1 /* 2 * INET An implementation of the TCP/IP protocol suite for the LINUX 3 * operating system. INET is implemented using the BSD Socket 4 * interface as the means of communication with the user level. 5 * 6 * Generic socket support routines. Memory allocators, socket lock/release 7 * handler for protocols to use and generic option handler. 8 * 9 * 10 * Authors: Ross Biro 11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 12 * Florian La Roche, <flla@stud.uni-sb.de> 13 * Alan Cox, <A.Cox@swansea.ac.uk> 14 * 15 * Fixes: 16 * Alan Cox : Numerous verify_area() problems 17 * Alan Cox : Connecting on a connecting socket 18 * now returns an error for tcp. 19 * Alan Cox : sock->protocol is set correctly. 20 * and is not sometimes left as 0. 21 * Alan Cox : connect handles icmp errors on a 22 * connect properly. Unfortunately there 23 * is a restart syscall nasty there. I 24 * can't match BSD without hacking the C 25 * library. Ideas urgently sought! 26 * Alan Cox : Disallow bind() to addresses that are 27 * not ours - especially broadcast ones!! 28 * Alan Cox : Socket 1024 _IS_ ok for users. (fencepost) 29 * Alan Cox : sock_wfree/sock_rfree don't destroy sockets, 30 * instead they leave that for the DESTROY timer. 31 * Alan Cox : Clean up error flag in accept 32 * Alan Cox : TCP ack handling is buggy, the DESTROY timer 33 * was buggy. Put a remove_sock() in the handler 34 * for memory when we hit 0. Also altered the timer 35 * code. The ACK stuff can wait and needs major 36 * TCP layer surgery. 37 * Alan Cox : Fixed TCP ack bug, removed remove sock 38 * and fixed timer/inet_bh race. 39 * Alan Cox : Added zapped flag for TCP 40 * Alan Cox : Move kfree_skb into skbuff.c and tidied up surplus code 41 * Alan Cox : for new sk_buff allocations wmalloc/rmalloc now call alloc_skb 42 * Alan Cox : kfree_s calls now are kfree_skbmem so we can track skb resources 43 * Alan Cox : Supports socket option broadcast now as does udp. Packet and raw need fixing. 44 * Alan Cox : Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so... 45 * Rick Sladkey : Relaxed UDP rules for matching packets. 46 * C.E.Hawkins : IFF_PROMISC/SIOCGHWADDR support 47 * Pauline Middelink : identd support 48 * Alan Cox : Fixed connect() taking signals I think. 49 * Alan Cox : SO_LINGER supported 50 * Alan Cox : Error reporting fixes 51 * Anonymous : inet_create tidied up (sk->reuse setting) 52 * Alan Cox : inet sockets don't set sk->type! 53 * Alan Cox : Split socket option code 54 * Alan Cox : Callbacks 55 * Alan Cox : Nagle flag for Charles & Johannes stuff 56 * Alex : Removed restriction on inet fioctl 57 * Alan Cox : Splitting INET from NET core 58 * Alan Cox : Fixed bogus SO_TYPE handling in getsockopt() 59 * Adam Caldwell : Missing return in SO_DONTROUTE/SO_DEBUG code 60 * Alan Cox : Split IP from generic code 61 * Alan Cox : New kfree_skbmem() 62 * Alan Cox : Make SO_DEBUG superuser only. 63 * Alan Cox : Allow anyone to clear SO_DEBUG 64 * (compatibility fix) 65 * Alan Cox : Added optimistic memory grabbing for AF_UNIX throughput. 66 * Alan Cox : Allocator for a socket is settable. 67 * Alan Cox : SO_ERROR includes soft errors. 68 * Alan Cox : Allow NULL arguments on some SO_ opts 69 * Alan Cox : Generic socket allocation to make hooks 70 * easier (suggested by Craig Metz). 71 * Michael Pall : SO_ERROR returns positive errno again 72 * Steve Whitehouse: Added default destructor to free 73 * protocol private data. 74 * Steve Whitehouse: Added various other default routines 75 * common to several socket families. 76 * Chris Evans : Call suser() check last on F_SETOWN 77 * Jay Schulist : Added SO_ATTACH_FILTER and SO_DETACH_FILTER. 78 * Andi Kleen : Add sock_kmalloc()/sock_kfree_s() 79 * Andi Kleen : Fix write_space callback 80 * Chris Evans : Security fixes - signedness again 81 * Arnaldo C. Melo : cleanups, use skb_queue_purge 82 * 83 * To Fix: 84 * 85 * 86 * This program is free software; you can redistribute it and/or 87 * modify it under the terms of the GNU General Public License 88 * as published by the Free Software Foundation; either version 89 * 2 of the License, or (at your option) any later version. 90 */ 91 92 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 93 94 #include <linux/capability.h> 95 #include <linux/errno.h> 96 #include <linux/errqueue.h> 97 #include <linux/types.h> 98 #include <linux/socket.h> 99 #include <linux/in.h> 100 #include <linux/kernel.h> 101 #include <linux/module.h> 102 #include <linux/proc_fs.h> 103 #include <linux/seq_file.h> 104 #include <linux/sched.h> 105 #include <linux/timer.h> 106 #include <linux/string.h> 107 #include <linux/sockios.h> 108 #include <linux/net.h> 109 #include <linux/mm.h> 110 #include <linux/slab.h> 111 #include <linux/interrupt.h> 112 #include <linux/poll.h> 113 #include <linux/tcp.h> 114 #include <linux/init.h> 115 #include <linux/highmem.h> 116 #include <linux/user_namespace.h> 117 #include <linux/static_key.h> 118 #include <linux/memcontrol.h> 119 #include <linux/prefetch.h> 120 121 #include <asm/uaccess.h> 122 123 #include <linux/netdevice.h> 124 #include <net/protocol.h> 125 #include <linux/skbuff.h> 126 #include <net/net_namespace.h> 127 #include <net/request_sock.h> 128 #include <net/sock.h> 129 #include <linux/net_tstamp.h> 130 #include <net/xfrm.h> 131 #include <linux/ipsec.h> 132 #include <net/cls_cgroup.h> 133 #include <net/netprio_cgroup.h> 134 135 #include <linux/filter.h> 136 137 #include <trace/events/sock.h> 138 139 #ifdef CONFIG_INET 140 #include <net/tcp.h> 141 #endif 142 143 #include <net/busy_poll.h> 144 145 static DEFINE_MUTEX(proto_list_mutex); 146 static LIST_HEAD(proto_list); 147 148 /** 149 * sk_ns_capable - General socket capability test 150 * @sk: Socket to use a capability on or through 151 * @user_ns: The user namespace of the capability to use 152 * @cap: The capability to use 153 * 154 * Test to see if the opener of the socket had when the socket was 155 * created and the current process has the capability @cap in the user 156 * namespace @user_ns. 157 */ 158 bool sk_ns_capable(const struct sock *sk, 159 struct user_namespace *user_ns, int cap) 160 { 161 return file_ns_capable(sk->sk_socket->file, user_ns, cap) && 162 ns_capable(user_ns, cap); 163 } 164 EXPORT_SYMBOL(sk_ns_capable); 165 166 /** 167 * sk_capable - Socket global capability test 168 * @sk: Socket to use a capability on or through 169 * @cap: The global capbility to use 170 * 171 * Test to see if the opener of the socket had when the socket was 172 * created and the current process has the capability @cap in all user 173 * namespaces. 174 */ 175 bool sk_capable(const struct sock *sk, int cap) 176 { 177 return sk_ns_capable(sk, &init_user_ns, cap); 178 } 179 EXPORT_SYMBOL(sk_capable); 180 181 /** 182 * sk_net_capable - Network namespace socket capability test 183 * @sk: Socket to use a capability on or through 184 * @cap: The capability to use 185 * 186 * Test to see if the opener of the socket had when the socke was created 187 * and the current process has the capability @cap over the network namespace 188 * the socket is a member of. 189 */ 190 bool sk_net_capable(const struct sock *sk, int cap) 191 { 192 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap); 193 } 194 EXPORT_SYMBOL(sk_net_capable); 195 196 197 #ifdef CONFIG_MEMCG_KMEM 198 int mem_cgroup_sockets_init(struct mem_cgroup *memcg, struct cgroup_subsys *ss) 199 { 200 struct proto *proto; 201 int ret = 0; 202 203 mutex_lock(&proto_list_mutex); 204 list_for_each_entry(proto, &proto_list, node) { 205 if (proto->init_cgroup) { 206 ret = proto->init_cgroup(memcg, ss); 207 if (ret) 208 goto out; 209 } 210 } 211 212 mutex_unlock(&proto_list_mutex); 213 return ret; 214 out: 215 list_for_each_entry_continue_reverse(proto, &proto_list, node) 216 if (proto->destroy_cgroup) 217 proto->destroy_cgroup(memcg); 218 mutex_unlock(&proto_list_mutex); 219 return ret; 220 } 221 222 void mem_cgroup_sockets_destroy(struct mem_cgroup *memcg) 223 { 224 struct proto *proto; 225 226 mutex_lock(&proto_list_mutex); 227 list_for_each_entry_reverse(proto, &proto_list, node) 228 if (proto->destroy_cgroup) 229 proto->destroy_cgroup(memcg); 230 mutex_unlock(&proto_list_mutex); 231 } 232 #endif 233 234 /* 235 * Each address family might have different locking rules, so we have 236 * one slock key per address family: 237 */ 238 static struct lock_class_key af_family_keys[AF_MAX]; 239 static struct lock_class_key af_family_slock_keys[AF_MAX]; 240 241 #if defined(CONFIG_MEMCG_KMEM) 242 struct static_key memcg_socket_limit_enabled; 243 EXPORT_SYMBOL(memcg_socket_limit_enabled); 244 #endif 245 246 /* 247 * Make lock validator output more readable. (we pre-construct these 248 * strings build-time, so that runtime initialization of socket 249 * locks is fast): 250 */ 251 static const char *const af_family_key_strings[AF_MAX+1] = { 252 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" , 253 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK", 254 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" , 255 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" , 256 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" , 257 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" , 258 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" , 259 "sk_lock-AF_RDS" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" , 260 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" , 261 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" , 262 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" , 263 "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN" , "sk_lock-AF_PHONET" , 264 "sk_lock-AF_IEEE802154", "sk_lock-AF_CAIF" , "sk_lock-AF_ALG" , 265 "sk_lock-AF_NFC" , "sk_lock-AF_VSOCK" , "sk_lock-AF_MAX" 266 }; 267 static const char *const af_family_slock_key_strings[AF_MAX+1] = { 268 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" , 269 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK", 270 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" , 271 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" , 272 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" , 273 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" , 274 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" , 275 "slock-AF_RDS" , "slock-AF_SNA" , "slock-AF_IRDA" , 276 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" , 277 "slock-27" , "slock-28" , "slock-AF_CAN" , 278 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" , 279 "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" , 280 "slock-AF_IEEE802154", "slock-AF_CAIF" , "slock-AF_ALG" , 281 "slock-AF_NFC" , "slock-AF_VSOCK" ,"slock-AF_MAX" 282 }; 283 static const char *const af_family_clock_key_strings[AF_MAX+1] = { 284 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" , 285 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK", 286 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" , 287 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" , 288 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" , 289 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" , 290 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" , 291 "clock-AF_RDS" , "clock-AF_SNA" , "clock-AF_IRDA" , 292 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" , 293 "clock-27" , "clock-28" , "clock-AF_CAN" , 294 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" , 295 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" , 296 "clock-AF_IEEE802154", "clock-AF_CAIF" , "clock-AF_ALG" , 297 "clock-AF_NFC" , "clock-AF_VSOCK" , "clock-AF_MAX" 298 }; 299 300 /* 301 * sk_callback_lock locking rules are per-address-family, 302 * so split the lock classes by using a per-AF key: 303 */ 304 static struct lock_class_key af_callback_keys[AF_MAX]; 305 306 /* Take into consideration the size of the struct sk_buff overhead in the 307 * determination of these values, since that is non-constant across 308 * platforms. This makes socket queueing behavior and performance 309 * not depend upon such differences. 310 */ 311 #define _SK_MEM_PACKETS 256 312 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256) 313 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 314 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 315 316 /* Run time adjustable parameters. */ 317 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX; 318 EXPORT_SYMBOL(sysctl_wmem_max); 319 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX; 320 EXPORT_SYMBOL(sysctl_rmem_max); 321 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX; 322 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX; 323 324 /* Maximal space eaten by iovec or ancillary data plus some space */ 325 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512); 326 EXPORT_SYMBOL(sysctl_optmem_max); 327 328 struct static_key memalloc_socks = STATIC_KEY_INIT_FALSE; 329 EXPORT_SYMBOL_GPL(memalloc_socks); 330 331 /** 332 * sk_set_memalloc - sets %SOCK_MEMALLOC 333 * @sk: socket to set it on 334 * 335 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves. 336 * It's the responsibility of the admin to adjust min_free_kbytes 337 * to meet the requirements 338 */ 339 void sk_set_memalloc(struct sock *sk) 340 { 341 sock_set_flag(sk, SOCK_MEMALLOC); 342 sk->sk_allocation |= __GFP_MEMALLOC; 343 static_key_slow_inc(&memalloc_socks); 344 } 345 EXPORT_SYMBOL_GPL(sk_set_memalloc); 346 347 void sk_clear_memalloc(struct sock *sk) 348 { 349 sock_reset_flag(sk, SOCK_MEMALLOC); 350 sk->sk_allocation &= ~__GFP_MEMALLOC; 351 static_key_slow_dec(&memalloc_socks); 352 353 /* 354 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward 355 * progress of swapping. However, if SOCK_MEMALLOC is cleared while 356 * it has rmem allocations there is a risk that the user of the 357 * socket cannot make forward progress due to exceeding the rmem 358 * limits. By rights, sk_clear_memalloc() should only be called 359 * on sockets being torn down but warn and reset the accounting if 360 * that assumption breaks. 361 */ 362 if (WARN_ON(sk->sk_forward_alloc)) 363 sk_mem_reclaim(sk); 364 } 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 int skb_len; 441 unsigned long flags; 442 struct sk_buff_head *list = &sk->sk_receive_queue; 443 444 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) { 445 atomic_inc(&sk->sk_drops); 446 trace_sock_rcvqueue_full(sk, skb); 447 return -ENOMEM; 448 } 449 450 err = sk_filter(sk, skb); 451 if (err) 452 return err; 453 454 if (!sk_rmem_schedule(sk, skb, skb->truesize)) { 455 atomic_inc(&sk->sk_drops); 456 return -ENOBUFS; 457 } 458 459 skb->dev = NULL; 460 skb_set_owner_r(skb, sk); 461 462 /* Cache the SKB length before we tack it onto the receive 463 * queue. Once it is added it no longer belongs to us and 464 * may be freed by other threads of control pulling packets 465 * from the queue. 466 */ 467 skb_len = skb->len; 468 469 /* we escape from rcu protected region, make sure we dont leak 470 * a norefcounted dst 471 */ 472 skb_dst_force(skb); 473 474 spin_lock_irqsave(&list->lock, flags); 475 skb->dropcount = atomic_read(&sk->sk_drops); 476 __skb_queue_tail(list, skb); 477 spin_unlock_irqrestore(&list->lock, flags); 478 479 if (!sock_flag(sk, SOCK_DEAD)) 480 sk->sk_data_ready(sk); 481 return 0; 482 } 483 EXPORT_SYMBOL(sock_queue_rcv_skb); 484 485 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested) 486 { 487 int rc = NET_RX_SUCCESS; 488 489 if (sk_filter(sk, skb)) 490 goto discard_and_relse; 491 492 skb->dev = NULL; 493 494 if (sk_rcvqueues_full(sk, skb, sk->sk_rcvbuf)) { 495 atomic_inc(&sk->sk_drops); 496 goto discard_and_relse; 497 } 498 if (nested) 499 bh_lock_sock_nested(sk); 500 else 501 bh_lock_sock(sk); 502 if (!sock_owned_by_user(sk)) { 503 /* 504 * trylock + unlock semantics: 505 */ 506 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_); 507 508 rc = sk_backlog_rcv(sk, skb); 509 510 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_); 511 } else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) { 512 bh_unlock_sock(sk); 513 atomic_inc(&sk->sk_drops); 514 goto discard_and_relse; 515 } 516 517 bh_unlock_sock(sk); 518 out: 519 sock_put(sk); 520 return rc; 521 discard_and_relse: 522 kfree_skb(skb); 523 goto out; 524 } 525 EXPORT_SYMBOL(sk_receive_skb); 526 527 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie) 528 { 529 struct dst_entry *dst = __sk_dst_get(sk); 530 531 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) { 532 sk_tx_queue_clear(sk); 533 RCU_INIT_POINTER(sk->sk_dst_cache, NULL); 534 dst_release(dst); 535 return NULL; 536 } 537 538 return dst; 539 } 540 EXPORT_SYMBOL(__sk_dst_check); 541 542 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie) 543 { 544 struct dst_entry *dst = sk_dst_get(sk); 545 546 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) { 547 sk_dst_reset(sk); 548 dst_release(dst); 549 return NULL; 550 } 551 552 return dst; 553 } 554 EXPORT_SYMBOL(sk_dst_check); 555 556 static int sock_setbindtodevice(struct sock *sk, char __user *optval, 557 int optlen) 558 { 559 int ret = -ENOPROTOOPT; 560 #ifdef CONFIG_NETDEVICES 561 struct net *net = sock_net(sk); 562 char devname[IFNAMSIZ]; 563 int index; 564 565 /* Sorry... */ 566 ret = -EPERM; 567 if (!ns_capable(net->user_ns, CAP_NET_RAW)) 568 goto out; 569 570 ret = -EINVAL; 571 if (optlen < 0) 572 goto out; 573 574 /* Bind this socket to a particular device like "eth0", 575 * as specified in the passed interface name. If the 576 * name is "" or the option length is zero the socket 577 * is not bound. 578 */ 579 if (optlen > IFNAMSIZ - 1) 580 optlen = IFNAMSIZ - 1; 581 memset(devname, 0, sizeof(devname)); 582 583 ret = -EFAULT; 584 if (copy_from_user(devname, optval, optlen)) 585 goto out; 586 587 index = 0; 588 if (devname[0] != '\0') { 589 struct net_device *dev; 590 591 rcu_read_lock(); 592 dev = dev_get_by_name_rcu(net, devname); 593 if (dev) 594 index = dev->ifindex; 595 rcu_read_unlock(); 596 ret = -ENODEV; 597 if (!dev) 598 goto out; 599 } 600 601 lock_sock(sk); 602 sk->sk_bound_dev_if = index; 603 sk_dst_reset(sk); 604 release_sock(sk); 605 606 ret = 0; 607 608 out: 609 #endif 610 611 return ret; 612 } 613 614 static int sock_getbindtodevice(struct sock *sk, char __user *optval, 615 int __user *optlen, int len) 616 { 617 int ret = -ENOPROTOOPT; 618 #ifdef CONFIG_NETDEVICES 619 struct net *net = sock_net(sk); 620 char devname[IFNAMSIZ]; 621 622 if (sk->sk_bound_dev_if == 0) { 623 len = 0; 624 goto zero; 625 } 626 627 ret = -EINVAL; 628 if (len < IFNAMSIZ) 629 goto out; 630 631 ret = netdev_get_name(net, devname, sk->sk_bound_dev_if); 632 if (ret) 633 goto out; 634 635 len = strlen(devname) + 1; 636 637 ret = -EFAULT; 638 if (copy_to_user(optval, devname, len)) 639 goto out; 640 641 zero: 642 ret = -EFAULT; 643 if (put_user(len, optlen)) 644 goto out; 645 646 ret = 0; 647 648 out: 649 #endif 650 651 return ret; 652 } 653 654 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool) 655 { 656 if (valbool) 657 sock_set_flag(sk, bit); 658 else 659 sock_reset_flag(sk, bit); 660 } 661 662 /* 663 * This is meant for all protocols to use and covers goings on 664 * at the socket level. Everything here is generic. 665 */ 666 667 int sock_setsockopt(struct socket *sock, int level, int optname, 668 char __user *optval, unsigned int optlen) 669 { 670 struct sock *sk = sock->sk; 671 int val; 672 int valbool; 673 struct linger ling; 674 int ret = 0; 675 676 /* 677 * Options without arguments 678 */ 679 680 if (optname == SO_BINDTODEVICE) 681 return sock_setbindtodevice(sk, optval, optlen); 682 683 if (optlen < sizeof(int)) 684 return -EINVAL; 685 686 if (get_user(val, (int __user *)optval)) 687 return -EFAULT; 688 689 valbool = val ? 1 : 0; 690 691 lock_sock(sk); 692 693 switch (optname) { 694 case SO_DEBUG: 695 if (val && !capable(CAP_NET_ADMIN)) 696 ret = -EACCES; 697 else 698 sock_valbool_flag(sk, SOCK_DBG, valbool); 699 break; 700 case SO_REUSEADDR: 701 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE); 702 break; 703 case SO_REUSEPORT: 704 sk->sk_reuseport = valbool; 705 break; 706 case SO_TYPE: 707 case SO_PROTOCOL: 708 case SO_DOMAIN: 709 case SO_ERROR: 710 ret = -ENOPROTOOPT; 711 break; 712 case SO_DONTROUTE: 713 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool); 714 break; 715 case SO_BROADCAST: 716 sock_valbool_flag(sk, SOCK_BROADCAST, valbool); 717 break; 718 case SO_SNDBUF: 719 /* Don't error on this BSD doesn't and if you think 720 * about it this is right. Otherwise apps have to 721 * play 'guess the biggest size' games. RCVBUF/SNDBUF 722 * are treated in BSD as hints 723 */ 724 val = min_t(u32, val, sysctl_wmem_max); 725 set_sndbuf: 726 sk->sk_userlocks |= SOCK_SNDBUF_LOCK; 727 sk->sk_sndbuf = max_t(u32, val * 2, SOCK_MIN_SNDBUF); 728 /* Wake up sending tasks if we upped the value. */ 729 sk->sk_write_space(sk); 730 break; 731 732 case SO_SNDBUFFORCE: 733 if (!capable(CAP_NET_ADMIN)) { 734 ret = -EPERM; 735 break; 736 } 737 goto set_sndbuf; 738 739 case SO_RCVBUF: 740 /* Don't error on this BSD doesn't and if you think 741 * about it this is right. Otherwise apps have to 742 * play 'guess the biggest size' games. RCVBUF/SNDBUF 743 * are treated in BSD as hints 744 */ 745 val = min_t(u32, val, sysctl_rmem_max); 746 set_rcvbuf: 747 sk->sk_userlocks |= SOCK_RCVBUF_LOCK; 748 /* 749 * We double it on the way in to account for 750 * "struct sk_buff" etc. overhead. Applications 751 * assume that the SO_RCVBUF setting they make will 752 * allow that much actual data to be received on that 753 * socket. 754 * 755 * Applications are unaware that "struct sk_buff" and 756 * other overheads allocate from the receive buffer 757 * during socket buffer allocation. 758 * 759 * And after considering the possible alternatives, 760 * returning the value we actually used in getsockopt 761 * is the most desirable behavior. 762 */ 763 sk->sk_rcvbuf = max_t(u32, val * 2, SOCK_MIN_RCVBUF); 764 break; 765 766 case SO_RCVBUFFORCE: 767 if (!capable(CAP_NET_ADMIN)) { 768 ret = -EPERM; 769 break; 770 } 771 goto set_rcvbuf; 772 773 case SO_KEEPALIVE: 774 #ifdef CONFIG_INET 775 if (sk->sk_protocol == IPPROTO_TCP && 776 sk->sk_type == SOCK_STREAM) 777 tcp_set_keepalive(sk, valbool); 778 #endif 779 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool); 780 break; 781 782 case SO_OOBINLINE: 783 sock_valbool_flag(sk, SOCK_URGINLINE, valbool); 784 break; 785 786 case SO_NO_CHECK: 787 sk->sk_no_check_tx = valbool; 788 break; 789 790 case SO_PRIORITY: 791 if ((val >= 0 && val <= 6) || 792 ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) 793 sk->sk_priority = val; 794 else 795 ret = -EPERM; 796 break; 797 798 case SO_LINGER: 799 if (optlen < sizeof(ling)) { 800 ret = -EINVAL; /* 1003.1g */ 801 break; 802 } 803 if (copy_from_user(&ling, optval, sizeof(ling))) { 804 ret = -EFAULT; 805 break; 806 } 807 if (!ling.l_onoff) 808 sock_reset_flag(sk, SOCK_LINGER); 809 else { 810 #if (BITS_PER_LONG == 32) 811 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ) 812 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT; 813 else 814 #endif 815 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ; 816 sock_set_flag(sk, SOCK_LINGER); 817 } 818 break; 819 820 case SO_BSDCOMPAT: 821 sock_warn_obsolete_bsdism("setsockopt"); 822 break; 823 824 case SO_PASSCRED: 825 if (valbool) 826 set_bit(SOCK_PASSCRED, &sock->flags); 827 else 828 clear_bit(SOCK_PASSCRED, &sock->flags); 829 break; 830 831 case SO_TIMESTAMP: 832 case SO_TIMESTAMPNS: 833 if (valbool) { 834 if (optname == SO_TIMESTAMP) 835 sock_reset_flag(sk, SOCK_RCVTSTAMPNS); 836 else 837 sock_set_flag(sk, SOCK_RCVTSTAMPNS); 838 sock_set_flag(sk, SOCK_RCVTSTAMP); 839 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 840 } else { 841 sock_reset_flag(sk, SOCK_RCVTSTAMP); 842 sock_reset_flag(sk, SOCK_RCVTSTAMPNS); 843 } 844 break; 845 846 case SO_TIMESTAMPING: 847 if (val & ~SOF_TIMESTAMPING_MASK) { 848 ret = -EINVAL; 849 break; 850 } 851 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE, 852 val & SOF_TIMESTAMPING_TX_HARDWARE); 853 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE, 854 val & SOF_TIMESTAMPING_TX_SOFTWARE); 855 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE, 856 val & SOF_TIMESTAMPING_RX_HARDWARE); 857 if (val & SOF_TIMESTAMPING_RX_SOFTWARE) 858 sock_enable_timestamp(sk, 859 SOCK_TIMESTAMPING_RX_SOFTWARE); 860 else 861 sock_disable_timestamp(sk, 862 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE)); 863 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SOFTWARE, 864 val & SOF_TIMESTAMPING_SOFTWARE); 865 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE, 866 val & SOF_TIMESTAMPING_SYS_HARDWARE); 867 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE, 868 val & SOF_TIMESTAMPING_RAW_HARDWARE); 869 break; 870 871 case SO_RCVLOWAT: 872 if (val < 0) 873 val = INT_MAX; 874 sk->sk_rcvlowat = val ? : 1; 875 break; 876 877 case SO_RCVTIMEO: 878 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen); 879 break; 880 881 case SO_SNDTIMEO: 882 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen); 883 break; 884 885 case SO_ATTACH_FILTER: 886 ret = -EINVAL; 887 if (optlen == sizeof(struct sock_fprog)) { 888 struct sock_fprog fprog; 889 890 ret = -EFAULT; 891 if (copy_from_user(&fprog, optval, sizeof(fprog))) 892 break; 893 894 ret = sk_attach_filter(&fprog, sk); 895 } 896 break; 897 898 case SO_DETACH_FILTER: 899 ret = sk_detach_filter(sk); 900 break; 901 902 case SO_LOCK_FILTER: 903 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool) 904 ret = -EPERM; 905 else 906 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool); 907 break; 908 909 case SO_PASSSEC: 910 if (valbool) 911 set_bit(SOCK_PASSSEC, &sock->flags); 912 else 913 clear_bit(SOCK_PASSSEC, &sock->flags); 914 break; 915 case SO_MARK: 916 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) 917 ret = -EPERM; 918 else 919 sk->sk_mark = val; 920 break; 921 922 /* We implement the SO_SNDLOWAT etc to 923 not be settable (1003.1g 5.3) */ 924 case SO_RXQ_OVFL: 925 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool); 926 break; 927 928 case SO_WIFI_STATUS: 929 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool); 930 break; 931 932 case SO_PEEK_OFF: 933 if (sock->ops->set_peek_off) 934 ret = sock->ops->set_peek_off(sk, val); 935 else 936 ret = -EOPNOTSUPP; 937 break; 938 939 case SO_NOFCS: 940 sock_valbool_flag(sk, SOCK_NOFCS, valbool); 941 break; 942 943 case SO_SELECT_ERR_QUEUE: 944 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool); 945 break; 946 947 #ifdef CONFIG_NET_RX_BUSY_POLL 948 case SO_BUSY_POLL: 949 /* allow unprivileged users to decrease the value */ 950 if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN)) 951 ret = -EPERM; 952 else { 953 if (val < 0) 954 ret = -EINVAL; 955 else 956 sk->sk_ll_usec = val; 957 } 958 break; 959 #endif 960 961 case SO_MAX_PACING_RATE: 962 sk->sk_max_pacing_rate = val; 963 sk->sk_pacing_rate = min(sk->sk_pacing_rate, 964 sk->sk_max_pacing_rate); 965 break; 966 967 default: 968 ret = -ENOPROTOOPT; 969 break; 970 } 971 release_sock(sk); 972 return ret; 973 } 974 EXPORT_SYMBOL(sock_setsockopt); 975 976 977 static void cred_to_ucred(struct pid *pid, const struct cred *cred, 978 struct ucred *ucred) 979 { 980 ucred->pid = pid_vnr(pid); 981 ucred->uid = ucred->gid = -1; 982 if (cred) { 983 struct user_namespace *current_ns = current_user_ns(); 984 985 ucred->uid = from_kuid_munged(current_ns, cred->euid); 986 ucred->gid = from_kgid_munged(current_ns, cred->egid); 987 } 988 } 989 990 int sock_getsockopt(struct socket *sock, int level, int optname, 991 char __user *optval, int __user *optlen) 992 { 993 struct sock *sk = sock->sk; 994 995 union { 996 int val; 997 struct linger ling; 998 struct timeval tm; 999 } v; 1000 1001 int lv = sizeof(int); 1002 int len; 1003 1004 if (get_user(len, optlen)) 1005 return -EFAULT; 1006 if (len < 0) 1007 return -EINVAL; 1008 1009 memset(&v, 0, sizeof(v)); 1010 1011 switch (optname) { 1012 case SO_DEBUG: 1013 v.val = sock_flag(sk, SOCK_DBG); 1014 break; 1015 1016 case SO_DONTROUTE: 1017 v.val = sock_flag(sk, SOCK_LOCALROUTE); 1018 break; 1019 1020 case SO_BROADCAST: 1021 v.val = sock_flag(sk, SOCK_BROADCAST); 1022 break; 1023 1024 case SO_SNDBUF: 1025 v.val = sk->sk_sndbuf; 1026 break; 1027 1028 case SO_RCVBUF: 1029 v.val = sk->sk_rcvbuf; 1030 break; 1031 1032 case SO_REUSEADDR: 1033 v.val = sk->sk_reuse; 1034 break; 1035 1036 case SO_REUSEPORT: 1037 v.val = sk->sk_reuseport; 1038 break; 1039 1040 case SO_KEEPALIVE: 1041 v.val = sock_flag(sk, SOCK_KEEPOPEN); 1042 break; 1043 1044 case SO_TYPE: 1045 v.val = sk->sk_type; 1046 break; 1047 1048 case SO_PROTOCOL: 1049 v.val = sk->sk_protocol; 1050 break; 1051 1052 case SO_DOMAIN: 1053 v.val = sk->sk_family; 1054 break; 1055 1056 case SO_ERROR: 1057 v.val = -sock_error(sk); 1058 if (v.val == 0) 1059 v.val = xchg(&sk->sk_err_soft, 0); 1060 break; 1061 1062 case SO_OOBINLINE: 1063 v.val = sock_flag(sk, SOCK_URGINLINE); 1064 break; 1065 1066 case SO_NO_CHECK: 1067 v.val = sk->sk_no_check_tx; 1068 break; 1069 1070 case SO_PRIORITY: 1071 v.val = sk->sk_priority; 1072 break; 1073 1074 case SO_LINGER: 1075 lv = sizeof(v.ling); 1076 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER); 1077 v.ling.l_linger = sk->sk_lingertime / HZ; 1078 break; 1079 1080 case SO_BSDCOMPAT: 1081 sock_warn_obsolete_bsdism("getsockopt"); 1082 break; 1083 1084 case SO_TIMESTAMP: 1085 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && 1086 !sock_flag(sk, SOCK_RCVTSTAMPNS); 1087 break; 1088 1089 case SO_TIMESTAMPNS: 1090 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS); 1091 break; 1092 1093 case SO_TIMESTAMPING: 1094 v.val = 0; 1095 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE)) 1096 v.val |= SOF_TIMESTAMPING_TX_HARDWARE; 1097 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE)) 1098 v.val |= SOF_TIMESTAMPING_TX_SOFTWARE; 1099 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE)) 1100 v.val |= SOF_TIMESTAMPING_RX_HARDWARE; 1101 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE)) 1102 v.val |= SOF_TIMESTAMPING_RX_SOFTWARE; 1103 if (sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE)) 1104 v.val |= SOF_TIMESTAMPING_SOFTWARE; 1105 if (sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE)) 1106 v.val |= SOF_TIMESTAMPING_SYS_HARDWARE; 1107 if (sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE)) 1108 v.val |= SOF_TIMESTAMPING_RAW_HARDWARE; 1109 break; 1110 1111 case SO_RCVTIMEO: 1112 lv = sizeof(struct timeval); 1113 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) { 1114 v.tm.tv_sec = 0; 1115 v.tm.tv_usec = 0; 1116 } else { 1117 v.tm.tv_sec = sk->sk_rcvtimeo / HZ; 1118 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ; 1119 } 1120 break; 1121 1122 case SO_SNDTIMEO: 1123 lv = sizeof(struct timeval); 1124 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) { 1125 v.tm.tv_sec = 0; 1126 v.tm.tv_usec = 0; 1127 } else { 1128 v.tm.tv_sec = sk->sk_sndtimeo / HZ; 1129 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ; 1130 } 1131 break; 1132 1133 case SO_RCVLOWAT: 1134 v.val = sk->sk_rcvlowat; 1135 break; 1136 1137 case SO_SNDLOWAT: 1138 v.val = 1; 1139 break; 1140 1141 case SO_PASSCRED: 1142 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags); 1143 break; 1144 1145 case SO_PEERCRED: 1146 { 1147 struct ucred peercred; 1148 if (len > sizeof(peercred)) 1149 len = sizeof(peercred); 1150 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred); 1151 if (copy_to_user(optval, &peercred, len)) 1152 return -EFAULT; 1153 goto lenout; 1154 } 1155 1156 case SO_PEERNAME: 1157 { 1158 char address[128]; 1159 1160 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2)) 1161 return -ENOTCONN; 1162 if (lv < len) 1163 return -EINVAL; 1164 if (copy_to_user(optval, address, len)) 1165 return -EFAULT; 1166 goto lenout; 1167 } 1168 1169 /* Dubious BSD thing... Probably nobody even uses it, but 1170 * the UNIX standard wants it for whatever reason... -DaveM 1171 */ 1172 case SO_ACCEPTCONN: 1173 v.val = sk->sk_state == TCP_LISTEN; 1174 break; 1175 1176 case SO_PASSSEC: 1177 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags); 1178 break; 1179 1180 case SO_PEERSEC: 1181 return security_socket_getpeersec_stream(sock, optval, optlen, len); 1182 1183 case SO_MARK: 1184 v.val = sk->sk_mark; 1185 break; 1186 1187 case SO_RXQ_OVFL: 1188 v.val = sock_flag(sk, SOCK_RXQ_OVFL); 1189 break; 1190 1191 case SO_WIFI_STATUS: 1192 v.val = sock_flag(sk, SOCK_WIFI_STATUS); 1193 break; 1194 1195 case SO_PEEK_OFF: 1196 if (!sock->ops->set_peek_off) 1197 return -EOPNOTSUPP; 1198 1199 v.val = sk->sk_peek_off; 1200 break; 1201 case SO_NOFCS: 1202 v.val = sock_flag(sk, SOCK_NOFCS); 1203 break; 1204 1205 case SO_BINDTODEVICE: 1206 return sock_getbindtodevice(sk, optval, optlen, len); 1207 1208 case SO_GET_FILTER: 1209 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len); 1210 if (len < 0) 1211 return len; 1212 1213 goto lenout; 1214 1215 case SO_LOCK_FILTER: 1216 v.val = sock_flag(sk, SOCK_FILTER_LOCKED); 1217 break; 1218 1219 case SO_BPF_EXTENSIONS: 1220 v.val = bpf_tell_extensions(); 1221 break; 1222 1223 case SO_SELECT_ERR_QUEUE: 1224 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE); 1225 break; 1226 1227 #ifdef CONFIG_NET_RX_BUSY_POLL 1228 case SO_BUSY_POLL: 1229 v.val = sk->sk_ll_usec; 1230 break; 1231 #endif 1232 1233 case SO_MAX_PACING_RATE: 1234 v.val = sk->sk_max_pacing_rate; 1235 break; 1236 1237 default: 1238 return -ENOPROTOOPT; 1239 } 1240 1241 if (len > lv) 1242 len = lv; 1243 if (copy_to_user(optval, &v, len)) 1244 return -EFAULT; 1245 lenout: 1246 if (put_user(len, optlen)) 1247 return -EFAULT; 1248 return 0; 1249 } 1250 1251 /* 1252 * Initialize an sk_lock. 1253 * 1254 * (We also register the sk_lock with the lock validator.) 1255 */ 1256 static inline void sock_lock_init(struct sock *sk) 1257 { 1258 sock_lock_init_class_and_name(sk, 1259 af_family_slock_key_strings[sk->sk_family], 1260 af_family_slock_keys + sk->sk_family, 1261 af_family_key_strings[sk->sk_family], 1262 af_family_keys + sk->sk_family); 1263 } 1264 1265 /* 1266 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet, 1267 * even temporarly, because of RCU lookups. sk_node should also be left as is. 1268 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end 1269 */ 1270 static void sock_copy(struct sock *nsk, const struct sock *osk) 1271 { 1272 #ifdef CONFIG_SECURITY_NETWORK 1273 void *sptr = nsk->sk_security; 1274 #endif 1275 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin)); 1276 1277 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end, 1278 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end)); 1279 1280 #ifdef CONFIG_SECURITY_NETWORK 1281 nsk->sk_security = sptr; 1282 security_sk_clone(osk, nsk); 1283 #endif 1284 } 1285 1286 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size) 1287 { 1288 unsigned long nulls1, nulls2; 1289 1290 nulls1 = offsetof(struct sock, __sk_common.skc_node.next); 1291 nulls2 = offsetof(struct sock, __sk_common.skc_portaddr_node.next); 1292 if (nulls1 > nulls2) 1293 swap(nulls1, nulls2); 1294 1295 if (nulls1 != 0) 1296 memset((char *)sk, 0, nulls1); 1297 memset((char *)sk + nulls1 + sizeof(void *), 0, 1298 nulls2 - nulls1 - sizeof(void *)); 1299 memset((char *)sk + nulls2 + sizeof(void *), 0, 1300 size - nulls2 - sizeof(void *)); 1301 } 1302 EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls); 1303 1304 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority, 1305 int family) 1306 { 1307 struct sock *sk; 1308 struct kmem_cache *slab; 1309 1310 slab = prot->slab; 1311 if (slab != NULL) { 1312 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO); 1313 if (!sk) 1314 return sk; 1315 if (priority & __GFP_ZERO) { 1316 if (prot->clear_sk) 1317 prot->clear_sk(sk, prot->obj_size); 1318 else 1319 sk_prot_clear_nulls(sk, prot->obj_size); 1320 } 1321 } else 1322 sk = kmalloc(prot->obj_size, priority); 1323 1324 if (sk != NULL) { 1325 kmemcheck_annotate_bitfield(sk, flags); 1326 1327 if (security_sk_alloc(sk, family, priority)) 1328 goto out_free; 1329 1330 if (!try_module_get(prot->owner)) 1331 goto out_free_sec; 1332 sk_tx_queue_clear(sk); 1333 } 1334 1335 return sk; 1336 1337 out_free_sec: 1338 security_sk_free(sk); 1339 out_free: 1340 if (slab != NULL) 1341 kmem_cache_free(slab, sk); 1342 else 1343 kfree(sk); 1344 return NULL; 1345 } 1346 1347 static void sk_prot_free(struct proto *prot, struct sock *sk) 1348 { 1349 struct kmem_cache *slab; 1350 struct module *owner; 1351 1352 owner = prot->owner; 1353 slab = prot->slab; 1354 1355 security_sk_free(sk); 1356 if (slab != NULL) 1357 kmem_cache_free(slab, sk); 1358 else 1359 kfree(sk); 1360 module_put(owner); 1361 } 1362 1363 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO) 1364 void sock_update_netprioidx(struct sock *sk) 1365 { 1366 if (in_interrupt()) 1367 return; 1368 1369 sk->sk_cgrp_prioidx = task_netprioidx(current); 1370 } 1371 EXPORT_SYMBOL_GPL(sock_update_netprioidx); 1372 #endif 1373 1374 /** 1375 * sk_alloc - All socket objects are allocated here 1376 * @net: the applicable net namespace 1377 * @family: protocol family 1378 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 1379 * @prot: struct proto associated with this new sock instance 1380 */ 1381 struct sock *sk_alloc(struct net *net, int family, gfp_t priority, 1382 struct proto *prot) 1383 { 1384 struct sock *sk; 1385 1386 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family); 1387 if (sk) { 1388 sk->sk_family = family; 1389 /* 1390 * See comment in struct sock definition to understand 1391 * why we need sk_prot_creator -acme 1392 */ 1393 sk->sk_prot = sk->sk_prot_creator = prot; 1394 sock_lock_init(sk); 1395 sock_net_set(sk, get_net(net)); 1396 atomic_set(&sk->sk_wmem_alloc, 1); 1397 1398 sock_update_classid(sk); 1399 sock_update_netprioidx(sk); 1400 } 1401 1402 return sk; 1403 } 1404 EXPORT_SYMBOL(sk_alloc); 1405 1406 static void __sk_free(struct sock *sk) 1407 { 1408 struct sk_filter *filter; 1409 1410 if (sk->sk_destruct) 1411 sk->sk_destruct(sk); 1412 1413 filter = rcu_dereference_check(sk->sk_filter, 1414 atomic_read(&sk->sk_wmem_alloc) == 0); 1415 if (filter) { 1416 sk_filter_uncharge(sk, filter); 1417 RCU_INIT_POINTER(sk->sk_filter, NULL); 1418 } 1419 1420 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP); 1421 1422 if (atomic_read(&sk->sk_omem_alloc)) 1423 pr_debug("%s: optmem leakage (%d bytes) detected\n", 1424 __func__, atomic_read(&sk->sk_omem_alloc)); 1425 1426 if (sk->sk_peer_cred) 1427 put_cred(sk->sk_peer_cred); 1428 put_pid(sk->sk_peer_pid); 1429 put_net(sock_net(sk)); 1430 sk_prot_free(sk->sk_prot_creator, sk); 1431 } 1432 1433 void sk_free(struct sock *sk) 1434 { 1435 /* 1436 * We subtract one from sk_wmem_alloc and can know if 1437 * some packets are still in some tx queue. 1438 * If not null, sock_wfree() will call __sk_free(sk) later 1439 */ 1440 if (atomic_dec_and_test(&sk->sk_wmem_alloc)) 1441 __sk_free(sk); 1442 } 1443 EXPORT_SYMBOL(sk_free); 1444 1445 /* 1446 * Last sock_put should drop reference to sk->sk_net. It has already 1447 * been dropped in sk_change_net. Taking reference to stopping namespace 1448 * is not an option. 1449 * Take reference to a socket to remove it from hash _alive_ and after that 1450 * destroy it in the context of init_net. 1451 */ 1452 void sk_release_kernel(struct sock *sk) 1453 { 1454 if (sk == NULL || sk->sk_socket == NULL) 1455 return; 1456 1457 sock_hold(sk); 1458 sock_release(sk->sk_socket); 1459 release_net(sock_net(sk)); 1460 sock_net_set(sk, get_net(&init_net)); 1461 sock_put(sk); 1462 } 1463 EXPORT_SYMBOL(sk_release_kernel); 1464 1465 static void sk_update_clone(const struct sock *sk, struct sock *newsk) 1466 { 1467 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) 1468 sock_update_memcg(newsk); 1469 } 1470 1471 /** 1472 * sk_clone_lock - clone a socket, and lock its clone 1473 * @sk: the socket to clone 1474 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 1475 * 1476 * Caller must unlock socket even in error path (bh_unlock_sock(newsk)) 1477 */ 1478 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority) 1479 { 1480 struct sock *newsk; 1481 1482 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family); 1483 if (newsk != NULL) { 1484 struct sk_filter *filter; 1485 1486 sock_copy(newsk, sk); 1487 1488 /* SANITY */ 1489 get_net(sock_net(newsk)); 1490 sk_node_init(&newsk->sk_node); 1491 sock_lock_init(newsk); 1492 bh_lock_sock(newsk); 1493 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL; 1494 newsk->sk_backlog.len = 0; 1495 1496 atomic_set(&newsk->sk_rmem_alloc, 0); 1497 /* 1498 * sk_wmem_alloc set to one (see sk_free() and sock_wfree()) 1499 */ 1500 atomic_set(&newsk->sk_wmem_alloc, 1); 1501 atomic_set(&newsk->sk_omem_alloc, 0); 1502 skb_queue_head_init(&newsk->sk_receive_queue); 1503 skb_queue_head_init(&newsk->sk_write_queue); 1504 #ifdef CONFIG_NET_DMA 1505 skb_queue_head_init(&newsk->sk_async_wait_queue); 1506 #endif 1507 1508 spin_lock_init(&newsk->sk_dst_lock); 1509 rwlock_init(&newsk->sk_callback_lock); 1510 lockdep_set_class_and_name(&newsk->sk_callback_lock, 1511 af_callback_keys + newsk->sk_family, 1512 af_family_clock_key_strings[newsk->sk_family]); 1513 1514 newsk->sk_dst_cache = NULL; 1515 newsk->sk_wmem_queued = 0; 1516 newsk->sk_forward_alloc = 0; 1517 newsk->sk_send_head = NULL; 1518 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK; 1519 1520 sock_reset_flag(newsk, SOCK_DONE); 1521 skb_queue_head_init(&newsk->sk_error_queue); 1522 1523 filter = rcu_dereference_protected(newsk->sk_filter, 1); 1524 if (filter != NULL) 1525 sk_filter_charge(newsk, filter); 1526 1527 if (unlikely(xfrm_sk_clone_policy(newsk))) { 1528 /* It is still raw copy of parent, so invalidate 1529 * destructor and make plain sk_free() */ 1530 newsk->sk_destruct = NULL; 1531 bh_unlock_sock(newsk); 1532 sk_free(newsk); 1533 newsk = NULL; 1534 goto out; 1535 } 1536 1537 newsk->sk_err = 0; 1538 newsk->sk_priority = 0; 1539 /* 1540 * Before updating sk_refcnt, we must commit prior changes to memory 1541 * (Documentation/RCU/rculist_nulls.txt for details) 1542 */ 1543 smp_wmb(); 1544 atomic_set(&newsk->sk_refcnt, 2); 1545 1546 /* 1547 * Increment the counter in the same struct proto as the master 1548 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that 1549 * is the same as sk->sk_prot->socks, as this field was copied 1550 * with memcpy). 1551 * 1552 * This _changes_ the previous behaviour, where 1553 * tcp_create_openreq_child always was incrementing the 1554 * equivalent to tcp_prot->socks (inet_sock_nr), so this have 1555 * to be taken into account in all callers. -acme 1556 */ 1557 sk_refcnt_debug_inc(newsk); 1558 sk_set_socket(newsk, NULL); 1559 newsk->sk_wq = NULL; 1560 1561 sk_update_clone(sk, newsk); 1562 1563 if (newsk->sk_prot->sockets_allocated) 1564 sk_sockets_allocated_inc(newsk); 1565 1566 if (newsk->sk_flags & SK_FLAGS_TIMESTAMP) 1567 net_enable_timestamp(); 1568 } 1569 out: 1570 return newsk; 1571 } 1572 EXPORT_SYMBOL_GPL(sk_clone_lock); 1573 1574 void sk_setup_caps(struct sock *sk, struct dst_entry *dst) 1575 { 1576 __sk_dst_set(sk, dst); 1577 sk->sk_route_caps = dst->dev->features; 1578 if (sk->sk_route_caps & NETIF_F_GSO) 1579 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE; 1580 sk->sk_route_caps &= ~sk->sk_route_nocaps; 1581 if (sk_can_gso(sk)) { 1582 if (dst->header_len) { 1583 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 1584 } else { 1585 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM; 1586 sk->sk_gso_max_size = dst->dev->gso_max_size; 1587 sk->sk_gso_max_segs = dst->dev->gso_max_segs; 1588 } 1589 } 1590 } 1591 EXPORT_SYMBOL_GPL(sk_setup_caps); 1592 1593 /* 1594 * Simple resource managers for sockets. 1595 */ 1596 1597 1598 /* 1599 * Write buffer destructor automatically called from kfree_skb. 1600 */ 1601 void sock_wfree(struct sk_buff *skb) 1602 { 1603 struct sock *sk = skb->sk; 1604 unsigned int len = skb->truesize; 1605 1606 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) { 1607 /* 1608 * Keep a reference on sk_wmem_alloc, this will be released 1609 * after sk_write_space() call 1610 */ 1611 atomic_sub(len - 1, &sk->sk_wmem_alloc); 1612 sk->sk_write_space(sk); 1613 len = 1; 1614 } 1615 /* 1616 * if sk_wmem_alloc reaches 0, we must finish what sk_free() 1617 * could not do because of in-flight packets 1618 */ 1619 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc)) 1620 __sk_free(sk); 1621 } 1622 EXPORT_SYMBOL(sock_wfree); 1623 1624 void skb_orphan_partial(struct sk_buff *skb) 1625 { 1626 /* TCP stack sets skb->ooo_okay based on sk_wmem_alloc, 1627 * so we do not completely orphan skb, but transfert all 1628 * accounted bytes but one, to avoid unexpected reorders. 1629 */ 1630 if (skb->destructor == sock_wfree 1631 #ifdef CONFIG_INET 1632 || skb->destructor == tcp_wfree 1633 #endif 1634 ) { 1635 atomic_sub(skb->truesize - 1, &skb->sk->sk_wmem_alloc); 1636 skb->truesize = 1; 1637 } else { 1638 skb_orphan(skb); 1639 } 1640 } 1641 EXPORT_SYMBOL(skb_orphan_partial); 1642 1643 /* 1644 * Read buffer destructor automatically called from kfree_skb. 1645 */ 1646 void sock_rfree(struct sk_buff *skb) 1647 { 1648 struct sock *sk = skb->sk; 1649 unsigned int len = skb->truesize; 1650 1651 atomic_sub(len, &sk->sk_rmem_alloc); 1652 sk_mem_uncharge(sk, len); 1653 } 1654 EXPORT_SYMBOL(sock_rfree); 1655 1656 void sock_edemux(struct sk_buff *skb) 1657 { 1658 struct sock *sk = skb->sk; 1659 1660 #ifdef CONFIG_INET 1661 if (sk->sk_state == TCP_TIME_WAIT) 1662 inet_twsk_put(inet_twsk(sk)); 1663 else 1664 #endif 1665 sock_put(sk); 1666 } 1667 EXPORT_SYMBOL(sock_edemux); 1668 1669 kuid_t sock_i_uid(struct sock *sk) 1670 { 1671 kuid_t uid; 1672 1673 read_lock_bh(&sk->sk_callback_lock); 1674 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID; 1675 read_unlock_bh(&sk->sk_callback_lock); 1676 return uid; 1677 } 1678 EXPORT_SYMBOL(sock_i_uid); 1679 1680 unsigned long sock_i_ino(struct sock *sk) 1681 { 1682 unsigned long ino; 1683 1684 read_lock_bh(&sk->sk_callback_lock); 1685 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0; 1686 read_unlock_bh(&sk->sk_callback_lock); 1687 return ino; 1688 } 1689 EXPORT_SYMBOL(sock_i_ino); 1690 1691 /* 1692 * Allocate a skb from the socket's send buffer. 1693 */ 1694 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 1695 gfp_t priority) 1696 { 1697 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) { 1698 struct sk_buff *skb = alloc_skb(size, priority); 1699 if (skb) { 1700 skb_set_owner_w(skb, sk); 1701 return skb; 1702 } 1703 } 1704 return NULL; 1705 } 1706 EXPORT_SYMBOL(sock_wmalloc); 1707 1708 /* 1709 * Allocate a memory block from the socket's option memory buffer. 1710 */ 1711 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority) 1712 { 1713 if ((unsigned int)size <= sysctl_optmem_max && 1714 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) { 1715 void *mem; 1716 /* First do the add, to avoid the race if kmalloc 1717 * might sleep. 1718 */ 1719 atomic_add(size, &sk->sk_omem_alloc); 1720 mem = kmalloc(size, priority); 1721 if (mem) 1722 return mem; 1723 atomic_sub(size, &sk->sk_omem_alloc); 1724 } 1725 return NULL; 1726 } 1727 EXPORT_SYMBOL(sock_kmalloc); 1728 1729 /* 1730 * Free an option memory block. 1731 */ 1732 void sock_kfree_s(struct sock *sk, void *mem, int size) 1733 { 1734 kfree(mem); 1735 atomic_sub(size, &sk->sk_omem_alloc); 1736 } 1737 EXPORT_SYMBOL(sock_kfree_s); 1738 1739 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock. 1740 I think, these locks should be removed for datagram sockets. 1741 */ 1742 static long sock_wait_for_wmem(struct sock *sk, long timeo) 1743 { 1744 DEFINE_WAIT(wait); 1745 1746 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags); 1747 for (;;) { 1748 if (!timeo) 1749 break; 1750 if (signal_pending(current)) 1751 break; 1752 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 1753 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 1754 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) 1755 break; 1756 if (sk->sk_shutdown & SEND_SHUTDOWN) 1757 break; 1758 if (sk->sk_err) 1759 break; 1760 timeo = schedule_timeout(timeo); 1761 } 1762 finish_wait(sk_sleep(sk), &wait); 1763 return timeo; 1764 } 1765 1766 1767 /* 1768 * Generic send/receive buffer handlers 1769 */ 1770 1771 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 1772 unsigned long data_len, int noblock, 1773 int *errcode, int max_page_order) 1774 { 1775 struct sk_buff *skb = NULL; 1776 unsigned long chunk; 1777 gfp_t gfp_mask; 1778 long timeo; 1779 int err; 1780 int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT; 1781 struct page *page; 1782 int i; 1783 1784 err = -EMSGSIZE; 1785 if (npages > MAX_SKB_FRAGS) 1786 goto failure; 1787 1788 timeo = sock_sndtimeo(sk, noblock); 1789 while (!skb) { 1790 err = sock_error(sk); 1791 if (err != 0) 1792 goto failure; 1793 1794 err = -EPIPE; 1795 if (sk->sk_shutdown & SEND_SHUTDOWN) 1796 goto failure; 1797 1798 if (atomic_read(&sk->sk_wmem_alloc) >= sk->sk_sndbuf) { 1799 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags); 1800 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 1801 err = -EAGAIN; 1802 if (!timeo) 1803 goto failure; 1804 if (signal_pending(current)) 1805 goto interrupted; 1806 timeo = sock_wait_for_wmem(sk, timeo); 1807 continue; 1808 } 1809 1810 err = -ENOBUFS; 1811 gfp_mask = sk->sk_allocation; 1812 if (gfp_mask & __GFP_WAIT) 1813 gfp_mask |= __GFP_REPEAT; 1814 1815 skb = alloc_skb(header_len, gfp_mask); 1816 if (!skb) 1817 goto failure; 1818 1819 skb->truesize += data_len; 1820 1821 for (i = 0; npages > 0; i++) { 1822 int order = max_page_order; 1823 1824 while (order) { 1825 if (npages >= 1 << order) { 1826 page = alloc_pages(sk->sk_allocation | 1827 __GFP_COMP | 1828 __GFP_NOWARN | 1829 __GFP_NORETRY, 1830 order); 1831 if (page) 1832 goto fill_page; 1833 } 1834 order--; 1835 } 1836 page = alloc_page(sk->sk_allocation); 1837 if (!page) 1838 goto failure; 1839 fill_page: 1840 chunk = min_t(unsigned long, data_len, 1841 PAGE_SIZE << order); 1842 skb_fill_page_desc(skb, i, page, 0, chunk); 1843 data_len -= chunk; 1844 npages -= 1 << order; 1845 } 1846 } 1847 1848 skb_set_owner_w(skb, sk); 1849 return skb; 1850 1851 interrupted: 1852 err = sock_intr_errno(timeo); 1853 failure: 1854 kfree_skb(skb); 1855 *errcode = err; 1856 return NULL; 1857 } 1858 EXPORT_SYMBOL(sock_alloc_send_pskb); 1859 1860 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size, 1861 int noblock, int *errcode) 1862 { 1863 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0); 1864 } 1865 EXPORT_SYMBOL(sock_alloc_send_skb); 1866 1867 /* On 32bit arches, an skb frag is limited to 2^15 */ 1868 #define SKB_FRAG_PAGE_ORDER get_order(32768) 1869 1870 /** 1871 * skb_page_frag_refill - check that a page_frag contains enough room 1872 * @sz: minimum size of the fragment we want to get 1873 * @pfrag: pointer to page_frag 1874 * @prio: priority for memory allocation 1875 * 1876 * Note: While this allocator tries to use high order pages, there is 1877 * no guarantee that allocations succeed. Therefore, @sz MUST be 1878 * less or equal than PAGE_SIZE. 1879 */ 1880 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio) 1881 { 1882 int order; 1883 1884 if (pfrag->page) { 1885 if (atomic_read(&pfrag->page->_count) == 1) { 1886 pfrag->offset = 0; 1887 return true; 1888 } 1889 if (pfrag->offset + sz <= pfrag->size) 1890 return true; 1891 put_page(pfrag->page); 1892 } 1893 1894 order = SKB_FRAG_PAGE_ORDER; 1895 do { 1896 gfp_t gfp = prio; 1897 1898 if (order) 1899 gfp |= __GFP_COMP | __GFP_NOWARN | __GFP_NORETRY; 1900 pfrag->page = alloc_pages(gfp, order); 1901 if (likely(pfrag->page)) { 1902 pfrag->offset = 0; 1903 pfrag->size = PAGE_SIZE << order; 1904 return true; 1905 } 1906 } while (--order >= 0); 1907 1908 return false; 1909 } 1910 EXPORT_SYMBOL(skb_page_frag_refill); 1911 1912 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag) 1913 { 1914 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation))) 1915 return true; 1916 1917 sk_enter_memory_pressure(sk); 1918 sk_stream_moderate_sndbuf(sk); 1919 return false; 1920 } 1921 EXPORT_SYMBOL(sk_page_frag_refill); 1922 1923 static void __lock_sock(struct sock *sk) 1924 __releases(&sk->sk_lock.slock) 1925 __acquires(&sk->sk_lock.slock) 1926 { 1927 DEFINE_WAIT(wait); 1928 1929 for (;;) { 1930 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait, 1931 TASK_UNINTERRUPTIBLE); 1932 spin_unlock_bh(&sk->sk_lock.slock); 1933 schedule(); 1934 spin_lock_bh(&sk->sk_lock.slock); 1935 if (!sock_owned_by_user(sk)) 1936 break; 1937 } 1938 finish_wait(&sk->sk_lock.wq, &wait); 1939 } 1940 1941 static void __release_sock(struct sock *sk) 1942 __releases(&sk->sk_lock.slock) 1943 __acquires(&sk->sk_lock.slock) 1944 { 1945 struct sk_buff *skb = sk->sk_backlog.head; 1946 1947 do { 1948 sk->sk_backlog.head = sk->sk_backlog.tail = NULL; 1949 bh_unlock_sock(sk); 1950 1951 do { 1952 struct sk_buff *next = skb->next; 1953 1954 prefetch(next); 1955 WARN_ON_ONCE(skb_dst_is_noref(skb)); 1956 skb->next = NULL; 1957 sk_backlog_rcv(sk, skb); 1958 1959 /* 1960 * We are in process context here with softirqs 1961 * disabled, use cond_resched_softirq() to preempt. 1962 * This is safe to do because we've taken the backlog 1963 * queue private: 1964 */ 1965 cond_resched_softirq(); 1966 1967 skb = next; 1968 } while (skb != NULL); 1969 1970 bh_lock_sock(sk); 1971 } while ((skb = sk->sk_backlog.head) != NULL); 1972 1973 /* 1974 * Doing the zeroing here guarantee we can not loop forever 1975 * while a wild producer attempts to flood us. 1976 */ 1977 sk->sk_backlog.len = 0; 1978 } 1979 1980 /** 1981 * sk_wait_data - wait for data to arrive at sk_receive_queue 1982 * @sk: sock to wait on 1983 * @timeo: for how long 1984 * 1985 * Now socket state including sk->sk_err is changed only under lock, 1986 * hence we may omit checks after joining wait queue. 1987 * We check receive queue before schedule() only as optimization; 1988 * it is very likely that release_sock() added new data. 1989 */ 1990 int sk_wait_data(struct sock *sk, long *timeo) 1991 { 1992 int rc; 1993 DEFINE_WAIT(wait); 1994 1995 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 1996 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags); 1997 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue)); 1998 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags); 1999 finish_wait(sk_sleep(sk), &wait); 2000 return rc; 2001 } 2002 EXPORT_SYMBOL(sk_wait_data); 2003 2004 /** 2005 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated 2006 * @sk: socket 2007 * @size: memory size to allocate 2008 * @kind: allocation type 2009 * 2010 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means 2011 * rmem allocation. This function assumes that protocols which have 2012 * memory_pressure use sk_wmem_queued as write buffer accounting. 2013 */ 2014 int __sk_mem_schedule(struct sock *sk, int size, int kind) 2015 { 2016 struct proto *prot = sk->sk_prot; 2017 int amt = sk_mem_pages(size); 2018 long allocated; 2019 int parent_status = UNDER_LIMIT; 2020 2021 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM; 2022 2023 allocated = sk_memory_allocated_add(sk, amt, &parent_status); 2024 2025 /* Under limit. */ 2026 if (parent_status == UNDER_LIMIT && 2027 allocated <= sk_prot_mem_limits(sk, 0)) { 2028 sk_leave_memory_pressure(sk); 2029 return 1; 2030 } 2031 2032 /* Under pressure. (we or our parents) */ 2033 if ((parent_status > SOFT_LIMIT) || 2034 allocated > sk_prot_mem_limits(sk, 1)) 2035 sk_enter_memory_pressure(sk); 2036 2037 /* Over hard limit (we or our parents) */ 2038 if ((parent_status == OVER_LIMIT) || 2039 (allocated > sk_prot_mem_limits(sk, 2))) 2040 goto suppress_allocation; 2041 2042 /* guarantee minimum buffer size under pressure */ 2043 if (kind == SK_MEM_RECV) { 2044 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0]) 2045 return 1; 2046 2047 } else { /* SK_MEM_SEND */ 2048 if (sk->sk_type == SOCK_STREAM) { 2049 if (sk->sk_wmem_queued < prot->sysctl_wmem[0]) 2050 return 1; 2051 } else if (atomic_read(&sk->sk_wmem_alloc) < 2052 prot->sysctl_wmem[0]) 2053 return 1; 2054 } 2055 2056 if (sk_has_memory_pressure(sk)) { 2057 int alloc; 2058 2059 if (!sk_under_memory_pressure(sk)) 2060 return 1; 2061 alloc = sk_sockets_allocated_read_positive(sk); 2062 if (sk_prot_mem_limits(sk, 2) > alloc * 2063 sk_mem_pages(sk->sk_wmem_queued + 2064 atomic_read(&sk->sk_rmem_alloc) + 2065 sk->sk_forward_alloc)) 2066 return 1; 2067 } 2068 2069 suppress_allocation: 2070 2071 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) { 2072 sk_stream_moderate_sndbuf(sk); 2073 2074 /* Fail only if socket is _under_ its sndbuf. 2075 * In this case we cannot block, so that we have to fail. 2076 */ 2077 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf) 2078 return 1; 2079 } 2080 2081 trace_sock_exceed_buf_limit(sk, prot, allocated); 2082 2083 /* Alas. Undo changes. */ 2084 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM; 2085 2086 sk_memory_allocated_sub(sk, amt); 2087 2088 return 0; 2089 } 2090 EXPORT_SYMBOL(__sk_mem_schedule); 2091 2092 /** 2093 * __sk_reclaim - reclaim memory_allocated 2094 * @sk: socket 2095 */ 2096 void __sk_mem_reclaim(struct sock *sk) 2097 { 2098 sk_memory_allocated_sub(sk, 2099 sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT); 2100 sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1; 2101 2102 if (sk_under_memory_pressure(sk) && 2103 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0))) 2104 sk_leave_memory_pressure(sk); 2105 } 2106 EXPORT_SYMBOL(__sk_mem_reclaim); 2107 2108 2109 /* 2110 * Set of default routines for initialising struct proto_ops when 2111 * the protocol does not support a particular function. In certain 2112 * cases where it makes no sense for a protocol to have a "do nothing" 2113 * function, some default processing is provided. 2114 */ 2115 2116 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len) 2117 { 2118 return -EOPNOTSUPP; 2119 } 2120 EXPORT_SYMBOL(sock_no_bind); 2121 2122 int sock_no_connect(struct socket *sock, struct sockaddr *saddr, 2123 int len, int flags) 2124 { 2125 return -EOPNOTSUPP; 2126 } 2127 EXPORT_SYMBOL(sock_no_connect); 2128 2129 int sock_no_socketpair(struct socket *sock1, struct socket *sock2) 2130 { 2131 return -EOPNOTSUPP; 2132 } 2133 EXPORT_SYMBOL(sock_no_socketpair); 2134 2135 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags) 2136 { 2137 return -EOPNOTSUPP; 2138 } 2139 EXPORT_SYMBOL(sock_no_accept); 2140 2141 int sock_no_getname(struct socket *sock, struct sockaddr *saddr, 2142 int *len, int peer) 2143 { 2144 return -EOPNOTSUPP; 2145 } 2146 EXPORT_SYMBOL(sock_no_getname); 2147 2148 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt) 2149 { 2150 return 0; 2151 } 2152 EXPORT_SYMBOL(sock_no_poll); 2153 2154 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 2155 { 2156 return -EOPNOTSUPP; 2157 } 2158 EXPORT_SYMBOL(sock_no_ioctl); 2159 2160 int sock_no_listen(struct socket *sock, int backlog) 2161 { 2162 return -EOPNOTSUPP; 2163 } 2164 EXPORT_SYMBOL(sock_no_listen); 2165 2166 int sock_no_shutdown(struct socket *sock, int how) 2167 { 2168 return -EOPNOTSUPP; 2169 } 2170 EXPORT_SYMBOL(sock_no_shutdown); 2171 2172 int sock_no_setsockopt(struct socket *sock, int level, int optname, 2173 char __user *optval, unsigned int optlen) 2174 { 2175 return -EOPNOTSUPP; 2176 } 2177 EXPORT_SYMBOL(sock_no_setsockopt); 2178 2179 int sock_no_getsockopt(struct socket *sock, int level, int optname, 2180 char __user *optval, int __user *optlen) 2181 { 2182 return -EOPNOTSUPP; 2183 } 2184 EXPORT_SYMBOL(sock_no_getsockopt); 2185 2186 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m, 2187 size_t len) 2188 { 2189 return -EOPNOTSUPP; 2190 } 2191 EXPORT_SYMBOL(sock_no_sendmsg); 2192 2193 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m, 2194 size_t len, int flags) 2195 { 2196 return -EOPNOTSUPP; 2197 } 2198 EXPORT_SYMBOL(sock_no_recvmsg); 2199 2200 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma) 2201 { 2202 /* Mirror missing mmap method error code */ 2203 return -ENODEV; 2204 } 2205 EXPORT_SYMBOL(sock_no_mmap); 2206 2207 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags) 2208 { 2209 ssize_t res; 2210 struct msghdr msg = {.msg_flags = flags}; 2211 struct kvec iov; 2212 char *kaddr = kmap(page); 2213 iov.iov_base = kaddr + offset; 2214 iov.iov_len = size; 2215 res = kernel_sendmsg(sock, &msg, &iov, 1, size); 2216 kunmap(page); 2217 return res; 2218 } 2219 EXPORT_SYMBOL(sock_no_sendpage); 2220 2221 /* 2222 * Default Socket Callbacks 2223 */ 2224 2225 static void sock_def_wakeup(struct sock *sk) 2226 { 2227 struct socket_wq *wq; 2228 2229 rcu_read_lock(); 2230 wq = rcu_dereference(sk->sk_wq); 2231 if (wq_has_sleeper(wq)) 2232 wake_up_interruptible_all(&wq->wait); 2233 rcu_read_unlock(); 2234 } 2235 2236 static void sock_def_error_report(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_poll(&wq->wait, POLLERR); 2244 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR); 2245 rcu_read_unlock(); 2246 } 2247 2248 static void sock_def_readable(struct sock *sk) 2249 { 2250 struct socket_wq *wq; 2251 2252 rcu_read_lock(); 2253 wq = rcu_dereference(sk->sk_wq); 2254 if (wq_has_sleeper(wq)) 2255 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI | 2256 POLLRDNORM | POLLRDBAND); 2257 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN); 2258 rcu_read_unlock(); 2259 } 2260 2261 static void sock_def_write_space(struct sock *sk) 2262 { 2263 struct socket_wq *wq; 2264 2265 rcu_read_lock(); 2266 2267 /* Do not wake up a writer until he can make "significant" 2268 * progress. --DaveM 2269 */ 2270 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) { 2271 wq = rcu_dereference(sk->sk_wq); 2272 if (wq_has_sleeper(wq)) 2273 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT | 2274 POLLWRNORM | POLLWRBAND); 2275 2276 /* Should agree with poll, otherwise some programs break */ 2277 if (sock_writeable(sk)) 2278 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT); 2279 } 2280 2281 rcu_read_unlock(); 2282 } 2283 2284 static void sock_def_destruct(struct sock *sk) 2285 { 2286 kfree(sk->sk_protinfo); 2287 } 2288 2289 void sk_send_sigurg(struct sock *sk) 2290 { 2291 if (sk->sk_socket && sk->sk_socket->file) 2292 if (send_sigurg(&sk->sk_socket->file->f_owner)) 2293 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI); 2294 } 2295 EXPORT_SYMBOL(sk_send_sigurg); 2296 2297 void sk_reset_timer(struct sock *sk, struct timer_list* timer, 2298 unsigned long expires) 2299 { 2300 if (!mod_timer(timer, expires)) 2301 sock_hold(sk); 2302 } 2303 EXPORT_SYMBOL(sk_reset_timer); 2304 2305 void sk_stop_timer(struct sock *sk, struct timer_list* timer) 2306 { 2307 if (del_timer(timer)) 2308 __sock_put(sk); 2309 } 2310 EXPORT_SYMBOL(sk_stop_timer); 2311 2312 void sock_init_data(struct socket *sock, struct sock *sk) 2313 { 2314 skb_queue_head_init(&sk->sk_receive_queue); 2315 skb_queue_head_init(&sk->sk_write_queue); 2316 skb_queue_head_init(&sk->sk_error_queue); 2317 #ifdef CONFIG_NET_DMA 2318 skb_queue_head_init(&sk->sk_async_wait_queue); 2319 #endif 2320 2321 sk->sk_send_head = NULL; 2322 2323 init_timer(&sk->sk_timer); 2324 2325 sk->sk_allocation = GFP_KERNEL; 2326 sk->sk_rcvbuf = sysctl_rmem_default; 2327 sk->sk_sndbuf = sysctl_wmem_default; 2328 sk->sk_state = TCP_CLOSE; 2329 sk_set_socket(sk, sock); 2330 2331 sock_set_flag(sk, SOCK_ZAPPED); 2332 2333 if (sock) { 2334 sk->sk_type = sock->type; 2335 sk->sk_wq = sock->wq; 2336 sock->sk = sk; 2337 } else 2338 sk->sk_wq = NULL; 2339 2340 spin_lock_init(&sk->sk_dst_lock); 2341 rwlock_init(&sk->sk_callback_lock); 2342 lockdep_set_class_and_name(&sk->sk_callback_lock, 2343 af_callback_keys + sk->sk_family, 2344 af_family_clock_key_strings[sk->sk_family]); 2345 2346 sk->sk_state_change = sock_def_wakeup; 2347 sk->sk_data_ready = sock_def_readable; 2348 sk->sk_write_space = sock_def_write_space; 2349 sk->sk_error_report = sock_def_error_report; 2350 sk->sk_destruct = sock_def_destruct; 2351 2352 sk->sk_frag.page = NULL; 2353 sk->sk_frag.offset = 0; 2354 sk->sk_peek_off = -1; 2355 2356 sk->sk_peer_pid = NULL; 2357 sk->sk_peer_cred = NULL; 2358 sk->sk_write_pending = 0; 2359 sk->sk_rcvlowat = 1; 2360 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; 2361 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT; 2362 2363 sk->sk_stamp = ktime_set(-1L, 0); 2364 2365 #ifdef CONFIG_NET_RX_BUSY_POLL 2366 sk->sk_napi_id = 0; 2367 sk->sk_ll_usec = sysctl_net_busy_read; 2368 #endif 2369 2370 sk->sk_max_pacing_rate = ~0U; 2371 sk->sk_pacing_rate = ~0U; 2372 /* 2373 * Before updating sk_refcnt, we must commit prior changes to memory 2374 * (Documentation/RCU/rculist_nulls.txt for details) 2375 */ 2376 smp_wmb(); 2377 atomic_set(&sk->sk_refcnt, 1); 2378 atomic_set(&sk->sk_drops, 0); 2379 } 2380 EXPORT_SYMBOL(sock_init_data); 2381 2382 void lock_sock_nested(struct sock *sk, int subclass) 2383 { 2384 might_sleep(); 2385 spin_lock_bh(&sk->sk_lock.slock); 2386 if (sk->sk_lock.owned) 2387 __lock_sock(sk); 2388 sk->sk_lock.owned = 1; 2389 spin_unlock(&sk->sk_lock.slock); 2390 /* 2391 * The sk_lock has mutex_lock() semantics here: 2392 */ 2393 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_); 2394 local_bh_enable(); 2395 } 2396 EXPORT_SYMBOL(lock_sock_nested); 2397 2398 void release_sock(struct sock *sk) 2399 { 2400 /* 2401 * The sk_lock has mutex_unlock() semantics: 2402 */ 2403 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_); 2404 2405 spin_lock_bh(&sk->sk_lock.slock); 2406 if (sk->sk_backlog.tail) 2407 __release_sock(sk); 2408 2409 /* Warning : release_cb() might need to release sk ownership, 2410 * ie call sock_release_ownership(sk) before us. 2411 */ 2412 if (sk->sk_prot->release_cb) 2413 sk->sk_prot->release_cb(sk); 2414 2415 sock_release_ownership(sk); 2416 if (waitqueue_active(&sk->sk_lock.wq)) 2417 wake_up(&sk->sk_lock.wq); 2418 spin_unlock_bh(&sk->sk_lock.slock); 2419 } 2420 EXPORT_SYMBOL(release_sock); 2421 2422 /** 2423 * lock_sock_fast - fast version of lock_sock 2424 * @sk: socket 2425 * 2426 * This version should be used for very small section, where process wont block 2427 * return false if fast path is taken 2428 * sk_lock.slock locked, owned = 0, BH disabled 2429 * return true if slow path is taken 2430 * sk_lock.slock unlocked, owned = 1, BH enabled 2431 */ 2432 bool lock_sock_fast(struct sock *sk) 2433 { 2434 might_sleep(); 2435 spin_lock_bh(&sk->sk_lock.slock); 2436 2437 if (!sk->sk_lock.owned) 2438 /* 2439 * Note : We must disable BH 2440 */ 2441 return false; 2442 2443 __lock_sock(sk); 2444 sk->sk_lock.owned = 1; 2445 spin_unlock(&sk->sk_lock.slock); 2446 /* 2447 * The sk_lock has mutex_lock() semantics here: 2448 */ 2449 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_); 2450 local_bh_enable(); 2451 return true; 2452 } 2453 EXPORT_SYMBOL(lock_sock_fast); 2454 2455 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp) 2456 { 2457 struct timeval tv; 2458 if (!sock_flag(sk, SOCK_TIMESTAMP)) 2459 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 2460 tv = ktime_to_timeval(sk->sk_stamp); 2461 if (tv.tv_sec == -1) 2462 return -ENOENT; 2463 if (tv.tv_sec == 0) { 2464 sk->sk_stamp = ktime_get_real(); 2465 tv = ktime_to_timeval(sk->sk_stamp); 2466 } 2467 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0; 2468 } 2469 EXPORT_SYMBOL(sock_get_timestamp); 2470 2471 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp) 2472 { 2473 struct timespec ts; 2474 if (!sock_flag(sk, SOCK_TIMESTAMP)) 2475 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 2476 ts = ktime_to_timespec(sk->sk_stamp); 2477 if (ts.tv_sec == -1) 2478 return -ENOENT; 2479 if (ts.tv_sec == 0) { 2480 sk->sk_stamp = ktime_get_real(); 2481 ts = ktime_to_timespec(sk->sk_stamp); 2482 } 2483 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0; 2484 } 2485 EXPORT_SYMBOL(sock_get_timestampns); 2486 2487 void sock_enable_timestamp(struct sock *sk, int flag) 2488 { 2489 if (!sock_flag(sk, flag)) { 2490 unsigned long previous_flags = sk->sk_flags; 2491 2492 sock_set_flag(sk, flag); 2493 /* 2494 * we just set one of the two flags which require net 2495 * time stamping, but time stamping might have been on 2496 * already because of the other one 2497 */ 2498 if (!(previous_flags & SK_FLAGS_TIMESTAMP)) 2499 net_enable_timestamp(); 2500 } 2501 } 2502 2503 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, 2504 int level, int type) 2505 { 2506 struct sock_exterr_skb *serr; 2507 struct sk_buff *skb, *skb2; 2508 int copied, err; 2509 2510 err = -EAGAIN; 2511 skb = skb_dequeue(&sk->sk_error_queue); 2512 if (skb == NULL) 2513 goto out; 2514 2515 copied = skb->len; 2516 if (copied > len) { 2517 msg->msg_flags |= MSG_TRUNC; 2518 copied = len; 2519 } 2520 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied); 2521 if (err) 2522 goto out_free_skb; 2523 2524 sock_recv_timestamp(msg, sk, skb); 2525 2526 serr = SKB_EXT_ERR(skb); 2527 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee); 2528 2529 msg->msg_flags |= MSG_ERRQUEUE; 2530 err = copied; 2531 2532 /* Reset and regenerate socket error */ 2533 spin_lock_bh(&sk->sk_error_queue.lock); 2534 sk->sk_err = 0; 2535 if ((skb2 = skb_peek(&sk->sk_error_queue)) != NULL) { 2536 sk->sk_err = SKB_EXT_ERR(skb2)->ee.ee_errno; 2537 spin_unlock_bh(&sk->sk_error_queue.lock); 2538 sk->sk_error_report(sk); 2539 } else 2540 spin_unlock_bh(&sk->sk_error_queue.lock); 2541 2542 out_free_skb: 2543 kfree_skb(skb); 2544 out: 2545 return err; 2546 } 2547 EXPORT_SYMBOL(sock_recv_errqueue); 2548 2549 /* 2550 * Get a socket option on an socket. 2551 * 2552 * FIX: POSIX 1003.1g is very ambiguous here. It states that 2553 * asynchronous errors should be reported by getsockopt. We assume 2554 * this means if you specify SO_ERROR (otherwise whats the point of it). 2555 */ 2556 int sock_common_getsockopt(struct socket *sock, int level, int optname, 2557 char __user *optval, int __user *optlen) 2558 { 2559 struct sock *sk = sock->sk; 2560 2561 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen); 2562 } 2563 EXPORT_SYMBOL(sock_common_getsockopt); 2564 2565 #ifdef CONFIG_COMPAT 2566 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname, 2567 char __user *optval, int __user *optlen) 2568 { 2569 struct sock *sk = sock->sk; 2570 2571 if (sk->sk_prot->compat_getsockopt != NULL) 2572 return sk->sk_prot->compat_getsockopt(sk, level, optname, 2573 optval, optlen); 2574 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen); 2575 } 2576 EXPORT_SYMBOL(compat_sock_common_getsockopt); 2577 #endif 2578 2579 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock, 2580 struct msghdr *msg, size_t size, int flags) 2581 { 2582 struct sock *sk = sock->sk; 2583 int addr_len = 0; 2584 int err; 2585 2586 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT, 2587 flags & ~MSG_DONTWAIT, &addr_len); 2588 if (err >= 0) 2589 msg->msg_namelen = addr_len; 2590 return err; 2591 } 2592 EXPORT_SYMBOL(sock_common_recvmsg); 2593 2594 /* 2595 * Set socket options on an inet socket. 2596 */ 2597 int sock_common_setsockopt(struct socket *sock, int level, int optname, 2598 char __user *optval, unsigned int optlen) 2599 { 2600 struct sock *sk = sock->sk; 2601 2602 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen); 2603 } 2604 EXPORT_SYMBOL(sock_common_setsockopt); 2605 2606 #ifdef CONFIG_COMPAT 2607 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname, 2608 char __user *optval, unsigned int optlen) 2609 { 2610 struct sock *sk = sock->sk; 2611 2612 if (sk->sk_prot->compat_setsockopt != NULL) 2613 return sk->sk_prot->compat_setsockopt(sk, level, optname, 2614 optval, optlen); 2615 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen); 2616 } 2617 EXPORT_SYMBOL(compat_sock_common_setsockopt); 2618 #endif 2619 2620 void sk_common_release(struct sock *sk) 2621 { 2622 if (sk->sk_prot->destroy) 2623 sk->sk_prot->destroy(sk); 2624 2625 /* 2626 * Observation: when sock_common_release is called, processes have 2627 * no access to socket. But net still has. 2628 * Step one, detach it from networking: 2629 * 2630 * A. Remove from hash tables. 2631 */ 2632 2633 sk->sk_prot->unhash(sk); 2634 2635 /* 2636 * In this point socket cannot receive new packets, but it is possible 2637 * that some packets are in flight because some CPU runs receiver and 2638 * did hash table lookup before we unhashed socket. They will achieve 2639 * receive queue and will be purged by socket destructor. 2640 * 2641 * Also we still have packets pending on receive queue and probably, 2642 * our own packets waiting in device queues. sock_destroy will drain 2643 * receive queue, but transmitted packets will delay socket destruction 2644 * until the last reference will be released. 2645 */ 2646 2647 sock_orphan(sk); 2648 2649 xfrm_sk_free_policy(sk); 2650 2651 sk_refcnt_debug_release(sk); 2652 2653 if (sk->sk_frag.page) { 2654 put_page(sk->sk_frag.page); 2655 sk->sk_frag.page = NULL; 2656 } 2657 2658 sock_put(sk); 2659 } 2660 EXPORT_SYMBOL(sk_common_release); 2661 2662 #ifdef CONFIG_PROC_FS 2663 #define PROTO_INUSE_NR 64 /* should be enough for the first time */ 2664 struct prot_inuse { 2665 int val[PROTO_INUSE_NR]; 2666 }; 2667 2668 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR); 2669 2670 #ifdef CONFIG_NET_NS 2671 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val) 2672 { 2673 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val); 2674 } 2675 EXPORT_SYMBOL_GPL(sock_prot_inuse_add); 2676 2677 int sock_prot_inuse_get(struct net *net, struct proto *prot) 2678 { 2679 int cpu, idx = prot->inuse_idx; 2680 int res = 0; 2681 2682 for_each_possible_cpu(cpu) 2683 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx]; 2684 2685 return res >= 0 ? res : 0; 2686 } 2687 EXPORT_SYMBOL_GPL(sock_prot_inuse_get); 2688 2689 static int __net_init sock_inuse_init_net(struct net *net) 2690 { 2691 net->core.inuse = alloc_percpu(struct prot_inuse); 2692 return net->core.inuse ? 0 : -ENOMEM; 2693 } 2694 2695 static void __net_exit sock_inuse_exit_net(struct net *net) 2696 { 2697 free_percpu(net->core.inuse); 2698 } 2699 2700 static struct pernet_operations net_inuse_ops = { 2701 .init = sock_inuse_init_net, 2702 .exit = sock_inuse_exit_net, 2703 }; 2704 2705 static __init int net_inuse_init(void) 2706 { 2707 if (register_pernet_subsys(&net_inuse_ops)) 2708 panic("Cannot initialize net inuse counters"); 2709 2710 return 0; 2711 } 2712 2713 core_initcall(net_inuse_init); 2714 #else 2715 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse); 2716 2717 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val) 2718 { 2719 __this_cpu_add(prot_inuse.val[prot->inuse_idx], val); 2720 } 2721 EXPORT_SYMBOL_GPL(sock_prot_inuse_add); 2722 2723 int sock_prot_inuse_get(struct net *net, struct proto *prot) 2724 { 2725 int cpu, idx = prot->inuse_idx; 2726 int res = 0; 2727 2728 for_each_possible_cpu(cpu) 2729 res += per_cpu(prot_inuse, cpu).val[idx]; 2730 2731 return res >= 0 ? res : 0; 2732 } 2733 EXPORT_SYMBOL_GPL(sock_prot_inuse_get); 2734 #endif 2735 2736 static void assign_proto_idx(struct proto *prot) 2737 { 2738 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR); 2739 2740 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) { 2741 pr_err("PROTO_INUSE_NR exhausted\n"); 2742 return; 2743 } 2744 2745 set_bit(prot->inuse_idx, proto_inuse_idx); 2746 } 2747 2748 static void release_proto_idx(struct proto *prot) 2749 { 2750 if (prot->inuse_idx != PROTO_INUSE_NR - 1) 2751 clear_bit(prot->inuse_idx, proto_inuse_idx); 2752 } 2753 #else 2754 static inline void assign_proto_idx(struct proto *prot) 2755 { 2756 } 2757 2758 static inline void release_proto_idx(struct proto *prot) 2759 { 2760 } 2761 #endif 2762 2763 int proto_register(struct proto *prot, int alloc_slab) 2764 { 2765 if (alloc_slab) { 2766 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0, 2767 SLAB_HWCACHE_ALIGN | prot->slab_flags, 2768 NULL); 2769 2770 if (prot->slab == NULL) { 2771 pr_crit("%s: Can't create sock SLAB cache!\n", 2772 prot->name); 2773 goto out; 2774 } 2775 2776 if (prot->rsk_prot != NULL) { 2777 prot->rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", prot->name); 2778 if (prot->rsk_prot->slab_name == NULL) 2779 goto out_free_sock_slab; 2780 2781 prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name, 2782 prot->rsk_prot->obj_size, 0, 2783 SLAB_HWCACHE_ALIGN, NULL); 2784 2785 if (prot->rsk_prot->slab == NULL) { 2786 pr_crit("%s: Can't create request sock SLAB cache!\n", 2787 prot->name); 2788 goto out_free_request_sock_slab_name; 2789 } 2790 } 2791 2792 if (prot->twsk_prot != NULL) { 2793 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name); 2794 2795 if (prot->twsk_prot->twsk_slab_name == NULL) 2796 goto out_free_request_sock_slab; 2797 2798 prot->twsk_prot->twsk_slab = 2799 kmem_cache_create(prot->twsk_prot->twsk_slab_name, 2800 prot->twsk_prot->twsk_obj_size, 2801 0, 2802 SLAB_HWCACHE_ALIGN | 2803 prot->slab_flags, 2804 NULL); 2805 if (prot->twsk_prot->twsk_slab == NULL) 2806 goto out_free_timewait_sock_slab_name; 2807 } 2808 } 2809 2810 mutex_lock(&proto_list_mutex); 2811 list_add(&prot->node, &proto_list); 2812 assign_proto_idx(prot); 2813 mutex_unlock(&proto_list_mutex); 2814 return 0; 2815 2816 out_free_timewait_sock_slab_name: 2817 kfree(prot->twsk_prot->twsk_slab_name); 2818 out_free_request_sock_slab: 2819 if (prot->rsk_prot && prot->rsk_prot->slab) { 2820 kmem_cache_destroy(prot->rsk_prot->slab); 2821 prot->rsk_prot->slab = NULL; 2822 } 2823 out_free_request_sock_slab_name: 2824 if (prot->rsk_prot) 2825 kfree(prot->rsk_prot->slab_name); 2826 out_free_sock_slab: 2827 kmem_cache_destroy(prot->slab); 2828 prot->slab = NULL; 2829 out: 2830 return -ENOBUFS; 2831 } 2832 EXPORT_SYMBOL(proto_register); 2833 2834 void proto_unregister(struct proto *prot) 2835 { 2836 mutex_lock(&proto_list_mutex); 2837 release_proto_idx(prot); 2838 list_del(&prot->node); 2839 mutex_unlock(&proto_list_mutex); 2840 2841 if (prot->slab != NULL) { 2842 kmem_cache_destroy(prot->slab); 2843 prot->slab = NULL; 2844 } 2845 2846 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) { 2847 kmem_cache_destroy(prot->rsk_prot->slab); 2848 kfree(prot->rsk_prot->slab_name); 2849 prot->rsk_prot->slab = NULL; 2850 } 2851 2852 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) { 2853 kmem_cache_destroy(prot->twsk_prot->twsk_slab); 2854 kfree(prot->twsk_prot->twsk_slab_name); 2855 prot->twsk_prot->twsk_slab = NULL; 2856 } 2857 } 2858 EXPORT_SYMBOL(proto_unregister); 2859 2860 #ifdef CONFIG_PROC_FS 2861 static void *proto_seq_start(struct seq_file *seq, loff_t *pos) 2862 __acquires(proto_list_mutex) 2863 { 2864 mutex_lock(&proto_list_mutex); 2865 return seq_list_start_head(&proto_list, *pos); 2866 } 2867 2868 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos) 2869 { 2870 return seq_list_next(v, &proto_list, pos); 2871 } 2872 2873 static void proto_seq_stop(struct seq_file *seq, void *v) 2874 __releases(proto_list_mutex) 2875 { 2876 mutex_unlock(&proto_list_mutex); 2877 } 2878 2879 static char proto_method_implemented(const void *method) 2880 { 2881 return method == NULL ? 'n' : 'y'; 2882 } 2883 static long sock_prot_memory_allocated(struct proto *proto) 2884 { 2885 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L; 2886 } 2887 2888 static char *sock_prot_memory_pressure(struct proto *proto) 2889 { 2890 return proto->memory_pressure != NULL ? 2891 proto_memory_pressure(proto) ? "yes" : "no" : "NI"; 2892 } 2893 2894 static void proto_seq_printf(struct seq_file *seq, struct proto *proto) 2895 { 2896 2897 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s " 2898 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n", 2899 proto->name, 2900 proto->obj_size, 2901 sock_prot_inuse_get(seq_file_net(seq), proto), 2902 sock_prot_memory_allocated(proto), 2903 sock_prot_memory_pressure(proto), 2904 proto->max_header, 2905 proto->slab == NULL ? "no" : "yes", 2906 module_name(proto->owner), 2907 proto_method_implemented(proto->close), 2908 proto_method_implemented(proto->connect), 2909 proto_method_implemented(proto->disconnect), 2910 proto_method_implemented(proto->accept), 2911 proto_method_implemented(proto->ioctl), 2912 proto_method_implemented(proto->init), 2913 proto_method_implemented(proto->destroy), 2914 proto_method_implemented(proto->shutdown), 2915 proto_method_implemented(proto->setsockopt), 2916 proto_method_implemented(proto->getsockopt), 2917 proto_method_implemented(proto->sendmsg), 2918 proto_method_implemented(proto->recvmsg), 2919 proto_method_implemented(proto->sendpage), 2920 proto_method_implemented(proto->bind), 2921 proto_method_implemented(proto->backlog_rcv), 2922 proto_method_implemented(proto->hash), 2923 proto_method_implemented(proto->unhash), 2924 proto_method_implemented(proto->get_port), 2925 proto_method_implemented(proto->enter_memory_pressure)); 2926 } 2927 2928 static int proto_seq_show(struct seq_file *seq, void *v) 2929 { 2930 if (v == &proto_list) 2931 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s", 2932 "protocol", 2933 "size", 2934 "sockets", 2935 "memory", 2936 "press", 2937 "maxhdr", 2938 "slab", 2939 "module", 2940 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n"); 2941 else 2942 proto_seq_printf(seq, list_entry(v, struct proto, node)); 2943 return 0; 2944 } 2945 2946 static const struct seq_operations proto_seq_ops = { 2947 .start = proto_seq_start, 2948 .next = proto_seq_next, 2949 .stop = proto_seq_stop, 2950 .show = proto_seq_show, 2951 }; 2952 2953 static int proto_seq_open(struct inode *inode, struct file *file) 2954 { 2955 return seq_open_net(inode, file, &proto_seq_ops, 2956 sizeof(struct seq_net_private)); 2957 } 2958 2959 static const struct file_operations proto_seq_fops = { 2960 .owner = THIS_MODULE, 2961 .open = proto_seq_open, 2962 .read = seq_read, 2963 .llseek = seq_lseek, 2964 .release = seq_release_net, 2965 }; 2966 2967 static __net_init int proto_init_net(struct net *net) 2968 { 2969 if (!proc_create("protocols", S_IRUGO, net->proc_net, &proto_seq_fops)) 2970 return -ENOMEM; 2971 2972 return 0; 2973 } 2974 2975 static __net_exit void proto_exit_net(struct net *net) 2976 { 2977 remove_proc_entry("protocols", net->proc_net); 2978 } 2979 2980 2981 static __net_initdata struct pernet_operations proto_net_ops = { 2982 .init = proto_init_net, 2983 .exit = proto_exit_net, 2984 }; 2985 2986 static int __init proto_init(void) 2987 { 2988 return register_pernet_subsys(&proto_net_ops); 2989 } 2990 2991 subsys_initcall(proto_init); 2992 2993 #endif /* PROC_FS */ 2994