1 /* 2 * INET An implementation of the TCP/IP protocol suite for the LINUX 3 * operating system. INET is implemented using the BSD Socket 4 * interface as the means of communication with the user level. 5 * 6 * Generic socket support routines. Memory allocators, socket lock/release 7 * handler for protocols to use and generic option handler. 8 * 9 * 10 * Authors: Ross Biro 11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 12 * Florian La Roche, <flla@stud.uni-sb.de> 13 * Alan Cox, <A.Cox@swansea.ac.uk> 14 * 15 * Fixes: 16 * Alan Cox : Numerous verify_area() problems 17 * Alan Cox : Connecting on a connecting socket 18 * now returns an error for tcp. 19 * Alan Cox : sock->protocol is set correctly. 20 * and is not sometimes left as 0. 21 * Alan Cox : connect handles icmp errors on a 22 * connect properly. Unfortunately there 23 * is a restart syscall nasty there. I 24 * can't match BSD without hacking the C 25 * library. Ideas urgently sought! 26 * Alan Cox : Disallow bind() to addresses that are 27 * not ours - especially broadcast ones!! 28 * Alan Cox : Socket 1024 _IS_ ok for users. (fencepost) 29 * Alan Cox : sock_wfree/sock_rfree don't destroy sockets, 30 * instead they leave that for the DESTROY timer. 31 * Alan Cox : Clean up error flag in accept 32 * Alan Cox : TCP ack handling is buggy, the DESTROY timer 33 * was buggy. Put a remove_sock() in the handler 34 * for memory when we hit 0. Also altered the timer 35 * code. The ACK stuff can wait and needs major 36 * TCP layer surgery. 37 * Alan Cox : Fixed TCP ack bug, removed remove sock 38 * and fixed timer/inet_bh race. 39 * Alan Cox : Added zapped flag for TCP 40 * Alan Cox : Move kfree_skb into skbuff.c and tidied up surplus code 41 * Alan Cox : for new sk_buff allocations wmalloc/rmalloc now call alloc_skb 42 * Alan Cox : kfree_s calls now are kfree_skbmem so we can track skb resources 43 * Alan Cox : Supports socket option broadcast now as does udp. Packet and raw need fixing. 44 * Alan Cox : Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so... 45 * Rick Sladkey : Relaxed UDP rules for matching packets. 46 * C.E.Hawkins : IFF_PROMISC/SIOCGHWADDR support 47 * Pauline Middelink : identd support 48 * Alan Cox : Fixed connect() taking signals I think. 49 * Alan Cox : SO_LINGER supported 50 * Alan Cox : Error reporting fixes 51 * Anonymous : inet_create tidied up (sk->reuse setting) 52 * Alan Cox : inet sockets don't set sk->type! 53 * Alan Cox : Split socket option code 54 * Alan Cox : Callbacks 55 * Alan Cox : Nagle flag for Charles & Johannes stuff 56 * Alex : Removed restriction on inet fioctl 57 * Alan Cox : Splitting INET from NET core 58 * Alan Cox : Fixed bogus SO_TYPE handling in getsockopt() 59 * Adam Caldwell : Missing return in SO_DONTROUTE/SO_DEBUG code 60 * Alan Cox : Split IP from generic code 61 * Alan Cox : New kfree_skbmem() 62 * Alan Cox : Make SO_DEBUG superuser only. 63 * Alan Cox : Allow anyone to clear SO_DEBUG 64 * (compatibility fix) 65 * Alan Cox : Added optimistic memory grabbing for AF_UNIX throughput. 66 * Alan Cox : Allocator for a socket is settable. 67 * Alan Cox : SO_ERROR includes soft errors. 68 * Alan Cox : Allow NULL arguments on some SO_ opts 69 * Alan Cox : Generic socket allocation to make hooks 70 * easier (suggested by Craig Metz). 71 * Michael Pall : SO_ERROR returns positive errno again 72 * Steve Whitehouse: Added default destructor to free 73 * protocol private data. 74 * Steve Whitehouse: Added various other default routines 75 * common to several socket families. 76 * Chris Evans : Call suser() check last on F_SETOWN 77 * Jay Schulist : Added SO_ATTACH_FILTER and SO_DETACH_FILTER. 78 * Andi Kleen : Add sock_kmalloc()/sock_kfree_s() 79 * Andi Kleen : Fix write_space callback 80 * Chris Evans : Security fixes - signedness again 81 * Arnaldo C. Melo : cleanups, use skb_queue_purge 82 * 83 * To Fix: 84 * 85 * 86 * This program is free software; you can redistribute it and/or 87 * modify it under the terms of the GNU General Public License 88 * as published by the Free Software Foundation; either version 89 * 2 of the License, or (at your option) any later version. 90 */ 91 92 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 93 94 #include <linux/capability.h> 95 #include <linux/errno.h> 96 #include <linux/errqueue.h> 97 #include <linux/types.h> 98 #include <linux/socket.h> 99 #include <linux/in.h> 100 #include <linux/kernel.h> 101 #include <linux/module.h> 102 #include <linux/proc_fs.h> 103 #include <linux/seq_file.h> 104 #include <linux/sched.h> 105 #include <linux/timer.h> 106 #include <linux/string.h> 107 #include <linux/sockios.h> 108 #include <linux/net.h> 109 #include <linux/mm.h> 110 #include <linux/slab.h> 111 #include <linux/interrupt.h> 112 #include <linux/poll.h> 113 #include <linux/tcp.h> 114 #include <linux/init.h> 115 #include <linux/highmem.h> 116 #include <linux/user_namespace.h> 117 #include <linux/static_key.h> 118 #include <linux/memcontrol.h> 119 #include <linux/prefetch.h> 120 121 #include <asm/uaccess.h> 122 123 #include <linux/netdevice.h> 124 #include <net/protocol.h> 125 #include <linux/skbuff.h> 126 #include <net/net_namespace.h> 127 #include <net/request_sock.h> 128 #include <net/sock.h> 129 #include <linux/net_tstamp.h> 130 #include <net/xfrm.h> 131 #include <linux/ipsec.h> 132 #include <net/cls_cgroup.h> 133 #include <net/netprio_cgroup.h> 134 #include <linux/sock_diag.h> 135 136 #include <linux/filter.h> 137 #include <net/sock_reuseport.h> 138 139 #include <trace/events/sock.h> 140 141 #ifdef CONFIG_INET 142 #include <net/tcp.h> 143 #endif 144 145 #include <net/busy_poll.h> 146 147 static DEFINE_MUTEX(proto_list_mutex); 148 static LIST_HEAD(proto_list); 149 150 /** 151 * sk_ns_capable - General socket capability test 152 * @sk: Socket to use a capability on or through 153 * @user_ns: The user namespace of the capability to use 154 * @cap: The capability to use 155 * 156 * Test to see if the opener of the socket had when the socket was 157 * created and the current process has the capability @cap in the user 158 * namespace @user_ns. 159 */ 160 bool sk_ns_capable(const struct sock *sk, 161 struct user_namespace *user_ns, int cap) 162 { 163 return file_ns_capable(sk->sk_socket->file, user_ns, cap) && 164 ns_capable(user_ns, cap); 165 } 166 EXPORT_SYMBOL(sk_ns_capable); 167 168 /** 169 * sk_capable - Socket global capability test 170 * @sk: Socket to use a capability on or through 171 * @cap: The global capability to use 172 * 173 * Test to see if the opener of the socket had when the socket was 174 * created and the current process has the capability @cap in all user 175 * namespaces. 176 */ 177 bool sk_capable(const struct sock *sk, int cap) 178 { 179 return sk_ns_capable(sk, &init_user_ns, cap); 180 } 181 EXPORT_SYMBOL(sk_capable); 182 183 /** 184 * sk_net_capable - Network namespace socket capability test 185 * @sk: Socket to use a capability on or through 186 * @cap: The capability to use 187 * 188 * Test to see if the opener of the socket had when the socket was created 189 * and the current process has the capability @cap over the network namespace 190 * the socket is a member of. 191 */ 192 bool sk_net_capable(const struct sock *sk, int cap) 193 { 194 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap); 195 } 196 EXPORT_SYMBOL(sk_net_capable); 197 198 /* 199 * Each address family might have different locking rules, so we have 200 * one slock key per address family: 201 */ 202 static struct lock_class_key af_family_keys[AF_MAX]; 203 static struct lock_class_key af_family_slock_keys[AF_MAX]; 204 205 /* 206 * Make lock validator output more readable. (we pre-construct these 207 * strings build-time, so that runtime initialization of socket 208 * locks is fast): 209 */ 210 static const char *const af_family_key_strings[AF_MAX+1] = { 211 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" , 212 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK", 213 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" , 214 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" , 215 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" , 216 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" , 217 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" , 218 "sk_lock-AF_RDS" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" , 219 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" , 220 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" , 221 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" , 222 "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN" , "sk_lock-AF_PHONET" , 223 "sk_lock-AF_IEEE802154", "sk_lock-AF_CAIF" , "sk_lock-AF_ALG" , 224 "sk_lock-AF_NFC" , "sk_lock-AF_VSOCK" , "sk_lock-AF_KCM" , 225 "sk_lock-AF_MAX" 226 }; 227 static const char *const af_family_slock_key_strings[AF_MAX+1] = { 228 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" , 229 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK", 230 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" , 231 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" , 232 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" , 233 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" , 234 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" , 235 "slock-AF_RDS" , "slock-AF_SNA" , "slock-AF_IRDA" , 236 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" , 237 "slock-27" , "slock-28" , "slock-AF_CAN" , 238 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" , 239 "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" , 240 "slock-AF_IEEE802154", "slock-AF_CAIF" , "slock-AF_ALG" , 241 "slock-AF_NFC" , "slock-AF_VSOCK" ,"slock-AF_KCM" , 242 "slock-AF_MAX" 243 }; 244 static const char *const af_family_clock_key_strings[AF_MAX+1] = { 245 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" , 246 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK", 247 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" , 248 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" , 249 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" , 250 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" , 251 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" , 252 "clock-AF_RDS" , "clock-AF_SNA" , "clock-AF_IRDA" , 253 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" , 254 "clock-27" , "clock-28" , "clock-AF_CAN" , 255 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" , 256 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" , 257 "clock-AF_IEEE802154", "clock-AF_CAIF" , "clock-AF_ALG" , 258 "clock-AF_NFC" , "clock-AF_VSOCK" , "clock-AF_KCM" , 259 "clock-AF_MAX" 260 }; 261 262 /* 263 * sk_callback_lock locking rules are per-address-family, 264 * so split the lock classes by using a per-AF key: 265 */ 266 static struct lock_class_key af_callback_keys[AF_MAX]; 267 268 /* Take into consideration the size of the struct sk_buff overhead in the 269 * determination of these values, since that is non-constant across 270 * platforms. This makes socket queueing behavior and performance 271 * not depend upon such differences. 272 */ 273 #define _SK_MEM_PACKETS 256 274 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256) 275 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 276 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 277 278 /* Run time adjustable parameters. */ 279 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX; 280 EXPORT_SYMBOL(sysctl_wmem_max); 281 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX; 282 EXPORT_SYMBOL(sysctl_rmem_max); 283 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX; 284 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX; 285 286 /* Maximal space eaten by iovec or ancillary data plus some space */ 287 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512); 288 EXPORT_SYMBOL(sysctl_optmem_max); 289 290 int sysctl_tstamp_allow_data __read_mostly = 1; 291 292 struct static_key memalloc_socks = STATIC_KEY_INIT_FALSE; 293 EXPORT_SYMBOL_GPL(memalloc_socks); 294 295 /** 296 * sk_set_memalloc - sets %SOCK_MEMALLOC 297 * @sk: socket to set it on 298 * 299 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves. 300 * It's the responsibility of the admin to adjust min_free_kbytes 301 * to meet the requirements 302 */ 303 void sk_set_memalloc(struct sock *sk) 304 { 305 sock_set_flag(sk, SOCK_MEMALLOC); 306 sk->sk_allocation |= __GFP_MEMALLOC; 307 static_key_slow_inc(&memalloc_socks); 308 } 309 EXPORT_SYMBOL_GPL(sk_set_memalloc); 310 311 void sk_clear_memalloc(struct sock *sk) 312 { 313 sock_reset_flag(sk, SOCK_MEMALLOC); 314 sk->sk_allocation &= ~__GFP_MEMALLOC; 315 static_key_slow_dec(&memalloc_socks); 316 317 /* 318 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward 319 * progress of swapping. SOCK_MEMALLOC may be cleared while 320 * it has rmem allocations due to the last swapfile being deactivated 321 * but there is a risk that the socket is unusable due to exceeding 322 * the rmem limits. Reclaim the reserves and obey rmem limits again. 323 */ 324 sk_mem_reclaim(sk); 325 } 326 EXPORT_SYMBOL_GPL(sk_clear_memalloc); 327 328 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb) 329 { 330 int ret; 331 unsigned long pflags = current->flags; 332 333 /* these should have been dropped before queueing */ 334 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC)); 335 336 current->flags |= PF_MEMALLOC; 337 ret = sk->sk_backlog_rcv(sk, skb); 338 tsk_restore_flags(current, pflags, PF_MEMALLOC); 339 340 return ret; 341 } 342 EXPORT_SYMBOL(__sk_backlog_rcv); 343 344 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen) 345 { 346 struct timeval tv; 347 348 if (optlen < sizeof(tv)) 349 return -EINVAL; 350 if (copy_from_user(&tv, optval, sizeof(tv))) 351 return -EFAULT; 352 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC) 353 return -EDOM; 354 355 if (tv.tv_sec < 0) { 356 static int warned __read_mostly; 357 358 *timeo_p = 0; 359 if (warned < 10 && net_ratelimit()) { 360 warned++; 361 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n", 362 __func__, current->comm, task_pid_nr(current)); 363 } 364 return 0; 365 } 366 *timeo_p = MAX_SCHEDULE_TIMEOUT; 367 if (tv.tv_sec == 0 && tv.tv_usec == 0) 368 return 0; 369 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1)) 370 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ); 371 return 0; 372 } 373 374 static void sock_warn_obsolete_bsdism(const char *name) 375 { 376 static int warned; 377 static char warncomm[TASK_COMM_LEN]; 378 if (strcmp(warncomm, current->comm) && warned < 5) { 379 strcpy(warncomm, current->comm); 380 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n", 381 warncomm, name); 382 warned++; 383 } 384 } 385 386 static bool sock_needs_netstamp(const struct sock *sk) 387 { 388 switch (sk->sk_family) { 389 case AF_UNSPEC: 390 case AF_UNIX: 391 return false; 392 default: 393 return true; 394 } 395 } 396 397 static void sock_disable_timestamp(struct sock *sk, unsigned long flags) 398 { 399 if (sk->sk_flags & flags) { 400 sk->sk_flags &= ~flags; 401 if (sock_needs_netstamp(sk) && 402 !(sk->sk_flags & SK_FLAGS_TIMESTAMP)) 403 net_disable_timestamp(); 404 } 405 } 406 407 408 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 409 { 410 unsigned long flags; 411 struct sk_buff_head *list = &sk->sk_receive_queue; 412 413 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) { 414 atomic_inc(&sk->sk_drops); 415 trace_sock_rcvqueue_full(sk, skb); 416 return -ENOMEM; 417 } 418 419 if (!sk_rmem_schedule(sk, skb, skb->truesize)) { 420 atomic_inc(&sk->sk_drops); 421 return -ENOBUFS; 422 } 423 424 skb->dev = NULL; 425 skb_set_owner_r(skb, sk); 426 427 /* we escape from rcu protected region, make sure we dont leak 428 * a norefcounted dst 429 */ 430 skb_dst_force(skb); 431 432 spin_lock_irqsave(&list->lock, flags); 433 sock_skb_set_dropcount(sk, skb); 434 __skb_queue_tail(list, skb); 435 spin_unlock_irqrestore(&list->lock, flags); 436 437 if (!sock_flag(sk, SOCK_DEAD)) 438 sk->sk_data_ready(sk); 439 return 0; 440 } 441 EXPORT_SYMBOL(__sock_queue_rcv_skb); 442 443 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 444 { 445 int err; 446 447 err = sk_filter(sk, skb); 448 if (err) 449 return err; 450 451 return __sock_queue_rcv_skb(sk, skb); 452 } 453 EXPORT_SYMBOL(sock_queue_rcv_skb); 454 455 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, 456 const int nested, unsigned int trim_cap, bool refcounted) 457 { 458 int rc = NET_RX_SUCCESS; 459 460 if (sk_filter_trim_cap(sk, skb, trim_cap)) 461 goto discard_and_relse; 462 463 skb->dev = NULL; 464 465 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) { 466 atomic_inc(&sk->sk_drops); 467 goto discard_and_relse; 468 } 469 if (nested) 470 bh_lock_sock_nested(sk); 471 else 472 bh_lock_sock(sk); 473 if (!sock_owned_by_user(sk)) { 474 /* 475 * trylock + unlock semantics: 476 */ 477 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_); 478 479 rc = sk_backlog_rcv(sk, skb); 480 481 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_); 482 } else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) { 483 bh_unlock_sock(sk); 484 atomic_inc(&sk->sk_drops); 485 goto discard_and_relse; 486 } 487 488 bh_unlock_sock(sk); 489 out: 490 if (refcounted) 491 sock_put(sk); 492 return rc; 493 discard_and_relse: 494 kfree_skb(skb); 495 goto out; 496 } 497 EXPORT_SYMBOL(__sk_receive_skb); 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_tx_queue_clear(sk); 505 RCU_INIT_POINTER(sk->sk_dst_cache, NULL); 506 dst_release(dst); 507 return NULL; 508 } 509 510 return dst; 511 } 512 EXPORT_SYMBOL(__sk_dst_check); 513 514 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie) 515 { 516 struct dst_entry *dst = sk_dst_get(sk); 517 518 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) { 519 sk_dst_reset(sk); 520 dst_release(dst); 521 return NULL; 522 } 523 524 return dst; 525 } 526 EXPORT_SYMBOL(sk_dst_check); 527 528 static int sock_setbindtodevice(struct sock *sk, char __user *optval, 529 int optlen) 530 { 531 int ret = -ENOPROTOOPT; 532 #ifdef CONFIG_NETDEVICES 533 struct net *net = sock_net(sk); 534 char devname[IFNAMSIZ]; 535 int index; 536 537 /* Sorry... */ 538 ret = -EPERM; 539 if (!ns_capable(net->user_ns, CAP_NET_RAW)) 540 goto out; 541 542 ret = -EINVAL; 543 if (optlen < 0) 544 goto out; 545 546 /* Bind this socket to a particular device like "eth0", 547 * as specified in the passed interface name. If the 548 * name is "" or the option length is zero the socket 549 * is not bound. 550 */ 551 if (optlen > IFNAMSIZ - 1) 552 optlen = IFNAMSIZ - 1; 553 memset(devname, 0, sizeof(devname)); 554 555 ret = -EFAULT; 556 if (copy_from_user(devname, optval, optlen)) 557 goto out; 558 559 index = 0; 560 if (devname[0] != '\0') { 561 struct net_device *dev; 562 563 rcu_read_lock(); 564 dev = dev_get_by_name_rcu(net, devname); 565 if (dev) 566 index = dev->ifindex; 567 rcu_read_unlock(); 568 ret = -ENODEV; 569 if (!dev) 570 goto out; 571 } 572 573 lock_sock(sk); 574 sk->sk_bound_dev_if = index; 575 sk_dst_reset(sk); 576 release_sock(sk); 577 578 ret = 0; 579 580 out: 581 #endif 582 583 return ret; 584 } 585 586 static int sock_getbindtodevice(struct sock *sk, char __user *optval, 587 int __user *optlen, int len) 588 { 589 int ret = -ENOPROTOOPT; 590 #ifdef CONFIG_NETDEVICES 591 struct net *net = sock_net(sk); 592 char devname[IFNAMSIZ]; 593 594 if (sk->sk_bound_dev_if == 0) { 595 len = 0; 596 goto zero; 597 } 598 599 ret = -EINVAL; 600 if (len < IFNAMSIZ) 601 goto out; 602 603 ret = netdev_get_name(net, devname, sk->sk_bound_dev_if); 604 if (ret) 605 goto out; 606 607 len = strlen(devname) + 1; 608 609 ret = -EFAULT; 610 if (copy_to_user(optval, devname, len)) 611 goto out; 612 613 zero: 614 ret = -EFAULT; 615 if (put_user(len, optlen)) 616 goto out; 617 618 ret = 0; 619 620 out: 621 #endif 622 623 return ret; 624 } 625 626 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool) 627 { 628 if (valbool) 629 sock_set_flag(sk, bit); 630 else 631 sock_reset_flag(sk, bit); 632 } 633 634 bool sk_mc_loop(struct sock *sk) 635 { 636 if (dev_recursion_level()) 637 return false; 638 if (!sk) 639 return true; 640 switch (sk->sk_family) { 641 case AF_INET: 642 return inet_sk(sk)->mc_loop; 643 #if IS_ENABLED(CONFIG_IPV6) 644 case AF_INET6: 645 return inet6_sk(sk)->mc_loop; 646 #endif 647 } 648 WARN_ON(1); 649 return true; 650 } 651 EXPORT_SYMBOL(sk_mc_loop); 652 653 /* 654 * This is meant for all protocols to use and covers goings on 655 * at the socket level. Everything here is generic. 656 */ 657 658 int sock_setsockopt(struct socket *sock, int level, int optname, 659 char __user *optval, unsigned int optlen) 660 { 661 struct sock *sk = sock->sk; 662 int val; 663 int valbool; 664 struct linger ling; 665 int ret = 0; 666 667 /* 668 * Options without arguments 669 */ 670 671 if (optname == SO_BINDTODEVICE) 672 return sock_setbindtodevice(sk, optval, optlen); 673 674 if (optlen < sizeof(int)) 675 return -EINVAL; 676 677 if (get_user(val, (int __user *)optval)) 678 return -EFAULT; 679 680 valbool = val ? 1 : 0; 681 682 lock_sock(sk); 683 684 switch (optname) { 685 case SO_DEBUG: 686 if (val && !capable(CAP_NET_ADMIN)) 687 ret = -EACCES; 688 else 689 sock_valbool_flag(sk, SOCK_DBG, valbool); 690 break; 691 case SO_REUSEADDR: 692 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE); 693 break; 694 case SO_REUSEPORT: 695 sk->sk_reuseport = valbool; 696 break; 697 case SO_TYPE: 698 case SO_PROTOCOL: 699 case SO_DOMAIN: 700 case SO_ERROR: 701 ret = -ENOPROTOOPT; 702 break; 703 case SO_DONTROUTE: 704 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool); 705 break; 706 case SO_BROADCAST: 707 sock_valbool_flag(sk, SOCK_BROADCAST, valbool); 708 break; 709 case SO_SNDBUF: 710 /* Don't error on this BSD doesn't and if you think 711 * about it this is right. Otherwise apps have to 712 * play 'guess the biggest size' games. RCVBUF/SNDBUF 713 * are treated in BSD as hints 714 */ 715 val = min_t(u32, val, sysctl_wmem_max); 716 set_sndbuf: 717 sk->sk_userlocks |= SOCK_SNDBUF_LOCK; 718 sk->sk_sndbuf = max_t(int, val * 2, SOCK_MIN_SNDBUF); 719 /* Wake up sending tasks if we upped the value. */ 720 sk->sk_write_space(sk); 721 break; 722 723 case SO_SNDBUFFORCE: 724 if (!capable(CAP_NET_ADMIN)) { 725 ret = -EPERM; 726 break; 727 } 728 goto set_sndbuf; 729 730 case SO_RCVBUF: 731 /* Don't error on this BSD doesn't and if you think 732 * about it this is right. Otherwise apps have to 733 * play 'guess the biggest size' games. RCVBUF/SNDBUF 734 * are treated in BSD as hints 735 */ 736 val = min_t(u32, val, sysctl_rmem_max); 737 set_rcvbuf: 738 sk->sk_userlocks |= SOCK_RCVBUF_LOCK; 739 /* 740 * We double it on the way in to account for 741 * "struct sk_buff" etc. overhead. Applications 742 * assume that the SO_RCVBUF setting they make will 743 * allow that much actual data to be received on that 744 * socket. 745 * 746 * Applications are unaware that "struct sk_buff" and 747 * other overheads allocate from the receive buffer 748 * during socket buffer allocation. 749 * 750 * And after considering the possible alternatives, 751 * returning the value we actually used in getsockopt 752 * is the most desirable behavior. 753 */ 754 sk->sk_rcvbuf = max_t(int, val * 2, SOCK_MIN_RCVBUF); 755 break; 756 757 case SO_RCVBUFFORCE: 758 if (!capable(CAP_NET_ADMIN)) { 759 ret = -EPERM; 760 break; 761 } 762 goto set_rcvbuf; 763 764 case SO_KEEPALIVE: 765 #ifdef CONFIG_INET 766 if (sk->sk_protocol == IPPROTO_TCP && 767 sk->sk_type == SOCK_STREAM) 768 tcp_set_keepalive(sk, valbool); 769 #endif 770 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool); 771 break; 772 773 case SO_OOBINLINE: 774 sock_valbool_flag(sk, SOCK_URGINLINE, valbool); 775 break; 776 777 case SO_NO_CHECK: 778 sk->sk_no_check_tx = valbool; 779 break; 780 781 case SO_PRIORITY: 782 if ((val >= 0 && val <= 6) || 783 ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) 784 sk->sk_priority = val; 785 else 786 ret = -EPERM; 787 break; 788 789 case SO_LINGER: 790 if (optlen < sizeof(ling)) { 791 ret = -EINVAL; /* 1003.1g */ 792 break; 793 } 794 if (copy_from_user(&ling, optval, sizeof(ling))) { 795 ret = -EFAULT; 796 break; 797 } 798 if (!ling.l_onoff) 799 sock_reset_flag(sk, SOCK_LINGER); 800 else { 801 #if (BITS_PER_LONG == 32) 802 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ) 803 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT; 804 else 805 #endif 806 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ; 807 sock_set_flag(sk, SOCK_LINGER); 808 } 809 break; 810 811 case SO_BSDCOMPAT: 812 sock_warn_obsolete_bsdism("setsockopt"); 813 break; 814 815 case SO_PASSCRED: 816 if (valbool) 817 set_bit(SOCK_PASSCRED, &sock->flags); 818 else 819 clear_bit(SOCK_PASSCRED, &sock->flags); 820 break; 821 822 case SO_TIMESTAMP: 823 case SO_TIMESTAMPNS: 824 if (valbool) { 825 if (optname == SO_TIMESTAMP) 826 sock_reset_flag(sk, SOCK_RCVTSTAMPNS); 827 else 828 sock_set_flag(sk, SOCK_RCVTSTAMPNS); 829 sock_set_flag(sk, SOCK_RCVTSTAMP); 830 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 831 } else { 832 sock_reset_flag(sk, SOCK_RCVTSTAMP); 833 sock_reset_flag(sk, SOCK_RCVTSTAMPNS); 834 } 835 break; 836 837 case SO_TIMESTAMPING: 838 if (val & ~SOF_TIMESTAMPING_MASK) { 839 ret = -EINVAL; 840 break; 841 } 842 843 if (val & SOF_TIMESTAMPING_OPT_ID && 844 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) { 845 if (sk->sk_protocol == IPPROTO_TCP && 846 sk->sk_type == SOCK_STREAM) { 847 if ((1 << sk->sk_state) & 848 (TCPF_CLOSE | TCPF_LISTEN)) { 849 ret = -EINVAL; 850 break; 851 } 852 sk->sk_tskey = tcp_sk(sk)->snd_una; 853 } else { 854 sk->sk_tskey = 0; 855 } 856 } 857 sk->sk_tsflags = val; 858 if (val & SOF_TIMESTAMPING_RX_SOFTWARE) 859 sock_enable_timestamp(sk, 860 SOCK_TIMESTAMPING_RX_SOFTWARE); 861 else 862 sock_disable_timestamp(sk, 863 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE)); 864 break; 865 866 case SO_RCVLOWAT: 867 if (val < 0) 868 val = INT_MAX; 869 sk->sk_rcvlowat = val ? : 1; 870 break; 871 872 case SO_RCVTIMEO: 873 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen); 874 break; 875 876 case SO_SNDTIMEO: 877 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen); 878 break; 879 880 case SO_ATTACH_FILTER: 881 ret = -EINVAL; 882 if (optlen == sizeof(struct sock_fprog)) { 883 struct sock_fprog fprog; 884 885 ret = -EFAULT; 886 if (copy_from_user(&fprog, optval, sizeof(fprog))) 887 break; 888 889 ret = sk_attach_filter(&fprog, sk); 890 } 891 break; 892 893 case SO_ATTACH_BPF: 894 ret = -EINVAL; 895 if (optlen == sizeof(u32)) { 896 u32 ufd; 897 898 ret = -EFAULT; 899 if (copy_from_user(&ufd, optval, sizeof(ufd))) 900 break; 901 902 ret = sk_attach_bpf(ufd, sk); 903 } 904 break; 905 906 case SO_ATTACH_REUSEPORT_CBPF: 907 ret = -EINVAL; 908 if (optlen == sizeof(struct sock_fprog)) { 909 struct sock_fprog fprog; 910 911 ret = -EFAULT; 912 if (copy_from_user(&fprog, optval, sizeof(fprog))) 913 break; 914 915 ret = sk_reuseport_attach_filter(&fprog, sk); 916 } 917 break; 918 919 case SO_ATTACH_REUSEPORT_EBPF: 920 ret = -EINVAL; 921 if (optlen == sizeof(u32)) { 922 u32 ufd; 923 924 ret = -EFAULT; 925 if (copy_from_user(&ufd, optval, sizeof(ufd))) 926 break; 927 928 ret = sk_reuseport_attach_bpf(ufd, sk); 929 } 930 break; 931 932 case SO_DETACH_FILTER: 933 ret = sk_detach_filter(sk); 934 break; 935 936 case SO_LOCK_FILTER: 937 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool) 938 ret = -EPERM; 939 else 940 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool); 941 break; 942 943 case SO_PASSSEC: 944 if (valbool) 945 set_bit(SOCK_PASSSEC, &sock->flags); 946 else 947 clear_bit(SOCK_PASSSEC, &sock->flags); 948 break; 949 case SO_MARK: 950 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) 951 ret = -EPERM; 952 else 953 sk->sk_mark = val; 954 break; 955 956 case SO_RXQ_OVFL: 957 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool); 958 break; 959 960 case SO_WIFI_STATUS: 961 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool); 962 break; 963 964 case SO_PEEK_OFF: 965 if (sock->ops->set_peek_off) 966 ret = sock->ops->set_peek_off(sk, val); 967 else 968 ret = -EOPNOTSUPP; 969 break; 970 971 case SO_NOFCS: 972 sock_valbool_flag(sk, SOCK_NOFCS, valbool); 973 break; 974 975 case SO_SELECT_ERR_QUEUE: 976 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool); 977 break; 978 979 #ifdef CONFIG_NET_RX_BUSY_POLL 980 case SO_BUSY_POLL: 981 /* allow unprivileged users to decrease the value */ 982 if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN)) 983 ret = -EPERM; 984 else { 985 if (val < 0) 986 ret = -EINVAL; 987 else 988 sk->sk_ll_usec = val; 989 } 990 break; 991 #endif 992 993 case SO_MAX_PACING_RATE: 994 sk->sk_max_pacing_rate = val; 995 sk->sk_pacing_rate = min(sk->sk_pacing_rate, 996 sk->sk_max_pacing_rate); 997 break; 998 999 case SO_INCOMING_CPU: 1000 sk->sk_incoming_cpu = val; 1001 break; 1002 1003 case SO_CNX_ADVICE: 1004 if (val == 1) 1005 dst_negative_advice(sk); 1006 break; 1007 default: 1008 ret = -ENOPROTOOPT; 1009 break; 1010 } 1011 release_sock(sk); 1012 return ret; 1013 } 1014 EXPORT_SYMBOL(sock_setsockopt); 1015 1016 1017 static void cred_to_ucred(struct pid *pid, const struct cred *cred, 1018 struct ucred *ucred) 1019 { 1020 ucred->pid = pid_vnr(pid); 1021 ucred->uid = ucred->gid = -1; 1022 if (cred) { 1023 struct user_namespace *current_ns = current_user_ns(); 1024 1025 ucred->uid = from_kuid_munged(current_ns, cred->euid); 1026 ucred->gid = from_kgid_munged(current_ns, cred->egid); 1027 } 1028 } 1029 1030 int sock_getsockopt(struct socket *sock, int level, int optname, 1031 char __user *optval, int __user *optlen) 1032 { 1033 struct sock *sk = sock->sk; 1034 1035 union { 1036 int val; 1037 struct linger ling; 1038 struct timeval tm; 1039 } v; 1040 1041 int lv = sizeof(int); 1042 int len; 1043 1044 if (get_user(len, optlen)) 1045 return -EFAULT; 1046 if (len < 0) 1047 return -EINVAL; 1048 1049 memset(&v, 0, sizeof(v)); 1050 1051 switch (optname) { 1052 case SO_DEBUG: 1053 v.val = sock_flag(sk, SOCK_DBG); 1054 break; 1055 1056 case SO_DONTROUTE: 1057 v.val = sock_flag(sk, SOCK_LOCALROUTE); 1058 break; 1059 1060 case SO_BROADCAST: 1061 v.val = sock_flag(sk, SOCK_BROADCAST); 1062 break; 1063 1064 case SO_SNDBUF: 1065 v.val = sk->sk_sndbuf; 1066 break; 1067 1068 case SO_RCVBUF: 1069 v.val = sk->sk_rcvbuf; 1070 break; 1071 1072 case SO_REUSEADDR: 1073 v.val = sk->sk_reuse; 1074 break; 1075 1076 case SO_REUSEPORT: 1077 v.val = sk->sk_reuseport; 1078 break; 1079 1080 case SO_KEEPALIVE: 1081 v.val = sock_flag(sk, SOCK_KEEPOPEN); 1082 break; 1083 1084 case SO_TYPE: 1085 v.val = sk->sk_type; 1086 break; 1087 1088 case SO_PROTOCOL: 1089 v.val = sk->sk_protocol; 1090 break; 1091 1092 case SO_DOMAIN: 1093 v.val = sk->sk_family; 1094 break; 1095 1096 case SO_ERROR: 1097 v.val = -sock_error(sk); 1098 if (v.val == 0) 1099 v.val = xchg(&sk->sk_err_soft, 0); 1100 break; 1101 1102 case SO_OOBINLINE: 1103 v.val = sock_flag(sk, SOCK_URGINLINE); 1104 break; 1105 1106 case SO_NO_CHECK: 1107 v.val = sk->sk_no_check_tx; 1108 break; 1109 1110 case SO_PRIORITY: 1111 v.val = sk->sk_priority; 1112 break; 1113 1114 case SO_LINGER: 1115 lv = sizeof(v.ling); 1116 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER); 1117 v.ling.l_linger = sk->sk_lingertime / HZ; 1118 break; 1119 1120 case SO_BSDCOMPAT: 1121 sock_warn_obsolete_bsdism("getsockopt"); 1122 break; 1123 1124 case SO_TIMESTAMP: 1125 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && 1126 !sock_flag(sk, SOCK_RCVTSTAMPNS); 1127 break; 1128 1129 case SO_TIMESTAMPNS: 1130 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS); 1131 break; 1132 1133 case SO_TIMESTAMPING: 1134 v.val = sk->sk_tsflags; 1135 break; 1136 1137 case SO_RCVTIMEO: 1138 lv = sizeof(struct timeval); 1139 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) { 1140 v.tm.tv_sec = 0; 1141 v.tm.tv_usec = 0; 1142 } else { 1143 v.tm.tv_sec = sk->sk_rcvtimeo / HZ; 1144 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ; 1145 } 1146 break; 1147 1148 case SO_SNDTIMEO: 1149 lv = sizeof(struct timeval); 1150 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) { 1151 v.tm.tv_sec = 0; 1152 v.tm.tv_usec = 0; 1153 } else { 1154 v.tm.tv_sec = sk->sk_sndtimeo / HZ; 1155 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ; 1156 } 1157 break; 1158 1159 case SO_RCVLOWAT: 1160 v.val = sk->sk_rcvlowat; 1161 break; 1162 1163 case SO_SNDLOWAT: 1164 v.val = 1; 1165 break; 1166 1167 case SO_PASSCRED: 1168 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags); 1169 break; 1170 1171 case SO_PEERCRED: 1172 { 1173 struct ucred peercred; 1174 if (len > sizeof(peercred)) 1175 len = sizeof(peercred); 1176 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred); 1177 if (copy_to_user(optval, &peercred, len)) 1178 return -EFAULT; 1179 goto lenout; 1180 } 1181 1182 case SO_PEERNAME: 1183 { 1184 char address[128]; 1185 1186 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2)) 1187 return -ENOTCONN; 1188 if (lv < len) 1189 return -EINVAL; 1190 if (copy_to_user(optval, address, len)) 1191 return -EFAULT; 1192 goto lenout; 1193 } 1194 1195 /* Dubious BSD thing... Probably nobody even uses it, but 1196 * the UNIX standard wants it for whatever reason... -DaveM 1197 */ 1198 case SO_ACCEPTCONN: 1199 v.val = sk->sk_state == TCP_LISTEN; 1200 break; 1201 1202 case SO_PASSSEC: 1203 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags); 1204 break; 1205 1206 case SO_PEERSEC: 1207 return security_socket_getpeersec_stream(sock, optval, optlen, len); 1208 1209 case SO_MARK: 1210 v.val = sk->sk_mark; 1211 break; 1212 1213 case SO_RXQ_OVFL: 1214 v.val = sock_flag(sk, SOCK_RXQ_OVFL); 1215 break; 1216 1217 case SO_WIFI_STATUS: 1218 v.val = sock_flag(sk, SOCK_WIFI_STATUS); 1219 break; 1220 1221 case SO_PEEK_OFF: 1222 if (!sock->ops->set_peek_off) 1223 return -EOPNOTSUPP; 1224 1225 v.val = sk->sk_peek_off; 1226 break; 1227 case SO_NOFCS: 1228 v.val = sock_flag(sk, SOCK_NOFCS); 1229 break; 1230 1231 case SO_BINDTODEVICE: 1232 return sock_getbindtodevice(sk, optval, optlen, len); 1233 1234 case SO_GET_FILTER: 1235 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len); 1236 if (len < 0) 1237 return len; 1238 1239 goto lenout; 1240 1241 case SO_LOCK_FILTER: 1242 v.val = sock_flag(sk, SOCK_FILTER_LOCKED); 1243 break; 1244 1245 case SO_BPF_EXTENSIONS: 1246 v.val = bpf_tell_extensions(); 1247 break; 1248 1249 case SO_SELECT_ERR_QUEUE: 1250 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE); 1251 break; 1252 1253 #ifdef CONFIG_NET_RX_BUSY_POLL 1254 case SO_BUSY_POLL: 1255 v.val = sk->sk_ll_usec; 1256 break; 1257 #endif 1258 1259 case SO_MAX_PACING_RATE: 1260 v.val = sk->sk_max_pacing_rate; 1261 break; 1262 1263 case SO_INCOMING_CPU: 1264 v.val = sk->sk_incoming_cpu; 1265 break; 1266 1267 default: 1268 /* We implement the SO_SNDLOWAT etc to not be settable 1269 * (1003.1g 7). 1270 */ 1271 return -ENOPROTOOPT; 1272 } 1273 1274 if (len > lv) 1275 len = lv; 1276 if (copy_to_user(optval, &v, len)) 1277 return -EFAULT; 1278 lenout: 1279 if (put_user(len, optlen)) 1280 return -EFAULT; 1281 return 0; 1282 } 1283 1284 /* 1285 * Initialize an sk_lock. 1286 * 1287 * (We also register the sk_lock with the lock validator.) 1288 */ 1289 static inline void sock_lock_init(struct sock *sk) 1290 { 1291 sock_lock_init_class_and_name(sk, 1292 af_family_slock_key_strings[sk->sk_family], 1293 af_family_slock_keys + sk->sk_family, 1294 af_family_key_strings[sk->sk_family], 1295 af_family_keys + sk->sk_family); 1296 } 1297 1298 /* 1299 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet, 1300 * even temporarly, because of RCU lookups. sk_node should also be left as is. 1301 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end 1302 */ 1303 static void sock_copy(struct sock *nsk, const struct sock *osk) 1304 { 1305 #ifdef CONFIG_SECURITY_NETWORK 1306 void *sptr = nsk->sk_security; 1307 #endif 1308 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin)); 1309 1310 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end, 1311 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end)); 1312 1313 #ifdef CONFIG_SECURITY_NETWORK 1314 nsk->sk_security = sptr; 1315 security_sk_clone(osk, nsk); 1316 #endif 1317 } 1318 1319 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority, 1320 int family) 1321 { 1322 struct sock *sk; 1323 struct kmem_cache *slab; 1324 1325 slab = prot->slab; 1326 if (slab != NULL) { 1327 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO); 1328 if (!sk) 1329 return sk; 1330 if (priority & __GFP_ZERO) 1331 sk_prot_clear_nulls(sk, prot->obj_size); 1332 } else 1333 sk = kmalloc(prot->obj_size, priority); 1334 1335 if (sk != NULL) { 1336 kmemcheck_annotate_bitfield(sk, flags); 1337 1338 if (security_sk_alloc(sk, family, priority)) 1339 goto out_free; 1340 1341 if (!try_module_get(prot->owner)) 1342 goto out_free_sec; 1343 sk_tx_queue_clear(sk); 1344 } 1345 1346 return sk; 1347 1348 out_free_sec: 1349 security_sk_free(sk); 1350 out_free: 1351 if (slab != NULL) 1352 kmem_cache_free(slab, sk); 1353 else 1354 kfree(sk); 1355 return NULL; 1356 } 1357 1358 static void sk_prot_free(struct proto *prot, struct sock *sk) 1359 { 1360 struct kmem_cache *slab; 1361 struct module *owner; 1362 1363 owner = prot->owner; 1364 slab = prot->slab; 1365 1366 cgroup_sk_free(&sk->sk_cgrp_data); 1367 mem_cgroup_sk_free(sk); 1368 security_sk_free(sk); 1369 if (slab != NULL) 1370 kmem_cache_free(slab, sk); 1371 else 1372 kfree(sk); 1373 module_put(owner); 1374 } 1375 1376 /** 1377 * sk_alloc - All socket objects are allocated here 1378 * @net: the applicable net namespace 1379 * @family: protocol family 1380 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 1381 * @prot: struct proto associated with this new sock instance 1382 * @kern: is this to be a kernel socket? 1383 */ 1384 struct sock *sk_alloc(struct net *net, int family, gfp_t priority, 1385 struct proto *prot, int kern) 1386 { 1387 struct sock *sk; 1388 1389 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family); 1390 if (sk) { 1391 sk->sk_family = family; 1392 /* 1393 * See comment in struct sock definition to understand 1394 * why we need sk_prot_creator -acme 1395 */ 1396 sk->sk_prot = sk->sk_prot_creator = prot; 1397 sock_lock_init(sk); 1398 sk->sk_net_refcnt = kern ? 0 : 1; 1399 if (likely(sk->sk_net_refcnt)) 1400 get_net(net); 1401 sock_net_set(sk, net); 1402 atomic_set(&sk->sk_wmem_alloc, 1); 1403 1404 mem_cgroup_sk_alloc(sk); 1405 cgroup_sk_alloc(&sk->sk_cgrp_data); 1406 sock_update_classid(&sk->sk_cgrp_data); 1407 sock_update_netprioidx(&sk->sk_cgrp_data); 1408 } 1409 1410 return sk; 1411 } 1412 EXPORT_SYMBOL(sk_alloc); 1413 1414 /* Sockets having SOCK_RCU_FREE will call this function after one RCU 1415 * grace period. This is the case for UDP sockets and TCP listeners. 1416 */ 1417 static void __sk_destruct(struct rcu_head *head) 1418 { 1419 struct sock *sk = container_of(head, struct sock, sk_rcu); 1420 struct sk_filter *filter; 1421 1422 if (sk->sk_destruct) 1423 sk->sk_destruct(sk); 1424 1425 filter = rcu_dereference_check(sk->sk_filter, 1426 atomic_read(&sk->sk_wmem_alloc) == 0); 1427 if (filter) { 1428 sk_filter_uncharge(sk, filter); 1429 RCU_INIT_POINTER(sk->sk_filter, NULL); 1430 } 1431 if (rcu_access_pointer(sk->sk_reuseport_cb)) 1432 reuseport_detach_sock(sk); 1433 1434 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP); 1435 1436 if (atomic_read(&sk->sk_omem_alloc)) 1437 pr_debug("%s: optmem leakage (%d bytes) detected\n", 1438 __func__, atomic_read(&sk->sk_omem_alloc)); 1439 1440 if (sk->sk_peer_cred) 1441 put_cred(sk->sk_peer_cred); 1442 put_pid(sk->sk_peer_pid); 1443 if (likely(sk->sk_net_refcnt)) 1444 put_net(sock_net(sk)); 1445 sk_prot_free(sk->sk_prot_creator, sk); 1446 } 1447 1448 void sk_destruct(struct sock *sk) 1449 { 1450 if (sock_flag(sk, SOCK_RCU_FREE)) 1451 call_rcu(&sk->sk_rcu, __sk_destruct); 1452 else 1453 __sk_destruct(&sk->sk_rcu); 1454 } 1455 1456 static void __sk_free(struct sock *sk) 1457 { 1458 if (unlikely(sock_diag_has_destroy_listeners(sk) && sk->sk_net_refcnt)) 1459 sock_diag_broadcast_destroy(sk); 1460 else 1461 sk_destruct(sk); 1462 } 1463 1464 void sk_free(struct sock *sk) 1465 { 1466 /* 1467 * We subtract one from sk_wmem_alloc and can know if 1468 * some packets are still in some tx queue. 1469 * If not null, sock_wfree() will call __sk_free(sk) later 1470 */ 1471 if (atomic_dec_and_test(&sk->sk_wmem_alloc)) 1472 __sk_free(sk); 1473 } 1474 EXPORT_SYMBOL(sk_free); 1475 1476 /** 1477 * sk_clone_lock - clone a socket, and lock its clone 1478 * @sk: the socket to clone 1479 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 1480 * 1481 * Caller must unlock socket even in error path (bh_unlock_sock(newsk)) 1482 */ 1483 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority) 1484 { 1485 struct sock *newsk; 1486 bool is_charged = true; 1487 1488 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family); 1489 if (newsk != NULL) { 1490 struct sk_filter *filter; 1491 1492 sock_copy(newsk, sk); 1493 1494 /* SANITY */ 1495 if (likely(newsk->sk_net_refcnt)) 1496 get_net(sock_net(newsk)); 1497 sk_node_init(&newsk->sk_node); 1498 sock_lock_init(newsk); 1499 bh_lock_sock(newsk); 1500 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL; 1501 newsk->sk_backlog.len = 0; 1502 1503 atomic_set(&newsk->sk_rmem_alloc, 0); 1504 /* 1505 * sk_wmem_alloc set to one (see sk_free() and sock_wfree()) 1506 */ 1507 atomic_set(&newsk->sk_wmem_alloc, 1); 1508 atomic_set(&newsk->sk_omem_alloc, 0); 1509 skb_queue_head_init(&newsk->sk_receive_queue); 1510 skb_queue_head_init(&newsk->sk_write_queue); 1511 1512 rwlock_init(&newsk->sk_callback_lock); 1513 lockdep_set_class_and_name(&newsk->sk_callback_lock, 1514 af_callback_keys + newsk->sk_family, 1515 af_family_clock_key_strings[newsk->sk_family]); 1516 1517 newsk->sk_dst_cache = NULL; 1518 newsk->sk_wmem_queued = 0; 1519 newsk->sk_forward_alloc = 0; 1520 atomic_set(&newsk->sk_drops, 0); 1521 newsk->sk_send_head = NULL; 1522 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK; 1523 1524 sock_reset_flag(newsk, SOCK_DONE); 1525 skb_queue_head_init(&newsk->sk_error_queue); 1526 1527 filter = rcu_dereference_protected(newsk->sk_filter, 1); 1528 if (filter != NULL) 1529 /* though it's an empty new sock, the charging may fail 1530 * if sysctl_optmem_max was changed between creation of 1531 * original socket and cloning 1532 */ 1533 is_charged = sk_filter_charge(newsk, filter); 1534 1535 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) { 1536 /* It is still raw copy of parent, so invalidate 1537 * destructor and make plain sk_free() */ 1538 newsk->sk_destruct = NULL; 1539 bh_unlock_sock(newsk); 1540 sk_free(newsk); 1541 newsk = NULL; 1542 goto out; 1543 } 1544 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL); 1545 1546 newsk->sk_err = 0; 1547 newsk->sk_err_soft = 0; 1548 newsk->sk_priority = 0; 1549 newsk->sk_incoming_cpu = raw_smp_processor_id(); 1550 atomic64_set(&newsk->sk_cookie, 0); 1551 1552 mem_cgroup_sk_alloc(newsk); 1553 cgroup_sk_alloc(&newsk->sk_cgrp_data); 1554 1555 /* 1556 * Before updating sk_refcnt, we must commit prior changes to memory 1557 * (Documentation/RCU/rculist_nulls.txt for details) 1558 */ 1559 smp_wmb(); 1560 atomic_set(&newsk->sk_refcnt, 2); 1561 1562 /* 1563 * Increment the counter in the same struct proto as the master 1564 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that 1565 * is the same as sk->sk_prot->socks, as this field was copied 1566 * with memcpy). 1567 * 1568 * This _changes_ the previous behaviour, where 1569 * tcp_create_openreq_child always was incrementing the 1570 * equivalent to tcp_prot->socks (inet_sock_nr), so this have 1571 * to be taken into account in all callers. -acme 1572 */ 1573 sk_refcnt_debug_inc(newsk); 1574 sk_set_socket(newsk, NULL); 1575 newsk->sk_wq = NULL; 1576 1577 if (newsk->sk_prot->sockets_allocated) 1578 sk_sockets_allocated_inc(newsk); 1579 1580 if (sock_needs_netstamp(sk) && 1581 newsk->sk_flags & SK_FLAGS_TIMESTAMP) 1582 net_enable_timestamp(); 1583 } 1584 out: 1585 return newsk; 1586 } 1587 EXPORT_SYMBOL_GPL(sk_clone_lock); 1588 1589 void sk_setup_caps(struct sock *sk, struct dst_entry *dst) 1590 { 1591 u32 max_segs = 1; 1592 1593 sk_dst_set(sk, dst); 1594 sk->sk_route_caps = dst->dev->features; 1595 if (sk->sk_route_caps & NETIF_F_GSO) 1596 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE; 1597 sk->sk_route_caps &= ~sk->sk_route_nocaps; 1598 if (sk_can_gso(sk)) { 1599 if (dst->header_len) { 1600 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 1601 } else { 1602 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM; 1603 sk->sk_gso_max_size = dst->dev->gso_max_size; 1604 max_segs = max_t(u32, dst->dev->gso_max_segs, 1); 1605 } 1606 } 1607 sk->sk_gso_max_segs = max_segs; 1608 } 1609 EXPORT_SYMBOL_GPL(sk_setup_caps); 1610 1611 /* 1612 * Simple resource managers for sockets. 1613 */ 1614 1615 1616 /* 1617 * Write buffer destructor automatically called from kfree_skb. 1618 */ 1619 void sock_wfree(struct sk_buff *skb) 1620 { 1621 struct sock *sk = skb->sk; 1622 unsigned int len = skb->truesize; 1623 1624 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) { 1625 /* 1626 * Keep a reference on sk_wmem_alloc, this will be released 1627 * after sk_write_space() call 1628 */ 1629 atomic_sub(len - 1, &sk->sk_wmem_alloc); 1630 sk->sk_write_space(sk); 1631 len = 1; 1632 } 1633 /* 1634 * if sk_wmem_alloc reaches 0, we must finish what sk_free() 1635 * could not do because of in-flight packets 1636 */ 1637 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc)) 1638 __sk_free(sk); 1639 } 1640 EXPORT_SYMBOL(sock_wfree); 1641 1642 /* This variant of sock_wfree() is used by TCP, 1643 * since it sets SOCK_USE_WRITE_QUEUE. 1644 */ 1645 void __sock_wfree(struct sk_buff *skb) 1646 { 1647 struct sock *sk = skb->sk; 1648 1649 if (atomic_sub_and_test(skb->truesize, &sk->sk_wmem_alloc)) 1650 __sk_free(sk); 1651 } 1652 1653 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk) 1654 { 1655 skb_orphan(skb); 1656 skb->sk = sk; 1657 #ifdef CONFIG_INET 1658 if (unlikely(!sk_fullsock(sk))) { 1659 skb->destructor = sock_edemux; 1660 sock_hold(sk); 1661 return; 1662 } 1663 #endif 1664 skb->destructor = sock_wfree; 1665 skb_set_hash_from_sk(skb, sk); 1666 /* 1667 * We used to take a refcount on sk, but following operation 1668 * is enough to guarantee sk_free() wont free this sock until 1669 * all in-flight packets are completed 1670 */ 1671 atomic_add(skb->truesize, &sk->sk_wmem_alloc); 1672 } 1673 EXPORT_SYMBOL(skb_set_owner_w); 1674 1675 /* This helper is used by netem, as it can hold packets in its 1676 * delay queue. We want to allow the owner socket to send more 1677 * packets, as if they were already TX completed by a typical driver. 1678 * But we also want to keep skb->sk set because some packet schedulers 1679 * rely on it (sch_fq for example). So we set skb->truesize to a small 1680 * amount (1) and decrease sk_wmem_alloc accordingly. 1681 */ 1682 void skb_orphan_partial(struct sk_buff *skb) 1683 { 1684 /* If this skb is a TCP pure ACK or already went here, 1685 * we have nothing to do. 2 is already a very small truesize. 1686 */ 1687 if (skb->truesize <= 2) 1688 return; 1689 1690 /* TCP stack sets skb->ooo_okay based on sk_wmem_alloc, 1691 * so we do not completely orphan skb, but transfert all 1692 * accounted bytes but one, to avoid unexpected reorders. 1693 */ 1694 if (skb->destructor == sock_wfree 1695 #ifdef CONFIG_INET 1696 || skb->destructor == tcp_wfree 1697 #endif 1698 ) { 1699 atomic_sub(skb->truesize - 1, &skb->sk->sk_wmem_alloc); 1700 skb->truesize = 1; 1701 } else { 1702 skb_orphan(skb); 1703 } 1704 } 1705 EXPORT_SYMBOL(skb_orphan_partial); 1706 1707 /* 1708 * Read buffer destructor automatically called from kfree_skb. 1709 */ 1710 void sock_rfree(struct sk_buff *skb) 1711 { 1712 struct sock *sk = skb->sk; 1713 unsigned int len = skb->truesize; 1714 1715 atomic_sub(len, &sk->sk_rmem_alloc); 1716 sk_mem_uncharge(sk, len); 1717 } 1718 EXPORT_SYMBOL(sock_rfree); 1719 1720 /* 1721 * Buffer destructor for skbs that are not used directly in read or write 1722 * path, e.g. for error handler skbs. Automatically called from kfree_skb. 1723 */ 1724 void sock_efree(struct sk_buff *skb) 1725 { 1726 sock_put(skb->sk); 1727 } 1728 EXPORT_SYMBOL(sock_efree); 1729 1730 kuid_t sock_i_uid(struct sock *sk) 1731 { 1732 kuid_t uid; 1733 1734 read_lock_bh(&sk->sk_callback_lock); 1735 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID; 1736 read_unlock_bh(&sk->sk_callback_lock); 1737 return uid; 1738 } 1739 EXPORT_SYMBOL(sock_i_uid); 1740 1741 unsigned long sock_i_ino(struct sock *sk) 1742 { 1743 unsigned long ino; 1744 1745 read_lock_bh(&sk->sk_callback_lock); 1746 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0; 1747 read_unlock_bh(&sk->sk_callback_lock); 1748 return ino; 1749 } 1750 EXPORT_SYMBOL(sock_i_ino); 1751 1752 /* 1753 * Allocate a skb from the socket's send buffer. 1754 */ 1755 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 1756 gfp_t priority) 1757 { 1758 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) { 1759 struct sk_buff *skb = alloc_skb(size, priority); 1760 if (skb) { 1761 skb_set_owner_w(skb, sk); 1762 return skb; 1763 } 1764 } 1765 return NULL; 1766 } 1767 EXPORT_SYMBOL(sock_wmalloc); 1768 1769 /* 1770 * Allocate a memory block from the socket's option memory buffer. 1771 */ 1772 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority) 1773 { 1774 if ((unsigned int)size <= sysctl_optmem_max && 1775 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) { 1776 void *mem; 1777 /* First do the add, to avoid the race if kmalloc 1778 * might sleep. 1779 */ 1780 atomic_add(size, &sk->sk_omem_alloc); 1781 mem = kmalloc(size, priority); 1782 if (mem) 1783 return mem; 1784 atomic_sub(size, &sk->sk_omem_alloc); 1785 } 1786 return NULL; 1787 } 1788 EXPORT_SYMBOL(sock_kmalloc); 1789 1790 /* Free an option memory block. Note, we actually want the inline 1791 * here as this allows gcc to detect the nullify and fold away the 1792 * condition entirely. 1793 */ 1794 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size, 1795 const bool nullify) 1796 { 1797 if (WARN_ON_ONCE(!mem)) 1798 return; 1799 if (nullify) 1800 kzfree(mem); 1801 else 1802 kfree(mem); 1803 atomic_sub(size, &sk->sk_omem_alloc); 1804 } 1805 1806 void sock_kfree_s(struct sock *sk, void *mem, int size) 1807 { 1808 __sock_kfree_s(sk, mem, size, false); 1809 } 1810 EXPORT_SYMBOL(sock_kfree_s); 1811 1812 void sock_kzfree_s(struct sock *sk, void *mem, int size) 1813 { 1814 __sock_kfree_s(sk, mem, size, true); 1815 } 1816 EXPORT_SYMBOL(sock_kzfree_s); 1817 1818 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock. 1819 I think, these locks should be removed for datagram sockets. 1820 */ 1821 static long sock_wait_for_wmem(struct sock *sk, long timeo) 1822 { 1823 DEFINE_WAIT(wait); 1824 1825 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk); 1826 for (;;) { 1827 if (!timeo) 1828 break; 1829 if (signal_pending(current)) 1830 break; 1831 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 1832 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 1833 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) 1834 break; 1835 if (sk->sk_shutdown & SEND_SHUTDOWN) 1836 break; 1837 if (sk->sk_err) 1838 break; 1839 timeo = schedule_timeout(timeo); 1840 } 1841 finish_wait(sk_sleep(sk), &wait); 1842 return timeo; 1843 } 1844 1845 1846 /* 1847 * Generic send/receive buffer handlers 1848 */ 1849 1850 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 1851 unsigned long data_len, int noblock, 1852 int *errcode, int max_page_order) 1853 { 1854 struct sk_buff *skb; 1855 long timeo; 1856 int err; 1857 1858 timeo = sock_sndtimeo(sk, noblock); 1859 for (;;) { 1860 err = sock_error(sk); 1861 if (err != 0) 1862 goto failure; 1863 1864 err = -EPIPE; 1865 if (sk->sk_shutdown & SEND_SHUTDOWN) 1866 goto failure; 1867 1868 if (sk_wmem_alloc_get(sk) < sk->sk_sndbuf) 1869 break; 1870 1871 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk); 1872 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 1873 err = -EAGAIN; 1874 if (!timeo) 1875 goto failure; 1876 if (signal_pending(current)) 1877 goto interrupted; 1878 timeo = sock_wait_for_wmem(sk, timeo); 1879 } 1880 skb = alloc_skb_with_frags(header_len, data_len, max_page_order, 1881 errcode, sk->sk_allocation); 1882 if (skb) 1883 skb_set_owner_w(skb, sk); 1884 return skb; 1885 1886 interrupted: 1887 err = sock_intr_errno(timeo); 1888 failure: 1889 *errcode = err; 1890 return NULL; 1891 } 1892 EXPORT_SYMBOL(sock_alloc_send_pskb); 1893 1894 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size, 1895 int noblock, int *errcode) 1896 { 1897 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0); 1898 } 1899 EXPORT_SYMBOL(sock_alloc_send_skb); 1900 1901 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg, 1902 struct sockcm_cookie *sockc) 1903 { 1904 u32 tsflags; 1905 1906 switch (cmsg->cmsg_type) { 1907 case SO_MARK: 1908 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) 1909 return -EPERM; 1910 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 1911 return -EINVAL; 1912 sockc->mark = *(u32 *)CMSG_DATA(cmsg); 1913 break; 1914 case SO_TIMESTAMPING: 1915 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 1916 return -EINVAL; 1917 1918 tsflags = *(u32 *)CMSG_DATA(cmsg); 1919 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK) 1920 return -EINVAL; 1921 1922 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK; 1923 sockc->tsflags |= tsflags; 1924 break; 1925 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */ 1926 case SCM_RIGHTS: 1927 case SCM_CREDENTIALS: 1928 break; 1929 default: 1930 return -EINVAL; 1931 } 1932 return 0; 1933 } 1934 EXPORT_SYMBOL(__sock_cmsg_send); 1935 1936 int sock_cmsg_send(struct sock *sk, struct msghdr *msg, 1937 struct sockcm_cookie *sockc) 1938 { 1939 struct cmsghdr *cmsg; 1940 int ret; 1941 1942 for_each_cmsghdr(cmsg, msg) { 1943 if (!CMSG_OK(msg, cmsg)) 1944 return -EINVAL; 1945 if (cmsg->cmsg_level != SOL_SOCKET) 1946 continue; 1947 ret = __sock_cmsg_send(sk, msg, cmsg, sockc); 1948 if (ret) 1949 return ret; 1950 } 1951 return 0; 1952 } 1953 EXPORT_SYMBOL(sock_cmsg_send); 1954 1955 /* On 32bit arches, an skb frag is limited to 2^15 */ 1956 #define SKB_FRAG_PAGE_ORDER get_order(32768) 1957 1958 /** 1959 * skb_page_frag_refill - check that a page_frag contains enough room 1960 * @sz: minimum size of the fragment we want to get 1961 * @pfrag: pointer to page_frag 1962 * @gfp: priority for memory allocation 1963 * 1964 * Note: While this allocator tries to use high order pages, there is 1965 * no guarantee that allocations succeed. Therefore, @sz MUST be 1966 * less or equal than PAGE_SIZE. 1967 */ 1968 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp) 1969 { 1970 if (pfrag->page) { 1971 if (page_ref_count(pfrag->page) == 1) { 1972 pfrag->offset = 0; 1973 return true; 1974 } 1975 if (pfrag->offset + sz <= pfrag->size) 1976 return true; 1977 put_page(pfrag->page); 1978 } 1979 1980 pfrag->offset = 0; 1981 if (SKB_FRAG_PAGE_ORDER) { 1982 /* Avoid direct reclaim but allow kswapd to wake */ 1983 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) | 1984 __GFP_COMP | __GFP_NOWARN | 1985 __GFP_NORETRY, 1986 SKB_FRAG_PAGE_ORDER); 1987 if (likely(pfrag->page)) { 1988 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER; 1989 return true; 1990 } 1991 } 1992 pfrag->page = alloc_page(gfp); 1993 if (likely(pfrag->page)) { 1994 pfrag->size = PAGE_SIZE; 1995 return true; 1996 } 1997 return false; 1998 } 1999 EXPORT_SYMBOL(skb_page_frag_refill); 2000 2001 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag) 2002 { 2003 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation))) 2004 return true; 2005 2006 sk_enter_memory_pressure(sk); 2007 sk_stream_moderate_sndbuf(sk); 2008 return false; 2009 } 2010 EXPORT_SYMBOL(sk_page_frag_refill); 2011 2012 static void __lock_sock(struct sock *sk) 2013 __releases(&sk->sk_lock.slock) 2014 __acquires(&sk->sk_lock.slock) 2015 { 2016 DEFINE_WAIT(wait); 2017 2018 for (;;) { 2019 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait, 2020 TASK_UNINTERRUPTIBLE); 2021 spin_unlock_bh(&sk->sk_lock.slock); 2022 schedule(); 2023 spin_lock_bh(&sk->sk_lock.slock); 2024 if (!sock_owned_by_user(sk)) 2025 break; 2026 } 2027 finish_wait(&sk->sk_lock.wq, &wait); 2028 } 2029 2030 static void __release_sock(struct sock *sk) 2031 __releases(&sk->sk_lock.slock) 2032 __acquires(&sk->sk_lock.slock) 2033 { 2034 struct sk_buff *skb, *next; 2035 2036 while ((skb = sk->sk_backlog.head) != NULL) { 2037 sk->sk_backlog.head = sk->sk_backlog.tail = NULL; 2038 2039 spin_unlock_bh(&sk->sk_lock.slock); 2040 2041 do { 2042 next = skb->next; 2043 prefetch(next); 2044 WARN_ON_ONCE(skb_dst_is_noref(skb)); 2045 skb->next = NULL; 2046 sk_backlog_rcv(sk, skb); 2047 2048 cond_resched(); 2049 2050 skb = next; 2051 } while (skb != NULL); 2052 2053 spin_lock_bh(&sk->sk_lock.slock); 2054 } 2055 2056 /* 2057 * Doing the zeroing here guarantee we can not loop forever 2058 * while a wild producer attempts to flood us. 2059 */ 2060 sk->sk_backlog.len = 0; 2061 } 2062 2063 void __sk_flush_backlog(struct sock *sk) 2064 { 2065 spin_lock_bh(&sk->sk_lock.slock); 2066 __release_sock(sk); 2067 spin_unlock_bh(&sk->sk_lock.slock); 2068 } 2069 2070 /** 2071 * sk_wait_data - wait for data to arrive at sk_receive_queue 2072 * @sk: sock to wait on 2073 * @timeo: for how long 2074 * @skb: last skb seen on sk_receive_queue 2075 * 2076 * Now socket state including sk->sk_err is changed only under lock, 2077 * hence we may omit checks after joining wait queue. 2078 * We check receive queue before schedule() only as optimization; 2079 * it is very likely that release_sock() added new data. 2080 */ 2081 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb) 2082 { 2083 int rc; 2084 DEFINE_WAIT(wait); 2085 2086 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 2087 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk); 2088 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb); 2089 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk); 2090 finish_wait(sk_sleep(sk), &wait); 2091 return rc; 2092 } 2093 EXPORT_SYMBOL(sk_wait_data); 2094 2095 /** 2096 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated 2097 * @sk: socket 2098 * @size: memory size to allocate 2099 * @kind: allocation type 2100 * 2101 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means 2102 * rmem allocation. This function assumes that protocols which have 2103 * memory_pressure use sk_wmem_queued as write buffer accounting. 2104 */ 2105 int __sk_mem_schedule(struct sock *sk, int size, int kind) 2106 { 2107 struct proto *prot = sk->sk_prot; 2108 int amt = sk_mem_pages(size); 2109 long allocated; 2110 2111 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM; 2112 2113 allocated = sk_memory_allocated_add(sk, amt); 2114 2115 if (mem_cgroup_sockets_enabled && sk->sk_memcg && 2116 !mem_cgroup_charge_skmem(sk->sk_memcg, amt)) 2117 goto suppress_allocation; 2118 2119 /* Under limit. */ 2120 if (allocated <= sk_prot_mem_limits(sk, 0)) { 2121 sk_leave_memory_pressure(sk); 2122 return 1; 2123 } 2124 2125 /* Under pressure. */ 2126 if (allocated > sk_prot_mem_limits(sk, 1)) 2127 sk_enter_memory_pressure(sk); 2128 2129 /* Over hard limit. */ 2130 if (allocated > sk_prot_mem_limits(sk, 2)) 2131 goto suppress_allocation; 2132 2133 /* guarantee minimum buffer size under pressure */ 2134 if (kind == SK_MEM_RECV) { 2135 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0]) 2136 return 1; 2137 2138 } else { /* SK_MEM_SEND */ 2139 if (sk->sk_type == SOCK_STREAM) { 2140 if (sk->sk_wmem_queued < prot->sysctl_wmem[0]) 2141 return 1; 2142 } else if (atomic_read(&sk->sk_wmem_alloc) < 2143 prot->sysctl_wmem[0]) 2144 return 1; 2145 } 2146 2147 if (sk_has_memory_pressure(sk)) { 2148 int alloc; 2149 2150 if (!sk_under_memory_pressure(sk)) 2151 return 1; 2152 alloc = sk_sockets_allocated_read_positive(sk); 2153 if (sk_prot_mem_limits(sk, 2) > alloc * 2154 sk_mem_pages(sk->sk_wmem_queued + 2155 atomic_read(&sk->sk_rmem_alloc) + 2156 sk->sk_forward_alloc)) 2157 return 1; 2158 } 2159 2160 suppress_allocation: 2161 2162 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) { 2163 sk_stream_moderate_sndbuf(sk); 2164 2165 /* Fail only if socket is _under_ its sndbuf. 2166 * In this case we cannot block, so that we have to fail. 2167 */ 2168 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf) 2169 return 1; 2170 } 2171 2172 trace_sock_exceed_buf_limit(sk, prot, allocated); 2173 2174 /* Alas. Undo changes. */ 2175 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM; 2176 2177 sk_memory_allocated_sub(sk, amt); 2178 2179 if (mem_cgroup_sockets_enabled && sk->sk_memcg) 2180 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt); 2181 2182 return 0; 2183 } 2184 EXPORT_SYMBOL(__sk_mem_schedule); 2185 2186 /** 2187 * __sk_mem_reclaim - reclaim memory_allocated 2188 * @sk: socket 2189 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple) 2190 */ 2191 void __sk_mem_reclaim(struct sock *sk, int amount) 2192 { 2193 amount >>= SK_MEM_QUANTUM_SHIFT; 2194 sk_memory_allocated_sub(sk, amount); 2195 sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT; 2196 2197 if (mem_cgroup_sockets_enabled && sk->sk_memcg) 2198 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount); 2199 2200 if (sk_under_memory_pressure(sk) && 2201 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0))) 2202 sk_leave_memory_pressure(sk); 2203 } 2204 EXPORT_SYMBOL(__sk_mem_reclaim); 2205 2206 int sk_set_peek_off(struct sock *sk, int val) 2207 { 2208 if (val < 0) 2209 return -EINVAL; 2210 2211 sk->sk_peek_off = val; 2212 return 0; 2213 } 2214 EXPORT_SYMBOL_GPL(sk_set_peek_off); 2215 2216 /* 2217 * Set of default routines for initialising struct proto_ops when 2218 * the protocol does not support a particular function. In certain 2219 * cases where it makes no sense for a protocol to have a "do nothing" 2220 * function, some default processing is provided. 2221 */ 2222 2223 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len) 2224 { 2225 return -EOPNOTSUPP; 2226 } 2227 EXPORT_SYMBOL(sock_no_bind); 2228 2229 int sock_no_connect(struct socket *sock, struct sockaddr *saddr, 2230 int len, int flags) 2231 { 2232 return -EOPNOTSUPP; 2233 } 2234 EXPORT_SYMBOL(sock_no_connect); 2235 2236 int sock_no_socketpair(struct socket *sock1, struct socket *sock2) 2237 { 2238 return -EOPNOTSUPP; 2239 } 2240 EXPORT_SYMBOL(sock_no_socketpair); 2241 2242 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags) 2243 { 2244 return -EOPNOTSUPP; 2245 } 2246 EXPORT_SYMBOL(sock_no_accept); 2247 2248 int sock_no_getname(struct socket *sock, struct sockaddr *saddr, 2249 int *len, int peer) 2250 { 2251 return -EOPNOTSUPP; 2252 } 2253 EXPORT_SYMBOL(sock_no_getname); 2254 2255 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt) 2256 { 2257 return 0; 2258 } 2259 EXPORT_SYMBOL(sock_no_poll); 2260 2261 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 2262 { 2263 return -EOPNOTSUPP; 2264 } 2265 EXPORT_SYMBOL(sock_no_ioctl); 2266 2267 int sock_no_listen(struct socket *sock, int backlog) 2268 { 2269 return -EOPNOTSUPP; 2270 } 2271 EXPORT_SYMBOL(sock_no_listen); 2272 2273 int sock_no_shutdown(struct socket *sock, int how) 2274 { 2275 return -EOPNOTSUPP; 2276 } 2277 EXPORT_SYMBOL(sock_no_shutdown); 2278 2279 int sock_no_setsockopt(struct socket *sock, int level, int optname, 2280 char __user *optval, unsigned int optlen) 2281 { 2282 return -EOPNOTSUPP; 2283 } 2284 EXPORT_SYMBOL(sock_no_setsockopt); 2285 2286 int sock_no_getsockopt(struct socket *sock, int level, int optname, 2287 char __user *optval, int __user *optlen) 2288 { 2289 return -EOPNOTSUPP; 2290 } 2291 EXPORT_SYMBOL(sock_no_getsockopt); 2292 2293 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len) 2294 { 2295 return -EOPNOTSUPP; 2296 } 2297 EXPORT_SYMBOL(sock_no_sendmsg); 2298 2299 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len, 2300 int flags) 2301 { 2302 return -EOPNOTSUPP; 2303 } 2304 EXPORT_SYMBOL(sock_no_recvmsg); 2305 2306 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma) 2307 { 2308 /* Mirror missing mmap method error code */ 2309 return -ENODEV; 2310 } 2311 EXPORT_SYMBOL(sock_no_mmap); 2312 2313 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags) 2314 { 2315 ssize_t res; 2316 struct msghdr msg = {.msg_flags = flags}; 2317 struct kvec iov; 2318 char *kaddr = kmap(page); 2319 iov.iov_base = kaddr + offset; 2320 iov.iov_len = size; 2321 res = kernel_sendmsg(sock, &msg, &iov, 1, size); 2322 kunmap(page); 2323 return res; 2324 } 2325 EXPORT_SYMBOL(sock_no_sendpage); 2326 2327 /* 2328 * Default Socket Callbacks 2329 */ 2330 2331 static void sock_def_wakeup(struct sock *sk) 2332 { 2333 struct socket_wq *wq; 2334 2335 rcu_read_lock(); 2336 wq = rcu_dereference(sk->sk_wq); 2337 if (skwq_has_sleeper(wq)) 2338 wake_up_interruptible_all(&wq->wait); 2339 rcu_read_unlock(); 2340 } 2341 2342 static void sock_def_error_report(struct sock *sk) 2343 { 2344 struct socket_wq *wq; 2345 2346 rcu_read_lock(); 2347 wq = rcu_dereference(sk->sk_wq); 2348 if (skwq_has_sleeper(wq)) 2349 wake_up_interruptible_poll(&wq->wait, POLLERR); 2350 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR); 2351 rcu_read_unlock(); 2352 } 2353 2354 static void sock_def_readable(struct sock *sk) 2355 { 2356 struct socket_wq *wq; 2357 2358 rcu_read_lock(); 2359 wq = rcu_dereference(sk->sk_wq); 2360 if (skwq_has_sleeper(wq)) 2361 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI | 2362 POLLRDNORM | POLLRDBAND); 2363 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN); 2364 rcu_read_unlock(); 2365 } 2366 2367 static void sock_def_write_space(struct sock *sk) 2368 { 2369 struct socket_wq *wq; 2370 2371 rcu_read_lock(); 2372 2373 /* Do not wake up a writer until he can make "significant" 2374 * progress. --DaveM 2375 */ 2376 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) { 2377 wq = rcu_dereference(sk->sk_wq); 2378 if (skwq_has_sleeper(wq)) 2379 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT | 2380 POLLWRNORM | POLLWRBAND); 2381 2382 /* Should agree with poll, otherwise some programs break */ 2383 if (sock_writeable(sk)) 2384 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT); 2385 } 2386 2387 rcu_read_unlock(); 2388 } 2389 2390 static void sock_def_destruct(struct sock *sk) 2391 { 2392 } 2393 2394 void sk_send_sigurg(struct sock *sk) 2395 { 2396 if (sk->sk_socket && sk->sk_socket->file) 2397 if (send_sigurg(&sk->sk_socket->file->f_owner)) 2398 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI); 2399 } 2400 EXPORT_SYMBOL(sk_send_sigurg); 2401 2402 void sk_reset_timer(struct sock *sk, struct timer_list* timer, 2403 unsigned long expires) 2404 { 2405 if (!mod_timer(timer, expires)) 2406 sock_hold(sk); 2407 } 2408 EXPORT_SYMBOL(sk_reset_timer); 2409 2410 void sk_stop_timer(struct sock *sk, struct timer_list* timer) 2411 { 2412 if (del_timer(timer)) 2413 __sock_put(sk); 2414 } 2415 EXPORT_SYMBOL(sk_stop_timer); 2416 2417 void sock_init_data(struct socket *sock, struct sock *sk) 2418 { 2419 skb_queue_head_init(&sk->sk_receive_queue); 2420 skb_queue_head_init(&sk->sk_write_queue); 2421 skb_queue_head_init(&sk->sk_error_queue); 2422 2423 sk->sk_send_head = NULL; 2424 2425 init_timer(&sk->sk_timer); 2426 2427 sk->sk_allocation = GFP_KERNEL; 2428 sk->sk_rcvbuf = sysctl_rmem_default; 2429 sk->sk_sndbuf = sysctl_wmem_default; 2430 sk->sk_state = TCP_CLOSE; 2431 sk_set_socket(sk, sock); 2432 2433 sock_set_flag(sk, SOCK_ZAPPED); 2434 2435 if (sock) { 2436 sk->sk_type = sock->type; 2437 sk->sk_wq = sock->wq; 2438 sock->sk = sk; 2439 } else 2440 sk->sk_wq = NULL; 2441 2442 rwlock_init(&sk->sk_callback_lock); 2443 lockdep_set_class_and_name(&sk->sk_callback_lock, 2444 af_callback_keys + sk->sk_family, 2445 af_family_clock_key_strings[sk->sk_family]); 2446 2447 sk->sk_state_change = sock_def_wakeup; 2448 sk->sk_data_ready = sock_def_readable; 2449 sk->sk_write_space = sock_def_write_space; 2450 sk->sk_error_report = sock_def_error_report; 2451 sk->sk_destruct = sock_def_destruct; 2452 2453 sk->sk_frag.page = NULL; 2454 sk->sk_frag.offset = 0; 2455 sk->sk_peek_off = -1; 2456 2457 sk->sk_peer_pid = NULL; 2458 sk->sk_peer_cred = NULL; 2459 sk->sk_write_pending = 0; 2460 sk->sk_rcvlowat = 1; 2461 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; 2462 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT; 2463 2464 sk->sk_stamp = ktime_set(-1L, 0); 2465 2466 #ifdef CONFIG_NET_RX_BUSY_POLL 2467 sk->sk_napi_id = 0; 2468 sk->sk_ll_usec = sysctl_net_busy_read; 2469 #endif 2470 2471 sk->sk_max_pacing_rate = ~0U; 2472 sk->sk_pacing_rate = ~0U; 2473 sk->sk_incoming_cpu = -1; 2474 /* 2475 * Before updating sk_refcnt, we must commit prior changes to memory 2476 * (Documentation/RCU/rculist_nulls.txt for details) 2477 */ 2478 smp_wmb(); 2479 atomic_set(&sk->sk_refcnt, 1); 2480 atomic_set(&sk->sk_drops, 0); 2481 } 2482 EXPORT_SYMBOL(sock_init_data); 2483 2484 void lock_sock_nested(struct sock *sk, int subclass) 2485 { 2486 might_sleep(); 2487 spin_lock_bh(&sk->sk_lock.slock); 2488 if (sk->sk_lock.owned) 2489 __lock_sock(sk); 2490 sk->sk_lock.owned = 1; 2491 spin_unlock(&sk->sk_lock.slock); 2492 /* 2493 * The sk_lock has mutex_lock() semantics here: 2494 */ 2495 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_); 2496 local_bh_enable(); 2497 } 2498 EXPORT_SYMBOL(lock_sock_nested); 2499 2500 void release_sock(struct sock *sk) 2501 { 2502 spin_lock_bh(&sk->sk_lock.slock); 2503 if (sk->sk_backlog.tail) 2504 __release_sock(sk); 2505 2506 /* Warning : release_cb() might need to release sk ownership, 2507 * ie call sock_release_ownership(sk) before us. 2508 */ 2509 if (sk->sk_prot->release_cb) 2510 sk->sk_prot->release_cb(sk); 2511 2512 sock_release_ownership(sk); 2513 if (waitqueue_active(&sk->sk_lock.wq)) 2514 wake_up(&sk->sk_lock.wq); 2515 spin_unlock_bh(&sk->sk_lock.slock); 2516 } 2517 EXPORT_SYMBOL(release_sock); 2518 2519 /** 2520 * lock_sock_fast - fast version of lock_sock 2521 * @sk: socket 2522 * 2523 * This version should be used for very small section, where process wont block 2524 * return false if fast path is taken 2525 * sk_lock.slock locked, owned = 0, BH disabled 2526 * return true if slow path is taken 2527 * sk_lock.slock unlocked, owned = 1, BH enabled 2528 */ 2529 bool lock_sock_fast(struct sock *sk) 2530 { 2531 might_sleep(); 2532 spin_lock_bh(&sk->sk_lock.slock); 2533 2534 if (!sk->sk_lock.owned) 2535 /* 2536 * Note : We must disable BH 2537 */ 2538 return false; 2539 2540 __lock_sock(sk); 2541 sk->sk_lock.owned = 1; 2542 spin_unlock(&sk->sk_lock.slock); 2543 /* 2544 * The sk_lock has mutex_lock() semantics here: 2545 */ 2546 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_); 2547 local_bh_enable(); 2548 return true; 2549 } 2550 EXPORT_SYMBOL(lock_sock_fast); 2551 2552 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp) 2553 { 2554 struct timeval tv; 2555 if (!sock_flag(sk, SOCK_TIMESTAMP)) 2556 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 2557 tv = ktime_to_timeval(sk->sk_stamp); 2558 if (tv.tv_sec == -1) 2559 return -ENOENT; 2560 if (tv.tv_sec == 0) { 2561 sk->sk_stamp = ktime_get_real(); 2562 tv = ktime_to_timeval(sk->sk_stamp); 2563 } 2564 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0; 2565 } 2566 EXPORT_SYMBOL(sock_get_timestamp); 2567 2568 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp) 2569 { 2570 struct timespec ts; 2571 if (!sock_flag(sk, SOCK_TIMESTAMP)) 2572 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 2573 ts = ktime_to_timespec(sk->sk_stamp); 2574 if (ts.tv_sec == -1) 2575 return -ENOENT; 2576 if (ts.tv_sec == 0) { 2577 sk->sk_stamp = ktime_get_real(); 2578 ts = ktime_to_timespec(sk->sk_stamp); 2579 } 2580 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0; 2581 } 2582 EXPORT_SYMBOL(sock_get_timestampns); 2583 2584 void sock_enable_timestamp(struct sock *sk, int flag) 2585 { 2586 if (!sock_flag(sk, flag)) { 2587 unsigned long previous_flags = sk->sk_flags; 2588 2589 sock_set_flag(sk, flag); 2590 /* 2591 * we just set one of the two flags which require net 2592 * time stamping, but time stamping might have been on 2593 * already because of the other one 2594 */ 2595 if (sock_needs_netstamp(sk) && 2596 !(previous_flags & SK_FLAGS_TIMESTAMP)) 2597 net_enable_timestamp(); 2598 } 2599 } 2600 2601 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, 2602 int level, int type) 2603 { 2604 struct sock_exterr_skb *serr; 2605 struct sk_buff *skb; 2606 int copied, err; 2607 2608 err = -EAGAIN; 2609 skb = sock_dequeue_err_skb(sk); 2610 if (skb == NULL) 2611 goto out; 2612 2613 copied = skb->len; 2614 if (copied > len) { 2615 msg->msg_flags |= MSG_TRUNC; 2616 copied = len; 2617 } 2618 err = skb_copy_datagram_msg(skb, 0, msg, copied); 2619 if (err) 2620 goto out_free_skb; 2621 2622 sock_recv_timestamp(msg, sk, skb); 2623 2624 serr = SKB_EXT_ERR(skb); 2625 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee); 2626 2627 msg->msg_flags |= MSG_ERRQUEUE; 2628 err = copied; 2629 2630 out_free_skb: 2631 kfree_skb(skb); 2632 out: 2633 return err; 2634 } 2635 EXPORT_SYMBOL(sock_recv_errqueue); 2636 2637 /* 2638 * Get a socket option on an socket. 2639 * 2640 * FIX: POSIX 1003.1g is very ambiguous here. It states that 2641 * asynchronous errors should be reported by getsockopt. We assume 2642 * this means if you specify SO_ERROR (otherwise whats the point of it). 2643 */ 2644 int sock_common_getsockopt(struct socket *sock, int level, int optname, 2645 char __user *optval, int __user *optlen) 2646 { 2647 struct sock *sk = sock->sk; 2648 2649 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen); 2650 } 2651 EXPORT_SYMBOL(sock_common_getsockopt); 2652 2653 #ifdef CONFIG_COMPAT 2654 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname, 2655 char __user *optval, int __user *optlen) 2656 { 2657 struct sock *sk = sock->sk; 2658 2659 if (sk->sk_prot->compat_getsockopt != NULL) 2660 return sk->sk_prot->compat_getsockopt(sk, level, optname, 2661 optval, optlen); 2662 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen); 2663 } 2664 EXPORT_SYMBOL(compat_sock_common_getsockopt); 2665 #endif 2666 2667 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 2668 int flags) 2669 { 2670 struct sock *sk = sock->sk; 2671 int addr_len = 0; 2672 int err; 2673 2674 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT, 2675 flags & ~MSG_DONTWAIT, &addr_len); 2676 if (err >= 0) 2677 msg->msg_namelen = addr_len; 2678 return err; 2679 } 2680 EXPORT_SYMBOL(sock_common_recvmsg); 2681 2682 /* 2683 * Set socket options on an inet socket. 2684 */ 2685 int sock_common_setsockopt(struct socket *sock, int level, int optname, 2686 char __user *optval, unsigned int optlen) 2687 { 2688 struct sock *sk = sock->sk; 2689 2690 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen); 2691 } 2692 EXPORT_SYMBOL(sock_common_setsockopt); 2693 2694 #ifdef CONFIG_COMPAT 2695 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname, 2696 char __user *optval, unsigned int optlen) 2697 { 2698 struct sock *sk = sock->sk; 2699 2700 if (sk->sk_prot->compat_setsockopt != NULL) 2701 return sk->sk_prot->compat_setsockopt(sk, level, optname, 2702 optval, optlen); 2703 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen); 2704 } 2705 EXPORT_SYMBOL(compat_sock_common_setsockopt); 2706 #endif 2707 2708 void sk_common_release(struct sock *sk) 2709 { 2710 if (sk->sk_prot->destroy) 2711 sk->sk_prot->destroy(sk); 2712 2713 /* 2714 * Observation: when sock_common_release is called, processes have 2715 * no access to socket. But net still has. 2716 * Step one, detach it from networking: 2717 * 2718 * A. Remove from hash tables. 2719 */ 2720 2721 sk->sk_prot->unhash(sk); 2722 2723 /* 2724 * In this point socket cannot receive new packets, but it is possible 2725 * that some packets are in flight because some CPU runs receiver and 2726 * did hash table lookup before we unhashed socket. They will achieve 2727 * receive queue and will be purged by socket destructor. 2728 * 2729 * Also we still have packets pending on receive queue and probably, 2730 * our own packets waiting in device queues. sock_destroy will drain 2731 * receive queue, but transmitted packets will delay socket destruction 2732 * until the last reference will be released. 2733 */ 2734 2735 sock_orphan(sk); 2736 2737 xfrm_sk_free_policy(sk); 2738 2739 sk_refcnt_debug_release(sk); 2740 2741 if (sk->sk_frag.page) { 2742 put_page(sk->sk_frag.page); 2743 sk->sk_frag.page = NULL; 2744 } 2745 2746 sock_put(sk); 2747 } 2748 EXPORT_SYMBOL(sk_common_release); 2749 2750 #ifdef CONFIG_PROC_FS 2751 #define PROTO_INUSE_NR 64 /* should be enough for the first time */ 2752 struct prot_inuse { 2753 int val[PROTO_INUSE_NR]; 2754 }; 2755 2756 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR); 2757 2758 #ifdef CONFIG_NET_NS 2759 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val) 2760 { 2761 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val); 2762 } 2763 EXPORT_SYMBOL_GPL(sock_prot_inuse_add); 2764 2765 int sock_prot_inuse_get(struct net *net, struct proto *prot) 2766 { 2767 int cpu, idx = prot->inuse_idx; 2768 int res = 0; 2769 2770 for_each_possible_cpu(cpu) 2771 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx]; 2772 2773 return res >= 0 ? res : 0; 2774 } 2775 EXPORT_SYMBOL_GPL(sock_prot_inuse_get); 2776 2777 static int __net_init sock_inuse_init_net(struct net *net) 2778 { 2779 net->core.inuse = alloc_percpu(struct prot_inuse); 2780 return net->core.inuse ? 0 : -ENOMEM; 2781 } 2782 2783 static void __net_exit sock_inuse_exit_net(struct net *net) 2784 { 2785 free_percpu(net->core.inuse); 2786 } 2787 2788 static struct pernet_operations net_inuse_ops = { 2789 .init = sock_inuse_init_net, 2790 .exit = sock_inuse_exit_net, 2791 }; 2792 2793 static __init int net_inuse_init(void) 2794 { 2795 if (register_pernet_subsys(&net_inuse_ops)) 2796 panic("Cannot initialize net inuse counters"); 2797 2798 return 0; 2799 } 2800 2801 core_initcall(net_inuse_init); 2802 #else 2803 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse); 2804 2805 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val) 2806 { 2807 __this_cpu_add(prot_inuse.val[prot->inuse_idx], val); 2808 } 2809 EXPORT_SYMBOL_GPL(sock_prot_inuse_add); 2810 2811 int sock_prot_inuse_get(struct net *net, struct proto *prot) 2812 { 2813 int cpu, idx = prot->inuse_idx; 2814 int res = 0; 2815 2816 for_each_possible_cpu(cpu) 2817 res += per_cpu(prot_inuse, cpu).val[idx]; 2818 2819 return res >= 0 ? res : 0; 2820 } 2821 EXPORT_SYMBOL_GPL(sock_prot_inuse_get); 2822 #endif 2823 2824 static void assign_proto_idx(struct proto *prot) 2825 { 2826 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR); 2827 2828 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) { 2829 pr_err("PROTO_INUSE_NR exhausted\n"); 2830 return; 2831 } 2832 2833 set_bit(prot->inuse_idx, proto_inuse_idx); 2834 } 2835 2836 static void release_proto_idx(struct proto *prot) 2837 { 2838 if (prot->inuse_idx != PROTO_INUSE_NR - 1) 2839 clear_bit(prot->inuse_idx, proto_inuse_idx); 2840 } 2841 #else 2842 static inline void assign_proto_idx(struct proto *prot) 2843 { 2844 } 2845 2846 static inline void release_proto_idx(struct proto *prot) 2847 { 2848 } 2849 #endif 2850 2851 static void req_prot_cleanup(struct request_sock_ops *rsk_prot) 2852 { 2853 if (!rsk_prot) 2854 return; 2855 kfree(rsk_prot->slab_name); 2856 rsk_prot->slab_name = NULL; 2857 kmem_cache_destroy(rsk_prot->slab); 2858 rsk_prot->slab = NULL; 2859 } 2860 2861 static int req_prot_init(const struct proto *prot) 2862 { 2863 struct request_sock_ops *rsk_prot = prot->rsk_prot; 2864 2865 if (!rsk_prot) 2866 return 0; 2867 2868 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", 2869 prot->name); 2870 if (!rsk_prot->slab_name) 2871 return -ENOMEM; 2872 2873 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name, 2874 rsk_prot->obj_size, 0, 2875 prot->slab_flags, NULL); 2876 2877 if (!rsk_prot->slab) { 2878 pr_crit("%s: Can't create request sock SLAB cache!\n", 2879 prot->name); 2880 return -ENOMEM; 2881 } 2882 return 0; 2883 } 2884 2885 int proto_register(struct proto *prot, int alloc_slab) 2886 { 2887 if (alloc_slab) { 2888 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0, 2889 SLAB_HWCACHE_ALIGN | prot->slab_flags, 2890 NULL); 2891 2892 if (prot->slab == NULL) { 2893 pr_crit("%s: Can't create sock SLAB cache!\n", 2894 prot->name); 2895 goto out; 2896 } 2897 2898 if (req_prot_init(prot)) 2899 goto out_free_request_sock_slab; 2900 2901 if (prot->twsk_prot != NULL) { 2902 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name); 2903 2904 if (prot->twsk_prot->twsk_slab_name == NULL) 2905 goto out_free_request_sock_slab; 2906 2907 prot->twsk_prot->twsk_slab = 2908 kmem_cache_create(prot->twsk_prot->twsk_slab_name, 2909 prot->twsk_prot->twsk_obj_size, 2910 0, 2911 prot->slab_flags, 2912 NULL); 2913 if (prot->twsk_prot->twsk_slab == NULL) 2914 goto out_free_timewait_sock_slab_name; 2915 } 2916 } 2917 2918 mutex_lock(&proto_list_mutex); 2919 list_add(&prot->node, &proto_list); 2920 assign_proto_idx(prot); 2921 mutex_unlock(&proto_list_mutex); 2922 return 0; 2923 2924 out_free_timewait_sock_slab_name: 2925 kfree(prot->twsk_prot->twsk_slab_name); 2926 out_free_request_sock_slab: 2927 req_prot_cleanup(prot->rsk_prot); 2928 2929 kmem_cache_destroy(prot->slab); 2930 prot->slab = NULL; 2931 out: 2932 return -ENOBUFS; 2933 } 2934 EXPORT_SYMBOL(proto_register); 2935 2936 void proto_unregister(struct proto *prot) 2937 { 2938 mutex_lock(&proto_list_mutex); 2939 release_proto_idx(prot); 2940 list_del(&prot->node); 2941 mutex_unlock(&proto_list_mutex); 2942 2943 kmem_cache_destroy(prot->slab); 2944 prot->slab = NULL; 2945 2946 req_prot_cleanup(prot->rsk_prot); 2947 2948 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) { 2949 kmem_cache_destroy(prot->twsk_prot->twsk_slab); 2950 kfree(prot->twsk_prot->twsk_slab_name); 2951 prot->twsk_prot->twsk_slab = NULL; 2952 } 2953 } 2954 EXPORT_SYMBOL(proto_unregister); 2955 2956 #ifdef CONFIG_PROC_FS 2957 static void *proto_seq_start(struct seq_file *seq, loff_t *pos) 2958 __acquires(proto_list_mutex) 2959 { 2960 mutex_lock(&proto_list_mutex); 2961 return seq_list_start_head(&proto_list, *pos); 2962 } 2963 2964 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos) 2965 { 2966 return seq_list_next(v, &proto_list, pos); 2967 } 2968 2969 static void proto_seq_stop(struct seq_file *seq, void *v) 2970 __releases(proto_list_mutex) 2971 { 2972 mutex_unlock(&proto_list_mutex); 2973 } 2974 2975 static char proto_method_implemented(const void *method) 2976 { 2977 return method == NULL ? 'n' : 'y'; 2978 } 2979 static long sock_prot_memory_allocated(struct proto *proto) 2980 { 2981 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L; 2982 } 2983 2984 static char *sock_prot_memory_pressure(struct proto *proto) 2985 { 2986 return proto->memory_pressure != NULL ? 2987 proto_memory_pressure(proto) ? "yes" : "no" : "NI"; 2988 } 2989 2990 static void proto_seq_printf(struct seq_file *seq, struct proto *proto) 2991 { 2992 2993 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s " 2994 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n", 2995 proto->name, 2996 proto->obj_size, 2997 sock_prot_inuse_get(seq_file_net(seq), proto), 2998 sock_prot_memory_allocated(proto), 2999 sock_prot_memory_pressure(proto), 3000 proto->max_header, 3001 proto->slab == NULL ? "no" : "yes", 3002 module_name(proto->owner), 3003 proto_method_implemented(proto->close), 3004 proto_method_implemented(proto->connect), 3005 proto_method_implemented(proto->disconnect), 3006 proto_method_implemented(proto->accept), 3007 proto_method_implemented(proto->ioctl), 3008 proto_method_implemented(proto->init), 3009 proto_method_implemented(proto->destroy), 3010 proto_method_implemented(proto->shutdown), 3011 proto_method_implemented(proto->setsockopt), 3012 proto_method_implemented(proto->getsockopt), 3013 proto_method_implemented(proto->sendmsg), 3014 proto_method_implemented(proto->recvmsg), 3015 proto_method_implemented(proto->sendpage), 3016 proto_method_implemented(proto->bind), 3017 proto_method_implemented(proto->backlog_rcv), 3018 proto_method_implemented(proto->hash), 3019 proto_method_implemented(proto->unhash), 3020 proto_method_implemented(proto->get_port), 3021 proto_method_implemented(proto->enter_memory_pressure)); 3022 } 3023 3024 static int proto_seq_show(struct seq_file *seq, void *v) 3025 { 3026 if (v == &proto_list) 3027 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s", 3028 "protocol", 3029 "size", 3030 "sockets", 3031 "memory", 3032 "press", 3033 "maxhdr", 3034 "slab", 3035 "module", 3036 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n"); 3037 else 3038 proto_seq_printf(seq, list_entry(v, struct proto, node)); 3039 return 0; 3040 } 3041 3042 static const struct seq_operations proto_seq_ops = { 3043 .start = proto_seq_start, 3044 .next = proto_seq_next, 3045 .stop = proto_seq_stop, 3046 .show = proto_seq_show, 3047 }; 3048 3049 static int proto_seq_open(struct inode *inode, struct file *file) 3050 { 3051 return seq_open_net(inode, file, &proto_seq_ops, 3052 sizeof(struct seq_net_private)); 3053 } 3054 3055 static const struct file_operations proto_seq_fops = { 3056 .owner = THIS_MODULE, 3057 .open = proto_seq_open, 3058 .read = seq_read, 3059 .llseek = seq_lseek, 3060 .release = seq_release_net, 3061 }; 3062 3063 static __net_init int proto_init_net(struct net *net) 3064 { 3065 if (!proc_create("protocols", S_IRUGO, net->proc_net, &proto_seq_fops)) 3066 return -ENOMEM; 3067 3068 return 0; 3069 } 3070 3071 static __net_exit void proto_exit_net(struct net *net) 3072 { 3073 remove_proc_entry("protocols", net->proc_net); 3074 } 3075 3076 3077 static __net_initdata struct pernet_operations proto_net_ops = { 3078 .init = proto_init_net, 3079 .exit = proto_exit_net, 3080 }; 3081 3082 static int __init proto_init(void) 3083 { 3084 return register_pernet_subsys(&proto_net_ops); 3085 } 3086 3087 subsys_initcall(proto_init); 3088 3089 #endif /* PROC_FS */ 3090