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 #include <net/sock_reuseport.h> 138 139 #include <trace/events/sock.h> 140 141 #ifdef CONFIG_INET 142 #include <net/tcp.h> 143 #endif 144 145 #include <net/busy_poll.h> 146 147 static DEFINE_MUTEX(proto_list_mutex); 148 static LIST_HEAD(proto_list); 149 150 /** 151 * sk_ns_capable - General socket capability test 152 * @sk: Socket to use a capability on or through 153 * @user_ns: The user namespace of the capability to use 154 * @cap: The capability to use 155 * 156 * Test to see if the opener of the socket had when the socket was 157 * created and the current process has the capability @cap in the user 158 * namespace @user_ns. 159 */ 160 bool sk_ns_capable(const struct sock *sk, 161 struct user_namespace *user_ns, int cap) 162 { 163 return file_ns_capable(sk->sk_socket->file, user_ns, cap) && 164 ns_capable(user_ns, cap); 165 } 166 EXPORT_SYMBOL(sk_ns_capable); 167 168 /** 169 * sk_capable - Socket global capability test 170 * @sk: Socket to use a capability on or through 171 * @cap: The global capability to use 172 * 173 * Test to see if the opener of the socket had when the socket was 174 * created and the current process has the capability @cap in all user 175 * namespaces. 176 */ 177 bool sk_capable(const struct sock *sk, int cap) 178 { 179 return sk_ns_capable(sk, &init_user_ns, cap); 180 } 181 EXPORT_SYMBOL(sk_capable); 182 183 /** 184 * sk_net_capable - Network namespace socket capability test 185 * @sk: Socket to use a capability on or through 186 * @cap: The capability to use 187 * 188 * Test to see if the opener of the socket had when the socket was created 189 * and the current process has the capability @cap over the network namespace 190 * the socket is a member of. 191 */ 192 bool sk_net_capable(const struct sock *sk, int cap) 193 { 194 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap); 195 } 196 EXPORT_SYMBOL(sk_net_capable); 197 198 199 #ifdef CONFIG_MEMCG_KMEM 200 int mem_cgroup_sockets_init(struct mem_cgroup *memcg, struct cgroup_subsys *ss) 201 { 202 struct proto *proto; 203 int ret = 0; 204 205 mutex_lock(&proto_list_mutex); 206 list_for_each_entry(proto, &proto_list, node) { 207 if (proto->init_cgroup) { 208 ret = proto->init_cgroup(memcg, ss); 209 if (ret) 210 goto out; 211 } 212 } 213 214 mutex_unlock(&proto_list_mutex); 215 return ret; 216 out: 217 list_for_each_entry_continue_reverse(proto, &proto_list, node) 218 if (proto->destroy_cgroup) 219 proto->destroy_cgroup(memcg); 220 mutex_unlock(&proto_list_mutex); 221 return ret; 222 } 223 224 void mem_cgroup_sockets_destroy(struct mem_cgroup *memcg) 225 { 226 struct proto *proto; 227 228 mutex_lock(&proto_list_mutex); 229 list_for_each_entry_reverse(proto, &proto_list, node) 230 if (proto->destroy_cgroup) 231 proto->destroy_cgroup(memcg); 232 mutex_unlock(&proto_list_mutex); 233 } 234 #endif 235 236 /* 237 * Each address family might have different locking rules, so we have 238 * one slock key per address family: 239 */ 240 static struct lock_class_key af_family_keys[AF_MAX]; 241 static struct lock_class_key af_family_slock_keys[AF_MAX]; 242 243 #if defined(CONFIG_MEMCG_KMEM) 244 struct static_key memcg_socket_limit_enabled; 245 EXPORT_SYMBOL(memcg_socket_limit_enabled); 246 #endif 247 248 /* 249 * Make lock validator output more readable. (we pre-construct these 250 * strings build-time, so that runtime initialization of socket 251 * locks is fast): 252 */ 253 static const char *const af_family_key_strings[AF_MAX+1] = { 254 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" , 255 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK", 256 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" , 257 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" , 258 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" , 259 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" , 260 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" , 261 "sk_lock-AF_RDS" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" , 262 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" , 263 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" , 264 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" , 265 "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN" , "sk_lock-AF_PHONET" , 266 "sk_lock-AF_IEEE802154", "sk_lock-AF_CAIF" , "sk_lock-AF_ALG" , 267 "sk_lock-AF_NFC" , "sk_lock-AF_VSOCK" , "sk_lock-AF_MAX" 268 }; 269 static const char *const af_family_slock_key_strings[AF_MAX+1] = { 270 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" , 271 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK", 272 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" , 273 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" , 274 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" , 275 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" , 276 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" , 277 "slock-AF_RDS" , "slock-AF_SNA" , "slock-AF_IRDA" , 278 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" , 279 "slock-27" , "slock-28" , "slock-AF_CAN" , 280 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" , 281 "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" , 282 "slock-AF_IEEE802154", "slock-AF_CAIF" , "slock-AF_ALG" , 283 "slock-AF_NFC" , "slock-AF_VSOCK" ,"slock-AF_MAX" 284 }; 285 static const char *const af_family_clock_key_strings[AF_MAX+1] = { 286 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" , 287 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK", 288 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" , 289 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" , 290 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" , 291 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" , 292 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" , 293 "clock-AF_RDS" , "clock-AF_SNA" , "clock-AF_IRDA" , 294 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" , 295 "clock-27" , "clock-28" , "clock-AF_CAN" , 296 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" , 297 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" , 298 "clock-AF_IEEE802154", "clock-AF_CAIF" , "clock-AF_ALG" , 299 "clock-AF_NFC" , "clock-AF_VSOCK" , "clock-AF_MAX" 300 }; 301 302 /* 303 * sk_callback_lock locking rules are per-address-family, 304 * so split the lock classes by using a per-AF key: 305 */ 306 static struct lock_class_key af_callback_keys[AF_MAX]; 307 308 /* Take into consideration the size of the struct sk_buff overhead in the 309 * determination of these values, since that is non-constant across 310 * platforms. This makes socket queueing behavior and performance 311 * not depend upon such differences. 312 */ 313 #define _SK_MEM_PACKETS 256 314 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256) 315 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 316 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 317 318 /* Run time adjustable parameters. */ 319 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX; 320 EXPORT_SYMBOL(sysctl_wmem_max); 321 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX; 322 EXPORT_SYMBOL(sysctl_rmem_max); 323 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX; 324 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX; 325 326 /* Maximal space eaten by iovec or ancillary data plus some space */ 327 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512); 328 EXPORT_SYMBOL(sysctl_optmem_max); 329 330 int sysctl_tstamp_allow_data __read_mostly = 1; 331 332 struct static_key memalloc_socks = STATIC_KEY_INIT_FALSE; 333 EXPORT_SYMBOL_GPL(memalloc_socks); 334 335 /** 336 * sk_set_memalloc - sets %SOCK_MEMALLOC 337 * @sk: socket to set it on 338 * 339 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves. 340 * It's the responsibility of the admin to adjust min_free_kbytes 341 * to meet the requirements 342 */ 343 void sk_set_memalloc(struct sock *sk) 344 { 345 sock_set_flag(sk, SOCK_MEMALLOC); 346 sk->sk_allocation |= __GFP_MEMALLOC; 347 static_key_slow_inc(&memalloc_socks); 348 } 349 EXPORT_SYMBOL_GPL(sk_set_memalloc); 350 351 void sk_clear_memalloc(struct sock *sk) 352 { 353 sock_reset_flag(sk, SOCK_MEMALLOC); 354 sk->sk_allocation &= ~__GFP_MEMALLOC; 355 static_key_slow_dec(&memalloc_socks); 356 357 /* 358 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward 359 * progress of swapping. SOCK_MEMALLOC may be cleared while 360 * it has rmem allocations due to the last swapfile being deactivated 361 * but there is a risk that the socket is unusable due to exceeding 362 * the rmem limits. Reclaim the reserves and obey rmem limits again. 363 */ 364 sk_mem_reclaim(sk); 365 } 366 EXPORT_SYMBOL_GPL(sk_clear_memalloc); 367 368 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb) 369 { 370 int ret; 371 unsigned long pflags = current->flags; 372 373 /* these should have been dropped before queueing */ 374 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC)); 375 376 current->flags |= PF_MEMALLOC; 377 ret = sk->sk_backlog_rcv(sk, skb); 378 tsk_restore_flags(current, pflags, PF_MEMALLOC); 379 380 return ret; 381 } 382 EXPORT_SYMBOL(__sk_backlog_rcv); 383 384 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen) 385 { 386 struct timeval tv; 387 388 if (optlen < sizeof(tv)) 389 return -EINVAL; 390 if (copy_from_user(&tv, optval, sizeof(tv))) 391 return -EFAULT; 392 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC) 393 return -EDOM; 394 395 if (tv.tv_sec < 0) { 396 static int warned __read_mostly; 397 398 *timeo_p = 0; 399 if (warned < 10 && net_ratelimit()) { 400 warned++; 401 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n", 402 __func__, current->comm, task_pid_nr(current)); 403 } 404 return 0; 405 } 406 *timeo_p = MAX_SCHEDULE_TIMEOUT; 407 if (tv.tv_sec == 0 && tv.tv_usec == 0) 408 return 0; 409 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1)) 410 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ); 411 return 0; 412 } 413 414 static void sock_warn_obsolete_bsdism(const char *name) 415 { 416 static int warned; 417 static char warncomm[TASK_COMM_LEN]; 418 if (strcmp(warncomm, current->comm) && warned < 5) { 419 strcpy(warncomm, current->comm); 420 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n", 421 warncomm, name); 422 warned++; 423 } 424 } 425 426 static bool sock_needs_netstamp(const struct sock *sk) 427 { 428 switch (sk->sk_family) { 429 case AF_UNSPEC: 430 case AF_UNIX: 431 return false; 432 default: 433 return true; 434 } 435 } 436 437 static void sock_disable_timestamp(struct sock *sk, unsigned long flags) 438 { 439 if (sk->sk_flags & flags) { 440 sk->sk_flags &= ~flags; 441 if (sock_needs_netstamp(sk) && 442 !(sk->sk_flags & SK_FLAGS_TIMESTAMP)) 443 net_disable_timestamp(); 444 } 445 } 446 447 448 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 449 { 450 int err; 451 unsigned long flags; 452 struct sk_buff_head *list = &sk->sk_receive_queue; 453 454 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) { 455 atomic_inc(&sk->sk_drops); 456 trace_sock_rcvqueue_full(sk, skb); 457 return -ENOMEM; 458 } 459 460 err = sk_filter(sk, skb); 461 if (err) 462 return err; 463 464 if (!sk_rmem_schedule(sk, skb, skb->truesize)) { 465 atomic_inc(&sk->sk_drops); 466 return -ENOBUFS; 467 } 468 469 skb->dev = NULL; 470 skb_set_owner_r(skb, sk); 471 472 /* we escape from rcu protected region, make sure we dont leak 473 * a norefcounted dst 474 */ 475 skb_dst_force(skb); 476 477 spin_lock_irqsave(&list->lock, flags); 478 sock_skb_set_dropcount(sk, skb); 479 __skb_queue_tail(list, skb); 480 spin_unlock_irqrestore(&list->lock, flags); 481 482 if (!sock_flag(sk, SOCK_DEAD)) 483 sk->sk_data_ready(sk); 484 return 0; 485 } 486 EXPORT_SYMBOL(sock_queue_rcv_skb); 487 488 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested) 489 { 490 int rc = NET_RX_SUCCESS; 491 492 if (sk_filter(sk, skb)) 493 goto discard_and_relse; 494 495 skb->dev = NULL; 496 497 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) { 498 atomic_inc(&sk->sk_drops); 499 goto discard_and_relse; 500 } 501 if (nested) 502 bh_lock_sock_nested(sk); 503 else 504 bh_lock_sock(sk); 505 if (!sock_owned_by_user(sk)) { 506 /* 507 * trylock + unlock semantics: 508 */ 509 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_); 510 511 rc = sk_backlog_rcv(sk, skb); 512 513 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_); 514 } else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) { 515 bh_unlock_sock(sk); 516 atomic_inc(&sk->sk_drops); 517 goto discard_and_relse; 518 } 519 520 bh_unlock_sock(sk); 521 out: 522 sock_put(sk); 523 return rc; 524 discard_and_relse: 525 kfree_skb(skb); 526 goto out; 527 } 528 EXPORT_SYMBOL(sk_receive_skb); 529 530 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie) 531 { 532 struct dst_entry *dst = __sk_dst_get(sk); 533 534 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) { 535 sk_tx_queue_clear(sk); 536 RCU_INIT_POINTER(sk->sk_dst_cache, NULL); 537 dst_release(dst); 538 return NULL; 539 } 540 541 return dst; 542 } 543 EXPORT_SYMBOL(__sk_dst_check); 544 545 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie) 546 { 547 struct dst_entry *dst = sk_dst_get(sk); 548 549 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) { 550 sk_dst_reset(sk); 551 dst_release(dst); 552 return NULL; 553 } 554 555 return dst; 556 } 557 EXPORT_SYMBOL(sk_dst_check); 558 559 static int sock_setbindtodevice(struct sock *sk, char __user *optval, 560 int optlen) 561 { 562 int ret = -ENOPROTOOPT; 563 #ifdef CONFIG_NETDEVICES 564 struct net *net = sock_net(sk); 565 char devname[IFNAMSIZ]; 566 int index; 567 568 /* Sorry... */ 569 ret = -EPERM; 570 if (!ns_capable(net->user_ns, CAP_NET_RAW)) 571 goto out; 572 573 ret = -EINVAL; 574 if (optlen < 0) 575 goto out; 576 577 /* Bind this socket to a particular device like "eth0", 578 * as specified in the passed interface name. If the 579 * name is "" or the option length is zero the socket 580 * is not bound. 581 */ 582 if (optlen > IFNAMSIZ - 1) 583 optlen = IFNAMSIZ - 1; 584 memset(devname, 0, sizeof(devname)); 585 586 ret = -EFAULT; 587 if (copy_from_user(devname, optval, optlen)) 588 goto out; 589 590 index = 0; 591 if (devname[0] != '\0') { 592 struct net_device *dev; 593 594 rcu_read_lock(); 595 dev = dev_get_by_name_rcu(net, devname); 596 if (dev) 597 index = dev->ifindex; 598 rcu_read_unlock(); 599 ret = -ENODEV; 600 if (!dev) 601 goto out; 602 } 603 604 lock_sock(sk); 605 sk->sk_bound_dev_if = index; 606 sk_dst_reset(sk); 607 release_sock(sk); 608 609 ret = 0; 610 611 out: 612 #endif 613 614 return ret; 615 } 616 617 static int sock_getbindtodevice(struct sock *sk, char __user *optval, 618 int __user *optlen, int len) 619 { 620 int ret = -ENOPROTOOPT; 621 #ifdef CONFIG_NETDEVICES 622 struct net *net = sock_net(sk); 623 char devname[IFNAMSIZ]; 624 625 if (sk->sk_bound_dev_if == 0) { 626 len = 0; 627 goto zero; 628 } 629 630 ret = -EINVAL; 631 if (len < IFNAMSIZ) 632 goto out; 633 634 ret = netdev_get_name(net, devname, sk->sk_bound_dev_if); 635 if (ret) 636 goto out; 637 638 len = strlen(devname) + 1; 639 640 ret = -EFAULT; 641 if (copy_to_user(optval, devname, len)) 642 goto out; 643 644 zero: 645 ret = -EFAULT; 646 if (put_user(len, optlen)) 647 goto out; 648 649 ret = 0; 650 651 out: 652 #endif 653 654 return ret; 655 } 656 657 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool) 658 { 659 if (valbool) 660 sock_set_flag(sk, bit); 661 else 662 sock_reset_flag(sk, bit); 663 } 664 665 bool sk_mc_loop(struct sock *sk) 666 { 667 if (dev_recursion_level()) 668 return false; 669 if (!sk) 670 return true; 671 switch (sk->sk_family) { 672 case AF_INET: 673 return inet_sk(sk)->mc_loop; 674 #if IS_ENABLED(CONFIG_IPV6) 675 case AF_INET6: 676 return inet6_sk(sk)->mc_loop; 677 #endif 678 } 679 WARN_ON(1); 680 return true; 681 } 682 EXPORT_SYMBOL(sk_mc_loop); 683 684 /* 685 * This is meant for all protocols to use and covers goings on 686 * at the socket level. Everything here is generic. 687 */ 688 689 int sock_setsockopt(struct socket *sock, int level, int optname, 690 char __user *optval, unsigned int optlen) 691 { 692 struct sock *sk = sock->sk; 693 int val; 694 int valbool; 695 struct linger ling; 696 int ret = 0; 697 698 /* 699 * Options without arguments 700 */ 701 702 if (optname == SO_BINDTODEVICE) 703 return sock_setbindtodevice(sk, optval, optlen); 704 705 if (optlen < sizeof(int)) 706 return -EINVAL; 707 708 if (get_user(val, (int __user *)optval)) 709 return -EFAULT; 710 711 valbool = val ? 1 : 0; 712 713 lock_sock(sk); 714 715 switch (optname) { 716 case SO_DEBUG: 717 if (val && !capable(CAP_NET_ADMIN)) 718 ret = -EACCES; 719 else 720 sock_valbool_flag(sk, SOCK_DBG, valbool); 721 break; 722 case SO_REUSEADDR: 723 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE); 724 break; 725 case SO_REUSEPORT: 726 sk->sk_reuseport = valbool; 727 break; 728 case SO_TYPE: 729 case SO_PROTOCOL: 730 case SO_DOMAIN: 731 case SO_ERROR: 732 ret = -ENOPROTOOPT; 733 break; 734 case SO_DONTROUTE: 735 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool); 736 break; 737 case SO_BROADCAST: 738 sock_valbool_flag(sk, SOCK_BROADCAST, valbool); 739 break; 740 case SO_SNDBUF: 741 /* Don't error on this BSD doesn't and if you think 742 * about it this is right. Otherwise apps have to 743 * play 'guess the biggest size' games. RCVBUF/SNDBUF 744 * are treated in BSD as hints 745 */ 746 val = min_t(u32, val, sysctl_wmem_max); 747 set_sndbuf: 748 sk->sk_userlocks |= SOCK_SNDBUF_LOCK; 749 sk->sk_sndbuf = max_t(u32, val * 2, SOCK_MIN_SNDBUF); 750 /* Wake up sending tasks if we upped the value. */ 751 sk->sk_write_space(sk); 752 break; 753 754 case SO_SNDBUFFORCE: 755 if (!capable(CAP_NET_ADMIN)) { 756 ret = -EPERM; 757 break; 758 } 759 goto set_sndbuf; 760 761 case SO_RCVBUF: 762 /* Don't error on this BSD doesn't and if you think 763 * about it this is right. Otherwise apps have to 764 * play 'guess the biggest size' games. RCVBUF/SNDBUF 765 * are treated in BSD as hints 766 */ 767 val = min_t(u32, val, sysctl_rmem_max); 768 set_rcvbuf: 769 sk->sk_userlocks |= SOCK_RCVBUF_LOCK; 770 /* 771 * We double it on the way in to account for 772 * "struct sk_buff" etc. overhead. Applications 773 * assume that the SO_RCVBUF setting they make will 774 * allow that much actual data to be received on that 775 * socket. 776 * 777 * Applications are unaware that "struct sk_buff" and 778 * other overheads allocate from the receive buffer 779 * during socket buffer allocation. 780 * 781 * And after considering the possible alternatives, 782 * returning the value we actually used in getsockopt 783 * is the most desirable behavior. 784 */ 785 sk->sk_rcvbuf = max_t(u32, val * 2, SOCK_MIN_RCVBUF); 786 break; 787 788 case SO_RCVBUFFORCE: 789 if (!capable(CAP_NET_ADMIN)) { 790 ret = -EPERM; 791 break; 792 } 793 goto set_rcvbuf; 794 795 case SO_KEEPALIVE: 796 #ifdef CONFIG_INET 797 if (sk->sk_protocol == IPPROTO_TCP && 798 sk->sk_type == SOCK_STREAM) 799 tcp_set_keepalive(sk, valbool); 800 #endif 801 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool); 802 break; 803 804 case SO_OOBINLINE: 805 sock_valbool_flag(sk, SOCK_URGINLINE, valbool); 806 break; 807 808 case SO_NO_CHECK: 809 sk->sk_no_check_tx = valbool; 810 break; 811 812 case SO_PRIORITY: 813 if ((val >= 0 && val <= 6) || 814 ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) 815 sk->sk_priority = val; 816 else 817 ret = -EPERM; 818 break; 819 820 case SO_LINGER: 821 if (optlen < sizeof(ling)) { 822 ret = -EINVAL; /* 1003.1g */ 823 break; 824 } 825 if (copy_from_user(&ling, optval, sizeof(ling))) { 826 ret = -EFAULT; 827 break; 828 } 829 if (!ling.l_onoff) 830 sock_reset_flag(sk, SOCK_LINGER); 831 else { 832 #if (BITS_PER_LONG == 32) 833 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ) 834 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT; 835 else 836 #endif 837 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ; 838 sock_set_flag(sk, SOCK_LINGER); 839 } 840 break; 841 842 case SO_BSDCOMPAT: 843 sock_warn_obsolete_bsdism("setsockopt"); 844 break; 845 846 case SO_PASSCRED: 847 if (valbool) 848 set_bit(SOCK_PASSCRED, &sock->flags); 849 else 850 clear_bit(SOCK_PASSCRED, &sock->flags); 851 break; 852 853 case SO_TIMESTAMP: 854 case SO_TIMESTAMPNS: 855 if (valbool) { 856 if (optname == SO_TIMESTAMP) 857 sock_reset_flag(sk, SOCK_RCVTSTAMPNS); 858 else 859 sock_set_flag(sk, SOCK_RCVTSTAMPNS); 860 sock_set_flag(sk, SOCK_RCVTSTAMP); 861 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 862 } else { 863 sock_reset_flag(sk, SOCK_RCVTSTAMP); 864 sock_reset_flag(sk, SOCK_RCVTSTAMPNS); 865 } 866 break; 867 868 case SO_TIMESTAMPING: 869 if (val & ~SOF_TIMESTAMPING_MASK) { 870 ret = -EINVAL; 871 break; 872 } 873 874 if (val & SOF_TIMESTAMPING_OPT_ID && 875 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) { 876 if (sk->sk_protocol == IPPROTO_TCP && 877 sk->sk_type == SOCK_STREAM) { 878 if (sk->sk_state != TCP_ESTABLISHED) { 879 ret = -EINVAL; 880 break; 881 } 882 sk->sk_tskey = tcp_sk(sk)->snd_una; 883 } else { 884 sk->sk_tskey = 0; 885 } 886 } 887 sk->sk_tsflags = val; 888 if (val & SOF_TIMESTAMPING_RX_SOFTWARE) 889 sock_enable_timestamp(sk, 890 SOCK_TIMESTAMPING_RX_SOFTWARE); 891 else 892 sock_disable_timestamp(sk, 893 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE)); 894 break; 895 896 case SO_RCVLOWAT: 897 if (val < 0) 898 val = INT_MAX; 899 sk->sk_rcvlowat = val ? : 1; 900 break; 901 902 case SO_RCVTIMEO: 903 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen); 904 break; 905 906 case SO_SNDTIMEO: 907 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen); 908 break; 909 910 case SO_ATTACH_FILTER: 911 ret = -EINVAL; 912 if (optlen == sizeof(struct sock_fprog)) { 913 struct sock_fprog fprog; 914 915 ret = -EFAULT; 916 if (copy_from_user(&fprog, optval, sizeof(fprog))) 917 break; 918 919 ret = sk_attach_filter(&fprog, sk); 920 } 921 break; 922 923 case SO_ATTACH_BPF: 924 ret = -EINVAL; 925 if (optlen == sizeof(u32)) { 926 u32 ufd; 927 928 ret = -EFAULT; 929 if (copy_from_user(&ufd, optval, sizeof(ufd))) 930 break; 931 932 ret = sk_attach_bpf(ufd, sk); 933 } 934 break; 935 936 case SO_ATTACH_REUSEPORT_CBPF: 937 ret = -EINVAL; 938 if (optlen == sizeof(struct sock_fprog)) { 939 struct sock_fprog fprog; 940 941 ret = -EFAULT; 942 if (copy_from_user(&fprog, optval, sizeof(fprog))) 943 break; 944 945 ret = sk_reuseport_attach_filter(&fprog, sk); 946 } 947 break; 948 949 case SO_ATTACH_REUSEPORT_EBPF: 950 ret = -EINVAL; 951 if (optlen == sizeof(u32)) { 952 u32 ufd; 953 954 ret = -EFAULT; 955 if (copy_from_user(&ufd, optval, sizeof(ufd))) 956 break; 957 958 ret = sk_reuseport_attach_bpf(ufd, sk); 959 } 960 break; 961 962 case SO_DETACH_FILTER: 963 ret = sk_detach_filter(sk); 964 break; 965 966 case SO_LOCK_FILTER: 967 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool) 968 ret = -EPERM; 969 else 970 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool); 971 break; 972 973 case SO_PASSSEC: 974 if (valbool) 975 set_bit(SOCK_PASSSEC, &sock->flags); 976 else 977 clear_bit(SOCK_PASSSEC, &sock->flags); 978 break; 979 case SO_MARK: 980 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) 981 ret = -EPERM; 982 else 983 sk->sk_mark = val; 984 break; 985 986 case SO_RXQ_OVFL: 987 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool); 988 break; 989 990 case SO_WIFI_STATUS: 991 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool); 992 break; 993 994 case SO_PEEK_OFF: 995 if (sock->ops->set_peek_off) 996 ret = sock->ops->set_peek_off(sk, val); 997 else 998 ret = -EOPNOTSUPP; 999 break; 1000 1001 case SO_NOFCS: 1002 sock_valbool_flag(sk, SOCK_NOFCS, valbool); 1003 break; 1004 1005 case SO_SELECT_ERR_QUEUE: 1006 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool); 1007 break; 1008 1009 #ifdef CONFIG_NET_RX_BUSY_POLL 1010 case SO_BUSY_POLL: 1011 /* allow unprivileged users to decrease the value */ 1012 if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN)) 1013 ret = -EPERM; 1014 else { 1015 if (val < 0) 1016 ret = -EINVAL; 1017 else 1018 sk->sk_ll_usec = val; 1019 } 1020 break; 1021 #endif 1022 1023 case SO_MAX_PACING_RATE: 1024 sk->sk_max_pacing_rate = val; 1025 sk->sk_pacing_rate = min(sk->sk_pacing_rate, 1026 sk->sk_max_pacing_rate); 1027 break; 1028 1029 case SO_INCOMING_CPU: 1030 sk->sk_incoming_cpu = val; 1031 break; 1032 1033 default: 1034 ret = -ENOPROTOOPT; 1035 break; 1036 } 1037 release_sock(sk); 1038 return ret; 1039 } 1040 EXPORT_SYMBOL(sock_setsockopt); 1041 1042 1043 static void cred_to_ucred(struct pid *pid, const struct cred *cred, 1044 struct ucred *ucred) 1045 { 1046 ucred->pid = pid_vnr(pid); 1047 ucred->uid = ucred->gid = -1; 1048 if (cred) { 1049 struct user_namespace *current_ns = current_user_ns(); 1050 1051 ucred->uid = from_kuid_munged(current_ns, cred->euid); 1052 ucred->gid = from_kgid_munged(current_ns, cred->egid); 1053 } 1054 } 1055 1056 int sock_getsockopt(struct socket *sock, int level, int optname, 1057 char __user *optval, int __user *optlen) 1058 { 1059 struct sock *sk = sock->sk; 1060 1061 union { 1062 int val; 1063 struct linger ling; 1064 struct timeval tm; 1065 } v; 1066 1067 int lv = sizeof(int); 1068 int len; 1069 1070 if (get_user(len, optlen)) 1071 return -EFAULT; 1072 if (len < 0) 1073 return -EINVAL; 1074 1075 memset(&v, 0, sizeof(v)); 1076 1077 switch (optname) { 1078 case SO_DEBUG: 1079 v.val = sock_flag(sk, SOCK_DBG); 1080 break; 1081 1082 case SO_DONTROUTE: 1083 v.val = sock_flag(sk, SOCK_LOCALROUTE); 1084 break; 1085 1086 case SO_BROADCAST: 1087 v.val = sock_flag(sk, SOCK_BROADCAST); 1088 break; 1089 1090 case SO_SNDBUF: 1091 v.val = sk->sk_sndbuf; 1092 break; 1093 1094 case SO_RCVBUF: 1095 v.val = sk->sk_rcvbuf; 1096 break; 1097 1098 case SO_REUSEADDR: 1099 v.val = sk->sk_reuse; 1100 break; 1101 1102 case SO_REUSEPORT: 1103 v.val = sk->sk_reuseport; 1104 break; 1105 1106 case SO_KEEPALIVE: 1107 v.val = sock_flag(sk, SOCK_KEEPOPEN); 1108 break; 1109 1110 case SO_TYPE: 1111 v.val = sk->sk_type; 1112 break; 1113 1114 case SO_PROTOCOL: 1115 v.val = sk->sk_protocol; 1116 break; 1117 1118 case SO_DOMAIN: 1119 v.val = sk->sk_family; 1120 break; 1121 1122 case SO_ERROR: 1123 v.val = -sock_error(sk); 1124 if (v.val == 0) 1125 v.val = xchg(&sk->sk_err_soft, 0); 1126 break; 1127 1128 case SO_OOBINLINE: 1129 v.val = sock_flag(sk, SOCK_URGINLINE); 1130 break; 1131 1132 case SO_NO_CHECK: 1133 v.val = sk->sk_no_check_tx; 1134 break; 1135 1136 case SO_PRIORITY: 1137 v.val = sk->sk_priority; 1138 break; 1139 1140 case SO_LINGER: 1141 lv = sizeof(v.ling); 1142 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER); 1143 v.ling.l_linger = sk->sk_lingertime / HZ; 1144 break; 1145 1146 case SO_BSDCOMPAT: 1147 sock_warn_obsolete_bsdism("getsockopt"); 1148 break; 1149 1150 case SO_TIMESTAMP: 1151 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && 1152 !sock_flag(sk, SOCK_RCVTSTAMPNS); 1153 break; 1154 1155 case SO_TIMESTAMPNS: 1156 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS); 1157 break; 1158 1159 case SO_TIMESTAMPING: 1160 v.val = sk->sk_tsflags; 1161 break; 1162 1163 case SO_RCVTIMEO: 1164 lv = sizeof(struct timeval); 1165 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) { 1166 v.tm.tv_sec = 0; 1167 v.tm.tv_usec = 0; 1168 } else { 1169 v.tm.tv_sec = sk->sk_rcvtimeo / HZ; 1170 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ; 1171 } 1172 break; 1173 1174 case SO_SNDTIMEO: 1175 lv = sizeof(struct timeval); 1176 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) { 1177 v.tm.tv_sec = 0; 1178 v.tm.tv_usec = 0; 1179 } else { 1180 v.tm.tv_sec = sk->sk_sndtimeo / HZ; 1181 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ; 1182 } 1183 break; 1184 1185 case SO_RCVLOWAT: 1186 v.val = sk->sk_rcvlowat; 1187 break; 1188 1189 case SO_SNDLOWAT: 1190 v.val = 1; 1191 break; 1192 1193 case SO_PASSCRED: 1194 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags); 1195 break; 1196 1197 case SO_PEERCRED: 1198 { 1199 struct ucred peercred; 1200 if (len > sizeof(peercred)) 1201 len = sizeof(peercred); 1202 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred); 1203 if (copy_to_user(optval, &peercred, len)) 1204 return -EFAULT; 1205 goto lenout; 1206 } 1207 1208 case SO_PEERNAME: 1209 { 1210 char address[128]; 1211 1212 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2)) 1213 return -ENOTCONN; 1214 if (lv < len) 1215 return -EINVAL; 1216 if (copy_to_user(optval, address, len)) 1217 return -EFAULT; 1218 goto lenout; 1219 } 1220 1221 /* Dubious BSD thing... Probably nobody even uses it, but 1222 * the UNIX standard wants it for whatever reason... -DaveM 1223 */ 1224 case SO_ACCEPTCONN: 1225 v.val = sk->sk_state == TCP_LISTEN; 1226 break; 1227 1228 case SO_PASSSEC: 1229 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags); 1230 break; 1231 1232 case SO_PEERSEC: 1233 return security_socket_getpeersec_stream(sock, optval, optlen, len); 1234 1235 case SO_MARK: 1236 v.val = sk->sk_mark; 1237 break; 1238 1239 case SO_RXQ_OVFL: 1240 v.val = sock_flag(sk, SOCK_RXQ_OVFL); 1241 break; 1242 1243 case SO_WIFI_STATUS: 1244 v.val = sock_flag(sk, SOCK_WIFI_STATUS); 1245 break; 1246 1247 case SO_PEEK_OFF: 1248 if (!sock->ops->set_peek_off) 1249 return -EOPNOTSUPP; 1250 1251 v.val = sk->sk_peek_off; 1252 break; 1253 case SO_NOFCS: 1254 v.val = sock_flag(sk, SOCK_NOFCS); 1255 break; 1256 1257 case SO_BINDTODEVICE: 1258 return sock_getbindtodevice(sk, optval, optlen, len); 1259 1260 case SO_GET_FILTER: 1261 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len); 1262 if (len < 0) 1263 return len; 1264 1265 goto lenout; 1266 1267 case SO_LOCK_FILTER: 1268 v.val = sock_flag(sk, SOCK_FILTER_LOCKED); 1269 break; 1270 1271 case SO_BPF_EXTENSIONS: 1272 v.val = bpf_tell_extensions(); 1273 break; 1274 1275 case SO_SELECT_ERR_QUEUE: 1276 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE); 1277 break; 1278 1279 #ifdef CONFIG_NET_RX_BUSY_POLL 1280 case SO_BUSY_POLL: 1281 v.val = sk->sk_ll_usec; 1282 break; 1283 #endif 1284 1285 case SO_MAX_PACING_RATE: 1286 v.val = sk->sk_max_pacing_rate; 1287 break; 1288 1289 case SO_INCOMING_CPU: 1290 v.val = sk->sk_incoming_cpu; 1291 break; 1292 1293 default: 1294 /* We implement the SO_SNDLOWAT etc to not be settable 1295 * (1003.1g 7). 1296 */ 1297 return -ENOPROTOOPT; 1298 } 1299 1300 if (len > lv) 1301 len = lv; 1302 if (copy_to_user(optval, &v, len)) 1303 return -EFAULT; 1304 lenout: 1305 if (put_user(len, optlen)) 1306 return -EFAULT; 1307 return 0; 1308 } 1309 1310 /* 1311 * Initialize an sk_lock. 1312 * 1313 * (We also register the sk_lock with the lock validator.) 1314 */ 1315 static inline void sock_lock_init(struct sock *sk) 1316 { 1317 sock_lock_init_class_and_name(sk, 1318 af_family_slock_key_strings[sk->sk_family], 1319 af_family_slock_keys + sk->sk_family, 1320 af_family_key_strings[sk->sk_family], 1321 af_family_keys + sk->sk_family); 1322 } 1323 1324 /* 1325 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet, 1326 * even temporarly, because of RCU lookups. sk_node should also be left as is. 1327 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end 1328 */ 1329 static void sock_copy(struct sock *nsk, const struct sock *osk) 1330 { 1331 #ifdef CONFIG_SECURITY_NETWORK 1332 void *sptr = nsk->sk_security; 1333 #endif 1334 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin)); 1335 1336 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end, 1337 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end)); 1338 1339 #ifdef CONFIG_SECURITY_NETWORK 1340 nsk->sk_security = sptr; 1341 security_sk_clone(osk, nsk); 1342 #endif 1343 } 1344 1345 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size) 1346 { 1347 unsigned long nulls1, nulls2; 1348 1349 nulls1 = offsetof(struct sock, __sk_common.skc_node.next); 1350 nulls2 = offsetof(struct sock, __sk_common.skc_portaddr_node.next); 1351 if (nulls1 > nulls2) 1352 swap(nulls1, nulls2); 1353 1354 if (nulls1 != 0) 1355 memset((char *)sk, 0, nulls1); 1356 memset((char *)sk + nulls1 + sizeof(void *), 0, 1357 nulls2 - nulls1 - sizeof(void *)); 1358 memset((char *)sk + nulls2 + sizeof(void *), 0, 1359 size - nulls2 - sizeof(void *)); 1360 } 1361 EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls); 1362 1363 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority, 1364 int family) 1365 { 1366 struct sock *sk; 1367 struct kmem_cache *slab; 1368 1369 slab = prot->slab; 1370 if (slab != NULL) { 1371 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO); 1372 if (!sk) 1373 return sk; 1374 if (priority & __GFP_ZERO) { 1375 if (prot->clear_sk) 1376 prot->clear_sk(sk, prot->obj_size); 1377 else 1378 sk_prot_clear_nulls(sk, prot->obj_size); 1379 } 1380 } else 1381 sk = kmalloc(prot->obj_size, priority); 1382 1383 if (sk != NULL) { 1384 kmemcheck_annotate_bitfield(sk, flags); 1385 1386 if (security_sk_alloc(sk, family, priority)) 1387 goto out_free; 1388 1389 if (!try_module_get(prot->owner)) 1390 goto out_free_sec; 1391 sk_tx_queue_clear(sk); 1392 cgroup_sk_alloc(&sk->sk_cgrp_data); 1393 } 1394 1395 return sk; 1396 1397 out_free_sec: 1398 security_sk_free(sk); 1399 out_free: 1400 if (slab != NULL) 1401 kmem_cache_free(slab, sk); 1402 else 1403 kfree(sk); 1404 return NULL; 1405 } 1406 1407 static void sk_prot_free(struct proto *prot, struct sock *sk) 1408 { 1409 struct kmem_cache *slab; 1410 struct module *owner; 1411 1412 owner = prot->owner; 1413 slab = prot->slab; 1414 1415 cgroup_sk_free(&sk->sk_cgrp_data); 1416 security_sk_free(sk); 1417 if (slab != NULL) 1418 kmem_cache_free(slab, sk); 1419 else 1420 kfree(sk); 1421 module_put(owner); 1422 } 1423 1424 /** 1425 * sk_alloc - All socket objects are allocated here 1426 * @net: the applicable net namespace 1427 * @family: protocol family 1428 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 1429 * @prot: struct proto associated with this new sock instance 1430 * @kern: is this to be a kernel socket? 1431 */ 1432 struct sock *sk_alloc(struct net *net, int family, gfp_t priority, 1433 struct proto *prot, int kern) 1434 { 1435 struct sock *sk; 1436 1437 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family); 1438 if (sk) { 1439 sk->sk_family = family; 1440 /* 1441 * See comment in struct sock definition to understand 1442 * why we need sk_prot_creator -acme 1443 */ 1444 sk->sk_prot = sk->sk_prot_creator = prot; 1445 sock_lock_init(sk); 1446 sk->sk_net_refcnt = kern ? 0 : 1; 1447 if (likely(sk->sk_net_refcnt)) 1448 get_net(net); 1449 sock_net_set(sk, net); 1450 atomic_set(&sk->sk_wmem_alloc, 1); 1451 1452 sock_update_classid(&sk->sk_cgrp_data); 1453 sock_update_netprioidx(&sk->sk_cgrp_data); 1454 } 1455 1456 return sk; 1457 } 1458 EXPORT_SYMBOL(sk_alloc); 1459 1460 void sk_destruct(struct sock *sk) 1461 { 1462 struct sk_filter *filter; 1463 1464 if (sk->sk_destruct) 1465 sk->sk_destruct(sk); 1466 1467 filter = rcu_dereference_check(sk->sk_filter, 1468 atomic_read(&sk->sk_wmem_alloc) == 0); 1469 if (filter) { 1470 sk_filter_uncharge(sk, filter); 1471 RCU_INIT_POINTER(sk->sk_filter, NULL); 1472 } 1473 if (rcu_access_pointer(sk->sk_reuseport_cb)) 1474 reuseport_detach_sock(sk); 1475 1476 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP); 1477 1478 if (atomic_read(&sk->sk_omem_alloc)) 1479 pr_debug("%s: optmem leakage (%d bytes) detected\n", 1480 __func__, atomic_read(&sk->sk_omem_alloc)); 1481 1482 if (sk->sk_peer_cred) 1483 put_cred(sk->sk_peer_cred); 1484 put_pid(sk->sk_peer_pid); 1485 if (likely(sk->sk_net_refcnt)) 1486 put_net(sock_net(sk)); 1487 sk_prot_free(sk->sk_prot_creator, sk); 1488 } 1489 1490 static void __sk_free(struct sock *sk) 1491 { 1492 if (unlikely(sock_diag_has_destroy_listeners(sk) && sk->sk_net_refcnt)) 1493 sock_diag_broadcast_destroy(sk); 1494 else 1495 sk_destruct(sk); 1496 } 1497 1498 void sk_free(struct sock *sk) 1499 { 1500 /* 1501 * We subtract one from sk_wmem_alloc and can know if 1502 * some packets are still in some tx queue. 1503 * If not null, sock_wfree() will call __sk_free(sk) later 1504 */ 1505 if (atomic_dec_and_test(&sk->sk_wmem_alloc)) 1506 __sk_free(sk); 1507 } 1508 EXPORT_SYMBOL(sk_free); 1509 1510 static void sk_update_clone(const struct sock *sk, struct sock *newsk) 1511 { 1512 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) 1513 sock_update_memcg(newsk); 1514 } 1515 1516 /** 1517 * sk_clone_lock - clone a socket, and lock its clone 1518 * @sk: the socket to clone 1519 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 1520 * 1521 * Caller must unlock socket even in error path (bh_unlock_sock(newsk)) 1522 */ 1523 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority) 1524 { 1525 struct sock *newsk; 1526 bool is_charged = true; 1527 1528 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family); 1529 if (newsk != NULL) { 1530 struct sk_filter *filter; 1531 1532 sock_copy(newsk, sk); 1533 1534 /* SANITY */ 1535 if (likely(newsk->sk_net_refcnt)) 1536 get_net(sock_net(newsk)); 1537 sk_node_init(&newsk->sk_node); 1538 sock_lock_init(newsk); 1539 bh_lock_sock(newsk); 1540 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL; 1541 newsk->sk_backlog.len = 0; 1542 1543 atomic_set(&newsk->sk_rmem_alloc, 0); 1544 /* 1545 * sk_wmem_alloc set to one (see sk_free() and sock_wfree()) 1546 */ 1547 atomic_set(&newsk->sk_wmem_alloc, 1); 1548 atomic_set(&newsk->sk_omem_alloc, 0); 1549 skb_queue_head_init(&newsk->sk_receive_queue); 1550 skb_queue_head_init(&newsk->sk_write_queue); 1551 1552 rwlock_init(&newsk->sk_callback_lock); 1553 lockdep_set_class_and_name(&newsk->sk_callback_lock, 1554 af_callback_keys + newsk->sk_family, 1555 af_family_clock_key_strings[newsk->sk_family]); 1556 1557 newsk->sk_dst_cache = NULL; 1558 newsk->sk_wmem_queued = 0; 1559 newsk->sk_forward_alloc = 0; 1560 newsk->sk_send_head = NULL; 1561 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK; 1562 1563 sock_reset_flag(newsk, SOCK_DONE); 1564 skb_queue_head_init(&newsk->sk_error_queue); 1565 1566 filter = rcu_dereference_protected(newsk->sk_filter, 1); 1567 if (filter != NULL) 1568 /* though it's an empty new sock, the charging may fail 1569 * if sysctl_optmem_max was changed between creation of 1570 * original socket and cloning 1571 */ 1572 is_charged = sk_filter_charge(newsk, filter); 1573 1574 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) { 1575 /* It is still raw copy of parent, so invalidate 1576 * destructor and make plain sk_free() */ 1577 newsk->sk_destruct = NULL; 1578 bh_unlock_sock(newsk); 1579 sk_free(newsk); 1580 newsk = NULL; 1581 goto out; 1582 } 1583 1584 newsk->sk_err = 0; 1585 newsk->sk_priority = 0; 1586 newsk->sk_incoming_cpu = raw_smp_processor_id(); 1587 atomic64_set(&newsk->sk_cookie, 0); 1588 /* 1589 * Before updating sk_refcnt, we must commit prior changes to memory 1590 * (Documentation/RCU/rculist_nulls.txt for details) 1591 */ 1592 smp_wmb(); 1593 atomic_set(&newsk->sk_refcnt, 2); 1594 1595 /* 1596 * Increment the counter in the same struct proto as the master 1597 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that 1598 * is the same as sk->sk_prot->socks, as this field was copied 1599 * with memcpy). 1600 * 1601 * This _changes_ the previous behaviour, where 1602 * tcp_create_openreq_child always was incrementing the 1603 * equivalent to tcp_prot->socks (inet_sock_nr), so this have 1604 * to be taken into account in all callers. -acme 1605 */ 1606 sk_refcnt_debug_inc(newsk); 1607 sk_set_socket(newsk, NULL); 1608 newsk->sk_wq = NULL; 1609 1610 sk_update_clone(sk, newsk); 1611 1612 if (newsk->sk_prot->sockets_allocated) 1613 sk_sockets_allocated_inc(newsk); 1614 1615 if (sock_needs_netstamp(sk) && 1616 newsk->sk_flags & SK_FLAGS_TIMESTAMP) 1617 net_enable_timestamp(); 1618 } 1619 out: 1620 return newsk; 1621 } 1622 EXPORT_SYMBOL_GPL(sk_clone_lock); 1623 1624 void sk_setup_caps(struct sock *sk, struct dst_entry *dst) 1625 { 1626 u32 max_segs = 1; 1627 1628 sk_dst_set(sk, dst); 1629 sk->sk_route_caps = dst->dev->features; 1630 if (sk->sk_route_caps & NETIF_F_GSO) 1631 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE; 1632 sk->sk_route_caps &= ~sk->sk_route_nocaps; 1633 if (sk_can_gso(sk)) { 1634 if (dst->header_len) { 1635 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 1636 } else { 1637 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM; 1638 sk->sk_gso_max_size = dst->dev->gso_max_size; 1639 max_segs = max_t(u32, dst->dev->gso_max_segs, 1); 1640 } 1641 } 1642 sk->sk_gso_max_segs = max_segs; 1643 } 1644 EXPORT_SYMBOL_GPL(sk_setup_caps); 1645 1646 /* 1647 * Simple resource managers for sockets. 1648 */ 1649 1650 1651 /* 1652 * Write buffer destructor automatically called from kfree_skb. 1653 */ 1654 void sock_wfree(struct sk_buff *skb) 1655 { 1656 struct sock *sk = skb->sk; 1657 unsigned int len = skb->truesize; 1658 1659 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) { 1660 /* 1661 * Keep a reference on sk_wmem_alloc, this will be released 1662 * after sk_write_space() call 1663 */ 1664 atomic_sub(len - 1, &sk->sk_wmem_alloc); 1665 sk->sk_write_space(sk); 1666 len = 1; 1667 } 1668 /* 1669 * if sk_wmem_alloc reaches 0, we must finish what sk_free() 1670 * could not do because of in-flight packets 1671 */ 1672 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc)) 1673 __sk_free(sk); 1674 } 1675 EXPORT_SYMBOL(sock_wfree); 1676 1677 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk) 1678 { 1679 skb_orphan(skb); 1680 skb->sk = sk; 1681 #ifdef CONFIG_INET 1682 if (unlikely(!sk_fullsock(sk))) { 1683 skb->destructor = sock_edemux; 1684 sock_hold(sk); 1685 return; 1686 } 1687 #endif 1688 skb->destructor = sock_wfree; 1689 skb_set_hash_from_sk(skb, sk); 1690 /* 1691 * We used to take a refcount on sk, but following operation 1692 * is enough to guarantee sk_free() wont free this sock until 1693 * all in-flight packets are completed 1694 */ 1695 atomic_add(skb->truesize, &sk->sk_wmem_alloc); 1696 } 1697 EXPORT_SYMBOL(skb_set_owner_w); 1698 1699 void skb_orphan_partial(struct sk_buff *skb) 1700 { 1701 /* TCP stack sets skb->ooo_okay based on sk_wmem_alloc, 1702 * so we do not completely orphan skb, but transfert all 1703 * accounted bytes but one, to avoid unexpected reorders. 1704 */ 1705 if (skb->destructor == sock_wfree 1706 #ifdef CONFIG_INET 1707 || skb->destructor == tcp_wfree 1708 #endif 1709 ) { 1710 atomic_sub(skb->truesize - 1, &skb->sk->sk_wmem_alloc); 1711 skb->truesize = 1; 1712 } else { 1713 skb_orphan(skb); 1714 } 1715 } 1716 EXPORT_SYMBOL(skb_orphan_partial); 1717 1718 /* 1719 * Read buffer destructor automatically called from kfree_skb. 1720 */ 1721 void sock_rfree(struct sk_buff *skb) 1722 { 1723 struct sock *sk = skb->sk; 1724 unsigned int len = skb->truesize; 1725 1726 atomic_sub(len, &sk->sk_rmem_alloc); 1727 sk_mem_uncharge(sk, len); 1728 } 1729 EXPORT_SYMBOL(sock_rfree); 1730 1731 /* 1732 * Buffer destructor for skbs that are not used directly in read or write 1733 * path, e.g. for error handler skbs. Automatically called from kfree_skb. 1734 */ 1735 void sock_efree(struct sk_buff *skb) 1736 { 1737 sock_put(skb->sk); 1738 } 1739 EXPORT_SYMBOL(sock_efree); 1740 1741 kuid_t sock_i_uid(struct sock *sk) 1742 { 1743 kuid_t uid; 1744 1745 read_lock_bh(&sk->sk_callback_lock); 1746 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID; 1747 read_unlock_bh(&sk->sk_callback_lock); 1748 return uid; 1749 } 1750 EXPORT_SYMBOL(sock_i_uid); 1751 1752 unsigned long sock_i_ino(struct sock *sk) 1753 { 1754 unsigned long ino; 1755 1756 read_lock_bh(&sk->sk_callback_lock); 1757 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0; 1758 read_unlock_bh(&sk->sk_callback_lock); 1759 return ino; 1760 } 1761 EXPORT_SYMBOL(sock_i_ino); 1762 1763 /* 1764 * Allocate a skb from the socket's send buffer. 1765 */ 1766 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 1767 gfp_t priority) 1768 { 1769 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) { 1770 struct sk_buff *skb = alloc_skb(size, priority); 1771 if (skb) { 1772 skb_set_owner_w(skb, sk); 1773 return skb; 1774 } 1775 } 1776 return NULL; 1777 } 1778 EXPORT_SYMBOL(sock_wmalloc); 1779 1780 /* 1781 * Allocate a memory block from the socket's option memory buffer. 1782 */ 1783 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority) 1784 { 1785 if ((unsigned int)size <= sysctl_optmem_max && 1786 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) { 1787 void *mem; 1788 /* First do the add, to avoid the race if kmalloc 1789 * might sleep. 1790 */ 1791 atomic_add(size, &sk->sk_omem_alloc); 1792 mem = kmalloc(size, priority); 1793 if (mem) 1794 return mem; 1795 atomic_sub(size, &sk->sk_omem_alloc); 1796 } 1797 return NULL; 1798 } 1799 EXPORT_SYMBOL(sock_kmalloc); 1800 1801 /* Free an option memory block. Note, we actually want the inline 1802 * here as this allows gcc to detect the nullify and fold away the 1803 * condition entirely. 1804 */ 1805 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size, 1806 const bool nullify) 1807 { 1808 if (WARN_ON_ONCE(!mem)) 1809 return; 1810 if (nullify) 1811 kzfree(mem); 1812 else 1813 kfree(mem); 1814 atomic_sub(size, &sk->sk_omem_alloc); 1815 } 1816 1817 void sock_kfree_s(struct sock *sk, void *mem, int size) 1818 { 1819 __sock_kfree_s(sk, mem, size, false); 1820 } 1821 EXPORT_SYMBOL(sock_kfree_s); 1822 1823 void sock_kzfree_s(struct sock *sk, void *mem, int size) 1824 { 1825 __sock_kfree_s(sk, mem, size, true); 1826 } 1827 EXPORT_SYMBOL(sock_kzfree_s); 1828 1829 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock. 1830 I think, these locks should be removed for datagram sockets. 1831 */ 1832 static long sock_wait_for_wmem(struct sock *sk, long timeo) 1833 { 1834 DEFINE_WAIT(wait); 1835 1836 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk); 1837 for (;;) { 1838 if (!timeo) 1839 break; 1840 if (signal_pending(current)) 1841 break; 1842 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 1843 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 1844 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) 1845 break; 1846 if (sk->sk_shutdown & SEND_SHUTDOWN) 1847 break; 1848 if (sk->sk_err) 1849 break; 1850 timeo = schedule_timeout(timeo); 1851 } 1852 finish_wait(sk_sleep(sk), &wait); 1853 return timeo; 1854 } 1855 1856 1857 /* 1858 * Generic send/receive buffer handlers 1859 */ 1860 1861 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 1862 unsigned long data_len, int noblock, 1863 int *errcode, int max_page_order) 1864 { 1865 struct sk_buff *skb; 1866 long timeo; 1867 int err; 1868 1869 timeo = sock_sndtimeo(sk, noblock); 1870 for (;;) { 1871 err = sock_error(sk); 1872 if (err != 0) 1873 goto failure; 1874 1875 err = -EPIPE; 1876 if (sk->sk_shutdown & SEND_SHUTDOWN) 1877 goto failure; 1878 1879 if (sk_wmem_alloc_get(sk) < sk->sk_sndbuf) 1880 break; 1881 1882 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk); 1883 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 1884 err = -EAGAIN; 1885 if (!timeo) 1886 goto failure; 1887 if (signal_pending(current)) 1888 goto interrupted; 1889 timeo = sock_wait_for_wmem(sk, timeo); 1890 } 1891 skb = alloc_skb_with_frags(header_len, data_len, max_page_order, 1892 errcode, sk->sk_allocation); 1893 if (skb) 1894 skb_set_owner_w(skb, sk); 1895 return skb; 1896 1897 interrupted: 1898 err = sock_intr_errno(timeo); 1899 failure: 1900 *errcode = err; 1901 return NULL; 1902 } 1903 EXPORT_SYMBOL(sock_alloc_send_pskb); 1904 1905 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size, 1906 int noblock, int *errcode) 1907 { 1908 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0); 1909 } 1910 EXPORT_SYMBOL(sock_alloc_send_skb); 1911 1912 int sock_cmsg_send(struct sock *sk, struct msghdr *msg, 1913 struct sockcm_cookie *sockc) 1914 { 1915 struct cmsghdr *cmsg; 1916 1917 for_each_cmsghdr(cmsg, msg) { 1918 if (!CMSG_OK(msg, cmsg)) 1919 return -EINVAL; 1920 if (cmsg->cmsg_level != SOL_SOCKET) 1921 continue; 1922 switch (cmsg->cmsg_type) { 1923 case SO_MARK: 1924 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) 1925 return -EPERM; 1926 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 1927 return -EINVAL; 1928 sockc->mark = *(u32 *)CMSG_DATA(cmsg); 1929 break; 1930 default: 1931 return -EINVAL; 1932 } 1933 } 1934 return 0; 1935 } 1936 EXPORT_SYMBOL(sock_cmsg_send); 1937 1938 /* On 32bit arches, an skb frag is limited to 2^15 */ 1939 #define SKB_FRAG_PAGE_ORDER get_order(32768) 1940 1941 /** 1942 * skb_page_frag_refill - check that a page_frag contains enough room 1943 * @sz: minimum size of the fragment we want to get 1944 * @pfrag: pointer to page_frag 1945 * @gfp: priority for memory allocation 1946 * 1947 * Note: While this allocator tries to use high order pages, there is 1948 * no guarantee that allocations succeed. Therefore, @sz MUST be 1949 * less or equal than PAGE_SIZE. 1950 */ 1951 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp) 1952 { 1953 if (pfrag->page) { 1954 if (atomic_read(&pfrag->page->_count) == 1) { 1955 pfrag->offset = 0; 1956 return true; 1957 } 1958 if (pfrag->offset + sz <= pfrag->size) 1959 return true; 1960 put_page(pfrag->page); 1961 } 1962 1963 pfrag->offset = 0; 1964 if (SKB_FRAG_PAGE_ORDER) { 1965 /* Avoid direct reclaim but allow kswapd to wake */ 1966 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) | 1967 __GFP_COMP | __GFP_NOWARN | 1968 __GFP_NORETRY, 1969 SKB_FRAG_PAGE_ORDER); 1970 if (likely(pfrag->page)) { 1971 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER; 1972 return true; 1973 } 1974 } 1975 pfrag->page = alloc_page(gfp); 1976 if (likely(pfrag->page)) { 1977 pfrag->size = PAGE_SIZE; 1978 return true; 1979 } 1980 return false; 1981 } 1982 EXPORT_SYMBOL(skb_page_frag_refill); 1983 1984 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag) 1985 { 1986 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation))) 1987 return true; 1988 1989 sk_enter_memory_pressure(sk); 1990 sk_stream_moderate_sndbuf(sk); 1991 return false; 1992 } 1993 EXPORT_SYMBOL(sk_page_frag_refill); 1994 1995 static void __lock_sock(struct sock *sk) 1996 __releases(&sk->sk_lock.slock) 1997 __acquires(&sk->sk_lock.slock) 1998 { 1999 DEFINE_WAIT(wait); 2000 2001 for (;;) { 2002 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait, 2003 TASK_UNINTERRUPTIBLE); 2004 spin_unlock_bh(&sk->sk_lock.slock); 2005 schedule(); 2006 spin_lock_bh(&sk->sk_lock.slock); 2007 if (!sock_owned_by_user(sk)) 2008 break; 2009 } 2010 finish_wait(&sk->sk_lock.wq, &wait); 2011 } 2012 2013 static void __release_sock(struct sock *sk) 2014 __releases(&sk->sk_lock.slock) 2015 __acquires(&sk->sk_lock.slock) 2016 { 2017 struct sk_buff *skb = sk->sk_backlog.head; 2018 2019 do { 2020 sk->sk_backlog.head = sk->sk_backlog.tail = NULL; 2021 bh_unlock_sock(sk); 2022 2023 do { 2024 struct sk_buff *next = skb->next; 2025 2026 prefetch(next); 2027 WARN_ON_ONCE(skb_dst_is_noref(skb)); 2028 skb->next = NULL; 2029 sk_backlog_rcv(sk, skb); 2030 2031 /* 2032 * We are in process context here with softirqs 2033 * disabled, use cond_resched_softirq() to preempt. 2034 * This is safe to do because we've taken the backlog 2035 * queue private: 2036 */ 2037 cond_resched_softirq(); 2038 2039 skb = next; 2040 } while (skb != NULL); 2041 2042 bh_lock_sock(sk); 2043 } while ((skb = sk->sk_backlog.head) != NULL); 2044 2045 /* 2046 * Doing the zeroing here guarantee we can not loop forever 2047 * while a wild producer attempts to flood us. 2048 */ 2049 sk->sk_backlog.len = 0; 2050 } 2051 2052 /** 2053 * sk_wait_data - wait for data to arrive at sk_receive_queue 2054 * @sk: sock to wait on 2055 * @timeo: for how long 2056 * @skb: last skb seen on sk_receive_queue 2057 * 2058 * Now socket state including sk->sk_err is changed only under lock, 2059 * hence we may omit checks after joining wait queue. 2060 * We check receive queue before schedule() only as optimization; 2061 * it is very likely that release_sock() added new data. 2062 */ 2063 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb) 2064 { 2065 int rc; 2066 DEFINE_WAIT(wait); 2067 2068 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 2069 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk); 2070 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb); 2071 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk); 2072 finish_wait(sk_sleep(sk), &wait); 2073 return rc; 2074 } 2075 EXPORT_SYMBOL(sk_wait_data); 2076 2077 /** 2078 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated 2079 * @sk: socket 2080 * @size: memory size to allocate 2081 * @kind: allocation type 2082 * 2083 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means 2084 * rmem allocation. This function assumes that protocols which have 2085 * memory_pressure use sk_wmem_queued as write buffer accounting. 2086 */ 2087 int __sk_mem_schedule(struct sock *sk, int size, int kind) 2088 { 2089 struct proto *prot = sk->sk_prot; 2090 int amt = sk_mem_pages(size); 2091 long allocated; 2092 int parent_status = UNDER_LIMIT; 2093 2094 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM; 2095 2096 allocated = sk_memory_allocated_add(sk, amt, &parent_status); 2097 2098 /* Under limit. */ 2099 if (parent_status == UNDER_LIMIT && 2100 allocated <= sk_prot_mem_limits(sk, 0)) { 2101 sk_leave_memory_pressure(sk); 2102 return 1; 2103 } 2104 2105 /* Under pressure. (we or our parents) */ 2106 if ((parent_status > SOFT_LIMIT) || 2107 allocated > sk_prot_mem_limits(sk, 1)) 2108 sk_enter_memory_pressure(sk); 2109 2110 /* Over hard limit (we or our parents) */ 2111 if ((parent_status == OVER_LIMIT) || 2112 (allocated > sk_prot_mem_limits(sk, 2))) 2113 goto suppress_allocation; 2114 2115 /* guarantee minimum buffer size under pressure */ 2116 if (kind == SK_MEM_RECV) { 2117 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0]) 2118 return 1; 2119 2120 } else { /* SK_MEM_SEND */ 2121 if (sk->sk_type == SOCK_STREAM) { 2122 if (sk->sk_wmem_queued < prot->sysctl_wmem[0]) 2123 return 1; 2124 } else if (atomic_read(&sk->sk_wmem_alloc) < 2125 prot->sysctl_wmem[0]) 2126 return 1; 2127 } 2128 2129 if (sk_has_memory_pressure(sk)) { 2130 int alloc; 2131 2132 if (!sk_under_memory_pressure(sk)) 2133 return 1; 2134 alloc = sk_sockets_allocated_read_positive(sk); 2135 if (sk_prot_mem_limits(sk, 2) > alloc * 2136 sk_mem_pages(sk->sk_wmem_queued + 2137 atomic_read(&sk->sk_rmem_alloc) + 2138 sk->sk_forward_alloc)) 2139 return 1; 2140 } 2141 2142 suppress_allocation: 2143 2144 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) { 2145 sk_stream_moderate_sndbuf(sk); 2146 2147 /* Fail only if socket is _under_ its sndbuf. 2148 * In this case we cannot block, so that we have to fail. 2149 */ 2150 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf) 2151 return 1; 2152 } 2153 2154 trace_sock_exceed_buf_limit(sk, prot, allocated); 2155 2156 /* Alas. Undo changes. */ 2157 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM; 2158 2159 sk_memory_allocated_sub(sk, amt); 2160 2161 return 0; 2162 } 2163 EXPORT_SYMBOL(__sk_mem_schedule); 2164 2165 /** 2166 * __sk_mem_reclaim - reclaim memory_allocated 2167 * @sk: socket 2168 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple) 2169 */ 2170 void __sk_mem_reclaim(struct sock *sk, int amount) 2171 { 2172 amount >>= SK_MEM_QUANTUM_SHIFT; 2173 sk_memory_allocated_sub(sk, amount); 2174 sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT; 2175 2176 if (sk_under_memory_pressure(sk) && 2177 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0))) 2178 sk_leave_memory_pressure(sk); 2179 } 2180 EXPORT_SYMBOL(__sk_mem_reclaim); 2181 2182 2183 /* 2184 * Set of default routines for initialising struct proto_ops when 2185 * the protocol does not support a particular function. In certain 2186 * cases where it makes no sense for a protocol to have a "do nothing" 2187 * function, some default processing is provided. 2188 */ 2189 2190 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len) 2191 { 2192 return -EOPNOTSUPP; 2193 } 2194 EXPORT_SYMBOL(sock_no_bind); 2195 2196 int sock_no_connect(struct socket *sock, struct sockaddr *saddr, 2197 int len, int flags) 2198 { 2199 return -EOPNOTSUPP; 2200 } 2201 EXPORT_SYMBOL(sock_no_connect); 2202 2203 int sock_no_socketpair(struct socket *sock1, struct socket *sock2) 2204 { 2205 return -EOPNOTSUPP; 2206 } 2207 EXPORT_SYMBOL(sock_no_socketpair); 2208 2209 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags) 2210 { 2211 return -EOPNOTSUPP; 2212 } 2213 EXPORT_SYMBOL(sock_no_accept); 2214 2215 int sock_no_getname(struct socket *sock, struct sockaddr *saddr, 2216 int *len, int peer) 2217 { 2218 return -EOPNOTSUPP; 2219 } 2220 EXPORT_SYMBOL(sock_no_getname); 2221 2222 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt) 2223 { 2224 return 0; 2225 } 2226 EXPORT_SYMBOL(sock_no_poll); 2227 2228 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 2229 { 2230 return -EOPNOTSUPP; 2231 } 2232 EXPORT_SYMBOL(sock_no_ioctl); 2233 2234 int sock_no_listen(struct socket *sock, int backlog) 2235 { 2236 return -EOPNOTSUPP; 2237 } 2238 EXPORT_SYMBOL(sock_no_listen); 2239 2240 int sock_no_shutdown(struct socket *sock, int how) 2241 { 2242 return -EOPNOTSUPP; 2243 } 2244 EXPORT_SYMBOL(sock_no_shutdown); 2245 2246 int sock_no_setsockopt(struct socket *sock, int level, int optname, 2247 char __user *optval, unsigned int optlen) 2248 { 2249 return -EOPNOTSUPP; 2250 } 2251 EXPORT_SYMBOL(sock_no_setsockopt); 2252 2253 int sock_no_getsockopt(struct socket *sock, int level, int optname, 2254 char __user *optval, int __user *optlen) 2255 { 2256 return -EOPNOTSUPP; 2257 } 2258 EXPORT_SYMBOL(sock_no_getsockopt); 2259 2260 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len) 2261 { 2262 return -EOPNOTSUPP; 2263 } 2264 EXPORT_SYMBOL(sock_no_sendmsg); 2265 2266 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len, 2267 int flags) 2268 { 2269 return -EOPNOTSUPP; 2270 } 2271 EXPORT_SYMBOL(sock_no_recvmsg); 2272 2273 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma) 2274 { 2275 /* Mirror missing mmap method error code */ 2276 return -ENODEV; 2277 } 2278 EXPORT_SYMBOL(sock_no_mmap); 2279 2280 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags) 2281 { 2282 ssize_t res; 2283 struct msghdr msg = {.msg_flags = flags}; 2284 struct kvec iov; 2285 char *kaddr = kmap(page); 2286 iov.iov_base = kaddr + offset; 2287 iov.iov_len = size; 2288 res = kernel_sendmsg(sock, &msg, &iov, 1, size); 2289 kunmap(page); 2290 return res; 2291 } 2292 EXPORT_SYMBOL(sock_no_sendpage); 2293 2294 /* 2295 * Default Socket Callbacks 2296 */ 2297 2298 static void sock_def_wakeup(struct sock *sk) 2299 { 2300 struct socket_wq *wq; 2301 2302 rcu_read_lock(); 2303 wq = rcu_dereference(sk->sk_wq); 2304 if (skwq_has_sleeper(wq)) 2305 wake_up_interruptible_all(&wq->wait); 2306 rcu_read_unlock(); 2307 } 2308 2309 static void sock_def_error_report(struct sock *sk) 2310 { 2311 struct socket_wq *wq; 2312 2313 rcu_read_lock(); 2314 wq = rcu_dereference(sk->sk_wq); 2315 if (skwq_has_sleeper(wq)) 2316 wake_up_interruptible_poll(&wq->wait, POLLERR); 2317 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR); 2318 rcu_read_unlock(); 2319 } 2320 2321 static void sock_def_readable(struct sock *sk) 2322 { 2323 struct socket_wq *wq; 2324 2325 rcu_read_lock(); 2326 wq = rcu_dereference(sk->sk_wq); 2327 if (skwq_has_sleeper(wq)) 2328 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI | 2329 POLLRDNORM | POLLRDBAND); 2330 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN); 2331 rcu_read_unlock(); 2332 } 2333 2334 static void sock_def_write_space(struct sock *sk) 2335 { 2336 struct socket_wq *wq; 2337 2338 rcu_read_lock(); 2339 2340 /* Do not wake up a writer until he can make "significant" 2341 * progress. --DaveM 2342 */ 2343 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) { 2344 wq = rcu_dereference(sk->sk_wq); 2345 if (skwq_has_sleeper(wq)) 2346 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT | 2347 POLLWRNORM | POLLWRBAND); 2348 2349 /* Should agree with poll, otherwise some programs break */ 2350 if (sock_writeable(sk)) 2351 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT); 2352 } 2353 2354 rcu_read_unlock(); 2355 } 2356 2357 static void sock_def_destruct(struct sock *sk) 2358 { 2359 } 2360 2361 void sk_send_sigurg(struct sock *sk) 2362 { 2363 if (sk->sk_socket && sk->sk_socket->file) 2364 if (send_sigurg(&sk->sk_socket->file->f_owner)) 2365 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI); 2366 } 2367 EXPORT_SYMBOL(sk_send_sigurg); 2368 2369 void sk_reset_timer(struct sock *sk, struct timer_list* timer, 2370 unsigned long expires) 2371 { 2372 if (!mod_timer(timer, expires)) 2373 sock_hold(sk); 2374 } 2375 EXPORT_SYMBOL(sk_reset_timer); 2376 2377 void sk_stop_timer(struct sock *sk, struct timer_list* timer) 2378 { 2379 if (del_timer(timer)) 2380 __sock_put(sk); 2381 } 2382 EXPORT_SYMBOL(sk_stop_timer); 2383 2384 void sock_init_data(struct socket *sock, struct sock *sk) 2385 { 2386 skb_queue_head_init(&sk->sk_receive_queue); 2387 skb_queue_head_init(&sk->sk_write_queue); 2388 skb_queue_head_init(&sk->sk_error_queue); 2389 2390 sk->sk_send_head = NULL; 2391 2392 init_timer(&sk->sk_timer); 2393 2394 sk->sk_allocation = GFP_KERNEL; 2395 sk->sk_rcvbuf = sysctl_rmem_default; 2396 sk->sk_sndbuf = sysctl_wmem_default; 2397 sk->sk_state = TCP_CLOSE; 2398 sk_set_socket(sk, sock); 2399 2400 sock_set_flag(sk, SOCK_ZAPPED); 2401 2402 if (sock) { 2403 sk->sk_type = sock->type; 2404 sk->sk_wq = sock->wq; 2405 sock->sk = sk; 2406 } else 2407 sk->sk_wq = NULL; 2408 2409 rwlock_init(&sk->sk_callback_lock); 2410 lockdep_set_class_and_name(&sk->sk_callback_lock, 2411 af_callback_keys + sk->sk_family, 2412 af_family_clock_key_strings[sk->sk_family]); 2413 2414 sk->sk_state_change = sock_def_wakeup; 2415 sk->sk_data_ready = sock_def_readable; 2416 sk->sk_write_space = sock_def_write_space; 2417 sk->sk_error_report = sock_def_error_report; 2418 sk->sk_destruct = sock_def_destruct; 2419 2420 sk->sk_frag.page = NULL; 2421 sk->sk_frag.offset = 0; 2422 sk->sk_peek_off = -1; 2423 2424 sk->sk_peer_pid = NULL; 2425 sk->sk_peer_cred = NULL; 2426 sk->sk_write_pending = 0; 2427 sk->sk_rcvlowat = 1; 2428 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; 2429 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT; 2430 2431 sk->sk_stamp = ktime_set(-1L, 0); 2432 2433 #ifdef CONFIG_NET_RX_BUSY_POLL 2434 sk->sk_napi_id = 0; 2435 sk->sk_ll_usec = sysctl_net_busy_read; 2436 #endif 2437 2438 sk->sk_max_pacing_rate = ~0U; 2439 sk->sk_pacing_rate = ~0U; 2440 sk->sk_incoming_cpu = -1; 2441 /* 2442 * Before updating sk_refcnt, we must commit prior changes to memory 2443 * (Documentation/RCU/rculist_nulls.txt for details) 2444 */ 2445 smp_wmb(); 2446 atomic_set(&sk->sk_refcnt, 1); 2447 atomic_set(&sk->sk_drops, 0); 2448 } 2449 EXPORT_SYMBOL(sock_init_data); 2450 2451 void lock_sock_nested(struct sock *sk, int subclass) 2452 { 2453 might_sleep(); 2454 spin_lock_bh(&sk->sk_lock.slock); 2455 if (sk->sk_lock.owned) 2456 __lock_sock(sk); 2457 sk->sk_lock.owned = 1; 2458 spin_unlock(&sk->sk_lock.slock); 2459 /* 2460 * The sk_lock has mutex_lock() semantics here: 2461 */ 2462 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_); 2463 local_bh_enable(); 2464 } 2465 EXPORT_SYMBOL(lock_sock_nested); 2466 2467 void release_sock(struct sock *sk) 2468 { 2469 /* 2470 * The sk_lock has mutex_unlock() semantics: 2471 */ 2472 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_); 2473 2474 spin_lock_bh(&sk->sk_lock.slock); 2475 if (sk->sk_backlog.tail) 2476 __release_sock(sk); 2477 2478 /* Warning : release_cb() might need to release sk ownership, 2479 * ie call sock_release_ownership(sk) before us. 2480 */ 2481 if (sk->sk_prot->release_cb) 2482 sk->sk_prot->release_cb(sk); 2483 2484 sock_release_ownership(sk); 2485 if (waitqueue_active(&sk->sk_lock.wq)) 2486 wake_up(&sk->sk_lock.wq); 2487 spin_unlock_bh(&sk->sk_lock.slock); 2488 } 2489 EXPORT_SYMBOL(release_sock); 2490 2491 /** 2492 * lock_sock_fast - fast version of lock_sock 2493 * @sk: socket 2494 * 2495 * This version should be used for very small section, where process wont block 2496 * return false if fast path is taken 2497 * sk_lock.slock locked, owned = 0, BH disabled 2498 * return true if slow path is taken 2499 * sk_lock.slock unlocked, owned = 1, BH enabled 2500 */ 2501 bool lock_sock_fast(struct sock *sk) 2502 { 2503 might_sleep(); 2504 spin_lock_bh(&sk->sk_lock.slock); 2505 2506 if (!sk->sk_lock.owned) 2507 /* 2508 * Note : We must disable BH 2509 */ 2510 return false; 2511 2512 __lock_sock(sk); 2513 sk->sk_lock.owned = 1; 2514 spin_unlock(&sk->sk_lock.slock); 2515 /* 2516 * The sk_lock has mutex_lock() semantics here: 2517 */ 2518 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_); 2519 local_bh_enable(); 2520 return true; 2521 } 2522 EXPORT_SYMBOL(lock_sock_fast); 2523 2524 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp) 2525 { 2526 struct timeval tv; 2527 if (!sock_flag(sk, SOCK_TIMESTAMP)) 2528 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 2529 tv = ktime_to_timeval(sk->sk_stamp); 2530 if (tv.tv_sec == -1) 2531 return -ENOENT; 2532 if (tv.tv_sec == 0) { 2533 sk->sk_stamp = ktime_get_real(); 2534 tv = ktime_to_timeval(sk->sk_stamp); 2535 } 2536 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0; 2537 } 2538 EXPORT_SYMBOL(sock_get_timestamp); 2539 2540 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp) 2541 { 2542 struct timespec ts; 2543 if (!sock_flag(sk, SOCK_TIMESTAMP)) 2544 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 2545 ts = ktime_to_timespec(sk->sk_stamp); 2546 if (ts.tv_sec == -1) 2547 return -ENOENT; 2548 if (ts.tv_sec == 0) { 2549 sk->sk_stamp = ktime_get_real(); 2550 ts = ktime_to_timespec(sk->sk_stamp); 2551 } 2552 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0; 2553 } 2554 EXPORT_SYMBOL(sock_get_timestampns); 2555 2556 void sock_enable_timestamp(struct sock *sk, int flag) 2557 { 2558 if (!sock_flag(sk, flag)) { 2559 unsigned long previous_flags = sk->sk_flags; 2560 2561 sock_set_flag(sk, flag); 2562 /* 2563 * we just set one of the two flags which require net 2564 * time stamping, but time stamping might have been on 2565 * already because of the other one 2566 */ 2567 if (sock_needs_netstamp(sk) && 2568 !(previous_flags & SK_FLAGS_TIMESTAMP)) 2569 net_enable_timestamp(); 2570 } 2571 } 2572 2573 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, 2574 int level, int type) 2575 { 2576 struct sock_exterr_skb *serr; 2577 struct sk_buff *skb; 2578 int copied, err; 2579 2580 err = -EAGAIN; 2581 skb = sock_dequeue_err_skb(sk); 2582 if (skb == NULL) 2583 goto out; 2584 2585 copied = skb->len; 2586 if (copied > len) { 2587 msg->msg_flags |= MSG_TRUNC; 2588 copied = len; 2589 } 2590 err = skb_copy_datagram_msg(skb, 0, msg, copied); 2591 if (err) 2592 goto out_free_skb; 2593 2594 sock_recv_timestamp(msg, sk, skb); 2595 2596 serr = SKB_EXT_ERR(skb); 2597 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee); 2598 2599 msg->msg_flags |= MSG_ERRQUEUE; 2600 err = copied; 2601 2602 out_free_skb: 2603 kfree_skb(skb); 2604 out: 2605 return err; 2606 } 2607 EXPORT_SYMBOL(sock_recv_errqueue); 2608 2609 /* 2610 * Get a socket option on an socket. 2611 * 2612 * FIX: POSIX 1003.1g is very ambiguous here. It states that 2613 * asynchronous errors should be reported by getsockopt. We assume 2614 * this means if you specify SO_ERROR (otherwise whats the point of it). 2615 */ 2616 int sock_common_getsockopt(struct socket *sock, int level, int optname, 2617 char __user *optval, int __user *optlen) 2618 { 2619 struct sock *sk = sock->sk; 2620 2621 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen); 2622 } 2623 EXPORT_SYMBOL(sock_common_getsockopt); 2624 2625 #ifdef CONFIG_COMPAT 2626 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname, 2627 char __user *optval, int __user *optlen) 2628 { 2629 struct sock *sk = sock->sk; 2630 2631 if (sk->sk_prot->compat_getsockopt != NULL) 2632 return sk->sk_prot->compat_getsockopt(sk, level, optname, 2633 optval, optlen); 2634 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen); 2635 } 2636 EXPORT_SYMBOL(compat_sock_common_getsockopt); 2637 #endif 2638 2639 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 2640 int flags) 2641 { 2642 struct sock *sk = sock->sk; 2643 int addr_len = 0; 2644 int err; 2645 2646 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT, 2647 flags & ~MSG_DONTWAIT, &addr_len); 2648 if (err >= 0) 2649 msg->msg_namelen = addr_len; 2650 return err; 2651 } 2652 EXPORT_SYMBOL(sock_common_recvmsg); 2653 2654 /* 2655 * Set socket options on an inet socket. 2656 */ 2657 int sock_common_setsockopt(struct socket *sock, int level, int optname, 2658 char __user *optval, unsigned int optlen) 2659 { 2660 struct sock *sk = sock->sk; 2661 2662 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen); 2663 } 2664 EXPORT_SYMBOL(sock_common_setsockopt); 2665 2666 #ifdef CONFIG_COMPAT 2667 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname, 2668 char __user *optval, unsigned int optlen) 2669 { 2670 struct sock *sk = sock->sk; 2671 2672 if (sk->sk_prot->compat_setsockopt != NULL) 2673 return sk->sk_prot->compat_setsockopt(sk, level, optname, 2674 optval, optlen); 2675 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen); 2676 } 2677 EXPORT_SYMBOL(compat_sock_common_setsockopt); 2678 #endif 2679 2680 void sk_common_release(struct sock *sk) 2681 { 2682 if (sk->sk_prot->destroy) 2683 sk->sk_prot->destroy(sk); 2684 2685 /* 2686 * Observation: when sock_common_release is called, processes have 2687 * no access to socket. But net still has. 2688 * Step one, detach it from networking: 2689 * 2690 * A. Remove from hash tables. 2691 */ 2692 2693 sk->sk_prot->unhash(sk); 2694 2695 /* 2696 * In this point socket cannot receive new packets, but it is possible 2697 * that some packets are in flight because some CPU runs receiver and 2698 * did hash table lookup before we unhashed socket. They will achieve 2699 * receive queue and will be purged by socket destructor. 2700 * 2701 * Also we still have packets pending on receive queue and probably, 2702 * our own packets waiting in device queues. sock_destroy will drain 2703 * receive queue, but transmitted packets will delay socket destruction 2704 * until the last reference will be released. 2705 */ 2706 2707 sock_orphan(sk); 2708 2709 xfrm_sk_free_policy(sk); 2710 2711 sk_refcnt_debug_release(sk); 2712 2713 if (sk->sk_frag.page) { 2714 put_page(sk->sk_frag.page); 2715 sk->sk_frag.page = NULL; 2716 } 2717 2718 sock_put(sk); 2719 } 2720 EXPORT_SYMBOL(sk_common_release); 2721 2722 #ifdef CONFIG_PROC_FS 2723 #define PROTO_INUSE_NR 64 /* should be enough for the first time */ 2724 struct prot_inuse { 2725 int val[PROTO_INUSE_NR]; 2726 }; 2727 2728 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR); 2729 2730 #ifdef CONFIG_NET_NS 2731 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val) 2732 { 2733 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val); 2734 } 2735 EXPORT_SYMBOL_GPL(sock_prot_inuse_add); 2736 2737 int sock_prot_inuse_get(struct net *net, struct proto *prot) 2738 { 2739 int cpu, idx = prot->inuse_idx; 2740 int res = 0; 2741 2742 for_each_possible_cpu(cpu) 2743 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx]; 2744 2745 return res >= 0 ? res : 0; 2746 } 2747 EXPORT_SYMBOL_GPL(sock_prot_inuse_get); 2748 2749 static int __net_init sock_inuse_init_net(struct net *net) 2750 { 2751 net->core.inuse = alloc_percpu(struct prot_inuse); 2752 return net->core.inuse ? 0 : -ENOMEM; 2753 } 2754 2755 static void __net_exit sock_inuse_exit_net(struct net *net) 2756 { 2757 free_percpu(net->core.inuse); 2758 } 2759 2760 static struct pernet_operations net_inuse_ops = { 2761 .init = sock_inuse_init_net, 2762 .exit = sock_inuse_exit_net, 2763 }; 2764 2765 static __init int net_inuse_init(void) 2766 { 2767 if (register_pernet_subsys(&net_inuse_ops)) 2768 panic("Cannot initialize net inuse counters"); 2769 2770 return 0; 2771 } 2772 2773 core_initcall(net_inuse_init); 2774 #else 2775 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse); 2776 2777 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val) 2778 { 2779 __this_cpu_add(prot_inuse.val[prot->inuse_idx], val); 2780 } 2781 EXPORT_SYMBOL_GPL(sock_prot_inuse_add); 2782 2783 int sock_prot_inuse_get(struct net *net, struct proto *prot) 2784 { 2785 int cpu, idx = prot->inuse_idx; 2786 int res = 0; 2787 2788 for_each_possible_cpu(cpu) 2789 res += per_cpu(prot_inuse, cpu).val[idx]; 2790 2791 return res >= 0 ? res : 0; 2792 } 2793 EXPORT_SYMBOL_GPL(sock_prot_inuse_get); 2794 #endif 2795 2796 static void assign_proto_idx(struct proto *prot) 2797 { 2798 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR); 2799 2800 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) { 2801 pr_err("PROTO_INUSE_NR exhausted\n"); 2802 return; 2803 } 2804 2805 set_bit(prot->inuse_idx, proto_inuse_idx); 2806 } 2807 2808 static void release_proto_idx(struct proto *prot) 2809 { 2810 if (prot->inuse_idx != PROTO_INUSE_NR - 1) 2811 clear_bit(prot->inuse_idx, proto_inuse_idx); 2812 } 2813 #else 2814 static inline void assign_proto_idx(struct proto *prot) 2815 { 2816 } 2817 2818 static inline void release_proto_idx(struct proto *prot) 2819 { 2820 } 2821 #endif 2822 2823 static void req_prot_cleanup(struct request_sock_ops *rsk_prot) 2824 { 2825 if (!rsk_prot) 2826 return; 2827 kfree(rsk_prot->slab_name); 2828 rsk_prot->slab_name = NULL; 2829 kmem_cache_destroy(rsk_prot->slab); 2830 rsk_prot->slab = NULL; 2831 } 2832 2833 static int req_prot_init(const struct proto *prot) 2834 { 2835 struct request_sock_ops *rsk_prot = prot->rsk_prot; 2836 2837 if (!rsk_prot) 2838 return 0; 2839 2840 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", 2841 prot->name); 2842 if (!rsk_prot->slab_name) 2843 return -ENOMEM; 2844 2845 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name, 2846 rsk_prot->obj_size, 0, 2847 prot->slab_flags, NULL); 2848 2849 if (!rsk_prot->slab) { 2850 pr_crit("%s: Can't create request sock SLAB cache!\n", 2851 prot->name); 2852 return -ENOMEM; 2853 } 2854 return 0; 2855 } 2856 2857 int proto_register(struct proto *prot, int alloc_slab) 2858 { 2859 if (alloc_slab) { 2860 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0, 2861 SLAB_HWCACHE_ALIGN | prot->slab_flags, 2862 NULL); 2863 2864 if (prot->slab == NULL) { 2865 pr_crit("%s: Can't create sock SLAB cache!\n", 2866 prot->name); 2867 goto out; 2868 } 2869 2870 if (req_prot_init(prot)) 2871 goto out_free_request_sock_slab; 2872 2873 if (prot->twsk_prot != NULL) { 2874 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name); 2875 2876 if (prot->twsk_prot->twsk_slab_name == NULL) 2877 goto out_free_request_sock_slab; 2878 2879 prot->twsk_prot->twsk_slab = 2880 kmem_cache_create(prot->twsk_prot->twsk_slab_name, 2881 prot->twsk_prot->twsk_obj_size, 2882 0, 2883 prot->slab_flags, 2884 NULL); 2885 if (prot->twsk_prot->twsk_slab == NULL) 2886 goto out_free_timewait_sock_slab_name; 2887 } 2888 } 2889 2890 mutex_lock(&proto_list_mutex); 2891 list_add(&prot->node, &proto_list); 2892 assign_proto_idx(prot); 2893 mutex_unlock(&proto_list_mutex); 2894 return 0; 2895 2896 out_free_timewait_sock_slab_name: 2897 kfree(prot->twsk_prot->twsk_slab_name); 2898 out_free_request_sock_slab: 2899 req_prot_cleanup(prot->rsk_prot); 2900 2901 kmem_cache_destroy(prot->slab); 2902 prot->slab = NULL; 2903 out: 2904 return -ENOBUFS; 2905 } 2906 EXPORT_SYMBOL(proto_register); 2907 2908 void proto_unregister(struct proto *prot) 2909 { 2910 mutex_lock(&proto_list_mutex); 2911 release_proto_idx(prot); 2912 list_del(&prot->node); 2913 mutex_unlock(&proto_list_mutex); 2914 2915 kmem_cache_destroy(prot->slab); 2916 prot->slab = NULL; 2917 2918 req_prot_cleanup(prot->rsk_prot); 2919 2920 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) { 2921 kmem_cache_destroy(prot->twsk_prot->twsk_slab); 2922 kfree(prot->twsk_prot->twsk_slab_name); 2923 prot->twsk_prot->twsk_slab = NULL; 2924 } 2925 } 2926 EXPORT_SYMBOL(proto_unregister); 2927 2928 #ifdef CONFIG_PROC_FS 2929 static void *proto_seq_start(struct seq_file *seq, loff_t *pos) 2930 __acquires(proto_list_mutex) 2931 { 2932 mutex_lock(&proto_list_mutex); 2933 return seq_list_start_head(&proto_list, *pos); 2934 } 2935 2936 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos) 2937 { 2938 return seq_list_next(v, &proto_list, pos); 2939 } 2940 2941 static void proto_seq_stop(struct seq_file *seq, void *v) 2942 __releases(proto_list_mutex) 2943 { 2944 mutex_unlock(&proto_list_mutex); 2945 } 2946 2947 static char proto_method_implemented(const void *method) 2948 { 2949 return method == NULL ? 'n' : 'y'; 2950 } 2951 static long sock_prot_memory_allocated(struct proto *proto) 2952 { 2953 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L; 2954 } 2955 2956 static char *sock_prot_memory_pressure(struct proto *proto) 2957 { 2958 return proto->memory_pressure != NULL ? 2959 proto_memory_pressure(proto) ? "yes" : "no" : "NI"; 2960 } 2961 2962 static void proto_seq_printf(struct seq_file *seq, struct proto *proto) 2963 { 2964 2965 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s " 2966 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n", 2967 proto->name, 2968 proto->obj_size, 2969 sock_prot_inuse_get(seq_file_net(seq), proto), 2970 sock_prot_memory_allocated(proto), 2971 sock_prot_memory_pressure(proto), 2972 proto->max_header, 2973 proto->slab == NULL ? "no" : "yes", 2974 module_name(proto->owner), 2975 proto_method_implemented(proto->close), 2976 proto_method_implemented(proto->connect), 2977 proto_method_implemented(proto->disconnect), 2978 proto_method_implemented(proto->accept), 2979 proto_method_implemented(proto->ioctl), 2980 proto_method_implemented(proto->init), 2981 proto_method_implemented(proto->destroy), 2982 proto_method_implemented(proto->shutdown), 2983 proto_method_implemented(proto->setsockopt), 2984 proto_method_implemented(proto->getsockopt), 2985 proto_method_implemented(proto->sendmsg), 2986 proto_method_implemented(proto->recvmsg), 2987 proto_method_implemented(proto->sendpage), 2988 proto_method_implemented(proto->bind), 2989 proto_method_implemented(proto->backlog_rcv), 2990 proto_method_implemented(proto->hash), 2991 proto_method_implemented(proto->unhash), 2992 proto_method_implemented(proto->get_port), 2993 proto_method_implemented(proto->enter_memory_pressure)); 2994 } 2995 2996 static int proto_seq_show(struct seq_file *seq, void *v) 2997 { 2998 if (v == &proto_list) 2999 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s", 3000 "protocol", 3001 "size", 3002 "sockets", 3003 "memory", 3004 "press", 3005 "maxhdr", 3006 "slab", 3007 "module", 3008 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n"); 3009 else 3010 proto_seq_printf(seq, list_entry(v, struct proto, node)); 3011 return 0; 3012 } 3013 3014 static const struct seq_operations proto_seq_ops = { 3015 .start = proto_seq_start, 3016 .next = proto_seq_next, 3017 .stop = proto_seq_stop, 3018 .show = proto_seq_show, 3019 }; 3020 3021 static int proto_seq_open(struct inode *inode, struct file *file) 3022 { 3023 return seq_open_net(inode, file, &proto_seq_ops, 3024 sizeof(struct seq_net_private)); 3025 } 3026 3027 static const struct file_operations proto_seq_fops = { 3028 .owner = THIS_MODULE, 3029 .open = proto_seq_open, 3030 .read = seq_read, 3031 .llseek = seq_lseek, 3032 .release = seq_release_net, 3033 }; 3034 3035 static __net_init int proto_init_net(struct net *net) 3036 { 3037 if (!proc_create("protocols", S_IRUGO, net->proc_net, &proto_seq_fops)) 3038 return -ENOMEM; 3039 3040 return 0; 3041 } 3042 3043 static __net_exit void proto_exit_net(struct net *net) 3044 { 3045 remove_proc_entry("protocols", net->proc_net); 3046 } 3047 3048 3049 static __net_initdata struct pernet_operations proto_net_ops = { 3050 .init = proto_init_net, 3051 .exit = proto_exit_net, 3052 }; 3053 3054 static int __init proto_init(void) 3055 { 3056 return register_pernet_subsys(&proto_net_ops); 3057 } 3058 3059 subsys_initcall(proto_init); 3060 3061 #endif /* PROC_FS */ 3062