int flags) 2042{ 2043 return -EOPNOTSUPP; 2044} 2045EXPORT_SYMBOL(sock_no_recvmsg); 2046 2047int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma) 2048{ 2049 /* Mirror missing mmap method error code */ 2050 return -ENODEV; 2051} 2052EXPORT_SYMBOL(sock_no_mmap); 2053 2054ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags) 2055{ 2056 ssize_t res; 2057 struct msghdr msg = {.msg_flags = flags}; 2058 struct kvec iov; 2059 char *kaddr = kmap(page); 2060 iov.iov_base = kaddr + offset; 2061 iov.iov_len = size; 2062 res = kernel_sendmsg(sock, &msg, &iov, 1, size); 2063 kunmap(page); 2064 return res; 2065} 2066EXPORT_SYMBOL(sock_no_sendpage); 2067 2068/* 2069 * Default Socket Callbacks 2070 */ 2071 2072static void sock_def_wakeup(struct sock *sk) 2073{ 2074 struct socket_wq *wq; 2075 2076 rcu_read_lock(); 2077 wq = rcu_dereference(sk->sk_wq); 2078 if (wq_has_sleeper(wq)) 2079 wake_up_interruptible_all(&wq->wait); 2080 rcu_read_unlock(); 2081} 2082 2083static void sock_def_error_report(struct sock *sk) 2084{ 2085 struct socket_wq *wq; 2086 2087 rcu_read_lock(); 2088 wq = rcu_dereference(sk->sk_wq); 2089 if (wq_has_sleeper(wq)) 2090 wake_up_interruptible_poll(&wq->wait, POLLERR); 2091 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR); 2092 rcu_read_unlock(); 2093} 2094 2095static void sock_def_readable(struct sock *sk, int len) 2096{ 2097 struct socket_wq *wq; 2098 2099 rcu_read_lock(); 2100 wq = rcu_dereference(sk->sk_wq); 2101 if (wq_has_sleeper(wq)) 2102 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI | 2103 POLLRDNORM | POLLRDBAND); 2104 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN); 2105 rcu_read_unlock(); 2106} 2107 2108static void sock_def_write_space(struct sock *sk) 2109{ 2110 struct socket_wq *wq; 2111 2112 rcu_read_lock(); 2113 2114 /* Do not wake up a writer until he can make "significant" 2115 * progress. --DaveM 2116 */ 2117 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) { 2118 wq = rcu_dereference(sk->sk_wq); 2119 if (wq_has_sleeper(wq)) 2120 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT | 2121 POLLWRNORM | POLLWRBAND); 2122 2123 /* Should agree with poll, otherwise some programs break */ 2124 if (sock_writeable(sk)) 2125 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT); 2126 } 2127 2128 rcu_read_unlock(); 2129} 2130 2131static void sock_def_destruct(struct sock *sk) 2132{ 2133 kfree(sk->sk_protinfo); 2134} 2135 2136void sk_send_sigurg(struct sock *sk) 2137{ 2138 if (sk->sk_socket && sk->sk_socket->file) 2139 if (send_sigurg(&sk->sk_socket->file->f_owner)) 2140 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI); 2141} 2142EXPORT_SYMBOL(sk_send_sigurg); 2143 2144void sk_reset_timer(struct sock *sk, struct timer_list* timer, 2145 unsigned long expires) 2146{ 2147 if (!mod_timer(timer, expires)) 2148 sock_hold(sk); 2149} 2150EXPORT_SYMBOL(sk_reset_timer); 2151 2152void sk_stop_timer(struct sock *sk, struct timer_list* timer) 2153{ 2154 if (timer_pending(timer) && del_timer(timer)) 2155 __sock_put(sk); 2156} 2157EXPORT_SYMBOL(sk_stop_timer); 2158 2159void sock_init_data(struct socket *sock, struct sock *sk) 2160{ 2161 skb_queue_head_init(&sk->sk_receive_queue); 2162 skb_queue_head_init(&sk->sk_write_queue); 2163 skb_queue_head_init(&sk->sk_error_queue); 2164#ifdef CONFIG_NET_DMA 2165 skb_queue_head_init(&sk->sk_async_wait_queue); 2166#endif 2167 2168 sk->sk_send_head = NULL; 2169 2170 init_timer(&sk->sk_timer); 2171 2172 sk->sk_allocation = GFP_KERNEL; 2173 sk->sk_rcvbuf = sysctl_rmem_default; 2174 sk->sk_sndbuf = sysctl_wmem_default; 2175 sk->sk_state = TCP_CLOSE; 2176 sk_set_socket(sk, sock); 2177 2178 sock_set_flag(sk, SOCK_ZAPPED); 2179 2180 if (sock) { 2181 sk->sk_type = sock->type; 2182 sk->sk_wq = sock->wq; 2183 sock->sk = sk; 2184 } else 2185 sk->sk_wq = NULL; 2186 2187 spin_lock_init(&sk->sk_dst_lock); 2188 rwlock_init(&sk->sk_callback_lock); 2189 lockdep_set_class_and_name(&sk->sk_callback_lock, 2190 af_callback_keys + sk->sk_family, 2191 af_family_clock_key_strings[sk->sk_family]); 2192 2193 sk->sk_state_change = sock_def_wakeup; 2194 sk->sk_data_ready = sock_def_readable; 2195 sk->sk_write_space = sock_def_write_space; 2196 sk->sk_error_report = sock_def_error_report; 2197 sk->sk_destruct = sock_def_destruct; 2198 2199 sk->sk_frag.page = NULL; 2200 sk->sk_frag.offset = 0; 2201 sk->sk_peek_off = -1; 2202 2203 sk->sk_peer_pid = NULL; 2204 sk->sk_peer_cred = NULL; 2205 sk->sk_write_pending = 0; 2206 sk->sk_rcvlowat = 1; 2207 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; 2208 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT; 2209 2210 sk->sk_stamp = ktime_set(-1L, 0); 2211 2212 /* 2213 * Before updating sk_refcnt, we must commit prior changes to memory 2214 * (Documentation/RCU/rculist_nulls.txt for details) 2215 */ 2216 smp_wmb(); 2217 atomic_set(&sk->sk_refcnt, 1); 2218 atomic_set(&sk->sk_drops, 0); 2219} 2220EXPORT_SYMBOL(sock_init_data); 2221 2222void lock_sock_nested(struct sock *sk, int subclass) 2223{ 2224 might_sleep(); 2225 spin_lock_bh(&sk->sk_lock.slock); 2226 if (sk->sk_lock.owned) 2227 __lock_sock(sk); 2228 sk->sk_lock.owned = 1; 2229 spin_unlock(&sk->sk_lock.slock); 2230 /* 2231 * The sk_lock has mutex_lock() semantics here: 2232 */ 2233 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_); 2234 local_bh_enable(); 2235} 2236EXPORT_SYMBOL(lock_sock_nested); 2237 2238void release_sock(struct sock *sk) 2239{ 2240 /* 2241 * The sk_lock has mutex_unlock() semantics: 2242 */ 2243 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_); 2244 2245 spin_lock_bh(&sk->sk_lock.slock); 2246 if (sk->sk_backlog.tail) 2247 __release_sock(sk); 2248 2249 if (sk->sk_prot->release_cb) 2250 sk->sk_prot->release_cb(sk); 2251 2252 sk->sk_lock.owned = 0; 2253 if (waitqueue_active(&sk->sk_lock.wq)) 2254 wake_up(&sk->sk_lock.wq); 2255 spin_unlock_bh(&sk->sk_lock.slock); 2256} 2257EXPORT_SYMBOL(release_sock); 2258 2259/** 2260 * lock_sock_fast - fast version of lock_sock 2261 * @sk: socket 2262 * 2263 * This version should be used for very small section, where process wont block 2264 * return false if fast path is taken 2265 * sk_lock.slock locked, owned = 0, BH disabled 2266 * return true if slow path is taken 2267 * sk_lock.slock unlocked, owned = 1, BH enabled 2268 */ 2269bool lock_sock_fast(struct sock *sk) 2270{ 2271 might_sleep(); 2272 spin_lock_bh(&sk->sk_lock.slock); 2273 2274 if (!sk->sk_lock.owned) 2275 /* 2276 * Note : We must disable BH 2277 */ 2278 return false; 2279 2280 __lock_sock(sk); 2281 sk->sk_lock.owned = 1; 2282 spin_unlock(&sk->sk_lock.slock); 2283 /* 2284 * The sk_lock has mutex_lock() semantics here: 2285 */ 2286 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_); 2287 local_bh_enable(); 2288 return true; 2289} 2290EXPORT_SYMBOL(lock_sock_fast); 2291 2292int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp) 2293{ 2294 struct timeval tv; 2295 if (!sock_flag(sk, SOCK_TIMESTAMP)) 2296 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 2297 tv = ktime_to_timeval(sk->sk_stamp); 2298 if (tv.tv_sec == -1) 2299 return -ENOENT; 2300 if (tv.tv_sec == 0) { 2301 sk->sk_stamp = ktime_get_real(); 2302 tv = ktime_to_timeval(sk->sk_stamp); 2303 } 2304 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0; 2305} 2306EXPORT_SYMBOL(sock_get_timestamp); 2307 2308int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp) 2309{ 2310 struct timespec ts; 2311 if (!sock_flag(sk, SOCK_TIMESTAMP)) 2312 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 2313 ts = ktime_to_timespec(sk->sk_stamp); 2314 if (ts.tv_sec == -1) 2315 return -ENOENT; 2316 if (ts.tv_sec == 0) { 2317 sk->sk_stamp = ktime_get_real(); 2318 ts = ktime_to_timespec(sk->sk_stamp); 2319 } 2320 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0; 2321} 2322EXPORT_SYMBOL(sock_get_timestampns); 2323 2324void sock_enable_timestamp(struct sock *sk, int flag) 2325{ 2326 if (!sock_flag(sk, flag)) { 2327 unsigned long previous_flags = sk->sk_flags; 2328 2329 sock_set_flag(sk, flag); 2330 /* 2331 * we just set one of the two flags which require net 2332 * time stamping, but time stamping might have been on 2333 * already because of the other one 2334 */ 2335 if (!(previous_flags & SK_FLAGS_TIMESTAMP)) 2336 net_enable_timestamp(); 2337 } 2338} 2339 2340/* 2341 * Get a socket option on an socket. 2342 * 2343 * FIX: POSIX 1003.1g is very ambiguous here. It states that 2344 * asynchronous errors should be reported by getsockopt. We assume 2345 * this means if you specify SO_ERROR (otherwise whats the point of it). 2346 */ 2347int sock_common_getsockopt(struct socket *sock, int level, int optname, 2348 char __user *optval, int __user *optlen) 2349{ 2350 struct sock *sk = sock->sk; 2351 2352 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen); 2353} 2354EXPORT_SYMBOL(sock_common_getsockopt); 2355 2356#ifdef CONFIG_COMPAT 2357int compat_sock_common_getsockopt(struct socket *sock, int level, int optname, 2358 char __user *optval, int __user *optlen) 2359{ 2360 struct sock *sk = sock->sk; 2361 2362 if (sk->sk_prot->compat_getsockopt != NULL) 2363 return sk->sk_prot->compat_getsockopt(sk, level, optname, 2364 optval, optlen); 2365 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen); 2366} 2367EXPORT_SYMBOL(compat_sock_common_getsockopt); 2368#endif 2369 2370int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock, 2371 struct msghdr *msg, size_t size, int flags) 2372{ 2373 struct sock *sk = sock->sk; 2374 int addr_len = 0; 2375 int err; 2376 2377 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT, 2378 flags & ~MSG_DONTWAIT, &addr_len); 2379 if (err >= 0) 2380 msg->msg_namelen = addr_len; 2381 return err; 2382} 2383EXPORT_SYMBOL(sock_common_recvmsg); 2384 2385/* 2386 * Set socket options on an inet socket. 2387 */ 2388int sock_common_setsockopt(struct socket *sock, int level, int optname, 2389 char __user *optval, unsigned int optlen) 2390{ 2391 struct sock *sk = sock->sk; 2392 2393 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen); 2394} 2395EXPORT_SYMBOL(sock_common_setsockopt); 2396 2397#ifdef CONFIG_COMPAT 2398int compat_sock_common_setsockopt(struct socket *sock, int level, int optname, 2399 char __user *optval, unsigned int optlen) 2400{ 2401 struct sock *sk = sock->sk; 2402 2403 if (sk->sk_prot->compat_setsockopt != NULL) 2404 return sk->sk_prot->compat_setsockopt(sk, level, optname, 2405 optval, optlen); 2406 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen); 2407} 2408EXPORT_SYMBOL(compat_sock_common_setsockopt); 2409#endif 2410 2411void sk_common_release(struct sock *sk) 2412{ 2413 if (sk->sk_prot->destroy) 2414 sk->sk_prot->destroy(sk); 2415 2416 /* 2417 * Observation: when sock_common_release is called, processes have 2418 * no access to socket. But net still has. 2419 * Step one, detach it from networking: 2420 * 2421 * A. Remove from hash tables. 2422 */ 2423 2424 sk->sk_prot->unhash(sk); 2425 2426 /* 2427 * In this point socket cannot receive new packets, but it is possible 2428 * that some packets are in flight because some CPU runs receiver and 2429 * did hash table lookup before we unhashed socket. They will achieve 2430 * receive queue and will be purged by socket destructor. 2431 * 2432 * Also we still have packets pending on receive queue and probably, 2433 * our own packets waiting in device queues. sock_destroy will drain 2434 * receive queue, but transmitted packets will delay socket destruction 2435 * until the last reference will be released. 2436 */ 2437 2438 sock_orphan(sk); 2439 2440 xfrm_sk_free_policy(sk); 2441 2442 sk_refcnt_debug_release(sk); 2443 2444 if (sk->sk_frag.page) { 2445 put_page(sk->sk_frag.page); 2446 sk->sk_frag.page = NULL; 2447 } 2448 2449 sock_put(sk); 2450} 2451EXPORT_SYMBOL(sk_common_release); 2452 2453#ifdef CONFIG_PROC_FS 2454#define PROTO_INUSE_NR 64 /* should be enough for the first time */ 2455struct prot_inuse { 2456 int val[PROTO_INUSE_NR]; 2457}; 2458 2459static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR); 2460 2461#ifdef CONFIG_NET_NS 2462void sock_prot_inuse_add(struct net *net, struct proto *prot, int val) 2463{ 2464 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val); 2465} 2466EXPORT_SYMBOL_GPL(sock_prot_inuse_add); 2467 2468int sock_prot_inuse_get(struct net *net, struct proto *prot) 2469{ 2470 int cpu, idx = prot->inuse_idx; 2471 int res = 0; 2472 2473 for_each_possible_cpu(cpu) 2474 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx]; 2475 2476 return res >= 0 ? res : 0; 2477} 2478EXPORT_SYMBOL_GPL(sock_prot_inuse_get); 2479 2480static int __net_init sock_inuse_init_net(struct net *net) 2481{ 2482 net->core.inuse = alloc_percpu(struct prot_inuse); 2483 return net->core.inuse ? 0 : -ENOMEM; 2484} 2485 2486static void __net_exit sock_inuse_exit_net(struct net *net) 2487{ 2488 free_percpu(net->core.inuse); 2489} 2490 2491static struct pernet_operations net_inuse_ops = { 2492 .init = sock_inuse_init_net, 2493 .exit = sock_inuse_exit_net, 2494}; 2495 2496static __init int net_inuse_init(void) 2497{ 2498 if (register_pernet_subsys(&net_inuse_ops)) 2499 panic("Cannot initialize net inuse counters"); 2500 2501 return 0; 2502} 2503 2504core_initcall(net_inuse_init); 2505#else 2506static DEFINE_PER_CPU(struct prot_inuse, prot_inuse); 2507 2508void sock_prot_inuse_add(struct net *net, struct proto *prot, int val) 2509{ 2510 __this_cpu_add(prot_inuse.val[prot->inuse_idx], val); 2511} 2512EXPORT_SYMBOL_GPL(sock_prot_inuse_add); 2513 2514int sock_prot_inuse_get(struct net *net, struct proto *prot) 2515{ 2516 int cpu, idx = prot->inuse_idx; 2517 int res = 0; 2518 2519 for_each_possible_cpu(cpu) 2520 res += per_cpu(prot_inuse, cpu).val[idx]; 2521 2522 return res >= 0 ? res : 0; 2523} 2524EXPORT_SYMBOL_GPL(sock_prot_inuse_get); 2525#endif 2526 2527static void assign_proto_idx(struct proto *prot) 2528{ 2529 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR); 2530 2531 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) { 2532 pr_err("PROTO_INUSE_NR exhausted\n"); 2533 return; 2534 } 2535 2536 set_bit(prot->inuse_idx, proto_inuse_idx); 2537} 2538 2539static void release_proto_idx(struct proto *prot) 2540{ 2541 if (prot->inuse_idx != PROTO_INUSE_NR - 1) 2542 clear_bit(prot->inuse_idx, proto_inuse_idx); 2543} 2544#else 2545static inline void assign_proto_idx(struct proto *prot) 2546{ 2547} 2548 2549static inline void release_proto_idx(struct proto *prot) 2550{ 2551} 2552#endif 2553 2554int proto_register(struct proto *prot, int alloc_slab) 2555{ 2556 if (alloc_slab) { 2557 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0, 2558 SLAB_HWCACHE_ALIGN | prot->slab_flags, 2559 NULL); 2560 2561 if (prot->slab == NULL) { 2562 pr_crit("%s: Can't create sock SLAB cache!\n", 2563 prot->name); 2564 goto out; 2565 } 2566 2567 if (prot->rsk_prot != NULL) { 2568 prot->rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", prot->name); 2569 if (prot->rsk_prot->slab_name == NULL) 2570 goto out_free_sock_slab; 2571 2572 prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name, 2573 prot->rsk_prot->obj_size, 0, 2574 SLAB_HWCACHE_ALIGN, NULL); 2575 2576 if (prot->rsk_prot->slab == NULL) { 2577 pr_crit("%s: Can't create request sock SLAB cache!\n", 2578 prot->name); 2579 goto out_free_request_sock_slab_name; 2580 } 2581 } 2582 2583 if (prot->twsk_prot != NULL) { 2584 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name); 2585 2586 if (prot->twsk_prot->twsk_slab_name == NULL) 2587 goto out_free_request_sock_slab; 2588 2589 prot->twsk_prot->twsk_slab = 2590 kmem_cache_create(prot->twsk_prot->twsk_slab_name, 2591 prot->twsk_prot->twsk_obj_size, 2592 0, 2593 SLAB_HWCACHE_ALIGN | 2594 prot->slab_flags, 2595 NULL); 2596 if (prot->twsk_prot->twsk_slab == NULL) 2597 goto out_free_timewait_sock_slab_name; 2598 } 2599 } 2600 2601 mutex_lock(&proto_list_mutex); 2602 list_add(&prot->node, &proto_list); 2603 assign_proto_idx(prot); 2604 mutex_unlock(&proto_list_mutex); 2605 return 0; 2606 2607out_free_timewait_sock_slab_name: 2608 kfree(prot->twsk_prot->twsk_slab_name); 2609out_free_request_sock_slab: 2610 if (prot->rsk_prot && prot->rsk_prot->slab) { 2611 kmem_cache_destroy(prot->rsk_prot->slab); 2612 prot->rsk_prot->slab = NULL; 2613 } 2614out_free_request_sock_slab_name: 2615 if (prot->rsk_prot) 2616 kfree(prot->rsk_prot->slab_name); 2617out_free_sock_slab: 2618 kmem_cache_destroy(prot->slab); 2619 prot->slab = NULL; 2620out: 2621 return -ENOBUFS; 2622} 2623EXPORT_SYMBOL(proto_register); 2624 2625void proto_unregister(struct proto *prot) 2626{ 2627 mutex_lock(&proto_list_mutex); 2628 release_proto_idx(prot); 2629 list_del(&prot->node); 2630 mutex_unlock(&proto_list_mutex); 2631 2632 if (prot->slab != NULL) { 2633 kmem_cache_destroy(prot->slab); 2634 prot->slab = NULL; 2635 } 2636 2637 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) { 2638 kmem_cache_destroy(prot->rsk_prot->slab); 2639 kfree(prot->rsk_prot->slab_name); 2640 prot->rsk_prot->slab = NULL; 2641 } 2642 2643 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) { 2644 kmem_cache_destroy(prot->twsk_prot->twsk_slab); 2645 kfree(prot->twsk_prot->twsk_slab_name); 2646 prot->twsk_prot->twsk_slab = NULL; 2647 } 2648} 2649EXPORT_SYMBOL(proto_unregister); 2650 2651#ifdef CONFIG_PROC_FS 2652static void *proto_seq_start(struct seq_file *seq, loff_t *pos) 2653 __acquires(proto_list_mutex) 2654{ 2655 mutex_lock(&proto_list_mutex); 2656 return seq_list_start_head(&proto_list, *pos); 2657} 2658 2659static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos) 2660{ 2661 return seq_list_next(v, &proto_list, pos); 2662} 2663 2664static void proto_seq_stop(struct seq_file *seq, void *v) 2665 __releases(proto_list_mutex) 2666{ 2667 mutex_unlock(&proto_list_mutex); 2668} 2669 2670static char proto_method_implemented(const void *method) 2671{ 2672 return method == NULL ? 'n' : 'y'; 2673} 2674static long sock_prot_memory_allocated(struct proto *proto) 2675{ 2676 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L; 2677} 2678 2679static char *sock_prot_memory_pressure(struct proto *proto) 2680{ 2681 return proto->memory_pressure != NULL ? 2682 proto_memory_pressure(proto) ? "yes" : "no" : "NI"; 2683} 2684 2685static void proto_seq_printf(struct seq_file *seq, struct proto *proto) 2686{ 2687 2688 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s " 2689 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n", 2690 proto->name, 2691 proto->obj_size, 2692 sock_prot_inuse_get(seq_file_net(seq), proto), 2693 sock_prot_memory_allocated(proto), 2694 sock_prot_memory_pressure(proto), 2695 proto->max_header, 2696 proto->slab == NULL ? "no" : "yes", 2697 module_name(proto->owner), 2698 proto_method_implemented(proto->close), 2699 proto_method_implemented(proto->connect), 2700 proto_method_implemented(proto->disconnect), 2701 proto_method_implemented(proto->accept), 2702 proto_method_implemented(proto->ioctl), 2703 proto_method_implemented(proto->init), 2704 proto_method_implemented(proto->destroy), 2705 proto_method_implemented(proto->shutdown), 2706 proto_method_implemented(proto->setsockopt), 2707 proto_method_implemented(proto->getsockopt), 2708 proto_method_implemented(proto->sendmsg), 2709 proto_method_implemented(proto->recvmsg), 2710 proto_method_implemented(proto->sendpage), 2711 proto_method_implemented(proto->bind), 2712 proto_method_implemented(proto->backlog_rcv), 2713 proto_method_implemented(proto->hash), 2714 proto_method_implemented(proto->unhash), 2715 proto_method_implemented(proto->get_port), 2716 proto_method_implemented(proto->enter_memory_pressure)); 2717} 2718 2719static int proto_seq_show(struct seq_file *seq, void *v) 2720{ 2721 if (v == &proto_list) 2722 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s", 2723 "protocol", 2724 "size", 2725 "sockets", 2726 "memory", 2727 "press", 2728 "maxhdr", 2729 "slab", 2730 "module", 2731 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n"); 2732 else 2733 proto_seq_printf(seq, list_entry(v, struct proto, node)); 2734 return 0; 2735} 2736 2737static const struct seq_operations proto_seq_ops = { 2738 .start = proto_seq_start, 2739 .next = proto_seq_next, 2740 .stop = proto_seq_stop, 2741 .show = proto_seq_show, 2742}; 2743 2744static int proto_seq_open(struct inode *inode, struct file *file) 2745{ 2746 return seq_open_net(inode, file, &proto_seq_ops, 2747 sizeof(struct seq_net_private)); 2748} 2749 2750static const struct file_operations proto_seq_fops = { 2751 .owner = THIS_MODULE, 2752 .open = proto_seq_open, 2753 .read = seq_read, 2754 .llseek = seq_lseek, 2755 .release = seq_release_net, 2756}; 2757 2758static __net_init int proto_init_net(struct net *net) 2759{ 2760 if (!proc_net_fops_create(net, "protocols", S_IRUGO, &proto_seq_fops)) 2761 return -ENOMEM; 2762 2763 return 0; 2764} 2765 2766static __net_exit void proto_exit_net(struct net *net) 2767{ 2768 proc_net_remove(net, "protocols"); 2769} 2770 2771 2772static __net_initdata struct pernet_operations proto_net_ops = { 2773 .init = proto_init_net, 2774 .exit = proto_exit_net, 2775}; 2776 2777static int __init proto_init(void) 2778{ 2779 return register_pernet_subsys(&proto_net_ops); 2780} 2781 2782subsys_initcall(proto_init); 2783 2784#endif /* PROC_FS */ 2785