sock.c revision 884cf705c7e60bc6ade7ddafcbe943af4dc84604
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/errqueue.h> 97#include <linux/types.h> 98#include <linux/socket.h> 99#include <linux/in.h> 100#include <linux/kernel.h> 101#include <linux/module.h> 102#include <linux/proc_fs.h> 103#include <linux/seq_file.h> 104#include <linux/sched.h> 105#include <linux/timer.h> 106#include <linux/string.h> 107#include <linux/sockios.h> 108#include <linux/net.h> 109#include <linux/mm.h> 110#include <linux/slab.h> 111#include <linux/interrupt.h> 112#include <linux/poll.h> 113#include <linux/tcp.h> 114#include <linux/init.h> 115#include <linux/highmem.h> 116#include <linux/user_namespace.h> 117#include <linux/static_key.h> 118#include <linux/memcontrol.h> 119#include <linux/prefetch.h> 120 121#include <asm/uaccess.h> 122 123#include <linux/netdevice.h> 124#include <net/protocol.h> 125#include <linux/skbuff.h> 126#include <net/net_namespace.h> 127#include <net/request_sock.h> 128#include <net/sock.h> 129#include <linux/net_tstamp.h> 130#include <net/xfrm.h> 131#include <linux/ipsec.h> 132#include <net/cls_cgroup.h> 133#include <net/netprio_cgroup.h> 134 135#include <linux/filter.h> 136 137#include <trace/events/sock.h> 138 139#ifdef CONFIG_INET 140#include <net/tcp.h> 141#endif 142 143#include <net/busy_poll.h> 144 145static DEFINE_MUTEX(proto_list_mutex); 146static LIST_HEAD(proto_list); 147 148/** 149 * sk_ns_capable - General socket capability test 150 * @sk: Socket to use a capability on or through 151 * @user_ns: The user namespace of the capability to use 152 * @cap: The capability to use 153 * 154 * Test to see if the opener of the socket had when the socket was 155 * created and the current process has the capability @cap in the user 156 * namespace @user_ns. 157 */ 158bool sk_ns_capable(const struct sock *sk, 159 struct user_namespace *user_ns, int cap) 160{ 161 return file_ns_capable(sk->sk_socket->file, user_ns, cap) && 162 ns_capable(user_ns, cap); 163} 164EXPORT_SYMBOL(sk_ns_capable); 165 166/** 167 * sk_capable - Socket global capability test 168 * @sk: Socket to use a capability on or through 169 * @cap: The global capbility to use 170 * 171 * Test to see if the opener of the socket had when the socket was 172 * created and the current process has the capability @cap in all user 173 * namespaces. 174 */ 175bool sk_capable(const struct sock *sk, int cap) 176{ 177 return sk_ns_capable(sk, &init_user_ns, cap); 178} 179EXPORT_SYMBOL(sk_capable); 180 181/** 182 * sk_net_capable - Network namespace socket capability test 183 * @sk: Socket to use a capability on or through 184 * @cap: The capability to use 185 * 186 * Test to see if the opener of the socket had when the socke was created 187 * and the current process has the capability @cap over the network namespace 188 * the socket is a member of. 189 */ 190bool sk_net_capable(const struct sock *sk, int cap) 191{ 192 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap); 193} 194EXPORT_SYMBOL(sk_net_capable); 195 196 197#ifdef CONFIG_MEMCG_KMEM 198int mem_cgroup_sockets_init(struct mem_cgroup *memcg, struct cgroup_subsys *ss) 199{ 200 struct proto *proto; 201 int ret = 0; 202 203 mutex_lock(&proto_list_mutex); 204 list_for_each_entry(proto, &proto_list, node) { 205 if (proto->init_cgroup) { 206 ret = proto->init_cgroup(memcg, ss); 207 if (ret) 208 goto out; 209 } 210 } 211 212 mutex_unlock(&proto_list_mutex); 213 return ret; 214out: 215 list_for_each_entry_continue_reverse(proto, &proto_list, node) 216 if (proto->destroy_cgroup) 217 proto->destroy_cgroup(memcg); 218 mutex_unlock(&proto_list_mutex); 219 return ret; 220} 221 222void mem_cgroup_sockets_destroy(struct mem_cgroup *memcg) 223{ 224 struct proto *proto; 225 226 mutex_lock(&proto_list_mutex); 227 list_for_each_entry_reverse(proto, &proto_list, node) 228 if (proto->destroy_cgroup) 229 proto->destroy_cgroup(memcg); 230 mutex_unlock(&proto_list_mutex); 231} 232#endif 233 234/* 235 * Each address family might have different locking rules, so we have 236 * one slock key per address family: 237 */ 238static struct lock_class_key af_family_keys[AF_MAX]; 239static struct lock_class_key af_family_slock_keys[AF_MAX]; 240 241#if defined(CONFIG_MEMCG_KMEM) 242struct static_key memcg_socket_limit_enabled; 243EXPORT_SYMBOL(memcg_socket_limit_enabled); 244#endif 245 246/* 247 * Make lock validator output more readable. (we pre-construct these 248 * strings build-time, so that runtime initialization of socket 249 * locks is fast): 250 */ 251static const char *const af_family_key_strings[AF_MAX+1] = { 252 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" , 253 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK", 254 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" , 255 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" , 256 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" , 257 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" , 258 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" , 259 "sk_lock-AF_RDS" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" , 260 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" , 261 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" , 262 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" , 263 "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN" , "sk_lock-AF_PHONET" , 264 "sk_lock-AF_IEEE802154", "sk_lock-AF_CAIF" , "sk_lock-AF_ALG" , 265 "sk_lock-AF_NFC" , "sk_lock-AF_VSOCK" , "sk_lock-AF_MAX" 266}; 267static const char *const af_family_slock_key_strings[AF_MAX+1] = { 268 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" , 269 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK", 270 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" , 271 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" , 272 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" , 273 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" , 274 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" , 275 "slock-AF_RDS" , "slock-AF_SNA" , "slock-AF_IRDA" , 276 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" , 277 "slock-27" , "slock-28" , "slock-AF_CAN" , 278 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" , 279 "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" , 280 "slock-AF_IEEE802154", "slock-AF_CAIF" , "slock-AF_ALG" , 281 "slock-AF_NFC" , "slock-AF_VSOCK" ,"slock-AF_MAX" 282}; 283static const char *const af_family_clock_key_strings[AF_MAX+1] = { 284 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" , 285 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK", 286 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" , 287 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" , 288 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" , 289 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" , 290 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" , 291 "clock-AF_RDS" , "clock-AF_SNA" , "clock-AF_IRDA" , 292 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" , 293 "clock-27" , "clock-28" , "clock-AF_CAN" , 294 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" , 295 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" , 296 "clock-AF_IEEE802154", "clock-AF_CAIF" , "clock-AF_ALG" , 297 "clock-AF_NFC" , "clock-AF_VSOCK" , "clock-AF_MAX" 298}; 299 300/* 301 * sk_callback_lock locking rules are per-address-family, 302 * so split the lock classes by using a per-AF key: 303 */ 304static struct lock_class_key af_callback_keys[AF_MAX]; 305 306/* Take into consideration the size of the struct sk_buff overhead in the 307 * determination of these values, since that is non-constant across 308 * platforms. This makes socket queueing behavior and performance 309 * not depend upon such differences. 310 */ 311#define _SK_MEM_PACKETS 256 312#define _SK_MEM_OVERHEAD SKB_TRUESIZE(256) 313#define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 314#define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 315 316/* Run time adjustable parameters. */ 317__u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX; 318EXPORT_SYMBOL(sysctl_wmem_max); 319__u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX; 320EXPORT_SYMBOL(sysctl_rmem_max); 321__u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX; 322__u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX; 323 324/* Maximal space eaten by iovec or ancillary data plus some space */ 325int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512); 326EXPORT_SYMBOL(sysctl_optmem_max); 327 328struct static_key memalloc_socks = STATIC_KEY_INIT_FALSE; 329EXPORT_SYMBOL_GPL(memalloc_socks); 330 331/** 332 * sk_set_memalloc - sets %SOCK_MEMALLOC 333 * @sk: socket to set it on 334 * 335 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves. 336 * It's the responsibility of the admin to adjust min_free_kbytes 337 * to meet the requirements 338 */ 339void sk_set_memalloc(struct sock *sk) 340{ 341 sock_set_flag(sk, SOCK_MEMALLOC); 342 sk->sk_allocation |= __GFP_MEMALLOC; 343 static_key_slow_inc(&memalloc_socks); 344} 345EXPORT_SYMBOL_GPL(sk_set_memalloc); 346 347void sk_clear_memalloc(struct sock *sk) 348{ 349 sock_reset_flag(sk, SOCK_MEMALLOC); 350 sk->sk_allocation &= ~__GFP_MEMALLOC; 351 static_key_slow_dec(&memalloc_socks); 352 353 /* 354 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward 355 * progress of swapping. However, if SOCK_MEMALLOC is cleared while 356 * it has rmem allocations there is a risk that the user of the 357 * socket cannot make forward progress due to exceeding the rmem 358 * limits. By rights, sk_clear_memalloc() should only be called 359 * on sockets being torn down but warn and reset the accounting if 360 * that assumption breaks. 361 */ 362 if (WARN_ON(sk->sk_forward_alloc)) 363 sk_mem_reclaim(sk); 364} 365EXPORT_SYMBOL_GPL(sk_clear_memalloc); 366 367int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb) 368{ 369 int ret; 370 unsigned long pflags = current->flags; 371 372 /* these should have been dropped before queueing */ 373 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC)); 374 375 current->flags |= PF_MEMALLOC; 376 ret = sk->sk_backlog_rcv(sk, skb); 377 tsk_restore_flags(current, pflags, PF_MEMALLOC); 378 379 return ret; 380} 381EXPORT_SYMBOL(__sk_backlog_rcv); 382 383static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen) 384{ 385 struct timeval tv; 386 387 if (optlen < sizeof(tv)) 388 return -EINVAL; 389 if (copy_from_user(&tv, optval, sizeof(tv))) 390 return -EFAULT; 391 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC) 392 return -EDOM; 393 394 if (tv.tv_sec < 0) { 395 static int warned __read_mostly; 396 397 *timeo_p = 0; 398 if (warned < 10 && net_ratelimit()) { 399 warned++; 400 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n", 401 __func__, current->comm, task_pid_nr(current)); 402 } 403 return 0; 404 } 405 *timeo_p = MAX_SCHEDULE_TIMEOUT; 406 if (tv.tv_sec == 0 && tv.tv_usec == 0) 407 return 0; 408 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1)) 409 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ); 410 return 0; 411} 412 413static void sock_warn_obsolete_bsdism(const char *name) 414{ 415 static int warned; 416 static char warncomm[TASK_COMM_LEN]; 417 if (strcmp(warncomm, current->comm) && warned < 5) { 418 strcpy(warncomm, current->comm); 419 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n", 420 warncomm, name); 421 warned++; 422 } 423} 424 425#define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE)) 426 427static void sock_disable_timestamp(struct sock *sk, unsigned long flags) 428{ 429 if (sk->sk_flags & flags) { 430 sk->sk_flags &= ~flags; 431 if (!(sk->sk_flags & SK_FLAGS_TIMESTAMP)) 432 net_disable_timestamp(); 433 } 434} 435 436 437int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 438{ 439 int err; 440 unsigned long flags; 441 struct sk_buff_head *list = &sk->sk_receive_queue; 442 443 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) { 444 atomic_inc(&sk->sk_drops); 445 trace_sock_rcvqueue_full(sk, skb); 446 return -ENOMEM; 447 } 448 449 err = sk_filter(sk, skb); 450 if (err) 451 return err; 452 453 if (!sk_rmem_schedule(sk, skb, skb->truesize)) { 454 atomic_inc(&sk->sk_drops); 455 return -ENOBUFS; 456 } 457 458 skb->dev = NULL; 459 skb_set_owner_r(skb, sk); 460 461 /* we escape from rcu protected region, make sure we dont leak 462 * a norefcounted dst 463 */ 464 skb_dst_force(skb); 465 466 spin_lock_irqsave(&list->lock, flags); 467 skb->dropcount = atomic_read(&sk->sk_drops); 468 __skb_queue_tail(list, skb); 469 spin_unlock_irqrestore(&list->lock, flags); 470 471 if (!sock_flag(sk, SOCK_DEAD)) 472 sk->sk_data_ready(sk); 473 return 0; 474} 475EXPORT_SYMBOL(sock_queue_rcv_skb); 476 477int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested) 478{ 479 int rc = NET_RX_SUCCESS; 480 481 if (sk_filter(sk, skb)) 482 goto discard_and_relse; 483 484 skb->dev = NULL; 485 486 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) { 487 atomic_inc(&sk->sk_drops); 488 goto discard_and_relse; 489 } 490 if (nested) 491 bh_lock_sock_nested(sk); 492 else 493 bh_lock_sock(sk); 494 if (!sock_owned_by_user(sk)) { 495 /* 496 * trylock + unlock semantics: 497 */ 498 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_); 499 500 rc = sk_backlog_rcv(sk, skb); 501 502 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_); 503 } else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) { 504 bh_unlock_sock(sk); 505 atomic_inc(&sk->sk_drops); 506 goto discard_and_relse; 507 } 508 509 bh_unlock_sock(sk); 510out: 511 sock_put(sk); 512 return rc; 513discard_and_relse: 514 kfree_skb(skb); 515 goto out; 516} 517EXPORT_SYMBOL(sk_receive_skb); 518 519struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie) 520{ 521 struct dst_entry *dst = __sk_dst_get(sk); 522 523 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) { 524 sk_tx_queue_clear(sk); 525 RCU_INIT_POINTER(sk->sk_dst_cache, NULL); 526 dst_release(dst); 527 return NULL; 528 } 529 530 return dst; 531} 532EXPORT_SYMBOL(__sk_dst_check); 533 534struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie) 535{ 536 struct dst_entry *dst = sk_dst_get(sk); 537 538 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) { 539 sk_dst_reset(sk); 540 dst_release(dst); 541 return NULL; 542 } 543 544 return dst; 545} 546EXPORT_SYMBOL(sk_dst_check); 547 548static int sock_setbindtodevice(struct sock *sk, char __user *optval, 549 int optlen) 550{ 551 int ret = -ENOPROTOOPT; 552#ifdef CONFIG_NETDEVICES 553 struct net *net = sock_net(sk); 554 char devname[IFNAMSIZ]; 555 int index; 556 557 /* Sorry... */ 558 ret = -EPERM; 559 if (!ns_capable(net->user_ns, CAP_NET_RAW)) 560 goto out; 561 562 ret = -EINVAL; 563 if (optlen < 0) 564 goto out; 565 566 /* Bind this socket to a particular device like "eth0", 567 * as specified in the passed interface name. If the 568 * name is "" or the option length is zero the socket 569 * is not bound. 570 */ 571 if (optlen > IFNAMSIZ - 1) 572 optlen = IFNAMSIZ - 1; 573 memset(devname, 0, sizeof(devname)); 574 575 ret = -EFAULT; 576 if (copy_from_user(devname, optval, optlen)) 577 goto out; 578 579 index = 0; 580 if (devname[0] != '\0') { 581 struct net_device *dev; 582 583 rcu_read_lock(); 584 dev = dev_get_by_name_rcu(net, devname); 585 if (dev) 586 index = dev->ifindex; 587 rcu_read_unlock(); 588 ret = -ENODEV; 589 if (!dev) 590 goto out; 591 } 592 593 lock_sock(sk); 594 sk->sk_bound_dev_if = index; 595 sk_dst_reset(sk); 596 release_sock(sk); 597 598 ret = 0; 599 600out: 601#endif 602 603 return ret; 604} 605 606static int sock_getbindtodevice(struct sock *sk, char __user *optval, 607 int __user *optlen, int len) 608{ 609 int ret = -ENOPROTOOPT; 610#ifdef CONFIG_NETDEVICES 611 struct net *net = sock_net(sk); 612 char devname[IFNAMSIZ]; 613 614 if (sk->sk_bound_dev_if == 0) { 615 len = 0; 616 goto zero; 617 } 618 619 ret = -EINVAL; 620 if (len < IFNAMSIZ) 621 goto out; 622 623 ret = netdev_get_name(net, devname, sk->sk_bound_dev_if); 624 if (ret) 625 goto out; 626 627 len = strlen(devname) + 1; 628 629 ret = -EFAULT; 630 if (copy_to_user(optval, devname, len)) 631 goto out; 632 633zero: 634 ret = -EFAULT; 635 if (put_user(len, optlen)) 636 goto out; 637 638 ret = 0; 639 640out: 641#endif 642 643 return ret; 644} 645 646static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool) 647{ 648 if (valbool) 649 sock_set_flag(sk, bit); 650 else 651 sock_reset_flag(sk, bit); 652} 653 654/* 655 * This is meant for all protocols to use and covers goings on 656 * at the socket level. Everything here is generic. 657 */ 658 659int sock_setsockopt(struct socket *sock, int level, int optname, 660 char __user *optval, unsigned int optlen) 661{ 662 struct sock *sk = sock->sk; 663 int val; 664 int valbool; 665 struct linger ling; 666 int ret = 0; 667 668 /* 669 * Options without arguments 670 */ 671 672 if (optname == SO_BINDTODEVICE) 673 return sock_setbindtodevice(sk, optval, optlen); 674 675 if (optlen < sizeof(int)) 676 return -EINVAL; 677 678 if (get_user(val, (int __user *)optval)) 679 return -EFAULT; 680 681 valbool = val ? 1 : 0; 682 683 lock_sock(sk); 684 685 switch (optname) { 686 case SO_DEBUG: 687 if (val && !capable(CAP_NET_ADMIN)) 688 ret = -EACCES; 689 else 690 sock_valbool_flag(sk, SOCK_DBG, valbool); 691 break; 692 case SO_REUSEADDR: 693 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE); 694 break; 695 case SO_REUSEPORT: 696 sk->sk_reuseport = valbool; 697 break; 698 case SO_TYPE: 699 case SO_PROTOCOL: 700 case SO_DOMAIN: 701 case SO_ERROR: 702 ret = -ENOPROTOOPT; 703 break; 704 case SO_DONTROUTE: 705 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool); 706 break; 707 case SO_BROADCAST: 708 sock_valbool_flag(sk, SOCK_BROADCAST, valbool); 709 break; 710 case SO_SNDBUF: 711 /* Don't error on this BSD doesn't and if you think 712 * about it this is right. Otherwise apps have to 713 * play 'guess the biggest size' games. RCVBUF/SNDBUF 714 * are treated in BSD as hints 715 */ 716 val = min_t(u32, val, sysctl_wmem_max); 717set_sndbuf: 718 sk->sk_userlocks |= SOCK_SNDBUF_LOCK; 719 sk->sk_sndbuf = max_t(u32, val * 2, SOCK_MIN_SNDBUF); 720 /* Wake up sending tasks if we upped the value. */ 721 sk->sk_write_space(sk); 722 break; 723 724 case SO_SNDBUFFORCE: 725 if (!capable(CAP_NET_ADMIN)) { 726 ret = -EPERM; 727 break; 728 } 729 goto set_sndbuf; 730 731 case SO_RCVBUF: 732 /* Don't error on this BSD doesn't and if you think 733 * about it this is right. Otherwise apps have to 734 * play 'guess the biggest size' games. RCVBUF/SNDBUF 735 * are treated in BSD as hints 736 */ 737 val = min_t(u32, val, sysctl_rmem_max); 738set_rcvbuf: 739 sk->sk_userlocks |= SOCK_RCVBUF_LOCK; 740 /* 741 * We double it on the way in to account for 742 * "struct sk_buff" etc. overhead. Applications 743 * assume that the SO_RCVBUF setting they make will 744 * allow that much actual data to be received on that 745 * socket. 746 * 747 * Applications are unaware that "struct sk_buff" and 748 * other overheads allocate from the receive buffer 749 * during socket buffer allocation. 750 * 751 * And after considering the possible alternatives, 752 * returning the value we actually used in getsockopt 753 * is the most desirable behavior. 754 */ 755 sk->sk_rcvbuf = max_t(u32, val * 2, SOCK_MIN_RCVBUF); 756 break; 757 758 case SO_RCVBUFFORCE: 759 if (!capable(CAP_NET_ADMIN)) { 760 ret = -EPERM; 761 break; 762 } 763 goto set_rcvbuf; 764 765 case SO_KEEPALIVE: 766#ifdef CONFIG_INET 767 if (sk->sk_protocol == IPPROTO_TCP && 768 sk->sk_type == SOCK_STREAM) 769 tcp_set_keepalive(sk, valbool); 770#endif 771 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool); 772 break; 773 774 case SO_OOBINLINE: 775 sock_valbool_flag(sk, SOCK_URGINLINE, valbool); 776 break; 777 778 case SO_NO_CHECK: 779 sk->sk_no_check_tx = valbool; 780 break; 781 782 case SO_PRIORITY: 783 if ((val >= 0 && val <= 6) || 784 ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) 785 sk->sk_priority = val; 786 else 787 ret = -EPERM; 788 break; 789 790 case SO_LINGER: 791 if (optlen < sizeof(ling)) { 792 ret = -EINVAL; /* 1003.1g */ 793 break; 794 } 795 if (copy_from_user(&ling, optval, sizeof(ling))) { 796 ret = -EFAULT; 797 break; 798 } 799 if (!ling.l_onoff) 800 sock_reset_flag(sk, SOCK_LINGER); 801 else { 802#if (BITS_PER_LONG == 32) 803 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ) 804 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT; 805 else 806#endif 807 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ; 808 sock_set_flag(sk, SOCK_LINGER); 809 } 810 break; 811 812 case SO_BSDCOMPAT: 813 sock_warn_obsolete_bsdism("setsockopt"); 814 break; 815 816 case SO_PASSCRED: 817 if (valbool) 818 set_bit(SOCK_PASSCRED, &sock->flags); 819 else 820 clear_bit(SOCK_PASSCRED, &sock->flags); 821 break; 822 823 case SO_TIMESTAMP: 824 case SO_TIMESTAMPNS: 825 if (valbool) { 826 if (optname == SO_TIMESTAMP) 827 sock_reset_flag(sk, SOCK_RCVTSTAMPNS); 828 else 829 sock_set_flag(sk, SOCK_RCVTSTAMPNS); 830 sock_set_flag(sk, SOCK_RCVTSTAMP); 831 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 832 } else { 833 sock_reset_flag(sk, SOCK_RCVTSTAMP); 834 sock_reset_flag(sk, SOCK_RCVTSTAMPNS); 835 } 836 break; 837 838 case SO_TIMESTAMPING: 839 if (val & ~SOF_TIMESTAMPING_MASK) { 840 ret = -EINVAL; 841 break; 842 } 843 if (val & SOF_TIMESTAMPING_OPT_ID && 844 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) { 845 if (sk->sk_protocol == IPPROTO_TCP) { 846 if (sk->sk_state != TCP_ESTABLISHED) { 847 ret = -EINVAL; 848 break; 849 } 850 sk->sk_tskey = tcp_sk(sk)->snd_una; 851 } else { 852 sk->sk_tskey = 0; 853 } 854 } 855 sk->sk_tsflags = val; 856 if (val & SOF_TIMESTAMPING_RX_SOFTWARE) 857 sock_enable_timestamp(sk, 858 SOCK_TIMESTAMPING_RX_SOFTWARE); 859 else 860 sock_disable_timestamp(sk, 861 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE)); 862 break; 863 864 case SO_RCVLOWAT: 865 if (val < 0) 866 val = INT_MAX; 867 sk->sk_rcvlowat = val ? : 1; 868 break; 869 870 case SO_RCVTIMEO: 871 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen); 872 break; 873 874 case SO_SNDTIMEO: 875 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen); 876 break; 877 878 case SO_ATTACH_FILTER: 879 ret = -EINVAL; 880 if (optlen == sizeof(struct sock_fprog)) { 881 struct sock_fprog fprog; 882 883 ret = -EFAULT; 884 if (copy_from_user(&fprog, optval, sizeof(fprog))) 885 break; 886 887 ret = sk_attach_filter(&fprog, sk); 888 } 889 break; 890 891 case SO_DETACH_FILTER: 892 ret = sk_detach_filter(sk); 893 break; 894 895 case SO_LOCK_FILTER: 896 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool) 897 ret = -EPERM; 898 else 899 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool); 900 break; 901 902 case SO_PASSSEC: 903 if (valbool) 904 set_bit(SOCK_PASSSEC, &sock->flags); 905 else 906 clear_bit(SOCK_PASSSEC, &sock->flags); 907 break; 908 case SO_MARK: 909 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) 910 ret = -EPERM; 911 else 912 sk->sk_mark = val; 913 break; 914 915 /* We implement the SO_SNDLOWAT etc to 916 not be settable (1003.1g 5.3) */ 917 case SO_RXQ_OVFL: 918 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool); 919 break; 920 921 case SO_WIFI_STATUS: 922 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool); 923 break; 924 925 case SO_PEEK_OFF: 926 if (sock->ops->set_peek_off) 927 ret = sock->ops->set_peek_off(sk, val); 928 else 929 ret = -EOPNOTSUPP; 930 break; 931 932 case SO_NOFCS: 933 sock_valbool_flag(sk, SOCK_NOFCS, valbool); 934 break; 935 936 case SO_SELECT_ERR_QUEUE: 937 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool); 938 break; 939 940#ifdef CONFIG_NET_RX_BUSY_POLL 941 case SO_BUSY_POLL: 942 /* allow unprivileged users to decrease the value */ 943 if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN)) 944 ret = -EPERM; 945 else { 946 if (val < 0) 947 ret = -EINVAL; 948 else 949 sk->sk_ll_usec = val; 950 } 951 break; 952#endif 953 954 case SO_MAX_PACING_RATE: 955 sk->sk_max_pacing_rate = val; 956 sk->sk_pacing_rate = min(sk->sk_pacing_rate, 957 sk->sk_max_pacing_rate); 958 break; 959 960 default: 961 ret = -ENOPROTOOPT; 962 break; 963 } 964 release_sock(sk); 965 return ret; 966} 967EXPORT_SYMBOL(sock_setsockopt); 968 969 970static void cred_to_ucred(struct pid *pid, const struct cred *cred, 971 struct ucred *ucred) 972{ 973 ucred->pid = pid_vnr(pid); 974 ucred->uid = ucred->gid = -1; 975 if (cred) { 976 struct user_namespace *current_ns = current_user_ns(); 977 978 ucred->uid = from_kuid_munged(current_ns, cred->euid); 979 ucred->gid = from_kgid_munged(current_ns, cred->egid); 980 } 981} 982 983int sock_getsockopt(struct socket *sock, int level, int optname, 984 char __user *optval, int __user *optlen) 985{ 986 struct sock *sk = sock->sk; 987 988 union { 989 int val; 990 struct linger ling; 991 struct timeval tm; 992 } v; 993 994 int lv = sizeof(int); 995 int len; 996 997 if (get_user(len, optlen)) 998 return -EFAULT; 999 if (len < 0) 1000 return -EINVAL; 1001 1002 memset(&v, 0, sizeof(v)); 1003 1004 switch (optname) { 1005 case SO_DEBUG: 1006 v.val = sock_flag(sk, SOCK_DBG); 1007 break; 1008 1009 case SO_DONTROUTE: 1010 v.val = sock_flag(sk, SOCK_LOCALROUTE); 1011 break; 1012 1013 case SO_BROADCAST: 1014 v.val = sock_flag(sk, SOCK_BROADCAST); 1015 break; 1016 1017 case SO_SNDBUF: 1018 v.val = sk->sk_sndbuf; 1019 break; 1020 1021 case SO_RCVBUF: 1022 v.val = sk->sk_rcvbuf; 1023 break; 1024 1025 case SO_REUSEADDR: 1026 v.val = sk->sk_reuse; 1027 break; 1028 1029 case SO_REUSEPORT: 1030 v.val = sk->sk_reuseport; 1031 break; 1032 1033 case SO_KEEPALIVE: 1034 v.val = sock_flag(sk, SOCK_KEEPOPEN); 1035 break; 1036 1037 case SO_TYPE: 1038 v.val = sk->sk_type; 1039 break; 1040 1041 case SO_PROTOCOL: 1042 v.val = sk->sk_protocol; 1043 break; 1044 1045 case SO_DOMAIN: 1046 v.val = sk->sk_family; 1047 break; 1048 1049 case SO_ERROR: 1050 v.val = -sock_error(sk); 1051 if (v.val == 0) 1052 v.val = xchg(&sk->sk_err_soft, 0); 1053 break; 1054 1055 case SO_OOBINLINE: 1056 v.val = sock_flag(sk, SOCK_URGINLINE); 1057 break; 1058 1059 case SO_NO_CHECK: 1060 v.val = sk->sk_no_check_tx; 1061 break; 1062 1063 case SO_PRIORITY: 1064 v.val = sk->sk_priority; 1065 break; 1066 1067 case SO_LINGER: 1068 lv = sizeof(v.ling); 1069 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER); 1070 v.ling.l_linger = sk->sk_lingertime / HZ; 1071 break; 1072 1073 case SO_BSDCOMPAT: 1074 sock_warn_obsolete_bsdism("getsockopt"); 1075 break; 1076 1077 case SO_TIMESTAMP: 1078 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && 1079 !sock_flag(sk, SOCK_RCVTSTAMPNS); 1080 break; 1081 1082 case SO_TIMESTAMPNS: 1083 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS); 1084 break; 1085 1086 case SO_TIMESTAMPING: 1087 v.val = sk->sk_tsflags; 1088 break; 1089 1090 case SO_RCVTIMEO: 1091 lv = sizeof(struct timeval); 1092 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) { 1093 v.tm.tv_sec = 0; 1094 v.tm.tv_usec = 0; 1095 } else { 1096 v.tm.tv_sec = sk->sk_rcvtimeo / HZ; 1097 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ; 1098 } 1099 break; 1100 1101 case SO_SNDTIMEO: 1102 lv = sizeof(struct timeval); 1103 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) { 1104 v.tm.tv_sec = 0; 1105 v.tm.tv_usec = 0; 1106 } else { 1107 v.tm.tv_sec = sk->sk_sndtimeo / HZ; 1108 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ; 1109 } 1110 break; 1111 1112 case SO_RCVLOWAT: 1113 v.val = sk->sk_rcvlowat; 1114 break; 1115 1116 case SO_SNDLOWAT: 1117 v.val = 1; 1118 break; 1119 1120 case SO_PASSCRED: 1121 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags); 1122 break; 1123 1124 case SO_PEERCRED: 1125 { 1126 struct ucred peercred; 1127 if (len > sizeof(peercred)) 1128 len = sizeof(peercred); 1129 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred); 1130 if (copy_to_user(optval, &peercred, len)) 1131 return -EFAULT; 1132 goto lenout; 1133 } 1134 1135 case SO_PEERNAME: 1136 { 1137 char address[128]; 1138 1139 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2)) 1140 return -ENOTCONN; 1141 if (lv < len) 1142 return -EINVAL; 1143 if (copy_to_user(optval, address, len)) 1144 return -EFAULT; 1145 goto lenout; 1146 } 1147 1148 /* Dubious BSD thing... Probably nobody even uses it, but 1149 * the UNIX standard wants it for whatever reason... -DaveM 1150 */ 1151 case SO_ACCEPTCONN: 1152 v.val = sk->sk_state == TCP_LISTEN; 1153 break; 1154 1155 case SO_PASSSEC: 1156 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags); 1157 break; 1158 1159 case SO_PEERSEC: 1160 return security_socket_getpeersec_stream(sock, optval, optlen, len); 1161 1162 case SO_MARK: 1163 v.val = sk->sk_mark; 1164 break; 1165 1166 case SO_RXQ_OVFL: 1167 v.val = sock_flag(sk, SOCK_RXQ_OVFL); 1168 break; 1169 1170 case SO_WIFI_STATUS: 1171 v.val = sock_flag(sk, SOCK_WIFI_STATUS); 1172 break; 1173 1174 case SO_PEEK_OFF: 1175 if (!sock->ops->set_peek_off) 1176 return -EOPNOTSUPP; 1177 1178 v.val = sk->sk_peek_off; 1179 break; 1180 case SO_NOFCS: 1181 v.val = sock_flag(sk, SOCK_NOFCS); 1182 break; 1183 1184 case SO_BINDTODEVICE: 1185 return sock_getbindtodevice(sk, optval, optlen, len); 1186 1187 case SO_GET_FILTER: 1188 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len); 1189 if (len < 0) 1190 return len; 1191 1192 goto lenout; 1193 1194 case SO_LOCK_FILTER: 1195 v.val = sock_flag(sk, SOCK_FILTER_LOCKED); 1196 break; 1197 1198 case SO_BPF_EXTENSIONS: 1199 v.val = bpf_tell_extensions(); 1200 break; 1201 1202 case SO_SELECT_ERR_QUEUE: 1203 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE); 1204 break; 1205 1206#ifdef CONFIG_NET_RX_BUSY_POLL 1207 case SO_BUSY_POLL: 1208 v.val = sk->sk_ll_usec; 1209 break; 1210#endif 1211 1212 case SO_MAX_PACING_RATE: 1213 v.val = sk->sk_max_pacing_rate; 1214 break; 1215 1216 default: 1217 return -ENOPROTOOPT; 1218 } 1219 1220 if (len > lv) 1221 len = lv; 1222 if (copy_to_user(optval, &v, len)) 1223 return -EFAULT; 1224lenout: 1225 if (put_user(len, optlen)) 1226 return -EFAULT; 1227 return 0; 1228} 1229 1230/* 1231 * Initialize an sk_lock. 1232 * 1233 * (We also register the sk_lock with the lock validator.) 1234 */ 1235static inline void sock_lock_init(struct sock *sk) 1236{ 1237 sock_lock_init_class_and_name(sk, 1238 af_family_slock_key_strings[sk->sk_family], 1239 af_family_slock_keys + sk->sk_family, 1240 af_family_key_strings[sk->sk_family], 1241 af_family_keys + sk->sk_family); 1242} 1243 1244/* 1245 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet, 1246 * even temporarly, because of RCU lookups. sk_node should also be left as is. 1247 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end 1248 */ 1249static void sock_copy(struct sock *nsk, const struct sock *osk) 1250{ 1251#ifdef CONFIG_SECURITY_NETWORK 1252 void *sptr = nsk->sk_security; 1253#endif 1254 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin)); 1255 1256 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end, 1257 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end)); 1258 1259#ifdef CONFIG_SECURITY_NETWORK 1260 nsk->sk_security = sptr; 1261 security_sk_clone(osk, nsk); 1262#endif 1263} 1264 1265void sk_prot_clear_portaddr_nulls(struct sock *sk, int size) 1266{ 1267 unsigned long nulls1, nulls2; 1268 1269 nulls1 = offsetof(struct sock, __sk_common.skc_node.next); 1270 nulls2 = offsetof(struct sock, __sk_common.skc_portaddr_node.next); 1271 if (nulls1 > nulls2) 1272 swap(nulls1, nulls2); 1273 1274 if (nulls1 != 0) 1275 memset((char *)sk, 0, nulls1); 1276 memset((char *)sk + nulls1 + sizeof(void *), 0, 1277 nulls2 - nulls1 - sizeof(void *)); 1278 memset((char *)sk + nulls2 + sizeof(void *), 0, 1279 size - nulls2 - sizeof(void *)); 1280} 1281EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls); 1282 1283static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority, 1284 int family) 1285{ 1286 struct sock *sk; 1287 struct kmem_cache *slab; 1288 1289 slab = prot->slab; 1290 if (slab != NULL) { 1291 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO); 1292 if (!sk) 1293 return sk; 1294 if (priority & __GFP_ZERO) { 1295 if (prot->clear_sk) 1296 prot->clear_sk(sk, prot->obj_size); 1297 else 1298 sk_prot_clear_nulls(sk, prot->obj_size); 1299 } 1300 } else 1301 sk = kmalloc(prot->obj_size, priority); 1302 1303 if (sk != NULL) { 1304 kmemcheck_annotate_bitfield(sk, flags); 1305 1306 if (security_sk_alloc(sk, family, priority)) 1307 goto out_free; 1308 1309 if (!try_module_get(prot->owner)) 1310 goto out_free_sec; 1311 sk_tx_queue_clear(sk); 1312 } 1313 1314 return sk; 1315 1316out_free_sec: 1317 security_sk_free(sk); 1318out_free: 1319 if (slab != NULL) 1320 kmem_cache_free(slab, sk); 1321 else 1322 kfree(sk); 1323 return NULL; 1324} 1325 1326static void sk_prot_free(struct proto *prot, struct sock *sk) 1327{ 1328 struct kmem_cache *slab; 1329 struct module *owner; 1330 1331 owner = prot->owner; 1332 slab = prot->slab; 1333 1334 security_sk_free(sk); 1335 if (slab != NULL) 1336 kmem_cache_free(slab, sk); 1337 else 1338 kfree(sk); 1339 module_put(owner); 1340} 1341 1342#if IS_ENABLED(CONFIG_CGROUP_NET_PRIO) 1343void sock_update_netprioidx(struct sock *sk) 1344{ 1345 if (in_interrupt()) 1346 return; 1347 1348 sk->sk_cgrp_prioidx = task_netprioidx(current); 1349} 1350EXPORT_SYMBOL_GPL(sock_update_netprioidx); 1351#endif 1352 1353/** 1354 * sk_alloc - All socket objects are allocated here 1355 * @net: the applicable net namespace 1356 * @family: protocol family 1357 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 1358 * @prot: struct proto associated with this new sock instance 1359 */ 1360struct sock *sk_alloc(struct net *net, int family, gfp_t priority, 1361 struct proto *prot) 1362{ 1363 struct sock *sk; 1364 1365 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family); 1366 if (sk) { 1367 sk->sk_family = family; 1368 /* 1369 * See comment in struct sock definition to understand 1370 * why we need sk_prot_creator -acme 1371 */ 1372 sk->sk_prot = sk->sk_prot_creator = prot; 1373 sock_lock_init(sk); 1374 sock_net_set(sk, get_net(net)); 1375 atomic_set(&sk->sk_wmem_alloc, 1); 1376 1377 sock_update_classid(sk); 1378 sock_update_netprioidx(sk); 1379 } 1380 1381 return sk; 1382} 1383EXPORT_SYMBOL(sk_alloc); 1384 1385static void __sk_free(struct sock *sk) 1386{ 1387 struct sk_filter *filter; 1388 1389 if (sk->sk_destruct) 1390 sk->sk_destruct(sk); 1391 1392 filter = rcu_dereference_check(sk->sk_filter, 1393 atomic_read(&sk->sk_wmem_alloc) == 0); 1394 if (filter) { 1395 sk_filter_uncharge(sk, filter); 1396 RCU_INIT_POINTER(sk->sk_filter, NULL); 1397 } 1398 1399 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP); 1400 1401 if (atomic_read(&sk->sk_omem_alloc)) 1402 pr_debug("%s: optmem leakage (%d bytes) detected\n", 1403 __func__, atomic_read(&sk->sk_omem_alloc)); 1404 1405 if (sk->sk_peer_cred) 1406 put_cred(sk->sk_peer_cred); 1407 put_pid(sk->sk_peer_pid); 1408 put_net(sock_net(sk)); 1409 sk_prot_free(sk->sk_prot_creator, sk); 1410} 1411 1412void sk_free(struct sock *sk) 1413{ 1414 /* 1415 * We subtract one from sk_wmem_alloc and can know if 1416 * some packets are still in some tx queue. 1417 * If not null, sock_wfree() will call __sk_free(sk) later 1418 */ 1419 if (atomic_dec_and_test(&sk->sk_wmem_alloc)) 1420 __sk_free(sk); 1421} 1422EXPORT_SYMBOL(sk_free); 1423 1424/* 1425 * Last sock_put should drop reference to sk->sk_net. It has already 1426 * been dropped in sk_change_net. Taking reference to stopping namespace 1427 * is not an option. 1428 * Take reference to a socket to remove it from hash _alive_ and after that 1429 * destroy it in the context of init_net. 1430 */ 1431void sk_release_kernel(struct sock *sk) 1432{ 1433 if (sk == NULL || sk->sk_socket == NULL) 1434 return; 1435 1436 sock_hold(sk); 1437 sock_release(sk->sk_socket); 1438 release_net(sock_net(sk)); 1439 sock_net_set(sk, get_net(&init_net)); 1440 sock_put(sk); 1441} 1442EXPORT_SYMBOL(sk_release_kernel); 1443 1444static void sk_update_clone(const struct sock *sk, struct sock *newsk) 1445{ 1446 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) 1447 sock_update_memcg(newsk); 1448} 1449 1450/** 1451 * sk_clone_lock - clone a socket, and lock its clone 1452 * @sk: the socket to clone 1453 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 1454 * 1455 * Caller must unlock socket even in error path (bh_unlock_sock(newsk)) 1456 */ 1457struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority) 1458{ 1459 struct sock *newsk; 1460 bool is_charged = true; 1461 1462 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family); 1463 if (newsk != NULL) { 1464 struct sk_filter *filter; 1465 1466 sock_copy(newsk, sk); 1467 1468 /* SANITY */ 1469 get_net(sock_net(newsk)); 1470 sk_node_init(&newsk->sk_node); 1471 sock_lock_init(newsk); 1472 bh_lock_sock(newsk); 1473 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL; 1474 newsk->sk_backlog.len = 0; 1475 1476 atomic_set(&newsk->sk_rmem_alloc, 0); 1477 /* 1478 * sk_wmem_alloc set to one (see sk_free() and sock_wfree()) 1479 */ 1480 atomic_set(&newsk->sk_wmem_alloc, 1); 1481 atomic_set(&newsk->sk_omem_alloc, 0); 1482 skb_queue_head_init(&newsk->sk_receive_queue); 1483 skb_queue_head_init(&newsk->sk_write_queue); 1484#ifdef CONFIG_NET_DMA 1485 skb_queue_head_init(&newsk->sk_async_wait_queue); 1486#endif 1487 1488 spin_lock_init(&newsk->sk_dst_lock); 1489 rwlock_init(&newsk->sk_callback_lock); 1490 lockdep_set_class_and_name(&newsk->sk_callback_lock, 1491 af_callback_keys + newsk->sk_family, 1492 af_family_clock_key_strings[newsk->sk_family]); 1493 1494 newsk->sk_dst_cache = NULL; 1495 newsk->sk_wmem_queued = 0; 1496 newsk->sk_forward_alloc = 0; 1497 newsk->sk_send_head = NULL; 1498 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK; 1499 1500 sock_reset_flag(newsk, SOCK_DONE); 1501 skb_queue_head_init(&newsk->sk_error_queue); 1502 1503 filter = rcu_dereference_protected(newsk->sk_filter, 1); 1504 if (filter != NULL) 1505 /* though it's an empty new sock, the charging may fail 1506 * if sysctl_optmem_max was changed between creation of 1507 * original socket and cloning 1508 */ 1509 is_charged = sk_filter_charge(newsk, filter); 1510 1511 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk))) { 1512 /* It is still raw copy of parent, so invalidate 1513 * destructor and make plain sk_free() */ 1514 newsk->sk_destruct = NULL; 1515 bh_unlock_sock(newsk); 1516 sk_free(newsk); 1517 newsk = NULL; 1518 goto out; 1519 } 1520 1521 newsk->sk_err = 0; 1522 newsk->sk_priority = 0; 1523 /* 1524 * Before updating sk_refcnt, we must commit prior changes to memory 1525 * (Documentation/RCU/rculist_nulls.txt for details) 1526 */ 1527 smp_wmb(); 1528 atomic_set(&newsk->sk_refcnt, 2); 1529 1530 /* 1531 * Increment the counter in the same struct proto as the master 1532 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that 1533 * is the same as sk->sk_prot->socks, as this field was copied 1534 * with memcpy). 1535 * 1536 * This _changes_ the previous behaviour, where 1537 * tcp_create_openreq_child always was incrementing the 1538 * equivalent to tcp_prot->socks (inet_sock_nr), so this have 1539 * to be taken into account in all callers. -acme 1540 */ 1541 sk_refcnt_debug_inc(newsk); 1542 sk_set_socket(newsk, NULL); 1543 newsk->sk_wq = NULL; 1544 1545 sk_update_clone(sk, newsk); 1546 1547 if (newsk->sk_prot->sockets_allocated) 1548 sk_sockets_allocated_inc(newsk); 1549 1550 if (newsk->sk_flags & SK_FLAGS_TIMESTAMP) 1551 net_enable_timestamp(); 1552 } 1553out: 1554 return newsk; 1555} 1556EXPORT_SYMBOL_GPL(sk_clone_lock); 1557 1558void sk_setup_caps(struct sock *sk, struct dst_entry *dst) 1559{ 1560 __sk_dst_set(sk, dst); 1561 sk->sk_route_caps = dst->dev->features; 1562 if (sk->sk_route_caps & NETIF_F_GSO) 1563 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE; 1564 sk->sk_route_caps &= ~sk->sk_route_nocaps; 1565 if (sk_can_gso(sk)) { 1566 if (dst->header_len) { 1567 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 1568 } else { 1569 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM; 1570 sk->sk_gso_max_size = dst->dev->gso_max_size; 1571 sk->sk_gso_max_segs = dst->dev->gso_max_segs; 1572 } 1573 } 1574} 1575EXPORT_SYMBOL_GPL(sk_setup_caps); 1576 1577/* 1578 * Simple resource managers for sockets. 1579 */ 1580 1581 1582/* 1583 * Write buffer destructor automatically called from kfree_skb. 1584 */ 1585void sock_wfree(struct sk_buff *skb) 1586{ 1587 struct sock *sk = skb->sk; 1588 unsigned int len = skb->truesize; 1589 1590 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) { 1591 /* 1592 * Keep a reference on sk_wmem_alloc, this will be released 1593 * after sk_write_space() call 1594 */ 1595 atomic_sub(len - 1, &sk->sk_wmem_alloc); 1596 sk->sk_write_space(sk); 1597 len = 1; 1598 } 1599 /* 1600 * if sk_wmem_alloc reaches 0, we must finish what sk_free() 1601 * could not do because of in-flight packets 1602 */ 1603 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc)) 1604 __sk_free(sk); 1605} 1606EXPORT_SYMBOL(sock_wfree); 1607 1608void skb_orphan_partial(struct sk_buff *skb) 1609{ 1610 /* TCP stack sets skb->ooo_okay based on sk_wmem_alloc, 1611 * so we do not completely orphan skb, but transfert all 1612 * accounted bytes but one, to avoid unexpected reorders. 1613 */ 1614 if (skb->destructor == sock_wfree 1615#ifdef CONFIG_INET 1616 || skb->destructor == tcp_wfree 1617#endif 1618 ) { 1619 atomic_sub(skb->truesize - 1, &skb->sk->sk_wmem_alloc); 1620 skb->truesize = 1; 1621 } else { 1622 skb_orphan(skb); 1623 } 1624} 1625EXPORT_SYMBOL(skb_orphan_partial); 1626 1627/* 1628 * Read buffer destructor automatically called from kfree_skb. 1629 */ 1630void sock_rfree(struct sk_buff *skb) 1631{ 1632 struct sock *sk = skb->sk; 1633 unsigned int len = skb->truesize; 1634 1635 atomic_sub(len, &sk->sk_rmem_alloc); 1636 sk_mem_uncharge(sk, len); 1637} 1638EXPORT_SYMBOL(sock_rfree); 1639 1640void sock_edemux(struct sk_buff *skb) 1641{ 1642 struct sock *sk = skb->sk; 1643 1644#ifdef CONFIG_INET 1645 if (sk->sk_state == TCP_TIME_WAIT) 1646 inet_twsk_put(inet_twsk(sk)); 1647 else 1648#endif 1649 sock_put(sk); 1650} 1651EXPORT_SYMBOL(sock_edemux); 1652 1653kuid_t sock_i_uid(struct sock *sk) 1654{ 1655 kuid_t uid; 1656 1657 read_lock_bh(&sk->sk_callback_lock); 1658 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID; 1659 read_unlock_bh(&sk->sk_callback_lock); 1660 return uid; 1661} 1662EXPORT_SYMBOL(sock_i_uid); 1663 1664unsigned long sock_i_ino(struct sock *sk) 1665{ 1666 unsigned long ino; 1667 1668 read_lock_bh(&sk->sk_callback_lock); 1669 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0; 1670 read_unlock_bh(&sk->sk_callback_lock); 1671 return ino; 1672} 1673EXPORT_SYMBOL(sock_i_ino); 1674 1675/* 1676 * Allocate a skb from the socket's send buffer. 1677 */ 1678struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 1679 gfp_t priority) 1680{ 1681 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) { 1682 struct sk_buff *skb = alloc_skb(size, priority); 1683 if (skb) { 1684 skb_set_owner_w(skb, sk); 1685 return skb; 1686 } 1687 } 1688 return NULL; 1689} 1690EXPORT_SYMBOL(sock_wmalloc); 1691 1692/* 1693 * Allocate a memory block from the socket's option memory buffer. 1694 */ 1695void *sock_kmalloc(struct sock *sk, int size, gfp_t priority) 1696{ 1697 if ((unsigned int)size <= sysctl_optmem_max && 1698 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) { 1699 void *mem; 1700 /* First do the add, to avoid the race if kmalloc 1701 * might sleep. 1702 */ 1703 atomic_add(size, &sk->sk_omem_alloc); 1704 mem = kmalloc(size, priority); 1705 if (mem) 1706 return mem; 1707 atomic_sub(size, &sk->sk_omem_alloc); 1708 } 1709 return NULL; 1710} 1711EXPORT_SYMBOL(sock_kmalloc); 1712 1713/* 1714 * Free an option memory block. 1715 */ 1716void sock_kfree_s(struct sock *sk, void *mem, int size) 1717{ 1718 kfree(mem); 1719 atomic_sub(size, &sk->sk_omem_alloc); 1720} 1721EXPORT_SYMBOL(sock_kfree_s); 1722 1723/* It is almost wait_for_tcp_memory minus release_sock/lock_sock. 1724 I think, these locks should be removed for datagram sockets. 1725 */ 1726static long sock_wait_for_wmem(struct sock *sk, long timeo) 1727{ 1728 DEFINE_WAIT(wait); 1729 1730 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags); 1731 for (;;) { 1732 if (!timeo) 1733 break; 1734 if (signal_pending(current)) 1735 break; 1736 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 1737 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 1738 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) 1739 break; 1740 if (sk->sk_shutdown & SEND_SHUTDOWN) 1741 break; 1742 if (sk->sk_err) 1743 break; 1744 timeo = schedule_timeout(timeo); 1745 } 1746 finish_wait(sk_sleep(sk), &wait); 1747 return timeo; 1748} 1749 1750 1751/* 1752 * Generic send/receive buffer handlers 1753 */ 1754 1755struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 1756 unsigned long data_len, int noblock, 1757 int *errcode, int max_page_order) 1758{ 1759 struct sk_buff *skb = NULL; 1760 unsigned long chunk; 1761 gfp_t gfp_mask; 1762 long timeo; 1763 int err; 1764 int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT; 1765 struct page *page; 1766 int i; 1767 1768 err = -EMSGSIZE; 1769 if (npages > MAX_SKB_FRAGS) 1770 goto failure; 1771 1772 timeo = sock_sndtimeo(sk, noblock); 1773 while (!skb) { 1774 err = sock_error(sk); 1775 if (err != 0) 1776 goto failure; 1777 1778 err = -EPIPE; 1779 if (sk->sk_shutdown & SEND_SHUTDOWN) 1780 goto failure; 1781 1782 if (atomic_read(&sk->sk_wmem_alloc) >= sk->sk_sndbuf) { 1783 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags); 1784 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 1785 err = -EAGAIN; 1786 if (!timeo) 1787 goto failure; 1788 if (signal_pending(current)) 1789 goto interrupted; 1790 timeo = sock_wait_for_wmem(sk, timeo); 1791 continue; 1792 } 1793 1794 err = -ENOBUFS; 1795 gfp_mask = sk->sk_allocation; 1796 if (gfp_mask & __GFP_WAIT) 1797 gfp_mask |= __GFP_REPEAT; 1798 1799 skb = alloc_skb(header_len, gfp_mask); 1800 if (!skb) 1801 goto failure; 1802 1803 skb->truesize += data_len; 1804 1805 for (i = 0; npages > 0; i++) { 1806 int order = max_page_order; 1807 1808 while (order) { 1809 if (npages >= 1 << order) { 1810 page = alloc_pages(sk->sk_allocation | 1811 __GFP_COMP | 1812 __GFP_NOWARN | 1813 __GFP_NORETRY, 1814 order); 1815 if (page) 1816 goto fill_page; 1817 } 1818 order--; 1819 } 1820 page = alloc_page(sk->sk_allocation); 1821 if (!page) 1822 goto failure; 1823fill_page: 1824 chunk = min_t(unsigned long, data_len, 1825 PAGE_SIZE << order); 1826 skb_fill_page_desc(skb, i, page, 0, chunk); 1827 data_len -= chunk; 1828 npages -= 1 << order; 1829 } 1830 } 1831 1832 skb_set_owner_w(skb, sk); 1833 return skb; 1834 1835interrupted: 1836 err = sock_intr_errno(timeo); 1837failure: 1838 kfree_skb(skb); 1839 *errcode = err; 1840 return NULL; 1841} 1842EXPORT_SYMBOL(sock_alloc_send_pskb); 1843 1844struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size, 1845 int noblock, int *errcode) 1846{ 1847 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0); 1848} 1849EXPORT_SYMBOL(sock_alloc_send_skb); 1850 1851/* On 32bit arches, an skb frag is limited to 2^15 */ 1852#define SKB_FRAG_PAGE_ORDER get_order(32768) 1853 1854/** 1855 * skb_page_frag_refill - check that a page_frag contains enough room 1856 * @sz: minimum size of the fragment we want to get 1857 * @pfrag: pointer to page_frag 1858 * @prio: priority for memory allocation 1859 * 1860 * Note: While this allocator tries to use high order pages, there is 1861 * no guarantee that allocations succeed. Therefore, @sz MUST be 1862 * less or equal than PAGE_SIZE. 1863 */ 1864bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio) 1865{ 1866 int order; 1867 1868 if (pfrag->page) { 1869 if (atomic_read(&pfrag->page->_count) == 1) { 1870 pfrag->offset = 0; 1871 return true; 1872 } 1873 if (pfrag->offset + sz <= pfrag->size) 1874 return true; 1875 put_page(pfrag->page); 1876 } 1877 1878 order = SKB_FRAG_PAGE_ORDER; 1879 do { 1880 gfp_t gfp = prio; 1881 1882 if (order) 1883 gfp |= __GFP_COMP | __GFP_NOWARN | __GFP_NORETRY; 1884 pfrag->page = alloc_pages(gfp, order); 1885 if (likely(pfrag->page)) { 1886 pfrag->offset = 0; 1887 pfrag->size = PAGE_SIZE << order; 1888 return true; 1889 } 1890 } while (--order >= 0); 1891 1892 return false; 1893} 1894EXPORT_SYMBOL(skb_page_frag_refill); 1895 1896bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag) 1897{ 1898 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation))) 1899 return true; 1900 1901 sk_enter_memory_pressure(sk); 1902 sk_stream_moderate_sndbuf(sk); 1903 return false; 1904} 1905EXPORT_SYMBOL(sk_page_frag_refill); 1906 1907static void __lock_sock(struct sock *sk) 1908 __releases(&sk->sk_lock.slock) 1909 __acquires(&sk->sk_lock.slock) 1910{ 1911 DEFINE_WAIT(wait); 1912 1913 for (;;) { 1914 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait, 1915 TASK_UNINTERRUPTIBLE); 1916 spin_unlock_bh(&sk->sk_lock.slock); 1917 schedule(); 1918 spin_lock_bh(&sk->sk_lock.slock); 1919 if (!sock_owned_by_user(sk)) 1920 break; 1921 } 1922 finish_wait(&sk->sk_lock.wq, &wait); 1923} 1924 1925static void __release_sock(struct sock *sk) 1926 __releases(&sk->sk_lock.slock) 1927 __acquires(&sk->sk_lock.slock) 1928{ 1929 struct sk_buff *skb = sk->sk_backlog.head; 1930 1931 do { 1932 sk->sk_backlog.head = sk->sk_backlog.tail = NULL; 1933 bh_unlock_sock(sk); 1934 1935 do { 1936 struct sk_buff *next = skb->next; 1937 1938 prefetch(next); 1939 WARN_ON_ONCE(skb_dst_is_noref(skb)); 1940 skb->next = NULL; 1941 sk_backlog_rcv(sk, skb); 1942 1943 /* 1944 * We are in process context here with softirqs 1945 * disabled, use cond_resched_softirq() to preempt. 1946 * This is safe to do because we've taken the backlog 1947 * queue private: 1948 */ 1949 cond_resched_softirq(); 1950 1951 skb = next; 1952 } while (skb != NULL); 1953 1954 bh_lock_sock(sk); 1955 } while ((skb = sk->sk_backlog.head) != NULL); 1956 1957 /* 1958 * Doing the zeroing here guarantee we can not loop forever 1959 * while a wild producer attempts to flood us. 1960 */ 1961 sk->sk_backlog.len = 0; 1962} 1963 1964/** 1965 * sk_wait_data - wait for data to arrive at sk_receive_queue 1966 * @sk: sock to wait on 1967 * @timeo: for how long 1968 * 1969 * Now socket state including sk->sk_err is changed only under lock, 1970 * hence we may omit checks after joining wait queue. 1971 * We check receive queue before schedule() only as optimization; 1972 * it is very likely that release_sock() added new data. 1973 */ 1974int sk_wait_data(struct sock *sk, long *timeo) 1975{ 1976 int rc; 1977 DEFINE_WAIT(wait); 1978 1979 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 1980 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags); 1981 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue)); 1982 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags); 1983 finish_wait(sk_sleep(sk), &wait); 1984 return rc; 1985} 1986EXPORT_SYMBOL(sk_wait_data); 1987 1988/** 1989 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated 1990 * @sk: socket 1991 * @size: memory size to allocate 1992 * @kind: allocation type 1993 * 1994 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means 1995 * rmem allocation. This function assumes that protocols which have 1996 * memory_pressure use sk_wmem_queued as write buffer accounting. 1997 */ 1998int __sk_mem_schedule(struct sock *sk, int size, int kind) 1999{ 2000 struct proto *prot = sk->sk_prot; 2001 int amt = sk_mem_pages(size); 2002 long allocated; 2003 int parent_status = UNDER_LIMIT; 2004 2005 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM; 2006 2007 allocated = sk_memory_allocated_add(sk, amt, &parent_status); 2008 2009 /* Under limit. */ 2010 if (parent_status == UNDER_LIMIT && 2011 allocated <= sk_prot_mem_limits(sk, 0)) { 2012 sk_leave_memory_pressure(sk); 2013 return 1; 2014 } 2015 2016 /* Under pressure. (we or our parents) */ 2017 if ((parent_status > SOFT_LIMIT) || 2018 allocated > sk_prot_mem_limits(sk, 1)) 2019 sk_enter_memory_pressure(sk); 2020 2021 /* Over hard limit (we or our parents) */ 2022 if ((parent_status == OVER_LIMIT) || 2023 (allocated > sk_prot_mem_limits(sk, 2))) 2024 goto suppress_allocation; 2025 2026 /* guarantee minimum buffer size under pressure */ 2027 if (kind == SK_MEM_RECV) { 2028 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0]) 2029 return 1; 2030 2031 } else { /* SK_MEM_SEND */ 2032 if (sk->sk_type == SOCK_STREAM) { 2033 if (sk->sk_wmem_queued < prot->sysctl_wmem[0]) 2034 return 1; 2035 } else if (atomic_read(&sk->sk_wmem_alloc) < 2036 prot->sysctl_wmem[0]) 2037 return 1; 2038 } 2039 2040 if (sk_has_memory_pressure(sk)) { 2041 int alloc; 2042 2043 if (!sk_under_memory_pressure(sk)) 2044 return 1; 2045 alloc = sk_sockets_allocated_read_positive(sk); 2046 if (sk_prot_mem_limits(sk, 2) > alloc * 2047 sk_mem_pages(sk->sk_wmem_queued + 2048 atomic_read(&sk->sk_rmem_alloc) + 2049 sk->sk_forward_alloc)) 2050 return 1; 2051 } 2052 2053suppress_allocation: 2054 2055 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) { 2056 sk_stream_moderate_sndbuf(sk); 2057 2058 /* Fail only if socket is _under_ its sndbuf. 2059 * In this case we cannot block, so that we have to fail. 2060 */ 2061 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf) 2062 return 1; 2063 } 2064 2065 trace_sock_exceed_buf_limit(sk, prot, allocated); 2066 2067 /* Alas. Undo changes. */ 2068 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM; 2069 2070 sk_memory_allocated_sub(sk, amt); 2071 2072 return 0; 2073} 2074EXPORT_SYMBOL(__sk_mem_schedule); 2075 2076/** 2077 * __sk_reclaim - reclaim memory_allocated 2078 * @sk: socket 2079 */ 2080void __sk_mem_reclaim(struct sock *sk) 2081{ 2082 sk_memory_allocated_sub(sk, 2083 sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT); 2084 sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1; 2085 2086 if (sk_under_memory_pressure(sk) && 2087 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0))) 2088 sk_leave_memory_pressure(sk); 2089} 2090EXPORT_SYMBOL(__sk_mem_reclaim); 2091 2092 2093/* 2094 * Set of default routines for initialising struct proto_ops when 2095 * the protocol does not support a particular function. In certain 2096 * cases where it makes no sense for a protocol to have a "do nothing" 2097 * function, some default processing is provided. 2098 */ 2099 2100int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len) 2101{ 2102 return -EOPNOTSUPP; 2103} 2104EXPORT_SYMBOL(sock_no_bind); 2105 2106int sock_no_connect(struct socket *sock, struct sockaddr *saddr, 2107 int len, int flags) 2108{ 2109 return -EOPNOTSUPP; 2110} 2111EXPORT_SYMBOL(sock_no_connect); 2112 2113int sock_no_socketpair(struct socket *sock1, struct socket *sock2) 2114{ 2115 return -EOPNOTSUPP; 2116} 2117EXPORT_SYMBOL(sock_no_socketpair); 2118 2119int sock_no_accept(struct socket *sock, struct socket *newsock, int flags) 2120{ 2121 return -EOPNOTSUPP; 2122} 2123EXPORT_SYMBOL(sock_no_accept); 2124 2125int sock_no_getname(struct socket *sock, struct sockaddr *saddr, 2126 int *len, int peer) 2127{ 2128 return -EOPNOTSUPP; 2129} 2130EXPORT_SYMBOL(sock_no_getname); 2131 2132unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt) 2133{ 2134 return 0; 2135} 2136EXPORT_SYMBOL(sock_no_poll); 2137 2138int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 2139{ 2140 return -EOPNOTSUPP; 2141} 2142EXPORT_SYMBOL(sock_no_ioctl); 2143 2144int sock_no_listen(struct socket *sock, int backlog) 2145{ 2146 return -EOPNOTSUPP; 2147} 2148EXPORT_SYMBOL(sock_no_listen); 2149 2150int sock_no_shutdown(struct socket *sock, int how) 2151{ 2152 return -EOPNOTSUPP; 2153} 2154EXPORT_SYMBOL(sock_no_shutdown); 2155 2156int sock_no_setsockopt(struct socket *sock, int level, int optname, 2157 char __user *optval, unsigned int optlen) 2158{ 2159 return -EOPNOTSUPP; 2160} 2161EXPORT_SYMBOL(sock_no_setsockopt); 2162 2163int sock_no_getsockopt(struct socket *sock, int level, int optname, 2164 char __user *optval, int __user *optlen) 2165{ 2166 return -EOPNOTSUPP; 2167} 2168EXPORT_SYMBOL(sock_no_getsockopt); 2169 2170int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m, 2171 size_t len) 2172{ 2173 return -EOPNOTSUPP; 2174} 2175EXPORT_SYMBOL(sock_no_sendmsg); 2176 2177int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m, 2178 size_t len, int flags) 2179{ 2180 return -EOPNOTSUPP; 2181} 2182EXPORT_SYMBOL(sock_no_recvmsg); 2183 2184int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma) 2185{ 2186 /* Mirror missing mmap method error code */ 2187 return -ENODEV; 2188} 2189EXPORT_SYMBOL(sock_no_mmap); 2190 2191ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags) 2192{ 2193 ssize_t res; 2194 struct msghdr msg = {.msg_flags = flags}; 2195 struct kvec iov; 2196 char *kaddr = kmap(page); 2197 iov.iov_base = kaddr + offset; 2198 iov.iov_len = size; 2199 res = kernel_sendmsg(sock, &msg, &iov, 1, size); 2200 kunmap(page); 2201 return res; 2202} 2203EXPORT_SYMBOL(sock_no_sendpage); 2204 2205/* 2206 * Default Socket Callbacks 2207 */ 2208 2209static void sock_def_wakeup(struct sock *sk) 2210{ 2211 struct socket_wq *wq; 2212 2213 rcu_read_lock(); 2214 wq = rcu_dereference(sk->sk_wq); 2215 if (wq_has_sleeper(wq)) 2216 wake_up_interruptible_all(&wq->wait); 2217 rcu_read_unlock(); 2218} 2219 2220static void sock_def_error_report(struct sock *sk) 2221{ 2222 struct socket_wq *wq; 2223 2224 rcu_read_lock(); 2225 wq = rcu_dereference(sk->sk_wq); 2226 if (wq_has_sleeper(wq)) 2227 wake_up_interruptible_poll(&wq->wait, POLLERR); 2228 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR); 2229 rcu_read_unlock(); 2230} 2231 2232static void sock_def_readable(struct sock *sk) 2233{ 2234 struct socket_wq *wq; 2235 2236 rcu_read_lock(); 2237 wq = rcu_dereference(sk->sk_wq); 2238 if (wq_has_sleeper(wq)) 2239 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI | 2240 POLLRDNORM | POLLRDBAND); 2241 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN); 2242 rcu_read_unlock(); 2243} 2244 2245static void sock_def_write_space(struct sock *sk) 2246{ 2247 struct socket_wq *wq; 2248 2249 rcu_read_lock(); 2250 2251 /* Do not wake up a writer until he can make "significant" 2252 * progress. --DaveM 2253 */ 2254 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) { 2255 wq = rcu_dereference(sk->sk_wq); 2256 if (wq_has_sleeper(wq)) 2257 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT | 2258 POLLWRNORM | POLLWRBAND); 2259 2260 /* Should agree with poll, otherwise some programs break */ 2261 if (sock_writeable(sk)) 2262 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT); 2263 } 2264 2265 rcu_read_unlock(); 2266} 2267 2268static void sock_def_destruct(struct sock *sk) 2269{ 2270 kfree(sk->sk_protinfo); 2271} 2272 2273void sk_send_sigurg(struct sock *sk) 2274{ 2275 if (sk->sk_socket && sk->sk_socket->file) 2276 if (send_sigurg(&sk->sk_socket->file->f_owner)) 2277 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI); 2278} 2279EXPORT_SYMBOL(sk_send_sigurg); 2280 2281void sk_reset_timer(struct sock *sk, struct timer_list* timer, 2282 unsigned long expires) 2283{ 2284 if (!mod_timer(timer, expires)) 2285 sock_hold(sk); 2286} 2287EXPORT_SYMBOL(sk_reset_timer); 2288 2289void sk_stop_timer(struct sock *sk, struct timer_list* timer) 2290{ 2291 if (del_timer(timer)) 2292 __sock_put(sk); 2293} 2294EXPORT_SYMBOL(sk_stop_timer); 2295 2296void sock_init_data(struct socket *sock, struct sock *sk) 2297{ 2298 skb_queue_head_init(&sk->sk_receive_queue); 2299 skb_queue_head_init(&sk->sk_write_queue); 2300 skb_queue_head_init(&sk->sk_error_queue); 2301#ifdef CONFIG_NET_DMA 2302 skb_queue_head_init(&sk->sk_async_wait_queue); 2303#endif 2304 2305 sk->sk_send_head = NULL; 2306 2307 init_timer(&sk->sk_timer); 2308 2309 sk->sk_allocation = GFP_KERNEL; 2310 sk->sk_rcvbuf = sysctl_rmem_default; 2311 sk->sk_sndbuf = sysctl_wmem_default; 2312 sk->sk_state = TCP_CLOSE; 2313 sk_set_socket(sk, sock); 2314 2315 sock_set_flag(sk, SOCK_ZAPPED); 2316 2317 if (sock) { 2318 sk->sk_type = sock->type; 2319 sk->sk_wq = sock->wq; 2320 sock->sk = sk; 2321 } else 2322 sk->sk_wq = NULL; 2323 2324 spin_lock_init(&sk->sk_dst_lock); 2325 rwlock_init(&sk->sk_callback_lock); 2326 lockdep_set_class_and_name(&sk->sk_callback_lock, 2327 af_callback_keys + sk->sk_family, 2328 af_family_clock_key_strings[sk->sk_family]); 2329 2330 sk->sk_state_change = sock_def_wakeup; 2331 sk->sk_data_ready = sock_def_readable; 2332 sk->sk_write_space = sock_def_write_space; 2333 sk->sk_error_report = sock_def_error_report; 2334 sk->sk_destruct = sock_def_destruct; 2335 2336 sk->sk_frag.page = NULL; 2337 sk->sk_frag.offset = 0; 2338 sk->sk_peek_off = -1; 2339 2340 sk->sk_peer_pid = NULL; 2341 sk->sk_peer_cred = NULL; 2342 sk->sk_write_pending = 0; 2343 sk->sk_rcvlowat = 1; 2344 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; 2345 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT; 2346 2347 sk->sk_stamp = ktime_set(-1L, 0); 2348 2349#ifdef CONFIG_NET_RX_BUSY_POLL 2350 sk->sk_napi_id = 0; 2351 sk->sk_ll_usec = sysctl_net_busy_read; 2352#endif 2353 2354 sk->sk_max_pacing_rate = ~0U; 2355 sk->sk_pacing_rate = ~0U; 2356 /* 2357 * Before updating sk_refcnt, we must commit prior changes to memory 2358 * (Documentation/RCU/rculist_nulls.txt for details) 2359 */ 2360 smp_wmb(); 2361 atomic_set(&sk->sk_refcnt, 1); 2362 atomic_set(&sk->sk_drops, 0); 2363} 2364EXPORT_SYMBOL(sock_init_data); 2365 2366void lock_sock_nested(struct sock *sk, int subclass) 2367{ 2368 might_sleep(); 2369 spin_lock_bh(&sk->sk_lock.slock); 2370 if (sk->sk_lock.owned) 2371 __lock_sock(sk); 2372 sk->sk_lock.owned = 1; 2373 spin_unlock(&sk->sk_lock.slock); 2374 /* 2375 * The sk_lock has mutex_lock() semantics here: 2376 */ 2377 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_); 2378 local_bh_enable(); 2379} 2380EXPORT_SYMBOL(lock_sock_nested); 2381 2382void release_sock(struct sock *sk) 2383{ 2384 /* 2385 * The sk_lock has mutex_unlock() semantics: 2386 */ 2387 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_); 2388 2389 spin_lock_bh(&sk->sk_lock.slock); 2390 if (sk->sk_backlog.tail) 2391 __release_sock(sk); 2392 2393 /* Warning : release_cb() might need to release sk ownership, 2394 * ie call sock_release_ownership(sk) before us. 2395 */ 2396 if (sk->sk_prot->release_cb) 2397 sk->sk_prot->release_cb(sk); 2398 2399 sock_release_ownership(sk); 2400 if (waitqueue_active(&sk->sk_lock.wq)) 2401 wake_up(&sk->sk_lock.wq); 2402 spin_unlock_bh(&sk->sk_lock.slock); 2403} 2404EXPORT_SYMBOL(release_sock); 2405 2406/** 2407 * lock_sock_fast - fast version of lock_sock 2408 * @sk: socket 2409 * 2410 * This version should be used for very small section, where process wont block 2411 * return false if fast path is taken 2412 * sk_lock.slock locked, owned = 0, BH disabled 2413 * return true if slow path is taken 2414 * sk_lock.slock unlocked, owned = 1, BH enabled 2415 */ 2416bool lock_sock_fast(struct sock *sk) 2417{ 2418 might_sleep(); 2419 spin_lock_bh(&sk->sk_lock.slock); 2420 2421 if (!sk->sk_lock.owned) 2422 /* 2423 * Note : We must disable BH 2424 */ 2425 return false; 2426 2427 __lock_sock(sk); 2428 sk->sk_lock.owned = 1; 2429 spin_unlock(&sk->sk_lock.slock); 2430 /* 2431 * The sk_lock has mutex_lock() semantics here: 2432 */ 2433 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_); 2434 local_bh_enable(); 2435 return true; 2436} 2437EXPORT_SYMBOL(lock_sock_fast); 2438 2439int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp) 2440{ 2441 struct timeval tv; 2442 if (!sock_flag(sk, SOCK_TIMESTAMP)) 2443 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 2444 tv = ktime_to_timeval(sk->sk_stamp); 2445 if (tv.tv_sec == -1) 2446 return -ENOENT; 2447 if (tv.tv_sec == 0) { 2448 sk->sk_stamp = ktime_get_real(); 2449 tv = ktime_to_timeval(sk->sk_stamp); 2450 } 2451 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0; 2452} 2453EXPORT_SYMBOL(sock_get_timestamp); 2454 2455int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp) 2456{ 2457 struct timespec ts; 2458 if (!sock_flag(sk, SOCK_TIMESTAMP)) 2459 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 2460 ts = ktime_to_timespec(sk->sk_stamp); 2461 if (ts.tv_sec == -1) 2462 return -ENOENT; 2463 if (ts.tv_sec == 0) { 2464 sk->sk_stamp = ktime_get_real(); 2465 ts = ktime_to_timespec(sk->sk_stamp); 2466 } 2467 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0; 2468} 2469EXPORT_SYMBOL(sock_get_timestampns); 2470 2471void sock_enable_timestamp(struct sock *sk, int flag) 2472{ 2473 if (!sock_flag(sk, flag)) { 2474 unsigned long previous_flags = sk->sk_flags; 2475 2476 sock_set_flag(sk, flag); 2477 /* 2478 * we just set one of the two flags which require net 2479 * time stamping, but time stamping might have been on 2480 * already because of the other one 2481 */ 2482 if (!(previous_flags & SK_FLAGS_TIMESTAMP)) 2483 net_enable_timestamp(); 2484 } 2485} 2486 2487int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, 2488 int level, int type) 2489{ 2490 struct sock_exterr_skb *serr; 2491 struct sk_buff *skb, *skb2; 2492 int copied, err; 2493 2494 err = -EAGAIN; 2495 skb = skb_dequeue(&sk->sk_error_queue); 2496 if (skb == NULL) 2497 goto out; 2498 2499 copied = skb->len; 2500 if (copied > len) { 2501 msg->msg_flags |= MSG_TRUNC; 2502 copied = len; 2503 } 2504 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied); 2505 if (err) 2506 goto out_free_skb; 2507 2508 sock_recv_timestamp(msg, sk, skb); 2509 2510 serr = SKB_EXT_ERR(skb); 2511 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee); 2512 2513 msg->msg_flags |= MSG_ERRQUEUE; 2514 err = copied; 2515 2516 /* Reset and regenerate socket error */ 2517 spin_lock_bh(&sk->sk_error_queue.lock); 2518 sk->sk_err = 0; 2519 if ((skb2 = skb_peek(&sk->sk_error_queue)) != NULL) { 2520 sk->sk_err = SKB_EXT_ERR(skb2)->ee.ee_errno; 2521 spin_unlock_bh(&sk->sk_error_queue.lock); 2522 sk->sk_error_report(sk); 2523 } else 2524 spin_unlock_bh(&sk->sk_error_queue.lock); 2525 2526out_free_skb: 2527 kfree_skb(skb); 2528out: 2529 return err; 2530} 2531EXPORT_SYMBOL(sock_recv_errqueue); 2532 2533/* 2534 * Get a socket option on an socket. 2535 * 2536 * FIX: POSIX 1003.1g is very ambiguous here. It states that 2537 * asynchronous errors should be reported by getsockopt. We assume 2538 * this means if you specify SO_ERROR (otherwise whats the point of it). 2539 */ 2540int sock_common_getsockopt(struct socket *sock, int level, int optname, 2541 char __user *optval, int __user *optlen) 2542{ 2543 struct sock *sk = sock->sk; 2544 2545 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen); 2546} 2547EXPORT_SYMBOL(sock_common_getsockopt); 2548 2549#ifdef CONFIG_COMPAT 2550int compat_sock_common_getsockopt(struct socket *sock, int level, int optname, 2551 char __user *optval, int __user *optlen) 2552{ 2553 struct sock *sk = sock->sk; 2554 2555 if (sk->sk_prot->compat_getsockopt != NULL) 2556 return sk->sk_prot->compat_getsockopt(sk, level, optname, 2557 optval, optlen); 2558 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen); 2559} 2560EXPORT_SYMBOL(compat_sock_common_getsockopt); 2561#endif 2562 2563int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock, 2564 struct msghdr *msg, size_t size, int flags) 2565{ 2566 struct sock *sk = sock->sk; 2567 int addr_len = 0; 2568 int err; 2569 2570 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT, 2571 flags & ~MSG_DONTWAIT, &addr_len); 2572 if (err >= 0) 2573 msg->msg_namelen = addr_len; 2574 return err; 2575} 2576EXPORT_SYMBOL(sock_common_recvmsg); 2577 2578/* 2579 * Set socket options on an inet socket. 2580 */ 2581int sock_common_setsockopt(struct socket *sock, int level, int optname, 2582 char __user *optval, unsigned int optlen) 2583{ 2584 struct sock *sk = sock->sk; 2585 2586 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen); 2587} 2588EXPORT_SYMBOL(sock_common_setsockopt); 2589 2590#ifdef CONFIG_COMPAT 2591int compat_sock_common_setsockopt(struct socket *sock, int level, int optname, 2592 char __user *optval, unsigned int optlen) 2593{ 2594 struct sock *sk = sock->sk; 2595 2596 if (sk->sk_prot->compat_setsockopt != NULL) 2597 return sk->sk_prot->compat_setsockopt(sk, level, optname, 2598 optval, optlen); 2599 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen); 2600} 2601EXPORT_SYMBOL(compat_sock_common_setsockopt); 2602#endif 2603 2604void sk_common_release(struct sock *sk) 2605{ 2606 if (sk->sk_prot->destroy) 2607 sk->sk_prot->destroy(sk); 2608 2609 /* 2610 * Observation: when sock_common_release is called, processes have 2611 * no access to socket. But net still has. 2612 * Step one, detach it from networking: 2613 * 2614 * A. Remove from hash tables. 2615 */ 2616 2617 sk->sk_prot->unhash(sk); 2618 2619 /* 2620 * In this point socket cannot receive new packets, but it is possible 2621 * that some packets are in flight because some CPU runs receiver and 2622 * did hash table lookup before we unhashed socket. They will achieve 2623 * receive queue and will be purged by socket destructor. 2624 * 2625 * Also we still have packets pending on receive queue and probably, 2626 * our own packets waiting in device queues. sock_destroy will drain 2627 * receive queue, but transmitted packets will delay socket destruction 2628 * until the last reference will be released. 2629 */ 2630 2631 sock_orphan(sk); 2632 2633 xfrm_sk_free_policy(sk); 2634 2635 sk_refcnt_debug_release(sk); 2636 2637 if (sk->sk_frag.page) { 2638 put_page(sk->sk_frag.page); 2639 sk->sk_frag.page = NULL; 2640 } 2641 2642 sock_put(sk); 2643} 2644EXPORT_SYMBOL(sk_common_release); 2645 2646#ifdef CONFIG_PROC_FS 2647#define PROTO_INUSE_NR 64 /* should be enough for the first time */ 2648struct prot_inuse { 2649 int val[PROTO_INUSE_NR]; 2650}; 2651 2652static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR); 2653 2654#ifdef CONFIG_NET_NS 2655void sock_prot_inuse_add(struct net *net, struct proto *prot, int val) 2656{ 2657 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val); 2658} 2659EXPORT_SYMBOL_GPL(sock_prot_inuse_add); 2660 2661int sock_prot_inuse_get(struct net *net, struct proto *prot) 2662{ 2663 int cpu, idx = prot->inuse_idx; 2664 int res = 0; 2665 2666 for_each_possible_cpu(cpu) 2667 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx]; 2668 2669 return res >= 0 ? res : 0; 2670} 2671EXPORT_SYMBOL_GPL(sock_prot_inuse_get); 2672 2673static int __net_init sock_inuse_init_net(struct net *net) 2674{ 2675 net->core.inuse = alloc_percpu(struct prot_inuse); 2676 return net->core.inuse ? 0 : -ENOMEM; 2677} 2678 2679static void __net_exit sock_inuse_exit_net(struct net *net) 2680{ 2681 free_percpu(net->core.inuse); 2682} 2683 2684static struct pernet_operations net_inuse_ops = { 2685 .init = sock_inuse_init_net, 2686 .exit = sock_inuse_exit_net, 2687}; 2688 2689static __init int net_inuse_init(void) 2690{ 2691 if (register_pernet_subsys(&net_inuse_ops)) 2692 panic("Cannot initialize net inuse counters"); 2693 2694 return 0; 2695} 2696 2697core_initcall(net_inuse_init); 2698#else 2699static DEFINE_PER_CPU(struct prot_inuse, prot_inuse); 2700 2701void sock_prot_inuse_add(struct net *net, struct proto *prot, int val) 2702{ 2703 __this_cpu_add(prot_inuse.val[prot->inuse_idx], val); 2704} 2705EXPORT_SYMBOL_GPL(sock_prot_inuse_add); 2706 2707int sock_prot_inuse_get(struct net *net, struct proto *prot) 2708{ 2709 int cpu, idx = prot->inuse_idx; 2710 int res = 0; 2711 2712 for_each_possible_cpu(cpu) 2713 res += per_cpu(prot_inuse, cpu).val[idx]; 2714 2715 return res >= 0 ? res : 0; 2716} 2717EXPORT_SYMBOL_GPL(sock_prot_inuse_get); 2718#endif 2719 2720static void assign_proto_idx(struct proto *prot) 2721{ 2722 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR); 2723 2724 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) { 2725 pr_err("PROTO_INUSE_NR exhausted\n"); 2726 return; 2727 } 2728 2729 set_bit(prot->inuse_idx, proto_inuse_idx); 2730} 2731 2732static void release_proto_idx(struct proto *prot) 2733{ 2734 if (prot->inuse_idx != PROTO_INUSE_NR - 1) 2735 clear_bit(prot->inuse_idx, proto_inuse_idx); 2736} 2737#else 2738static inline void assign_proto_idx(struct proto *prot) 2739{ 2740} 2741 2742static inline void release_proto_idx(struct proto *prot) 2743{ 2744} 2745#endif 2746 2747int proto_register(struct proto *prot, int alloc_slab) 2748{ 2749 if (alloc_slab) { 2750 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0, 2751 SLAB_HWCACHE_ALIGN | prot->slab_flags, 2752 NULL); 2753 2754 if (prot->slab == NULL) { 2755 pr_crit("%s: Can't create sock SLAB cache!\n", 2756 prot->name); 2757 goto out; 2758 } 2759 2760 if (prot->rsk_prot != NULL) { 2761 prot->rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", prot->name); 2762 if (prot->rsk_prot->slab_name == NULL) 2763 goto out_free_sock_slab; 2764 2765 prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name, 2766 prot->rsk_prot->obj_size, 0, 2767 SLAB_HWCACHE_ALIGN, NULL); 2768 2769 if (prot->rsk_prot->slab == NULL) { 2770 pr_crit("%s: Can't create request sock SLAB cache!\n", 2771 prot->name); 2772 goto out_free_request_sock_slab_name; 2773 } 2774 } 2775 2776 if (prot->twsk_prot != NULL) { 2777 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name); 2778 2779 if (prot->twsk_prot->twsk_slab_name == NULL) 2780 goto out_free_request_sock_slab; 2781 2782 prot->twsk_prot->twsk_slab = 2783 kmem_cache_create(prot->twsk_prot->twsk_slab_name, 2784 prot->twsk_prot->twsk_obj_size, 2785 0, 2786 SLAB_HWCACHE_ALIGN | 2787 prot->slab_flags, 2788 NULL); 2789 if (prot->twsk_prot->twsk_slab == NULL) 2790 goto out_free_timewait_sock_slab_name; 2791 } 2792 } 2793 2794 mutex_lock(&proto_list_mutex); 2795 list_add(&prot->node, &proto_list); 2796 assign_proto_idx(prot); 2797 mutex_unlock(&proto_list_mutex); 2798 return 0; 2799 2800out_free_timewait_sock_slab_name: 2801 kfree(prot->twsk_prot->twsk_slab_name); 2802out_free_request_sock_slab: 2803 if (prot->rsk_prot && prot->rsk_prot->slab) { 2804 kmem_cache_destroy(prot->rsk_prot->slab); 2805 prot->rsk_prot->slab = NULL; 2806 } 2807out_free_request_sock_slab_name: 2808 if (prot->rsk_prot) 2809 kfree(prot->rsk_prot->slab_name); 2810out_free_sock_slab: 2811 kmem_cache_destroy(prot->slab); 2812 prot->slab = NULL; 2813out: 2814 return -ENOBUFS; 2815} 2816EXPORT_SYMBOL(proto_register); 2817 2818void proto_unregister(struct proto *prot) 2819{ 2820 mutex_lock(&proto_list_mutex); 2821 release_proto_idx(prot); 2822 list_del(&prot->node); 2823 mutex_unlock(&proto_list_mutex); 2824 2825 if (prot->slab != NULL) { 2826 kmem_cache_destroy(prot->slab); 2827 prot->slab = NULL; 2828 } 2829 2830 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) { 2831 kmem_cache_destroy(prot->rsk_prot->slab); 2832 kfree(prot->rsk_prot->slab_name); 2833 prot->rsk_prot->slab = NULL; 2834 } 2835 2836 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) { 2837 kmem_cache_destroy(prot->twsk_prot->twsk_slab); 2838 kfree(prot->twsk_prot->twsk_slab_name); 2839 prot->twsk_prot->twsk_slab = NULL; 2840 } 2841} 2842EXPORT_SYMBOL(proto_unregister); 2843 2844#ifdef CONFIG_PROC_FS 2845static void *proto_seq_start(struct seq_file *seq, loff_t *pos) 2846 __acquires(proto_list_mutex) 2847{ 2848 mutex_lock(&proto_list_mutex); 2849 return seq_list_start_head(&proto_list, *pos); 2850} 2851 2852static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos) 2853{ 2854 return seq_list_next(v, &proto_list, pos); 2855} 2856 2857static void proto_seq_stop(struct seq_file *seq, void *v) 2858 __releases(proto_list_mutex) 2859{ 2860 mutex_unlock(&proto_list_mutex); 2861} 2862 2863static char proto_method_implemented(const void *method) 2864{ 2865 return method == NULL ? 'n' : 'y'; 2866} 2867static long sock_prot_memory_allocated(struct proto *proto) 2868{ 2869 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L; 2870} 2871 2872static char *sock_prot_memory_pressure(struct proto *proto) 2873{ 2874 return proto->memory_pressure != NULL ? 2875 proto_memory_pressure(proto) ? "yes" : "no" : "NI"; 2876} 2877 2878static void proto_seq_printf(struct seq_file *seq, struct proto *proto) 2879{ 2880 2881 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s " 2882 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n", 2883 proto->name, 2884 proto->obj_size, 2885 sock_prot_inuse_get(seq_file_net(seq), proto), 2886 sock_prot_memory_allocated(proto), 2887 sock_prot_memory_pressure(proto), 2888 proto->max_header, 2889 proto->slab == NULL ? "no" : "yes", 2890 module_name(proto->owner), 2891 proto_method_implemented(proto->close), 2892 proto_method_implemented(proto->connect), 2893 proto_method_implemented(proto->disconnect), 2894 proto_method_implemented(proto->accept), 2895 proto_method_implemented(proto->ioctl), 2896 proto_method_implemented(proto->init), 2897 proto_method_implemented(proto->destroy), 2898 proto_method_implemented(proto->shutdown), 2899 proto_method_implemented(proto->setsockopt), 2900 proto_method_implemented(proto->getsockopt), 2901 proto_method_implemented(proto->sendmsg), 2902 proto_method_implemented(proto->recvmsg), 2903 proto_method_implemented(proto->sendpage), 2904 proto_method_implemented(proto->bind), 2905 proto_method_implemented(proto->backlog_rcv), 2906 proto_method_implemented(proto->hash), 2907 proto_method_implemented(proto->unhash), 2908 proto_method_implemented(proto->get_port), 2909 proto_method_implemented(proto->enter_memory_pressure)); 2910} 2911 2912static int proto_seq_show(struct seq_file *seq, void *v) 2913{ 2914 if (v == &proto_list) 2915 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s", 2916 "protocol", 2917 "size", 2918 "sockets", 2919 "memory", 2920 "press", 2921 "maxhdr", 2922 "slab", 2923 "module", 2924 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n"); 2925 else 2926 proto_seq_printf(seq, list_entry(v, struct proto, node)); 2927 return 0; 2928} 2929 2930static const struct seq_operations proto_seq_ops = { 2931 .start = proto_seq_start, 2932 .next = proto_seq_next, 2933 .stop = proto_seq_stop, 2934 .show = proto_seq_show, 2935}; 2936 2937static int proto_seq_open(struct inode *inode, struct file *file) 2938{ 2939 return seq_open_net(inode, file, &proto_seq_ops, 2940 sizeof(struct seq_net_private)); 2941} 2942 2943static const struct file_operations proto_seq_fops = { 2944 .owner = THIS_MODULE, 2945 .open = proto_seq_open, 2946 .read = seq_read, 2947 .llseek = seq_lseek, 2948 .release = seq_release_net, 2949}; 2950 2951static __net_init int proto_init_net(struct net *net) 2952{ 2953 if (!proc_create("protocols", S_IRUGO, net->proc_net, &proto_seq_fops)) 2954 return -ENOMEM; 2955 2956 return 0; 2957} 2958 2959static __net_exit void proto_exit_net(struct net *net) 2960{ 2961 remove_proc_entry("protocols", net->proc_net); 2962} 2963 2964 2965static __net_initdata struct pernet_operations proto_net_ops = { 2966 .init = proto_init_net, 2967 .exit = proto_exit_net, 2968}; 2969 2970static int __init proto_init(void) 2971{ 2972 return register_pernet_subsys(&proto_net_ops); 2973} 2974 2975subsys_initcall(proto_init); 2976 2977#endif /* PROC_FS */ 2978