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/sched/mm.h> 106 #include <linux/timer.h> 107 #include <linux/string.h> 108 #include <linux/sockios.h> 109 #include <linux/net.h> 110 #include <linux/mm.h> 111 #include <linux/slab.h> 112 #include <linux/interrupt.h> 113 #include <linux/poll.h> 114 #include <linux/tcp.h> 115 #include <linux/init.h> 116 #include <linux/highmem.h> 117 #include <linux/user_namespace.h> 118 #include <linux/static_key.h> 119 #include <linux/memcontrol.h> 120 #include <linux/prefetch.h> 121 122 #include <linux/uaccess.h> 123 124 #include <linux/netdevice.h> 125 #include <net/protocol.h> 126 #include <linux/skbuff.h> 127 #include <net/net_namespace.h> 128 #include <net/request_sock.h> 129 #include <net/sock.h> 130 #include <linux/net_tstamp.h> 131 #include <net/xfrm.h> 132 #include <linux/ipsec.h> 133 #include <net/cls_cgroup.h> 134 #include <net/netprio_cgroup.h> 135 #include <linux/sock_diag.h> 136 137 #include <linux/filter.h> 138 #include <net/sock_reuseport.h> 139 140 #include <trace/events/sock.h> 141 142 #ifdef CONFIG_INET 143 #include <net/tcp.h> 144 #endif 145 146 #include <net/busy_poll.h> 147 148 static DEFINE_MUTEX(proto_list_mutex); 149 static LIST_HEAD(proto_list); 150 151 /** 152 * sk_ns_capable - General socket capability test 153 * @sk: Socket to use a capability on or through 154 * @user_ns: The user namespace of the capability to use 155 * @cap: The capability to use 156 * 157 * Test to see if the opener of the socket had when the socket was 158 * created and the current process has the capability @cap in the user 159 * namespace @user_ns. 160 */ 161 bool sk_ns_capable(const struct sock *sk, 162 struct user_namespace *user_ns, int cap) 163 { 164 return file_ns_capable(sk->sk_socket->file, user_ns, cap) && 165 ns_capable(user_ns, cap); 166 } 167 EXPORT_SYMBOL(sk_ns_capable); 168 169 /** 170 * sk_capable - Socket global capability test 171 * @sk: Socket to use a capability on or through 172 * @cap: The global capability to use 173 * 174 * Test to see if the opener of the socket had when the socket was 175 * created and the current process has the capability @cap in all user 176 * namespaces. 177 */ 178 bool sk_capable(const struct sock *sk, int cap) 179 { 180 return sk_ns_capable(sk, &init_user_ns, cap); 181 } 182 EXPORT_SYMBOL(sk_capable); 183 184 /** 185 * sk_net_capable - Network namespace socket capability test 186 * @sk: Socket to use a capability on or through 187 * @cap: The capability to use 188 * 189 * Test to see if the opener of the socket had when the socket was created 190 * and the current process has the capability @cap over the network namespace 191 * the socket is a member of. 192 */ 193 bool sk_net_capable(const struct sock *sk, int cap) 194 { 195 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap); 196 } 197 EXPORT_SYMBOL(sk_net_capable); 198 199 /* 200 * Each address family might have different locking rules, so we have 201 * one slock key per address family and separate keys for internal and 202 * userspace sockets. 203 */ 204 static struct lock_class_key af_family_keys[AF_MAX]; 205 static struct lock_class_key af_family_kern_keys[AF_MAX]; 206 static struct lock_class_key af_family_slock_keys[AF_MAX]; 207 static struct lock_class_key af_family_kern_slock_keys[AF_MAX]; 208 209 /* 210 * Make lock validator output more readable. (we pre-construct these 211 * strings build-time, so that runtime initialization of socket 212 * locks is fast): 213 */ 214 215 #define _sock_locks(x) \ 216 x "AF_UNSPEC", x "AF_UNIX" , x "AF_INET" , \ 217 x "AF_AX25" , x "AF_IPX" , x "AF_APPLETALK", \ 218 x "AF_NETROM", x "AF_BRIDGE" , x "AF_ATMPVC" , \ 219 x "AF_X25" , x "AF_INET6" , x "AF_ROSE" , \ 220 x "AF_DECnet", x "AF_NETBEUI" , x "AF_SECURITY" , \ 221 x "AF_KEY" , x "AF_NETLINK" , x "AF_PACKET" , \ 222 x "AF_ASH" , x "AF_ECONET" , x "AF_ATMSVC" , \ 223 x "AF_RDS" , x "AF_SNA" , x "AF_IRDA" , \ 224 x "AF_PPPOX" , x "AF_WANPIPE" , x "AF_LLC" , \ 225 x "27" , x "28" , x "AF_CAN" , \ 226 x "AF_TIPC" , x "AF_BLUETOOTH", x "IUCV" , \ 227 x "AF_RXRPC" , x "AF_ISDN" , x "AF_PHONET" , \ 228 x "AF_IEEE802154", x "AF_CAIF" , x "AF_ALG" , \ 229 x "AF_NFC" , x "AF_VSOCK" , x "AF_KCM" , \ 230 x "AF_QIPCRTR", x "AF_SMC" , x "AF_MAX" 231 232 static const char *const af_family_key_strings[AF_MAX+1] = { 233 _sock_locks("sk_lock-") 234 }; 235 static const char *const af_family_slock_key_strings[AF_MAX+1] = { 236 _sock_locks("slock-") 237 }; 238 static const char *const af_family_clock_key_strings[AF_MAX+1] = { 239 _sock_locks("clock-") 240 }; 241 242 static const char *const af_family_kern_key_strings[AF_MAX+1] = { 243 _sock_locks("k-sk_lock-") 244 }; 245 static const char *const af_family_kern_slock_key_strings[AF_MAX+1] = { 246 _sock_locks("k-slock-") 247 }; 248 static const char *const af_family_kern_clock_key_strings[AF_MAX+1] = { 249 _sock_locks("k-clock-") 250 }; 251 static const char *const af_family_rlock_key_strings[AF_MAX+1] = { 252 "rlock-AF_UNSPEC", "rlock-AF_UNIX" , "rlock-AF_INET" , 253 "rlock-AF_AX25" , "rlock-AF_IPX" , "rlock-AF_APPLETALK", 254 "rlock-AF_NETROM", "rlock-AF_BRIDGE" , "rlock-AF_ATMPVC" , 255 "rlock-AF_X25" , "rlock-AF_INET6" , "rlock-AF_ROSE" , 256 "rlock-AF_DECnet", "rlock-AF_NETBEUI" , "rlock-AF_SECURITY" , 257 "rlock-AF_KEY" , "rlock-AF_NETLINK" , "rlock-AF_PACKET" , 258 "rlock-AF_ASH" , "rlock-AF_ECONET" , "rlock-AF_ATMSVC" , 259 "rlock-AF_RDS" , "rlock-AF_SNA" , "rlock-AF_IRDA" , 260 "rlock-AF_PPPOX" , "rlock-AF_WANPIPE" , "rlock-AF_LLC" , 261 "rlock-27" , "rlock-28" , "rlock-AF_CAN" , 262 "rlock-AF_TIPC" , "rlock-AF_BLUETOOTH", "rlock-AF_IUCV" , 263 "rlock-AF_RXRPC" , "rlock-AF_ISDN" , "rlock-AF_PHONET" , 264 "rlock-AF_IEEE802154", "rlock-AF_CAIF" , "rlock-AF_ALG" , 265 "rlock-AF_NFC" , "rlock-AF_VSOCK" , "rlock-AF_KCM" , 266 "rlock-AF_QIPCRTR", "rlock-AF_SMC" , "rlock-AF_MAX" 267 }; 268 static const char *const af_family_wlock_key_strings[AF_MAX+1] = { 269 "wlock-AF_UNSPEC", "wlock-AF_UNIX" , "wlock-AF_INET" , 270 "wlock-AF_AX25" , "wlock-AF_IPX" , "wlock-AF_APPLETALK", 271 "wlock-AF_NETROM", "wlock-AF_BRIDGE" , "wlock-AF_ATMPVC" , 272 "wlock-AF_X25" , "wlock-AF_INET6" , "wlock-AF_ROSE" , 273 "wlock-AF_DECnet", "wlock-AF_NETBEUI" , "wlock-AF_SECURITY" , 274 "wlock-AF_KEY" , "wlock-AF_NETLINK" , "wlock-AF_PACKET" , 275 "wlock-AF_ASH" , "wlock-AF_ECONET" , "wlock-AF_ATMSVC" , 276 "wlock-AF_RDS" , "wlock-AF_SNA" , "wlock-AF_IRDA" , 277 "wlock-AF_PPPOX" , "wlock-AF_WANPIPE" , "wlock-AF_LLC" , 278 "wlock-27" , "wlock-28" , "wlock-AF_CAN" , 279 "wlock-AF_TIPC" , "wlock-AF_BLUETOOTH", "wlock-AF_IUCV" , 280 "wlock-AF_RXRPC" , "wlock-AF_ISDN" , "wlock-AF_PHONET" , 281 "wlock-AF_IEEE802154", "wlock-AF_CAIF" , "wlock-AF_ALG" , 282 "wlock-AF_NFC" , "wlock-AF_VSOCK" , "wlock-AF_KCM" , 283 "wlock-AF_QIPCRTR", "wlock-AF_SMC" , "wlock-AF_MAX" 284 }; 285 static const char *const af_family_elock_key_strings[AF_MAX+1] = { 286 "elock-AF_UNSPEC", "elock-AF_UNIX" , "elock-AF_INET" , 287 "elock-AF_AX25" , "elock-AF_IPX" , "elock-AF_APPLETALK", 288 "elock-AF_NETROM", "elock-AF_BRIDGE" , "elock-AF_ATMPVC" , 289 "elock-AF_X25" , "elock-AF_INET6" , "elock-AF_ROSE" , 290 "elock-AF_DECnet", "elock-AF_NETBEUI" , "elock-AF_SECURITY" , 291 "elock-AF_KEY" , "elock-AF_NETLINK" , "elock-AF_PACKET" , 292 "elock-AF_ASH" , "elock-AF_ECONET" , "elock-AF_ATMSVC" , 293 "elock-AF_RDS" , "elock-AF_SNA" , "elock-AF_IRDA" , 294 "elock-AF_PPPOX" , "elock-AF_WANPIPE" , "elock-AF_LLC" , 295 "elock-27" , "elock-28" , "elock-AF_CAN" , 296 "elock-AF_TIPC" , "elock-AF_BLUETOOTH", "elock-AF_IUCV" , 297 "elock-AF_RXRPC" , "elock-AF_ISDN" , "elock-AF_PHONET" , 298 "elock-AF_IEEE802154", "elock-AF_CAIF" , "elock-AF_ALG" , 299 "elock-AF_NFC" , "elock-AF_VSOCK" , "elock-AF_KCM" , 300 "elock-AF_QIPCRTR", "elock-AF_SMC" , "elock-AF_MAX" 301 }; 302 303 /* 304 * sk_callback_lock and sk queues locking rules are per-address-family, 305 * so split the lock classes by using a per-AF key: 306 */ 307 static struct lock_class_key af_callback_keys[AF_MAX]; 308 static struct lock_class_key af_rlock_keys[AF_MAX]; 309 static struct lock_class_key af_wlock_keys[AF_MAX]; 310 static struct lock_class_key af_elock_keys[AF_MAX]; 311 static struct lock_class_key af_kern_callback_keys[AF_MAX]; 312 313 /* Take into consideration the size of the struct sk_buff overhead in the 314 * determination of these values, since that is non-constant across 315 * platforms. This makes socket queueing behavior and performance 316 * not depend upon such differences. 317 */ 318 #define _SK_MEM_PACKETS 256 319 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256) 320 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 321 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 322 323 /* Run time adjustable parameters. */ 324 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX; 325 EXPORT_SYMBOL(sysctl_wmem_max); 326 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX; 327 EXPORT_SYMBOL(sysctl_rmem_max); 328 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX; 329 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX; 330 331 /* Maximal space eaten by iovec or ancillary data plus some space */ 332 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512); 333 EXPORT_SYMBOL(sysctl_optmem_max); 334 335 int sysctl_tstamp_allow_data __read_mostly = 1; 336 337 struct static_key memalloc_socks = STATIC_KEY_INIT_FALSE; 338 EXPORT_SYMBOL_GPL(memalloc_socks); 339 340 /** 341 * sk_set_memalloc - sets %SOCK_MEMALLOC 342 * @sk: socket to set it on 343 * 344 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves. 345 * It's the responsibility of the admin to adjust min_free_kbytes 346 * to meet the requirements 347 */ 348 void sk_set_memalloc(struct sock *sk) 349 { 350 sock_set_flag(sk, SOCK_MEMALLOC); 351 sk->sk_allocation |= __GFP_MEMALLOC; 352 static_key_slow_inc(&memalloc_socks); 353 } 354 EXPORT_SYMBOL_GPL(sk_set_memalloc); 355 356 void sk_clear_memalloc(struct sock *sk) 357 { 358 sock_reset_flag(sk, SOCK_MEMALLOC); 359 sk->sk_allocation &= ~__GFP_MEMALLOC; 360 static_key_slow_dec(&memalloc_socks); 361 362 /* 363 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward 364 * progress of swapping. SOCK_MEMALLOC may be cleared while 365 * it has rmem allocations due to the last swapfile being deactivated 366 * but there is a risk that the socket is unusable due to exceeding 367 * the rmem limits. Reclaim the reserves and obey rmem limits again. 368 */ 369 sk_mem_reclaim(sk); 370 } 371 EXPORT_SYMBOL_GPL(sk_clear_memalloc); 372 373 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb) 374 { 375 int ret; 376 unsigned int noreclaim_flag; 377 378 /* these should have been dropped before queueing */ 379 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC)); 380 381 noreclaim_flag = memalloc_noreclaim_save(); 382 ret = sk->sk_backlog_rcv(sk, skb); 383 memalloc_noreclaim_restore(noreclaim_flag); 384 385 return ret; 386 } 387 EXPORT_SYMBOL(__sk_backlog_rcv); 388 389 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen) 390 { 391 struct timeval tv; 392 393 if (optlen < sizeof(tv)) 394 return -EINVAL; 395 if (copy_from_user(&tv, optval, sizeof(tv))) 396 return -EFAULT; 397 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC) 398 return -EDOM; 399 400 if (tv.tv_sec < 0) { 401 static int warned __read_mostly; 402 403 *timeo_p = 0; 404 if (warned < 10 && net_ratelimit()) { 405 warned++; 406 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n", 407 __func__, current->comm, task_pid_nr(current)); 408 } 409 return 0; 410 } 411 *timeo_p = MAX_SCHEDULE_TIMEOUT; 412 if (tv.tv_sec == 0 && tv.tv_usec == 0) 413 return 0; 414 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1)) 415 *timeo_p = tv.tv_sec * HZ + DIV_ROUND_UP(tv.tv_usec, USEC_PER_SEC / HZ); 416 return 0; 417 } 418 419 static void sock_warn_obsolete_bsdism(const char *name) 420 { 421 static int warned; 422 static char warncomm[TASK_COMM_LEN]; 423 if (strcmp(warncomm, current->comm) && warned < 5) { 424 strcpy(warncomm, current->comm); 425 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n", 426 warncomm, name); 427 warned++; 428 } 429 } 430 431 static bool sock_needs_netstamp(const struct sock *sk) 432 { 433 switch (sk->sk_family) { 434 case AF_UNSPEC: 435 case AF_UNIX: 436 return false; 437 default: 438 return true; 439 } 440 } 441 442 static void sock_disable_timestamp(struct sock *sk, unsigned long flags) 443 { 444 if (sk->sk_flags & flags) { 445 sk->sk_flags &= ~flags; 446 if (sock_needs_netstamp(sk) && 447 !(sk->sk_flags & SK_FLAGS_TIMESTAMP)) 448 net_disable_timestamp(); 449 } 450 } 451 452 453 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 454 { 455 unsigned long flags; 456 struct sk_buff_head *list = &sk->sk_receive_queue; 457 458 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) { 459 atomic_inc(&sk->sk_drops); 460 trace_sock_rcvqueue_full(sk, skb); 461 return -ENOMEM; 462 } 463 464 if (!sk_rmem_schedule(sk, skb, skb->truesize)) { 465 atomic_inc(&sk->sk_drops); 466 return -ENOBUFS; 467 } 468 469 skb->dev = NULL; 470 skb_set_owner_r(skb, sk); 471 472 /* we escape from rcu protected region, make sure we dont leak 473 * a norefcounted dst 474 */ 475 skb_dst_force(skb); 476 477 spin_lock_irqsave(&list->lock, flags); 478 sock_skb_set_dropcount(sk, skb); 479 __skb_queue_tail(list, skb); 480 spin_unlock_irqrestore(&list->lock, flags); 481 482 if (!sock_flag(sk, SOCK_DEAD)) 483 sk->sk_data_ready(sk); 484 return 0; 485 } 486 EXPORT_SYMBOL(__sock_queue_rcv_skb); 487 488 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 489 { 490 int err; 491 492 err = sk_filter(sk, skb); 493 if (err) 494 return err; 495 496 return __sock_queue_rcv_skb(sk, skb); 497 } 498 EXPORT_SYMBOL(sock_queue_rcv_skb); 499 500 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, 501 const int nested, unsigned int trim_cap, bool refcounted) 502 { 503 int rc = NET_RX_SUCCESS; 504 505 if (sk_filter_trim_cap(sk, skb, trim_cap)) 506 goto discard_and_relse; 507 508 skb->dev = NULL; 509 510 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) { 511 atomic_inc(&sk->sk_drops); 512 goto discard_and_relse; 513 } 514 if (nested) 515 bh_lock_sock_nested(sk); 516 else 517 bh_lock_sock(sk); 518 if (!sock_owned_by_user(sk)) { 519 /* 520 * trylock + unlock semantics: 521 */ 522 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_); 523 524 rc = sk_backlog_rcv(sk, skb); 525 526 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_); 527 } else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) { 528 bh_unlock_sock(sk); 529 atomic_inc(&sk->sk_drops); 530 goto discard_and_relse; 531 } 532 533 bh_unlock_sock(sk); 534 out: 535 if (refcounted) 536 sock_put(sk); 537 return rc; 538 discard_and_relse: 539 kfree_skb(skb); 540 goto out; 541 } 542 EXPORT_SYMBOL(__sk_receive_skb); 543 544 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie) 545 { 546 struct dst_entry *dst = __sk_dst_get(sk); 547 548 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) { 549 sk_tx_queue_clear(sk); 550 sk->sk_dst_pending_confirm = 0; 551 RCU_INIT_POINTER(sk->sk_dst_cache, NULL); 552 dst_release(dst); 553 return NULL; 554 } 555 556 return dst; 557 } 558 EXPORT_SYMBOL(__sk_dst_check); 559 560 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie) 561 { 562 struct dst_entry *dst = sk_dst_get(sk); 563 564 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) { 565 sk_dst_reset(sk); 566 dst_release(dst); 567 return NULL; 568 } 569 570 return dst; 571 } 572 EXPORT_SYMBOL(sk_dst_check); 573 574 static int sock_setbindtodevice(struct sock *sk, char __user *optval, 575 int optlen) 576 { 577 int ret = -ENOPROTOOPT; 578 #ifdef CONFIG_NETDEVICES 579 struct net *net = sock_net(sk); 580 char devname[IFNAMSIZ]; 581 int index; 582 583 /* Sorry... */ 584 ret = -EPERM; 585 if (!ns_capable(net->user_ns, CAP_NET_RAW)) 586 goto out; 587 588 ret = -EINVAL; 589 if (optlen < 0) 590 goto out; 591 592 /* Bind this socket to a particular device like "eth0", 593 * as specified in the passed interface name. If the 594 * name is "" or the option length is zero the socket 595 * is not bound. 596 */ 597 if (optlen > IFNAMSIZ - 1) 598 optlen = IFNAMSIZ - 1; 599 memset(devname, 0, sizeof(devname)); 600 601 ret = -EFAULT; 602 if (copy_from_user(devname, optval, optlen)) 603 goto out; 604 605 index = 0; 606 if (devname[0] != '\0') { 607 struct net_device *dev; 608 609 rcu_read_lock(); 610 dev = dev_get_by_name_rcu(net, devname); 611 if (dev) 612 index = dev->ifindex; 613 rcu_read_unlock(); 614 ret = -ENODEV; 615 if (!dev) 616 goto out; 617 } 618 619 lock_sock(sk); 620 sk->sk_bound_dev_if = index; 621 sk_dst_reset(sk); 622 release_sock(sk); 623 624 ret = 0; 625 626 out: 627 #endif 628 629 return ret; 630 } 631 632 static int sock_getbindtodevice(struct sock *sk, char __user *optval, 633 int __user *optlen, int len) 634 { 635 int ret = -ENOPROTOOPT; 636 #ifdef CONFIG_NETDEVICES 637 struct net *net = sock_net(sk); 638 char devname[IFNAMSIZ]; 639 640 if (sk->sk_bound_dev_if == 0) { 641 len = 0; 642 goto zero; 643 } 644 645 ret = -EINVAL; 646 if (len < IFNAMSIZ) 647 goto out; 648 649 ret = netdev_get_name(net, devname, sk->sk_bound_dev_if); 650 if (ret) 651 goto out; 652 653 len = strlen(devname) + 1; 654 655 ret = -EFAULT; 656 if (copy_to_user(optval, devname, len)) 657 goto out; 658 659 zero: 660 ret = -EFAULT; 661 if (put_user(len, optlen)) 662 goto out; 663 664 ret = 0; 665 666 out: 667 #endif 668 669 return ret; 670 } 671 672 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool) 673 { 674 if (valbool) 675 sock_set_flag(sk, bit); 676 else 677 sock_reset_flag(sk, bit); 678 } 679 680 bool sk_mc_loop(struct sock *sk) 681 { 682 if (dev_recursion_level()) 683 return false; 684 if (!sk) 685 return true; 686 switch (sk->sk_family) { 687 case AF_INET: 688 return inet_sk(sk)->mc_loop; 689 #if IS_ENABLED(CONFIG_IPV6) 690 case AF_INET6: 691 return inet6_sk(sk)->mc_loop; 692 #endif 693 } 694 WARN_ON(1); 695 return true; 696 } 697 EXPORT_SYMBOL(sk_mc_loop); 698 699 /* 700 * This is meant for all protocols to use and covers goings on 701 * at the socket level. Everything here is generic. 702 */ 703 704 int sock_setsockopt(struct socket *sock, int level, int optname, 705 char __user *optval, unsigned int optlen) 706 { 707 struct sock *sk = sock->sk; 708 int val; 709 int valbool; 710 struct linger ling; 711 int ret = 0; 712 713 /* 714 * Options without arguments 715 */ 716 717 if (optname == SO_BINDTODEVICE) 718 return sock_setbindtodevice(sk, optval, optlen); 719 720 if (optlen < sizeof(int)) 721 return -EINVAL; 722 723 if (get_user(val, (int __user *)optval)) 724 return -EFAULT; 725 726 valbool = val ? 1 : 0; 727 728 lock_sock(sk); 729 730 switch (optname) { 731 case SO_DEBUG: 732 if (val && !capable(CAP_NET_ADMIN)) 733 ret = -EACCES; 734 else 735 sock_valbool_flag(sk, SOCK_DBG, valbool); 736 break; 737 case SO_REUSEADDR: 738 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE); 739 break; 740 case SO_REUSEPORT: 741 sk->sk_reuseport = valbool; 742 break; 743 case SO_TYPE: 744 case SO_PROTOCOL: 745 case SO_DOMAIN: 746 case SO_ERROR: 747 ret = -ENOPROTOOPT; 748 break; 749 case SO_DONTROUTE: 750 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool); 751 break; 752 case SO_BROADCAST: 753 sock_valbool_flag(sk, SOCK_BROADCAST, valbool); 754 break; 755 case SO_SNDBUF: 756 /* Don't error on this BSD doesn't and if you think 757 * about it this is right. Otherwise apps have to 758 * play 'guess the biggest size' games. RCVBUF/SNDBUF 759 * are treated in BSD as hints 760 */ 761 val = min_t(u32, val, sysctl_wmem_max); 762 set_sndbuf: 763 sk->sk_userlocks |= SOCK_SNDBUF_LOCK; 764 sk->sk_sndbuf = max_t(int, val * 2, SOCK_MIN_SNDBUF); 765 /* Wake up sending tasks if we upped the value. */ 766 sk->sk_write_space(sk); 767 break; 768 769 case SO_SNDBUFFORCE: 770 if (!capable(CAP_NET_ADMIN)) { 771 ret = -EPERM; 772 break; 773 } 774 goto set_sndbuf; 775 776 case SO_RCVBUF: 777 /* Don't error on this BSD doesn't and if you think 778 * about it this is right. Otherwise apps have to 779 * play 'guess the biggest size' games. RCVBUF/SNDBUF 780 * are treated in BSD as hints 781 */ 782 val = min_t(u32, val, sysctl_rmem_max); 783 set_rcvbuf: 784 sk->sk_userlocks |= SOCK_RCVBUF_LOCK; 785 /* 786 * We double it on the way in to account for 787 * "struct sk_buff" etc. overhead. Applications 788 * assume that the SO_RCVBUF setting they make will 789 * allow that much actual data to be received on that 790 * socket. 791 * 792 * Applications are unaware that "struct sk_buff" and 793 * other overheads allocate from the receive buffer 794 * during socket buffer allocation. 795 * 796 * And after considering the possible alternatives, 797 * returning the value we actually used in getsockopt 798 * is the most desirable behavior. 799 */ 800 sk->sk_rcvbuf = max_t(int, val * 2, SOCK_MIN_RCVBUF); 801 break; 802 803 case SO_RCVBUFFORCE: 804 if (!capable(CAP_NET_ADMIN)) { 805 ret = -EPERM; 806 break; 807 } 808 goto set_rcvbuf; 809 810 case SO_KEEPALIVE: 811 if (sk->sk_prot->keepalive) 812 sk->sk_prot->keepalive(sk, valbool); 813 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool); 814 break; 815 816 case SO_OOBINLINE: 817 sock_valbool_flag(sk, SOCK_URGINLINE, valbool); 818 break; 819 820 case SO_NO_CHECK: 821 sk->sk_no_check_tx = valbool; 822 break; 823 824 case SO_PRIORITY: 825 if ((val >= 0 && val <= 6) || 826 ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) 827 sk->sk_priority = val; 828 else 829 ret = -EPERM; 830 break; 831 832 case SO_LINGER: 833 if (optlen < sizeof(ling)) { 834 ret = -EINVAL; /* 1003.1g */ 835 break; 836 } 837 if (copy_from_user(&ling, optval, sizeof(ling))) { 838 ret = -EFAULT; 839 break; 840 } 841 if (!ling.l_onoff) 842 sock_reset_flag(sk, SOCK_LINGER); 843 else { 844 #if (BITS_PER_LONG == 32) 845 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ) 846 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT; 847 else 848 #endif 849 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ; 850 sock_set_flag(sk, SOCK_LINGER); 851 } 852 break; 853 854 case SO_BSDCOMPAT: 855 sock_warn_obsolete_bsdism("setsockopt"); 856 break; 857 858 case SO_PASSCRED: 859 if (valbool) 860 set_bit(SOCK_PASSCRED, &sock->flags); 861 else 862 clear_bit(SOCK_PASSCRED, &sock->flags); 863 break; 864 865 case SO_TIMESTAMP: 866 case SO_TIMESTAMPNS: 867 if (valbool) { 868 if (optname == SO_TIMESTAMP) 869 sock_reset_flag(sk, SOCK_RCVTSTAMPNS); 870 else 871 sock_set_flag(sk, SOCK_RCVTSTAMPNS); 872 sock_set_flag(sk, SOCK_RCVTSTAMP); 873 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 874 } else { 875 sock_reset_flag(sk, SOCK_RCVTSTAMP); 876 sock_reset_flag(sk, SOCK_RCVTSTAMPNS); 877 } 878 break; 879 880 case SO_TIMESTAMPING: 881 if (val & ~SOF_TIMESTAMPING_MASK) { 882 ret = -EINVAL; 883 break; 884 } 885 886 if (val & SOF_TIMESTAMPING_OPT_ID && 887 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) { 888 if (sk->sk_protocol == IPPROTO_TCP && 889 sk->sk_type == SOCK_STREAM) { 890 if ((1 << sk->sk_state) & 891 (TCPF_CLOSE | TCPF_LISTEN)) { 892 ret = -EINVAL; 893 break; 894 } 895 sk->sk_tskey = tcp_sk(sk)->snd_una; 896 } else { 897 sk->sk_tskey = 0; 898 } 899 } 900 901 if (val & SOF_TIMESTAMPING_OPT_STATS && 902 !(val & SOF_TIMESTAMPING_OPT_TSONLY)) { 903 ret = -EINVAL; 904 break; 905 } 906 907 sk->sk_tsflags = val; 908 if (val & SOF_TIMESTAMPING_RX_SOFTWARE) 909 sock_enable_timestamp(sk, 910 SOCK_TIMESTAMPING_RX_SOFTWARE); 911 else 912 sock_disable_timestamp(sk, 913 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE)); 914 break; 915 916 case SO_RCVLOWAT: 917 if (val < 0) 918 val = INT_MAX; 919 sk->sk_rcvlowat = val ? : 1; 920 break; 921 922 case SO_RCVTIMEO: 923 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen); 924 break; 925 926 case SO_SNDTIMEO: 927 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen); 928 break; 929 930 case SO_ATTACH_FILTER: 931 ret = -EINVAL; 932 if (optlen == sizeof(struct sock_fprog)) { 933 struct sock_fprog fprog; 934 935 ret = -EFAULT; 936 if (copy_from_user(&fprog, optval, sizeof(fprog))) 937 break; 938 939 ret = sk_attach_filter(&fprog, sk); 940 } 941 break; 942 943 case SO_ATTACH_BPF: 944 ret = -EINVAL; 945 if (optlen == sizeof(u32)) { 946 u32 ufd; 947 948 ret = -EFAULT; 949 if (copy_from_user(&ufd, optval, sizeof(ufd))) 950 break; 951 952 ret = sk_attach_bpf(ufd, sk); 953 } 954 break; 955 956 case SO_ATTACH_REUSEPORT_CBPF: 957 ret = -EINVAL; 958 if (optlen == sizeof(struct sock_fprog)) { 959 struct sock_fprog fprog; 960 961 ret = -EFAULT; 962 if (copy_from_user(&fprog, optval, sizeof(fprog))) 963 break; 964 965 ret = sk_reuseport_attach_filter(&fprog, sk); 966 } 967 break; 968 969 case SO_ATTACH_REUSEPORT_EBPF: 970 ret = -EINVAL; 971 if (optlen == sizeof(u32)) { 972 u32 ufd; 973 974 ret = -EFAULT; 975 if (copy_from_user(&ufd, optval, sizeof(ufd))) 976 break; 977 978 ret = sk_reuseport_attach_bpf(ufd, sk); 979 } 980 break; 981 982 case SO_DETACH_FILTER: 983 ret = sk_detach_filter(sk); 984 break; 985 986 case SO_LOCK_FILTER: 987 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool) 988 ret = -EPERM; 989 else 990 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool); 991 break; 992 993 case SO_PASSSEC: 994 if (valbool) 995 set_bit(SOCK_PASSSEC, &sock->flags); 996 else 997 clear_bit(SOCK_PASSSEC, &sock->flags); 998 break; 999 case SO_MARK: 1000 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) 1001 ret = -EPERM; 1002 else 1003 sk->sk_mark = val; 1004 break; 1005 1006 case SO_RXQ_OVFL: 1007 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool); 1008 break; 1009 1010 case SO_WIFI_STATUS: 1011 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool); 1012 break; 1013 1014 case SO_PEEK_OFF: 1015 if (sock->ops->set_peek_off) 1016 ret = sock->ops->set_peek_off(sk, val); 1017 else 1018 ret = -EOPNOTSUPP; 1019 break; 1020 1021 case SO_NOFCS: 1022 sock_valbool_flag(sk, SOCK_NOFCS, valbool); 1023 break; 1024 1025 case SO_SELECT_ERR_QUEUE: 1026 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool); 1027 break; 1028 1029 #ifdef CONFIG_NET_RX_BUSY_POLL 1030 case SO_BUSY_POLL: 1031 /* allow unprivileged users to decrease the value */ 1032 if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN)) 1033 ret = -EPERM; 1034 else { 1035 if (val < 0) 1036 ret = -EINVAL; 1037 else 1038 sk->sk_ll_usec = val; 1039 } 1040 break; 1041 #endif 1042 1043 case SO_MAX_PACING_RATE: 1044 sk->sk_max_pacing_rate = val; 1045 sk->sk_pacing_rate = min(sk->sk_pacing_rate, 1046 sk->sk_max_pacing_rate); 1047 break; 1048 1049 case SO_INCOMING_CPU: 1050 sk->sk_incoming_cpu = val; 1051 break; 1052 1053 case SO_CNX_ADVICE: 1054 if (val == 1) 1055 dst_negative_advice(sk); 1056 break; 1057 default: 1058 ret = -ENOPROTOOPT; 1059 break; 1060 } 1061 release_sock(sk); 1062 return ret; 1063 } 1064 EXPORT_SYMBOL(sock_setsockopt); 1065 1066 1067 static void cred_to_ucred(struct pid *pid, const struct cred *cred, 1068 struct ucred *ucred) 1069 { 1070 ucred->pid = pid_vnr(pid); 1071 ucred->uid = ucred->gid = -1; 1072 if (cred) { 1073 struct user_namespace *current_ns = current_user_ns(); 1074 1075 ucred->uid = from_kuid_munged(current_ns, cred->euid); 1076 ucred->gid = from_kgid_munged(current_ns, cred->egid); 1077 } 1078 } 1079 1080 int sock_getsockopt(struct socket *sock, int level, int optname, 1081 char __user *optval, int __user *optlen) 1082 { 1083 struct sock *sk = sock->sk; 1084 1085 union { 1086 int val; 1087 u64 val64; 1088 struct linger ling; 1089 struct timeval tm; 1090 } v; 1091 1092 int lv = sizeof(int); 1093 int len; 1094 1095 if (get_user(len, optlen)) 1096 return -EFAULT; 1097 if (len < 0) 1098 return -EINVAL; 1099 1100 memset(&v, 0, sizeof(v)); 1101 1102 switch (optname) { 1103 case SO_DEBUG: 1104 v.val = sock_flag(sk, SOCK_DBG); 1105 break; 1106 1107 case SO_DONTROUTE: 1108 v.val = sock_flag(sk, SOCK_LOCALROUTE); 1109 break; 1110 1111 case SO_BROADCAST: 1112 v.val = sock_flag(sk, SOCK_BROADCAST); 1113 break; 1114 1115 case SO_SNDBUF: 1116 v.val = sk->sk_sndbuf; 1117 break; 1118 1119 case SO_RCVBUF: 1120 v.val = sk->sk_rcvbuf; 1121 break; 1122 1123 case SO_REUSEADDR: 1124 v.val = sk->sk_reuse; 1125 break; 1126 1127 case SO_REUSEPORT: 1128 v.val = sk->sk_reuseport; 1129 break; 1130 1131 case SO_KEEPALIVE: 1132 v.val = sock_flag(sk, SOCK_KEEPOPEN); 1133 break; 1134 1135 case SO_TYPE: 1136 v.val = sk->sk_type; 1137 break; 1138 1139 case SO_PROTOCOL: 1140 v.val = sk->sk_protocol; 1141 break; 1142 1143 case SO_DOMAIN: 1144 v.val = sk->sk_family; 1145 break; 1146 1147 case SO_ERROR: 1148 v.val = -sock_error(sk); 1149 if (v.val == 0) 1150 v.val = xchg(&sk->sk_err_soft, 0); 1151 break; 1152 1153 case SO_OOBINLINE: 1154 v.val = sock_flag(sk, SOCK_URGINLINE); 1155 break; 1156 1157 case SO_NO_CHECK: 1158 v.val = sk->sk_no_check_tx; 1159 break; 1160 1161 case SO_PRIORITY: 1162 v.val = sk->sk_priority; 1163 break; 1164 1165 case SO_LINGER: 1166 lv = sizeof(v.ling); 1167 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER); 1168 v.ling.l_linger = sk->sk_lingertime / HZ; 1169 break; 1170 1171 case SO_BSDCOMPAT: 1172 sock_warn_obsolete_bsdism("getsockopt"); 1173 break; 1174 1175 case SO_TIMESTAMP: 1176 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && 1177 !sock_flag(sk, SOCK_RCVTSTAMPNS); 1178 break; 1179 1180 case SO_TIMESTAMPNS: 1181 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS); 1182 break; 1183 1184 case SO_TIMESTAMPING: 1185 v.val = sk->sk_tsflags; 1186 break; 1187 1188 case SO_RCVTIMEO: 1189 lv = sizeof(struct timeval); 1190 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) { 1191 v.tm.tv_sec = 0; 1192 v.tm.tv_usec = 0; 1193 } else { 1194 v.tm.tv_sec = sk->sk_rcvtimeo / HZ; 1195 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * USEC_PER_SEC) / HZ; 1196 } 1197 break; 1198 1199 case SO_SNDTIMEO: 1200 lv = sizeof(struct timeval); 1201 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) { 1202 v.tm.tv_sec = 0; 1203 v.tm.tv_usec = 0; 1204 } else { 1205 v.tm.tv_sec = sk->sk_sndtimeo / HZ; 1206 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * USEC_PER_SEC) / HZ; 1207 } 1208 break; 1209 1210 case SO_RCVLOWAT: 1211 v.val = sk->sk_rcvlowat; 1212 break; 1213 1214 case SO_SNDLOWAT: 1215 v.val = 1; 1216 break; 1217 1218 case SO_PASSCRED: 1219 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags); 1220 break; 1221 1222 case SO_PEERCRED: 1223 { 1224 struct ucred peercred; 1225 if (len > sizeof(peercred)) 1226 len = sizeof(peercred); 1227 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred); 1228 if (copy_to_user(optval, &peercred, len)) 1229 return -EFAULT; 1230 goto lenout; 1231 } 1232 1233 case SO_PEERNAME: 1234 { 1235 char address[128]; 1236 1237 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2)) 1238 return -ENOTCONN; 1239 if (lv < len) 1240 return -EINVAL; 1241 if (copy_to_user(optval, address, len)) 1242 return -EFAULT; 1243 goto lenout; 1244 } 1245 1246 /* Dubious BSD thing... Probably nobody even uses it, but 1247 * the UNIX standard wants it for whatever reason... -DaveM 1248 */ 1249 case SO_ACCEPTCONN: 1250 v.val = sk->sk_state == TCP_LISTEN; 1251 break; 1252 1253 case SO_PASSSEC: 1254 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags); 1255 break; 1256 1257 case SO_PEERSEC: 1258 return security_socket_getpeersec_stream(sock, optval, optlen, len); 1259 1260 case SO_MARK: 1261 v.val = sk->sk_mark; 1262 break; 1263 1264 case SO_RXQ_OVFL: 1265 v.val = sock_flag(sk, SOCK_RXQ_OVFL); 1266 break; 1267 1268 case SO_WIFI_STATUS: 1269 v.val = sock_flag(sk, SOCK_WIFI_STATUS); 1270 break; 1271 1272 case SO_PEEK_OFF: 1273 if (!sock->ops->set_peek_off) 1274 return -EOPNOTSUPP; 1275 1276 v.val = sk->sk_peek_off; 1277 break; 1278 case SO_NOFCS: 1279 v.val = sock_flag(sk, SOCK_NOFCS); 1280 break; 1281 1282 case SO_BINDTODEVICE: 1283 return sock_getbindtodevice(sk, optval, optlen, len); 1284 1285 case SO_GET_FILTER: 1286 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len); 1287 if (len < 0) 1288 return len; 1289 1290 goto lenout; 1291 1292 case SO_LOCK_FILTER: 1293 v.val = sock_flag(sk, SOCK_FILTER_LOCKED); 1294 break; 1295 1296 case SO_BPF_EXTENSIONS: 1297 v.val = bpf_tell_extensions(); 1298 break; 1299 1300 case SO_SELECT_ERR_QUEUE: 1301 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE); 1302 break; 1303 1304 #ifdef CONFIG_NET_RX_BUSY_POLL 1305 case SO_BUSY_POLL: 1306 v.val = sk->sk_ll_usec; 1307 break; 1308 #endif 1309 1310 case SO_MAX_PACING_RATE: 1311 v.val = sk->sk_max_pacing_rate; 1312 break; 1313 1314 case SO_INCOMING_CPU: 1315 v.val = sk->sk_incoming_cpu; 1316 break; 1317 1318 case SO_MEMINFO: 1319 { 1320 u32 meminfo[SK_MEMINFO_VARS]; 1321 1322 if (get_user(len, optlen)) 1323 return -EFAULT; 1324 1325 sk_get_meminfo(sk, meminfo); 1326 1327 len = min_t(unsigned int, len, sizeof(meminfo)); 1328 if (copy_to_user(optval, &meminfo, len)) 1329 return -EFAULT; 1330 1331 goto lenout; 1332 } 1333 1334 #ifdef CONFIG_NET_RX_BUSY_POLL 1335 case SO_INCOMING_NAPI_ID: 1336 v.val = READ_ONCE(sk->sk_napi_id); 1337 1338 /* aggregate non-NAPI IDs down to 0 */ 1339 if (v.val < MIN_NAPI_ID) 1340 v.val = 0; 1341 1342 break; 1343 #endif 1344 1345 case SO_COOKIE: 1346 lv = sizeof(u64); 1347 if (len < lv) 1348 return -EINVAL; 1349 v.val64 = sock_gen_cookie(sk); 1350 break; 1351 1352 default: 1353 /* We implement the SO_SNDLOWAT etc to not be settable 1354 * (1003.1g 7). 1355 */ 1356 return -ENOPROTOOPT; 1357 } 1358 1359 if (len > lv) 1360 len = lv; 1361 if (copy_to_user(optval, &v, len)) 1362 return -EFAULT; 1363 lenout: 1364 if (put_user(len, optlen)) 1365 return -EFAULT; 1366 return 0; 1367 } 1368 1369 /* 1370 * Initialize an sk_lock. 1371 * 1372 * (We also register the sk_lock with the lock validator.) 1373 */ 1374 static inline void sock_lock_init(struct sock *sk) 1375 { 1376 if (sk->sk_kern_sock) 1377 sock_lock_init_class_and_name( 1378 sk, 1379 af_family_kern_slock_key_strings[sk->sk_family], 1380 af_family_kern_slock_keys + sk->sk_family, 1381 af_family_kern_key_strings[sk->sk_family], 1382 af_family_kern_keys + sk->sk_family); 1383 else 1384 sock_lock_init_class_and_name( 1385 sk, 1386 af_family_slock_key_strings[sk->sk_family], 1387 af_family_slock_keys + sk->sk_family, 1388 af_family_key_strings[sk->sk_family], 1389 af_family_keys + sk->sk_family); 1390 } 1391 1392 /* 1393 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet, 1394 * even temporarly, because of RCU lookups. sk_node should also be left as is. 1395 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end 1396 */ 1397 static void sock_copy(struct sock *nsk, const struct sock *osk) 1398 { 1399 #ifdef CONFIG_SECURITY_NETWORK 1400 void *sptr = nsk->sk_security; 1401 #endif 1402 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin)); 1403 1404 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end, 1405 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end)); 1406 1407 #ifdef CONFIG_SECURITY_NETWORK 1408 nsk->sk_security = sptr; 1409 security_sk_clone(osk, nsk); 1410 #endif 1411 } 1412 1413 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority, 1414 int family) 1415 { 1416 struct sock *sk; 1417 struct kmem_cache *slab; 1418 1419 slab = prot->slab; 1420 if (slab != NULL) { 1421 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO); 1422 if (!sk) 1423 return sk; 1424 if (priority & __GFP_ZERO) 1425 sk_prot_clear_nulls(sk, prot->obj_size); 1426 } else 1427 sk = kmalloc(prot->obj_size, priority); 1428 1429 if (sk != NULL) { 1430 kmemcheck_annotate_bitfield(sk, flags); 1431 1432 if (security_sk_alloc(sk, family, priority)) 1433 goto out_free; 1434 1435 if (!try_module_get(prot->owner)) 1436 goto out_free_sec; 1437 sk_tx_queue_clear(sk); 1438 } 1439 1440 return sk; 1441 1442 out_free_sec: 1443 security_sk_free(sk); 1444 out_free: 1445 if (slab != NULL) 1446 kmem_cache_free(slab, sk); 1447 else 1448 kfree(sk); 1449 return NULL; 1450 } 1451 1452 static void sk_prot_free(struct proto *prot, struct sock *sk) 1453 { 1454 struct kmem_cache *slab; 1455 struct module *owner; 1456 1457 owner = prot->owner; 1458 slab = prot->slab; 1459 1460 cgroup_sk_free(&sk->sk_cgrp_data); 1461 mem_cgroup_sk_free(sk); 1462 security_sk_free(sk); 1463 if (slab != NULL) 1464 kmem_cache_free(slab, sk); 1465 else 1466 kfree(sk); 1467 module_put(owner); 1468 } 1469 1470 /** 1471 * sk_alloc - All socket objects are allocated here 1472 * @net: the applicable net namespace 1473 * @family: protocol family 1474 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 1475 * @prot: struct proto associated with this new sock instance 1476 * @kern: is this to be a kernel socket? 1477 */ 1478 struct sock *sk_alloc(struct net *net, int family, gfp_t priority, 1479 struct proto *prot, int kern) 1480 { 1481 struct sock *sk; 1482 1483 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family); 1484 if (sk) { 1485 sk->sk_family = family; 1486 /* 1487 * See comment in struct sock definition to understand 1488 * why we need sk_prot_creator -acme 1489 */ 1490 sk->sk_prot = sk->sk_prot_creator = prot; 1491 sk->sk_kern_sock = kern; 1492 sock_lock_init(sk); 1493 sk->sk_net_refcnt = kern ? 0 : 1; 1494 if (likely(sk->sk_net_refcnt)) 1495 get_net(net); 1496 sock_net_set(sk, net); 1497 atomic_set(&sk->sk_wmem_alloc, 1); 1498 1499 mem_cgroup_sk_alloc(sk); 1500 cgroup_sk_alloc(&sk->sk_cgrp_data); 1501 sock_update_classid(&sk->sk_cgrp_data); 1502 sock_update_netprioidx(&sk->sk_cgrp_data); 1503 } 1504 1505 return sk; 1506 } 1507 EXPORT_SYMBOL(sk_alloc); 1508 1509 /* Sockets having SOCK_RCU_FREE will call this function after one RCU 1510 * grace period. This is the case for UDP sockets and TCP listeners. 1511 */ 1512 static void __sk_destruct(struct rcu_head *head) 1513 { 1514 struct sock *sk = container_of(head, struct sock, sk_rcu); 1515 struct sk_filter *filter; 1516 1517 if (sk->sk_destruct) 1518 sk->sk_destruct(sk); 1519 1520 filter = rcu_dereference_check(sk->sk_filter, 1521 atomic_read(&sk->sk_wmem_alloc) == 0); 1522 if (filter) { 1523 sk_filter_uncharge(sk, filter); 1524 RCU_INIT_POINTER(sk->sk_filter, NULL); 1525 } 1526 if (rcu_access_pointer(sk->sk_reuseport_cb)) 1527 reuseport_detach_sock(sk); 1528 1529 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP); 1530 1531 if (atomic_read(&sk->sk_omem_alloc)) 1532 pr_debug("%s: optmem leakage (%d bytes) detected\n", 1533 __func__, atomic_read(&sk->sk_omem_alloc)); 1534 1535 if (sk->sk_frag.page) { 1536 put_page(sk->sk_frag.page); 1537 sk->sk_frag.page = NULL; 1538 } 1539 1540 if (sk->sk_peer_cred) 1541 put_cred(sk->sk_peer_cred); 1542 put_pid(sk->sk_peer_pid); 1543 if (likely(sk->sk_net_refcnt)) 1544 put_net(sock_net(sk)); 1545 sk_prot_free(sk->sk_prot_creator, sk); 1546 } 1547 1548 void sk_destruct(struct sock *sk) 1549 { 1550 if (sock_flag(sk, SOCK_RCU_FREE)) 1551 call_rcu(&sk->sk_rcu, __sk_destruct); 1552 else 1553 __sk_destruct(&sk->sk_rcu); 1554 } 1555 1556 static void __sk_free(struct sock *sk) 1557 { 1558 if (unlikely(sock_diag_has_destroy_listeners(sk) && sk->sk_net_refcnt)) 1559 sock_diag_broadcast_destroy(sk); 1560 else 1561 sk_destruct(sk); 1562 } 1563 1564 void sk_free(struct sock *sk) 1565 { 1566 /* 1567 * We subtract one from sk_wmem_alloc and can know if 1568 * some packets are still in some tx queue. 1569 * If not null, sock_wfree() will call __sk_free(sk) later 1570 */ 1571 if (atomic_dec_and_test(&sk->sk_wmem_alloc)) 1572 __sk_free(sk); 1573 } 1574 EXPORT_SYMBOL(sk_free); 1575 1576 static void sk_init_common(struct sock *sk) 1577 { 1578 skb_queue_head_init(&sk->sk_receive_queue); 1579 skb_queue_head_init(&sk->sk_write_queue); 1580 skb_queue_head_init(&sk->sk_error_queue); 1581 1582 rwlock_init(&sk->sk_callback_lock); 1583 lockdep_set_class_and_name(&sk->sk_receive_queue.lock, 1584 af_rlock_keys + sk->sk_family, 1585 af_family_rlock_key_strings[sk->sk_family]); 1586 lockdep_set_class_and_name(&sk->sk_write_queue.lock, 1587 af_wlock_keys + sk->sk_family, 1588 af_family_wlock_key_strings[sk->sk_family]); 1589 lockdep_set_class_and_name(&sk->sk_error_queue.lock, 1590 af_elock_keys + sk->sk_family, 1591 af_family_elock_key_strings[sk->sk_family]); 1592 lockdep_set_class_and_name(&sk->sk_callback_lock, 1593 af_callback_keys + sk->sk_family, 1594 af_family_clock_key_strings[sk->sk_family]); 1595 } 1596 1597 /** 1598 * sk_clone_lock - clone a socket, and lock its clone 1599 * @sk: the socket to clone 1600 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 1601 * 1602 * Caller must unlock socket even in error path (bh_unlock_sock(newsk)) 1603 */ 1604 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority) 1605 { 1606 struct sock *newsk; 1607 bool is_charged = true; 1608 1609 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family); 1610 if (newsk != NULL) { 1611 struct sk_filter *filter; 1612 1613 sock_copy(newsk, sk); 1614 1615 /* SANITY */ 1616 if (likely(newsk->sk_net_refcnt)) 1617 get_net(sock_net(newsk)); 1618 sk_node_init(&newsk->sk_node); 1619 sock_lock_init(newsk); 1620 bh_lock_sock(newsk); 1621 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL; 1622 newsk->sk_backlog.len = 0; 1623 1624 atomic_set(&newsk->sk_rmem_alloc, 0); 1625 /* 1626 * sk_wmem_alloc set to one (see sk_free() and sock_wfree()) 1627 */ 1628 atomic_set(&newsk->sk_wmem_alloc, 1); 1629 atomic_set(&newsk->sk_omem_alloc, 0); 1630 sk_init_common(newsk); 1631 1632 newsk->sk_dst_cache = NULL; 1633 newsk->sk_dst_pending_confirm = 0; 1634 newsk->sk_wmem_queued = 0; 1635 newsk->sk_forward_alloc = 0; 1636 atomic_set(&newsk->sk_drops, 0); 1637 newsk->sk_send_head = NULL; 1638 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK; 1639 1640 sock_reset_flag(newsk, SOCK_DONE); 1641 1642 filter = rcu_dereference_protected(newsk->sk_filter, 1); 1643 if (filter != NULL) 1644 /* though it's an empty new sock, the charging may fail 1645 * if sysctl_optmem_max was changed between creation of 1646 * original socket and cloning 1647 */ 1648 is_charged = sk_filter_charge(newsk, filter); 1649 1650 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) { 1651 /* We need to make sure that we don't uncharge the new 1652 * socket if we couldn't charge it in the first place 1653 * as otherwise we uncharge the parent's filter. 1654 */ 1655 if (!is_charged) 1656 RCU_INIT_POINTER(newsk->sk_filter, NULL); 1657 sk_free_unlock_clone(newsk); 1658 newsk = NULL; 1659 goto out; 1660 } 1661 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL); 1662 1663 newsk->sk_err = 0; 1664 newsk->sk_err_soft = 0; 1665 newsk->sk_priority = 0; 1666 newsk->sk_incoming_cpu = raw_smp_processor_id(); 1667 atomic64_set(&newsk->sk_cookie, 0); 1668 1669 mem_cgroup_sk_alloc(newsk); 1670 cgroup_sk_alloc(&newsk->sk_cgrp_data); 1671 1672 /* 1673 * Before updating sk_refcnt, we must commit prior changes to memory 1674 * (Documentation/RCU/rculist_nulls.txt for details) 1675 */ 1676 smp_wmb(); 1677 atomic_set(&newsk->sk_refcnt, 2); 1678 1679 /* 1680 * Increment the counter in the same struct proto as the master 1681 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that 1682 * is the same as sk->sk_prot->socks, as this field was copied 1683 * with memcpy). 1684 * 1685 * This _changes_ the previous behaviour, where 1686 * tcp_create_openreq_child always was incrementing the 1687 * equivalent to tcp_prot->socks (inet_sock_nr), so this have 1688 * to be taken into account in all callers. -acme 1689 */ 1690 sk_refcnt_debug_inc(newsk); 1691 sk_set_socket(newsk, NULL); 1692 newsk->sk_wq = NULL; 1693 1694 if (newsk->sk_prot->sockets_allocated) 1695 sk_sockets_allocated_inc(newsk); 1696 1697 if (sock_needs_netstamp(sk) && 1698 newsk->sk_flags & SK_FLAGS_TIMESTAMP) 1699 net_enable_timestamp(); 1700 } 1701 out: 1702 return newsk; 1703 } 1704 EXPORT_SYMBOL_GPL(sk_clone_lock); 1705 1706 void sk_free_unlock_clone(struct sock *sk) 1707 { 1708 /* It is still raw copy of parent, so invalidate 1709 * destructor and make plain sk_free() */ 1710 sk->sk_destruct = NULL; 1711 bh_unlock_sock(sk); 1712 sk_free(sk); 1713 } 1714 EXPORT_SYMBOL_GPL(sk_free_unlock_clone); 1715 1716 void sk_setup_caps(struct sock *sk, struct dst_entry *dst) 1717 { 1718 u32 max_segs = 1; 1719 1720 sk_dst_set(sk, dst); 1721 sk->sk_route_caps = dst->dev->features; 1722 if (sk->sk_route_caps & NETIF_F_GSO) 1723 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE; 1724 sk->sk_route_caps &= ~sk->sk_route_nocaps; 1725 if (sk_can_gso(sk)) { 1726 if (dst->header_len) { 1727 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 1728 } else { 1729 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM; 1730 sk->sk_gso_max_size = dst->dev->gso_max_size; 1731 max_segs = max_t(u32, dst->dev->gso_max_segs, 1); 1732 } 1733 } 1734 sk->sk_gso_max_segs = max_segs; 1735 } 1736 EXPORT_SYMBOL_GPL(sk_setup_caps); 1737 1738 /* 1739 * Simple resource managers for sockets. 1740 */ 1741 1742 1743 /* 1744 * Write buffer destructor automatically called from kfree_skb. 1745 */ 1746 void sock_wfree(struct sk_buff *skb) 1747 { 1748 struct sock *sk = skb->sk; 1749 unsigned int len = skb->truesize; 1750 1751 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) { 1752 /* 1753 * Keep a reference on sk_wmem_alloc, this will be released 1754 * after sk_write_space() call 1755 */ 1756 atomic_sub(len - 1, &sk->sk_wmem_alloc); 1757 sk->sk_write_space(sk); 1758 len = 1; 1759 } 1760 /* 1761 * if sk_wmem_alloc reaches 0, we must finish what sk_free() 1762 * could not do because of in-flight packets 1763 */ 1764 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc)) 1765 __sk_free(sk); 1766 } 1767 EXPORT_SYMBOL(sock_wfree); 1768 1769 /* This variant of sock_wfree() is used by TCP, 1770 * since it sets SOCK_USE_WRITE_QUEUE. 1771 */ 1772 void __sock_wfree(struct sk_buff *skb) 1773 { 1774 struct sock *sk = skb->sk; 1775 1776 if (atomic_sub_and_test(skb->truesize, &sk->sk_wmem_alloc)) 1777 __sk_free(sk); 1778 } 1779 1780 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk) 1781 { 1782 skb_orphan(skb); 1783 skb->sk = sk; 1784 #ifdef CONFIG_INET 1785 if (unlikely(!sk_fullsock(sk))) { 1786 skb->destructor = sock_edemux; 1787 sock_hold(sk); 1788 return; 1789 } 1790 #endif 1791 skb->destructor = sock_wfree; 1792 skb_set_hash_from_sk(skb, sk); 1793 /* 1794 * We used to take a refcount on sk, but following operation 1795 * is enough to guarantee sk_free() wont free this sock until 1796 * all in-flight packets are completed 1797 */ 1798 atomic_add(skb->truesize, &sk->sk_wmem_alloc); 1799 } 1800 EXPORT_SYMBOL(skb_set_owner_w); 1801 1802 /* This helper is used by netem, as it can hold packets in its 1803 * delay queue. We want to allow the owner socket to send more 1804 * packets, as if they were already TX completed by a typical driver. 1805 * But we also want to keep skb->sk set because some packet schedulers 1806 * rely on it (sch_fq for example). So we set skb->truesize to a small 1807 * amount (1) and decrease sk_wmem_alloc accordingly. 1808 */ 1809 void skb_orphan_partial(struct sk_buff *skb) 1810 { 1811 /* If this skb is a TCP pure ACK or already went here, 1812 * we have nothing to do. 2 is already a very small truesize. 1813 */ 1814 if (skb->truesize <= 2) 1815 return; 1816 1817 /* TCP stack sets skb->ooo_okay based on sk_wmem_alloc, 1818 * so we do not completely orphan skb, but transfert all 1819 * accounted bytes but one, to avoid unexpected reorders. 1820 */ 1821 if (skb->destructor == sock_wfree 1822 #ifdef CONFIG_INET 1823 || skb->destructor == tcp_wfree 1824 #endif 1825 ) { 1826 atomic_sub(skb->truesize - 1, &skb->sk->sk_wmem_alloc); 1827 skb->truesize = 1; 1828 } else { 1829 skb_orphan(skb); 1830 } 1831 } 1832 EXPORT_SYMBOL(skb_orphan_partial); 1833 1834 /* 1835 * Read buffer destructor automatically called from kfree_skb. 1836 */ 1837 void sock_rfree(struct sk_buff *skb) 1838 { 1839 struct sock *sk = skb->sk; 1840 unsigned int len = skb->truesize; 1841 1842 atomic_sub(len, &sk->sk_rmem_alloc); 1843 sk_mem_uncharge(sk, len); 1844 } 1845 EXPORT_SYMBOL(sock_rfree); 1846 1847 /* 1848 * Buffer destructor for skbs that are not used directly in read or write 1849 * path, e.g. for error handler skbs. Automatically called from kfree_skb. 1850 */ 1851 void sock_efree(struct sk_buff *skb) 1852 { 1853 sock_put(skb->sk); 1854 } 1855 EXPORT_SYMBOL(sock_efree); 1856 1857 kuid_t sock_i_uid(struct sock *sk) 1858 { 1859 kuid_t uid; 1860 1861 read_lock_bh(&sk->sk_callback_lock); 1862 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID; 1863 read_unlock_bh(&sk->sk_callback_lock); 1864 return uid; 1865 } 1866 EXPORT_SYMBOL(sock_i_uid); 1867 1868 unsigned long sock_i_ino(struct sock *sk) 1869 { 1870 unsigned long ino; 1871 1872 read_lock_bh(&sk->sk_callback_lock); 1873 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0; 1874 read_unlock_bh(&sk->sk_callback_lock); 1875 return ino; 1876 } 1877 EXPORT_SYMBOL(sock_i_ino); 1878 1879 /* 1880 * Allocate a skb from the socket's send buffer. 1881 */ 1882 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 1883 gfp_t priority) 1884 { 1885 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) { 1886 struct sk_buff *skb = alloc_skb(size, priority); 1887 if (skb) { 1888 skb_set_owner_w(skb, sk); 1889 return skb; 1890 } 1891 } 1892 return NULL; 1893 } 1894 EXPORT_SYMBOL(sock_wmalloc); 1895 1896 /* 1897 * Allocate a memory block from the socket's option memory buffer. 1898 */ 1899 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority) 1900 { 1901 if ((unsigned int)size <= sysctl_optmem_max && 1902 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) { 1903 void *mem; 1904 /* First do the add, to avoid the race if kmalloc 1905 * might sleep. 1906 */ 1907 atomic_add(size, &sk->sk_omem_alloc); 1908 mem = kmalloc(size, priority); 1909 if (mem) 1910 return mem; 1911 atomic_sub(size, &sk->sk_omem_alloc); 1912 } 1913 return NULL; 1914 } 1915 EXPORT_SYMBOL(sock_kmalloc); 1916 1917 /* Free an option memory block. Note, we actually want the inline 1918 * here as this allows gcc to detect the nullify and fold away the 1919 * condition entirely. 1920 */ 1921 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size, 1922 const bool nullify) 1923 { 1924 if (WARN_ON_ONCE(!mem)) 1925 return; 1926 if (nullify) 1927 kzfree(mem); 1928 else 1929 kfree(mem); 1930 atomic_sub(size, &sk->sk_omem_alloc); 1931 } 1932 1933 void sock_kfree_s(struct sock *sk, void *mem, int size) 1934 { 1935 __sock_kfree_s(sk, mem, size, false); 1936 } 1937 EXPORT_SYMBOL(sock_kfree_s); 1938 1939 void sock_kzfree_s(struct sock *sk, void *mem, int size) 1940 { 1941 __sock_kfree_s(sk, mem, size, true); 1942 } 1943 EXPORT_SYMBOL(sock_kzfree_s); 1944 1945 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock. 1946 I think, these locks should be removed for datagram sockets. 1947 */ 1948 static long sock_wait_for_wmem(struct sock *sk, long timeo) 1949 { 1950 DEFINE_WAIT(wait); 1951 1952 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk); 1953 for (;;) { 1954 if (!timeo) 1955 break; 1956 if (signal_pending(current)) 1957 break; 1958 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 1959 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 1960 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) 1961 break; 1962 if (sk->sk_shutdown & SEND_SHUTDOWN) 1963 break; 1964 if (sk->sk_err) 1965 break; 1966 timeo = schedule_timeout(timeo); 1967 } 1968 finish_wait(sk_sleep(sk), &wait); 1969 return timeo; 1970 } 1971 1972 1973 /* 1974 * Generic send/receive buffer handlers 1975 */ 1976 1977 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 1978 unsigned long data_len, int noblock, 1979 int *errcode, int max_page_order) 1980 { 1981 struct sk_buff *skb; 1982 long timeo; 1983 int err; 1984 1985 timeo = sock_sndtimeo(sk, noblock); 1986 for (;;) { 1987 err = sock_error(sk); 1988 if (err != 0) 1989 goto failure; 1990 1991 err = -EPIPE; 1992 if (sk->sk_shutdown & SEND_SHUTDOWN) 1993 goto failure; 1994 1995 if (sk_wmem_alloc_get(sk) < sk->sk_sndbuf) 1996 break; 1997 1998 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk); 1999 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 2000 err = -EAGAIN; 2001 if (!timeo) 2002 goto failure; 2003 if (signal_pending(current)) 2004 goto interrupted; 2005 timeo = sock_wait_for_wmem(sk, timeo); 2006 } 2007 skb = alloc_skb_with_frags(header_len, data_len, max_page_order, 2008 errcode, sk->sk_allocation); 2009 if (skb) 2010 skb_set_owner_w(skb, sk); 2011 return skb; 2012 2013 interrupted: 2014 err = sock_intr_errno(timeo); 2015 failure: 2016 *errcode = err; 2017 return NULL; 2018 } 2019 EXPORT_SYMBOL(sock_alloc_send_pskb); 2020 2021 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size, 2022 int noblock, int *errcode) 2023 { 2024 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0); 2025 } 2026 EXPORT_SYMBOL(sock_alloc_send_skb); 2027 2028 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg, 2029 struct sockcm_cookie *sockc) 2030 { 2031 u32 tsflags; 2032 2033 switch (cmsg->cmsg_type) { 2034 case SO_MARK: 2035 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) 2036 return -EPERM; 2037 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 2038 return -EINVAL; 2039 sockc->mark = *(u32 *)CMSG_DATA(cmsg); 2040 break; 2041 case SO_TIMESTAMPING: 2042 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 2043 return -EINVAL; 2044 2045 tsflags = *(u32 *)CMSG_DATA(cmsg); 2046 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK) 2047 return -EINVAL; 2048 2049 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK; 2050 sockc->tsflags |= tsflags; 2051 break; 2052 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */ 2053 case SCM_RIGHTS: 2054 case SCM_CREDENTIALS: 2055 break; 2056 default: 2057 return -EINVAL; 2058 } 2059 return 0; 2060 } 2061 EXPORT_SYMBOL(__sock_cmsg_send); 2062 2063 int sock_cmsg_send(struct sock *sk, struct msghdr *msg, 2064 struct sockcm_cookie *sockc) 2065 { 2066 struct cmsghdr *cmsg; 2067 int ret; 2068 2069 for_each_cmsghdr(cmsg, msg) { 2070 if (!CMSG_OK(msg, cmsg)) 2071 return -EINVAL; 2072 if (cmsg->cmsg_level != SOL_SOCKET) 2073 continue; 2074 ret = __sock_cmsg_send(sk, msg, cmsg, sockc); 2075 if (ret) 2076 return ret; 2077 } 2078 return 0; 2079 } 2080 EXPORT_SYMBOL(sock_cmsg_send); 2081 2082 /* On 32bit arches, an skb frag is limited to 2^15 */ 2083 #define SKB_FRAG_PAGE_ORDER get_order(32768) 2084 2085 /** 2086 * skb_page_frag_refill - check that a page_frag contains enough room 2087 * @sz: minimum size of the fragment we want to get 2088 * @pfrag: pointer to page_frag 2089 * @gfp: priority for memory allocation 2090 * 2091 * Note: While this allocator tries to use high order pages, there is 2092 * no guarantee that allocations succeed. Therefore, @sz MUST be 2093 * less or equal than PAGE_SIZE. 2094 */ 2095 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp) 2096 { 2097 if (pfrag->page) { 2098 if (page_ref_count(pfrag->page) == 1) { 2099 pfrag->offset = 0; 2100 return true; 2101 } 2102 if (pfrag->offset + sz <= pfrag->size) 2103 return true; 2104 put_page(pfrag->page); 2105 } 2106 2107 pfrag->offset = 0; 2108 if (SKB_FRAG_PAGE_ORDER) { 2109 /* Avoid direct reclaim but allow kswapd to wake */ 2110 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) | 2111 __GFP_COMP | __GFP_NOWARN | 2112 __GFP_NORETRY, 2113 SKB_FRAG_PAGE_ORDER); 2114 if (likely(pfrag->page)) { 2115 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER; 2116 return true; 2117 } 2118 } 2119 pfrag->page = alloc_page(gfp); 2120 if (likely(pfrag->page)) { 2121 pfrag->size = PAGE_SIZE; 2122 return true; 2123 } 2124 return false; 2125 } 2126 EXPORT_SYMBOL(skb_page_frag_refill); 2127 2128 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag) 2129 { 2130 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation))) 2131 return true; 2132 2133 sk_enter_memory_pressure(sk); 2134 sk_stream_moderate_sndbuf(sk); 2135 return false; 2136 } 2137 EXPORT_SYMBOL(sk_page_frag_refill); 2138 2139 static void __lock_sock(struct sock *sk) 2140 __releases(&sk->sk_lock.slock) 2141 __acquires(&sk->sk_lock.slock) 2142 { 2143 DEFINE_WAIT(wait); 2144 2145 for (;;) { 2146 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait, 2147 TASK_UNINTERRUPTIBLE); 2148 spin_unlock_bh(&sk->sk_lock.slock); 2149 schedule(); 2150 spin_lock_bh(&sk->sk_lock.slock); 2151 if (!sock_owned_by_user(sk)) 2152 break; 2153 } 2154 finish_wait(&sk->sk_lock.wq, &wait); 2155 } 2156 2157 static void __release_sock(struct sock *sk) 2158 __releases(&sk->sk_lock.slock) 2159 __acquires(&sk->sk_lock.slock) 2160 { 2161 struct sk_buff *skb, *next; 2162 2163 while ((skb = sk->sk_backlog.head) != NULL) { 2164 sk->sk_backlog.head = sk->sk_backlog.tail = NULL; 2165 2166 spin_unlock_bh(&sk->sk_lock.slock); 2167 2168 do { 2169 next = skb->next; 2170 prefetch(next); 2171 WARN_ON_ONCE(skb_dst_is_noref(skb)); 2172 skb->next = NULL; 2173 sk_backlog_rcv(sk, skb); 2174 2175 cond_resched(); 2176 2177 skb = next; 2178 } while (skb != NULL); 2179 2180 spin_lock_bh(&sk->sk_lock.slock); 2181 } 2182 2183 /* 2184 * Doing the zeroing here guarantee we can not loop forever 2185 * while a wild producer attempts to flood us. 2186 */ 2187 sk->sk_backlog.len = 0; 2188 } 2189 2190 void __sk_flush_backlog(struct sock *sk) 2191 { 2192 spin_lock_bh(&sk->sk_lock.slock); 2193 __release_sock(sk); 2194 spin_unlock_bh(&sk->sk_lock.slock); 2195 } 2196 2197 /** 2198 * sk_wait_data - wait for data to arrive at sk_receive_queue 2199 * @sk: sock to wait on 2200 * @timeo: for how long 2201 * @skb: last skb seen on sk_receive_queue 2202 * 2203 * Now socket state including sk->sk_err is changed only under lock, 2204 * hence we may omit checks after joining wait queue. 2205 * We check receive queue before schedule() only as optimization; 2206 * it is very likely that release_sock() added new data. 2207 */ 2208 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb) 2209 { 2210 DEFINE_WAIT_FUNC(wait, woken_wake_function); 2211 int rc; 2212 2213 add_wait_queue(sk_sleep(sk), &wait); 2214 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk); 2215 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait); 2216 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk); 2217 remove_wait_queue(sk_sleep(sk), &wait); 2218 return rc; 2219 } 2220 EXPORT_SYMBOL(sk_wait_data); 2221 2222 /** 2223 * __sk_mem_raise_allocated - increase memory_allocated 2224 * @sk: socket 2225 * @size: memory size to allocate 2226 * @amt: pages to allocate 2227 * @kind: allocation type 2228 * 2229 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc 2230 */ 2231 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind) 2232 { 2233 struct proto *prot = sk->sk_prot; 2234 long allocated = sk_memory_allocated_add(sk, amt); 2235 2236 if (mem_cgroup_sockets_enabled && sk->sk_memcg && 2237 !mem_cgroup_charge_skmem(sk->sk_memcg, amt)) 2238 goto suppress_allocation; 2239 2240 /* Under limit. */ 2241 if (allocated <= sk_prot_mem_limits(sk, 0)) { 2242 sk_leave_memory_pressure(sk); 2243 return 1; 2244 } 2245 2246 /* Under pressure. */ 2247 if (allocated > sk_prot_mem_limits(sk, 1)) 2248 sk_enter_memory_pressure(sk); 2249 2250 /* Over hard limit. */ 2251 if (allocated > sk_prot_mem_limits(sk, 2)) 2252 goto suppress_allocation; 2253 2254 /* guarantee minimum buffer size under pressure */ 2255 if (kind == SK_MEM_RECV) { 2256 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0]) 2257 return 1; 2258 2259 } else { /* SK_MEM_SEND */ 2260 if (sk->sk_type == SOCK_STREAM) { 2261 if (sk->sk_wmem_queued < prot->sysctl_wmem[0]) 2262 return 1; 2263 } else if (atomic_read(&sk->sk_wmem_alloc) < 2264 prot->sysctl_wmem[0]) 2265 return 1; 2266 } 2267 2268 if (sk_has_memory_pressure(sk)) { 2269 int alloc; 2270 2271 if (!sk_under_memory_pressure(sk)) 2272 return 1; 2273 alloc = sk_sockets_allocated_read_positive(sk); 2274 if (sk_prot_mem_limits(sk, 2) > alloc * 2275 sk_mem_pages(sk->sk_wmem_queued + 2276 atomic_read(&sk->sk_rmem_alloc) + 2277 sk->sk_forward_alloc)) 2278 return 1; 2279 } 2280 2281 suppress_allocation: 2282 2283 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) { 2284 sk_stream_moderate_sndbuf(sk); 2285 2286 /* Fail only if socket is _under_ its sndbuf. 2287 * In this case we cannot block, so that we have to fail. 2288 */ 2289 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf) 2290 return 1; 2291 } 2292 2293 trace_sock_exceed_buf_limit(sk, prot, allocated); 2294 2295 sk_memory_allocated_sub(sk, amt); 2296 2297 if (mem_cgroup_sockets_enabled && sk->sk_memcg) 2298 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt); 2299 2300 return 0; 2301 } 2302 EXPORT_SYMBOL(__sk_mem_raise_allocated); 2303 2304 /** 2305 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated 2306 * @sk: socket 2307 * @size: memory size to allocate 2308 * @kind: allocation type 2309 * 2310 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means 2311 * rmem allocation. This function assumes that protocols which have 2312 * memory_pressure use sk_wmem_queued as write buffer accounting. 2313 */ 2314 int __sk_mem_schedule(struct sock *sk, int size, int kind) 2315 { 2316 int ret, amt = sk_mem_pages(size); 2317 2318 sk->sk_forward_alloc += amt << SK_MEM_QUANTUM_SHIFT; 2319 ret = __sk_mem_raise_allocated(sk, size, amt, kind); 2320 if (!ret) 2321 sk->sk_forward_alloc -= amt << SK_MEM_QUANTUM_SHIFT; 2322 return ret; 2323 } 2324 EXPORT_SYMBOL(__sk_mem_schedule); 2325 2326 /** 2327 * __sk_mem_reduce_allocated - reclaim memory_allocated 2328 * @sk: socket 2329 * @amount: number of quanta 2330 * 2331 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc 2332 */ 2333 void __sk_mem_reduce_allocated(struct sock *sk, int amount) 2334 { 2335 sk_memory_allocated_sub(sk, amount); 2336 2337 if (mem_cgroup_sockets_enabled && sk->sk_memcg) 2338 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount); 2339 2340 if (sk_under_memory_pressure(sk) && 2341 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0))) 2342 sk_leave_memory_pressure(sk); 2343 } 2344 EXPORT_SYMBOL(__sk_mem_reduce_allocated); 2345 2346 /** 2347 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated 2348 * @sk: socket 2349 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple) 2350 */ 2351 void __sk_mem_reclaim(struct sock *sk, int amount) 2352 { 2353 amount >>= SK_MEM_QUANTUM_SHIFT; 2354 sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT; 2355 __sk_mem_reduce_allocated(sk, amount); 2356 } 2357 EXPORT_SYMBOL(__sk_mem_reclaim); 2358 2359 int sk_set_peek_off(struct sock *sk, int val) 2360 { 2361 if (val < 0) 2362 return -EINVAL; 2363 2364 sk->sk_peek_off = val; 2365 return 0; 2366 } 2367 EXPORT_SYMBOL_GPL(sk_set_peek_off); 2368 2369 /* 2370 * Set of default routines for initialising struct proto_ops when 2371 * the protocol does not support a particular function. In certain 2372 * cases where it makes no sense for a protocol to have a "do nothing" 2373 * function, some default processing is provided. 2374 */ 2375 2376 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len) 2377 { 2378 return -EOPNOTSUPP; 2379 } 2380 EXPORT_SYMBOL(sock_no_bind); 2381 2382 int sock_no_connect(struct socket *sock, struct sockaddr *saddr, 2383 int len, int flags) 2384 { 2385 return -EOPNOTSUPP; 2386 } 2387 EXPORT_SYMBOL(sock_no_connect); 2388 2389 int sock_no_socketpair(struct socket *sock1, struct socket *sock2) 2390 { 2391 return -EOPNOTSUPP; 2392 } 2393 EXPORT_SYMBOL(sock_no_socketpair); 2394 2395 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags, 2396 bool kern) 2397 { 2398 return -EOPNOTSUPP; 2399 } 2400 EXPORT_SYMBOL(sock_no_accept); 2401 2402 int sock_no_getname(struct socket *sock, struct sockaddr *saddr, 2403 int *len, int peer) 2404 { 2405 return -EOPNOTSUPP; 2406 } 2407 EXPORT_SYMBOL(sock_no_getname); 2408 2409 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt) 2410 { 2411 return 0; 2412 } 2413 EXPORT_SYMBOL(sock_no_poll); 2414 2415 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 2416 { 2417 return -EOPNOTSUPP; 2418 } 2419 EXPORT_SYMBOL(sock_no_ioctl); 2420 2421 int sock_no_listen(struct socket *sock, int backlog) 2422 { 2423 return -EOPNOTSUPP; 2424 } 2425 EXPORT_SYMBOL(sock_no_listen); 2426 2427 int sock_no_shutdown(struct socket *sock, int how) 2428 { 2429 return -EOPNOTSUPP; 2430 } 2431 EXPORT_SYMBOL(sock_no_shutdown); 2432 2433 int sock_no_setsockopt(struct socket *sock, int level, int optname, 2434 char __user *optval, unsigned int optlen) 2435 { 2436 return -EOPNOTSUPP; 2437 } 2438 EXPORT_SYMBOL(sock_no_setsockopt); 2439 2440 int sock_no_getsockopt(struct socket *sock, int level, int optname, 2441 char __user *optval, int __user *optlen) 2442 { 2443 return -EOPNOTSUPP; 2444 } 2445 EXPORT_SYMBOL(sock_no_getsockopt); 2446 2447 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len) 2448 { 2449 return -EOPNOTSUPP; 2450 } 2451 EXPORT_SYMBOL(sock_no_sendmsg); 2452 2453 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len, 2454 int flags) 2455 { 2456 return -EOPNOTSUPP; 2457 } 2458 EXPORT_SYMBOL(sock_no_recvmsg); 2459 2460 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma) 2461 { 2462 /* Mirror missing mmap method error code */ 2463 return -ENODEV; 2464 } 2465 EXPORT_SYMBOL(sock_no_mmap); 2466 2467 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags) 2468 { 2469 ssize_t res; 2470 struct msghdr msg = {.msg_flags = flags}; 2471 struct kvec iov; 2472 char *kaddr = kmap(page); 2473 iov.iov_base = kaddr + offset; 2474 iov.iov_len = size; 2475 res = kernel_sendmsg(sock, &msg, &iov, 1, size); 2476 kunmap(page); 2477 return res; 2478 } 2479 EXPORT_SYMBOL(sock_no_sendpage); 2480 2481 /* 2482 * Default Socket Callbacks 2483 */ 2484 2485 static void sock_def_wakeup(struct sock *sk) 2486 { 2487 struct socket_wq *wq; 2488 2489 rcu_read_lock(); 2490 wq = rcu_dereference(sk->sk_wq); 2491 if (skwq_has_sleeper(wq)) 2492 wake_up_interruptible_all(&wq->wait); 2493 rcu_read_unlock(); 2494 } 2495 2496 static void sock_def_error_report(struct sock *sk) 2497 { 2498 struct socket_wq *wq; 2499 2500 rcu_read_lock(); 2501 wq = rcu_dereference(sk->sk_wq); 2502 if (skwq_has_sleeper(wq)) 2503 wake_up_interruptible_poll(&wq->wait, POLLERR); 2504 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR); 2505 rcu_read_unlock(); 2506 } 2507 2508 static void sock_def_readable(struct sock *sk) 2509 { 2510 struct socket_wq *wq; 2511 2512 rcu_read_lock(); 2513 wq = rcu_dereference(sk->sk_wq); 2514 if (skwq_has_sleeper(wq)) 2515 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI | 2516 POLLRDNORM | POLLRDBAND); 2517 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN); 2518 rcu_read_unlock(); 2519 } 2520 2521 static void sock_def_write_space(struct sock *sk) 2522 { 2523 struct socket_wq *wq; 2524 2525 rcu_read_lock(); 2526 2527 /* Do not wake up a writer until he can make "significant" 2528 * progress. --DaveM 2529 */ 2530 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) { 2531 wq = rcu_dereference(sk->sk_wq); 2532 if (skwq_has_sleeper(wq)) 2533 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT | 2534 POLLWRNORM | POLLWRBAND); 2535 2536 /* Should agree with poll, otherwise some programs break */ 2537 if (sock_writeable(sk)) 2538 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT); 2539 } 2540 2541 rcu_read_unlock(); 2542 } 2543 2544 static void sock_def_destruct(struct sock *sk) 2545 { 2546 } 2547 2548 void sk_send_sigurg(struct sock *sk) 2549 { 2550 if (sk->sk_socket && sk->sk_socket->file) 2551 if (send_sigurg(&sk->sk_socket->file->f_owner)) 2552 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI); 2553 } 2554 EXPORT_SYMBOL(sk_send_sigurg); 2555 2556 void sk_reset_timer(struct sock *sk, struct timer_list* timer, 2557 unsigned long expires) 2558 { 2559 if (!mod_timer(timer, expires)) 2560 sock_hold(sk); 2561 } 2562 EXPORT_SYMBOL(sk_reset_timer); 2563 2564 void sk_stop_timer(struct sock *sk, struct timer_list* timer) 2565 { 2566 if (del_timer(timer)) 2567 __sock_put(sk); 2568 } 2569 EXPORT_SYMBOL(sk_stop_timer); 2570 2571 void sock_init_data(struct socket *sock, struct sock *sk) 2572 { 2573 sk_init_common(sk); 2574 sk->sk_send_head = NULL; 2575 2576 init_timer(&sk->sk_timer); 2577 2578 sk->sk_allocation = GFP_KERNEL; 2579 sk->sk_rcvbuf = sysctl_rmem_default; 2580 sk->sk_sndbuf = sysctl_wmem_default; 2581 sk->sk_state = TCP_CLOSE; 2582 sk_set_socket(sk, sock); 2583 2584 sock_set_flag(sk, SOCK_ZAPPED); 2585 2586 if (sock) { 2587 sk->sk_type = sock->type; 2588 sk->sk_wq = sock->wq; 2589 sock->sk = sk; 2590 sk->sk_uid = SOCK_INODE(sock)->i_uid; 2591 } else { 2592 sk->sk_wq = NULL; 2593 sk->sk_uid = make_kuid(sock_net(sk)->user_ns, 0); 2594 } 2595 2596 rwlock_init(&sk->sk_callback_lock); 2597 if (sk->sk_kern_sock) 2598 lockdep_set_class_and_name( 2599 &sk->sk_callback_lock, 2600 af_kern_callback_keys + sk->sk_family, 2601 af_family_kern_clock_key_strings[sk->sk_family]); 2602 else 2603 lockdep_set_class_and_name( 2604 &sk->sk_callback_lock, 2605 af_callback_keys + sk->sk_family, 2606 af_family_clock_key_strings[sk->sk_family]); 2607 2608 sk->sk_state_change = sock_def_wakeup; 2609 sk->sk_data_ready = sock_def_readable; 2610 sk->sk_write_space = sock_def_write_space; 2611 sk->sk_error_report = sock_def_error_report; 2612 sk->sk_destruct = sock_def_destruct; 2613 2614 sk->sk_frag.page = NULL; 2615 sk->sk_frag.offset = 0; 2616 sk->sk_peek_off = -1; 2617 2618 sk->sk_peer_pid = NULL; 2619 sk->sk_peer_cred = NULL; 2620 sk->sk_write_pending = 0; 2621 sk->sk_rcvlowat = 1; 2622 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; 2623 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT; 2624 2625 sk->sk_stamp = SK_DEFAULT_STAMP; 2626 2627 #ifdef CONFIG_NET_RX_BUSY_POLL 2628 sk->sk_napi_id = 0; 2629 sk->sk_ll_usec = sysctl_net_busy_read; 2630 #endif 2631 2632 sk->sk_max_pacing_rate = ~0U; 2633 sk->sk_pacing_rate = ~0U; 2634 sk->sk_incoming_cpu = -1; 2635 /* 2636 * Before updating sk_refcnt, we must commit prior changes to memory 2637 * (Documentation/RCU/rculist_nulls.txt for details) 2638 */ 2639 smp_wmb(); 2640 atomic_set(&sk->sk_refcnt, 1); 2641 atomic_set(&sk->sk_drops, 0); 2642 } 2643 EXPORT_SYMBOL(sock_init_data); 2644 2645 void lock_sock_nested(struct sock *sk, int subclass) 2646 { 2647 might_sleep(); 2648 spin_lock_bh(&sk->sk_lock.slock); 2649 if (sk->sk_lock.owned) 2650 __lock_sock(sk); 2651 sk->sk_lock.owned = 1; 2652 spin_unlock(&sk->sk_lock.slock); 2653 /* 2654 * The sk_lock has mutex_lock() semantics here: 2655 */ 2656 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_); 2657 local_bh_enable(); 2658 } 2659 EXPORT_SYMBOL(lock_sock_nested); 2660 2661 void release_sock(struct sock *sk) 2662 { 2663 spin_lock_bh(&sk->sk_lock.slock); 2664 if (sk->sk_backlog.tail) 2665 __release_sock(sk); 2666 2667 /* Warning : release_cb() might need to release sk ownership, 2668 * ie call sock_release_ownership(sk) before us. 2669 */ 2670 if (sk->sk_prot->release_cb) 2671 sk->sk_prot->release_cb(sk); 2672 2673 sock_release_ownership(sk); 2674 if (waitqueue_active(&sk->sk_lock.wq)) 2675 wake_up(&sk->sk_lock.wq); 2676 spin_unlock_bh(&sk->sk_lock.slock); 2677 } 2678 EXPORT_SYMBOL(release_sock); 2679 2680 /** 2681 * lock_sock_fast - fast version of lock_sock 2682 * @sk: socket 2683 * 2684 * This version should be used for very small section, where process wont block 2685 * return false if fast path is taken 2686 * sk_lock.slock locked, owned = 0, BH disabled 2687 * return true if slow path is taken 2688 * sk_lock.slock unlocked, owned = 1, BH enabled 2689 */ 2690 bool lock_sock_fast(struct sock *sk) 2691 { 2692 might_sleep(); 2693 spin_lock_bh(&sk->sk_lock.slock); 2694 2695 if (!sk->sk_lock.owned) 2696 /* 2697 * Note : We must disable BH 2698 */ 2699 return false; 2700 2701 __lock_sock(sk); 2702 sk->sk_lock.owned = 1; 2703 spin_unlock(&sk->sk_lock.slock); 2704 /* 2705 * The sk_lock has mutex_lock() semantics here: 2706 */ 2707 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_); 2708 local_bh_enable(); 2709 return true; 2710 } 2711 EXPORT_SYMBOL(lock_sock_fast); 2712 2713 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp) 2714 { 2715 struct timeval tv; 2716 if (!sock_flag(sk, SOCK_TIMESTAMP)) 2717 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 2718 tv = ktime_to_timeval(sk->sk_stamp); 2719 if (tv.tv_sec == -1) 2720 return -ENOENT; 2721 if (tv.tv_sec == 0) { 2722 sk->sk_stamp = ktime_get_real(); 2723 tv = ktime_to_timeval(sk->sk_stamp); 2724 } 2725 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0; 2726 } 2727 EXPORT_SYMBOL(sock_get_timestamp); 2728 2729 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp) 2730 { 2731 struct timespec ts; 2732 if (!sock_flag(sk, SOCK_TIMESTAMP)) 2733 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 2734 ts = ktime_to_timespec(sk->sk_stamp); 2735 if (ts.tv_sec == -1) 2736 return -ENOENT; 2737 if (ts.tv_sec == 0) { 2738 sk->sk_stamp = ktime_get_real(); 2739 ts = ktime_to_timespec(sk->sk_stamp); 2740 } 2741 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0; 2742 } 2743 EXPORT_SYMBOL(sock_get_timestampns); 2744 2745 void sock_enable_timestamp(struct sock *sk, int flag) 2746 { 2747 if (!sock_flag(sk, flag)) { 2748 unsigned long previous_flags = sk->sk_flags; 2749 2750 sock_set_flag(sk, flag); 2751 /* 2752 * we just set one of the two flags which require net 2753 * time stamping, but time stamping might have been on 2754 * already because of the other one 2755 */ 2756 if (sock_needs_netstamp(sk) && 2757 !(previous_flags & SK_FLAGS_TIMESTAMP)) 2758 net_enable_timestamp(); 2759 } 2760 } 2761 2762 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, 2763 int level, int type) 2764 { 2765 struct sock_exterr_skb *serr; 2766 struct sk_buff *skb; 2767 int copied, err; 2768 2769 err = -EAGAIN; 2770 skb = sock_dequeue_err_skb(sk); 2771 if (skb == NULL) 2772 goto out; 2773 2774 copied = skb->len; 2775 if (copied > len) { 2776 msg->msg_flags |= MSG_TRUNC; 2777 copied = len; 2778 } 2779 err = skb_copy_datagram_msg(skb, 0, msg, copied); 2780 if (err) 2781 goto out_free_skb; 2782 2783 sock_recv_timestamp(msg, sk, skb); 2784 2785 serr = SKB_EXT_ERR(skb); 2786 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee); 2787 2788 msg->msg_flags |= MSG_ERRQUEUE; 2789 err = copied; 2790 2791 out_free_skb: 2792 kfree_skb(skb); 2793 out: 2794 return err; 2795 } 2796 EXPORT_SYMBOL(sock_recv_errqueue); 2797 2798 /* 2799 * Get a socket option on an socket. 2800 * 2801 * FIX: POSIX 1003.1g is very ambiguous here. It states that 2802 * asynchronous errors should be reported by getsockopt. We assume 2803 * this means if you specify SO_ERROR (otherwise whats the point of it). 2804 */ 2805 int sock_common_getsockopt(struct socket *sock, int level, int optname, 2806 char __user *optval, int __user *optlen) 2807 { 2808 struct sock *sk = sock->sk; 2809 2810 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen); 2811 } 2812 EXPORT_SYMBOL(sock_common_getsockopt); 2813 2814 #ifdef CONFIG_COMPAT 2815 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname, 2816 char __user *optval, int __user *optlen) 2817 { 2818 struct sock *sk = sock->sk; 2819 2820 if (sk->sk_prot->compat_getsockopt != NULL) 2821 return sk->sk_prot->compat_getsockopt(sk, level, optname, 2822 optval, optlen); 2823 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen); 2824 } 2825 EXPORT_SYMBOL(compat_sock_common_getsockopt); 2826 #endif 2827 2828 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 2829 int flags) 2830 { 2831 struct sock *sk = sock->sk; 2832 int addr_len = 0; 2833 int err; 2834 2835 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT, 2836 flags & ~MSG_DONTWAIT, &addr_len); 2837 if (err >= 0) 2838 msg->msg_namelen = addr_len; 2839 return err; 2840 } 2841 EXPORT_SYMBOL(sock_common_recvmsg); 2842 2843 /* 2844 * Set socket options on an inet socket. 2845 */ 2846 int sock_common_setsockopt(struct socket *sock, int level, int optname, 2847 char __user *optval, unsigned int optlen) 2848 { 2849 struct sock *sk = sock->sk; 2850 2851 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen); 2852 } 2853 EXPORT_SYMBOL(sock_common_setsockopt); 2854 2855 #ifdef CONFIG_COMPAT 2856 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname, 2857 char __user *optval, unsigned int optlen) 2858 { 2859 struct sock *sk = sock->sk; 2860 2861 if (sk->sk_prot->compat_setsockopt != NULL) 2862 return sk->sk_prot->compat_setsockopt(sk, level, optname, 2863 optval, optlen); 2864 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen); 2865 } 2866 EXPORT_SYMBOL(compat_sock_common_setsockopt); 2867 #endif 2868 2869 void sk_common_release(struct sock *sk) 2870 { 2871 if (sk->sk_prot->destroy) 2872 sk->sk_prot->destroy(sk); 2873 2874 /* 2875 * Observation: when sock_common_release is called, processes have 2876 * no access to socket. But net still has. 2877 * Step one, detach it from networking: 2878 * 2879 * A. Remove from hash tables. 2880 */ 2881 2882 sk->sk_prot->unhash(sk); 2883 2884 /* 2885 * In this point socket cannot receive new packets, but it is possible 2886 * that some packets are in flight because some CPU runs receiver and 2887 * did hash table lookup before we unhashed socket. They will achieve 2888 * receive queue and will be purged by socket destructor. 2889 * 2890 * Also we still have packets pending on receive queue and probably, 2891 * our own packets waiting in device queues. sock_destroy will drain 2892 * receive queue, but transmitted packets will delay socket destruction 2893 * until the last reference will be released. 2894 */ 2895 2896 sock_orphan(sk); 2897 2898 xfrm_sk_free_policy(sk); 2899 2900 sk_refcnt_debug_release(sk); 2901 2902 sock_put(sk); 2903 } 2904 EXPORT_SYMBOL(sk_common_release); 2905 2906 void sk_get_meminfo(const struct sock *sk, u32 *mem) 2907 { 2908 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS); 2909 2910 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk); 2911 mem[SK_MEMINFO_RCVBUF] = sk->sk_rcvbuf; 2912 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk); 2913 mem[SK_MEMINFO_SNDBUF] = sk->sk_sndbuf; 2914 mem[SK_MEMINFO_FWD_ALLOC] = sk->sk_forward_alloc; 2915 mem[SK_MEMINFO_WMEM_QUEUED] = sk->sk_wmem_queued; 2916 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc); 2917 mem[SK_MEMINFO_BACKLOG] = sk->sk_backlog.len; 2918 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops); 2919 } 2920 2921 #ifdef CONFIG_PROC_FS 2922 #define PROTO_INUSE_NR 64 /* should be enough for the first time */ 2923 struct prot_inuse { 2924 int val[PROTO_INUSE_NR]; 2925 }; 2926 2927 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR); 2928 2929 #ifdef CONFIG_NET_NS 2930 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val) 2931 { 2932 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val); 2933 } 2934 EXPORT_SYMBOL_GPL(sock_prot_inuse_add); 2935 2936 int sock_prot_inuse_get(struct net *net, struct proto *prot) 2937 { 2938 int cpu, idx = prot->inuse_idx; 2939 int res = 0; 2940 2941 for_each_possible_cpu(cpu) 2942 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx]; 2943 2944 return res >= 0 ? res : 0; 2945 } 2946 EXPORT_SYMBOL_GPL(sock_prot_inuse_get); 2947 2948 static int __net_init sock_inuse_init_net(struct net *net) 2949 { 2950 net->core.inuse = alloc_percpu(struct prot_inuse); 2951 return net->core.inuse ? 0 : -ENOMEM; 2952 } 2953 2954 static void __net_exit sock_inuse_exit_net(struct net *net) 2955 { 2956 free_percpu(net->core.inuse); 2957 } 2958 2959 static struct pernet_operations net_inuse_ops = { 2960 .init = sock_inuse_init_net, 2961 .exit = sock_inuse_exit_net, 2962 }; 2963 2964 static __init int net_inuse_init(void) 2965 { 2966 if (register_pernet_subsys(&net_inuse_ops)) 2967 panic("Cannot initialize net inuse counters"); 2968 2969 return 0; 2970 } 2971 2972 core_initcall(net_inuse_init); 2973 #else 2974 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse); 2975 2976 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val) 2977 { 2978 __this_cpu_add(prot_inuse.val[prot->inuse_idx], val); 2979 } 2980 EXPORT_SYMBOL_GPL(sock_prot_inuse_add); 2981 2982 int sock_prot_inuse_get(struct net *net, struct proto *prot) 2983 { 2984 int cpu, idx = prot->inuse_idx; 2985 int res = 0; 2986 2987 for_each_possible_cpu(cpu) 2988 res += per_cpu(prot_inuse, cpu).val[idx]; 2989 2990 return res >= 0 ? res : 0; 2991 } 2992 EXPORT_SYMBOL_GPL(sock_prot_inuse_get); 2993 #endif 2994 2995 static void assign_proto_idx(struct proto *prot) 2996 { 2997 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR); 2998 2999 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) { 3000 pr_err("PROTO_INUSE_NR exhausted\n"); 3001 return; 3002 } 3003 3004 set_bit(prot->inuse_idx, proto_inuse_idx); 3005 } 3006 3007 static void release_proto_idx(struct proto *prot) 3008 { 3009 if (prot->inuse_idx != PROTO_INUSE_NR - 1) 3010 clear_bit(prot->inuse_idx, proto_inuse_idx); 3011 } 3012 #else 3013 static inline void assign_proto_idx(struct proto *prot) 3014 { 3015 } 3016 3017 static inline void release_proto_idx(struct proto *prot) 3018 { 3019 } 3020 #endif 3021 3022 static void req_prot_cleanup(struct request_sock_ops *rsk_prot) 3023 { 3024 if (!rsk_prot) 3025 return; 3026 kfree(rsk_prot->slab_name); 3027 rsk_prot->slab_name = NULL; 3028 kmem_cache_destroy(rsk_prot->slab); 3029 rsk_prot->slab = NULL; 3030 } 3031 3032 static int req_prot_init(const struct proto *prot) 3033 { 3034 struct request_sock_ops *rsk_prot = prot->rsk_prot; 3035 3036 if (!rsk_prot) 3037 return 0; 3038 3039 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", 3040 prot->name); 3041 if (!rsk_prot->slab_name) 3042 return -ENOMEM; 3043 3044 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name, 3045 rsk_prot->obj_size, 0, 3046 prot->slab_flags, NULL); 3047 3048 if (!rsk_prot->slab) { 3049 pr_crit("%s: Can't create request sock SLAB cache!\n", 3050 prot->name); 3051 return -ENOMEM; 3052 } 3053 return 0; 3054 } 3055 3056 int proto_register(struct proto *prot, int alloc_slab) 3057 { 3058 if (alloc_slab) { 3059 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0, 3060 SLAB_HWCACHE_ALIGN | prot->slab_flags, 3061 NULL); 3062 3063 if (prot->slab == NULL) { 3064 pr_crit("%s: Can't create sock SLAB cache!\n", 3065 prot->name); 3066 goto out; 3067 } 3068 3069 if (req_prot_init(prot)) 3070 goto out_free_request_sock_slab; 3071 3072 if (prot->twsk_prot != NULL) { 3073 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name); 3074 3075 if (prot->twsk_prot->twsk_slab_name == NULL) 3076 goto out_free_request_sock_slab; 3077 3078 prot->twsk_prot->twsk_slab = 3079 kmem_cache_create(prot->twsk_prot->twsk_slab_name, 3080 prot->twsk_prot->twsk_obj_size, 3081 0, 3082 prot->slab_flags, 3083 NULL); 3084 if (prot->twsk_prot->twsk_slab == NULL) 3085 goto out_free_timewait_sock_slab_name; 3086 } 3087 } 3088 3089 mutex_lock(&proto_list_mutex); 3090 list_add(&prot->node, &proto_list); 3091 assign_proto_idx(prot); 3092 mutex_unlock(&proto_list_mutex); 3093 return 0; 3094 3095 out_free_timewait_sock_slab_name: 3096 kfree(prot->twsk_prot->twsk_slab_name); 3097 out_free_request_sock_slab: 3098 req_prot_cleanup(prot->rsk_prot); 3099 3100 kmem_cache_destroy(prot->slab); 3101 prot->slab = NULL; 3102 out: 3103 return -ENOBUFS; 3104 } 3105 EXPORT_SYMBOL(proto_register); 3106 3107 void proto_unregister(struct proto *prot) 3108 { 3109 mutex_lock(&proto_list_mutex); 3110 release_proto_idx(prot); 3111 list_del(&prot->node); 3112 mutex_unlock(&proto_list_mutex); 3113 3114 kmem_cache_destroy(prot->slab); 3115 prot->slab = NULL; 3116 3117 req_prot_cleanup(prot->rsk_prot); 3118 3119 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) { 3120 kmem_cache_destroy(prot->twsk_prot->twsk_slab); 3121 kfree(prot->twsk_prot->twsk_slab_name); 3122 prot->twsk_prot->twsk_slab = NULL; 3123 } 3124 } 3125 EXPORT_SYMBOL(proto_unregister); 3126 3127 #ifdef CONFIG_PROC_FS 3128 static void *proto_seq_start(struct seq_file *seq, loff_t *pos) 3129 __acquires(proto_list_mutex) 3130 { 3131 mutex_lock(&proto_list_mutex); 3132 return seq_list_start_head(&proto_list, *pos); 3133 } 3134 3135 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos) 3136 { 3137 return seq_list_next(v, &proto_list, pos); 3138 } 3139 3140 static void proto_seq_stop(struct seq_file *seq, void *v) 3141 __releases(proto_list_mutex) 3142 { 3143 mutex_unlock(&proto_list_mutex); 3144 } 3145 3146 static char proto_method_implemented(const void *method) 3147 { 3148 return method == NULL ? 'n' : 'y'; 3149 } 3150 static long sock_prot_memory_allocated(struct proto *proto) 3151 { 3152 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L; 3153 } 3154 3155 static char *sock_prot_memory_pressure(struct proto *proto) 3156 { 3157 return proto->memory_pressure != NULL ? 3158 proto_memory_pressure(proto) ? "yes" : "no" : "NI"; 3159 } 3160 3161 static void proto_seq_printf(struct seq_file *seq, struct proto *proto) 3162 { 3163 3164 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s " 3165 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n", 3166 proto->name, 3167 proto->obj_size, 3168 sock_prot_inuse_get(seq_file_net(seq), proto), 3169 sock_prot_memory_allocated(proto), 3170 sock_prot_memory_pressure(proto), 3171 proto->max_header, 3172 proto->slab == NULL ? "no" : "yes", 3173 module_name(proto->owner), 3174 proto_method_implemented(proto->close), 3175 proto_method_implemented(proto->connect), 3176 proto_method_implemented(proto->disconnect), 3177 proto_method_implemented(proto->accept), 3178 proto_method_implemented(proto->ioctl), 3179 proto_method_implemented(proto->init), 3180 proto_method_implemented(proto->destroy), 3181 proto_method_implemented(proto->shutdown), 3182 proto_method_implemented(proto->setsockopt), 3183 proto_method_implemented(proto->getsockopt), 3184 proto_method_implemented(proto->sendmsg), 3185 proto_method_implemented(proto->recvmsg), 3186 proto_method_implemented(proto->sendpage), 3187 proto_method_implemented(proto->bind), 3188 proto_method_implemented(proto->backlog_rcv), 3189 proto_method_implemented(proto->hash), 3190 proto_method_implemented(proto->unhash), 3191 proto_method_implemented(proto->get_port), 3192 proto_method_implemented(proto->enter_memory_pressure)); 3193 } 3194 3195 static int proto_seq_show(struct seq_file *seq, void *v) 3196 { 3197 if (v == &proto_list) 3198 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s", 3199 "protocol", 3200 "size", 3201 "sockets", 3202 "memory", 3203 "press", 3204 "maxhdr", 3205 "slab", 3206 "module", 3207 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n"); 3208 else 3209 proto_seq_printf(seq, list_entry(v, struct proto, node)); 3210 return 0; 3211 } 3212 3213 static const struct seq_operations proto_seq_ops = { 3214 .start = proto_seq_start, 3215 .next = proto_seq_next, 3216 .stop = proto_seq_stop, 3217 .show = proto_seq_show, 3218 }; 3219 3220 static int proto_seq_open(struct inode *inode, struct file *file) 3221 { 3222 return seq_open_net(inode, file, &proto_seq_ops, 3223 sizeof(struct seq_net_private)); 3224 } 3225 3226 static const struct file_operations proto_seq_fops = { 3227 .owner = THIS_MODULE, 3228 .open = proto_seq_open, 3229 .read = seq_read, 3230 .llseek = seq_lseek, 3231 .release = seq_release_net, 3232 }; 3233 3234 static __net_init int proto_init_net(struct net *net) 3235 { 3236 if (!proc_create("protocols", S_IRUGO, net->proc_net, &proto_seq_fops)) 3237 return -ENOMEM; 3238 3239 return 0; 3240 } 3241 3242 static __net_exit void proto_exit_net(struct net *net) 3243 { 3244 remove_proc_entry("protocols", net->proc_net); 3245 } 3246 3247 3248 static __net_initdata struct pernet_operations proto_net_ops = { 3249 .init = proto_init_net, 3250 .exit = proto_exit_net, 3251 }; 3252 3253 static int __init proto_init(void) 3254 { 3255 return register_pernet_subsys(&proto_net_ops); 3256 } 3257 3258 subsys_initcall(proto_init); 3259 3260 #endif /* PROC_FS */ 3261 3262 #ifdef CONFIG_NET_RX_BUSY_POLL 3263 bool sk_busy_loop_end(void *p, unsigned long start_time) 3264 { 3265 struct sock *sk = p; 3266 3267 return !skb_queue_empty(&sk->sk_receive_queue) || 3268 sk_busy_loop_timeout(sk, start_time); 3269 } 3270 EXPORT_SYMBOL(sk_busy_loop_end); 3271 #endif /* CONFIG_NET_RX_BUSY_POLL */ 3272