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 <linux/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 * Each address family might have different locking rules, so we have 200 * one slock key per address family and separate keys for internal and 201 * userspace sockets. 202 */ 203 static struct lock_class_key af_family_keys[AF_MAX]; 204 static struct lock_class_key af_family_kern_keys[AF_MAX]; 205 static struct lock_class_key af_family_slock_keys[AF_MAX]; 206 static struct lock_class_key af_family_kern_slock_keys[AF_MAX]; 207 208 /* 209 * Make lock validator output more readable. (we pre-construct these 210 * strings build-time, so that runtime initialization of socket 211 * locks is fast): 212 */ 213 214 #define _sock_locks(x) \ 215 x "AF_UNSPEC", x "AF_UNIX" , x "AF_INET" , \ 216 x "AF_AX25" , x "AF_IPX" , x "AF_APPLETALK", \ 217 x "AF_NETROM", x "AF_BRIDGE" , x "AF_ATMPVC" , \ 218 x "AF_X25" , x "AF_INET6" , x "AF_ROSE" , \ 219 x "AF_DECnet", x "AF_NETBEUI" , x "AF_SECURITY" , \ 220 x "AF_KEY" , x "AF_NETLINK" , x "AF_PACKET" , \ 221 x "AF_ASH" , x "AF_ECONET" , x "AF_ATMSVC" , \ 222 x "AF_RDS" , x "AF_SNA" , x "AF_IRDA" , \ 223 x "AF_PPPOX" , x "AF_WANPIPE" , x "AF_LLC" , \ 224 x "27" , x "28" , x "AF_CAN" , \ 225 x "AF_TIPC" , x "AF_BLUETOOTH", x "IUCV" , \ 226 x "AF_RXRPC" , x "AF_ISDN" , x "AF_PHONET" , \ 227 x "AF_IEEE802154", x "AF_CAIF" , x "AF_ALG" , \ 228 x "AF_NFC" , x "AF_VSOCK" , x "AF_KCM" , \ 229 x "AF_QIPCRTR", x "AF_SMC" , x "AF_MAX" 230 231 static const char *const af_family_key_strings[AF_MAX+1] = { 232 _sock_locks("sk_lock-") 233 }; 234 static const char *const af_family_slock_key_strings[AF_MAX+1] = { 235 _sock_locks("slock-") 236 }; 237 static const char *const af_family_clock_key_strings[AF_MAX+1] = { 238 _sock_locks("clock-") 239 }; 240 241 static const char *const af_family_kern_key_strings[AF_MAX+1] = { 242 _sock_locks("k-sk_lock-") 243 }; 244 static const char *const af_family_kern_slock_key_strings[AF_MAX+1] = { 245 _sock_locks("k-slock-") 246 }; 247 static const char *const af_family_kern_clock_key_strings[AF_MAX+1] = { 248 _sock_locks("k-clock-") 249 }; 250 static const char *const af_family_rlock_key_strings[AF_MAX+1] = { 251 "rlock-AF_UNSPEC", "rlock-AF_UNIX" , "rlock-AF_INET" , 252 "rlock-AF_AX25" , "rlock-AF_IPX" , "rlock-AF_APPLETALK", 253 "rlock-AF_NETROM", "rlock-AF_BRIDGE" , "rlock-AF_ATMPVC" , 254 "rlock-AF_X25" , "rlock-AF_INET6" , "rlock-AF_ROSE" , 255 "rlock-AF_DECnet", "rlock-AF_NETBEUI" , "rlock-AF_SECURITY" , 256 "rlock-AF_KEY" , "rlock-AF_NETLINK" , "rlock-AF_PACKET" , 257 "rlock-AF_ASH" , "rlock-AF_ECONET" , "rlock-AF_ATMSVC" , 258 "rlock-AF_RDS" , "rlock-AF_SNA" , "rlock-AF_IRDA" , 259 "rlock-AF_PPPOX" , "rlock-AF_WANPIPE" , "rlock-AF_LLC" , 260 "rlock-27" , "rlock-28" , "rlock-AF_CAN" , 261 "rlock-AF_TIPC" , "rlock-AF_BLUETOOTH", "rlock-AF_IUCV" , 262 "rlock-AF_RXRPC" , "rlock-AF_ISDN" , "rlock-AF_PHONET" , 263 "rlock-AF_IEEE802154", "rlock-AF_CAIF" , "rlock-AF_ALG" , 264 "rlock-AF_NFC" , "rlock-AF_VSOCK" , "rlock-AF_KCM" , 265 "rlock-AF_QIPCRTR", "rlock-AF_SMC" , "rlock-AF_MAX" 266 }; 267 static const char *const af_family_wlock_key_strings[AF_MAX+1] = { 268 "wlock-AF_UNSPEC", "wlock-AF_UNIX" , "wlock-AF_INET" , 269 "wlock-AF_AX25" , "wlock-AF_IPX" , "wlock-AF_APPLETALK", 270 "wlock-AF_NETROM", "wlock-AF_BRIDGE" , "wlock-AF_ATMPVC" , 271 "wlock-AF_X25" , "wlock-AF_INET6" , "wlock-AF_ROSE" , 272 "wlock-AF_DECnet", "wlock-AF_NETBEUI" , "wlock-AF_SECURITY" , 273 "wlock-AF_KEY" , "wlock-AF_NETLINK" , "wlock-AF_PACKET" , 274 "wlock-AF_ASH" , "wlock-AF_ECONET" , "wlock-AF_ATMSVC" , 275 "wlock-AF_RDS" , "wlock-AF_SNA" , "wlock-AF_IRDA" , 276 "wlock-AF_PPPOX" , "wlock-AF_WANPIPE" , "wlock-AF_LLC" , 277 "wlock-27" , "wlock-28" , "wlock-AF_CAN" , 278 "wlock-AF_TIPC" , "wlock-AF_BLUETOOTH", "wlock-AF_IUCV" , 279 "wlock-AF_RXRPC" , "wlock-AF_ISDN" , "wlock-AF_PHONET" , 280 "wlock-AF_IEEE802154", "wlock-AF_CAIF" , "wlock-AF_ALG" , 281 "wlock-AF_NFC" , "wlock-AF_VSOCK" , "wlock-AF_KCM" , 282 "wlock-AF_QIPCRTR", "wlock-AF_SMC" , "wlock-AF_MAX" 283 }; 284 static const char *const af_family_elock_key_strings[AF_MAX+1] = { 285 "elock-AF_UNSPEC", "elock-AF_UNIX" , "elock-AF_INET" , 286 "elock-AF_AX25" , "elock-AF_IPX" , "elock-AF_APPLETALK", 287 "elock-AF_NETROM", "elock-AF_BRIDGE" , "elock-AF_ATMPVC" , 288 "elock-AF_X25" , "elock-AF_INET6" , "elock-AF_ROSE" , 289 "elock-AF_DECnet", "elock-AF_NETBEUI" , "elock-AF_SECURITY" , 290 "elock-AF_KEY" , "elock-AF_NETLINK" , "elock-AF_PACKET" , 291 "elock-AF_ASH" , "elock-AF_ECONET" , "elock-AF_ATMSVC" , 292 "elock-AF_RDS" , "elock-AF_SNA" , "elock-AF_IRDA" , 293 "elock-AF_PPPOX" , "elock-AF_WANPIPE" , "elock-AF_LLC" , 294 "elock-27" , "elock-28" , "elock-AF_CAN" , 295 "elock-AF_TIPC" , "elock-AF_BLUETOOTH", "elock-AF_IUCV" , 296 "elock-AF_RXRPC" , "elock-AF_ISDN" , "elock-AF_PHONET" , 297 "elock-AF_IEEE802154", "elock-AF_CAIF" , "elock-AF_ALG" , 298 "elock-AF_NFC" , "elock-AF_VSOCK" , "elock-AF_KCM" , 299 "elock-AF_QIPCRTR", "elock-AF_SMC" , "elock-AF_MAX" 300 }; 301 302 /* 303 * sk_callback_lock and sk queues 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 static struct lock_class_key af_rlock_keys[AF_MAX]; 308 static struct lock_class_key af_wlock_keys[AF_MAX]; 309 static struct lock_class_key af_elock_keys[AF_MAX]; 310 static struct lock_class_key af_kern_callback_keys[AF_MAX]; 311 312 /* Take into consideration the size of the struct sk_buff overhead in the 313 * determination of these values, since that is non-constant across 314 * platforms. This makes socket queueing behavior and performance 315 * not depend upon such differences. 316 */ 317 #define _SK_MEM_PACKETS 256 318 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256) 319 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 320 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 321 322 /* Run time adjustable parameters. */ 323 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX; 324 EXPORT_SYMBOL(sysctl_wmem_max); 325 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX; 326 EXPORT_SYMBOL(sysctl_rmem_max); 327 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX; 328 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX; 329 330 /* Maximal space eaten by iovec or ancillary data plus some space */ 331 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512); 332 EXPORT_SYMBOL(sysctl_optmem_max); 333 334 int sysctl_tstamp_allow_data __read_mostly = 1; 335 336 struct static_key memalloc_socks = STATIC_KEY_INIT_FALSE; 337 EXPORT_SYMBOL_GPL(memalloc_socks); 338 339 /** 340 * sk_set_memalloc - sets %SOCK_MEMALLOC 341 * @sk: socket to set it on 342 * 343 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves. 344 * It's the responsibility of the admin to adjust min_free_kbytes 345 * to meet the requirements 346 */ 347 void sk_set_memalloc(struct sock *sk) 348 { 349 sock_set_flag(sk, SOCK_MEMALLOC); 350 sk->sk_allocation |= __GFP_MEMALLOC; 351 static_key_slow_inc(&memalloc_socks); 352 } 353 EXPORT_SYMBOL_GPL(sk_set_memalloc); 354 355 void sk_clear_memalloc(struct sock *sk) 356 { 357 sock_reset_flag(sk, SOCK_MEMALLOC); 358 sk->sk_allocation &= ~__GFP_MEMALLOC; 359 static_key_slow_dec(&memalloc_socks); 360 361 /* 362 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward 363 * progress of swapping. SOCK_MEMALLOC may be cleared while 364 * it has rmem allocations due to the last swapfile being deactivated 365 * but there is a risk that the socket is unusable due to exceeding 366 * the rmem limits. Reclaim the reserves and obey rmem limits again. 367 */ 368 sk_mem_reclaim(sk); 369 } 370 EXPORT_SYMBOL_GPL(sk_clear_memalloc); 371 372 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb) 373 { 374 int ret; 375 unsigned long pflags = current->flags; 376 377 /* these should have been dropped before queueing */ 378 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC)); 379 380 current->flags |= PF_MEMALLOC; 381 ret = sk->sk_backlog_rcv(sk, skb); 382 tsk_restore_flags(current, pflags, PF_MEMALLOC); 383 384 return ret; 385 } 386 EXPORT_SYMBOL(__sk_backlog_rcv); 387 388 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen) 389 { 390 struct timeval tv; 391 392 if (optlen < sizeof(tv)) 393 return -EINVAL; 394 if (copy_from_user(&tv, optval, sizeof(tv))) 395 return -EFAULT; 396 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC) 397 return -EDOM; 398 399 if (tv.tv_sec < 0) { 400 static int warned __read_mostly; 401 402 *timeo_p = 0; 403 if (warned < 10 && net_ratelimit()) { 404 warned++; 405 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n", 406 __func__, current->comm, task_pid_nr(current)); 407 } 408 return 0; 409 } 410 *timeo_p = MAX_SCHEDULE_TIMEOUT; 411 if (tv.tv_sec == 0 && tv.tv_usec == 0) 412 return 0; 413 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1)) 414 *timeo_p = tv.tv_sec * HZ + DIV_ROUND_UP(tv.tv_usec, USEC_PER_SEC / HZ); 415 return 0; 416 } 417 418 static void sock_warn_obsolete_bsdism(const char *name) 419 { 420 static int warned; 421 static char warncomm[TASK_COMM_LEN]; 422 if (strcmp(warncomm, current->comm) && warned < 5) { 423 strcpy(warncomm, current->comm); 424 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n", 425 warncomm, name); 426 warned++; 427 } 428 } 429 430 static bool sock_needs_netstamp(const struct sock *sk) 431 { 432 switch (sk->sk_family) { 433 case AF_UNSPEC: 434 case AF_UNIX: 435 return false; 436 default: 437 return true; 438 } 439 } 440 441 static void sock_disable_timestamp(struct sock *sk, unsigned long flags) 442 { 443 if (sk->sk_flags & flags) { 444 sk->sk_flags &= ~flags; 445 if (sock_needs_netstamp(sk) && 446 !(sk->sk_flags & SK_FLAGS_TIMESTAMP)) 447 net_disable_timestamp(); 448 } 449 } 450 451 452 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 453 { 454 unsigned long flags; 455 struct sk_buff_head *list = &sk->sk_receive_queue; 456 457 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) { 458 atomic_inc(&sk->sk_drops); 459 trace_sock_rcvqueue_full(sk, skb); 460 return -ENOMEM; 461 } 462 463 if (!sk_rmem_schedule(sk, skb, skb->truesize)) { 464 atomic_inc(&sk->sk_drops); 465 return -ENOBUFS; 466 } 467 468 skb->dev = NULL; 469 skb_set_owner_r(skb, sk); 470 471 /* we escape from rcu protected region, make sure we dont leak 472 * a norefcounted dst 473 */ 474 skb_dst_force(skb); 475 476 spin_lock_irqsave(&list->lock, flags); 477 sock_skb_set_dropcount(sk, skb); 478 __skb_queue_tail(list, skb); 479 spin_unlock_irqrestore(&list->lock, flags); 480 481 if (!sock_flag(sk, SOCK_DEAD)) 482 sk->sk_data_ready(sk); 483 return 0; 484 } 485 EXPORT_SYMBOL(__sock_queue_rcv_skb); 486 487 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 488 { 489 int err; 490 491 err = sk_filter(sk, skb); 492 if (err) 493 return err; 494 495 return __sock_queue_rcv_skb(sk, skb); 496 } 497 EXPORT_SYMBOL(sock_queue_rcv_skb); 498 499 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, 500 const int nested, unsigned int trim_cap, bool refcounted) 501 { 502 int rc = NET_RX_SUCCESS; 503 504 if (sk_filter_trim_cap(sk, skb, trim_cap)) 505 goto discard_and_relse; 506 507 skb->dev = NULL; 508 509 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) { 510 atomic_inc(&sk->sk_drops); 511 goto discard_and_relse; 512 } 513 if (nested) 514 bh_lock_sock_nested(sk); 515 else 516 bh_lock_sock(sk); 517 if (!sock_owned_by_user(sk)) { 518 /* 519 * trylock + unlock semantics: 520 */ 521 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_); 522 523 rc = sk_backlog_rcv(sk, skb); 524 525 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_); 526 } else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) { 527 bh_unlock_sock(sk); 528 atomic_inc(&sk->sk_drops); 529 goto discard_and_relse; 530 } 531 532 bh_unlock_sock(sk); 533 out: 534 if (refcounted) 535 sock_put(sk); 536 return rc; 537 discard_and_relse: 538 kfree_skb(skb); 539 goto out; 540 } 541 EXPORT_SYMBOL(__sk_receive_skb); 542 543 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie) 544 { 545 struct dst_entry *dst = __sk_dst_get(sk); 546 547 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) { 548 sk_tx_queue_clear(sk); 549 sk->sk_dst_pending_confirm = 0; 550 RCU_INIT_POINTER(sk->sk_dst_cache, NULL); 551 dst_release(dst); 552 return NULL; 553 } 554 555 return dst; 556 } 557 EXPORT_SYMBOL(__sk_dst_check); 558 559 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie) 560 { 561 struct dst_entry *dst = sk_dst_get(sk); 562 563 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) { 564 sk_dst_reset(sk); 565 dst_release(dst); 566 return NULL; 567 } 568 569 return dst; 570 } 571 EXPORT_SYMBOL(sk_dst_check); 572 573 static int sock_setbindtodevice(struct sock *sk, char __user *optval, 574 int optlen) 575 { 576 int ret = -ENOPROTOOPT; 577 #ifdef CONFIG_NETDEVICES 578 struct net *net = sock_net(sk); 579 char devname[IFNAMSIZ]; 580 int index; 581 582 /* Sorry... */ 583 ret = -EPERM; 584 if (!ns_capable(net->user_ns, CAP_NET_RAW)) 585 goto out; 586 587 ret = -EINVAL; 588 if (optlen < 0) 589 goto out; 590 591 /* Bind this socket to a particular device like "eth0", 592 * as specified in the passed interface name. If the 593 * name is "" or the option length is zero the socket 594 * is not bound. 595 */ 596 if (optlen > IFNAMSIZ - 1) 597 optlen = IFNAMSIZ - 1; 598 memset(devname, 0, sizeof(devname)); 599 600 ret = -EFAULT; 601 if (copy_from_user(devname, optval, optlen)) 602 goto out; 603 604 index = 0; 605 if (devname[0] != '\0') { 606 struct net_device *dev; 607 608 rcu_read_lock(); 609 dev = dev_get_by_name_rcu(net, devname); 610 if (dev) 611 index = dev->ifindex; 612 rcu_read_unlock(); 613 ret = -ENODEV; 614 if (!dev) 615 goto out; 616 } 617 618 lock_sock(sk); 619 sk->sk_bound_dev_if = index; 620 sk_dst_reset(sk); 621 release_sock(sk); 622 623 ret = 0; 624 625 out: 626 #endif 627 628 return ret; 629 } 630 631 static int sock_getbindtodevice(struct sock *sk, char __user *optval, 632 int __user *optlen, int len) 633 { 634 int ret = -ENOPROTOOPT; 635 #ifdef CONFIG_NETDEVICES 636 struct net *net = sock_net(sk); 637 char devname[IFNAMSIZ]; 638 639 if (sk->sk_bound_dev_if == 0) { 640 len = 0; 641 goto zero; 642 } 643 644 ret = -EINVAL; 645 if (len < IFNAMSIZ) 646 goto out; 647 648 ret = netdev_get_name(net, devname, sk->sk_bound_dev_if); 649 if (ret) 650 goto out; 651 652 len = strlen(devname) + 1; 653 654 ret = -EFAULT; 655 if (copy_to_user(optval, devname, len)) 656 goto out; 657 658 zero: 659 ret = -EFAULT; 660 if (put_user(len, optlen)) 661 goto out; 662 663 ret = 0; 664 665 out: 666 #endif 667 668 return ret; 669 } 670 671 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool) 672 { 673 if (valbool) 674 sock_set_flag(sk, bit); 675 else 676 sock_reset_flag(sk, bit); 677 } 678 679 bool sk_mc_loop(struct sock *sk) 680 { 681 if (dev_recursion_level()) 682 return false; 683 if (!sk) 684 return true; 685 switch (sk->sk_family) { 686 case AF_INET: 687 return inet_sk(sk)->mc_loop; 688 #if IS_ENABLED(CONFIG_IPV6) 689 case AF_INET6: 690 return inet6_sk(sk)->mc_loop; 691 #endif 692 } 693 WARN_ON(1); 694 return true; 695 } 696 EXPORT_SYMBOL(sk_mc_loop); 697 698 /* 699 * This is meant for all protocols to use and covers goings on 700 * at the socket level. Everything here is generic. 701 */ 702 703 int sock_setsockopt(struct socket *sock, int level, int optname, 704 char __user *optval, unsigned int optlen) 705 { 706 struct sock *sk = sock->sk; 707 int val; 708 int valbool; 709 struct linger ling; 710 int ret = 0; 711 712 /* 713 * Options without arguments 714 */ 715 716 if (optname == SO_BINDTODEVICE) 717 return sock_setbindtodevice(sk, optval, optlen); 718 719 if (optlen < sizeof(int)) 720 return -EINVAL; 721 722 if (get_user(val, (int __user *)optval)) 723 return -EFAULT; 724 725 valbool = val ? 1 : 0; 726 727 lock_sock(sk); 728 729 switch (optname) { 730 case SO_DEBUG: 731 if (val && !capable(CAP_NET_ADMIN)) 732 ret = -EACCES; 733 else 734 sock_valbool_flag(sk, SOCK_DBG, valbool); 735 break; 736 case SO_REUSEADDR: 737 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE); 738 break; 739 case SO_REUSEPORT: 740 sk->sk_reuseport = valbool; 741 break; 742 case SO_TYPE: 743 case SO_PROTOCOL: 744 case SO_DOMAIN: 745 case SO_ERROR: 746 ret = -ENOPROTOOPT; 747 break; 748 case SO_DONTROUTE: 749 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool); 750 break; 751 case SO_BROADCAST: 752 sock_valbool_flag(sk, SOCK_BROADCAST, valbool); 753 break; 754 case SO_SNDBUF: 755 /* Don't error on this BSD doesn't and if you think 756 * about it this is right. Otherwise apps have to 757 * play 'guess the biggest size' games. RCVBUF/SNDBUF 758 * are treated in BSD as hints 759 */ 760 val = min_t(u32, val, sysctl_wmem_max); 761 set_sndbuf: 762 sk->sk_userlocks |= SOCK_SNDBUF_LOCK; 763 sk->sk_sndbuf = max_t(int, val * 2, SOCK_MIN_SNDBUF); 764 /* Wake up sending tasks if we upped the value. */ 765 sk->sk_write_space(sk); 766 break; 767 768 case SO_SNDBUFFORCE: 769 if (!capable(CAP_NET_ADMIN)) { 770 ret = -EPERM; 771 break; 772 } 773 goto set_sndbuf; 774 775 case SO_RCVBUF: 776 /* Don't error on this BSD doesn't and if you think 777 * about it this is right. Otherwise apps have to 778 * play 'guess the biggest size' games. RCVBUF/SNDBUF 779 * are treated in BSD as hints 780 */ 781 val = min_t(u32, val, sysctl_rmem_max); 782 set_rcvbuf: 783 sk->sk_userlocks |= SOCK_RCVBUF_LOCK; 784 /* 785 * We double it on the way in to account for 786 * "struct sk_buff" etc. overhead. Applications 787 * assume that the SO_RCVBUF setting they make will 788 * allow that much actual data to be received on that 789 * socket. 790 * 791 * Applications are unaware that "struct sk_buff" and 792 * other overheads allocate from the receive buffer 793 * during socket buffer allocation. 794 * 795 * And after considering the possible alternatives, 796 * returning the value we actually used in getsockopt 797 * is the most desirable behavior. 798 */ 799 sk->sk_rcvbuf = max_t(int, val * 2, SOCK_MIN_RCVBUF); 800 break; 801 802 case SO_RCVBUFFORCE: 803 if (!capable(CAP_NET_ADMIN)) { 804 ret = -EPERM; 805 break; 806 } 807 goto set_rcvbuf; 808 809 case SO_KEEPALIVE: 810 if (sk->sk_prot->keepalive) 811 sk->sk_prot->keepalive(sk, valbool); 812 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool); 813 break; 814 815 case SO_OOBINLINE: 816 sock_valbool_flag(sk, SOCK_URGINLINE, valbool); 817 break; 818 819 case SO_NO_CHECK: 820 sk->sk_no_check_tx = valbool; 821 break; 822 823 case SO_PRIORITY: 824 if ((val >= 0 && val <= 6) || 825 ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) 826 sk->sk_priority = val; 827 else 828 ret = -EPERM; 829 break; 830 831 case SO_LINGER: 832 if (optlen < sizeof(ling)) { 833 ret = -EINVAL; /* 1003.1g */ 834 break; 835 } 836 if (copy_from_user(&ling, optval, sizeof(ling))) { 837 ret = -EFAULT; 838 break; 839 } 840 if (!ling.l_onoff) 841 sock_reset_flag(sk, SOCK_LINGER); 842 else { 843 #if (BITS_PER_LONG == 32) 844 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ) 845 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT; 846 else 847 #endif 848 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ; 849 sock_set_flag(sk, SOCK_LINGER); 850 } 851 break; 852 853 case SO_BSDCOMPAT: 854 sock_warn_obsolete_bsdism("setsockopt"); 855 break; 856 857 case SO_PASSCRED: 858 if (valbool) 859 set_bit(SOCK_PASSCRED, &sock->flags); 860 else 861 clear_bit(SOCK_PASSCRED, &sock->flags); 862 break; 863 864 case SO_TIMESTAMP: 865 case SO_TIMESTAMPNS: 866 if (valbool) { 867 if (optname == SO_TIMESTAMP) 868 sock_reset_flag(sk, SOCK_RCVTSTAMPNS); 869 else 870 sock_set_flag(sk, SOCK_RCVTSTAMPNS); 871 sock_set_flag(sk, SOCK_RCVTSTAMP); 872 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 873 } else { 874 sock_reset_flag(sk, SOCK_RCVTSTAMP); 875 sock_reset_flag(sk, SOCK_RCVTSTAMPNS); 876 } 877 break; 878 879 case SO_TIMESTAMPING: 880 if (val & ~SOF_TIMESTAMPING_MASK) { 881 ret = -EINVAL; 882 break; 883 } 884 885 if (val & SOF_TIMESTAMPING_OPT_ID && 886 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) { 887 if (sk->sk_protocol == IPPROTO_TCP && 888 sk->sk_type == SOCK_STREAM) { 889 if ((1 << sk->sk_state) & 890 (TCPF_CLOSE | TCPF_LISTEN)) { 891 ret = -EINVAL; 892 break; 893 } 894 sk->sk_tskey = tcp_sk(sk)->snd_una; 895 } else { 896 sk->sk_tskey = 0; 897 } 898 } 899 900 if (val & SOF_TIMESTAMPING_OPT_STATS && 901 !(val & SOF_TIMESTAMPING_OPT_TSONLY)) { 902 ret = -EINVAL; 903 break; 904 } 905 906 sk->sk_tsflags = val; 907 if (val & SOF_TIMESTAMPING_RX_SOFTWARE) 908 sock_enable_timestamp(sk, 909 SOCK_TIMESTAMPING_RX_SOFTWARE); 910 else 911 sock_disable_timestamp(sk, 912 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE)); 913 break; 914 915 case SO_RCVLOWAT: 916 if (val < 0) 917 val = INT_MAX; 918 sk->sk_rcvlowat = val ? : 1; 919 break; 920 921 case SO_RCVTIMEO: 922 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen); 923 break; 924 925 case SO_SNDTIMEO: 926 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen); 927 break; 928 929 case SO_ATTACH_FILTER: 930 ret = -EINVAL; 931 if (optlen == sizeof(struct sock_fprog)) { 932 struct sock_fprog fprog; 933 934 ret = -EFAULT; 935 if (copy_from_user(&fprog, optval, sizeof(fprog))) 936 break; 937 938 ret = sk_attach_filter(&fprog, sk); 939 } 940 break; 941 942 case SO_ATTACH_BPF: 943 ret = -EINVAL; 944 if (optlen == sizeof(u32)) { 945 u32 ufd; 946 947 ret = -EFAULT; 948 if (copy_from_user(&ufd, optval, sizeof(ufd))) 949 break; 950 951 ret = sk_attach_bpf(ufd, sk); 952 } 953 break; 954 955 case SO_ATTACH_REUSEPORT_CBPF: 956 ret = -EINVAL; 957 if (optlen == sizeof(struct sock_fprog)) { 958 struct sock_fprog fprog; 959 960 ret = -EFAULT; 961 if (copy_from_user(&fprog, optval, sizeof(fprog))) 962 break; 963 964 ret = sk_reuseport_attach_filter(&fprog, sk); 965 } 966 break; 967 968 case SO_ATTACH_REUSEPORT_EBPF: 969 ret = -EINVAL; 970 if (optlen == sizeof(u32)) { 971 u32 ufd; 972 973 ret = -EFAULT; 974 if (copy_from_user(&ufd, optval, sizeof(ufd))) 975 break; 976 977 ret = sk_reuseport_attach_bpf(ufd, sk); 978 } 979 break; 980 981 case SO_DETACH_FILTER: 982 ret = sk_detach_filter(sk); 983 break; 984 985 case SO_LOCK_FILTER: 986 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool) 987 ret = -EPERM; 988 else 989 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool); 990 break; 991 992 case SO_PASSSEC: 993 if (valbool) 994 set_bit(SOCK_PASSSEC, &sock->flags); 995 else 996 clear_bit(SOCK_PASSSEC, &sock->flags); 997 break; 998 case SO_MARK: 999 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) 1000 ret = -EPERM; 1001 else 1002 sk->sk_mark = val; 1003 break; 1004 1005 case SO_RXQ_OVFL: 1006 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool); 1007 break; 1008 1009 case SO_WIFI_STATUS: 1010 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool); 1011 break; 1012 1013 case SO_PEEK_OFF: 1014 if (sock->ops->set_peek_off) 1015 ret = sock->ops->set_peek_off(sk, val); 1016 else 1017 ret = -EOPNOTSUPP; 1018 break; 1019 1020 case SO_NOFCS: 1021 sock_valbool_flag(sk, SOCK_NOFCS, valbool); 1022 break; 1023 1024 case SO_SELECT_ERR_QUEUE: 1025 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool); 1026 break; 1027 1028 #ifdef CONFIG_NET_RX_BUSY_POLL 1029 case SO_BUSY_POLL: 1030 /* allow unprivileged users to decrease the value */ 1031 if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN)) 1032 ret = -EPERM; 1033 else { 1034 if (val < 0) 1035 ret = -EINVAL; 1036 else 1037 sk->sk_ll_usec = val; 1038 } 1039 break; 1040 #endif 1041 1042 case SO_MAX_PACING_RATE: 1043 sk->sk_max_pacing_rate = val; 1044 sk->sk_pacing_rate = min(sk->sk_pacing_rate, 1045 sk->sk_max_pacing_rate); 1046 break; 1047 1048 case SO_INCOMING_CPU: 1049 sk->sk_incoming_cpu = val; 1050 break; 1051 1052 case SO_CNX_ADVICE: 1053 if (val == 1) 1054 dst_negative_advice(sk); 1055 break; 1056 default: 1057 ret = -ENOPROTOOPT; 1058 break; 1059 } 1060 release_sock(sk); 1061 return ret; 1062 } 1063 EXPORT_SYMBOL(sock_setsockopt); 1064 1065 1066 static void cred_to_ucred(struct pid *pid, const struct cred *cred, 1067 struct ucred *ucred) 1068 { 1069 ucred->pid = pid_vnr(pid); 1070 ucred->uid = ucred->gid = -1; 1071 if (cred) { 1072 struct user_namespace *current_ns = current_user_ns(); 1073 1074 ucred->uid = from_kuid_munged(current_ns, cred->euid); 1075 ucred->gid = from_kgid_munged(current_ns, cred->egid); 1076 } 1077 } 1078 1079 int sock_getsockopt(struct socket *sock, int level, int optname, 1080 char __user *optval, int __user *optlen) 1081 { 1082 struct sock *sk = sock->sk; 1083 1084 union { 1085 int val; 1086 struct linger ling; 1087 struct timeval tm; 1088 } v; 1089 1090 int lv = sizeof(int); 1091 int len; 1092 1093 if (get_user(len, optlen)) 1094 return -EFAULT; 1095 if (len < 0) 1096 return -EINVAL; 1097 1098 memset(&v, 0, sizeof(v)); 1099 1100 switch (optname) { 1101 case SO_DEBUG: 1102 v.val = sock_flag(sk, SOCK_DBG); 1103 break; 1104 1105 case SO_DONTROUTE: 1106 v.val = sock_flag(sk, SOCK_LOCALROUTE); 1107 break; 1108 1109 case SO_BROADCAST: 1110 v.val = sock_flag(sk, SOCK_BROADCAST); 1111 break; 1112 1113 case SO_SNDBUF: 1114 v.val = sk->sk_sndbuf; 1115 break; 1116 1117 case SO_RCVBUF: 1118 v.val = sk->sk_rcvbuf; 1119 break; 1120 1121 case SO_REUSEADDR: 1122 v.val = sk->sk_reuse; 1123 break; 1124 1125 case SO_REUSEPORT: 1126 v.val = sk->sk_reuseport; 1127 break; 1128 1129 case SO_KEEPALIVE: 1130 v.val = sock_flag(sk, SOCK_KEEPOPEN); 1131 break; 1132 1133 case SO_TYPE: 1134 v.val = sk->sk_type; 1135 break; 1136 1137 case SO_PROTOCOL: 1138 v.val = sk->sk_protocol; 1139 break; 1140 1141 case SO_DOMAIN: 1142 v.val = sk->sk_family; 1143 break; 1144 1145 case SO_ERROR: 1146 v.val = -sock_error(sk); 1147 if (v.val == 0) 1148 v.val = xchg(&sk->sk_err_soft, 0); 1149 break; 1150 1151 case SO_OOBINLINE: 1152 v.val = sock_flag(sk, SOCK_URGINLINE); 1153 break; 1154 1155 case SO_NO_CHECK: 1156 v.val = sk->sk_no_check_tx; 1157 break; 1158 1159 case SO_PRIORITY: 1160 v.val = sk->sk_priority; 1161 break; 1162 1163 case SO_LINGER: 1164 lv = sizeof(v.ling); 1165 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER); 1166 v.ling.l_linger = sk->sk_lingertime / HZ; 1167 break; 1168 1169 case SO_BSDCOMPAT: 1170 sock_warn_obsolete_bsdism("getsockopt"); 1171 break; 1172 1173 case SO_TIMESTAMP: 1174 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && 1175 !sock_flag(sk, SOCK_RCVTSTAMPNS); 1176 break; 1177 1178 case SO_TIMESTAMPNS: 1179 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS); 1180 break; 1181 1182 case SO_TIMESTAMPING: 1183 v.val = sk->sk_tsflags; 1184 break; 1185 1186 case SO_RCVTIMEO: 1187 lv = sizeof(struct timeval); 1188 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) { 1189 v.tm.tv_sec = 0; 1190 v.tm.tv_usec = 0; 1191 } else { 1192 v.tm.tv_sec = sk->sk_rcvtimeo / HZ; 1193 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * USEC_PER_SEC) / HZ; 1194 } 1195 break; 1196 1197 case SO_SNDTIMEO: 1198 lv = sizeof(struct timeval); 1199 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) { 1200 v.tm.tv_sec = 0; 1201 v.tm.tv_usec = 0; 1202 } else { 1203 v.tm.tv_sec = sk->sk_sndtimeo / HZ; 1204 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * USEC_PER_SEC) / HZ; 1205 } 1206 break; 1207 1208 case SO_RCVLOWAT: 1209 v.val = sk->sk_rcvlowat; 1210 break; 1211 1212 case SO_SNDLOWAT: 1213 v.val = 1; 1214 break; 1215 1216 case SO_PASSCRED: 1217 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags); 1218 break; 1219 1220 case SO_PEERCRED: 1221 { 1222 struct ucred peercred; 1223 if (len > sizeof(peercred)) 1224 len = sizeof(peercred); 1225 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred); 1226 if (copy_to_user(optval, &peercred, len)) 1227 return -EFAULT; 1228 goto lenout; 1229 } 1230 1231 case SO_PEERNAME: 1232 { 1233 char address[128]; 1234 1235 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2)) 1236 return -ENOTCONN; 1237 if (lv < len) 1238 return -EINVAL; 1239 if (copy_to_user(optval, address, len)) 1240 return -EFAULT; 1241 goto lenout; 1242 } 1243 1244 /* Dubious BSD thing... Probably nobody even uses it, but 1245 * the UNIX standard wants it for whatever reason... -DaveM 1246 */ 1247 case SO_ACCEPTCONN: 1248 v.val = sk->sk_state == TCP_LISTEN; 1249 break; 1250 1251 case SO_PASSSEC: 1252 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags); 1253 break; 1254 1255 case SO_PEERSEC: 1256 return security_socket_getpeersec_stream(sock, optval, optlen, len); 1257 1258 case SO_MARK: 1259 v.val = sk->sk_mark; 1260 break; 1261 1262 case SO_RXQ_OVFL: 1263 v.val = sock_flag(sk, SOCK_RXQ_OVFL); 1264 break; 1265 1266 case SO_WIFI_STATUS: 1267 v.val = sock_flag(sk, SOCK_WIFI_STATUS); 1268 break; 1269 1270 case SO_PEEK_OFF: 1271 if (!sock->ops->set_peek_off) 1272 return -EOPNOTSUPP; 1273 1274 v.val = sk->sk_peek_off; 1275 break; 1276 case SO_NOFCS: 1277 v.val = sock_flag(sk, SOCK_NOFCS); 1278 break; 1279 1280 case SO_BINDTODEVICE: 1281 return sock_getbindtodevice(sk, optval, optlen, len); 1282 1283 case SO_GET_FILTER: 1284 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len); 1285 if (len < 0) 1286 return len; 1287 1288 goto lenout; 1289 1290 case SO_LOCK_FILTER: 1291 v.val = sock_flag(sk, SOCK_FILTER_LOCKED); 1292 break; 1293 1294 case SO_BPF_EXTENSIONS: 1295 v.val = bpf_tell_extensions(); 1296 break; 1297 1298 case SO_SELECT_ERR_QUEUE: 1299 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE); 1300 break; 1301 1302 #ifdef CONFIG_NET_RX_BUSY_POLL 1303 case SO_BUSY_POLL: 1304 v.val = sk->sk_ll_usec; 1305 break; 1306 #endif 1307 1308 case SO_MAX_PACING_RATE: 1309 v.val = sk->sk_max_pacing_rate; 1310 break; 1311 1312 case SO_INCOMING_CPU: 1313 v.val = sk->sk_incoming_cpu; 1314 break; 1315 1316 case SO_MEMINFO: 1317 { 1318 u32 meminfo[SK_MEMINFO_VARS]; 1319 1320 if (get_user(len, optlen)) 1321 return -EFAULT; 1322 1323 sk_get_meminfo(sk, meminfo); 1324 1325 len = min_t(unsigned int, len, sizeof(meminfo)); 1326 if (copy_to_user(optval, &meminfo, len)) 1327 return -EFAULT; 1328 1329 goto lenout; 1330 } 1331 1332 #ifdef CONFIG_NET_RX_BUSY_POLL 1333 case SO_INCOMING_NAPI_ID: 1334 v.val = READ_ONCE(sk->sk_napi_id); 1335 1336 /* aggregate non-NAPI IDs down to 0 */ 1337 if (v.val < MIN_NAPI_ID) 1338 v.val = 0; 1339 1340 break; 1341 #endif 1342 1343 default: 1344 /* We implement the SO_SNDLOWAT etc to not be settable 1345 * (1003.1g 7). 1346 */ 1347 return -ENOPROTOOPT; 1348 } 1349 1350 if (len > lv) 1351 len = lv; 1352 if (copy_to_user(optval, &v, len)) 1353 return -EFAULT; 1354 lenout: 1355 if (put_user(len, optlen)) 1356 return -EFAULT; 1357 return 0; 1358 } 1359 1360 /* 1361 * Initialize an sk_lock. 1362 * 1363 * (We also register the sk_lock with the lock validator.) 1364 */ 1365 static inline void sock_lock_init(struct sock *sk) 1366 { 1367 if (sk->sk_kern_sock) 1368 sock_lock_init_class_and_name( 1369 sk, 1370 af_family_kern_slock_key_strings[sk->sk_family], 1371 af_family_kern_slock_keys + sk->sk_family, 1372 af_family_kern_key_strings[sk->sk_family], 1373 af_family_kern_keys + sk->sk_family); 1374 else 1375 sock_lock_init_class_and_name( 1376 sk, 1377 af_family_slock_key_strings[sk->sk_family], 1378 af_family_slock_keys + sk->sk_family, 1379 af_family_key_strings[sk->sk_family], 1380 af_family_keys + sk->sk_family); 1381 } 1382 1383 /* 1384 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet, 1385 * even temporarly, because of RCU lookups. sk_node should also be left as is. 1386 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end 1387 */ 1388 static void sock_copy(struct sock *nsk, const struct sock *osk) 1389 { 1390 #ifdef CONFIG_SECURITY_NETWORK 1391 void *sptr = nsk->sk_security; 1392 #endif 1393 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin)); 1394 1395 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end, 1396 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end)); 1397 1398 #ifdef CONFIG_SECURITY_NETWORK 1399 nsk->sk_security = sptr; 1400 security_sk_clone(osk, nsk); 1401 #endif 1402 } 1403 1404 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority, 1405 int family) 1406 { 1407 struct sock *sk; 1408 struct kmem_cache *slab; 1409 1410 slab = prot->slab; 1411 if (slab != NULL) { 1412 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO); 1413 if (!sk) 1414 return sk; 1415 if (priority & __GFP_ZERO) 1416 sk_prot_clear_nulls(sk, prot->obj_size); 1417 } else 1418 sk = kmalloc(prot->obj_size, priority); 1419 1420 if (sk != NULL) { 1421 kmemcheck_annotate_bitfield(sk, flags); 1422 1423 if (security_sk_alloc(sk, family, priority)) 1424 goto out_free; 1425 1426 if (!try_module_get(prot->owner)) 1427 goto out_free_sec; 1428 sk_tx_queue_clear(sk); 1429 } 1430 1431 return sk; 1432 1433 out_free_sec: 1434 security_sk_free(sk); 1435 out_free: 1436 if (slab != NULL) 1437 kmem_cache_free(slab, sk); 1438 else 1439 kfree(sk); 1440 return NULL; 1441 } 1442 1443 static void sk_prot_free(struct proto *prot, struct sock *sk) 1444 { 1445 struct kmem_cache *slab; 1446 struct module *owner; 1447 1448 owner = prot->owner; 1449 slab = prot->slab; 1450 1451 cgroup_sk_free(&sk->sk_cgrp_data); 1452 mem_cgroup_sk_free(sk); 1453 security_sk_free(sk); 1454 if (slab != NULL) 1455 kmem_cache_free(slab, sk); 1456 else 1457 kfree(sk); 1458 module_put(owner); 1459 } 1460 1461 /** 1462 * sk_alloc - All socket objects are allocated here 1463 * @net: the applicable net namespace 1464 * @family: protocol family 1465 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 1466 * @prot: struct proto associated with this new sock instance 1467 * @kern: is this to be a kernel socket? 1468 */ 1469 struct sock *sk_alloc(struct net *net, int family, gfp_t priority, 1470 struct proto *prot, int kern) 1471 { 1472 struct sock *sk; 1473 1474 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family); 1475 if (sk) { 1476 sk->sk_family = family; 1477 /* 1478 * See comment in struct sock definition to understand 1479 * why we need sk_prot_creator -acme 1480 */ 1481 sk->sk_prot = sk->sk_prot_creator = prot; 1482 sk->sk_kern_sock = kern; 1483 sock_lock_init(sk); 1484 sk->sk_net_refcnt = kern ? 0 : 1; 1485 if (likely(sk->sk_net_refcnt)) 1486 get_net(net); 1487 sock_net_set(sk, net); 1488 atomic_set(&sk->sk_wmem_alloc, 1); 1489 1490 mem_cgroup_sk_alloc(sk); 1491 cgroup_sk_alloc(&sk->sk_cgrp_data); 1492 sock_update_classid(&sk->sk_cgrp_data); 1493 sock_update_netprioidx(&sk->sk_cgrp_data); 1494 } 1495 1496 return sk; 1497 } 1498 EXPORT_SYMBOL(sk_alloc); 1499 1500 /* Sockets having SOCK_RCU_FREE will call this function after one RCU 1501 * grace period. This is the case for UDP sockets and TCP listeners. 1502 */ 1503 static void __sk_destruct(struct rcu_head *head) 1504 { 1505 struct sock *sk = container_of(head, struct sock, sk_rcu); 1506 struct sk_filter *filter; 1507 1508 if (sk->sk_destruct) 1509 sk->sk_destruct(sk); 1510 1511 filter = rcu_dereference_check(sk->sk_filter, 1512 atomic_read(&sk->sk_wmem_alloc) == 0); 1513 if (filter) { 1514 sk_filter_uncharge(sk, filter); 1515 RCU_INIT_POINTER(sk->sk_filter, NULL); 1516 } 1517 if (rcu_access_pointer(sk->sk_reuseport_cb)) 1518 reuseport_detach_sock(sk); 1519 1520 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP); 1521 1522 if (atomic_read(&sk->sk_omem_alloc)) 1523 pr_debug("%s: optmem leakage (%d bytes) detected\n", 1524 __func__, atomic_read(&sk->sk_omem_alloc)); 1525 1526 if (sk->sk_frag.page) { 1527 put_page(sk->sk_frag.page); 1528 sk->sk_frag.page = NULL; 1529 } 1530 1531 if (sk->sk_peer_cred) 1532 put_cred(sk->sk_peer_cred); 1533 put_pid(sk->sk_peer_pid); 1534 if (likely(sk->sk_net_refcnt)) 1535 put_net(sock_net(sk)); 1536 sk_prot_free(sk->sk_prot_creator, sk); 1537 } 1538 1539 void sk_destruct(struct sock *sk) 1540 { 1541 if (sock_flag(sk, SOCK_RCU_FREE)) 1542 call_rcu(&sk->sk_rcu, __sk_destruct); 1543 else 1544 __sk_destruct(&sk->sk_rcu); 1545 } 1546 1547 static void __sk_free(struct sock *sk) 1548 { 1549 if (unlikely(sock_diag_has_destroy_listeners(sk) && sk->sk_net_refcnt)) 1550 sock_diag_broadcast_destroy(sk); 1551 else 1552 sk_destruct(sk); 1553 } 1554 1555 void sk_free(struct sock *sk) 1556 { 1557 /* 1558 * We subtract one from sk_wmem_alloc and can know if 1559 * some packets are still in some tx queue. 1560 * If not null, sock_wfree() will call __sk_free(sk) later 1561 */ 1562 if (atomic_dec_and_test(&sk->sk_wmem_alloc)) 1563 __sk_free(sk); 1564 } 1565 EXPORT_SYMBOL(sk_free); 1566 1567 static void sk_init_common(struct sock *sk) 1568 { 1569 skb_queue_head_init(&sk->sk_receive_queue); 1570 skb_queue_head_init(&sk->sk_write_queue); 1571 skb_queue_head_init(&sk->sk_error_queue); 1572 1573 rwlock_init(&sk->sk_callback_lock); 1574 lockdep_set_class_and_name(&sk->sk_receive_queue.lock, 1575 af_rlock_keys + sk->sk_family, 1576 af_family_rlock_key_strings[sk->sk_family]); 1577 lockdep_set_class_and_name(&sk->sk_write_queue.lock, 1578 af_wlock_keys + sk->sk_family, 1579 af_family_wlock_key_strings[sk->sk_family]); 1580 lockdep_set_class_and_name(&sk->sk_error_queue.lock, 1581 af_elock_keys + sk->sk_family, 1582 af_family_elock_key_strings[sk->sk_family]); 1583 lockdep_set_class_and_name(&sk->sk_callback_lock, 1584 af_callback_keys + sk->sk_family, 1585 af_family_clock_key_strings[sk->sk_family]); 1586 } 1587 1588 /** 1589 * sk_clone_lock - clone a socket, and lock its clone 1590 * @sk: the socket to clone 1591 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 1592 * 1593 * Caller must unlock socket even in error path (bh_unlock_sock(newsk)) 1594 */ 1595 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority) 1596 { 1597 struct sock *newsk; 1598 bool is_charged = true; 1599 1600 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family); 1601 if (newsk != NULL) { 1602 struct sk_filter *filter; 1603 1604 sock_copy(newsk, sk); 1605 1606 /* SANITY */ 1607 if (likely(newsk->sk_net_refcnt)) 1608 get_net(sock_net(newsk)); 1609 sk_node_init(&newsk->sk_node); 1610 sock_lock_init(newsk); 1611 bh_lock_sock(newsk); 1612 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL; 1613 newsk->sk_backlog.len = 0; 1614 1615 atomic_set(&newsk->sk_rmem_alloc, 0); 1616 /* 1617 * sk_wmem_alloc set to one (see sk_free() and sock_wfree()) 1618 */ 1619 atomic_set(&newsk->sk_wmem_alloc, 1); 1620 atomic_set(&newsk->sk_omem_alloc, 0); 1621 sk_init_common(newsk); 1622 1623 newsk->sk_dst_cache = NULL; 1624 newsk->sk_dst_pending_confirm = 0; 1625 newsk->sk_wmem_queued = 0; 1626 newsk->sk_forward_alloc = 0; 1627 atomic_set(&newsk->sk_drops, 0); 1628 newsk->sk_send_head = NULL; 1629 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK; 1630 1631 sock_reset_flag(newsk, SOCK_DONE); 1632 1633 filter = rcu_dereference_protected(newsk->sk_filter, 1); 1634 if (filter != NULL) 1635 /* though it's an empty new sock, the charging may fail 1636 * if sysctl_optmem_max was changed between creation of 1637 * original socket and cloning 1638 */ 1639 is_charged = sk_filter_charge(newsk, filter); 1640 1641 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) { 1642 /* We need to make sure that we don't uncharge the new 1643 * socket if we couldn't charge it in the first place 1644 * as otherwise we uncharge the parent's filter. 1645 */ 1646 if (!is_charged) 1647 RCU_INIT_POINTER(newsk->sk_filter, NULL); 1648 sk_free_unlock_clone(newsk); 1649 newsk = NULL; 1650 goto out; 1651 } 1652 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL); 1653 1654 newsk->sk_err = 0; 1655 newsk->sk_err_soft = 0; 1656 newsk->sk_priority = 0; 1657 newsk->sk_incoming_cpu = raw_smp_processor_id(); 1658 atomic64_set(&newsk->sk_cookie, 0); 1659 1660 mem_cgroup_sk_alloc(newsk); 1661 cgroup_sk_alloc(&newsk->sk_cgrp_data); 1662 1663 /* 1664 * Before updating sk_refcnt, we must commit prior changes to memory 1665 * (Documentation/RCU/rculist_nulls.txt for details) 1666 */ 1667 smp_wmb(); 1668 atomic_set(&newsk->sk_refcnt, 2); 1669 1670 /* 1671 * Increment the counter in the same struct proto as the master 1672 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that 1673 * is the same as sk->sk_prot->socks, as this field was copied 1674 * with memcpy). 1675 * 1676 * This _changes_ the previous behaviour, where 1677 * tcp_create_openreq_child always was incrementing the 1678 * equivalent to tcp_prot->socks (inet_sock_nr), so this have 1679 * to be taken into account in all callers. -acme 1680 */ 1681 sk_refcnt_debug_inc(newsk); 1682 sk_set_socket(newsk, NULL); 1683 newsk->sk_wq = NULL; 1684 1685 if (newsk->sk_prot->sockets_allocated) 1686 sk_sockets_allocated_inc(newsk); 1687 1688 if (sock_needs_netstamp(sk) && 1689 newsk->sk_flags & SK_FLAGS_TIMESTAMP) 1690 net_enable_timestamp(); 1691 } 1692 out: 1693 return newsk; 1694 } 1695 EXPORT_SYMBOL_GPL(sk_clone_lock); 1696 1697 void sk_free_unlock_clone(struct sock *sk) 1698 { 1699 /* It is still raw copy of parent, so invalidate 1700 * destructor and make plain sk_free() */ 1701 sk->sk_destruct = NULL; 1702 bh_unlock_sock(sk); 1703 sk_free(sk); 1704 } 1705 EXPORT_SYMBOL_GPL(sk_free_unlock_clone); 1706 1707 void sk_setup_caps(struct sock *sk, struct dst_entry *dst) 1708 { 1709 u32 max_segs = 1; 1710 1711 sk_dst_set(sk, dst); 1712 sk->sk_route_caps = dst->dev->features; 1713 if (sk->sk_route_caps & NETIF_F_GSO) 1714 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE; 1715 sk->sk_route_caps &= ~sk->sk_route_nocaps; 1716 if (sk_can_gso(sk)) { 1717 if (dst->header_len) { 1718 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 1719 } else { 1720 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM; 1721 sk->sk_gso_max_size = dst->dev->gso_max_size; 1722 max_segs = max_t(u32, dst->dev->gso_max_segs, 1); 1723 } 1724 } 1725 sk->sk_gso_max_segs = max_segs; 1726 } 1727 EXPORT_SYMBOL_GPL(sk_setup_caps); 1728 1729 /* 1730 * Simple resource managers for sockets. 1731 */ 1732 1733 1734 /* 1735 * Write buffer destructor automatically called from kfree_skb. 1736 */ 1737 void sock_wfree(struct sk_buff *skb) 1738 { 1739 struct sock *sk = skb->sk; 1740 unsigned int len = skb->truesize; 1741 1742 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) { 1743 /* 1744 * Keep a reference on sk_wmem_alloc, this will be released 1745 * after sk_write_space() call 1746 */ 1747 atomic_sub(len - 1, &sk->sk_wmem_alloc); 1748 sk->sk_write_space(sk); 1749 len = 1; 1750 } 1751 /* 1752 * if sk_wmem_alloc reaches 0, we must finish what sk_free() 1753 * could not do because of in-flight packets 1754 */ 1755 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc)) 1756 __sk_free(sk); 1757 } 1758 EXPORT_SYMBOL(sock_wfree); 1759 1760 /* This variant of sock_wfree() is used by TCP, 1761 * since it sets SOCK_USE_WRITE_QUEUE. 1762 */ 1763 void __sock_wfree(struct sk_buff *skb) 1764 { 1765 struct sock *sk = skb->sk; 1766 1767 if (atomic_sub_and_test(skb->truesize, &sk->sk_wmem_alloc)) 1768 __sk_free(sk); 1769 } 1770 1771 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk) 1772 { 1773 skb_orphan(skb); 1774 skb->sk = sk; 1775 #ifdef CONFIG_INET 1776 if (unlikely(!sk_fullsock(sk))) { 1777 skb->destructor = sock_edemux; 1778 sock_hold(sk); 1779 return; 1780 } 1781 #endif 1782 skb->destructor = sock_wfree; 1783 skb_set_hash_from_sk(skb, sk); 1784 /* 1785 * We used to take a refcount on sk, but following operation 1786 * is enough to guarantee sk_free() wont free this sock until 1787 * all in-flight packets are completed 1788 */ 1789 atomic_add(skb->truesize, &sk->sk_wmem_alloc); 1790 } 1791 EXPORT_SYMBOL(skb_set_owner_w); 1792 1793 /* This helper is used by netem, as it can hold packets in its 1794 * delay queue. We want to allow the owner socket to send more 1795 * packets, as if they were already TX completed by a typical driver. 1796 * But we also want to keep skb->sk set because some packet schedulers 1797 * rely on it (sch_fq for example). So we set skb->truesize to a small 1798 * amount (1) and decrease sk_wmem_alloc accordingly. 1799 */ 1800 void skb_orphan_partial(struct sk_buff *skb) 1801 { 1802 /* If this skb is a TCP pure ACK or already went here, 1803 * we have nothing to do. 2 is already a very small truesize. 1804 */ 1805 if (skb->truesize <= 2) 1806 return; 1807 1808 /* TCP stack sets skb->ooo_okay based on sk_wmem_alloc, 1809 * so we do not completely orphan skb, but transfert all 1810 * accounted bytes but one, to avoid unexpected reorders. 1811 */ 1812 if (skb->destructor == sock_wfree 1813 #ifdef CONFIG_INET 1814 || skb->destructor == tcp_wfree 1815 #endif 1816 ) { 1817 atomic_sub(skb->truesize - 1, &skb->sk->sk_wmem_alloc); 1818 skb->truesize = 1; 1819 } else { 1820 skb_orphan(skb); 1821 } 1822 } 1823 EXPORT_SYMBOL(skb_orphan_partial); 1824 1825 /* 1826 * Read buffer destructor automatically called from kfree_skb. 1827 */ 1828 void sock_rfree(struct sk_buff *skb) 1829 { 1830 struct sock *sk = skb->sk; 1831 unsigned int len = skb->truesize; 1832 1833 atomic_sub(len, &sk->sk_rmem_alloc); 1834 sk_mem_uncharge(sk, len); 1835 } 1836 EXPORT_SYMBOL(sock_rfree); 1837 1838 /* 1839 * Buffer destructor for skbs that are not used directly in read or write 1840 * path, e.g. for error handler skbs. Automatically called from kfree_skb. 1841 */ 1842 void sock_efree(struct sk_buff *skb) 1843 { 1844 sock_put(skb->sk); 1845 } 1846 EXPORT_SYMBOL(sock_efree); 1847 1848 kuid_t sock_i_uid(struct sock *sk) 1849 { 1850 kuid_t uid; 1851 1852 read_lock_bh(&sk->sk_callback_lock); 1853 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID; 1854 read_unlock_bh(&sk->sk_callback_lock); 1855 return uid; 1856 } 1857 EXPORT_SYMBOL(sock_i_uid); 1858 1859 unsigned long sock_i_ino(struct sock *sk) 1860 { 1861 unsigned long ino; 1862 1863 read_lock_bh(&sk->sk_callback_lock); 1864 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0; 1865 read_unlock_bh(&sk->sk_callback_lock); 1866 return ino; 1867 } 1868 EXPORT_SYMBOL(sock_i_ino); 1869 1870 /* 1871 * Allocate a skb from the socket's send buffer. 1872 */ 1873 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 1874 gfp_t priority) 1875 { 1876 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) { 1877 struct sk_buff *skb = alloc_skb(size, priority); 1878 if (skb) { 1879 skb_set_owner_w(skb, sk); 1880 return skb; 1881 } 1882 } 1883 return NULL; 1884 } 1885 EXPORT_SYMBOL(sock_wmalloc); 1886 1887 /* 1888 * Allocate a memory block from the socket's option memory buffer. 1889 */ 1890 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority) 1891 { 1892 if ((unsigned int)size <= sysctl_optmem_max && 1893 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) { 1894 void *mem; 1895 /* First do the add, to avoid the race if kmalloc 1896 * might sleep. 1897 */ 1898 atomic_add(size, &sk->sk_omem_alloc); 1899 mem = kmalloc(size, priority); 1900 if (mem) 1901 return mem; 1902 atomic_sub(size, &sk->sk_omem_alloc); 1903 } 1904 return NULL; 1905 } 1906 EXPORT_SYMBOL(sock_kmalloc); 1907 1908 /* Free an option memory block. Note, we actually want the inline 1909 * here as this allows gcc to detect the nullify and fold away the 1910 * condition entirely. 1911 */ 1912 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size, 1913 const bool nullify) 1914 { 1915 if (WARN_ON_ONCE(!mem)) 1916 return; 1917 if (nullify) 1918 kzfree(mem); 1919 else 1920 kfree(mem); 1921 atomic_sub(size, &sk->sk_omem_alloc); 1922 } 1923 1924 void sock_kfree_s(struct sock *sk, void *mem, int size) 1925 { 1926 __sock_kfree_s(sk, mem, size, false); 1927 } 1928 EXPORT_SYMBOL(sock_kfree_s); 1929 1930 void sock_kzfree_s(struct sock *sk, void *mem, int size) 1931 { 1932 __sock_kfree_s(sk, mem, size, true); 1933 } 1934 EXPORT_SYMBOL(sock_kzfree_s); 1935 1936 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock. 1937 I think, these locks should be removed for datagram sockets. 1938 */ 1939 static long sock_wait_for_wmem(struct sock *sk, long timeo) 1940 { 1941 DEFINE_WAIT(wait); 1942 1943 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk); 1944 for (;;) { 1945 if (!timeo) 1946 break; 1947 if (signal_pending(current)) 1948 break; 1949 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 1950 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 1951 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) 1952 break; 1953 if (sk->sk_shutdown & SEND_SHUTDOWN) 1954 break; 1955 if (sk->sk_err) 1956 break; 1957 timeo = schedule_timeout(timeo); 1958 } 1959 finish_wait(sk_sleep(sk), &wait); 1960 return timeo; 1961 } 1962 1963 1964 /* 1965 * Generic send/receive buffer handlers 1966 */ 1967 1968 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 1969 unsigned long data_len, int noblock, 1970 int *errcode, int max_page_order) 1971 { 1972 struct sk_buff *skb; 1973 long timeo; 1974 int err; 1975 1976 timeo = sock_sndtimeo(sk, noblock); 1977 for (;;) { 1978 err = sock_error(sk); 1979 if (err != 0) 1980 goto failure; 1981 1982 err = -EPIPE; 1983 if (sk->sk_shutdown & SEND_SHUTDOWN) 1984 goto failure; 1985 1986 if (sk_wmem_alloc_get(sk) < sk->sk_sndbuf) 1987 break; 1988 1989 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk); 1990 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 1991 err = -EAGAIN; 1992 if (!timeo) 1993 goto failure; 1994 if (signal_pending(current)) 1995 goto interrupted; 1996 timeo = sock_wait_for_wmem(sk, timeo); 1997 } 1998 skb = alloc_skb_with_frags(header_len, data_len, max_page_order, 1999 errcode, sk->sk_allocation); 2000 if (skb) 2001 skb_set_owner_w(skb, sk); 2002 return skb; 2003 2004 interrupted: 2005 err = sock_intr_errno(timeo); 2006 failure: 2007 *errcode = err; 2008 return NULL; 2009 } 2010 EXPORT_SYMBOL(sock_alloc_send_pskb); 2011 2012 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size, 2013 int noblock, int *errcode) 2014 { 2015 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0); 2016 } 2017 EXPORT_SYMBOL(sock_alloc_send_skb); 2018 2019 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg, 2020 struct sockcm_cookie *sockc) 2021 { 2022 u32 tsflags; 2023 2024 switch (cmsg->cmsg_type) { 2025 case SO_MARK: 2026 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) 2027 return -EPERM; 2028 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 2029 return -EINVAL; 2030 sockc->mark = *(u32 *)CMSG_DATA(cmsg); 2031 break; 2032 case SO_TIMESTAMPING: 2033 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 2034 return -EINVAL; 2035 2036 tsflags = *(u32 *)CMSG_DATA(cmsg); 2037 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK) 2038 return -EINVAL; 2039 2040 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK; 2041 sockc->tsflags |= tsflags; 2042 break; 2043 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */ 2044 case SCM_RIGHTS: 2045 case SCM_CREDENTIALS: 2046 break; 2047 default: 2048 return -EINVAL; 2049 } 2050 return 0; 2051 } 2052 EXPORT_SYMBOL(__sock_cmsg_send); 2053 2054 int sock_cmsg_send(struct sock *sk, struct msghdr *msg, 2055 struct sockcm_cookie *sockc) 2056 { 2057 struct cmsghdr *cmsg; 2058 int ret; 2059 2060 for_each_cmsghdr(cmsg, msg) { 2061 if (!CMSG_OK(msg, cmsg)) 2062 return -EINVAL; 2063 if (cmsg->cmsg_level != SOL_SOCKET) 2064 continue; 2065 ret = __sock_cmsg_send(sk, msg, cmsg, sockc); 2066 if (ret) 2067 return ret; 2068 } 2069 return 0; 2070 } 2071 EXPORT_SYMBOL(sock_cmsg_send); 2072 2073 /* On 32bit arches, an skb frag is limited to 2^15 */ 2074 #define SKB_FRAG_PAGE_ORDER get_order(32768) 2075 2076 /** 2077 * skb_page_frag_refill - check that a page_frag contains enough room 2078 * @sz: minimum size of the fragment we want to get 2079 * @pfrag: pointer to page_frag 2080 * @gfp: priority for memory allocation 2081 * 2082 * Note: While this allocator tries to use high order pages, there is 2083 * no guarantee that allocations succeed. Therefore, @sz MUST be 2084 * less or equal than PAGE_SIZE. 2085 */ 2086 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp) 2087 { 2088 if (pfrag->page) { 2089 if (page_ref_count(pfrag->page) == 1) { 2090 pfrag->offset = 0; 2091 return true; 2092 } 2093 if (pfrag->offset + sz <= pfrag->size) 2094 return true; 2095 put_page(pfrag->page); 2096 } 2097 2098 pfrag->offset = 0; 2099 if (SKB_FRAG_PAGE_ORDER) { 2100 /* Avoid direct reclaim but allow kswapd to wake */ 2101 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) | 2102 __GFP_COMP | __GFP_NOWARN | 2103 __GFP_NORETRY, 2104 SKB_FRAG_PAGE_ORDER); 2105 if (likely(pfrag->page)) { 2106 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER; 2107 return true; 2108 } 2109 } 2110 pfrag->page = alloc_page(gfp); 2111 if (likely(pfrag->page)) { 2112 pfrag->size = PAGE_SIZE; 2113 return true; 2114 } 2115 return false; 2116 } 2117 EXPORT_SYMBOL(skb_page_frag_refill); 2118 2119 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag) 2120 { 2121 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation))) 2122 return true; 2123 2124 sk_enter_memory_pressure(sk); 2125 sk_stream_moderate_sndbuf(sk); 2126 return false; 2127 } 2128 EXPORT_SYMBOL(sk_page_frag_refill); 2129 2130 static void __lock_sock(struct sock *sk) 2131 __releases(&sk->sk_lock.slock) 2132 __acquires(&sk->sk_lock.slock) 2133 { 2134 DEFINE_WAIT(wait); 2135 2136 for (;;) { 2137 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait, 2138 TASK_UNINTERRUPTIBLE); 2139 spin_unlock_bh(&sk->sk_lock.slock); 2140 schedule(); 2141 spin_lock_bh(&sk->sk_lock.slock); 2142 if (!sock_owned_by_user(sk)) 2143 break; 2144 } 2145 finish_wait(&sk->sk_lock.wq, &wait); 2146 } 2147 2148 static void __release_sock(struct sock *sk) 2149 __releases(&sk->sk_lock.slock) 2150 __acquires(&sk->sk_lock.slock) 2151 { 2152 struct sk_buff *skb, *next; 2153 2154 while ((skb = sk->sk_backlog.head) != NULL) { 2155 sk->sk_backlog.head = sk->sk_backlog.tail = NULL; 2156 2157 spin_unlock_bh(&sk->sk_lock.slock); 2158 2159 do { 2160 next = skb->next; 2161 prefetch(next); 2162 WARN_ON_ONCE(skb_dst_is_noref(skb)); 2163 skb->next = NULL; 2164 sk_backlog_rcv(sk, skb); 2165 2166 cond_resched(); 2167 2168 skb = next; 2169 } while (skb != NULL); 2170 2171 spin_lock_bh(&sk->sk_lock.slock); 2172 } 2173 2174 /* 2175 * Doing the zeroing here guarantee we can not loop forever 2176 * while a wild producer attempts to flood us. 2177 */ 2178 sk->sk_backlog.len = 0; 2179 } 2180 2181 void __sk_flush_backlog(struct sock *sk) 2182 { 2183 spin_lock_bh(&sk->sk_lock.slock); 2184 __release_sock(sk); 2185 spin_unlock_bh(&sk->sk_lock.slock); 2186 } 2187 2188 /** 2189 * sk_wait_data - wait for data to arrive at sk_receive_queue 2190 * @sk: sock to wait on 2191 * @timeo: for how long 2192 * @skb: last skb seen on sk_receive_queue 2193 * 2194 * Now socket state including sk->sk_err is changed only under lock, 2195 * hence we may omit checks after joining wait queue. 2196 * We check receive queue before schedule() only as optimization; 2197 * it is very likely that release_sock() added new data. 2198 */ 2199 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb) 2200 { 2201 DEFINE_WAIT_FUNC(wait, woken_wake_function); 2202 int rc; 2203 2204 add_wait_queue(sk_sleep(sk), &wait); 2205 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk); 2206 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait); 2207 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk); 2208 remove_wait_queue(sk_sleep(sk), &wait); 2209 return rc; 2210 } 2211 EXPORT_SYMBOL(sk_wait_data); 2212 2213 /** 2214 * __sk_mem_raise_allocated - increase memory_allocated 2215 * @sk: socket 2216 * @size: memory size to allocate 2217 * @amt: pages to allocate 2218 * @kind: allocation type 2219 * 2220 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc 2221 */ 2222 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind) 2223 { 2224 struct proto *prot = sk->sk_prot; 2225 long allocated = sk_memory_allocated_add(sk, amt); 2226 2227 if (mem_cgroup_sockets_enabled && sk->sk_memcg && 2228 !mem_cgroup_charge_skmem(sk->sk_memcg, amt)) 2229 goto suppress_allocation; 2230 2231 /* Under limit. */ 2232 if (allocated <= sk_prot_mem_limits(sk, 0)) { 2233 sk_leave_memory_pressure(sk); 2234 return 1; 2235 } 2236 2237 /* Under pressure. */ 2238 if (allocated > sk_prot_mem_limits(sk, 1)) 2239 sk_enter_memory_pressure(sk); 2240 2241 /* Over hard limit. */ 2242 if (allocated > sk_prot_mem_limits(sk, 2)) 2243 goto suppress_allocation; 2244 2245 /* guarantee minimum buffer size under pressure */ 2246 if (kind == SK_MEM_RECV) { 2247 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0]) 2248 return 1; 2249 2250 } else { /* SK_MEM_SEND */ 2251 if (sk->sk_type == SOCK_STREAM) { 2252 if (sk->sk_wmem_queued < prot->sysctl_wmem[0]) 2253 return 1; 2254 } else if (atomic_read(&sk->sk_wmem_alloc) < 2255 prot->sysctl_wmem[0]) 2256 return 1; 2257 } 2258 2259 if (sk_has_memory_pressure(sk)) { 2260 int alloc; 2261 2262 if (!sk_under_memory_pressure(sk)) 2263 return 1; 2264 alloc = sk_sockets_allocated_read_positive(sk); 2265 if (sk_prot_mem_limits(sk, 2) > alloc * 2266 sk_mem_pages(sk->sk_wmem_queued + 2267 atomic_read(&sk->sk_rmem_alloc) + 2268 sk->sk_forward_alloc)) 2269 return 1; 2270 } 2271 2272 suppress_allocation: 2273 2274 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) { 2275 sk_stream_moderate_sndbuf(sk); 2276 2277 /* Fail only if socket is _under_ its sndbuf. 2278 * In this case we cannot block, so that we have to fail. 2279 */ 2280 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf) 2281 return 1; 2282 } 2283 2284 trace_sock_exceed_buf_limit(sk, prot, allocated); 2285 2286 sk_memory_allocated_sub(sk, amt); 2287 2288 if (mem_cgroup_sockets_enabled && sk->sk_memcg) 2289 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt); 2290 2291 return 0; 2292 } 2293 EXPORT_SYMBOL(__sk_mem_raise_allocated); 2294 2295 /** 2296 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated 2297 * @sk: socket 2298 * @size: memory size to allocate 2299 * @kind: allocation type 2300 * 2301 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means 2302 * rmem allocation. This function assumes that protocols which have 2303 * memory_pressure use sk_wmem_queued as write buffer accounting. 2304 */ 2305 int __sk_mem_schedule(struct sock *sk, int size, int kind) 2306 { 2307 int ret, amt = sk_mem_pages(size); 2308 2309 sk->sk_forward_alloc += amt << SK_MEM_QUANTUM_SHIFT; 2310 ret = __sk_mem_raise_allocated(sk, size, amt, kind); 2311 if (!ret) 2312 sk->sk_forward_alloc -= amt << SK_MEM_QUANTUM_SHIFT; 2313 return ret; 2314 } 2315 EXPORT_SYMBOL(__sk_mem_schedule); 2316 2317 /** 2318 * __sk_mem_reduce_allocated - reclaim memory_allocated 2319 * @sk: socket 2320 * @amount: number of quanta 2321 * 2322 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc 2323 */ 2324 void __sk_mem_reduce_allocated(struct sock *sk, int amount) 2325 { 2326 sk_memory_allocated_sub(sk, amount); 2327 2328 if (mem_cgroup_sockets_enabled && sk->sk_memcg) 2329 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount); 2330 2331 if (sk_under_memory_pressure(sk) && 2332 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0))) 2333 sk_leave_memory_pressure(sk); 2334 } 2335 EXPORT_SYMBOL(__sk_mem_reduce_allocated); 2336 2337 /** 2338 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated 2339 * @sk: socket 2340 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple) 2341 */ 2342 void __sk_mem_reclaim(struct sock *sk, int amount) 2343 { 2344 amount >>= SK_MEM_QUANTUM_SHIFT; 2345 sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT; 2346 __sk_mem_reduce_allocated(sk, amount); 2347 } 2348 EXPORT_SYMBOL(__sk_mem_reclaim); 2349 2350 int sk_set_peek_off(struct sock *sk, int val) 2351 { 2352 if (val < 0) 2353 return -EINVAL; 2354 2355 sk->sk_peek_off = val; 2356 return 0; 2357 } 2358 EXPORT_SYMBOL_GPL(sk_set_peek_off); 2359 2360 /* 2361 * Set of default routines for initialising struct proto_ops when 2362 * the protocol does not support a particular function. In certain 2363 * cases where it makes no sense for a protocol to have a "do nothing" 2364 * function, some default processing is provided. 2365 */ 2366 2367 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len) 2368 { 2369 return -EOPNOTSUPP; 2370 } 2371 EXPORT_SYMBOL(sock_no_bind); 2372 2373 int sock_no_connect(struct socket *sock, struct sockaddr *saddr, 2374 int len, int flags) 2375 { 2376 return -EOPNOTSUPP; 2377 } 2378 EXPORT_SYMBOL(sock_no_connect); 2379 2380 int sock_no_socketpair(struct socket *sock1, struct socket *sock2) 2381 { 2382 return -EOPNOTSUPP; 2383 } 2384 EXPORT_SYMBOL(sock_no_socketpair); 2385 2386 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags, 2387 bool kern) 2388 { 2389 return -EOPNOTSUPP; 2390 } 2391 EXPORT_SYMBOL(sock_no_accept); 2392 2393 int sock_no_getname(struct socket *sock, struct sockaddr *saddr, 2394 int *len, int peer) 2395 { 2396 return -EOPNOTSUPP; 2397 } 2398 EXPORT_SYMBOL(sock_no_getname); 2399 2400 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt) 2401 { 2402 return 0; 2403 } 2404 EXPORT_SYMBOL(sock_no_poll); 2405 2406 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 2407 { 2408 return -EOPNOTSUPP; 2409 } 2410 EXPORT_SYMBOL(sock_no_ioctl); 2411 2412 int sock_no_listen(struct socket *sock, int backlog) 2413 { 2414 return -EOPNOTSUPP; 2415 } 2416 EXPORT_SYMBOL(sock_no_listen); 2417 2418 int sock_no_shutdown(struct socket *sock, int how) 2419 { 2420 return -EOPNOTSUPP; 2421 } 2422 EXPORT_SYMBOL(sock_no_shutdown); 2423 2424 int sock_no_setsockopt(struct socket *sock, int level, int optname, 2425 char __user *optval, unsigned int optlen) 2426 { 2427 return -EOPNOTSUPP; 2428 } 2429 EXPORT_SYMBOL(sock_no_setsockopt); 2430 2431 int sock_no_getsockopt(struct socket *sock, int level, int optname, 2432 char __user *optval, int __user *optlen) 2433 { 2434 return -EOPNOTSUPP; 2435 } 2436 EXPORT_SYMBOL(sock_no_getsockopt); 2437 2438 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len) 2439 { 2440 return -EOPNOTSUPP; 2441 } 2442 EXPORT_SYMBOL(sock_no_sendmsg); 2443 2444 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len, 2445 int flags) 2446 { 2447 return -EOPNOTSUPP; 2448 } 2449 EXPORT_SYMBOL(sock_no_recvmsg); 2450 2451 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma) 2452 { 2453 /* Mirror missing mmap method error code */ 2454 return -ENODEV; 2455 } 2456 EXPORT_SYMBOL(sock_no_mmap); 2457 2458 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags) 2459 { 2460 ssize_t res; 2461 struct msghdr msg = {.msg_flags = flags}; 2462 struct kvec iov; 2463 char *kaddr = kmap(page); 2464 iov.iov_base = kaddr + offset; 2465 iov.iov_len = size; 2466 res = kernel_sendmsg(sock, &msg, &iov, 1, size); 2467 kunmap(page); 2468 return res; 2469 } 2470 EXPORT_SYMBOL(sock_no_sendpage); 2471 2472 /* 2473 * Default Socket Callbacks 2474 */ 2475 2476 static void sock_def_wakeup(struct sock *sk) 2477 { 2478 struct socket_wq *wq; 2479 2480 rcu_read_lock(); 2481 wq = rcu_dereference(sk->sk_wq); 2482 if (skwq_has_sleeper(wq)) 2483 wake_up_interruptible_all(&wq->wait); 2484 rcu_read_unlock(); 2485 } 2486 2487 static void sock_def_error_report(struct sock *sk) 2488 { 2489 struct socket_wq *wq; 2490 2491 rcu_read_lock(); 2492 wq = rcu_dereference(sk->sk_wq); 2493 if (skwq_has_sleeper(wq)) 2494 wake_up_interruptible_poll(&wq->wait, POLLERR); 2495 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR); 2496 rcu_read_unlock(); 2497 } 2498 2499 static void sock_def_readable(struct sock *sk) 2500 { 2501 struct socket_wq *wq; 2502 2503 rcu_read_lock(); 2504 wq = rcu_dereference(sk->sk_wq); 2505 if (skwq_has_sleeper(wq)) 2506 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI | 2507 POLLRDNORM | POLLRDBAND); 2508 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN); 2509 rcu_read_unlock(); 2510 } 2511 2512 static void sock_def_write_space(struct sock *sk) 2513 { 2514 struct socket_wq *wq; 2515 2516 rcu_read_lock(); 2517 2518 /* Do not wake up a writer until he can make "significant" 2519 * progress. --DaveM 2520 */ 2521 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) { 2522 wq = rcu_dereference(sk->sk_wq); 2523 if (skwq_has_sleeper(wq)) 2524 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT | 2525 POLLWRNORM | POLLWRBAND); 2526 2527 /* Should agree with poll, otherwise some programs break */ 2528 if (sock_writeable(sk)) 2529 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT); 2530 } 2531 2532 rcu_read_unlock(); 2533 } 2534 2535 static void sock_def_destruct(struct sock *sk) 2536 { 2537 } 2538 2539 void sk_send_sigurg(struct sock *sk) 2540 { 2541 if (sk->sk_socket && sk->sk_socket->file) 2542 if (send_sigurg(&sk->sk_socket->file->f_owner)) 2543 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI); 2544 } 2545 EXPORT_SYMBOL(sk_send_sigurg); 2546 2547 void sk_reset_timer(struct sock *sk, struct timer_list* timer, 2548 unsigned long expires) 2549 { 2550 if (!mod_timer(timer, expires)) 2551 sock_hold(sk); 2552 } 2553 EXPORT_SYMBOL(sk_reset_timer); 2554 2555 void sk_stop_timer(struct sock *sk, struct timer_list* timer) 2556 { 2557 if (del_timer(timer)) 2558 __sock_put(sk); 2559 } 2560 EXPORT_SYMBOL(sk_stop_timer); 2561 2562 void sock_init_data(struct socket *sock, struct sock *sk) 2563 { 2564 sk_init_common(sk); 2565 sk->sk_send_head = NULL; 2566 2567 init_timer(&sk->sk_timer); 2568 2569 sk->sk_allocation = GFP_KERNEL; 2570 sk->sk_rcvbuf = sysctl_rmem_default; 2571 sk->sk_sndbuf = sysctl_wmem_default; 2572 sk->sk_state = TCP_CLOSE; 2573 sk_set_socket(sk, sock); 2574 2575 sock_set_flag(sk, SOCK_ZAPPED); 2576 2577 if (sock) { 2578 sk->sk_type = sock->type; 2579 sk->sk_wq = sock->wq; 2580 sock->sk = sk; 2581 sk->sk_uid = SOCK_INODE(sock)->i_uid; 2582 } else { 2583 sk->sk_wq = NULL; 2584 sk->sk_uid = make_kuid(sock_net(sk)->user_ns, 0); 2585 } 2586 2587 rwlock_init(&sk->sk_callback_lock); 2588 if (sk->sk_kern_sock) 2589 lockdep_set_class_and_name( 2590 &sk->sk_callback_lock, 2591 af_kern_callback_keys + sk->sk_family, 2592 af_family_kern_clock_key_strings[sk->sk_family]); 2593 else 2594 lockdep_set_class_and_name( 2595 &sk->sk_callback_lock, 2596 af_callback_keys + sk->sk_family, 2597 af_family_clock_key_strings[sk->sk_family]); 2598 2599 sk->sk_state_change = sock_def_wakeup; 2600 sk->sk_data_ready = sock_def_readable; 2601 sk->sk_write_space = sock_def_write_space; 2602 sk->sk_error_report = sock_def_error_report; 2603 sk->sk_destruct = sock_def_destruct; 2604 2605 sk->sk_frag.page = NULL; 2606 sk->sk_frag.offset = 0; 2607 sk->sk_peek_off = -1; 2608 2609 sk->sk_peer_pid = NULL; 2610 sk->sk_peer_cred = NULL; 2611 sk->sk_write_pending = 0; 2612 sk->sk_rcvlowat = 1; 2613 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; 2614 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT; 2615 2616 sk->sk_stamp = SK_DEFAULT_STAMP; 2617 2618 #ifdef CONFIG_NET_RX_BUSY_POLL 2619 sk->sk_napi_id = 0; 2620 sk->sk_ll_usec = sysctl_net_busy_read; 2621 #endif 2622 2623 sk->sk_max_pacing_rate = ~0U; 2624 sk->sk_pacing_rate = ~0U; 2625 sk->sk_incoming_cpu = -1; 2626 /* 2627 * Before updating sk_refcnt, we must commit prior changes to memory 2628 * (Documentation/RCU/rculist_nulls.txt for details) 2629 */ 2630 smp_wmb(); 2631 atomic_set(&sk->sk_refcnt, 1); 2632 atomic_set(&sk->sk_drops, 0); 2633 } 2634 EXPORT_SYMBOL(sock_init_data); 2635 2636 void lock_sock_nested(struct sock *sk, int subclass) 2637 { 2638 might_sleep(); 2639 spin_lock_bh(&sk->sk_lock.slock); 2640 if (sk->sk_lock.owned) 2641 __lock_sock(sk); 2642 sk->sk_lock.owned = 1; 2643 spin_unlock(&sk->sk_lock.slock); 2644 /* 2645 * The sk_lock has mutex_lock() semantics here: 2646 */ 2647 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_); 2648 local_bh_enable(); 2649 } 2650 EXPORT_SYMBOL(lock_sock_nested); 2651 2652 void release_sock(struct sock *sk) 2653 { 2654 spin_lock_bh(&sk->sk_lock.slock); 2655 if (sk->sk_backlog.tail) 2656 __release_sock(sk); 2657 2658 /* Warning : release_cb() might need to release sk ownership, 2659 * ie call sock_release_ownership(sk) before us. 2660 */ 2661 if (sk->sk_prot->release_cb) 2662 sk->sk_prot->release_cb(sk); 2663 2664 sock_release_ownership(sk); 2665 if (waitqueue_active(&sk->sk_lock.wq)) 2666 wake_up(&sk->sk_lock.wq); 2667 spin_unlock_bh(&sk->sk_lock.slock); 2668 } 2669 EXPORT_SYMBOL(release_sock); 2670 2671 /** 2672 * lock_sock_fast - fast version of lock_sock 2673 * @sk: socket 2674 * 2675 * This version should be used for very small section, where process wont block 2676 * return false if fast path is taken 2677 * sk_lock.slock locked, owned = 0, BH disabled 2678 * return true if slow path is taken 2679 * sk_lock.slock unlocked, owned = 1, BH enabled 2680 */ 2681 bool lock_sock_fast(struct sock *sk) 2682 { 2683 might_sleep(); 2684 spin_lock_bh(&sk->sk_lock.slock); 2685 2686 if (!sk->sk_lock.owned) 2687 /* 2688 * Note : We must disable BH 2689 */ 2690 return false; 2691 2692 __lock_sock(sk); 2693 sk->sk_lock.owned = 1; 2694 spin_unlock(&sk->sk_lock.slock); 2695 /* 2696 * The sk_lock has mutex_lock() semantics here: 2697 */ 2698 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_); 2699 local_bh_enable(); 2700 return true; 2701 } 2702 EXPORT_SYMBOL(lock_sock_fast); 2703 2704 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp) 2705 { 2706 struct timeval tv; 2707 if (!sock_flag(sk, SOCK_TIMESTAMP)) 2708 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 2709 tv = ktime_to_timeval(sk->sk_stamp); 2710 if (tv.tv_sec == -1) 2711 return -ENOENT; 2712 if (tv.tv_sec == 0) { 2713 sk->sk_stamp = ktime_get_real(); 2714 tv = ktime_to_timeval(sk->sk_stamp); 2715 } 2716 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0; 2717 } 2718 EXPORT_SYMBOL(sock_get_timestamp); 2719 2720 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp) 2721 { 2722 struct timespec ts; 2723 if (!sock_flag(sk, SOCK_TIMESTAMP)) 2724 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 2725 ts = ktime_to_timespec(sk->sk_stamp); 2726 if (ts.tv_sec == -1) 2727 return -ENOENT; 2728 if (ts.tv_sec == 0) { 2729 sk->sk_stamp = ktime_get_real(); 2730 ts = ktime_to_timespec(sk->sk_stamp); 2731 } 2732 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0; 2733 } 2734 EXPORT_SYMBOL(sock_get_timestampns); 2735 2736 void sock_enable_timestamp(struct sock *sk, int flag) 2737 { 2738 if (!sock_flag(sk, flag)) { 2739 unsigned long previous_flags = sk->sk_flags; 2740 2741 sock_set_flag(sk, flag); 2742 /* 2743 * we just set one of the two flags which require net 2744 * time stamping, but time stamping might have been on 2745 * already because of the other one 2746 */ 2747 if (sock_needs_netstamp(sk) && 2748 !(previous_flags & SK_FLAGS_TIMESTAMP)) 2749 net_enable_timestamp(); 2750 } 2751 } 2752 2753 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, 2754 int level, int type) 2755 { 2756 struct sock_exterr_skb *serr; 2757 struct sk_buff *skb; 2758 int copied, err; 2759 2760 err = -EAGAIN; 2761 skb = sock_dequeue_err_skb(sk); 2762 if (skb == NULL) 2763 goto out; 2764 2765 copied = skb->len; 2766 if (copied > len) { 2767 msg->msg_flags |= MSG_TRUNC; 2768 copied = len; 2769 } 2770 err = skb_copy_datagram_msg(skb, 0, msg, copied); 2771 if (err) 2772 goto out_free_skb; 2773 2774 sock_recv_timestamp(msg, sk, skb); 2775 2776 serr = SKB_EXT_ERR(skb); 2777 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee); 2778 2779 msg->msg_flags |= MSG_ERRQUEUE; 2780 err = copied; 2781 2782 out_free_skb: 2783 kfree_skb(skb); 2784 out: 2785 return err; 2786 } 2787 EXPORT_SYMBOL(sock_recv_errqueue); 2788 2789 /* 2790 * Get a socket option on an socket. 2791 * 2792 * FIX: POSIX 1003.1g is very ambiguous here. It states that 2793 * asynchronous errors should be reported by getsockopt. We assume 2794 * this means if you specify SO_ERROR (otherwise whats the point of it). 2795 */ 2796 int sock_common_getsockopt(struct socket *sock, int level, int optname, 2797 char __user *optval, int __user *optlen) 2798 { 2799 struct sock *sk = sock->sk; 2800 2801 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen); 2802 } 2803 EXPORT_SYMBOL(sock_common_getsockopt); 2804 2805 #ifdef CONFIG_COMPAT 2806 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname, 2807 char __user *optval, int __user *optlen) 2808 { 2809 struct sock *sk = sock->sk; 2810 2811 if (sk->sk_prot->compat_getsockopt != NULL) 2812 return sk->sk_prot->compat_getsockopt(sk, level, optname, 2813 optval, optlen); 2814 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen); 2815 } 2816 EXPORT_SYMBOL(compat_sock_common_getsockopt); 2817 #endif 2818 2819 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 2820 int flags) 2821 { 2822 struct sock *sk = sock->sk; 2823 int addr_len = 0; 2824 int err; 2825 2826 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT, 2827 flags & ~MSG_DONTWAIT, &addr_len); 2828 if (err >= 0) 2829 msg->msg_namelen = addr_len; 2830 return err; 2831 } 2832 EXPORT_SYMBOL(sock_common_recvmsg); 2833 2834 /* 2835 * Set socket options on an inet socket. 2836 */ 2837 int sock_common_setsockopt(struct socket *sock, int level, int optname, 2838 char __user *optval, unsigned int optlen) 2839 { 2840 struct sock *sk = sock->sk; 2841 2842 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen); 2843 } 2844 EXPORT_SYMBOL(sock_common_setsockopt); 2845 2846 #ifdef CONFIG_COMPAT 2847 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname, 2848 char __user *optval, unsigned int optlen) 2849 { 2850 struct sock *sk = sock->sk; 2851 2852 if (sk->sk_prot->compat_setsockopt != NULL) 2853 return sk->sk_prot->compat_setsockopt(sk, level, optname, 2854 optval, optlen); 2855 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen); 2856 } 2857 EXPORT_SYMBOL(compat_sock_common_setsockopt); 2858 #endif 2859 2860 void sk_common_release(struct sock *sk) 2861 { 2862 if (sk->sk_prot->destroy) 2863 sk->sk_prot->destroy(sk); 2864 2865 /* 2866 * Observation: when sock_common_release is called, processes have 2867 * no access to socket. But net still has. 2868 * Step one, detach it from networking: 2869 * 2870 * A. Remove from hash tables. 2871 */ 2872 2873 sk->sk_prot->unhash(sk); 2874 2875 /* 2876 * In this point socket cannot receive new packets, but it is possible 2877 * that some packets are in flight because some CPU runs receiver and 2878 * did hash table lookup before we unhashed socket. They will achieve 2879 * receive queue and will be purged by socket destructor. 2880 * 2881 * Also we still have packets pending on receive queue and probably, 2882 * our own packets waiting in device queues. sock_destroy will drain 2883 * receive queue, but transmitted packets will delay socket destruction 2884 * until the last reference will be released. 2885 */ 2886 2887 sock_orphan(sk); 2888 2889 xfrm_sk_free_policy(sk); 2890 2891 sk_refcnt_debug_release(sk); 2892 2893 sock_put(sk); 2894 } 2895 EXPORT_SYMBOL(sk_common_release); 2896 2897 void sk_get_meminfo(const struct sock *sk, u32 *mem) 2898 { 2899 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS); 2900 2901 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk); 2902 mem[SK_MEMINFO_RCVBUF] = sk->sk_rcvbuf; 2903 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk); 2904 mem[SK_MEMINFO_SNDBUF] = sk->sk_sndbuf; 2905 mem[SK_MEMINFO_FWD_ALLOC] = sk->sk_forward_alloc; 2906 mem[SK_MEMINFO_WMEM_QUEUED] = sk->sk_wmem_queued; 2907 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc); 2908 mem[SK_MEMINFO_BACKLOG] = sk->sk_backlog.len; 2909 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops); 2910 } 2911 2912 #ifdef CONFIG_PROC_FS 2913 #define PROTO_INUSE_NR 64 /* should be enough for the first time */ 2914 struct prot_inuse { 2915 int val[PROTO_INUSE_NR]; 2916 }; 2917 2918 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR); 2919 2920 #ifdef CONFIG_NET_NS 2921 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val) 2922 { 2923 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val); 2924 } 2925 EXPORT_SYMBOL_GPL(sock_prot_inuse_add); 2926 2927 int sock_prot_inuse_get(struct net *net, struct proto *prot) 2928 { 2929 int cpu, idx = prot->inuse_idx; 2930 int res = 0; 2931 2932 for_each_possible_cpu(cpu) 2933 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx]; 2934 2935 return res >= 0 ? res : 0; 2936 } 2937 EXPORT_SYMBOL_GPL(sock_prot_inuse_get); 2938 2939 static int __net_init sock_inuse_init_net(struct net *net) 2940 { 2941 net->core.inuse = alloc_percpu(struct prot_inuse); 2942 return net->core.inuse ? 0 : -ENOMEM; 2943 } 2944 2945 static void __net_exit sock_inuse_exit_net(struct net *net) 2946 { 2947 free_percpu(net->core.inuse); 2948 } 2949 2950 static struct pernet_operations net_inuse_ops = { 2951 .init = sock_inuse_init_net, 2952 .exit = sock_inuse_exit_net, 2953 }; 2954 2955 static __init int net_inuse_init(void) 2956 { 2957 if (register_pernet_subsys(&net_inuse_ops)) 2958 panic("Cannot initialize net inuse counters"); 2959 2960 return 0; 2961 } 2962 2963 core_initcall(net_inuse_init); 2964 #else 2965 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse); 2966 2967 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val) 2968 { 2969 __this_cpu_add(prot_inuse.val[prot->inuse_idx], val); 2970 } 2971 EXPORT_SYMBOL_GPL(sock_prot_inuse_add); 2972 2973 int sock_prot_inuse_get(struct net *net, struct proto *prot) 2974 { 2975 int cpu, idx = prot->inuse_idx; 2976 int res = 0; 2977 2978 for_each_possible_cpu(cpu) 2979 res += per_cpu(prot_inuse, cpu).val[idx]; 2980 2981 return res >= 0 ? res : 0; 2982 } 2983 EXPORT_SYMBOL_GPL(sock_prot_inuse_get); 2984 #endif 2985 2986 static void assign_proto_idx(struct proto *prot) 2987 { 2988 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR); 2989 2990 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) { 2991 pr_err("PROTO_INUSE_NR exhausted\n"); 2992 return; 2993 } 2994 2995 set_bit(prot->inuse_idx, proto_inuse_idx); 2996 } 2997 2998 static void release_proto_idx(struct proto *prot) 2999 { 3000 if (prot->inuse_idx != PROTO_INUSE_NR - 1) 3001 clear_bit(prot->inuse_idx, proto_inuse_idx); 3002 } 3003 #else 3004 static inline void assign_proto_idx(struct proto *prot) 3005 { 3006 } 3007 3008 static inline void release_proto_idx(struct proto *prot) 3009 { 3010 } 3011 #endif 3012 3013 static void req_prot_cleanup(struct request_sock_ops *rsk_prot) 3014 { 3015 if (!rsk_prot) 3016 return; 3017 kfree(rsk_prot->slab_name); 3018 rsk_prot->slab_name = NULL; 3019 kmem_cache_destroy(rsk_prot->slab); 3020 rsk_prot->slab = NULL; 3021 } 3022 3023 static int req_prot_init(const struct proto *prot) 3024 { 3025 struct request_sock_ops *rsk_prot = prot->rsk_prot; 3026 3027 if (!rsk_prot) 3028 return 0; 3029 3030 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", 3031 prot->name); 3032 if (!rsk_prot->slab_name) 3033 return -ENOMEM; 3034 3035 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name, 3036 rsk_prot->obj_size, 0, 3037 prot->slab_flags, NULL); 3038 3039 if (!rsk_prot->slab) { 3040 pr_crit("%s: Can't create request sock SLAB cache!\n", 3041 prot->name); 3042 return -ENOMEM; 3043 } 3044 return 0; 3045 } 3046 3047 int proto_register(struct proto *prot, int alloc_slab) 3048 { 3049 if (alloc_slab) { 3050 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0, 3051 SLAB_HWCACHE_ALIGN | prot->slab_flags, 3052 NULL); 3053 3054 if (prot->slab == NULL) { 3055 pr_crit("%s: Can't create sock SLAB cache!\n", 3056 prot->name); 3057 goto out; 3058 } 3059 3060 if (req_prot_init(prot)) 3061 goto out_free_request_sock_slab; 3062 3063 if (prot->twsk_prot != NULL) { 3064 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name); 3065 3066 if (prot->twsk_prot->twsk_slab_name == NULL) 3067 goto out_free_request_sock_slab; 3068 3069 prot->twsk_prot->twsk_slab = 3070 kmem_cache_create(prot->twsk_prot->twsk_slab_name, 3071 prot->twsk_prot->twsk_obj_size, 3072 0, 3073 prot->slab_flags, 3074 NULL); 3075 if (prot->twsk_prot->twsk_slab == NULL) 3076 goto out_free_timewait_sock_slab_name; 3077 } 3078 } 3079 3080 mutex_lock(&proto_list_mutex); 3081 list_add(&prot->node, &proto_list); 3082 assign_proto_idx(prot); 3083 mutex_unlock(&proto_list_mutex); 3084 return 0; 3085 3086 out_free_timewait_sock_slab_name: 3087 kfree(prot->twsk_prot->twsk_slab_name); 3088 out_free_request_sock_slab: 3089 req_prot_cleanup(prot->rsk_prot); 3090 3091 kmem_cache_destroy(prot->slab); 3092 prot->slab = NULL; 3093 out: 3094 return -ENOBUFS; 3095 } 3096 EXPORT_SYMBOL(proto_register); 3097 3098 void proto_unregister(struct proto *prot) 3099 { 3100 mutex_lock(&proto_list_mutex); 3101 release_proto_idx(prot); 3102 list_del(&prot->node); 3103 mutex_unlock(&proto_list_mutex); 3104 3105 kmem_cache_destroy(prot->slab); 3106 prot->slab = NULL; 3107 3108 req_prot_cleanup(prot->rsk_prot); 3109 3110 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) { 3111 kmem_cache_destroy(prot->twsk_prot->twsk_slab); 3112 kfree(prot->twsk_prot->twsk_slab_name); 3113 prot->twsk_prot->twsk_slab = NULL; 3114 } 3115 } 3116 EXPORT_SYMBOL(proto_unregister); 3117 3118 #ifdef CONFIG_PROC_FS 3119 static void *proto_seq_start(struct seq_file *seq, loff_t *pos) 3120 __acquires(proto_list_mutex) 3121 { 3122 mutex_lock(&proto_list_mutex); 3123 return seq_list_start_head(&proto_list, *pos); 3124 } 3125 3126 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos) 3127 { 3128 return seq_list_next(v, &proto_list, pos); 3129 } 3130 3131 static void proto_seq_stop(struct seq_file *seq, void *v) 3132 __releases(proto_list_mutex) 3133 { 3134 mutex_unlock(&proto_list_mutex); 3135 } 3136 3137 static char proto_method_implemented(const void *method) 3138 { 3139 return method == NULL ? 'n' : 'y'; 3140 } 3141 static long sock_prot_memory_allocated(struct proto *proto) 3142 { 3143 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L; 3144 } 3145 3146 static char *sock_prot_memory_pressure(struct proto *proto) 3147 { 3148 return proto->memory_pressure != NULL ? 3149 proto_memory_pressure(proto) ? "yes" : "no" : "NI"; 3150 } 3151 3152 static void proto_seq_printf(struct seq_file *seq, struct proto *proto) 3153 { 3154 3155 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s " 3156 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n", 3157 proto->name, 3158 proto->obj_size, 3159 sock_prot_inuse_get(seq_file_net(seq), proto), 3160 sock_prot_memory_allocated(proto), 3161 sock_prot_memory_pressure(proto), 3162 proto->max_header, 3163 proto->slab == NULL ? "no" : "yes", 3164 module_name(proto->owner), 3165 proto_method_implemented(proto->close), 3166 proto_method_implemented(proto->connect), 3167 proto_method_implemented(proto->disconnect), 3168 proto_method_implemented(proto->accept), 3169 proto_method_implemented(proto->ioctl), 3170 proto_method_implemented(proto->init), 3171 proto_method_implemented(proto->destroy), 3172 proto_method_implemented(proto->shutdown), 3173 proto_method_implemented(proto->setsockopt), 3174 proto_method_implemented(proto->getsockopt), 3175 proto_method_implemented(proto->sendmsg), 3176 proto_method_implemented(proto->recvmsg), 3177 proto_method_implemented(proto->sendpage), 3178 proto_method_implemented(proto->bind), 3179 proto_method_implemented(proto->backlog_rcv), 3180 proto_method_implemented(proto->hash), 3181 proto_method_implemented(proto->unhash), 3182 proto_method_implemented(proto->get_port), 3183 proto_method_implemented(proto->enter_memory_pressure)); 3184 } 3185 3186 static int proto_seq_show(struct seq_file *seq, void *v) 3187 { 3188 if (v == &proto_list) 3189 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s", 3190 "protocol", 3191 "size", 3192 "sockets", 3193 "memory", 3194 "press", 3195 "maxhdr", 3196 "slab", 3197 "module", 3198 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n"); 3199 else 3200 proto_seq_printf(seq, list_entry(v, struct proto, node)); 3201 return 0; 3202 } 3203 3204 static const struct seq_operations proto_seq_ops = { 3205 .start = proto_seq_start, 3206 .next = proto_seq_next, 3207 .stop = proto_seq_stop, 3208 .show = proto_seq_show, 3209 }; 3210 3211 static int proto_seq_open(struct inode *inode, struct file *file) 3212 { 3213 return seq_open_net(inode, file, &proto_seq_ops, 3214 sizeof(struct seq_net_private)); 3215 } 3216 3217 static const struct file_operations proto_seq_fops = { 3218 .owner = THIS_MODULE, 3219 .open = proto_seq_open, 3220 .read = seq_read, 3221 .llseek = seq_lseek, 3222 .release = seq_release_net, 3223 }; 3224 3225 static __net_init int proto_init_net(struct net *net) 3226 { 3227 if (!proc_create("protocols", S_IRUGO, net->proc_net, &proto_seq_fops)) 3228 return -ENOMEM; 3229 3230 return 0; 3231 } 3232 3233 static __net_exit void proto_exit_net(struct net *net) 3234 { 3235 remove_proc_entry("protocols", net->proc_net); 3236 } 3237 3238 3239 static __net_initdata struct pernet_operations proto_net_ops = { 3240 .init = proto_init_net, 3241 .exit = proto_exit_net, 3242 }; 3243 3244 static int __init proto_init(void) 3245 { 3246 return register_pernet_subsys(&proto_net_ops); 3247 } 3248 3249 subsys_initcall(proto_init); 3250 3251 #endif /* PROC_FS */ 3252 3253 #ifdef CONFIG_NET_RX_BUSY_POLL 3254 bool sk_busy_loop_end(void *p, unsigned long start_time) 3255 { 3256 struct sock *sk = p; 3257 3258 return !skb_queue_empty(&sk->sk_receive_queue) || 3259 sk_busy_loop_timeout(sk, start_time); 3260 } 3261 EXPORT_SYMBOL(sk_busy_loop_end); 3262 #endif /* CONFIG_NET_RX_BUSY_POLL */ 3263