1 /* 2 * INET An implementation of the TCP/IP protocol suite for the LINUX 3 * operating system. INET is implemented using the BSD Socket 4 * interface as the means of communication with the user level. 5 * 6 * Generic socket support routines. Memory allocators, socket lock/release 7 * handler for protocols to use and generic option handler. 8 * 9 * 10 * Authors: Ross Biro 11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 12 * Florian La Roche, <flla@stud.uni-sb.de> 13 * Alan Cox, <A.Cox@swansea.ac.uk> 14 * 15 * Fixes: 16 * Alan Cox : Numerous verify_area() problems 17 * Alan Cox : Connecting on a connecting socket 18 * now returns an error for tcp. 19 * Alan Cox : sock->protocol is set correctly. 20 * and is not sometimes left as 0. 21 * Alan Cox : connect handles icmp errors on a 22 * connect properly. Unfortunately there 23 * is a restart syscall nasty there. I 24 * can't match BSD without hacking the C 25 * library. Ideas urgently sought! 26 * Alan Cox : Disallow bind() to addresses that are 27 * not ours - especially broadcast ones!! 28 * Alan Cox : Socket 1024 _IS_ ok for users. (fencepost) 29 * Alan Cox : sock_wfree/sock_rfree don't destroy sockets, 30 * instead they leave that for the DESTROY timer. 31 * Alan Cox : Clean up error flag in accept 32 * Alan Cox : TCP ack handling is buggy, the DESTROY timer 33 * was buggy. Put a remove_sock() in the handler 34 * for memory when we hit 0. Also altered the timer 35 * code. The ACK stuff can wait and needs major 36 * TCP layer surgery. 37 * Alan Cox : Fixed TCP ack bug, removed remove sock 38 * and fixed timer/inet_bh race. 39 * Alan Cox : Added zapped flag for TCP 40 * Alan Cox : Move kfree_skb into skbuff.c and tidied up surplus code 41 * Alan Cox : for new sk_buff allocations wmalloc/rmalloc now call alloc_skb 42 * Alan Cox : kfree_s calls now are kfree_skbmem so we can track skb resources 43 * Alan Cox : Supports socket option broadcast now as does udp. Packet and raw need fixing. 44 * Alan Cox : Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so... 45 * Rick Sladkey : Relaxed UDP rules for matching packets. 46 * C.E.Hawkins : IFF_PROMISC/SIOCGHWADDR support 47 * Pauline Middelink : identd support 48 * Alan Cox : Fixed connect() taking signals I think. 49 * Alan Cox : SO_LINGER supported 50 * Alan Cox : Error reporting fixes 51 * Anonymous : inet_create tidied up (sk->reuse setting) 52 * Alan Cox : inet sockets don't set sk->type! 53 * Alan Cox : Split socket option code 54 * Alan Cox : Callbacks 55 * Alan Cox : Nagle flag for Charles & Johannes stuff 56 * Alex : Removed restriction on inet fioctl 57 * Alan Cox : Splitting INET from NET core 58 * Alan Cox : Fixed bogus SO_TYPE handling in getsockopt() 59 * Adam Caldwell : Missing return in SO_DONTROUTE/SO_DEBUG code 60 * Alan Cox : Split IP from generic code 61 * Alan Cox : New kfree_skbmem() 62 * Alan Cox : Make SO_DEBUG superuser only. 63 * Alan Cox : Allow anyone to clear SO_DEBUG 64 * (compatibility fix) 65 * Alan Cox : Added optimistic memory grabbing for AF_UNIX throughput. 66 * Alan Cox : Allocator for a socket is settable. 67 * Alan Cox : SO_ERROR includes soft errors. 68 * Alan Cox : Allow NULL arguments on some SO_ opts 69 * Alan Cox : Generic socket allocation to make hooks 70 * easier (suggested by Craig Metz). 71 * Michael Pall : SO_ERROR returns positive errno again 72 * Steve Whitehouse: Added default destructor to free 73 * protocol private data. 74 * Steve Whitehouse: Added various other default routines 75 * common to several socket families. 76 * Chris Evans : Call suser() check last on F_SETOWN 77 * Jay Schulist : Added SO_ATTACH_FILTER and SO_DETACH_FILTER. 78 * Andi Kleen : Add sock_kmalloc()/sock_kfree_s() 79 * Andi Kleen : Fix write_space callback 80 * Chris Evans : Security fixes - signedness again 81 * Arnaldo C. Melo : cleanups, use skb_queue_purge 82 * 83 * To Fix: 84 * 85 * 86 * This program is free software; you can redistribute it and/or 87 * modify it under the terms of the GNU General Public License 88 * as published by the Free Software Foundation; either version 89 * 2 of the License, or (at your option) any later version. 90 */ 91 92 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 93 94 #include <linux/capability.h> 95 #include <linux/errno.h> 96 #include <linux/errqueue.h> 97 #include <linux/types.h> 98 #include <linux/socket.h> 99 #include <linux/in.h> 100 #include <linux/kernel.h> 101 #include <linux/module.h> 102 #include <linux/proc_fs.h> 103 #include <linux/seq_file.h> 104 #include <linux/sched.h> 105 #include <linux/timer.h> 106 #include <linux/string.h> 107 #include <linux/sockios.h> 108 #include <linux/net.h> 109 #include <linux/mm.h> 110 #include <linux/slab.h> 111 #include <linux/interrupt.h> 112 #include <linux/poll.h> 113 #include <linux/tcp.h> 114 #include <linux/init.h> 115 #include <linux/highmem.h> 116 #include <linux/user_namespace.h> 117 #include <linux/static_key.h> 118 #include <linux/memcontrol.h> 119 #include <linux/prefetch.h> 120 121 #include <asm/uaccess.h> 122 123 #include <linux/netdevice.h> 124 #include <net/protocol.h> 125 #include <linux/skbuff.h> 126 #include <net/net_namespace.h> 127 #include <net/request_sock.h> 128 #include <net/sock.h> 129 #include <linux/net_tstamp.h> 130 #include <net/xfrm.h> 131 #include <linux/ipsec.h> 132 #include <net/cls_cgroup.h> 133 #include <net/netprio_cgroup.h> 134 135 #include <linux/filter.h> 136 137 #include <trace/events/sock.h> 138 139 #ifdef CONFIG_INET 140 #include <net/tcp.h> 141 #endif 142 143 #include <net/busy_poll.h> 144 145 static DEFINE_MUTEX(proto_list_mutex); 146 static LIST_HEAD(proto_list); 147 148 #ifdef CONFIG_MEMCG_KMEM 149 int mem_cgroup_sockets_init(struct mem_cgroup *memcg, struct cgroup_subsys *ss) 150 { 151 struct proto *proto; 152 int ret = 0; 153 154 mutex_lock(&proto_list_mutex); 155 list_for_each_entry(proto, &proto_list, node) { 156 if (proto->init_cgroup) { 157 ret = proto->init_cgroup(memcg, ss); 158 if (ret) 159 goto out; 160 } 161 } 162 163 mutex_unlock(&proto_list_mutex); 164 return ret; 165 out: 166 list_for_each_entry_continue_reverse(proto, &proto_list, node) 167 if (proto->destroy_cgroup) 168 proto->destroy_cgroup(memcg); 169 mutex_unlock(&proto_list_mutex); 170 return ret; 171 } 172 173 void mem_cgroup_sockets_destroy(struct mem_cgroup *memcg) 174 { 175 struct proto *proto; 176 177 mutex_lock(&proto_list_mutex); 178 list_for_each_entry_reverse(proto, &proto_list, node) 179 if (proto->destroy_cgroup) 180 proto->destroy_cgroup(memcg); 181 mutex_unlock(&proto_list_mutex); 182 } 183 #endif 184 185 /* 186 * Each address family might have different locking rules, so we have 187 * one slock key per address family: 188 */ 189 static struct lock_class_key af_family_keys[AF_MAX]; 190 static struct lock_class_key af_family_slock_keys[AF_MAX]; 191 192 #if defined(CONFIG_MEMCG_KMEM) 193 struct static_key memcg_socket_limit_enabled; 194 EXPORT_SYMBOL(memcg_socket_limit_enabled); 195 #endif 196 197 /* 198 * Make lock validator output more readable. (we pre-construct these 199 * strings build-time, so that runtime initialization of socket 200 * locks is fast): 201 */ 202 static const char *const af_family_key_strings[AF_MAX+1] = { 203 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" , 204 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK", 205 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" , 206 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" , 207 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" , 208 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" , 209 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" , 210 "sk_lock-AF_RDS" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" , 211 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" , 212 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" , 213 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" , 214 "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN" , "sk_lock-AF_PHONET" , 215 "sk_lock-AF_IEEE802154", "sk_lock-AF_CAIF" , "sk_lock-AF_ALG" , 216 "sk_lock-AF_NFC" , "sk_lock-AF_VSOCK" , "sk_lock-AF_MAX" 217 }; 218 static const char *const af_family_slock_key_strings[AF_MAX+1] = { 219 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" , 220 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK", 221 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" , 222 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" , 223 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" , 224 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" , 225 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" , 226 "slock-AF_RDS" , "slock-AF_SNA" , "slock-AF_IRDA" , 227 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" , 228 "slock-27" , "slock-28" , "slock-AF_CAN" , 229 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" , 230 "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" , 231 "slock-AF_IEEE802154", "slock-AF_CAIF" , "slock-AF_ALG" , 232 "slock-AF_NFC" , "slock-AF_VSOCK" ,"slock-AF_MAX" 233 }; 234 static const char *const af_family_clock_key_strings[AF_MAX+1] = { 235 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" , 236 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK", 237 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" , 238 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" , 239 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" , 240 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" , 241 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" , 242 "clock-AF_RDS" , "clock-AF_SNA" , "clock-AF_IRDA" , 243 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" , 244 "clock-27" , "clock-28" , "clock-AF_CAN" , 245 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" , 246 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" , 247 "clock-AF_IEEE802154", "clock-AF_CAIF" , "clock-AF_ALG" , 248 "clock-AF_NFC" , "clock-AF_VSOCK" , "clock-AF_MAX" 249 }; 250 251 /* 252 * sk_callback_lock locking rules are per-address-family, 253 * so split the lock classes by using a per-AF key: 254 */ 255 static struct lock_class_key af_callback_keys[AF_MAX]; 256 257 /* Take into consideration the size of the struct sk_buff overhead in the 258 * determination of these values, since that is non-constant across 259 * platforms. This makes socket queueing behavior and performance 260 * not depend upon such differences. 261 */ 262 #define _SK_MEM_PACKETS 256 263 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256) 264 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 265 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 266 267 /* Run time adjustable parameters. */ 268 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX; 269 EXPORT_SYMBOL(sysctl_wmem_max); 270 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX; 271 EXPORT_SYMBOL(sysctl_rmem_max); 272 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX; 273 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX; 274 275 /* Maximal space eaten by iovec or ancillary data plus some space */ 276 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512); 277 EXPORT_SYMBOL(sysctl_optmem_max); 278 279 struct static_key memalloc_socks = STATIC_KEY_INIT_FALSE; 280 EXPORT_SYMBOL_GPL(memalloc_socks); 281 282 /** 283 * sk_set_memalloc - sets %SOCK_MEMALLOC 284 * @sk: socket to set it on 285 * 286 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves. 287 * It's the responsibility of the admin to adjust min_free_kbytes 288 * to meet the requirements 289 */ 290 void sk_set_memalloc(struct sock *sk) 291 { 292 sock_set_flag(sk, SOCK_MEMALLOC); 293 sk->sk_allocation |= __GFP_MEMALLOC; 294 static_key_slow_inc(&memalloc_socks); 295 } 296 EXPORT_SYMBOL_GPL(sk_set_memalloc); 297 298 void sk_clear_memalloc(struct sock *sk) 299 { 300 sock_reset_flag(sk, SOCK_MEMALLOC); 301 sk->sk_allocation &= ~__GFP_MEMALLOC; 302 static_key_slow_dec(&memalloc_socks); 303 304 /* 305 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward 306 * progress of swapping. However, if SOCK_MEMALLOC is cleared while 307 * it has rmem allocations there is a risk that the user of the 308 * socket cannot make forward progress due to exceeding the rmem 309 * limits. By rights, sk_clear_memalloc() should only be called 310 * on sockets being torn down but warn and reset the accounting if 311 * that assumption breaks. 312 */ 313 if (WARN_ON(sk->sk_forward_alloc)) 314 sk_mem_reclaim(sk); 315 } 316 EXPORT_SYMBOL_GPL(sk_clear_memalloc); 317 318 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb) 319 { 320 int ret; 321 unsigned long pflags = current->flags; 322 323 /* these should have been dropped before queueing */ 324 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC)); 325 326 current->flags |= PF_MEMALLOC; 327 ret = sk->sk_backlog_rcv(sk, skb); 328 tsk_restore_flags(current, pflags, PF_MEMALLOC); 329 330 return ret; 331 } 332 EXPORT_SYMBOL(__sk_backlog_rcv); 333 334 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen) 335 { 336 struct timeval tv; 337 338 if (optlen < sizeof(tv)) 339 return -EINVAL; 340 if (copy_from_user(&tv, optval, sizeof(tv))) 341 return -EFAULT; 342 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC) 343 return -EDOM; 344 345 if (tv.tv_sec < 0) { 346 static int warned __read_mostly; 347 348 *timeo_p = 0; 349 if (warned < 10 && net_ratelimit()) { 350 warned++; 351 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n", 352 __func__, current->comm, task_pid_nr(current)); 353 } 354 return 0; 355 } 356 *timeo_p = MAX_SCHEDULE_TIMEOUT; 357 if (tv.tv_sec == 0 && tv.tv_usec == 0) 358 return 0; 359 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1)) 360 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ); 361 return 0; 362 } 363 364 static void sock_warn_obsolete_bsdism(const char *name) 365 { 366 static int warned; 367 static char warncomm[TASK_COMM_LEN]; 368 if (strcmp(warncomm, current->comm) && warned < 5) { 369 strcpy(warncomm, current->comm); 370 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n", 371 warncomm, name); 372 warned++; 373 } 374 } 375 376 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE)) 377 378 static void sock_disable_timestamp(struct sock *sk, unsigned long flags) 379 { 380 if (sk->sk_flags & flags) { 381 sk->sk_flags &= ~flags; 382 if (!(sk->sk_flags & SK_FLAGS_TIMESTAMP)) 383 net_disable_timestamp(); 384 } 385 } 386 387 388 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 389 { 390 int err; 391 int skb_len; 392 unsigned long flags; 393 struct sk_buff_head *list = &sk->sk_receive_queue; 394 395 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) { 396 atomic_inc(&sk->sk_drops); 397 trace_sock_rcvqueue_full(sk, skb); 398 return -ENOMEM; 399 } 400 401 err = sk_filter(sk, skb); 402 if (err) 403 return err; 404 405 if (!sk_rmem_schedule(sk, skb, skb->truesize)) { 406 atomic_inc(&sk->sk_drops); 407 return -ENOBUFS; 408 } 409 410 skb->dev = NULL; 411 skb_set_owner_r(skb, sk); 412 413 /* Cache the SKB length before we tack it onto the receive 414 * queue. Once it is added it no longer belongs to us and 415 * may be freed by other threads of control pulling packets 416 * from the queue. 417 */ 418 skb_len = skb->len; 419 420 /* we escape from rcu protected region, make sure we dont leak 421 * a norefcounted dst 422 */ 423 skb_dst_force(skb); 424 425 spin_lock_irqsave(&list->lock, flags); 426 skb->dropcount = atomic_read(&sk->sk_drops); 427 __skb_queue_tail(list, skb); 428 spin_unlock_irqrestore(&list->lock, flags); 429 430 if (!sock_flag(sk, SOCK_DEAD)) 431 sk->sk_data_ready(sk, skb_len); 432 return 0; 433 } 434 EXPORT_SYMBOL(sock_queue_rcv_skb); 435 436 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested) 437 { 438 int rc = NET_RX_SUCCESS; 439 440 if (sk_filter(sk, skb)) 441 goto discard_and_relse; 442 443 skb->dev = NULL; 444 445 if (sk_rcvqueues_full(sk, skb, sk->sk_rcvbuf)) { 446 atomic_inc(&sk->sk_drops); 447 goto discard_and_relse; 448 } 449 if (nested) 450 bh_lock_sock_nested(sk); 451 else 452 bh_lock_sock(sk); 453 if (!sock_owned_by_user(sk)) { 454 /* 455 * trylock + unlock semantics: 456 */ 457 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_); 458 459 rc = sk_backlog_rcv(sk, skb); 460 461 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_); 462 } else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) { 463 bh_unlock_sock(sk); 464 atomic_inc(&sk->sk_drops); 465 goto discard_and_relse; 466 } 467 468 bh_unlock_sock(sk); 469 out: 470 sock_put(sk); 471 return rc; 472 discard_and_relse: 473 kfree_skb(skb); 474 goto out; 475 } 476 EXPORT_SYMBOL(sk_receive_skb); 477 478 void sk_reset_txq(struct sock *sk) 479 { 480 sk_tx_queue_clear(sk); 481 } 482 EXPORT_SYMBOL(sk_reset_txq); 483 484 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie) 485 { 486 struct dst_entry *dst = __sk_dst_get(sk); 487 488 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) { 489 sk_tx_queue_clear(sk); 490 RCU_INIT_POINTER(sk->sk_dst_cache, NULL); 491 dst_release(dst); 492 return NULL; 493 } 494 495 return dst; 496 } 497 EXPORT_SYMBOL(__sk_dst_check); 498 499 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie) 500 { 501 struct dst_entry *dst = sk_dst_get(sk); 502 503 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) { 504 sk_dst_reset(sk); 505 dst_release(dst); 506 return NULL; 507 } 508 509 return dst; 510 } 511 EXPORT_SYMBOL(sk_dst_check); 512 513 static int sock_setbindtodevice(struct sock *sk, char __user *optval, 514 int optlen) 515 { 516 int ret = -ENOPROTOOPT; 517 #ifdef CONFIG_NETDEVICES 518 struct net *net = sock_net(sk); 519 char devname[IFNAMSIZ]; 520 int index; 521 522 /* Sorry... */ 523 ret = -EPERM; 524 if (!ns_capable(net->user_ns, CAP_NET_RAW)) 525 goto out; 526 527 ret = -EINVAL; 528 if (optlen < 0) 529 goto out; 530 531 /* Bind this socket to a particular device like "eth0", 532 * as specified in the passed interface name. If the 533 * name is "" or the option length is zero the socket 534 * is not bound. 535 */ 536 if (optlen > IFNAMSIZ - 1) 537 optlen = IFNAMSIZ - 1; 538 memset(devname, 0, sizeof(devname)); 539 540 ret = -EFAULT; 541 if (copy_from_user(devname, optval, optlen)) 542 goto out; 543 544 index = 0; 545 if (devname[0] != '\0') { 546 struct net_device *dev; 547 548 rcu_read_lock(); 549 dev = dev_get_by_name_rcu(net, devname); 550 if (dev) 551 index = dev->ifindex; 552 rcu_read_unlock(); 553 ret = -ENODEV; 554 if (!dev) 555 goto out; 556 } 557 558 lock_sock(sk); 559 sk->sk_bound_dev_if = index; 560 sk_dst_reset(sk); 561 release_sock(sk); 562 563 ret = 0; 564 565 out: 566 #endif 567 568 return ret; 569 } 570 571 static int sock_getbindtodevice(struct sock *sk, char __user *optval, 572 int __user *optlen, int len) 573 { 574 int ret = -ENOPROTOOPT; 575 #ifdef CONFIG_NETDEVICES 576 struct net *net = sock_net(sk); 577 char devname[IFNAMSIZ]; 578 579 if (sk->sk_bound_dev_if == 0) { 580 len = 0; 581 goto zero; 582 } 583 584 ret = -EINVAL; 585 if (len < IFNAMSIZ) 586 goto out; 587 588 ret = netdev_get_name(net, devname, sk->sk_bound_dev_if); 589 if (ret) 590 goto out; 591 592 len = strlen(devname) + 1; 593 594 ret = -EFAULT; 595 if (copy_to_user(optval, devname, len)) 596 goto out; 597 598 zero: 599 ret = -EFAULT; 600 if (put_user(len, optlen)) 601 goto out; 602 603 ret = 0; 604 605 out: 606 #endif 607 608 return ret; 609 } 610 611 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool) 612 { 613 if (valbool) 614 sock_set_flag(sk, bit); 615 else 616 sock_reset_flag(sk, bit); 617 } 618 619 /* 620 * This is meant for all protocols to use and covers goings on 621 * at the socket level. Everything here is generic. 622 */ 623 624 int sock_setsockopt(struct socket *sock, int level, int optname, 625 char __user *optval, unsigned int optlen) 626 { 627 struct sock *sk = sock->sk; 628 int val; 629 int valbool; 630 struct linger ling; 631 int ret = 0; 632 633 /* 634 * Options without arguments 635 */ 636 637 if (optname == SO_BINDTODEVICE) 638 return sock_setbindtodevice(sk, optval, optlen); 639 640 if (optlen < sizeof(int)) 641 return -EINVAL; 642 643 if (get_user(val, (int __user *)optval)) 644 return -EFAULT; 645 646 valbool = val ? 1 : 0; 647 648 lock_sock(sk); 649 650 switch (optname) { 651 case SO_DEBUG: 652 if (val && !capable(CAP_NET_ADMIN)) 653 ret = -EACCES; 654 else 655 sock_valbool_flag(sk, SOCK_DBG, valbool); 656 break; 657 case SO_REUSEADDR: 658 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE); 659 break; 660 case SO_REUSEPORT: 661 sk->sk_reuseport = valbool; 662 break; 663 case SO_TYPE: 664 case SO_PROTOCOL: 665 case SO_DOMAIN: 666 case SO_ERROR: 667 ret = -ENOPROTOOPT; 668 break; 669 case SO_DONTROUTE: 670 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool); 671 break; 672 case SO_BROADCAST: 673 sock_valbool_flag(sk, SOCK_BROADCAST, valbool); 674 break; 675 case SO_SNDBUF: 676 /* Don't error on this BSD doesn't and if you think 677 * about it this is right. Otherwise apps have to 678 * play 'guess the biggest size' games. RCVBUF/SNDBUF 679 * are treated in BSD as hints 680 */ 681 val = min_t(u32, val, sysctl_wmem_max); 682 set_sndbuf: 683 sk->sk_userlocks |= SOCK_SNDBUF_LOCK; 684 sk->sk_sndbuf = max_t(u32, val * 2, SOCK_MIN_SNDBUF); 685 /* Wake up sending tasks if we upped the value. */ 686 sk->sk_write_space(sk); 687 break; 688 689 case SO_SNDBUFFORCE: 690 if (!capable(CAP_NET_ADMIN)) { 691 ret = -EPERM; 692 break; 693 } 694 goto set_sndbuf; 695 696 case SO_RCVBUF: 697 /* Don't error on this BSD doesn't and if you think 698 * about it this is right. Otherwise apps have to 699 * play 'guess the biggest size' games. RCVBUF/SNDBUF 700 * are treated in BSD as hints 701 */ 702 val = min_t(u32, val, sysctl_rmem_max); 703 set_rcvbuf: 704 sk->sk_userlocks |= SOCK_RCVBUF_LOCK; 705 /* 706 * We double it on the way in to account for 707 * "struct sk_buff" etc. overhead. Applications 708 * assume that the SO_RCVBUF setting they make will 709 * allow that much actual data to be received on that 710 * socket. 711 * 712 * Applications are unaware that "struct sk_buff" and 713 * other overheads allocate from the receive buffer 714 * during socket buffer allocation. 715 * 716 * And after considering the possible alternatives, 717 * returning the value we actually used in getsockopt 718 * is the most desirable behavior. 719 */ 720 sk->sk_rcvbuf = max_t(u32, val * 2, SOCK_MIN_RCVBUF); 721 break; 722 723 case SO_RCVBUFFORCE: 724 if (!capable(CAP_NET_ADMIN)) { 725 ret = -EPERM; 726 break; 727 } 728 goto set_rcvbuf; 729 730 case SO_KEEPALIVE: 731 #ifdef CONFIG_INET 732 if (sk->sk_protocol == IPPROTO_TCP && 733 sk->sk_type == SOCK_STREAM) 734 tcp_set_keepalive(sk, valbool); 735 #endif 736 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool); 737 break; 738 739 case SO_OOBINLINE: 740 sock_valbool_flag(sk, SOCK_URGINLINE, valbool); 741 break; 742 743 case SO_NO_CHECK: 744 sk->sk_no_check = valbool; 745 break; 746 747 case SO_PRIORITY: 748 if ((val >= 0 && val <= 6) || 749 ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) 750 sk->sk_priority = val; 751 else 752 ret = -EPERM; 753 break; 754 755 case SO_LINGER: 756 if (optlen < sizeof(ling)) { 757 ret = -EINVAL; /* 1003.1g */ 758 break; 759 } 760 if (copy_from_user(&ling, optval, sizeof(ling))) { 761 ret = -EFAULT; 762 break; 763 } 764 if (!ling.l_onoff) 765 sock_reset_flag(sk, SOCK_LINGER); 766 else { 767 #if (BITS_PER_LONG == 32) 768 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ) 769 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT; 770 else 771 #endif 772 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ; 773 sock_set_flag(sk, SOCK_LINGER); 774 } 775 break; 776 777 case SO_BSDCOMPAT: 778 sock_warn_obsolete_bsdism("setsockopt"); 779 break; 780 781 case SO_PASSCRED: 782 if (valbool) 783 set_bit(SOCK_PASSCRED, &sock->flags); 784 else 785 clear_bit(SOCK_PASSCRED, &sock->flags); 786 break; 787 788 case SO_TIMESTAMP: 789 case SO_TIMESTAMPNS: 790 if (valbool) { 791 if (optname == SO_TIMESTAMP) 792 sock_reset_flag(sk, SOCK_RCVTSTAMPNS); 793 else 794 sock_set_flag(sk, SOCK_RCVTSTAMPNS); 795 sock_set_flag(sk, SOCK_RCVTSTAMP); 796 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 797 } else { 798 sock_reset_flag(sk, SOCK_RCVTSTAMP); 799 sock_reset_flag(sk, SOCK_RCVTSTAMPNS); 800 } 801 break; 802 803 case SO_TIMESTAMPING: 804 if (val & ~SOF_TIMESTAMPING_MASK) { 805 ret = -EINVAL; 806 break; 807 } 808 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE, 809 val & SOF_TIMESTAMPING_TX_HARDWARE); 810 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE, 811 val & SOF_TIMESTAMPING_TX_SOFTWARE); 812 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE, 813 val & SOF_TIMESTAMPING_RX_HARDWARE); 814 if (val & SOF_TIMESTAMPING_RX_SOFTWARE) 815 sock_enable_timestamp(sk, 816 SOCK_TIMESTAMPING_RX_SOFTWARE); 817 else 818 sock_disable_timestamp(sk, 819 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE)); 820 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SOFTWARE, 821 val & SOF_TIMESTAMPING_SOFTWARE); 822 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE, 823 val & SOF_TIMESTAMPING_SYS_HARDWARE); 824 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE, 825 val & SOF_TIMESTAMPING_RAW_HARDWARE); 826 break; 827 828 case SO_RCVLOWAT: 829 if (val < 0) 830 val = INT_MAX; 831 sk->sk_rcvlowat = val ? : 1; 832 break; 833 834 case SO_RCVTIMEO: 835 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen); 836 break; 837 838 case SO_SNDTIMEO: 839 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen); 840 break; 841 842 case SO_ATTACH_FILTER: 843 ret = -EINVAL; 844 if (optlen == sizeof(struct sock_fprog)) { 845 struct sock_fprog fprog; 846 847 ret = -EFAULT; 848 if (copy_from_user(&fprog, optval, sizeof(fprog))) 849 break; 850 851 ret = sk_attach_filter(&fprog, sk); 852 } 853 break; 854 855 case SO_DETACH_FILTER: 856 ret = sk_detach_filter(sk); 857 break; 858 859 case SO_LOCK_FILTER: 860 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool) 861 ret = -EPERM; 862 else 863 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool); 864 break; 865 866 case SO_PASSSEC: 867 if (valbool) 868 set_bit(SOCK_PASSSEC, &sock->flags); 869 else 870 clear_bit(SOCK_PASSSEC, &sock->flags); 871 break; 872 case SO_MARK: 873 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) 874 ret = -EPERM; 875 else 876 sk->sk_mark = val; 877 break; 878 879 /* We implement the SO_SNDLOWAT etc to 880 not be settable (1003.1g 5.3) */ 881 case SO_RXQ_OVFL: 882 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool); 883 break; 884 885 case SO_WIFI_STATUS: 886 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool); 887 break; 888 889 case SO_PEEK_OFF: 890 if (sock->ops->set_peek_off) 891 sock->ops->set_peek_off(sk, val); 892 else 893 ret = -EOPNOTSUPP; 894 break; 895 896 case SO_NOFCS: 897 sock_valbool_flag(sk, SOCK_NOFCS, valbool); 898 break; 899 900 case SO_SELECT_ERR_QUEUE: 901 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool); 902 break; 903 904 #ifdef CONFIG_NET_LL_RX_POLL 905 case SO_BUSY_POLL: 906 /* allow unprivileged users to decrease the value */ 907 if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN)) 908 ret = -EPERM; 909 else { 910 if (val < 0) 911 ret = -EINVAL; 912 else 913 sk->sk_ll_usec = val; 914 } 915 break; 916 #endif 917 default: 918 ret = -ENOPROTOOPT; 919 break; 920 } 921 release_sock(sk); 922 return ret; 923 } 924 EXPORT_SYMBOL(sock_setsockopt); 925 926 927 void cred_to_ucred(struct pid *pid, const struct cred *cred, 928 struct ucred *ucred) 929 { 930 ucred->pid = pid_vnr(pid); 931 ucred->uid = ucred->gid = -1; 932 if (cred) { 933 struct user_namespace *current_ns = current_user_ns(); 934 935 ucred->uid = from_kuid_munged(current_ns, cred->euid); 936 ucred->gid = from_kgid_munged(current_ns, cred->egid); 937 } 938 } 939 EXPORT_SYMBOL_GPL(cred_to_ucred); 940 941 int sock_getsockopt(struct socket *sock, int level, int optname, 942 char __user *optval, int __user *optlen) 943 { 944 struct sock *sk = sock->sk; 945 946 union { 947 int val; 948 struct linger ling; 949 struct timeval tm; 950 } v; 951 952 int lv = sizeof(int); 953 int len; 954 955 if (get_user(len, optlen)) 956 return -EFAULT; 957 if (len < 0) 958 return -EINVAL; 959 960 memset(&v, 0, sizeof(v)); 961 962 switch (optname) { 963 case SO_DEBUG: 964 v.val = sock_flag(sk, SOCK_DBG); 965 break; 966 967 case SO_DONTROUTE: 968 v.val = sock_flag(sk, SOCK_LOCALROUTE); 969 break; 970 971 case SO_BROADCAST: 972 v.val = sock_flag(sk, SOCK_BROADCAST); 973 break; 974 975 case SO_SNDBUF: 976 v.val = sk->sk_sndbuf; 977 break; 978 979 case SO_RCVBUF: 980 v.val = sk->sk_rcvbuf; 981 break; 982 983 case SO_REUSEADDR: 984 v.val = sk->sk_reuse; 985 break; 986 987 case SO_REUSEPORT: 988 v.val = sk->sk_reuseport; 989 break; 990 991 case SO_KEEPALIVE: 992 v.val = sock_flag(sk, SOCK_KEEPOPEN); 993 break; 994 995 case SO_TYPE: 996 v.val = sk->sk_type; 997 break; 998 999 case SO_PROTOCOL: 1000 v.val = sk->sk_protocol; 1001 break; 1002 1003 case SO_DOMAIN: 1004 v.val = sk->sk_family; 1005 break; 1006 1007 case SO_ERROR: 1008 v.val = -sock_error(sk); 1009 if (v.val == 0) 1010 v.val = xchg(&sk->sk_err_soft, 0); 1011 break; 1012 1013 case SO_OOBINLINE: 1014 v.val = sock_flag(sk, SOCK_URGINLINE); 1015 break; 1016 1017 case SO_NO_CHECK: 1018 v.val = sk->sk_no_check; 1019 break; 1020 1021 case SO_PRIORITY: 1022 v.val = sk->sk_priority; 1023 break; 1024 1025 case SO_LINGER: 1026 lv = sizeof(v.ling); 1027 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER); 1028 v.ling.l_linger = sk->sk_lingertime / HZ; 1029 break; 1030 1031 case SO_BSDCOMPAT: 1032 sock_warn_obsolete_bsdism("getsockopt"); 1033 break; 1034 1035 case SO_TIMESTAMP: 1036 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && 1037 !sock_flag(sk, SOCK_RCVTSTAMPNS); 1038 break; 1039 1040 case SO_TIMESTAMPNS: 1041 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS); 1042 break; 1043 1044 case SO_TIMESTAMPING: 1045 v.val = 0; 1046 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE)) 1047 v.val |= SOF_TIMESTAMPING_TX_HARDWARE; 1048 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE)) 1049 v.val |= SOF_TIMESTAMPING_TX_SOFTWARE; 1050 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE)) 1051 v.val |= SOF_TIMESTAMPING_RX_HARDWARE; 1052 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE)) 1053 v.val |= SOF_TIMESTAMPING_RX_SOFTWARE; 1054 if (sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE)) 1055 v.val |= SOF_TIMESTAMPING_SOFTWARE; 1056 if (sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE)) 1057 v.val |= SOF_TIMESTAMPING_SYS_HARDWARE; 1058 if (sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE)) 1059 v.val |= SOF_TIMESTAMPING_RAW_HARDWARE; 1060 break; 1061 1062 case SO_RCVTIMEO: 1063 lv = sizeof(struct timeval); 1064 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) { 1065 v.tm.tv_sec = 0; 1066 v.tm.tv_usec = 0; 1067 } else { 1068 v.tm.tv_sec = sk->sk_rcvtimeo / HZ; 1069 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ; 1070 } 1071 break; 1072 1073 case SO_SNDTIMEO: 1074 lv = sizeof(struct timeval); 1075 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) { 1076 v.tm.tv_sec = 0; 1077 v.tm.tv_usec = 0; 1078 } else { 1079 v.tm.tv_sec = sk->sk_sndtimeo / HZ; 1080 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ; 1081 } 1082 break; 1083 1084 case SO_RCVLOWAT: 1085 v.val = sk->sk_rcvlowat; 1086 break; 1087 1088 case SO_SNDLOWAT: 1089 v.val = 1; 1090 break; 1091 1092 case SO_PASSCRED: 1093 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags); 1094 break; 1095 1096 case SO_PEERCRED: 1097 { 1098 struct ucred peercred; 1099 if (len > sizeof(peercred)) 1100 len = sizeof(peercred); 1101 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred); 1102 if (copy_to_user(optval, &peercred, len)) 1103 return -EFAULT; 1104 goto lenout; 1105 } 1106 1107 case SO_PEERNAME: 1108 { 1109 char address[128]; 1110 1111 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2)) 1112 return -ENOTCONN; 1113 if (lv < len) 1114 return -EINVAL; 1115 if (copy_to_user(optval, address, len)) 1116 return -EFAULT; 1117 goto lenout; 1118 } 1119 1120 /* Dubious BSD thing... Probably nobody even uses it, but 1121 * the UNIX standard wants it for whatever reason... -DaveM 1122 */ 1123 case SO_ACCEPTCONN: 1124 v.val = sk->sk_state == TCP_LISTEN; 1125 break; 1126 1127 case SO_PASSSEC: 1128 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags); 1129 break; 1130 1131 case SO_PEERSEC: 1132 return security_socket_getpeersec_stream(sock, optval, optlen, len); 1133 1134 case SO_MARK: 1135 v.val = sk->sk_mark; 1136 break; 1137 1138 case SO_RXQ_OVFL: 1139 v.val = sock_flag(sk, SOCK_RXQ_OVFL); 1140 break; 1141 1142 case SO_WIFI_STATUS: 1143 v.val = sock_flag(sk, SOCK_WIFI_STATUS); 1144 break; 1145 1146 case SO_PEEK_OFF: 1147 if (!sock->ops->set_peek_off) 1148 return -EOPNOTSUPP; 1149 1150 v.val = sk->sk_peek_off; 1151 break; 1152 case SO_NOFCS: 1153 v.val = sock_flag(sk, SOCK_NOFCS); 1154 break; 1155 1156 case SO_BINDTODEVICE: 1157 return sock_getbindtodevice(sk, optval, optlen, len); 1158 1159 case SO_GET_FILTER: 1160 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len); 1161 if (len < 0) 1162 return len; 1163 1164 goto lenout; 1165 1166 case SO_LOCK_FILTER: 1167 v.val = sock_flag(sk, SOCK_FILTER_LOCKED); 1168 break; 1169 1170 case SO_SELECT_ERR_QUEUE: 1171 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE); 1172 break; 1173 1174 #ifdef CONFIG_NET_LL_RX_POLL 1175 case SO_BUSY_POLL: 1176 v.val = sk->sk_ll_usec; 1177 break; 1178 #endif 1179 1180 default: 1181 return -ENOPROTOOPT; 1182 } 1183 1184 if (len > lv) 1185 len = lv; 1186 if (copy_to_user(optval, &v, len)) 1187 return -EFAULT; 1188 lenout: 1189 if (put_user(len, optlen)) 1190 return -EFAULT; 1191 return 0; 1192 } 1193 1194 /* 1195 * Initialize an sk_lock. 1196 * 1197 * (We also register the sk_lock with the lock validator.) 1198 */ 1199 static inline void sock_lock_init(struct sock *sk) 1200 { 1201 sock_lock_init_class_and_name(sk, 1202 af_family_slock_key_strings[sk->sk_family], 1203 af_family_slock_keys + sk->sk_family, 1204 af_family_key_strings[sk->sk_family], 1205 af_family_keys + sk->sk_family); 1206 } 1207 1208 /* 1209 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet, 1210 * even temporarly, because of RCU lookups. sk_node should also be left as is. 1211 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end 1212 */ 1213 static void sock_copy(struct sock *nsk, const struct sock *osk) 1214 { 1215 #ifdef CONFIG_SECURITY_NETWORK 1216 void *sptr = nsk->sk_security; 1217 #endif 1218 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin)); 1219 1220 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end, 1221 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end)); 1222 1223 #ifdef CONFIG_SECURITY_NETWORK 1224 nsk->sk_security = sptr; 1225 security_sk_clone(osk, nsk); 1226 #endif 1227 } 1228 1229 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size) 1230 { 1231 unsigned long nulls1, nulls2; 1232 1233 nulls1 = offsetof(struct sock, __sk_common.skc_node.next); 1234 nulls2 = offsetof(struct sock, __sk_common.skc_portaddr_node.next); 1235 if (nulls1 > nulls2) 1236 swap(nulls1, nulls2); 1237 1238 if (nulls1 != 0) 1239 memset((char *)sk, 0, nulls1); 1240 memset((char *)sk + nulls1 + sizeof(void *), 0, 1241 nulls2 - nulls1 - sizeof(void *)); 1242 memset((char *)sk + nulls2 + sizeof(void *), 0, 1243 size - nulls2 - sizeof(void *)); 1244 } 1245 EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls); 1246 1247 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority, 1248 int family) 1249 { 1250 struct sock *sk; 1251 struct kmem_cache *slab; 1252 1253 slab = prot->slab; 1254 if (slab != NULL) { 1255 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO); 1256 if (!sk) 1257 return sk; 1258 if (priority & __GFP_ZERO) { 1259 if (prot->clear_sk) 1260 prot->clear_sk(sk, prot->obj_size); 1261 else 1262 sk_prot_clear_nulls(sk, prot->obj_size); 1263 } 1264 } else 1265 sk = kmalloc(prot->obj_size, priority); 1266 1267 if (sk != NULL) { 1268 kmemcheck_annotate_bitfield(sk, flags); 1269 1270 if (security_sk_alloc(sk, family, priority)) 1271 goto out_free; 1272 1273 if (!try_module_get(prot->owner)) 1274 goto out_free_sec; 1275 sk_tx_queue_clear(sk); 1276 } 1277 1278 return sk; 1279 1280 out_free_sec: 1281 security_sk_free(sk); 1282 out_free: 1283 if (slab != NULL) 1284 kmem_cache_free(slab, sk); 1285 else 1286 kfree(sk); 1287 return NULL; 1288 } 1289 1290 static void sk_prot_free(struct proto *prot, struct sock *sk) 1291 { 1292 struct kmem_cache *slab; 1293 struct module *owner; 1294 1295 owner = prot->owner; 1296 slab = prot->slab; 1297 1298 security_sk_free(sk); 1299 if (slab != NULL) 1300 kmem_cache_free(slab, sk); 1301 else 1302 kfree(sk); 1303 module_put(owner); 1304 } 1305 1306 #if IS_ENABLED(CONFIG_NET_CLS_CGROUP) 1307 void sock_update_classid(struct sock *sk) 1308 { 1309 u32 classid; 1310 1311 classid = task_cls_classid(current); 1312 if (classid != sk->sk_classid) 1313 sk->sk_classid = classid; 1314 } 1315 EXPORT_SYMBOL(sock_update_classid); 1316 #endif 1317 1318 #if IS_ENABLED(CONFIG_NETPRIO_CGROUP) 1319 void sock_update_netprioidx(struct sock *sk) 1320 { 1321 if (in_interrupt()) 1322 return; 1323 1324 sk->sk_cgrp_prioidx = task_netprioidx(current); 1325 } 1326 EXPORT_SYMBOL_GPL(sock_update_netprioidx); 1327 #endif 1328 1329 /** 1330 * sk_alloc - All socket objects are allocated here 1331 * @net: the applicable net namespace 1332 * @family: protocol family 1333 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 1334 * @prot: struct proto associated with this new sock instance 1335 */ 1336 struct sock *sk_alloc(struct net *net, int family, gfp_t priority, 1337 struct proto *prot) 1338 { 1339 struct sock *sk; 1340 1341 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family); 1342 if (sk) { 1343 sk->sk_family = family; 1344 /* 1345 * See comment in struct sock definition to understand 1346 * why we need sk_prot_creator -acme 1347 */ 1348 sk->sk_prot = sk->sk_prot_creator = prot; 1349 sock_lock_init(sk); 1350 sock_net_set(sk, get_net(net)); 1351 atomic_set(&sk->sk_wmem_alloc, 1); 1352 1353 sock_update_classid(sk); 1354 sock_update_netprioidx(sk); 1355 } 1356 1357 return sk; 1358 } 1359 EXPORT_SYMBOL(sk_alloc); 1360 1361 static void __sk_free(struct sock *sk) 1362 { 1363 struct sk_filter *filter; 1364 1365 if (sk->sk_destruct) 1366 sk->sk_destruct(sk); 1367 1368 filter = rcu_dereference_check(sk->sk_filter, 1369 atomic_read(&sk->sk_wmem_alloc) == 0); 1370 if (filter) { 1371 sk_filter_uncharge(sk, filter); 1372 RCU_INIT_POINTER(sk->sk_filter, NULL); 1373 } 1374 1375 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP); 1376 1377 if (atomic_read(&sk->sk_omem_alloc)) 1378 pr_debug("%s: optmem leakage (%d bytes) detected\n", 1379 __func__, atomic_read(&sk->sk_omem_alloc)); 1380 1381 if (sk->sk_peer_cred) 1382 put_cred(sk->sk_peer_cred); 1383 put_pid(sk->sk_peer_pid); 1384 put_net(sock_net(sk)); 1385 sk_prot_free(sk->sk_prot_creator, sk); 1386 } 1387 1388 void sk_free(struct sock *sk) 1389 { 1390 /* 1391 * We subtract one from sk_wmem_alloc and can know if 1392 * some packets are still in some tx queue. 1393 * If not null, sock_wfree() will call __sk_free(sk) later 1394 */ 1395 if (atomic_dec_and_test(&sk->sk_wmem_alloc)) 1396 __sk_free(sk); 1397 } 1398 EXPORT_SYMBOL(sk_free); 1399 1400 /* 1401 * Last sock_put should drop reference to sk->sk_net. It has already 1402 * been dropped in sk_change_net. Taking reference to stopping namespace 1403 * is not an option. 1404 * Take reference to a socket to remove it from hash _alive_ and after that 1405 * destroy it in the context of init_net. 1406 */ 1407 void sk_release_kernel(struct sock *sk) 1408 { 1409 if (sk == NULL || sk->sk_socket == NULL) 1410 return; 1411 1412 sock_hold(sk); 1413 sock_release(sk->sk_socket); 1414 release_net(sock_net(sk)); 1415 sock_net_set(sk, get_net(&init_net)); 1416 sock_put(sk); 1417 } 1418 EXPORT_SYMBOL(sk_release_kernel); 1419 1420 static void sk_update_clone(const struct sock *sk, struct sock *newsk) 1421 { 1422 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) 1423 sock_update_memcg(newsk); 1424 } 1425 1426 /** 1427 * sk_clone_lock - clone a socket, and lock its clone 1428 * @sk: the socket to clone 1429 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 1430 * 1431 * Caller must unlock socket even in error path (bh_unlock_sock(newsk)) 1432 */ 1433 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority) 1434 { 1435 struct sock *newsk; 1436 1437 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family); 1438 if (newsk != NULL) { 1439 struct sk_filter *filter; 1440 1441 sock_copy(newsk, sk); 1442 1443 /* SANITY */ 1444 get_net(sock_net(newsk)); 1445 sk_node_init(&newsk->sk_node); 1446 sock_lock_init(newsk); 1447 bh_lock_sock(newsk); 1448 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL; 1449 newsk->sk_backlog.len = 0; 1450 1451 atomic_set(&newsk->sk_rmem_alloc, 0); 1452 /* 1453 * sk_wmem_alloc set to one (see sk_free() and sock_wfree()) 1454 */ 1455 atomic_set(&newsk->sk_wmem_alloc, 1); 1456 atomic_set(&newsk->sk_omem_alloc, 0); 1457 skb_queue_head_init(&newsk->sk_receive_queue); 1458 skb_queue_head_init(&newsk->sk_write_queue); 1459 #ifdef CONFIG_NET_DMA 1460 skb_queue_head_init(&newsk->sk_async_wait_queue); 1461 #endif 1462 1463 spin_lock_init(&newsk->sk_dst_lock); 1464 rwlock_init(&newsk->sk_callback_lock); 1465 lockdep_set_class_and_name(&newsk->sk_callback_lock, 1466 af_callback_keys + newsk->sk_family, 1467 af_family_clock_key_strings[newsk->sk_family]); 1468 1469 newsk->sk_dst_cache = NULL; 1470 newsk->sk_wmem_queued = 0; 1471 newsk->sk_forward_alloc = 0; 1472 newsk->sk_send_head = NULL; 1473 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK; 1474 1475 sock_reset_flag(newsk, SOCK_DONE); 1476 skb_queue_head_init(&newsk->sk_error_queue); 1477 1478 filter = rcu_dereference_protected(newsk->sk_filter, 1); 1479 if (filter != NULL) 1480 sk_filter_charge(newsk, filter); 1481 1482 if (unlikely(xfrm_sk_clone_policy(newsk))) { 1483 /* It is still raw copy of parent, so invalidate 1484 * destructor and make plain sk_free() */ 1485 newsk->sk_destruct = NULL; 1486 bh_unlock_sock(newsk); 1487 sk_free(newsk); 1488 newsk = NULL; 1489 goto out; 1490 } 1491 1492 newsk->sk_err = 0; 1493 newsk->sk_priority = 0; 1494 /* 1495 * Before updating sk_refcnt, we must commit prior changes to memory 1496 * (Documentation/RCU/rculist_nulls.txt for details) 1497 */ 1498 smp_wmb(); 1499 atomic_set(&newsk->sk_refcnt, 2); 1500 1501 /* 1502 * Increment the counter in the same struct proto as the master 1503 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that 1504 * is the same as sk->sk_prot->socks, as this field was copied 1505 * with memcpy). 1506 * 1507 * This _changes_ the previous behaviour, where 1508 * tcp_create_openreq_child always was incrementing the 1509 * equivalent to tcp_prot->socks (inet_sock_nr), so this have 1510 * to be taken into account in all callers. -acme 1511 */ 1512 sk_refcnt_debug_inc(newsk); 1513 sk_set_socket(newsk, NULL); 1514 newsk->sk_wq = NULL; 1515 1516 sk_update_clone(sk, newsk); 1517 1518 if (newsk->sk_prot->sockets_allocated) 1519 sk_sockets_allocated_inc(newsk); 1520 1521 if (newsk->sk_flags & SK_FLAGS_TIMESTAMP) 1522 net_enable_timestamp(); 1523 } 1524 out: 1525 return newsk; 1526 } 1527 EXPORT_SYMBOL_GPL(sk_clone_lock); 1528 1529 void sk_setup_caps(struct sock *sk, struct dst_entry *dst) 1530 { 1531 __sk_dst_set(sk, dst); 1532 sk->sk_route_caps = dst->dev->features; 1533 if (sk->sk_route_caps & NETIF_F_GSO) 1534 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE; 1535 sk->sk_route_caps &= ~sk->sk_route_nocaps; 1536 if (sk_can_gso(sk)) { 1537 if (dst->header_len) { 1538 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 1539 } else { 1540 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM; 1541 sk->sk_gso_max_size = dst->dev->gso_max_size; 1542 sk->sk_gso_max_segs = dst->dev->gso_max_segs; 1543 } 1544 } 1545 } 1546 EXPORT_SYMBOL_GPL(sk_setup_caps); 1547 1548 /* 1549 * Simple resource managers for sockets. 1550 */ 1551 1552 1553 /* 1554 * Write buffer destructor automatically called from kfree_skb. 1555 */ 1556 void sock_wfree(struct sk_buff *skb) 1557 { 1558 struct sock *sk = skb->sk; 1559 unsigned int len = skb->truesize; 1560 1561 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) { 1562 /* 1563 * Keep a reference on sk_wmem_alloc, this will be released 1564 * after sk_write_space() call 1565 */ 1566 atomic_sub(len - 1, &sk->sk_wmem_alloc); 1567 sk->sk_write_space(sk); 1568 len = 1; 1569 } 1570 /* 1571 * if sk_wmem_alloc reaches 0, we must finish what sk_free() 1572 * could not do because of in-flight packets 1573 */ 1574 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc)) 1575 __sk_free(sk); 1576 } 1577 EXPORT_SYMBOL(sock_wfree); 1578 1579 void skb_orphan_partial(struct sk_buff *skb) 1580 { 1581 /* TCP stack sets skb->ooo_okay based on sk_wmem_alloc, 1582 * so we do not completely orphan skb, but transfert all 1583 * accounted bytes but one, to avoid unexpected reorders. 1584 */ 1585 if (skb->destructor == sock_wfree 1586 #ifdef CONFIG_INET 1587 || skb->destructor == tcp_wfree 1588 #endif 1589 ) { 1590 atomic_sub(skb->truesize - 1, &skb->sk->sk_wmem_alloc); 1591 skb->truesize = 1; 1592 } else { 1593 skb_orphan(skb); 1594 } 1595 } 1596 EXPORT_SYMBOL(skb_orphan_partial); 1597 1598 /* 1599 * Read buffer destructor automatically called from kfree_skb. 1600 */ 1601 void sock_rfree(struct sk_buff *skb) 1602 { 1603 struct sock *sk = skb->sk; 1604 unsigned int len = skb->truesize; 1605 1606 atomic_sub(len, &sk->sk_rmem_alloc); 1607 sk_mem_uncharge(sk, len); 1608 } 1609 EXPORT_SYMBOL(sock_rfree); 1610 1611 void sock_edemux(struct sk_buff *skb) 1612 { 1613 struct sock *sk = skb->sk; 1614 1615 #ifdef CONFIG_INET 1616 if (sk->sk_state == TCP_TIME_WAIT) 1617 inet_twsk_put(inet_twsk(sk)); 1618 else 1619 #endif 1620 sock_put(sk); 1621 } 1622 EXPORT_SYMBOL(sock_edemux); 1623 1624 kuid_t sock_i_uid(struct sock *sk) 1625 { 1626 kuid_t uid; 1627 1628 read_lock_bh(&sk->sk_callback_lock); 1629 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID; 1630 read_unlock_bh(&sk->sk_callback_lock); 1631 return uid; 1632 } 1633 EXPORT_SYMBOL(sock_i_uid); 1634 1635 unsigned long sock_i_ino(struct sock *sk) 1636 { 1637 unsigned long ino; 1638 1639 read_lock_bh(&sk->sk_callback_lock); 1640 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0; 1641 read_unlock_bh(&sk->sk_callback_lock); 1642 return ino; 1643 } 1644 EXPORT_SYMBOL(sock_i_ino); 1645 1646 /* 1647 * Allocate a skb from the socket's send buffer. 1648 */ 1649 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 1650 gfp_t priority) 1651 { 1652 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) { 1653 struct sk_buff *skb = alloc_skb(size, priority); 1654 if (skb) { 1655 skb_set_owner_w(skb, sk); 1656 return skb; 1657 } 1658 } 1659 return NULL; 1660 } 1661 EXPORT_SYMBOL(sock_wmalloc); 1662 1663 /* 1664 * Allocate a skb from the socket's receive buffer. 1665 */ 1666 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force, 1667 gfp_t priority) 1668 { 1669 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) { 1670 struct sk_buff *skb = alloc_skb(size, priority); 1671 if (skb) { 1672 skb_set_owner_r(skb, sk); 1673 return skb; 1674 } 1675 } 1676 return NULL; 1677 } 1678 1679 /* 1680 * Allocate a memory block from the socket's option memory buffer. 1681 */ 1682 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority) 1683 { 1684 if ((unsigned int)size <= sysctl_optmem_max && 1685 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) { 1686 void *mem; 1687 /* First do the add, to avoid the race if kmalloc 1688 * might sleep. 1689 */ 1690 atomic_add(size, &sk->sk_omem_alloc); 1691 mem = kmalloc(size, priority); 1692 if (mem) 1693 return mem; 1694 atomic_sub(size, &sk->sk_omem_alloc); 1695 } 1696 return NULL; 1697 } 1698 EXPORT_SYMBOL(sock_kmalloc); 1699 1700 /* 1701 * Free an option memory block. 1702 */ 1703 void sock_kfree_s(struct sock *sk, void *mem, int size) 1704 { 1705 kfree(mem); 1706 atomic_sub(size, &sk->sk_omem_alloc); 1707 } 1708 EXPORT_SYMBOL(sock_kfree_s); 1709 1710 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock. 1711 I think, these locks should be removed for datagram sockets. 1712 */ 1713 static long sock_wait_for_wmem(struct sock *sk, long timeo) 1714 { 1715 DEFINE_WAIT(wait); 1716 1717 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags); 1718 for (;;) { 1719 if (!timeo) 1720 break; 1721 if (signal_pending(current)) 1722 break; 1723 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 1724 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 1725 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) 1726 break; 1727 if (sk->sk_shutdown & SEND_SHUTDOWN) 1728 break; 1729 if (sk->sk_err) 1730 break; 1731 timeo = schedule_timeout(timeo); 1732 } 1733 finish_wait(sk_sleep(sk), &wait); 1734 return timeo; 1735 } 1736 1737 1738 /* 1739 * Generic send/receive buffer handlers 1740 */ 1741 1742 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 1743 unsigned long data_len, int noblock, 1744 int *errcode) 1745 { 1746 struct sk_buff *skb; 1747 gfp_t gfp_mask; 1748 long timeo; 1749 int err; 1750 int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT; 1751 1752 err = -EMSGSIZE; 1753 if (npages > MAX_SKB_FRAGS) 1754 goto failure; 1755 1756 gfp_mask = sk->sk_allocation; 1757 if (gfp_mask & __GFP_WAIT) 1758 gfp_mask |= __GFP_REPEAT; 1759 1760 timeo = sock_sndtimeo(sk, noblock); 1761 while (1) { 1762 err = sock_error(sk); 1763 if (err != 0) 1764 goto failure; 1765 1766 err = -EPIPE; 1767 if (sk->sk_shutdown & SEND_SHUTDOWN) 1768 goto failure; 1769 1770 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) { 1771 skb = alloc_skb(header_len, gfp_mask); 1772 if (skb) { 1773 int i; 1774 1775 /* No pages, we're done... */ 1776 if (!data_len) 1777 break; 1778 1779 skb->truesize += data_len; 1780 skb_shinfo(skb)->nr_frags = npages; 1781 for (i = 0; i < npages; i++) { 1782 struct page *page; 1783 1784 page = alloc_pages(sk->sk_allocation, 0); 1785 if (!page) { 1786 err = -ENOBUFS; 1787 skb_shinfo(skb)->nr_frags = i; 1788 kfree_skb(skb); 1789 goto failure; 1790 } 1791 1792 __skb_fill_page_desc(skb, i, 1793 page, 0, 1794 (data_len >= PAGE_SIZE ? 1795 PAGE_SIZE : 1796 data_len)); 1797 data_len -= PAGE_SIZE; 1798 } 1799 1800 /* Full success... */ 1801 break; 1802 } 1803 err = -ENOBUFS; 1804 goto failure; 1805 } 1806 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags); 1807 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 1808 err = -EAGAIN; 1809 if (!timeo) 1810 goto failure; 1811 if (signal_pending(current)) 1812 goto interrupted; 1813 timeo = sock_wait_for_wmem(sk, timeo); 1814 } 1815 1816 skb_set_owner_w(skb, sk); 1817 return skb; 1818 1819 interrupted: 1820 err = sock_intr_errno(timeo); 1821 failure: 1822 *errcode = err; 1823 return NULL; 1824 } 1825 EXPORT_SYMBOL(sock_alloc_send_pskb); 1826 1827 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size, 1828 int noblock, int *errcode) 1829 { 1830 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode); 1831 } 1832 EXPORT_SYMBOL(sock_alloc_send_skb); 1833 1834 /* On 32bit arches, an skb frag is limited to 2^15 */ 1835 #define SKB_FRAG_PAGE_ORDER get_order(32768) 1836 1837 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag) 1838 { 1839 int order; 1840 1841 if (pfrag->page) { 1842 if (atomic_read(&pfrag->page->_count) == 1) { 1843 pfrag->offset = 0; 1844 return true; 1845 } 1846 if (pfrag->offset < pfrag->size) 1847 return true; 1848 put_page(pfrag->page); 1849 } 1850 1851 /* We restrict high order allocations to users that can afford to wait */ 1852 order = (sk->sk_allocation & __GFP_WAIT) ? SKB_FRAG_PAGE_ORDER : 0; 1853 1854 do { 1855 gfp_t gfp = sk->sk_allocation; 1856 1857 if (order) 1858 gfp |= __GFP_COMP | __GFP_NOWARN; 1859 pfrag->page = alloc_pages(gfp, order); 1860 if (likely(pfrag->page)) { 1861 pfrag->offset = 0; 1862 pfrag->size = PAGE_SIZE << order; 1863 return true; 1864 } 1865 } while (--order >= 0); 1866 1867 sk_enter_memory_pressure(sk); 1868 sk_stream_moderate_sndbuf(sk); 1869 return false; 1870 } 1871 EXPORT_SYMBOL(sk_page_frag_refill); 1872 1873 static void __lock_sock(struct sock *sk) 1874 __releases(&sk->sk_lock.slock) 1875 __acquires(&sk->sk_lock.slock) 1876 { 1877 DEFINE_WAIT(wait); 1878 1879 for (;;) { 1880 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait, 1881 TASK_UNINTERRUPTIBLE); 1882 spin_unlock_bh(&sk->sk_lock.slock); 1883 schedule(); 1884 spin_lock_bh(&sk->sk_lock.slock); 1885 if (!sock_owned_by_user(sk)) 1886 break; 1887 } 1888 finish_wait(&sk->sk_lock.wq, &wait); 1889 } 1890 1891 static void __release_sock(struct sock *sk) 1892 __releases(&sk->sk_lock.slock) 1893 __acquires(&sk->sk_lock.slock) 1894 { 1895 struct sk_buff *skb = sk->sk_backlog.head; 1896 1897 do { 1898 sk->sk_backlog.head = sk->sk_backlog.tail = NULL; 1899 bh_unlock_sock(sk); 1900 1901 do { 1902 struct sk_buff *next = skb->next; 1903 1904 prefetch(next); 1905 WARN_ON_ONCE(skb_dst_is_noref(skb)); 1906 skb->next = NULL; 1907 sk_backlog_rcv(sk, skb); 1908 1909 /* 1910 * We are in process context here with softirqs 1911 * disabled, use cond_resched_softirq() to preempt. 1912 * This is safe to do because we've taken the backlog 1913 * queue private: 1914 */ 1915 cond_resched_softirq(); 1916 1917 skb = next; 1918 } while (skb != NULL); 1919 1920 bh_lock_sock(sk); 1921 } while ((skb = sk->sk_backlog.head) != NULL); 1922 1923 /* 1924 * Doing the zeroing here guarantee we can not loop forever 1925 * while a wild producer attempts to flood us. 1926 */ 1927 sk->sk_backlog.len = 0; 1928 } 1929 1930 /** 1931 * sk_wait_data - wait for data to arrive at sk_receive_queue 1932 * @sk: sock to wait on 1933 * @timeo: for how long 1934 * 1935 * Now socket state including sk->sk_err is changed only under lock, 1936 * hence we may omit checks after joining wait queue. 1937 * We check receive queue before schedule() only as optimization; 1938 * it is very likely that release_sock() added new data. 1939 */ 1940 int sk_wait_data(struct sock *sk, long *timeo) 1941 { 1942 int rc; 1943 DEFINE_WAIT(wait); 1944 1945 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 1946 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags); 1947 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue)); 1948 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags); 1949 finish_wait(sk_sleep(sk), &wait); 1950 return rc; 1951 } 1952 EXPORT_SYMBOL(sk_wait_data); 1953 1954 /** 1955 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated 1956 * @sk: socket 1957 * @size: memory size to allocate 1958 * @kind: allocation type 1959 * 1960 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means 1961 * rmem allocation. This function assumes that protocols which have 1962 * memory_pressure use sk_wmem_queued as write buffer accounting. 1963 */ 1964 int __sk_mem_schedule(struct sock *sk, int size, int kind) 1965 { 1966 struct proto *prot = sk->sk_prot; 1967 int amt = sk_mem_pages(size); 1968 long allocated; 1969 int parent_status = UNDER_LIMIT; 1970 1971 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM; 1972 1973 allocated = sk_memory_allocated_add(sk, amt, &parent_status); 1974 1975 /* Under limit. */ 1976 if (parent_status == UNDER_LIMIT && 1977 allocated <= sk_prot_mem_limits(sk, 0)) { 1978 sk_leave_memory_pressure(sk); 1979 return 1; 1980 } 1981 1982 /* Under pressure. (we or our parents) */ 1983 if ((parent_status > SOFT_LIMIT) || 1984 allocated > sk_prot_mem_limits(sk, 1)) 1985 sk_enter_memory_pressure(sk); 1986 1987 /* Over hard limit (we or our parents) */ 1988 if ((parent_status == OVER_LIMIT) || 1989 (allocated > sk_prot_mem_limits(sk, 2))) 1990 goto suppress_allocation; 1991 1992 /* guarantee minimum buffer size under pressure */ 1993 if (kind == SK_MEM_RECV) { 1994 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0]) 1995 return 1; 1996 1997 } else { /* SK_MEM_SEND */ 1998 if (sk->sk_type == SOCK_STREAM) { 1999 if (sk->sk_wmem_queued < prot->sysctl_wmem[0]) 2000 return 1; 2001 } else if (atomic_read(&sk->sk_wmem_alloc) < 2002 prot->sysctl_wmem[0]) 2003 return 1; 2004 } 2005 2006 if (sk_has_memory_pressure(sk)) { 2007 int alloc; 2008 2009 if (!sk_under_memory_pressure(sk)) 2010 return 1; 2011 alloc = sk_sockets_allocated_read_positive(sk); 2012 if (sk_prot_mem_limits(sk, 2) > alloc * 2013 sk_mem_pages(sk->sk_wmem_queued + 2014 atomic_read(&sk->sk_rmem_alloc) + 2015 sk->sk_forward_alloc)) 2016 return 1; 2017 } 2018 2019 suppress_allocation: 2020 2021 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) { 2022 sk_stream_moderate_sndbuf(sk); 2023 2024 /* Fail only if socket is _under_ its sndbuf. 2025 * In this case we cannot block, so that we have to fail. 2026 */ 2027 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf) 2028 return 1; 2029 } 2030 2031 trace_sock_exceed_buf_limit(sk, prot, allocated); 2032 2033 /* Alas. Undo changes. */ 2034 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM; 2035 2036 sk_memory_allocated_sub(sk, amt); 2037 2038 return 0; 2039 } 2040 EXPORT_SYMBOL(__sk_mem_schedule); 2041 2042 /** 2043 * __sk_reclaim - reclaim memory_allocated 2044 * @sk: socket 2045 */ 2046 void __sk_mem_reclaim(struct sock *sk) 2047 { 2048 sk_memory_allocated_sub(sk, 2049 sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT); 2050 sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1; 2051 2052 if (sk_under_memory_pressure(sk) && 2053 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0))) 2054 sk_leave_memory_pressure(sk); 2055 } 2056 EXPORT_SYMBOL(__sk_mem_reclaim); 2057 2058 2059 /* 2060 * Set of default routines for initialising struct proto_ops when 2061 * the protocol does not support a particular function. In certain 2062 * cases where it makes no sense for a protocol to have a "do nothing" 2063 * function, some default processing is provided. 2064 */ 2065 2066 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len) 2067 { 2068 return -EOPNOTSUPP; 2069 } 2070 EXPORT_SYMBOL(sock_no_bind); 2071 2072 int sock_no_connect(struct socket *sock, struct sockaddr *saddr, 2073 int len, int flags) 2074 { 2075 return -EOPNOTSUPP; 2076 } 2077 EXPORT_SYMBOL(sock_no_connect); 2078 2079 int sock_no_socketpair(struct socket *sock1, struct socket *sock2) 2080 { 2081 return -EOPNOTSUPP; 2082 } 2083 EXPORT_SYMBOL(sock_no_socketpair); 2084 2085 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags) 2086 { 2087 return -EOPNOTSUPP; 2088 } 2089 EXPORT_SYMBOL(sock_no_accept); 2090 2091 int sock_no_getname(struct socket *sock, struct sockaddr *saddr, 2092 int *len, int peer) 2093 { 2094 return -EOPNOTSUPP; 2095 } 2096 EXPORT_SYMBOL(sock_no_getname); 2097 2098 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt) 2099 { 2100 return 0; 2101 } 2102 EXPORT_SYMBOL(sock_no_poll); 2103 2104 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 2105 { 2106 return -EOPNOTSUPP; 2107 } 2108 EXPORT_SYMBOL(sock_no_ioctl); 2109 2110 int sock_no_listen(struct socket *sock, int backlog) 2111 { 2112 return -EOPNOTSUPP; 2113 } 2114 EXPORT_SYMBOL(sock_no_listen); 2115 2116 int sock_no_shutdown(struct socket *sock, int how) 2117 { 2118 return -EOPNOTSUPP; 2119 } 2120 EXPORT_SYMBOL(sock_no_shutdown); 2121 2122 int sock_no_setsockopt(struct socket *sock, int level, int optname, 2123 char __user *optval, unsigned int optlen) 2124 { 2125 return -EOPNOTSUPP; 2126 } 2127 EXPORT_SYMBOL(sock_no_setsockopt); 2128 2129 int sock_no_getsockopt(struct socket *sock, int level, int optname, 2130 char __user *optval, int __user *optlen) 2131 { 2132 return -EOPNOTSUPP; 2133 } 2134 EXPORT_SYMBOL(sock_no_getsockopt); 2135 2136 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m, 2137 size_t len) 2138 { 2139 return -EOPNOTSUPP; 2140 } 2141 EXPORT_SYMBOL(sock_no_sendmsg); 2142 2143 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m, 2144 size_t len, int flags) 2145 { 2146 return -EOPNOTSUPP; 2147 } 2148 EXPORT_SYMBOL(sock_no_recvmsg); 2149 2150 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma) 2151 { 2152 /* Mirror missing mmap method error code */ 2153 return -ENODEV; 2154 } 2155 EXPORT_SYMBOL(sock_no_mmap); 2156 2157 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags) 2158 { 2159 ssize_t res; 2160 struct msghdr msg = {.msg_flags = flags}; 2161 struct kvec iov; 2162 char *kaddr = kmap(page); 2163 iov.iov_base = kaddr + offset; 2164 iov.iov_len = size; 2165 res = kernel_sendmsg(sock, &msg, &iov, 1, size); 2166 kunmap(page); 2167 return res; 2168 } 2169 EXPORT_SYMBOL(sock_no_sendpage); 2170 2171 /* 2172 * Default Socket Callbacks 2173 */ 2174 2175 static void sock_def_wakeup(struct sock *sk) 2176 { 2177 struct socket_wq *wq; 2178 2179 rcu_read_lock(); 2180 wq = rcu_dereference(sk->sk_wq); 2181 if (wq_has_sleeper(wq)) 2182 wake_up_interruptible_all(&wq->wait); 2183 rcu_read_unlock(); 2184 } 2185 2186 static void sock_def_error_report(struct sock *sk) 2187 { 2188 struct socket_wq *wq; 2189 2190 rcu_read_lock(); 2191 wq = rcu_dereference(sk->sk_wq); 2192 if (wq_has_sleeper(wq)) 2193 wake_up_interruptible_poll(&wq->wait, POLLERR); 2194 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR); 2195 rcu_read_unlock(); 2196 } 2197 2198 static void sock_def_readable(struct sock *sk, int len) 2199 { 2200 struct socket_wq *wq; 2201 2202 rcu_read_lock(); 2203 wq = rcu_dereference(sk->sk_wq); 2204 if (wq_has_sleeper(wq)) 2205 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI | 2206 POLLRDNORM | POLLRDBAND); 2207 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN); 2208 rcu_read_unlock(); 2209 } 2210 2211 static void sock_def_write_space(struct sock *sk) 2212 { 2213 struct socket_wq *wq; 2214 2215 rcu_read_lock(); 2216 2217 /* Do not wake up a writer until he can make "significant" 2218 * progress. --DaveM 2219 */ 2220 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) { 2221 wq = rcu_dereference(sk->sk_wq); 2222 if (wq_has_sleeper(wq)) 2223 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT | 2224 POLLWRNORM | POLLWRBAND); 2225 2226 /* Should agree with poll, otherwise some programs break */ 2227 if (sock_writeable(sk)) 2228 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT); 2229 } 2230 2231 rcu_read_unlock(); 2232 } 2233 2234 static void sock_def_destruct(struct sock *sk) 2235 { 2236 kfree(sk->sk_protinfo); 2237 } 2238 2239 void sk_send_sigurg(struct sock *sk) 2240 { 2241 if (sk->sk_socket && sk->sk_socket->file) 2242 if (send_sigurg(&sk->sk_socket->file->f_owner)) 2243 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI); 2244 } 2245 EXPORT_SYMBOL(sk_send_sigurg); 2246 2247 void sk_reset_timer(struct sock *sk, struct timer_list* timer, 2248 unsigned long expires) 2249 { 2250 if (!mod_timer(timer, expires)) 2251 sock_hold(sk); 2252 } 2253 EXPORT_SYMBOL(sk_reset_timer); 2254 2255 void sk_stop_timer(struct sock *sk, struct timer_list* timer) 2256 { 2257 if (del_timer(timer)) 2258 __sock_put(sk); 2259 } 2260 EXPORT_SYMBOL(sk_stop_timer); 2261 2262 void sock_init_data(struct socket *sock, struct sock *sk) 2263 { 2264 skb_queue_head_init(&sk->sk_receive_queue); 2265 skb_queue_head_init(&sk->sk_write_queue); 2266 skb_queue_head_init(&sk->sk_error_queue); 2267 #ifdef CONFIG_NET_DMA 2268 skb_queue_head_init(&sk->sk_async_wait_queue); 2269 #endif 2270 2271 sk->sk_send_head = NULL; 2272 2273 init_timer(&sk->sk_timer); 2274 2275 sk->sk_allocation = GFP_KERNEL; 2276 sk->sk_rcvbuf = sysctl_rmem_default; 2277 sk->sk_sndbuf = sysctl_wmem_default; 2278 sk->sk_state = TCP_CLOSE; 2279 sk_set_socket(sk, sock); 2280 2281 sock_set_flag(sk, SOCK_ZAPPED); 2282 2283 if (sock) { 2284 sk->sk_type = sock->type; 2285 sk->sk_wq = sock->wq; 2286 sock->sk = sk; 2287 } else 2288 sk->sk_wq = NULL; 2289 2290 spin_lock_init(&sk->sk_dst_lock); 2291 rwlock_init(&sk->sk_callback_lock); 2292 lockdep_set_class_and_name(&sk->sk_callback_lock, 2293 af_callback_keys + sk->sk_family, 2294 af_family_clock_key_strings[sk->sk_family]); 2295 2296 sk->sk_state_change = sock_def_wakeup; 2297 sk->sk_data_ready = sock_def_readable; 2298 sk->sk_write_space = sock_def_write_space; 2299 sk->sk_error_report = sock_def_error_report; 2300 sk->sk_destruct = sock_def_destruct; 2301 2302 sk->sk_frag.page = NULL; 2303 sk->sk_frag.offset = 0; 2304 sk->sk_peek_off = -1; 2305 2306 sk->sk_peer_pid = NULL; 2307 sk->sk_peer_cred = NULL; 2308 sk->sk_write_pending = 0; 2309 sk->sk_rcvlowat = 1; 2310 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; 2311 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT; 2312 2313 sk->sk_stamp = ktime_set(-1L, 0); 2314 2315 #ifdef CONFIG_NET_LL_RX_POLL 2316 sk->sk_napi_id = 0; 2317 sk->sk_ll_usec = sysctl_net_busy_read; 2318 #endif 2319 2320 /* 2321 * Before updating sk_refcnt, we must commit prior changes to memory 2322 * (Documentation/RCU/rculist_nulls.txt for details) 2323 */ 2324 smp_wmb(); 2325 atomic_set(&sk->sk_refcnt, 1); 2326 atomic_set(&sk->sk_drops, 0); 2327 } 2328 EXPORT_SYMBOL(sock_init_data); 2329 2330 void lock_sock_nested(struct sock *sk, int subclass) 2331 { 2332 might_sleep(); 2333 spin_lock_bh(&sk->sk_lock.slock); 2334 if (sk->sk_lock.owned) 2335 __lock_sock(sk); 2336 sk->sk_lock.owned = 1; 2337 spin_unlock(&sk->sk_lock.slock); 2338 /* 2339 * The sk_lock has mutex_lock() semantics here: 2340 */ 2341 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_); 2342 local_bh_enable(); 2343 } 2344 EXPORT_SYMBOL(lock_sock_nested); 2345 2346 void release_sock(struct sock *sk) 2347 { 2348 /* 2349 * The sk_lock has mutex_unlock() semantics: 2350 */ 2351 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_); 2352 2353 spin_lock_bh(&sk->sk_lock.slock); 2354 if (sk->sk_backlog.tail) 2355 __release_sock(sk); 2356 2357 if (sk->sk_prot->release_cb) 2358 sk->sk_prot->release_cb(sk); 2359 2360 sk->sk_lock.owned = 0; 2361 if (waitqueue_active(&sk->sk_lock.wq)) 2362 wake_up(&sk->sk_lock.wq); 2363 spin_unlock_bh(&sk->sk_lock.slock); 2364 } 2365 EXPORT_SYMBOL(release_sock); 2366 2367 /** 2368 * lock_sock_fast - fast version of lock_sock 2369 * @sk: socket 2370 * 2371 * This version should be used for very small section, where process wont block 2372 * return false if fast path is taken 2373 * sk_lock.slock locked, owned = 0, BH disabled 2374 * return true if slow path is taken 2375 * sk_lock.slock unlocked, owned = 1, BH enabled 2376 */ 2377 bool lock_sock_fast(struct sock *sk) 2378 { 2379 might_sleep(); 2380 spin_lock_bh(&sk->sk_lock.slock); 2381 2382 if (!sk->sk_lock.owned) 2383 /* 2384 * Note : We must disable BH 2385 */ 2386 return false; 2387 2388 __lock_sock(sk); 2389 sk->sk_lock.owned = 1; 2390 spin_unlock(&sk->sk_lock.slock); 2391 /* 2392 * The sk_lock has mutex_lock() semantics here: 2393 */ 2394 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_); 2395 local_bh_enable(); 2396 return true; 2397 } 2398 EXPORT_SYMBOL(lock_sock_fast); 2399 2400 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp) 2401 { 2402 struct timeval tv; 2403 if (!sock_flag(sk, SOCK_TIMESTAMP)) 2404 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 2405 tv = ktime_to_timeval(sk->sk_stamp); 2406 if (tv.tv_sec == -1) 2407 return -ENOENT; 2408 if (tv.tv_sec == 0) { 2409 sk->sk_stamp = ktime_get_real(); 2410 tv = ktime_to_timeval(sk->sk_stamp); 2411 } 2412 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0; 2413 } 2414 EXPORT_SYMBOL(sock_get_timestamp); 2415 2416 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp) 2417 { 2418 struct timespec ts; 2419 if (!sock_flag(sk, SOCK_TIMESTAMP)) 2420 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 2421 ts = ktime_to_timespec(sk->sk_stamp); 2422 if (ts.tv_sec == -1) 2423 return -ENOENT; 2424 if (ts.tv_sec == 0) { 2425 sk->sk_stamp = ktime_get_real(); 2426 ts = ktime_to_timespec(sk->sk_stamp); 2427 } 2428 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0; 2429 } 2430 EXPORT_SYMBOL(sock_get_timestampns); 2431 2432 void sock_enable_timestamp(struct sock *sk, int flag) 2433 { 2434 if (!sock_flag(sk, flag)) { 2435 unsigned long previous_flags = sk->sk_flags; 2436 2437 sock_set_flag(sk, flag); 2438 /* 2439 * we just set one of the two flags which require net 2440 * time stamping, but time stamping might have been on 2441 * already because of the other one 2442 */ 2443 if (!(previous_flags & SK_FLAGS_TIMESTAMP)) 2444 net_enable_timestamp(); 2445 } 2446 } 2447 2448 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, 2449 int level, int type) 2450 { 2451 struct sock_exterr_skb *serr; 2452 struct sk_buff *skb, *skb2; 2453 int copied, err; 2454 2455 err = -EAGAIN; 2456 skb = skb_dequeue(&sk->sk_error_queue); 2457 if (skb == NULL) 2458 goto out; 2459 2460 copied = skb->len; 2461 if (copied > len) { 2462 msg->msg_flags |= MSG_TRUNC; 2463 copied = len; 2464 } 2465 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied); 2466 if (err) 2467 goto out_free_skb; 2468 2469 sock_recv_timestamp(msg, sk, skb); 2470 2471 serr = SKB_EXT_ERR(skb); 2472 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee); 2473 2474 msg->msg_flags |= MSG_ERRQUEUE; 2475 err = copied; 2476 2477 /* Reset and regenerate socket error */ 2478 spin_lock_bh(&sk->sk_error_queue.lock); 2479 sk->sk_err = 0; 2480 if ((skb2 = skb_peek(&sk->sk_error_queue)) != NULL) { 2481 sk->sk_err = SKB_EXT_ERR(skb2)->ee.ee_errno; 2482 spin_unlock_bh(&sk->sk_error_queue.lock); 2483 sk->sk_error_report(sk); 2484 } else 2485 spin_unlock_bh(&sk->sk_error_queue.lock); 2486 2487 out_free_skb: 2488 kfree_skb(skb); 2489 out: 2490 return err; 2491 } 2492 EXPORT_SYMBOL(sock_recv_errqueue); 2493 2494 /* 2495 * Get a socket option on an socket. 2496 * 2497 * FIX: POSIX 1003.1g is very ambiguous here. It states that 2498 * asynchronous errors should be reported by getsockopt. We assume 2499 * this means if you specify SO_ERROR (otherwise whats the point of it). 2500 */ 2501 int sock_common_getsockopt(struct socket *sock, int level, int optname, 2502 char __user *optval, int __user *optlen) 2503 { 2504 struct sock *sk = sock->sk; 2505 2506 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen); 2507 } 2508 EXPORT_SYMBOL(sock_common_getsockopt); 2509 2510 #ifdef CONFIG_COMPAT 2511 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname, 2512 char __user *optval, int __user *optlen) 2513 { 2514 struct sock *sk = sock->sk; 2515 2516 if (sk->sk_prot->compat_getsockopt != NULL) 2517 return sk->sk_prot->compat_getsockopt(sk, level, optname, 2518 optval, optlen); 2519 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen); 2520 } 2521 EXPORT_SYMBOL(compat_sock_common_getsockopt); 2522 #endif 2523 2524 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock, 2525 struct msghdr *msg, size_t size, int flags) 2526 { 2527 struct sock *sk = sock->sk; 2528 int addr_len = 0; 2529 int err; 2530 2531 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT, 2532 flags & ~MSG_DONTWAIT, &addr_len); 2533 if (err >= 0) 2534 msg->msg_namelen = addr_len; 2535 return err; 2536 } 2537 EXPORT_SYMBOL(sock_common_recvmsg); 2538 2539 /* 2540 * Set socket options on an inet socket. 2541 */ 2542 int sock_common_setsockopt(struct socket *sock, int level, int optname, 2543 char __user *optval, unsigned int optlen) 2544 { 2545 struct sock *sk = sock->sk; 2546 2547 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen); 2548 } 2549 EXPORT_SYMBOL(sock_common_setsockopt); 2550 2551 #ifdef CONFIG_COMPAT 2552 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname, 2553 char __user *optval, unsigned int optlen) 2554 { 2555 struct sock *sk = sock->sk; 2556 2557 if (sk->sk_prot->compat_setsockopt != NULL) 2558 return sk->sk_prot->compat_setsockopt(sk, level, optname, 2559 optval, optlen); 2560 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen); 2561 } 2562 EXPORT_SYMBOL(compat_sock_common_setsockopt); 2563 #endif 2564 2565 void sk_common_release(struct sock *sk) 2566 { 2567 if (sk->sk_prot->destroy) 2568 sk->sk_prot->destroy(sk); 2569 2570 /* 2571 * Observation: when sock_common_release is called, processes have 2572 * no access to socket. But net still has. 2573 * Step one, detach it from networking: 2574 * 2575 * A. Remove from hash tables. 2576 */ 2577 2578 sk->sk_prot->unhash(sk); 2579 2580 /* 2581 * In this point socket cannot receive new packets, but it is possible 2582 * that some packets are in flight because some CPU runs receiver and 2583 * did hash table lookup before we unhashed socket. They will achieve 2584 * receive queue and will be purged by socket destructor. 2585 * 2586 * Also we still have packets pending on receive queue and probably, 2587 * our own packets waiting in device queues. sock_destroy will drain 2588 * receive queue, but transmitted packets will delay socket destruction 2589 * until the last reference will be released. 2590 */ 2591 2592 sock_orphan(sk); 2593 2594 xfrm_sk_free_policy(sk); 2595 2596 sk_refcnt_debug_release(sk); 2597 2598 if (sk->sk_frag.page) { 2599 put_page(sk->sk_frag.page); 2600 sk->sk_frag.page = NULL; 2601 } 2602 2603 sock_put(sk); 2604 } 2605 EXPORT_SYMBOL(sk_common_release); 2606 2607 #ifdef CONFIG_PROC_FS 2608 #define PROTO_INUSE_NR 64 /* should be enough for the first time */ 2609 struct prot_inuse { 2610 int val[PROTO_INUSE_NR]; 2611 }; 2612 2613 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR); 2614 2615 #ifdef CONFIG_NET_NS 2616 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val) 2617 { 2618 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val); 2619 } 2620 EXPORT_SYMBOL_GPL(sock_prot_inuse_add); 2621 2622 int sock_prot_inuse_get(struct net *net, struct proto *prot) 2623 { 2624 int cpu, idx = prot->inuse_idx; 2625 int res = 0; 2626 2627 for_each_possible_cpu(cpu) 2628 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx]; 2629 2630 return res >= 0 ? res : 0; 2631 } 2632 EXPORT_SYMBOL_GPL(sock_prot_inuse_get); 2633 2634 static int __net_init sock_inuse_init_net(struct net *net) 2635 { 2636 net->core.inuse = alloc_percpu(struct prot_inuse); 2637 return net->core.inuse ? 0 : -ENOMEM; 2638 } 2639 2640 static void __net_exit sock_inuse_exit_net(struct net *net) 2641 { 2642 free_percpu(net->core.inuse); 2643 } 2644 2645 static struct pernet_operations net_inuse_ops = { 2646 .init = sock_inuse_init_net, 2647 .exit = sock_inuse_exit_net, 2648 }; 2649 2650 static __init int net_inuse_init(void) 2651 { 2652 if (register_pernet_subsys(&net_inuse_ops)) 2653 panic("Cannot initialize net inuse counters"); 2654 2655 return 0; 2656 } 2657 2658 core_initcall(net_inuse_init); 2659 #else 2660 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse); 2661 2662 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val) 2663 { 2664 __this_cpu_add(prot_inuse.val[prot->inuse_idx], val); 2665 } 2666 EXPORT_SYMBOL_GPL(sock_prot_inuse_add); 2667 2668 int sock_prot_inuse_get(struct net *net, struct proto *prot) 2669 { 2670 int cpu, idx = prot->inuse_idx; 2671 int res = 0; 2672 2673 for_each_possible_cpu(cpu) 2674 res += per_cpu(prot_inuse, cpu).val[idx]; 2675 2676 return res >= 0 ? res : 0; 2677 } 2678 EXPORT_SYMBOL_GPL(sock_prot_inuse_get); 2679 #endif 2680 2681 static void assign_proto_idx(struct proto *prot) 2682 { 2683 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR); 2684 2685 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) { 2686 pr_err("PROTO_INUSE_NR exhausted\n"); 2687 return; 2688 } 2689 2690 set_bit(prot->inuse_idx, proto_inuse_idx); 2691 } 2692 2693 static void release_proto_idx(struct proto *prot) 2694 { 2695 if (prot->inuse_idx != PROTO_INUSE_NR - 1) 2696 clear_bit(prot->inuse_idx, proto_inuse_idx); 2697 } 2698 #else 2699 static inline void assign_proto_idx(struct proto *prot) 2700 { 2701 } 2702 2703 static inline void release_proto_idx(struct proto *prot) 2704 { 2705 } 2706 #endif 2707 2708 int proto_register(struct proto *prot, int alloc_slab) 2709 { 2710 if (alloc_slab) { 2711 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0, 2712 SLAB_HWCACHE_ALIGN | prot->slab_flags, 2713 NULL); 2714 2715 if (prot->slab == NULL) { 2716 pr_crit("%s: Can't create sock SLAB cache!\n", 2717 prot->name); 2718 goto out; 2719 } 2720 2721 if (prot->rsk_prot != NULL) { 2722 prot->rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", prot->name); 2723 if (prot->rsk_prot->slab_name == NULL) 2724 goto out_free_sock_slab; 2725 2726 prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name, 2727 prot->rsk_prot->obj_size, 0, 2728 SLAB_HWCACHE_ALIGN, NULL); 2729 2730 if (prot->rsk_prot->slab == NULL) { 2731 pr_crit("%s: Can't create request sock SLAB cache!\n", 2732 prot->name); 2733 goto out_free_request_sock_slab_name; 2734 } 2735 } 2736 2737 if (prot->twsk_prot != NULL) { 2738 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name); 2739 2740 if (prot->twsk_prot->twsk_slab_name == NULL) 2741 goto out_free_request_sock_slab; 2742 2743 prot->twsk_prot->twsk_slab = 2744 kmem_cache_create(prot->twsk_prot->twsk_slab_name, 2745 prot->twsk_prot->twsk_obj_size, 2746 0, 2747 SLAB_HWCACHE_ALIGN | 2748 prot->slab_flags, 2749 NULL); 2750 if (prot->twsk_prot->twsk_slab == NULL) 2751 goto out_free_timewait_sock_slab_name; 2752 } 2753 } 2754 2755 mutex_lock(&proto_list_mutex); 2756 list_add(&prot->node, &proto_list); 2757 assign_proto_idx(prot); 2758 mutex_unlock(&proto_list_mutex); 2759 return 0; 2760 2761 out_free_timewait_sock_slab_name: 2762 kfree(prot->twsk_prot->twsk_slab_name); 2763 out_free_request_sock_slab: 2764 if (prot->rsk_prot && prot->rsk_prot->slab) { 2765 kmem_cache_destroy(prot->rsk_prot->slab); 2766 prot->rsk_prot->slab = NULL; 2767 } 2768 out_free_request_sock_slab_name: 2769 if (prot->rsk_prot) 2770 kfree(prot->rsk_prot->slab_name); 2771 out_free_sock_slab: 2772 kmem_cache_destroy(prot->slab); 2773 prot->slab = NULL; 2774 out: 2775 return -ENOBUFS; 2776 } 2777 EXPORT_SYMBOL(proto_register); 2778 2779 void proto_unregister(struct proto *prot) 2780 { 2781 mutex_lock(&proto_list_mutex); 2782 release_proto_idx(prot); 2783 list_del(&prot->node); 2784 mutex_unlock(&proto_list_mutex); 2785 2786 if (prot->slab != NULL) { 2787 kmem_cache_destroy(prot->slab); 2788 prot->slab = NULL; 2789 } 2790 2791 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) { 2792 kmem_cache_destroy(prot->rsk_prot->slab); 2793 kfree(prot->rsk_prot->slab_name); 2794 prot->rsk_prot->slab = NULL; 2795 } 2796 2797 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) { 2798 kmem_cache_destroy(prot->twsk_prot->twsk_slab); 2799 kfree(prot->twsk_prot->twsk_slab_name); 2800 prot->twsk_prot->twsk_slab = NULL; 2801 } 2802 } 2803 EXPORT_SYMBOL(proto_unregister); 2804 2805 #ifdef CONFIG_PROC_FS 2806 static void *proto_seq_start(struct seq_file *seq, loff_t *pos) 2807 __acquires(proto_list_mutex) 2808 { 2809 mutex_lock(&proto_list_mutex); 2810 return seq_list_start_head(&proto_list, *pos); 2811 } 2812 2813 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos) 2814 { 2815 return seq_list_next(v, &proto_list, pos); 2816 } 2817 2818 static void proto_seq_stop(struct seq_file *seq, void *v) 2819 __releases(proto_list_mutex) 2820 { 2821 mutex_unlock(&proto_list_mutex); 2822 } 2823 2824 static char proto_method_implemented(const void *method) 2825 { 2826 return method == NULL ? 'n' : 'y'; 2827 } 2828 static long sock_prot_memory_allocated(struct proto *proto) 2829 { 2830 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L; 2831 } 2832 2833 static char *sock_prot_memory_pressure(struct proto *proto) 2834 { 2835 return proto->memory_pressure != NULL ? 2836 proto_memory_pressure(proto) ? "yes" : "no" : "NI"; 2837 } 2838 2839 static void proto_seq_printf(struct seq_file *seq, struct proto *proto) 2840 { 2841 2842 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s " 2843 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n", 2844 proto->name, 2845 proto->obj_size, 2846 sock_prot_inuse_get(seq_file_net(seq), proto), 2847 sock_prot_memory_allocated(proto), 2848 sock_prot_memory_pressure(proto), 2849 proto->max_header, 2850 proto->slab == NULL ? "no" : "yes", 2851 module_name(proto->owner), 2852 proto_method_implemented(proto->close), 2853 proto_method_implemented(proto->connect), 2854 proto_method_implemented(proto->disconnect), 2855 proto_method_implemented(proto->accept), 2856 proto_method_implemented(proto->ioctl), 2857 proto_method_implemented(proto->init), 2858 proto_method_implemented(proto->destroy), 2859 proto_method_implemented(proto->shutdown), 2860 proto_method_implemented(proto->setsockopt), 2861 proto_method_implemented(proto->getsockopt), 2862 proto_method_implemented(proto->sendmsg), 2863 proto_method_implemented(proto->recvmsg), 2864 proto_method_implemented(proto->sendpage), 2865 proto_method_implemented(proto->bind), 2866 proto_method_implemented(proto->backlog_rcv), 2867 proto_method_implemented(proto->hash), 2868 proto_method_implemented(proto->unhash), 2869 proto_method_implemented(proto->get_port), 2870 proto_method_implemented(proto->enter_memory_pressure)); 2871 } 2872 2873 static int proto_seq_show(struct seq_file *seq, void *v) 2874 { 2875 if (v == &proto_list) 2876 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s", 2877 "protocol", 2878 "size", 2879 "sockets", 2880 "memory", 2881 "press", 2882 "maxhdr", 2883 "slab", 2884 "module", 2885 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n"); 2886 else 2887 proto_seq_printf(seq, list_entry(v, struct proto, node)); 2888 return 0; 2889 } 2890 2891 static const struct seq_operations proto_seq_ops = { 2892 .start = proto_seq_start, 2893 .next = proto_seq_next, 2894 .stop = proto_seq_stop, 2895 .show = proto_seq_show, 2896 }; 2897 2898 static int proto_seq_open(struct inode *inode, struct file *file) 2899 { 2900 return seq_open_net(inode, file, &proto_seq_ops, 2901 sizeof(struct seq_net_private)); 2902 } 2903 2904 static const struct file_operations proto_seq_fops = { 2905 .owner = THIS_MODULE, 2906 .open = proto_seq_open, 2907 .read = seq_read, 2908 .llseek = seq_lseek, 2909 .release = seq_release_net, 2910 }; 2911 2912 static __net_init int proto_init_net(struct net *net) 2913 { 2914 if (!proc_create("protocols", S_IRUGO, net->proc_net, &proto_seq_fops)) 2915 return -ENOMEM; 2916 2917 return 0; 2918 } 2919 2920 static __net_exit void proto_exit_net(struct net *net) 2921 { 2922 remove_proc_entry("protocols", net->proc_net); 2923 } 2924 2925 2926 static __net_initdata struct pernet_operations proto_net_ops = { 2927 .init = proto_init_net, 2928 .exit = proto_exit_net, 2929 }; 2930 2931 static int __init proto_init(void) 2932 { 2933 return register_pernet_subsys(&proto_net_ops); 2934 } 2935 2936 subsys_initcall(proto_init); 2937 2938 #endif /* PROC_FS */ 2939