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 #include <linux/sock_diag.h> 135 136 #include <linux/filter.h> 137 138 #include <trace/events/sock.h> 139 140 #ifdef CONFIG_INET 141 #include <net/tcp.h> 142 #endif 143 144 #include <net/busy_poll.h> 145 146 static DEFINE_MUTEX(proto_list_mutex); 147 static LIST_HEAD(proto_list); 148 149 /** 150 * sk_ns_capable - General socket capability test 151 * @sk: Socket to use a capability on or through 152 * @user_ns: The user namespace of the capability to use 153 * @cap: The capability to use 154 * 155 * Test to see if the opener of the socket had when the socket was 156 * created and the current process has the capability @cap in the user 157 * namespace @user_ns. 158 */ 159 bool sk_ns_capable(const struct sock *sk, 160 struct user_namespace *user_ns, int cap) 161 { 162 return file_ns_capable(sk->sk_socket->file, user_ns, cap) && 163 ns_capable(user_ns, cap); 164 } 165 EXPORT_SYMBOL(sk_ns_capable); 166 167 /** 168 * sk_capable - Socket global capability test 169 * @sk: Socket to use a capability on or through 170 * @cap: The global capability to use 171 * 172 * Test to see if the opener of the socket had when the socket was 173 * created and the current process has the capability @cap in all user 174 * namespaces. 175 */ 176 bool sk_capable(const struct sock *sk, int cap) 177 { 178 return sk_ns_capable(sk, &init_user_ns, cap); 179 } 180 EXPORT_SYMBOL(sk_capable); 181 182 /** 183 * sk_net_capable - Network namespace socket capability test 184 * @sk: Socket to use a capability on or through 185 * @cap: The capability to use 186 * 187 * Test to see if the opener of the socket had when the socket was created 188 * and the current process has the capability @cap over the network namespace 189 * the socket is a member of. 190 */ 191 bool sk_net_capable(const struct sock *sk, int cap) 192 { 193 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap); 194 } 195 EXPORT_SYMBOL(sk_net_capable); 196 197 198 #ifdef CONFIG_MEMCG_KMEM 199 int mem_cgroup_sockets_init(struct mem_cgroup *memcg, struct cgroup_subsys *ss) 200 { 201 struct proto *proto; 202 int ret = 0; 203 204 mutex_lock(&proto_list_mutex); 205 list_for_each_entry(proto, &proto_list, node) { 206 if (proto->init_cgroup) { 207 ret = proto->init_cgroup(memcg, ss); 208 if (ret) 209 goto out; 210 } 211 } 212 213 mutex_unlock(&proto_list_mutex); 214 return ret; 215 out: 216 list_for_each_entry_continue_reverse(proto, &proto_list, node) 217 if (proto->destroy_cgroup) 218 proto->destroy_cgroup(memcg); 219 mutex_unlock(&proto_list_mutex); 220 return ret; 221 } 222 223 void mem_cgroup_sockets_destroy(struct mem_cgroup *memcg) 224 { 225 struct proto *proto; 226 227 mutex_lock(&proto_list_mutex); 228 list_for_each_entry_reverse(proto, &proto_list, node) 229 if (proto->destroy_cgroup) 230 proto->destroy_cgroup(memcg); 231 mutex_unlock(&proto_list_mutex); 232 } 233 #endif 234 235 /* 236 * Each address family might have different locking rules, so we have 237 * one slock key per address family: 238 */ 239 static struct lock_class_key af_family_keys[AF_MAX]; 240 static struct lock_class_key af_family_slock_keys[AF_MAX]; 241 242 #if defined(CONFIG_MEMCG_KMEM) 243 struct static_key memcg_socket_limit_enabled; 244 EXPORT_SYMBOL(memcg_socket_limit_enabled); 245 #endif 246 247 /* 248 * Make lock validator output more readable. (we pre-construct these 249 * strings build-time, so that runtime initialization of socket 250 * locks is fast): 251 */ 252 static const char *const af_family_key_strings[AF_MAX+1] = { 253 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" , 254 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK", 255 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" , 256 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" , 257 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" , 258 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" , 259 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" , 260 "sk_lock-AF_RDS" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" , 261 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" , 262 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" , 263 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" , 264 "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN" , "sk_lock-AF_PHONET" , 265 "sk_lock-AF_IEEE802154", "sk_lock-AF_CAIF" , "sk_lock-AF_ALG" , 266 "sk_lock-AF_NFC" , "sk_lock-AF_VSOCK" , "sk_lock-AF_MAX" 267 }; 268 static const char *const af_family_slock_key_strings[AF_MAX+1] = { 269 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" , 270 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK", 271 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" , 272 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" , 273 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" , 274 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" , 275 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" , 276 "slock-AF_RDS" , "slock-AF_SNA" , "slock-AF_IRDA" , 277 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" , 278 "slock-27" , "slock-28" , "slock-AF_CAN" , 279 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" , 280 "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" , 281 "slock-AF_IEEE802154", "slock-AF_CAIF" , "slock-AF_ALG" , 282 "slock-AF_NFC" , "slock-AF_VSOCK" ,"slock-AF_MAX" 283 }; 284 static const char *const af_family_clock_key_strings[AF_MAX+1] = { 285 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" , 286 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK", 287 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" , 288 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" , 289 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" , 290 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" , 291 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" , 292 "clock-AF_RDS" , "clock-AF_SNA" , "clock-AF_IRDA" , 293 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" , 294 "clock-27" , "clock-28" , "clock-AF_CAN" , 295 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" , 296 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" , 297 "clock-AF_IEEE802154", "clock-AF_CAIF" , "clock-AF_ALG" , 298 "clock-AF_NFC" , "clock-AF_VSOCK" , "clock-AF_MAX" 299 }; 300 301 /* 302 * sk_callback_lock locking rules are per-address-family, 303 * so split the lock classes by using a per-AF key: 304 */ 305 static struct lock_class_key af_callback_keys[AF_MAX]; 306 307 /* Take into consideration the size of the struct sk_buff overhead in the 308 * determination of these values, since that is non-constant across 309 * platforms. This makes socket queueing behavior and performance 310 * not depend upon such differences. 311 */ 312 #define _SK_MEM_PACKETS 256 313 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256) 314 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 315 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 316 317 /* Run time adjustable parameters. */ 318 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX; 319 EXPORT_SYMBOL(sysctl_wmem_max); 320 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX; 321 EXPORT_SYMBOL(sysctl_rmem_max); 322 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX; 323 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX; 324 325 /* Maximal space eaten by iovec or ancillary data plus some space */ 326 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512); 327 EXPORT_SYMBOL(sysctl_optmem_max); 328 329 int sysctl_tstamp_allow_data __read_mostly = 1; 330 331 struct static_key memalloc_socks = STATIC_KEY_INIT_FALSE; 332 EXPORT_SYMBOL_GPL(memalloc_socks); 333 334 /** 335 * sk_set_memalloc - sets %SOCK_MEMALLOC 336 * @sk: socket to set it on 337 * 338 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves. 339 * It's the responsibility of the admin to adjust min_free_kbytes 340 * to meet the requirements 341 */ 342 void sk_set_memalloc(struct sock *sk) 343 { 344 sock_set_flag(sk, SOCK_MEMALLOC); 345 sk->sk_allocation |= __GFP_MEMALLOC; 346 static_key_slow_inc(&memalloc_socks); 347 } 348 EXPORT_SYMBOL_GPL(sk_set_memalloc); 349 350 void sk_clear_memalloc(struct sock *sk) 351 { 352 sock_reset_flag(sk, SOCK_MEMALLOC); 353 sk->sk_allocation &= ~__GFP_MEMALLOC; 354 static_key_slow_dec(&memalloc_socks); 355 356 /* 357 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward 358 * progress of swapping. SOCK_MEMALLOC may be cleared while 359 * it has rmem allocations due to the last swapfile being deactivated 360 * but there is a risk that the socket is unusable due to exceeding 361 * the rmem limits. Reclaim the reserves and obey rmem limits again. 362 */ 363 sk_mem_reclaim(sk); 364 } 365 EXPORT_SYMBOL_GPL(sk_clear_memalloc); 366 367 int __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 } 381 EXPORT_SYMBOL(__sk_backlog_rcv); 382 383 static 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 } 412 413 static 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)) 426 427 static 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 } 435 436 437 int 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 sock_skb_set_dropcount(sk, skb); 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 } 475 EXPORT_SYMBOL(sock_queue_rcv_skb); 476 477 int 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); 510 out: 511 sock_put(sk); 512 return rc; 513 discard_and_relse: 514 kfree_skb(skb); 515 goto out; 516 } 517 EXPORT_SYMBOL(sk_receive_skb); 518 519 struct 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 } 532 EXPORT_SYMBOL(__sk_dst_check); 533 534 struct 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 } 546 EXPORT_SYMBOL(sk_dst_check); 547 548 static 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; 599 600 out: 601 #endif 602 603 return ret; 604 } 605 606 static 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; 632 633 zero: 634 ret = -EFAULT; 635 if (put_user(len, optlen)) 636 goto out; 637 638 ret = 0; 639 640 out: 641 #endif 642 643 return ret; 644 } 645 646 static 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 bool sk_mc_loop(struct sock *sk) 655 { 656 if (dev_recursion_level()) 657 return false; 658 if (!sk) 659 return true; 660 switch (sk->sk_family) { 661 case AF_INET: 662 return inet_sk(sk)->mc_loop; 663 #if IS_ENABLED(CONFIG_IPV6) 664 case AF_INET6: 665 return inet6_sk(sk)->mc_loop; 666 #endif 667 } 668 WARN_ON(1); 669 return true; 670 } 671 EXPORT_SYMBOL(sk_mc_loop); 672 673 /* 674 * This is meant for all protocols to use and covers goings on 675 * at the socket level. Everything here is generic. 676 */ 677 678 int sock_setsockopt(struct socket *sock, int level, int optname, 679 char __user *optval, unsigned int optlen) 680 { 681 struct sock *sk = sock->sk; 682 int val; 683 int valbool; 684 struct linger ling; 685 int ret = 0; 686 687 /* 688 * Options without arguments 689 */ 690 691 if (optname == SO_BINDTODEVICE) 692 return sock_setbindtodevice(sk, optval, optlen); 693 694 if (optlen < sizeof(int)) 695 return -EINVAL; 696 697 if (get_user(val, (int __user *)optval)) 698 return -EFAULT; 699 700 valbool = val ? 1 : 0; 701 702 lock_sock(sk); 703 704 switch (optname) { 705 case SO_DEBUG: 706 if (val && !capable(CAP_NET_ADMIN)) 707 ret = -EACCES; 708 else 709 sock_valbool_flag(sk, SOCK_DBG, valbool); 710 break; 711 case SO_REUSEADDR: 712 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE); 713 break; 714 case SO_REUSEPORT: 715 sk->sk_reuseport = valbool; 716 break; 717 case SO_TYPE: 718 case SO_PROTOCOL: 719 case SO_DOMAIN: 720 case SO_ERROR: 721 ret = -ENOPROTOOPT; 722 break; 723 case SO_DONTROUTE: 724 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool); 725 break; 726 case SO_BROADCAST: 727 sock_valbool_flag(sk, SOCK_BROADCAST, valbool); 728 break; 729 case SO_SNDBUF: 730 /* Don't error on this BSD doesn't and if you think 731 * about it this is right. Otherwise apps have to 732 * play 'guess the biggest size' games. RCVBUF/SNDBUF 733 * are treated in BSD as hints 734 */ 735 val = min_t(u32, val, sysctl_wmem_max); 736 set_sndbuf: 737 sk->sk_userlocks |= SOCK_SNDBUF_LOCK; 738 sk->sk_sndbuf = max_t(u32, val * 2, SOCK_MIN_SNDBUF); 739 /* Wake up sending tasks if we upped the value. */ 740 sk->sk_write_space(sk); 741 break; 742 743 case SO_SNDBUFFORCE: 744 if (!capable(CAP_NET_ADMIN)) { 745 ret = -EPERM; 746 break; 747 } 748 goto set_sndbuf; 749 750 case SO_RCVBUF: 751 /* Don't error on this BSD doesn't and if you think 752 * about it this is right. Otherwise apps have to 753 * play 'guess the biggest size' games. RCVBUF/SNDBUF 754 * are treated in BSD as hints 755 */ 756 val = min_t(u32, val, sysctl_rmem_max); 757 set_rcvbuf: 758 sk->sk_userlocks |= SOCK_RCVBUF_LOCK; 759 /* 760 * We double it on the way in to account for 761 * "struct sk_buff" etc. overhead. Applications 762 * assume that the SO_RCVBUF setting they make will 763 * allow that much actual data to be received on that 764 * socket. 765 * 766 * Applications are unaware that "struct sk_buff" and 767 * other overheads allocate from the receive buffer 768 * during socket buffer allocation. 769 * 770 * And after considering the possible alternatives, 771 * returning the value we actually used in getsockopt 772 * is the most desirable behavior. 773 */ 774 sk->sk_rcvbuf = max_t(u32, val * 2, SOCK_MIN_RCVBUF); 775 break; 776 777 case SO_RCVBUFFORCE: 778 if (!capable(CAP_NET_ADMIN)) { 779 ret = -EPERM; 780 break; 781 } 782 goto set_rcvbuf; 783 784 case SO_KEEPALIVE: 785 #ifdef CONFIG_INET 786 if (sk->sk_protocol == IPPROTO_TCP && 787 sk->sk_type == SOCK_STREAM) 788 tcp_set_keepalive(sk, valbool); 789 #endif 790 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool); 791 break; 792 793 case SO_OOBINLINE: 794 sock_valbool_flag(sk, SOCK_URGINLINE, valbool); 795 break; 796 797 case SO_NO_CHECK: 798 sk->sk_no_check_tx = valbool; 799 break; 800 801 case SO_PRIORITY: 802 if ((val >= 0 && val <= 6) || 803 ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) 804 sk->sk_priority = val; 805 else 806 ret = -EPERM; 807 break; 808 809 case SO_LINGER: 810 if (optlen < sizeof(ling)) { 811 ret = -EINVAL; /* 1003.1g */ 812 break; 813 } 814 if (copy_from_user(&ling, optval, sizeof(ling))) { 815 ret = -EFAULT; 816 break; 817 } 818 if (!ling.l_onoff) 819 sock_reset_flag(sk, SOCK_LINGER); 820 else { 821 #if (BITS_PER_LONG == 32) 822 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ) 823 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT; 824 else 825 #endif 826 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ; 827 sock_set_flag(sk, SOCK_LINGER); 828 } 829 break; 830 831 case SO_BSDCOMPAT: 832 sock_warn_obsolete_bsdism("setsockopt"); 833 break; 834 835 case SO_PASSCRED: 836 if (valbool) 837 set_bit(SOCK_PASSCRED, &sock->flags); 838 else 839 clear_bit(SOCK_PASSCRED, &sock->flags); 840 break; 841 842 case SO_TIMESTAMP: 843 case SO_TIMESTAMPNS: 844 if (valbool) { 845 if (optname == SO_TIMESTAMP) 846 sock_reset_flag(sk, SOCK_RCVTSTAMPNS); 847 else 848 sock_set_flag(sk, SOCK_RCVTSTAMPNS); 849 sock_set_flag(sk, SOCK_RCVTSTAMP); 850 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 851 } else { 852 sock_reset_flag(sk, SOCK_RCVTSTAMP); 853 sock_reset_flag(sk, SOCK_RCVTSTAMPNS); 854 } 855 break; 856 857 case SO_TIMESTAMPING: 858 if (val & ~SOF_TIMESTAMPING_MASK) { 859 ret = -EINVAL; 860 break; 861 } 862 863 if (val & SOF_TIMESTAMPING_OPT_ID && 864 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) { 865 if (sk->sk_protocol == IPPROTO_TCP) { 866 if (sk->sk_state != TCP_ESTABLISHED) { 867 ret = -EINVAL; 868 break; 869 } 870 sk->sk_tskey = tcp_sk(sk)->snd_una; 871 } else { 872 sk->sk_tskey = 0; 873 } 874 } 875 sk->sk_tsflags = val; 876 if (val & SOF_TIMESTAMPING_RX_SOFTWARE) 877 sock_enable_timestamp(sk, 878 SOCK_TIMESTAMPING_RX_SOFTWARE); 879 else 880 sock_disable_timestamp(sk, 881 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE)); 882 break; 883 884 case SO_RCVLOWAT: 885 if (val < 0) 886 val = INT_MAX; 887 sk->sk_rcvlowat = val ? : 1; 888 break; 889 890 case SO_RCVTIMEO: 891 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen); 892 break; 893 894 case SO_SNDTIMEO: 895 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen); 896 break; 897 898 case SO_ATTACH_FILTER: 899 ret = -EINVAL; 900 if (optlen == sizeof(struct sock_fprog)) { 901 struct sock_fprog fprog; 902 903 ret = -EFAULT; 904 if (copy_from_user(&fprog, optval, sizeof(fprog))) 905 break; 906 907 ret = sk_attach_filter(&fprog, sk); 908 } 909 break; 910 911 case SO_ATTACH_BPF: 912 ret = -EINVAL; 913 if (optlen == sizeof(u32)) { 914 u32 ufd; 915 916 ret = -EFAULT; 917 if (copy_from_user(&ufd, optval, sizeof(ufd))) 918 break; 919 920 ret = sk_attach_bpf(ufd, sk); 921 } 922 break; 923 924 case SO_DETACH_FILTER: 925 ret = sk_detach_filter(sk); 926 break; 927 928 case SO_LOCK_FILTER: 929 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool) 930 ret = -EPERM; 931 else 932 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool); 933 break; 934 935 case SO_PASSSEC: 936 if (valbool) 937 set_bit(SOCK_PASSSEC, &sock->flags); 938 else 939 clear_bit(SOCK_PASSSEC, &sock->flags); 940 break; 941 case SO_MARK: 942 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) 943 ret = -EPERM; 944 else 945 sk->sk_mark = val; 946 break; 947 948 case SO_RXQ_OVFL: 949 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool); 950 break; 951 952 case SO_WIFI_STATUS: 953 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool); 954 break; 955 956 case SO_PEEK_OFF: 957 if (sock->ops->set_peek_off) 958 ret = sock->ops->set_peek_off(sk, val); 959 else 960 ret = -EOPNOTSUPP; 961 break; 962 963 case SO_NOFCS: 964 sock_valbool_flag(sk, SOCK_NOFCS, valbool); 965 break; 966 967 case SO_SELECT_ERR_QUEUE: 968 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool); 969 break; 970 971 #ifdef CONFIG_NET_RX_BUSY_POLL 972 case SO_BUSY_POLL: 973 /* allow unprivileged users to decrease the value */ 974 if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN)) 975 ret = -EPERM; 976 else { 977 if (val < 0) 978 ret = -EINVAL; 979 else 980 sk->sk_ll_usec = val; 981 } 982 break; 983 #endif 984 985 case SO_MAX_PACING_RATE: 986 sk->sk_max_pacing_rate = val; 987 sk->sk_pacing_rate = min(sk->sk_pacing_rate, 988 sk->sk_max_pacing_rate); 989 break; 990 991 default: 992 ret = -ENOPROTOOPT; 993 break; 994 } 995 release_sock(sk); 996 return ret; 997 } 998 EXPORT_SYMBOL(sock_setsockopt); 999 1000 1001 static void cred_to_ucred(struct pid *pid, const struct cred *cred, 1002 struct ucred *ucred) 1003 { 1004 ucred->pid = pid_vnr(pid); 1005 ucred->uid = ucred->gid = -1; 1006 if (cred) { 1007 struct user_namespace *current_ns = current_user_ns(); 1008 1009 ucred->uid = from_kuid_munged(current_ns, cred->euid); 1010 ucred->gid = from_kgid_munged(current_ns, cred->egid); 1011 } 1012 } 1013 1014 int sock_getsockopt(struct socket *sock, int level, int optname, 1015 char __user *optval, int __user *optlen) 1016 { 1017 struct sock *sk = sock->sk; 1018 1019 union { 1020 int val; 1021 struct linger ling; 1022 struct timeval tm; 1023 } v; 1024 1025 int lv = sizeof(int); 1026 int len; 1027 1028 if (get_user(len, optlen)) 1029 return -EFAULT; 1030 if (len < 0) 1031 return -EINVAL; 1032 1033 memset(&v, 0, sizeof(v)); 1034 1035 switch (optname) { 1036 case SO_DEBUG: 1037 v.val = sock_flag(sk, SOCK_DBG); 1038 break; 1039 1040 case SO_DONTROUTE: 1041 v.val = sock_flag(sk, SOCK_LOCALROUTE); 1042 break; 1043 1044 case SO_BROADCAST: 1045 v.val = sock_flag(sk, SOCK_BROADCAST); 1046 break; 1047 1048 case SO_SNDBUF: 1049 v.val = sk->sk_sndbuf; 1050 break; 1051 1052 case SO_RCVBUF: 1053 v.val = sk->sk_rcvbuf; 1054 break; 1055 1056 case SO_REUSEADDR: 1057 v.val = sk->sk_reuse; 1058 break; 1059 1060 case SO_REUSEPORT: 1061 v.val = sk->sk_reuseport; 1062 break; 1063 1064 case SO_KEEPALIVE: 1065 v.val = sock_flag(sk, SOCK_KEEPOPEN); 1066 break; 1067 1068 case SO_TYPE: 1069 v.val = sk->sk_type; 1070 break; 1071 1072 case SO_PROTOCOL: 1073 v.val = sk->sk_protocol; 1074 break; 1075 1076 case SO_DOMAIN: 1077 v.val = sk->sk_family; 1078 break; 1079 1080 case SO_ERROR: 1081 v.val = -sock_error(sk); 1082 if (v.val == 0) 1083 v.val = xchg(&sk->sk_err_soft, 0); 1084 break; 1085 1086 case SO_OOBINLINE: 1087 v.val = sock_flag(sk, SOCK_URGINLINE); 1088 break; 1089 1090 case SO_NO_CHECK: 1091 v.val = sk->sk_no_check_tx; 1092 break; 1093 1094 case SO_PRIORITY: 1095 v.val = sk->sk_priority; 1096 break; 1097 1098 case SO_LINGER: 1099 lv = sizeof(v.ling); 1100 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER); 1101 v.ling.l_linger = sk->sk_lingertime / HZ; 1102 break; 1103 1104 case SO_BSDCOMPAT: 1105 sock_warn_obsolete_bsdism("getsockopt"); 1106 break; 1107 1108 case SO_TIMESTAMP: 1109 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && 1110 !sock_flag(sk, SOCK_RCVTSTAMPNS); 1111 break; 1112 1113 case SO_TIMESTAMPNS: 1114 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS); 1115 break; 1116 1117 case SO_TIMESTAMPING: 1118 v.val = sk->sk_tsflags; 1119 break; 1120 1121 case SO_RCVTIMEO: 1122 lv = sizeof(struct timeval); 1123 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) { 1124 v.tm.tv_sec = 0; 1125 v.tm.tv_usec = 0; 1126 } else { 1127 v.tm.tv_sec = sk->sk_rcvtimeo / HZ; 1128 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ; 1129 } 1130 break; 1131 1132 case SO_SNDTIMEO: 1133 lv = sizeof(struct timeval); 1134 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) { 1135 v.tm.tv_sec = 0; 1136 v.tm.tv_usec = 0; 1137 } else { 1138 v.tm.tv_sec = sk->sk_sndtimeo / HZ; 1139 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ; 1140 } 1141 break; 1142 1143 case SO_RCVLOWAT: 1144 v.val = sk->sk_rcvlowat; 1145 break; 1146 1147 case SO_SNDLOWAT: 1148 v.val = 1; 1149 break; 1150 1151 case SO_PASSCRED: 1152 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags); 1153 break; 1154 1155 case SO_PEERCRED: 1156 { 1157 struct ucred peercred; 1158 if (len > sizeof(peercred)) 1159 len = sizeof(peercred); 1160 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred); 1161 if (copy_to_user(optval, &peercred, len)) 1162 return -EFAULT; 1163 goto lenout; 1164 } 1165 1166 case SO_PEERNAME: 1167 { 1168 char address[128]; 1169 1170 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2)) 1171 return -ENOTCONN; 1172 if (lv < len) 1173 return -EINVAL; 1174 if (copy_to_user(optval, address, len)) 1175 return -EFAULT; 1176 goto lenout; 1177 } 1178 1179 /* Dubious BSD thing... Probably nobody even uses it, but 1180 * the UNIX standard wants it for whatever reason... -DaveM 1181 */ 1182 case SO_ACCEPTCONN: 1183 v.val = sk->sk_state == TCP_LISTEN; 1184 break; 1185 1186 case SO_PASSSEC: 1187 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags); 1188 break; 1189 1190 case SO_PEERSEC: 1191 return security_socket_getpeersec_stream(sock, optval, optlen, len); 1192 1193 case SO_MARK: 1194 v.val = sk->sk_mark; 1195 break; 1196 1197 case SO_RXQ_OVFL: 1198 v.val = sock_flag(sk, SOCK_RXQ_OVFL); 1199 break; 1200 1201 case SO_WIFI_STATUS: 1202 v.val = sock_flag(sk, SOCK_WIFI_STATUS); 1203 break; 1204 1205 case SO_PEEK_OFF: 1206 if (!sock->ops->set_peek_off) 1207 return -EOPNOTSUPP; 1208 1209 v.val = sk->sk_peek_off; 1210 break; 1211 case SO_NOFCS: 1212 v.val = sock_flag(sk, SOCK_NOFCS); 1213 break; 1214 1215 case SO_BINDTODEVICE: 1216 return sock_getbindtodevice(sk, optval, optlen, len); 1217 1218 case SO_GET_FILTER: 1219 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len); 1220 if (len < 0) 1221 return len; 1222 1223 goto lenout; 1224 1225 case SO_LOCK_FILTER: 1226 v.val = sock_flag(sk, SOCK_FILTER_LOCKED); 1227 break; 1228 1229 case SO_BPF_EXTENSIONS: 1230 v.val = bpf_tell_extensions(); 1231 break; 1232 1233 case SO_SELECT_ERR_QUEUE: 1234 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE); 1235 break; 1236 1237 #ifdef CONFIG_NET_RX_BUSY_POLL 1238 case SO_BUSY_POLL: 1239 v.val = sk->sk_ll_usec; 1240 break; 1241 #endif 1242 1243 case SO_MAX_PACING_RATE: 1244 v.val = sk->sk_max_pacing_rate; 1245 break; 1246 1247 case SO_INCOMING_CPU: 1248 v.val = sk->sk_incoming_cpu; 1249 break; 1250 1251 default: 1252 /* We implement the SO_SNDLOWAT etc to not be settable 1253 * (1003.1g 7). 1254 */ 1255 return -ENOPROTOOPT; 1256 } 1257 1258 if (len > lv) 1259 len = lv; 1260 if (copy_to_user(optval, &v, len)) 1261 return -EFAULT; 1262 lenout: 1263 if (put_user(len, optlen)) 1264 return -EFAULT; 1265 return 0; 1266 } 1267 1268 /* 1269 * Initialize an sk_lock. 1270 * 1271 * (We also register the sk_lock with the lock validator.) 1272 */ 1273 static inline void sock_lock_init(struct sock *sk) 1274 { 1275 sock_lock_init_class_and_name(sk, 1276 af_family_slock_key_strings[sk->sk_family], 1277 af_family_slock_keys + sk->sk_family, 1278 af_family_key_strings[sk->sk_family], 1279 af_family_keys + sk->sk_family); 1280 } 1281 1282 /* 1283 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet, 1284 * even temporarly, because of RCU lookups. sk_node should also be left as is. 1285 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end 1286 */ 1287 static void sock_copy(struct sock *nsk, const struct sock *osk) 1288 { 1289 #ifdef CONFIG_SECURITY_NETWORK 1290 void *sptr = nsk->sk_security; 1291 #endif 1292 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin)); 1293 1294 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end, 1295 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end)); 1296 1297 #ifdef CONFIG_SECURITY_NETWORK 1298 nsk->sk_security = sptr; 1299 security_sk_clone(osk, nsk); 1300 #endif 1301 } 1302 1303 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size) 1304 { 1305 unsigned long nulls1, nulls2; 1306 1307 nulls1 = offsetof(struct sock, __sk_common.skc_node.next); 1308 nulls2 = offsetof(struct sock, __sk_common.skc_portaddr_node.next); 1309 if (nulls1 > nulls2) 1310 swap(nulls1, nulls2); 1311 1312 if (nulls1 != 0) 1313 memset((char *)sk, 0, nulls1); 1314 memset((char *)sk + nulls1 + sizeof(void *), 0, 1315 nulls2 - nulls1 - sizeof(void *)); 1316 memset((char *)sk + nulls2 + sizeof(void *), 0, 1317 size - nulls2 - sizeof(void *)); 1318 } 1319 EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls); 1320 1321 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority, 1322 int family) 1323 { 1324 struct sock *sk; 1325 struct kmem_cache *slab; 1326 1327 slab = prot->slab; 1328 if (slab != NULL) { 1329 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO); 1330 if (!sk) 1331 return sk; 1332 if (priority & __GFP_ZERO) { 1333 if (prot->clear_sk) 1334 prot->clear_sk(sk, prot->obj_size); 1335 else 1336 sk_prot_clear_nulls(sk, prot->obj_size); 1337 } 1338 } else 1339 sk = kmalloc(prot->obj_size, priority); 1340 1341 if (sk != NULL) { 1342 kmemcheck_annotate_bitfield(sk, flags); 1343 1344 if (security_sk_alloc(sk, family, priority)) 1345 goto out_free; 1346 1347 if (!try_module_get(prot->owner)) 1348 goto out_free_sec; 1349 sk_tx_queue_clear(sk); 1350 } 1351 1352 return sk; 1353 1354 out_free_sec: 1355 security_sk_free(sk); 1356 out_free: 1357 if (slab != NULL) 1358 kmem_cache_free(slab, sk); 1359 else 1360 kfree(sk); 1361 return NULL; 1362 } 1363 1364 static void sk_prot_free(struct proto *prot, struct sock *sk) 1365 { 1366 struct kmem_cache *slab; 1367 struct module *owner; 1368 1369 owner = prot->owner; 1370 slab = prot->slab; 1371 1372 security_sk_free(sk); 1373 if (slab != NULL) 1374 kmem_cache_free(slab, sk); 1375 else 1376 kfree(sk); 1377 module_put(owner); 1378 } 1379 1380 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO) 1381 void sock_update_netprioidx(struct sock *sk) 1382 { 1383 if (in_interrupt()) 1384 return; 1385 1386 sk->sk_cgrp_prioidx = task_netprioidx(current); 1387 } 1388 EXPORT_SYMBOL_GPL(sock_update_netprioidx); 1389 #endif 1390 1391 /** 1392 * sk_alloc - All socket objects are allocated here 1393 * @net: the applicable net namespace 1394 * @family: protocol family 1395 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 1396 * @prot: struct proto associated with this new sock instance 1397 * @kern: is this to be a kernel socket? 1398 */ 1399 struct sock *sk_alloc(struct net *net, int family, gfp_t priority, 1400 struct proto *prot, int kern) 1401 { 1402 struct sock *sk; 1403 1404 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family); 1405 if (sk) { 1406 sk->sk_family = family; 1407 /* 1408 * See comment in struct sock definition to understand 1409 * why we need sk_prot_creator -acme 1410 */ 1411 sk->sk_prot = sk->sk_prot_creator = prot; 1412 sock_lock_init(sk); 1413 sk->sk_net_refcnt = kern ? 0 : 1; 1414 if (likely(sk->sk_net_refcnt)) 1415 get_net(net); 1416 sock_net_set(sk, net); 1417 atomic_set(&sk->sk_wmem_alloc, 1); 1418 1419 sock_update_classid(sk); 1420 sock_update_netprioidx(sk); 1421 } 1422 1423 return sk; 1424 } 1425 EXPORT_SYMBOL(sk_alloc); 1426 1427 void sk_destruct(struct sock *sk) 1428 { 1429 struct sk_filter *filter; 1430 1431 if (sk->sk_destruct) 1432 sk->sk_destruct(sk); 1433 1434 filter = rcu_dereference_check(sk->sk_filter, 1435 atomic_read(&sk->sk_wmem_alloc) == 0); 1436 if (filter) { 1437 sk_filter_uncharge(sk, filter); 1438 RCU_INIT_POINTER(sk->sk_filter, NULL); 1439 } 1440 1441 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP); 1442 1443 if (atomic_read(&sk->sk_omem_alloc)) 1444 pr_debug("%s: optmem leakage (%d bytes) detected\n", 1445 __func__, atomic_read(&sk->sk_omem_alloc)); 1446 1447 if (sk->sk_peer_cred) 1448 put_cred(sk->sk_peer_cred); 1449 put_pid(sk->sk_peer_pid); 1450 if (likely(sk->sk_net_refcnt)) 1451 put_net(sock_net(sk)); 1452 sk_prot_free(sk->sk_prot_creator, sk); 1453 } 1454 1455 static void __sk_free(struct sock *sk) 1456 { 1457 if (unlikely(sock_diag_has_destroy_listeners(sk) && sk->sk_net_refcnt)) 1458 sock_diag_broadcast_destroy(sk); 1459 else 1460 sk_destruct(sk); 1461 } 1462 1463 void sk_free(struct sock *sk) 1464 { 1465 /* 1466 * We subtract one from sk_wmem_alloc and can know if 1467 * some packets are still in some tx queue. 1468 * If not null, sock_wfree() will call __sk_free(sk) later 1469 */ 1470 if (atomic_dec_and_test(&sk->sk_wmem_alloc)) 1471 __sk_free(sk); 1472 } 1473 EXPORT_SYMBOL(sk_free); 1474 1475 static void sk_update_clone(const struct sock *sk, struct sock *newsk) 1476 { 1477 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) 1478 sock_update_memcg(newsk); 1479 } 1480 1481 /** 1482 * sk_clone_lock - clone a socket, and lock its clone 1483 * @sk: the socket to clone 1484 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 1485 * 1486 * Caller must unlock socket even in error path (bh_unlock_sock(newsk)) 1487 */ 1488 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority) 1489 { 1490 struct sock *newsk; 1491 bool is_charged = true; 1492 1493 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family); 1494 if (newsk != NULL) { 1495 struct sk_filter *filter; 1496 1497 sock_copy(newsk, sk); 1498 1499 /* SANITY */ 1500 if (likely(newsk->sk_net_refcnt)) 1501 get_net(sock_net(newsk)); 1502 sk_node_init(&newsk->sk_node); 1503 sock_lock_init(newsk); 1504 bh_lock_sock(newsk); 1505 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL; 1506 newsk->sk_backlog.len = 0; 1507 1508 atomic_set(&newsk->sk_rmem_alloc, 0); 1509 /* 1510 * sk_wmem_alloc set to one (see sk_free() and sock_wfree()) 1511 */ 1512 atomic_set(&newsk->sk_wmem_alloc, 1); 1513 atomic_set(&newsk->sk_omem_alloc, 0); 1514 skb_queue_head_init(&newsk->sk_receive_queue); 1515 skb_queue_head_init(&newsk->sk_write_queue); 1516 1517 spin_lock_init(&newsk->sk_dst_lock); 1518 rwlock_init(&newsk->sk_callback_lock); 1519 lockdep_set_class_and_name(&newsk->sk_callback_lock, 1520 af_callback_keys + newsk->sk_family, 1521 af_family_clock_key_strings[newsk->sk_family]); 1522 1523 newsk->sk_dst_cache = NULL; 1524 newsk->sk_wmem_queued = 0; 1525 newsk->sk_forward_alloc = 0; 1526 newsk->sk_send_head = NULL; 1527 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK; 1528 1529 sock_reset_flag(newsk, SOCK_DONE); 1530 skb_queue_head_init(&newsk->sk_error_queue); 1531 1532 filter = rcu_dereference_protected(newsk->sk_filter, 1); 1533 if (filter != NULL) 1534 /* though it's an empty new sock, the charging may fail 1535 * if sysctl_optmem_max was changed between creation of 1536 * original socket and cloning 1537 */ 1538 is_charged = sk_filter_charge(newsk, filter); 1539 1540 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk))) { 1541 /* It is still raw copy of parent, so invalidate 1542 * destructor and make plain sk_free() */ 1543 newsk->sk_destruct = NULL; 1544 bh_unlock_sock(newsk); 1545 sk_free(newsk); 1546 newsk = NULL; 1547 goto out; 1548 } 1549 1550 newsk->sk_err = 0; 1551 newsk->sk_priority = 0; 1552 newsk->sk_incoming_cpu = raw_smp_processor_id(); 1553 atomic64_set(&newsk->sk_cookie, 0); 1554 /* 1555 * Before updating sk_refcnt, we must commit prior changes to memory 1556 * (Documentation/RCU/rculist_nulls.txt for details) 1557 */ 1558 smp_wmb(); 1559 atomic_set(&newsk->sk_refcnt, 2); 1560 1561 /* 1562 * Increment the counter in the same struct proto as the master 1563 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that 1564 * is the same as sk->sk_prot->socks, as this field was copied 1565 * with memcpy). 1566 * 1567 * This _changes_ the previous behaviour, where 1568 * tcp_create_openreq_child always was incrementing the 1569 * equivalent to tcp_prot->socks (inet_sock_nr), so this have 1570 * to be taken into account in all callers. -acme 1571 */ 1572 sk_refcnt_debug_inc(newsk); 1573 sk_set_socket(newsk, NULL); 1574 newsk->sk_wq = NULL; 1575 1576 sk_update_clone(sk, newsk); 1577 1578 if (newsk->sk_prot->sockets_allocated) 1579 sk_sockets_allocated_inc(newsk); 1580 1581 if (newsk->sk_flags & SK_FLAGS_TIMESTAMP) 1582 net_enable_timestamp(); 1583 } 1584 out: 1585 return newsk; 1586 } 1587 EXPORT_SYMBOL_GPL(sk_clone_lock); 1588 1589 void sk_setup_caps(struct sock *sk, struct dst_entry *dst) 1590 { 1591 u32 max_segs = 1; 1592 1593 __sk_dst_set(sk, dst); 1594 sk->sk_route_caps = dst->dev->features; 1595 if (sk->sk_route_caps & NETIF_F_GSO) 1596 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE; 1597 sk->sk_route_caps &= ~sk->sk_route_nocaps; 1598 if (sk_can_gso(sk)) { 1599 if (dst->header_len) { 1600 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 1601 } else { 1602 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM; 1603 sk->sk_gso_max_size = dst->dev->gso_max_size; 1604 max_segs = max_t(u32, dst->dev->gso_max_segs, 1); 1605 } 1606 } 1607 sk->sk_gso_max_segs = max_segs; 1608 } 1609 EXPORT_SYMBOL_GPL(sk_setup_caps); 1610 1611 /* 1612 * Simple resource managers for sockets. 1613 */ 1614 1615 1616 /* 1617 * Write buffer destructor automatically called from kfree_skb. 1618 */ 1619 void sock_wfree(struct sk_buff *skb) 1620 { 1621 struct sock *sk = skb->sk; 1622 unsigned int len = skb->truesize; 1623 1624 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) { 1625 /* 1626 * Keep a reference on sk_wmem_alloc, this will be released 1627 * after sk_write_space() call 1628 */ 1629 atomic_sub(len - 1, &sk->sk_wmem_alloc); 1630 sk->sk_write_space(sk); 1631 len = 1; 1632 } 1633 /* 1634 * if sk_wmem_alloc reaches 0, we must finish what sk_free() 1635 * could not do because of in-flight packets 1636 */ 1637 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc)) 1638 __sk_free(sk); 1639 } 1640 EXPORT_SYMBOL(sock_wfree); 1641 1642 void skb_orphan_partial(struct sk_buff *skb) 1643 { 1644 /* TCP stack sets skb->ooo_okay based on sk_wmem_alloc, 1645 * so we do not completely orphan skb, but transfert all 1646 * accounted bytes but one, to avoid unexpected reorders. 1647 */ 1648 if (skb->destructor == sock_wfree 1649 #ifdef CONFIG_INET 1650 || skb->destructor == tcp_wfree 1651 #endif 1652 ) { 1653 atomic_sub(skb->truesize - 1, &skb->sk->sk_wmem_alloc); 1654 skb->truesize = 1; 1655 } else { 1656 skb_orphan(skb); 1657 } 1658 } 1659 EXPORT_SYMBOL(skb_orphan_partial); 1660 1661 /* 1662 * Read buffer destructor automatically called from kfree_skb. 1663 */ 1664 void sock_rfree(struct sk_buff *skb) 1665 { 1666 struct sock *sk = skb->sk; 1667 unsigned int len = skb->truesize; 1668 1669 atomic_sub(len, &sk->sk_rmem_alloc); 1670 sk_mem_uncharge(sk, len); 1671 } 1672 EXPORT_SYMBOL(sock_rfree); 1673 1674 /* 1675 * Buffer destructor for skbs that are not used directly in read or write 1676 * path, e.g. for error handler skbs. Automatically called from kfree_skb. 1677 */ 1678 void sock_efree(struct sk_buff *skb) 1679 { 1680 sock_put(skb->sk); 1681 } 1682 EXPORT_SYMBOL(sock_efree); 1683 1684 kuid_t sock_i_uid(struct sock *sk) 1685 { 1686 kuid_t uid; 1687 1688 read_lock_bh(&sk->sk_callback_lock); 1689 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID; 1690 read_unlock_bh(&sk->sk_callback_lock); 1691 return uid; 1692 } 1693 EXPORT_SYMBOL(sock_i_uid); 1694 1695 unsigned long sock_i_ino(struct sock *sk) 1696 { 1697 unsigned long ino; 1698 1699 read_lock_bh(&sk->sk_callback_lock); 1700 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0; 1701 read_unlock_bh(&sk->sk_callback_lock); 1702 return ino; 1703 } 1704 EXPORT_SYMBOL(sock_i_ino); 1705 1706 /* 1707 * Allocate a skb from the socket's send buffer. 1708 */ 1709 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 1710 gfp_t priority) 1711 { 1712 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) { 1713 struct sk_buff *skb = alloc_skb(size, priority); 1714 if (skb) { 1715 skb_set_owner_w(skb, sk); 1716 return skb; 1717 } 1718 } 1719 return NULL; 1720 } 1721 EXPORT_SYMBOL(sock_wmalloc); 1722 1723 /* 1724 * Allocate a memory block from the socket's option memory buffer. 1725 */ 1726 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority) 1727 { 1728 if ((unsigned int)size <= sysctl_optmem_max && 1729 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) { 1730 void *mem; 1731 /* First do the add, to avoid the race if kmalloc 1732 * might sleep. 1733 */ 1734 atomic_add(size, &sk->sk_omem_alloc); 1735 mem = kmalloc(size, priority); 1736 if (mem) 1737 return mem; 1738 atomic_sub(size, &sk->sk_omem_alloc); 1739 } 1740 return NULL; 1741 } 1742 EXPORT_SYMBOL(sock_kmalloc); 1743 1744 /* Free an option memory block. Note, we actually want the inline 1745 * here as this allows gcc to detect the nullify and fold away the 1746 * condition entirely. 1747 */ 1748 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size, 1749 const bool nullify) 1750 { 1751 if (WARN_ON_ONCE(!mem)) 1752 return; 1753 if (nullify) 1754 kzfree(mem); 1755 else 1756 kfree(mem); 1757 atomic_sub(size, &sk->sk_omem_alloc); 1758 } 1759 1760 void sock_kfree_s(struct sock *sk, void *mem, int size) 1761 { 1762 __sock_kfree_s(sk, mem, size, false); 1763 } 1764 EXPORT_SYMBOL(sock_kfree_s); 1765 1766 void sock_kzfree_s(struct sock *sk, void *mem, int size) 1767 { 1768 __sock_kfree_s(sk, mem, size, true); 1769 } 1770 EXPORT_SYMBOL(sock_kzfree_s); 1771 1772 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock. 1773 I think, these locks should be removed for datagram sockets. 1774 */ 1775 static long sock_wait_for_wmem(struct sock *sk, long timeo) 1776 { 1777 DEFINE_WAIT(wait); 1778 1779 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags); 1780 for (;;) { 1781 if (!timeo) 1782 break; 1783 if (signal_pending(current)) 1784 break; 1785 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 1786 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 1787 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) 1788 break; 1789 if (sk->sk_shutdown & SEND_SHUTDOWN) 1790 break; 1791 if (sk->sk_err) 1792 break; 1793 timeo = schedule_timeout(timeo); 1794 } 1795 finish_wait(sk_sleep(sk), &wait); 1796 return timeo; 1797 } 1798 1799 1800 /* 1801 * Generic send/receive buffer handlers 1802 */ 1803 1804 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 1805 unsigned long data_len, int noblock, 1806 int *errcode, int max_page_order) 1807 { 1808 struct sk_buff *skb; 1809 long timeo; 1810 int err; 1811 1812 timeo = sock_sndtimeo(sk, noblock); 1813 for (;;) { 1814 err = sock_error(sk); 1815 if (err != 0) 1816 goto failure; 1817 1818 err = -EPIPE; 1819 if (sk->sk_shutdown & SEND_SHUTDOWN) 1820 goto failure; 1821 1822 if (sk_wmem_alloc_get(sk) < sk->sk_sndbuf) 1823 break; 1824 1825 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags); 1826 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 1827 err = -EAGAIN; 1828 if (!timeo) 1829 goto failure; 1830 if (signal_pending(current)) 1831 goto interrupted; 1832 timeo = sock_wait_for_wmem(sk, timeo); 1833 } 1834 skb = alloc_skb_with_frags(header_len, data_len, max_page_order, 1835 errcode, sk->sk_allocation); 1836 if (skb) 1837 skb_set_owner_w(skb, sk); 1838 return skb; 1839 1840 interrupted: 1841 err = sock_intr_errno(timeo); 1842 failure: 1843 *errcode = err; 1844 return NULL; 1845 } 1846 EXPORT_SYMBOL(sock_alloc_send_pskb); 1847 1848 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size, 1849 int noblock, int *errcode) 1850 { 1851 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0); 1852 } 1853 EXPORT_SYMBOL(sock_alloc_send_skb); 1854 1855 /* On 32bit arches, an skb frag is limited to 2^15 */ 1856 #define SKB_FRAG_PAGE_ORDER get_order(32768) 1857 1858 /** 1859 * skb_page_frag_refill - check that a page_frag contains enough room 1860 * @sz: minimum size of the fragment we want to get 1861 * @pfrag: pointer to page_frag 1862 * @gfp: priority for memory allocation 1863 * 1864 * Note: While this allocator tries to use high order pages, there is 1865 * no guarantee that allocations succeed. Therefore, @sz MUST be 1866 * less or equal than PAGE_SIZE. 1867 */ 1868 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp) 1869 { 1870 if (pfrag->page) { 1871 if (atomic_read(&pfrag->page->_count) == 1) { 1872 pfrag->offset = 0; 1873 return true; 1874 } 1875 if (pfrag->offset + sz <= pfrag->size) 1876 return true; 1877 put_page(pfrag->page); 1878 } 1879 1880 pfrag->offset = 0; 1881 if (SKB_FRAG_PAGE_ORDER) { 1882 pfrag->page = alloc_pages((gfp & ~__GFP_WAIT) | __GFP_COMP | 1883 __GFP_NOWARN | __GFP_NORETRY, 1884 SKB_FRAG_PAGE_ORDER); 1885 if (likely(pfrag->page)) { 1886 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER; 1887 return true; 1888 } 1889 } 1890 pfrag->page = alloc_page(gfp); 1891 if (likely(pfrag->page)) { 1892 pfrag->size = PAGE_SIZE; 1893 return true; 1894 } 1895 return false; 1896 } 1897 EXPORT_SYMBOL(skb_page_frag_refill); 1898 1899 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag) 1900 { 1901 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation))) 1902 return true; 1903 1904 sk_enter_memory_pressure(sk); 1905 sk_stream_moderate_sndbuf(sk); 1906 return false; 1907 } 1908 EXPORT_SYMBOL(sk_page_frag_refill); 1909 1910 static void __lock_sock(struct sock *sk) 1911 __releases(&sk->sk_lock.slock) 1912 __acquires(&sk->sk_lock.slock) 1913 { 1914 DEFINE_WAIT(wait); 1915 1916 for (;;) { 1917 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait, 1918 TASK_UNINTERRUPTIBLE); 1919 spin_unlock_bh(&sk->sk_lock.slock); 1920 schedule(); 1921 spin_lock_bh(&sk->sk_lock.slock); 1922 if (!sock_owned_by_user(sk)) 1923 break; 1924 } 1925 finish_wait(&sk->sk_lock.wq, &wait); 1926 } 1927 1928 static void __release_sock(struct sock *sk) 1929 __releases(&sk->sk_lock.slock) 1930 __acquires(&sk->sk_lock.slock) 1931 { 1932 struct sk_buff *skb = sk->sk_backlog.head; 1933 1934 do { 1935 sk->sk_backlog.head = sk->sk_backlog.tail = NULL; 1936 bh_unlock_sock(sk); 1937 1938 do { 1939 struct sk_buff *next = skb->next; 1940 1941 prefetch(next); 1942 WARN_ON_ONCE(skb_dst_is_noref(skb)); 1943 skb->next = NULL; 1944 sk_backlog_rcv(sk, skb); 1945 1946 /* 1947 * We are in process context here with softirqs 1948 * disabled, use cond_resched_softirq() to preempt. 1949 * This is safe to do because we've taken the backlog 1950 * queue private: 1951 */ 1952 cond_resched_softirq(); 1953 1954 skb = next; 1955 } while (skb != NULL); 1956 1957 bh_lock_sock(sk); 1958 } while ((skb = sk->sk_backlog.head) != NULL); 1959 1960 /* 1961 * Doing the zeroing here guarantee we can not loop forever 1962 * while a wild producer attempts to flood us. 1963 */ 1964 sk->sk_backlog.len = 0; 1965 } 1966 1967 /** 1968 * sk_wait_data - wait for data to arrive at sk_receive_queue 1969 * @sk: sock to wait on 1970 * @timeo: for how long 1971 * @skb: last skb seen on sk_receive_queue 1972 * 1973 * Now socket state including sk->sk_err is changed only under lock, 1974 * hence we may omit checks after joining wait queue. 1975 * We check receive queue before schedule() only as optimization; 1976 * it is very likely that release_sock() added new data. 1977 */ 1978 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb) 1979 { 1980 int rc; 1981 DEFINE_WAIT(wait); 1982 1983 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 1984 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags); 1985 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb); 1986 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags); 1987 finish_wait(sk_sleep(sk), &wait); 1988 return rc; 1989 } 1990 EXPORT_SYMBOL(sk_wait_data); 1991 1992 /** 1993 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated 1994 * @sk: socket 1995 * @size: memory size to allocate 1996 * @kind: allocation type 1997 * 1998 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means 1999 * rmem allocation. This function assumes that protocols which have 2000 * memory_pressure use sk_wmem_queued as write buffer accounting. 2001 */ 2002 int __sk_mem_schedule(struct sock *sk, int size, int kind) 2003 { 2004 struct proto *prot = sk->sk_prot; 2005 int amt = sk_mem_pages(size); 2006 long allocated; 2007 int parent_status = UNDER_LIMIT; 2008 2009 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM; 2010 2011 allocated = sk_memory_allocated_add(sk, amt, &parent_status); 2012 2013 /* Under limit. */ 2014 if (parent_status == UNDER_LIMIT && 2015 allocated <= sk_prot_mem_limits(sk, 0)) { 2016 sk_leave_memory_pressure(sk); 2017 return 1; 2018 } 2019 2020 /* Under pressure. (we or our parents) */ 2021 if ((parent_status > SOFT_LIMIT) || 2022 allocated > sk_prot_mem_limits(sk, 1)) 2023 sk_enter_memory_pressure(sk); 2024 2025 /* Over hard limit (we or our parents) */ 2026 if ((parent_status == OVER_LIMIT) || 2027 (allocated > sk_prot_mem_limits(sk, 2))) 2028 goto suppress_allocation; 2029 2030 /* guarantee minimum buffer size under pressure */ 2031 if (kind == SK_MEM_RECV) { 2032 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0]) 2033 return 1; 2034 2035 } else { /* SK_MEM_SEND */ 2036 if (sk->sk_type == SOCK_STREAM) { 2037 if (sk->sk_wmem_queued < prot->sysctl_wmem[0]) 2038 return 1; 2039 } else if (atomic_read(&sk->sk_wmem_alloc) < 2040 prot->sysctl_wmem[0]) 2041 return 1; 2042 } 2043 2044 if (sk_has_memory_pressure(sk)) { 2045 int alloc; 2046 2047 if (!sk_under_memory_pressure(sk)) 2048 return 1; 2049 alloc = sk_sockets_allocated_read_positive(sk); 2050 if (sk_prot_mem_limits(sk, 2) > alloc * 2051 sk_mem_pages(sk->sk_wmem_queued + 2052 atomic_read(&sk->sk_rmem_alloc) + 2053 sk->sk_forward_alloc)) 2054 return 1; 2055 } 2056 2057 suppress_allocation: 2058 2059 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) { 2060 sk_stream_moderate_sndbuf(sk); 2061 2062 /* Fail only if socket is _under_ its sndbuf. 2063 * In this case we cannot block, so that we have to fail. 2064 */ 2065 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf) 2066 return 1; 2067 } 2068 2069 trace_sock_exceed_buf_limit(sk, prot, allocated); 2070 2071 /* Alas. Undo changes. */ 2072 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM; 2073 2074 sk_memory_allocated_sub(sk, amt); 2075 2076 return 0; 2077 } 2078 EXPORT_SYMBOL(__sk_mem_schedule); 2079 2080 /** 2081 * __sk_mem_reclaim - reclaim memory_allocated 2082 * @sk: socket 2083 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple) 2084 */ 2085 void __sk_mem_reclaim(struct sock *sk, int amount) 2086 { 2087 amount >>= SK_MEM_QUANTUM_SHIFT; 2088 sk_memory_allocated_sub(sk, amount); 2089 sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT; 2090 2091 if (sk_under_memory_pressure(sk) && 2092 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0))) 2093 sk_leave_memory_pressure(sk); 2094 } 2095 EXPORT_SYMBOL(__sk_mem_reclaim); 2096 2097 2098 /* 2099 * Set of default routines for initialising struct proto_ops when 2100 * the protocol does not support a particular function. In certain 2101 * cases where it makes no sense for a protocol to have a "do nothing" 2102 * function, some default processing is provided. 2103 */ 2104 2105 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len) 2106 { 2107 return -EOPNOTSUPP; 2108 } 2109 EXPORT_SYMBOL(sock_no_bind); 2110 2111 int sock_no_connect(struct socket *sock, struct sockaddr *saddr, 2112 int len, int flags) 2113 { 2114 return -EOPNOTSUPP; 2115 } 2116 EXPORT_SYMBOL(sock_no_connect); 2117 2118 int sock_no_socketpair(struct socket *sock1, struct socket *sock2) 2119 { 2120 return -EOPNOTSUPP; 2121 } 2122 EXPORT_SYMBOL(sock_no_socketpair); 2123 2124 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags) 2125 { 2126 return -EOPNOTSUPP; 2127 } 2128 EXPORT_SYMBOL(sock_no_accept); 2129 2130 int sock_no_getname(struct socket *sock, struct sockaddr *saddr, 2131 int *len, int peer) 2132 { 2133 return -EOPNOTSUPP; 2134 } 2135 EXPORT_SYMBOL(sock_no_getname); 2136 2137 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt) 2138 { 2139 return 0; 2140 } 2141 EXPORT_SYMBOL(sock_no_poll); 2142 2143 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 2144 { 2145 return -EOPNOTSUPP; 2146 } 2147 EXPORT_SYMBOL(sock_no_ioctl); 2148 2149 int sock_no_listen(struct socket *sock, int backlog) 2150 { 2151 return -EOPNOTSUPP; 2152 } 2153 EXPORT_SYMBOL(sock_no_listen); 2154 2155 int sock_no_shutdown(struct socket *sock, int how) 2156 { 2157 return -EOPNOTSUPP; 2158 } 2159 EXPORT_SYMBOL(sock_no_shutdown); 2160 2161 int sock_no_setsockopt(struct socket *sock, int level, int optname, 2162 char __user *optval, unsigned int optlen) 2163 { 2164 return -EOPNOTSUPP; 2165 } 2166 EXPORT_SYMBOL(sock_no_setsockopt); 2167 2168 int sock_no_getsockopt(struct socket *sock, int level, int optname, 2169 char __user *optval, int __user *optlen) 2170 { 2171 return -EOPNOTSUPP; 2172 } 2173 EXPORT_SYMBOL(sock_no_getsockopt); 2174 2175 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len) 2176 { 2177 return -EOPNOTSUPP; 2178 } 2179 EXPORT_SYMBOL(sock_no_sendmsg); 2180 2181 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len, 2182 int flags) 2183 { 2184 return -EOPNOTSUPP; 2185 } 2186 EXPORT_SYMBOL(sock_no_recvmsg); 2187 2188 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma) 2189 { 2190 /* Mirror missing mmap method error code */ 2191 return -ENODEV; 2192 } 2193 EXPORT_SYMBOL(sock_no_mmap); 2194 2195 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags) 2196 { 2197 ssize_t res; 2198 struct msghdr msg = {.msg_flags = flags}; 2199 struct kvec iov; 2200 char *kaddr = kmap(page); 2201 iov.iov_base = kaddr + offset; 2202 iov.iov_len = size; 2203 res = kernel_sendmsg(sock, &msg, &iov, 1, size); 2204 kunmap(page); 2205 return res; 2206 } 2207 EXPORT_SYMBOL(sock_no_sendpage); 2208 2209 /* 2210 * Default Socket Callbacks 2211 */ 2212 2213 static void sock_def_wakeup(struct sock *sk) 2214 { 2215 struct socket_wq *wq; 2216 2217 rcu_read_lock(); 2218 wq = rcu_dereference(sk->sk_wq); 2219 if (wq_has_sleeper(wq)) 2220 wake_up_interruptible_all(&wq->wait); 2221 rcu_read_unlock(); 2222 } 2223 2224 static void sock_def_error_report(struct sock *sk) 2225 { 2226 struct socket_wq *wq; 2227 2228 rcu_read_lock(); 2229 wq = rcu_dereference(sk->sk_wq); 2230 if (wq_has_sleeper(wq)) 2231 wake_up_interruptible_poll(&wq->wait, POLLERR); 2232 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR); 2233 rcu_read_unlock(); 2234 } 2235 2236 static void sock_def_readable(struct sock *sk) 2237 { 2238 struct socket_wq *wq; 2239 2240 rcu_read_lock(); 2241 wq = rcu_dereference(sk->sk_wq); 2242 if (wq_has_sleeper(wq)) 2243 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI | 2244 POLLRDNORM | POLLRDBAND); 2245 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN); 2246 rcu_read_unlock(); 2247 } 2248 2249 static void sock_def_write_space(struct sock *sk) 2250 { 2251 struct socket_wq *wq; 2252 2253 rcu_read_lock(); 2254 2255 /* Do not wake up a writer until he can make "significant" 2256 * progress. --DaveM 2257 */ 2258 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) { 2259 wq = rcu_dereference(sk->sk_wq); 2260 if (wq_has_sleeper(wq)) 2261 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT | 2262 POLLWRNORM | POLLWRBAND); 2263 2264 /* Should agree with poll, otherwise some programs break */ 2265 if (sock_writeable(sk)) 2266 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT); 2267 } 2268 2269 rcu_read_unlock(); 2270 } 2271 2272 static void sock_def_destruct(struct sock *sk) 2273 { 2274 } 2275 2276 void sk_send_sigurg(struct sock *sk) 2277 { 2278 if (sk->sk_socket && sk->sk_socket->file) 2279 if (send_sigurg(&sk->sk_socket->file->f_owner)) 2280 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI); 2281 } 2282 EXPORT_SYMBOL(sk_send_sigurg); 2283 2284 void sk_reset_timer(struct sock *sk, struct timer_list* timer, 2285 unsigned long expires) 2286 { 2287 if (!mod_timer(timer, expires)) 2288 sock_hold(sk); 2289 } 2290 EXPORT_SYMBOL(sk_reset_timer); 2291 2292 void sk_stop_timer(struct sock *sk, struct timer_list* timer) 2293 { 2294 if (del_timer(timer)) 2295 __sock_put(sk); 2296 } 2297 EXPORT_SYMBOL(sk_stop_timer); 2298 2299 void sock_init_data(struct socket *sock, struct sock *sk) 2300 { 2301 skb_queue_head_init(&sk->sk_receive_queue); 2302 skb_queue_head_init(&sk->sk_write_queue); 2303 skb_queue_head_init(&sk->sk_error_queue); 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 } 2364 EXPORT_SYMBOL(sock_init_data); 2365 2366 void 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 } 2380 EXPORT_SYMBOL(lock_sock_nested); 2381 2382 void 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 } 2404 EXPORT_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 */ 2416 bool 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 } 2437 EXPORT_SYMBOL(lock_sock_fast); 2438 2439 int 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 } 2453 EXPORT_SYMBOL(sock_get_timestamp); 2454 2455 int 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 } 2469 EXPORT_SYMBOL(sock_get_timestampns); 2470 2471 void 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 2487 int 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; 2492 int copied, err; 2493 2494 err = -EAGAIN; 2495 skb = sock_dequeue_err_skb(sk); 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_msg(skb, 0, msg, 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 out_free_skb: 2517 kfree_skb(skb); 2518 out: 2519 return err; 2520 } 2521 EXPORT_SYMBOL(sock_recv_errqueue); 2522 2523 /* 2524 * Get a socket option on an socket. 2525 * 2526 * FIX: POSIX 1003.1g is very ambiguous here. It states that 2527 * asynchronous errors should be reported by getsockopt. We assume 2528 * this means if you specify SO_ERROR (otherwise whats the point of it). 2529 */ 2530 int sock_common_getsockopt(struct socket *sock, int level, int optname, 2531 char __user *optval, int __user *optlen) 2532 { 2533 struct sock *sk = sock->sk; 2534 2535 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen); 2536 } 2537 EXPORT_SYMBOL(sock_common_getsockopt); 2538 2539 #ifdef CONFIG_COMPAT 2540 int compat_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 if (sk->sk_prot->compat_getsockopt != NULL) 2546 return sk->sk_prot->compat_getsockopt(sk, level, optname, 2547 optval, optlen); 2548 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen); 2549 } 2550 EXPORT_SYMBOL(compat_sock_common_getsockopt); 2551 #endif 2552 2553 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 2554 int flags) 2555 { 2556 struct sock *sk = sock->sk; 2557 int addr_len = 0; 2558 int err; 2559 2560 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT, 2561 flags & ~MSG_DONTWAIT, &addr_len); 2562 if (err >= 0) 2563 msg->msg_namelen = addr_len; 2564 return err; 2565 } 2566 EXPORT_SYMBOL(sock_common_recvmsg); 2567 2568 /* 2569 * Set socket options on an inet socket. 2570 */ 2571 int sock_common_setsockopt(struct socket *sock, int level, int optname, 2572 char __user *optval, unsigned int optlen) 2573 { 2574 struct sock *sk = sock->sk; 2575 2576 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen); 2577 } 2578 EXPORT_SYMBOL(sock_common_setsockopt); 2579 2580 #ifdef CONFIG_COMPAT 2581 int compat_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 if (sk->sk_prot->compat_setsockopt != NULL) 2587 return sk->sk_prot->compat_setsockopt(sk, level, optname, 2588 optval, optlen); 2589 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen); 2590 } 2591 EXPORT_SYMBOL(compat_sock_common_setsockopt); 2592 #endif 2593 2594 void sk_common_release(struct sock *sk) 2595 { 2596 if (sk->sk_prot->destroy) 2597 sk->sk_prot->destroy(sk); 2598 2599 /* 2600 * Observation: when sock_common_release is called, processes have 2601 * no access to socket. But net still has. 2602 * Step one, detach it from networking: 2603 * 2604 * A. Remove from hash tables. 2605 */ 2606 2607 sk->sk_prot->unhash(sk); 2608 2609 /* 2610 * In this point socket cannot receive new packets, but it is possible 2611 * that some packets are in flight because some CPU runs receiver and 2612 * did hash table lookup before we unhashed socket. They will achieve 2613 * receive queue and will be purged by socket destructor. 2614 * 2615 * Also we still have packets pending on receive queue and probably, 2616 * our own packets waiting in device queues. sock_destroy will drain 2617 * receive queue, but transmitted packets will delay socket destruction 2618 * until the last reference will be released. 2619 */ 2620 2621 sock_orphan(sk); 2622 2623 xfrm_sk_free_policy(sk); 2624 2625 sk_refcnt_debug_release(sk); 2626 2627 if (sk->sk_frag.page) { 2628 put_page(sk->sk_frag.page); 2629 sk->sk_frag.page = NULL; 2630 } 2631 2632 sock_put(sk); 2633 } 2634 EXPORT_SYMBOL(sk_common_release); 2635 2636 #ifdef CONFIG_PROC_FS 2637 #define PROTO_INUSE_NR 64 /* should be enough for the first time */ 2638 struct prot_inuse { 2639 int val[PROTO_INUSE_NR]; 2640 }; 2641 2642 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR); 2643 2644 #ifdef CONFIG_NET_NS 2645 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val) 2646 { 2647 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val); 2648 } 2649 EXPORT_SYMBOL_GPL(sock_prot_inuse_add); 2650 2651 int sock_prot_inuse_get(struct net *net, struct proto *prot) 2652 { 2653 int cpu, idx = prot->inuse_idx; 2654 int res = 0; 2655 2656 for_each_possible_cpu(cpu) 2657 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx]; 2658 2659 return res >= 0 ? res : 0; 2660 } 2661 EXPORT_SYMBOL_GPL(sock_prot_inuse_get); 2662 2663 static int __net_init sock_inuse_init_net(struct net *net) 2664 { 2665 net->core.inuse = alloc_percpu(struct prot_inuse); 2666 return net->core.inuse ? 0 : -ENOMEM; 2667 } 2668 2669 static void __net_exit sock_inuse_exit_net(struct net *net) 2670 { 2671 free_percpu(net->core.inuse); 2672 } 2673 2674 static struct pernet_operations net_inuse_ops = { 2675 .init = sock_inuse_init_net, 2676 .exit = sock_inuse_exit_net, 2677 }; 2678 2679 static __init int net_inuse_init(void) 2680 { 2681 if (register_pernet_subsys(&net_inuse_ops)) 2682 panic("Cannot initialize net inuse counters"); 2683 2684 return 0; 2685 } 2686 2687 core_initcall(net_inuse_init); 2688 #else 2689 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse); 2690 2691 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val) 2692 { 2693 __this_cpu_add(prot_inuse.val[prot->inuse_idx], val); 2694 } 2695 EXPORT_SYMBOL_GPL(sock_prot_inuse_add); 2696 2697 int sock_prot_inuse_get(struct net *net, struct proto *prot) 2698 { 2699 int cpu, idx = prot->inuse_idx; 2700 int res = 0; 2701 2702 for_each_possible_cpu(cpu) 2703 res += per_cpu(prot_inuse, cpu).val[idx]; 2704 2705 return res >= 0 ? res : 0; 2706 } 2707 EXPORT_SYMBOL_GPL(sock_prot_inuse_get); 2708 #endif 2709 2710 static void assign_proto_idx(struct proto *prot) 2711 { 2712 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR); 2713 2714 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) { 2715 pr_err("PROTO_INUSE_NR exhausted\n"); 2716 return; 2717 } 2718 2719 set_bit(prot->inuse_idx, proto_inuse_idx); 2720 } 2721 2722 static void release_proto_idx(struct proto *prot) 2723 { 2724 if (prot->inuse_idx != PROTO_INUSE_NR - 1) 2725 clear_bit(prot->inuse_idx, proto_inuse_idx); 2726 } 2727 #else 2728 static inline void assign_proto_idx(struct proto *prot) 2729 { 2730 } 2731 2732 static inline void release_proto_idx(struct proto *prot) 2733 { 2734 } 2735 #endif 2736 2737 static void req_prot_cleanup(struct request_sock_ops *rsk_prot) 2738 { 2739 if (!rsk_prot) 2740 return; 2741 kfree(rsk_prot->slab_name); 2742 rsk_prot->slab_name = NULL; 2743 kmem_cache_destroy(rsk_prot->slab); 2744 rsk_prot->slab = NULL; 2745 } 2746 2747 static int req_prot_init(const struct proto *prot) 2748 { 2749 struct request_sock_ops *rsk_prot = prot->rsk_prot; 2750 2751 if (!rsk_prot) 2752 return 0; 2753 2754 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", 2755 prot->name); 2756 if (!rsk_prot->slab_name) 2757 return -ENOMEM; 2758 2759 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name, 2760 rsk_prot->obj_size, 0, 2761 0, NULL); 2762 2763 if (!rsk_prot->slab) { 2764 pr_crit("%s: Can't create request sock SLAB cache!\n", 2765 prot->name); 2766 return -ENOMEM; 2767 } 2768 return 0; 2769 } 2770 2771 int proto_register(struct proto *prot, int alloc_slab) 2772 { 2773 if (alloc_slab) { 2774 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0, 2775 SLAB_HWCACHE_ALIGN | prot->slab_flags, 2776 NULL); 2777 2778 if (prot->slab == NULL) { 2779 pr_crit("%s: Can't create sock SLAB cache!\n", 2780 prot->name); 2781 goto out; 2782 } 2783 2784 if (req_prot_init(prot)) 2785 goto out_free_request_sock_slab; 2786 2787 if (prot->twsk_prot != NULL) { 2788 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name); 2789 2790 if (prot->twsk_prot->twsk_slab_name == NULL) 2791 goto out_free_request_sock_slab; 2792 2793 prot->twsk_prot->twsk_slab = 2794 kmem_cache_create(prot->twsk_prot->twsk_slab_name, 2795 prot->twsk_prot->twsk_obj_size, 2796 0, 2797 prot->slab_flags, 2798 NULL); 2799 if (prot->twsk_prot->twsk_slab == NULL) 2800 goto out_free_timewait_sock_slab_name; 2801 } 2802 } 2803 2804 mutex_lock(&proto_list_mutex); 2805 list_add(&prot->node, &proto_list); 2806 assign_proto_idx(prot); 2807 mutex_unlock(&proto_list_mutex); 2808 return 0; 2809 2810 out_free_timewait_sock_slab_name: 2811 kfree(prot->twsk_prot->twsk_slab_name); 2812 out_free_request_sock_slab: 2813 req_prot_cleanup(prot->rsk_prot); 2814 2815 kmem_cache_destroy(prot->slab); 2816 prot->slab = NULL; 2817 out: 2818 return -ENOBUFS; 2819 } 2820 EXPORT_SYMBOL(proto_register); 2821 2822 void proto_unregister(struct proto *prot) 2823 { 2824 mutex_lock(&proto_list_mutex); 2825 release_proto_idx(prot); 2826 list_del(&prot->node); 2827 mutex_unlock(&proto_list_mutex); 2828 2829 kmem_cache_destroy(prot->slab); 2830 prot->slab = NULL; 2831 2832 req_prot_cleanup(prot->rsk_prot); 2833 2834 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) { 2835 kmem_cache_destroy(prot->twsk_prot->twsk_slab); 2836 kfree(prot->twsk_prot->twsk_slab_name); 2837 prot->twsk_prot->twsk_slab = NULL; 2838 } 2839 } 2840 EXPORT_SYMBOL(proto_unregister); 2841 2842 #ifdef CONFIG_PROC_FS 2843 static void *proto_seq_start(struct seq_file *seq, loff_t *pos) 2844 __acquires(proto_list_mutex) 2845 { 2846 mutex_lock(&proto_list_mutex); 2847 return seq_list_start_head(&proto_list, *pos); 2848 } 2849 2850 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos) 2851 { 2852 return seq_list_next(v, &proto_list, pos); 2853 } 2854 2855 static void proto_seq_stop(struct seq_file *seq, void *v) 2856 __releases(proto_list_mutex) 2857 { 2858 mutex_unlock(&proto_list_mutex); 2859 } 2860 2861 static char proto_method_implemented(const void *method) 2862 { 2863 return method == NULL ? 'n' : 'y'; 2864 } 2865 static long sock_prot_memory_allocated(struct proto *proto) 2866 { 2867 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L; 2868 } 2869 2870 static char *sock_prot_memory_pressure(struct proto *proto) 2871 { 2872 return proto->memory_pressure != NULL ? 2873 proto_memory_pressure(proto) ? "yes" : "no" : "NI"; 2874 } 2875 2876 static void proto_seq_printf(struct seq_file *seq, struct proto *proto) 2877 { 2878 2879 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s " 2880 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n", 2881 proto->name, 2882 proto->obj_size, 2883 sock_prot_inuse_get(seq_file_net(seq), proto), 2884 sock_prot_memory_allocated(proto), 2885 sock_prot_memory_pressure(proto), 2886 proto->max_header, 2887 proto->slab == NULL ? "no" : "yes", 2888 module_name(proto->owner), 2889 proto_method_implemented(proto->close), 2890 proto_method_implemented(proto->connect), 2891 proto_method_implemented(proto->disconnect), 2892 proto_method_implemented(proto->accept), 2893 proto_method_implemented(proto->ioctl), 2894 proto_method_implemented(proto->init), 2895 proto_method_implemented(proto->destroy), 2896 proto_method_implemented(proto->shutdown), 2897 proto_method_implemented(proto->setsockopt), 2898 proto_method_implemented(proto->getsockopt), 2899 proto_method_implemented(proto->sendmsg), 2900 proto_method_implemented(proto->recvmsg), 2901 proto_method_implemented(proto->sendpage), 2902 proto_method_implemented(proto->bind), 2903 proto_method_implemented(proto->backlog_rcv), 2904 proto_method_implemented(proto->hash), 2905 proto_method_implemented(proto->unhash), 2906 proto_method_implemented(proto->get_port), 2907 proto_method_implemented(proto->enter_memory_pressure)); 2908 } 2909 2910 static int proto_seq_show(struct seq_file *seq, void *v) 2911 { 2912 if (v == &proto_list) 2913 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s", 2914 "protocol", 2915 "size", 2916 "sockets", 2917 "memory", 2918 "press", 2919 "maxhdr", 2920 "slab", 2921 "module", 2922 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n"); 2923 else 2924 proto_seq_printf(seq, list_entry(v, struct proto, node)); 2925 return 0; 2926 } 2927 2928 static const struct seq_operations proto_seq_ops = { 2929 .start = proto_seq_start, 2930 .next = proto_seq_next, 2931 .stop = proto_seq_stop, 2932 .show = proto_seq_show, 2933 }; 2934 2935 static int proto_seq_open(struct inode *inode, struct file *file) 2936 { 2937 return seq_open_net(inode, file, &proto_seq_ops, 2938 sizeof(struct seq_net_private)); 2939 } 2940 2941 static const struct file_operations proto_seq_fops = { 2942 .owner = THIS_MODULE, 2943 .open = proto_seq_open, 2944 .read = seq_read, 2945 .llseek = seq_lseek, 2946 .release = seq_release_net, 2947 }; 2948 2949 static __net_init int proto_init_net(struct net *net) 2950 { 2951 if (!proc_create("protocols", S_IRUGO, net->proc_net, &proto_seq_fops)) 2952 return -ENOMEM; 2953 2954 return 0; 2955 } 2956 2957 static __net_exit void proto_exit_net(struct net *net) 2958 { 2959 remove_proc_entry("protocols", net->proc_net); 2960 } 2961 2962 2963 static __net_initdata struct pernet_operations proto_net_ops = { 2964 .init = proto_init_net, 2965 .exit = proto_exit_net, 2966 }; 2967 2968 static int __init proto_init(void) 2969 { 2970 return register_pernet_subsys(&proto_net_ops); 2971 } 2972 2973 subsys_initcall(proto_init); 2974 2975 #endif /* PROC_FS */ 2976