1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * INET An implementation of the TCP/IP protocol suite for the LINUX 4 * operating system. INET is implemented using the BSD Socket 5 * interface as the means of communication with the user level. 6 * 7 * Generic socket support routines. Memory allocators, socket lock/release 8 * handler for protocols to use and generic option handler. 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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 87 88 #include <asm/unaligned.h> 89 #include <linux/capability.h> 90 #include <linux/errno.h> 91 #include <linux/errqueue.h> 92 #include <linux/types.h> 93 #include <linux/socket.h> 94 #include <linux/in.h> 95 #include <linux/kernel.h> 96 #include <linux/module.h> 97 #include <linux/proc_fs.h> 98 #include <linux/seq_file.h> 99 #include <linux/sched.h> 100 #include <linux/sched/mm.h> 101 #include <linux/timer.h> 102 #include <linux/string.h> 103 #include <linux/sockios.h> 104 #include <linux/net.h> 105 #include <linux/mm.h> 106 #include <linux/slab.h> 107 #include <linux/interrupt.h> 108 #include <linux/poll.h> 109 #include <linux/tcp.h> 110 #include <linux/init.h> 111 #include <linux/highmem.h> 112 #include <linux/user_namespace.h> 113 #include <linux/static_key.h> 114 #include <linux/memcontrol.h> 115 #include <linux/prefetch.h> 116 #include <linux/compat.h> 117 118 #include <linux/uaccess.h> 119 120 #include <linux/netdevice.h> 121 #include <net/protocol.h> 122 #include <linux/skbuff.h> 123 #include <net/net_namespace.h> 124 #include <net/request_sock.h> 125 #include <net/sock.h> 126 #include <linux/net_tstamp.h> 127 #include <net/xfrm.h> 128 #include <linux/ipsec.h> 129 #include <net/cls_cgroup.h> 130 #include <net/netprio_cgroup.h> 131 #include <linux/sock_diag.h> 132 133 #include <linux/filter.h> 134 #include <net/sock_reuseport.h> 135 #include <net/bpf_sk_storage.h> 136 137 #include <trace/events/sock.h> 138 139 #include <net/tcp.h> 140 #include <net/busy_poll.h> 141 142 #include <linux/ethtool.h> 143 144 static DEFINE_MUTEX(proto_list_mutex); 145 static LIST_HEAD(proto_list); 146 147 /** 148 * sk_ns_capable - General socket capability test 149 * @sk: Socket to use a capability on or through 150 * @user_ns: The user namespace of the capability to use 151 * @cap: The capability to use 152 * 153 * Test to see if the opener of the socket had when the socket was 154 * created and the current process has the capability @cap in the user 155 * namespace @user_ns. 156 */ 157 bool sk_ns_capable(const struct sock *sk, 158 struct user_namespace *user_ns, int cap) 159 { 160 return file_ns_capable(sk->sk_socket->file, user_ns, cap) && 161 ns_capable(user_ns, cap); 162 } 163 EXPORT_SYMBOL(sk_ns_capable); 164 165 /** 166 * sk_capable - Socket global capability test 167 * @sk: Socket to use a capability on or through 168 * @cap: The global capability to use 169 * 170 * Test to see if the opener of the socket had when the socket was 171 * created and the current process has the capability @cap in all user 172 * namespaces. 173 */ 174 bool sk_capable(const struct sock *sk, int cap) 175 { 176 return sk_ns_capable(sk, &init_user_ns, cap); 177 } 178 EXPORT_SYMBOL(sk_capable); 179 180 /** 181 * sk_net_capable - Network namespace socket capability test 182 * @sk: Socket to use a capability on or through 183 * @cap: The capability to use 184 * 185 * Test to see if the opener of the socket had when the socket was created 186 * and the current process has the capability @cap over the network namespace 187 * the socket is a member of. 188 */ 189 bool sk_net_capable(const struct sock *sk, int cap) 190 { 191 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap); 192 } 193 EXPORT_SYMBOL(sk_net_capable); 194 195 /* 196 * Each address family might have different locking rules, so we have 197 * one slock key per address family and separate keys for internal and 198 * userspace sockets. 199 */ 200 static struct lock_class_key af_family_keys[AF_MAX]; 201 static struct lock_class_key af_family_kern_keys[AF_MAX]; 202 static struct lock_class_key af_family_slock_keys[AF_MAX]; 203 static struct lock_class_key af_family_kern_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 */ 210 211 #define _sock_locks(x) \ 212 x "AF_UNSPEC", x "AF_UNIX" , x "AF_INET" , \ 213 x "AF_AX25" , x "AF_IPX" , x "AF_APPLETALK", \ 214 x "AF_NETROM", x "AF_BRIDGE" , x "AF_ATMPVC" , \ 215 x "AF_X25" , x "AF_INET6" , x "AF_ROSE" , \ 216 x "AF_DECnet", x "AF_NETBEUI" , x "AF_SECURITY" , \ 217 x "AF_KEY" , x "AF_NETLINK" , x "AF_PACKET" , \ 218 x "AF_ASH" , x "AF_ECONET" , x "AF_ATMSVC" , \ 219 x "AF_RDS" , x "AF_SNA" , x "AF_IRDA" , \ 220 x "AF_PPPOX" , x "AF_WANPIPE" , x "AF_LLC" , \ 221 x "27" , x "28" , x "AF_CAN" , \ 222 x "AF_TIPC" , x "AF_BLUETOOTH", x "IUCV" , \ 223 x "AF_RXRPC" , x "AF_ISDN" , x "AF_PHONET" , \ 224 x "AF_IEEE802154", x "AF_CAIF" , x "AF_ALG" , \ 225 x "AF_NFC" , x "AF_VSOCK" , x "AF_KCM" , \ 226 x "AF_QIPCRTR", x "AF_SMC" , x "AF_XDP" , \ 227 x "AF_MCTP" , \ 228 x "AF_MAX" 229 230 static const char *const af_family_key_strings[AF_MAX+1] = { 231 _sock_locks("sk_lock-") 232 }; 233 static const char *const af_family_slock_key_strings[AF_MAX+1] = { 234 _sock_locks("slock-") 235 }; 236 static const char *const af_family_clock_key_strings[AF_MAX+1] = { 237 _sock_locks("clock-") 238 }; 239 240 static const char *const af_family_kern_key_strings[AF_MAX+1] = { 241 _sock_locks("k-sk_lock-") 242 }; 243 static const char *const af_family_kern_slock_key_strings[AF_MAX+1] = { 244 _sock_locks("k-slock-") 245 }; 246 static const char *const af_family_kern_clock_key_strings[AF_MAX+1] = { 247 _sock_locks("k-clock-") 248 }; 249 static const char *const af_family_rlock_key_strings[AF_MAX+1] = { 250 _sock_locks("rlock-") 251 }; 252 static const char *const af_family_wlock_key_strings[AF_MAX+1] = { 253 _sock_locks("wlock-") 254 }; 255 static const char *const af_family_elock_key_strings[AF_MAX+1] = { 256 _sock_locks("elock-") 257 }; 258 259 /* 260 * sk_callback_lock and sk queues locking rules are per-address-family, 261 * so split the lock classes by using a per-AF key: 262 */ 263 static struct lock_class_key af_callback_keys[AF_MAX]; 264 static struct lock_class_key af_rlock_keys[AF_MAX]; 265 static struct lock_class_key af_wlock_keys[AF_MAX]; 266 static struct lock_class_key af_elock_keys[AF_MAX]; 267 static struct lock_class_key af_kern_callback_keys[AF_MAX]; 268 269 /* Run time adjustable parameters. */ 270 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX; 271 EXPORT_SYMBOL(sysctl_wmem_max); 272 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX; 273 EXPORT_SYMBOL(sysctl_rmem_max); 274 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX; 275 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX; 276 277 /* Maximal space eaten by iovec or ancillary data plus some space */ 278 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512); 279 EXPORT_SYMBOL(sysctl_optmem_max); 280 281 int sysctl_tstamp_allow_data __read_mostly = 1; 282 283 DEFINE_STATIC_KEY_FALSE(memalloc_socks_key); 284 EXPORT_SYMBOL_GPL(memalloc_socks_key); 285 286 /** 287 * sk_set_memalloc - sets %SOCK_MEMALLOC 288 * @sk: socket to set it on 289 * 290 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves. 291 * It's the responsibility of the admin to adjust min_free_kbytes 292 * to meet the requirements 293 */ 294 void sk_set_memalloc(struct sock *sk) 295 { 296 sock_set_flag(sk, SOCK_MEMALLOC); 297 sk->sk_allocation |= __GFP_MEMALLOC; 298 static_branch_inc(&memalloc_socks_key); 299 } 300 EXPORT_SYMBOL_GPL(sk_set_memalloc); 301 302 void sk_clear_memalloc(struct sock *sk) 303 { 304 sock_reset_flag(sk, SOCK_MEMALLOC); 305 sk->sk_allocation &= ~__GFP_MEMALLOC; 306 static_branch_dec(&memalloc_socks_key); 307 308 /* 309 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward 310 * progress of swapping. SOCK_MEMALLOC may be cleared while 311 * it has rmem allocations due to the last swapfile being deactivated 312 * but there is a risk that the socket is unusable due to exceeding 313 * the rmem limits. Reclaim the reserves and obey rmem limits again. 314 */ 315 sk_mem_reclaim(sk); 316 } 317 EXPORT_SYMBOL_GPL(sk_clear_memalloc); 318 319 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb) 320 { 321 int ret; 322 unsigned int noreclaim_flag; 323 324 /* these should have been dropped before queueing */ 325 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC)); 326 327 noreclaim_flag = memalloc_noreclaim_save(); 328 ret = INDIRECT_CALL_INET(sk->sk_backlog_rcv, 329 tcp_v6_do_rcv, 330 tcp_v4_do_rcv, 331 sk, skb); 332 memalloc_noreclaim_restore(noreclaim_flag); 333 334 return ret; 335 } 336 EXPORT_SYMBOL(__sk_backlog_rcv); 337 338 void sk_error_report(struct sock *sk) 339 { 340 sk->sk_error_report(sk); 341 342 switch (sk->sk_family) { 343 case AF_INET: 344 fallthrough; 345 case AF_INET6: 346 trace_inet_sk_error_report(sk); 347 break; 348 default: 349 break; 350 } 351 } 352 EXPORT_SYMBOL(sk_error_report); 353 354 int sock_get_timeout(long timeo, void *optval, bool old_timeval) 355 { 356 struct __kernel_sock_timeval tv; 357 358 if (timeo == MAX_SCHEDULE_TIMEOUT) { 359 tv.tv_sec = 0; 360 tv.tv_usec = 0; 361 } else { 362 tv.tv_sec = timeo / HZ; 363 tv.tv_usec = ((timeo % HZ) * USEC_PER_SEC) / HZ; 364 } 365 366 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) { 367 struct old_timeval32 tv32 = { tv.tv_sec, tv.tv_usec }; 368 *(struct old_timeval32 *)optval = tv32; 369 return sizeof(tv32); 370 } 371 372 if (old_timeval) { 373 struct __kernel_old_timeval old_tv; 374 old_tv.tv_sec = tv.tv_sec; 375 old_tv.tv_usec = tv.tv_usec; 376 *(struct __kernel_old_timeval *)optval = old_tv; 377 return sizeof(old_tv); 378 } 379 380 *(struct __kernel_sock_timeval *)optval = tv; 381 return sizeof(tv); 382 } 383 EXPORT_SYMBOL(sock_get_timeout); 384 385 int sock_copy_user_timeval(struct __kernel_sock_timeval *tv, 386 sockptr_t optval, int optlen, bool old_timeval) 387 { 388 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) { 389 struct old_timeval32 tv32; 390 391 if (optlen < sizeof(tv32)) 392 return -EINVAL; 393 394 if (copy_from_sockptr(&tv32, optval, sizeof(tv32))) 395 return -EFAULT; 396 tv->tv_sec = tv32.tv_sec; 397 tv->tv_usec = tv32.tv_usec; 398 } else if (old_timeval) { 399 struct __kernel_old_timeval old_tv; 400 401 if (optlen < sizeof(old_tv)) 402 return -EINVAL; 403 if (copy_from_sockptr(&old_tv, optval, sizeof(old_tv))) 404 return -EFAULT; 405 tv->tv_sec = old_tv.tv_sec; 406 tv->tv_usec = old_tv.tv_usec; 407 } else { 408 if (optlen < sizeof(*tv)) 409 return -EINVAL; 410 if (copy_from_sockptr(tv, optval, sizeof(*tv))) 411 return -EFAULT; 412 } 413 414 return 0; 415 } 416 EXPORT_SYMBOL(sock_copy_user_timeval); 417 418 static int sock_set_timeout(long *timeo_p, sockptr_t optval, int optlen, 419 bool old_timeval) 420 { 421 struct __kernel_sock_timeval tv; 422 int err = sock_copy_user_timeval(&tv, optval, optlen, old_timeval); 423 424 if (err) 425 return err; 426 427 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC) 428 return -EDOM; 429 430 if (tv.tv_sec < 0) { 431 static int warned __read_mostly; 432 433 *timeo_p = 0; 434 if (warned < 10 && net_ratelimit()) { 435 warned++; 436 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n", 437 __func__, current->comm, task_pid_nr(current)); 438 } 439 return 0; 440 } 441 *timeo_p = MAX_SCHEDULE_TIMEOUT; 442 if (tv.tv_sec == 0 && tv.tv_usec == 0) 443 return 0; 444 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1)) 445 *timeo_p = tv.tv_sec * HZ + DIV_ROUND_UP((unsigned long)tv.tv_usec, USEC_PER_SEC / HZ); 446 return 0; 447 } 448 449 static bool sock_needs_netstamp(const struct sock *sk) 450 { 451 switch (sk->sk_family) { 452 case AF_UNSPEC: 453 case AF_UNIX: 454 return false; 455 default: 456 return true; 457 } 458 } 459 460 static void sock_disable_timestamp(struct sock *sk, unsigned long flags) 461 { 462 if (sk->sk_flags & flags) { 463 sk->sk_flags &= ~flags; 464 if (sock_needs_netstamp(sk) && 465 !(sk->sk_flags & SK_FLAGS_TIMESTAMP)) 466 net_disable_timestamp(); 467 } 468 } 469 470 471 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 472 { 473 unsigned long flags; 474 struct sk_buff_head *list = &sk->sk_receive_queue; 475 476 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) { 477 atomic_inc(&sk->sk_drops); 478 trace_sock_rcvqueue_full(sk, skb); 479 return -ENOMEM; 480 } 481 482 if (!sk_rmem_schedule(sk, skb, skb->truesize)) { 483 atomic_inc(&sk->sk_drops); 484 return -ENOBUFS; 485 } 486 487 skb->dev = NULL; 488 skb_set_owner_r(skb, sk); 489 490 /* we escape from rcu protected region, make sure we dont leak 491 * a norefcounted dst 492 */ 493 skb_dst_force(skb); 494 495 spin_lock_irqsave(&list->lock, flags); 496 sock_skb_set_dropcount(sk, skb); 497 __skb_queue_tail(list, skb); 498 spin_unlock_irqrestore(&list->lock, flags); 499 500 if (!sock_flag(sk, SOCK_DEAD)) 501 sk->sk_data_ready(sk); 502 return 0; 503 } 504 EXPORT_SYMBOL(__sock_queue_rcv_skb); 505 506 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 507 { 508 int err; 509 510 err = sk_filter(sk, skb); 511 if (err) 512 return err; 513 514 return __sock_queue_rcv_skb(sk, skb); 515 } 516 EXPORT_SYMBOL(sock_queue_rcv_skb); 517 518 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, 519 const int nested, unsigned int trim_cap, bool refcounted) 520 { 521 int rc = NET_RX_SUCCESS; 522 523 if (sk_filter_trim_cap(sk, skb, trim_cap)) 524 goto discard_and_relse; 525 526 skb->dev = NULL; 527 528 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) { 529 atomic_inc(&sk->sk_drops); 530 goto discard_and_relse; 531 } 532 if (nested) 533 bh_lock_sock_nested(sk); 534 else 535 bh_lock_sock(sk); 536 if (!sock_owned_by_user(sk)) { 537 /* 538 * trylock + unlock semantics: 539 */ 540 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_); 541 542 rc = sk_backlog_rcv(sk, skb); 543 544 mutex_release(&sk->sk_lock.dep_map, _RET_IP_); 545 } else if (sk_add_backlog(sk, skb, READ_ONCE(sk->sk_rcvbuf))) { 546 bh_unlock_sock(sk); 547 atomic_inc(&sk->sk_drops); 548 goto discard_and_relse; 549 } 550 551 bh_unlock_sock(sk); 552 out: 553 if (refcounted) 554 sock_put(sk); 555 return rc; 556 discard_and_relse: 557 kfree_skb(skb); 558 goto out; 559 } 560 EXPORT_SYMBOL(__sk_receive_skb); 561 562 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ip6_dst_check(struct dst_entry *, 563 u32)); 564 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ipv4_dst_check(struct dst_entry *, 565 u32)); 566 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie) 567 { 568 struct dst_entry *dst = __sk_dst_get(sk); 569 570 if (dst && dst->obsolete && 571 INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check, 572 dst, cookie) == NULL) { 573 sk_tx_queue_clear(sk); 574 sk->sk_dst_pending_confirm = 0; 575 RCU_INIT_POINTER(sk->sk_dst_cache, NULL); 576 dst_release(dst); 577 return NULL; 578 } 579 580 return dst; 581 } 582 EXPORT_SYMBOL(__sk_dst_check); 583 584 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie) 585 { 586 struct dst_entry *dst = sk_dst_get(sk); 587 588 if (dst && dst->obsolete && 589 INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check, 590 dst, cookie) == NULL) { 591 sk_dst_reset(sk); 592 dst_release(dst); 593 return NULL; 594 } 595 596 return dst; 597 } 598 EXPORT_SYMBOL(sk_dst_check); 599 600 static int sock_bindtoindex_locked(struct sock *sk, int ifindex) 601 { 602 int ret = -ENOPROTOOPT; 603 #ifdef CONFIG_NETDEVICES 604 struct net *net = sock_net(sk); 605 606 /* Sorry... */ 607 ret = -EPERM; 608 if (sk->sk_bound_dev_if && !ns_capable(net->user_ns, CAP_NET_RAW)) 609 goto out; 610 611 ret = -EINVAL; 612 if (ifindex < 0) 613 goto out; 614 615 sk->sk_bound_dev_if = ifindex; 616 if (sk->sk_prot->rehash) 617 sk->sk_prot->rehash(sk); 618 sk_dst_reset(sk); 619 620 ret = 0; 621 622 out: 623 #endif 624 625 return ret; 626 } 627 628 int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk) 629 { 630 int ret; 631 632 if (lock_sk) 633 lock_sock(sk); 634 ret = sock_bindtoindex_locked(sk, ifindex); 635 if (lock_sk) 636 release_sock(sk); 637 638 return ret; 639 } 640 EXPORT_SYMBOL(sock_bindtoindex); 641 642 static int sock_setbindtodevice(struct sock *sk, sockptr_t optval, int optlen) 643 { 644 int ret = -ENOPROTOOPT; 645 #ifdef CONFIG_NETDEVICES 646 struct net *net = sock_net(sk); 647 char devname[IFNAMSIZ]; 648 int index; 649 650 ret = -EINVAL; 651 if (optlen < 0) 652 goto out; 653 654 /* Bind this socket to a particular device like "eth0", 655 * as specified in the passed interface name. If the 656 * name is "" or the option length is zero the socket 657 * is not bound. 658 */ 659 if (optlen > IFNAMSIZ - 1) 660 optlen = IFNAMSIZ - 1; 661 memset(devname, 0, sizeof(devname)); 662 663 ret = -EFAULT; 664 if (copy_from_sockptr(devname, optval, optlen)) 665 goto out; 666 667 index = 0; 668 if (devname[0] != '\0') { 669 struct net_device *dev; 670 671 rcu_read_lock(); 672 dev = dev_get_by_name_rcu(net, devname); 673 if (dev) 674 index = dev->ifindex; 675 rcu_read_unlock(); 676 ret = -ENODEV; 677 if (!dev) 678 goto out; 679 } 680 681 return sock_bindtoindex(sk, index, true); 682 out: 683 #endif 684 685 return ret; 686 } 687 688 static int sock_getbindtodevice(struct sock *sk, char __user *optval, 689 int __user *optlen, int len) 690 { 691 int ret = -ENOPROTOOPT; 692 #ifdef CONFIG_NETDEVICES 693 struct net *net = sock_net(sk); 694 char devname[IFNAMSIZ]; 695 696 if (sk->sk_bound_dev_if == 0) { 697 len = 0; 698 goto zero; 699 } 700 701 ret = -EINVAL; 702 if (len < IFNAMSIZ) 703 goto out; 704 705 ret = netdev_get_name(net, devname, sk->sk_bound_dev_if); 706 if (ret) 707 goto out; 708 709 len = strlen(devname) + 1; 710 711 ret = -EFAULT; 712 if (copy_to_user(optval, devname, len)) 713 goto out; 714 715 zero: 716 ret = -EFAULT; 717 if (put_user(len, optlen)) 718 goto out; 719 720 ret = 0; 721 722 out: 723 #endif 724 725 return ret; 726 } 727 728 bool sk_mc_loop(struct sock *sk) 729 { 730 if (dev_recursion_level()) 731 return false; 732 if (!sk) 733 return true; 734 switch (sk->sk_family) { 735 case AF_INET: 736 return inet_sk(sk)->mc_loop; 737 #if IS_ENABLED(CONFIG_IPV6) 738 case AF_INET6: 739 return inet6_sk(sk)->mc_loop; 740 #endif 741 } 742 WARN_ON_ONCE(1); 743 return true; 744 } 745 EXPORT_SYMBOL(sk_mc_loop); 746 747 void sock_set_reuseaddr(struct sock *sk) 748 { 749 lock_sock(sk); 750 sk->sk_reuse = SK_CAN_REUSE; 751 release_sock(sk); 752 } 753 EXPORT_SYMBOL(sock_set_reuseaddr); 754 755 void sock_set_reuseport(struct sock *sk) 756 { 757 lock_sock(sk); 758 sk->sk_reuseport = true; 759 release_sock(sk); 760 } 761 EXPORT_SYMBOL(sock_set_reuseport); 762 763 void sock_no_linger(struct sock *sk) 764 { 765 lock_sock(sk); 766 sk->sk_lingertime = 0; 767 sock_set_flag(sk, SOCK_LINGER); 768 release_sock(sk); 769 } 770 EXPORT_SYMBOL(sock_no_linger); 771 772 void sock_set_priority(struct sock *sk, u32 priority) 773 { 774 lock_sock(sk); 775 sk->sk_priority = priority; 776 release_sock(sk); 777 } 778 EXPORT_SYMBOL(sock_set_priority); 779 780 void sock_set_sndtimeo(struct sock *sk, s64 secs) 781 { 782 lock_sock(sk); 783 if (secs && secs < MAX_SCHEDULE_TIMEOUT / HZ - 1) 784 sk->sk_sndtimeo = secs * HZ; 785 else 786 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT; 787 release_sock(sk); 788 } 789 EXPORT_SYMBOL(sock_set_sndtimeo); 790 791 static void __sock_set_timestamps(struct sock *sk, bool val, bool new, bool ns) 792 { 793 if (val) { 794 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, new); 795 sock_valbool_flag(sk, SOCK_RCVTSTAMPNS, ns); 796 sock_set_flag(sk, SOCK_RCVTSTAMP); 797 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 798 } else { 799 sock_reset_flag(sk, SOCK_RCVTSTAMP); 800 sock_reset_flag(sk, SOCK_RCVTSTAMPNS); 801 } 802 } 803 804 void sock_enable_timestamps(struct sock *sk) 805 { 806 lock_sock(sk); 807 __sock_set_timestamps(sk, true, false, true); 808 release_sock(sk); 809 } 810 EXPORT_SYMBOL(sock_enable_timestamps); 811 812 void sock_set_timestamp(struct sock *sk, int optname, bool valbool) 813 { 814 switch (optname) { 815 case SO_TIMESTAMP_OLD: 816 __sock_set_timestamps(sk, valbool, false, false); 817 break; 818 case SO_TIMESTAMP_NEW: 819 __sock_set_timestamps(sk, valbool, true, false); 820 break; 821 case SO_TIMESTAMPNS_OLD: 822 __sock_set_timestamps(sk, valbool, false, true); 823 break; 824 case SO_TIMESTAMPNS_NEW: 825 __sock_set_timestamps(sk, valbool, true, true); 826 break; 827 } 828 } 829 830 static int sock_timestamping_bind_phc(struct sock *sk, int phc_index) 831 { 832 struct net *net = sock_net(sk); 833 struct net_device *dev = NULL; 834 bool match = false; 835 int *vclock_index; 836 int i, num; 837 838 if (sk->sk_bound_dev_if) 839 dev = dev_get_by_index(net, sk->sk_bound_dev_if); 840 841 if (!dev) { 842 pr_err("%s: sock not bind to device\n", __func__); 843 return -EOPNOTSUPP; 844 } 845 846 num = ethtool_get_phc_vclocks(dev, &vclock_index); 847 for (i = 0; i < num; i++) { 848 if (*(vclock_index + i) == phc_index) { 849 match = true; 850 break; 851 } 852 } 853 854 if (num > 0) 855 kfree(vclock_index); 856 857 if (!match) 858 return -EINVAL; 859 860 sk->sk_bind_phc = phc_index; 861 862 return 0; 863 } 864 865 int sock_set_timestamping(struct sock *sk, int optname, 866 struct so_timestamping timestamping) 867 { 868 int val = timestamping.flags; 869 int ret; 870 871 if (val & ~SOF_TIMESTAMPING_MASK) 872 return -EINVAL; 873 874 if (val & SOF_TIMESTAMPING_OPT_ID && 875 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) { 876 if (sk_is_tcp(sk)) { 877 if ((1 << sk->sk_state) & 878 (TCPF_CLOSE | TCPF_LISTEN)) 879 return -EINVAL; 880 sk->sk_tskey = tcp_sk(sk)->snd_una; 881 } else { 882 sk->sk_tskey = 0; 883 } 884 } 885 886 if (val & SOF_TIMESTAMPING_OPT_STATS && 887 !(val & SOF_TIMESTAMPING_OPT_TSONLY)) 888 return -EINVAL; 889 890 if (val & SOF_TIMESTAMPING_BIND_PHC) { 891 ret = sock_timestamping_bind_phc(sk, timestamping.bind_phc); 892 if (ret) 893 return ret; 894 } 895 896 sk->sk_tsflags = val; 897 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, optname == SO_TIMESTAMPING_NEW); 898 899 if (val & SOF_TIMESTAMPING_RX_SOFTWARE) 900 sock_enable_timestamp(sk, 901 SOCK_TIMESTAMPING_RX_SOFTWARE); 902 else 903 sock_disable_timestamp(sk, 904 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE)); 905 return 0; 906 } 907 908 void sock_set_keepalive(struct sock *sk) 909 { 910 lock_sock(sk); 911 if (sk->sk_prot->keepalive) 912 sk->sk_prot->keepalive(sk, true); 913 sock_valbool_flag(sk, SOCK_KEEPOPEN, true); 914 release_sock(sk); 915 } 916 EXPORT_SYMBOL(sock_set_keepalive); 917 918 static void __sock_set_rcvbuf(struct sock *sk, int val) 919 { 920 /* Ensure val * 2 fits into an int, to prevent max_t() from treating it 921 * as a negative value. 922 */ 923 val = min_t(int, val, INT_MAX / 2); 924 sk->sk_userlocks |= SOCK_RCVBUF_LOCK; 925 926 /* We double it on the way in to account for "struct sk_buff" etc. 927 * overhead. Applications assume that the SO_RCVBUF setting they make 928 * will allow that much actual data to be received on that socket. 929 * 930 * Applications are unaware that "struct sk_buff" and other overheads 931 * allocate from the receive buffer during socket buffer allocation. 932 * 933 * And after considering the possible alternatives, returning the value 934 * we actually used in getsockopt is the most desirable behavior. 935 */ 936 WRITE_ONCE(sk->sk_rcvbuf, max_t(int, val * 2, SOCK_MIN_RCVBUF)); 937 } 938 939 void sock_set_rcvbuf(struct sock *sk, int val) 940 { 941 lock_sock(sk); 942 __sock_set_rcvbuf(sk, val); 943 release_sock(sk); 944 } 945 EXPORT_SYMBOL(sock_set_rcvbuf); 946 947 static void __sock_set_mark(struct sock *sk, u32 val) 948 { 949 if (val != sk->sk_mark) { 950 sk->sk_mark = val; 951 sk_dst_reset(sk); 952 } 953 } 954 955 void sock_set_mark(struct sock *sk, u32 val) 956 { 957 lock_sock(sk); 958 __sock_set_mark(sk, val); 959 release_sock(sk); 960 } 961 EXPORT_SYMBOL(sock_set_mark); 962 963 static void sock_release_reserved_memory(struct sock *sk, int bytes) 964 { 965 /* Round down bytes to multiple of pages */ 966 bytes &= ~(SK_MEM_QUANTUM - 1); 967 968 WARN_ON(bytes > sk->sk_reserved_mem); 969 sk->sk_reserved_mem -= bytes; 970 sk_mem_reclaim(sk); 971 } 972 973 static int sock_reserve_memory(struct sock *sk, int bytes) 974 { 975 long allocated; 976 bool charged; 977 int pages; 978 979 if (!mem_cgroup_sockets_enabled || !sk->sk_memcg || !sk_has_account(sk)) 980 return -EOPNOTSUPP; 981 982 if (!bytes) 983 return 0; 984 985 pages = sk_mem_pages(bytes); 986 987 /* pre-charge to memcg */ 988 charged = mem_cgroup_charge_skmem(sk->sk_memcg, pages, 989 GFP_KERNEL | __GFP_RETRY_MAYFAIL); 990 if (!charged) 991 return -ENOMEM; 992 993 /* pre-charge to forward_alloc */ 994 allocated = sk_memory_allocated_add(sk, pages); 995 /* If the system goes into memory pressure with this 996 * precharge, give up and return error. 997 */ 998 if (allocated > sk_prot_mem_limits(sk, 1)) { 999 sk_memory_allocated_sub(sk, pages); 1000 mem_cgroup_uncharge_skmem(sk->sk_memcg, pages); 1001 return -ENOMEM; 1002 } 1003 sk->sk_forward_alloc += pages << SK_MEM_QUANTUM_SHIFT; 1004 1005 sk->sk_reserved_mem += pages << SK_MEM_QUANTUM_SHIFT; 1006 1007 return 0; 1008 } 1009 1010 /* 1011 * This is meant for all protocols to use and covers goings on 1012 * at the socket level. Everything here is generic. 1013 */ 1014 1015 int sock_setsockopt(struct socket *sock, int level, int optname, 1016 sockptr_t optval, unsigned int optlen) 1017 { 1018 struct so_timestamping timestamping; 1019 struct sock_txtime sk_txtime; 1020 struct sock *sk = sock->sk; 1021 int val; 1022 int valbool; 1023 struct linger ling; 1024 int ret = 0; 1025 1026 /* 1027 * Options without arguments 1028 */ 1029 1030 if (optname == SO_BINDTODEVICE) 1031 return sock_setbindtodevice(sk, optval, optlen); 1032 1033 if (optlen < sizeof(int)) 1034 return -EINVAL; 1035 1036 if (copy_from_sockptr(&val, optval, sizeof(val))) 1037 return -EFAULT; 1038 1039 valbool = val ? 1 : 0; 1040 1041 lock_sock(sk); 1042 1043 switch (optname) { 1044 case SO_DEBUG: 1045 if (val && !capable(CAP_NET_ADMIN)) 1046 ret = -EACCES; 1047 else 1048 sock_valbool_flag(sk, SOCK_DBG, valbool); 1049 break; 1050 case SO_REUSEADDR: 1051 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE); 1052 break; 1053 case SO_REUSEPORT: 1054 sk->sk_reuseport = valbool; 1055 break; 1056 case SO_TYPE: 1057 case SO_PROTOCOL: 1058 case SO_DOMAIN: 1059 case SO_ERROR: 1060 ret = -ENOPROTOOPT; 1061 break; 1062 case SO_DONTROUTE: 1063 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool); 1064 sk_dst_reset(sk); 1065 break; 1066 case SO_BROADCAST: 1067 sock_valbool_flag(sk, SOCK_BROADCAST, valbool); 1068 break; 1069 case SO_SNDBUF: 1070 /* Don't error on this BSD doesn't and if you think 1071 * about it this is right. Otherwise apps have to 1072 * play 'guess the biggest size' games. RCVBUF/SNDBUF 1073 * are treated in BSD as hints 1074 */ 1075 val = min_t(u32, val, sysctl_wmem_max); 1076 set_sndbuf: 1077 /* Ensure val * 2 fits into an int, to prevent max_t() 1078 * from treating it as a negative value. 1079 */ 1080 val = min_t(int, val, INT_MAX / 2); 1081 sk->sk_userlocks |= SOCK_SNDBUF_LOCK; 1082 WRITE_ONCE(sk->sk_sndbuf, 1083 max_t(int, val * 2, SOCK_MIN_SNDBUF)); 1084 /* Wake up sending tasks if we upped the value. */ 1085 sk->sk_write_space(sk); 1086 break; 1087 1088 case SO_SNDBUFFORCE: 1089 if (!capable(CAP_NET_ADMIN)) { 1090 ret = -EPERM; 1091 break; 1092 } 1093 1094 /* No negative values (to prevent underflow, as val will be 1095 * multiplied by 2). 1096 */ 1097 if (val < 0) 1098 val = 0; 1099 goto set_sndbuf; 1100 1101 case SO_RCVBUF: 1102 /* Don't error on this BSD doesn't and if you think 1103 * about it this is right. Otherwise apps have to 1104 * play 'guess the biggest size' games. RCVBUF/SNDBUF 1105 * are treated in BSD as hints 1106 */ 1107 __sock_set_rcvbuf(sk, min_t(u32, val, sysctl_rmem_max)); 1108 break; 1109 1110 case SO_RCVBUFFORCE: 1111 if (!capable(CAP_NET_ADMIN)) { 1112 ret = -EPERM; 1113 break; 1114 } 1115 1116 /* No negative values (to prevent underflow, as val will be 1117 * multiplied by 2). 1118 */ 1119 __sock_set_rcvbuf(sk, max(val, 0)); 1120 break; 1121 1122 case SO_KEEPALIVE: 1123 if (sk->sk_prot->keepalive) 1124 sk->sk_prot->keepalive(sk, valbool); 1125 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool); 1126 break; 1127 1128 case SO_OOBINLINE: 1129 sock_valbool_flag(sk, SOCK_URGINLINE, valbool); 1130 break; 1131 1132 case SO_NO_CHECK: 1133 sk->sk_no_check_tx = valbool; 1134 break; 1135 1136 case SO_PRIORITY: 1137 if ((val >= 0 && val <= 6) || 1138 ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) 1139 sk->sk_priority = val; 1140 else 1141 ret = -EPERM; 1142 break; 1143 1144 case SO_LINGER: 1145 if (optlen < sizeof(ling)) { 1146 ret = -EINVAL; /* 1003.1g */ 1147 break; 1148 } 1149 if (copy_from_sockptr(&ling, optval, sizeof(ling))) { 1150 ret = -EFAULT; 1151 break; 1152 } 1153 if (!ling.l_onoff) 1154 sock_reset_flag(sk, SOCK_LINGER); 1155 else { 1156 #if (BITS_PER_LONG == 32) 1157 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ) 1158 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT; 1159 else 1160 #endif 1161 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ; 1162 sock_set_flag(sk, SOCK_LINGER); 1163 } 1164 break; 1165 1166 case SO_BSDCOMPAT: 1167 break; 1168 1169 case SO_PASSCRED: 1170 if (valbool) 1171 set_bit(SOCK_PASSCRED, &sock->flags); 1172 else 1173 clear_bit(SOCK_PASSCRED, &sock->flags); 1174 break; 1175 1176 case SO_TIMESTAMP_OLD: 1177 case SO_TIMESTAMP_NEW: 1178 case SO_TIMESTAMPNS_OLD: 1179 case SO_TIMESTAMPNS_NEW: 1180 sock_set_timestamp(sk, optname, valbool); 1181 break; 1182 1183 case SO_TIMESTAMPING_NEW: 1184 case SO_TIMESTAMPING_OLD: 1185 if (optlen == sizeof(timestamping)) { 1186 if (copy_from_sockptr(×tamping, optval, 1187 sizeof(timestamping))) { 1188 ret = -EFAULT; 1189 break; 1190 } 1191 } else { 1192 memset(×tamping, 0, sizeof(timestamping)); 1193 timestamping.flags = val; 1194 } 1195 ret = sock_set_timestamping(sk, optname, timestamping); 1196 break; 1197 1198 case SO_RCVLOWAT: 1199 if (val < 0) 1200 val = INT_MAX; 1201 if (sock->ops->set_rcvlowat) 1202 ret = sock->ops->set_rcvlowat(sk, val); 1203 else 1204 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1); 1205 break; 1206 1207 case SO_RCVTIMEO_OLD: 1208 case SO_RCVTIMEO_NEW: 1209 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, 1210 optlen, optname == SO_RCVTIMEO_OLD); 1211 break; 1212 1213 case SO_SNDTIMEO_OLD: 1214 case SO_SNDTIMEO_NEW: 1215 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, 1216 optlen, optname == SO_SNDTIMEO_OLD); 1217 break; 1218 1219 case SO_ATTACH_FILTER: { 1220 struct sock_fprog fprog; 1221 1222 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen); 1223 if (!ret) 1224 ret = sk_attach_filter(&fprog, sk); 1225 break; 1226 } 1227 case SO_ATTACH_BPF: 1228 ret = -EINVAL; 1229 if (optlen == sizeof(u32)) { 1230 u32 ufd; 1231 1232 ret = -EFAULT; 1233 if (copy_from_sockptr(&ufd, optval, sizeof(ufd))) 1234 break; 1235 1236 ret = sk_attach_bpf(ufd, sk); 1237 } 1238 break; 1239 1240 case SO_ATTACH_REUSEPORT_CBPF: { 1241 struct sock_fprog fprog; 1242 1243 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen); 1244 if (!ret) 1245 ret = sk_reuseport_attach_filter(&fprog, sk); 1246 break; 1247 } 1248 case SO_ATTACH_REUSEPORT_EBPF: 1249 ret = -EINVAL; 1250 if (optlen == sizeof(u32)) { 1251 u32 ufd; 1252 1253 ret = -EFAULT; 1254 if (copy_from_sockptr(&ufd, optval, sizeof(ufd))) 1255 break; 1256 1257 ret = sk_reuseport_attach_bpf(ufd, sk); 1258 } 1259 break; 1260 1261 case SO_DETACH_REUSEPORT_BPF: 1262 ret = reuseport_detach_prog(sk); 1263 break; 1264 1265 case SO_DETACH_FILTER: 1266 ret = sk_detach_filter(sk); 1267 break; 1268 1269 case SO_LOCK_FILTER: 1270 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool) 1271 ret = -EPERM; 1272 else 1273 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool); 1274 break; 1275 1276 case SO_PASSSEC: 1277 if (valbool) 1278 set_bit(SOCK_PASSSEC, &sock->flags); 1279 else 1280 clear_bit(SOCK_PASSSEC, &sock->flags); 1281 break; 1282 case SO_MARK: 1283 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) { 1284 ret = -EPERM; 1285 break; 1286 } 1287 1288 __sock_set_mark(sk, val); 1289 break; 1290 1291 case SO_RXQ_OVFL: 1292 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool); 1293 break; 1294 1295 case SO_WIFI_STATUS: 1296 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool); 1297 break; 1298 1299 case SO_PEEK_OFF: 1300 if (sock->ops->set_peek_off) 1301 ret = sock->ops->set_peek_off(sk, val); 1302 else 1303 ret = -EOPNOTSUPP; 1304 break; 1305 1306 case SO_NOFCS: 1307 sock_valbool_flag(sk, SOCK_NOFCS, valbool); 1308 break; 1309 1310 case SO_SELECT_ERR_QUEUE: 1311 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool); 1312 break; 1313 1314 #ifdef CONFIG_NET_RX_BUSY_POLL 1315 case SO_BUSY_POLL: 1316 /* allow unprivileged users to decrease the value */ 1317 if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN)) 1318 ret = -EPERM; 1319 else { 1320 if (val < 0) 1321 ret = -EINVAL; 1322 else 1323 WRITE_ONCE(sk->sk_ll_usec, val); 1324 } 1325 break; 1326 case SO_PREFER_BUSY_POLL: 1327 if (valbool && !capable(CAP_NET_ADMIN)) 1328 ret = -EPERM; 1329 else 1330 WRITE_ONCE(sk->sk_prefer_busy_poll, valbool); 1331 break; 1332 case SO_BUSY_POLL_BUDGET: 1333 if (val > READ_ONCE(sk->sk_busy_poll_budget) && !capable(CAP_NET_ADMIN)) { 1334 ret = -EPERM; 1335 } else { 1336 if (val < 0 || val > U16_MAX) 1337 ret = -EINVAL; 1338 else 1339 WRITE_ONCE(sk->sk_busy_poll_budget, val); 1340 } 1341 break; 1342 #endif 1343 1344 case SO_MAX_PACING_RATE: 1345 { 1346 unsigned long ulval = (val == ~0U) ? ~0UL : (unsigned int)val; 1347 1348 if (sizeof(ulval) != sizeof(val) && 1349 optlen >= sizeof(ulval) && 1350 copy_from_sockptr(&ulval, optval, sizeof(ulval))) { 1351 ret = -EFAULT; 1352 break; 1353 } 1354 if (ulval != ~0UL) 1355 cmpxchg(&sk->sk_pacing_status, 1356 SK_PACING_NONE, 1357 SK_PACING_NEEDED); 1358 sk->sk_max_pacing_rate = ulval; 1359 sk->sk_pacing_rate = min(sk->sk_pacing_rate, ulval); 1360 break; 1361 } 1362 case SO_INCOMING_CPU: 1363 WRITE_ONCE(sk->sk_incoming_cpu, val); 1364 break; 1365 1366 case SO_CNX_ADVICE: 1367 if (val == 1) 1368 dst_negative_advice(sk); 1369 break; 1370 1371 case SO_ZEROCOPY: 1372 if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) { 1373 if (!(sk_is_tcp(sk) || 1374 (sk->sk_type == SOCK_DGRAM && 1375 sk->sk_protocol == IPPROTO_UDP))) 1376 ret = -ENOTSUPP; 1377 } else if (sk->sk_family != PF_RDS) { 1378 ret = -ENOTSUPP; 1379 } 1380 if (!ret) { 1381 if (val < 0 || val > 1) 1382 ret = -EINVAL; 1383 else 1384 sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool); 1385 } 1386 break; 1387 1388 case SO_TXTIME: 1389 if (optlen != sizeof(struct sock_txtime)) { 1390 ret = -EINVAL; 1391 break; 1392 } else if (copy_from_sockptr(&sk_txtime, optval, 1393 sizeof(struct sock_txtime))) { 1394 ret = -EFAULT; 1395 break; 1396 } else if (sk_txtime.flags & ~SOF_TXTIME_FLAGS_MASK) { 1397 ret = -EINVAL; 1398 break; 1399 } 1400 /* CLOCK_MONOTONIC is only used by sch_fq, and this packet 1401 * scheduler has enough safe guards. 1402 */ 1403 if (sk_txtime.clockid != CLOCK_MONOTONIC && 1404 !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) { 1405 ret = -EPERM; 1406 break; 1407 } 1408 sock_valbool_flag(sk, SOCK_TXTIME, true); 1409 sk->sk_clockid = sk_txtime.clockid; 1410 sk->sk_txtime_deadline_mode = 1411 !!(sk_txtime.flags & SOF_TXTIME_DEADLINE_MODE); 1412 sk->sk_txtime_report_errors = 1413 !!(sk_txtime.flags & SOF_TXTIME_REPORT_ERRORS); 1414 break; 1415 1416 case SO_BINDTOIFINDEX: 1417 ret = sock_bindtoindex_locked(sk, val); 1418 break; 1419 1420 case SO_BUF_LOCK: 1421 if (val & ~SOCK_BUF_LOCK_MASK) { 1422 ret = -EINVAL; 1423 break; 1424 } 1425 sk->sk_userlocks = val | (sk->sk_userlocks & 1426 ~SOCK_BUF_LOCK_MASK); 1427 break; 1428 1429 case SO_RESERVE_MEM: 1430 { 1431 int delta; 1432 1433 if (val < 0) { 1434 ret = -EINVAL; 1435 break; 1436 } 1437 1438 delta = val - sk->sk_reserved_mem; 1439 if (delta < 0) 1440 sock_release_reserved_memory(sk, -delta); 1441 else 1442 ret = sock_reserve_memory(sk, delta); 1443 break; 1444 } 1445 1446 default: 1447 ret = -ENOPROTOOPT; 1448 break; 1449 } 1450 release_sock(sk); 1451 return ret; 1452 } 1453 EXPORT_SYMBOL(sock_setsockopt); 1454 1455 static const struct cred *sk_get_peer_cred(struct sock *sk) 1456 { 1457 const struct cred *cred; 1458 1459 spin_lock(&sk->sk_peer_lock); 1460 cred = get_cred(sk->sk_peer_cred); 1461 spin_unlock(&sk->sk_peer_lock); 1462 1463 return cred; 1464 } 1465 1466 static void cred_to_ucred(struct pid *pid, const struct cred *cred, 1467 struct ucred *ucred) 1468 { 1469 ucred->pid = pid_vnr(pid); 1470 ucred->uid = ucred->gid = -1; 1471 if (cred) { 1472 struct user_namespace *current_ns = current_user_ns(); 1473 1474 ucred->uid = from_kuid_munged(current_ns, cred->euid); 1475 ucred->gid = from_kgid_munged(current_ns, cred->egid); 1476 } 1477 } 1478 1479 static int groups_to_user(gid_t __user *dst, const struct group_info *src) 1480 { 1481 struct user_namespace *user_ns = current_user_ns(); 1482 int i; 1483 1484 for (i = 0; i < src->ngroups; i++) 1485 if (put_user(from_kgid_munged(user_ns, src->gid[i]), dst + i)) 1486 return -EFAULT; 1487 1488 return 0; 1489 } 1490 1491 int sock_getsockopt(struct socket *sock, int level, int optname, 1492 char __user *optval, int __user *optlen) 1493 { 1494 struct sock *sk = sock->sk; 1495 1496 union { 1497 int val; 1498 u64 val64; 1499 unsigned long ulval; 1500 struct linger ling; 1501 struct old_timeval32 tm32; 1502 struct __kernel_old_timeval tm; 1503 struct __kernel_sock_timeval stm; 1504 struct sock_txtime txtime; 1505 struct so_timestamping timestamping; 1506 } v; 1507 1508 int lv = sizeof(int); 1509 int len; 1510 1511 if (get_user(len, optlen)) 1512 return -EFAULT; 1513 if (len < 0) 1514 return -EINVAL; 1515 1516 memset(&v, 0, sizeof(v)); 1517 1518 switch (optname) { 1519 case SO_DEBUG: 1520 v.val = sock_flag(sk, SOCK_DBG); 1521 break; 1522 1523 case SO_DONTROUTE: 1524 v.val = sock_flag(sk, SOCK_LOCALROUTE); 1525 break; 1526 1527 case SO_BROADCAST: 1528 v.val = sock_flag(sk, SOCK_BROADCAST); 1529 break; 1530 1531 case SO_SNDBUF: 1532 v.val = sk->sk_sndbuf; 1533 break; 1534 1535 case SO_RCVBUF: 1536 v.val = sk->sk_rcvbuf; 1537 break; 1538 1539 case SO_REUSEADDR: 1540 v.val = sk->sk_reuse; 1541 break; 1542 1543 case SO_REUSEPORT: 1544 v.val = sk->sk_reuseport; 1545 break; 1546 1547 case SO_KEEPALIVE: 1548 v.val = sock_flag(sk, SOCK_KEEPOPEN); 1549 break; 1550 1551 case SO_TYPE: 1552 v.val = sk->sk_type; 1553 break; 1554 1555 case SO_PROTOCOL: 1556 v.val = sk->sk_protocol; 1557 break; 1558 1559 case SO_DOMAIN: 1560 v.val = sk->sk_family; 1561 break; 1562 1563 case SO_ERROR: 1564 v.val = -sock_error(sk); 1565 if (v.val == 0) 1566 v.val = xchg(&sk->sk_err_soft, 0); 1567 break; 1568 1569 case SO_OOBINLINE: 1570 v.val = sock_flag(sk, SOCK_URGINLINE); 1571 break; 1572 1573 case SO_NO_CHECK: 1574 v.val = sk->sk_no_check_tx; 1575 break; 1576 1577 case SO_PRIORITY: 1578 v.val = sk->sk_priority; 1579 break; 1580 1581 case SO_LINGER: 1582 lv = sizeof(v.ling); 1583 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER); 1584 v.ling.l_linger = sk->sk_lingertime / HZ; 1585 break; 1586 1587 case SO_BSDCOMPAT: 1588 break; 1589 1590 case SO_TIMESTAMP_OLD: 1591 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && 1592 !sock_flag(sk, SOCK_TSTAMP_NEW) && 1593 !sock_flag(sk, SOCK_RCVTSTAMPNS); 1594 break; 1595 1596 case SO_TIMESTAMPNS_OLD: 1597 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && !sock_flag(sk, SOCK_TSTAMP_NEW); 1598 break; 1599 1600 case SO_TIMESTAMP_NEW: 1601 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && sock_flag(sk, SOCK_TSTAMP_NEW); 1602 break; 1603 1604 case SO_TIMESTAMPNS_NEW: 1605 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && sock_flag(sk, SOCK_TSTAMP_NEW); 1606 break; 1607 1608 case SO_TIMESTAMPING_OLD: 1609 lv = sizeof(v.timestamping); 1610 v.timestamping.flags = sk->sk_tsflags; 1611 v.timestamping.bind_phc = sk->sk_bind_phc; 1612 break; 1613 1614 case SO_RCVTIMEO_OLD: 1615 case SO_RCVTIMEO_NEW: 1616 lv = sock_get_timeout(sk->sk_rcvtimeo, &v, SO_RCVTIMEO_OLD == optname); 1617 break; 1618 1619 case SO_SNDTIMEO_OLD: 1620 case SO_SNDTIMEO_NEW: 1621 lv = sock_get_timeout(sk->sk_sndtimeo, &v, SO_SNDTIMEO_OLD == optname); 1622 break; 1623 1624 case SO_RCVLOWAT: 1625 v.val = sk->sk_rcvlowat; 1626 break; 1627 1628 case SO_SNDLOWAT: 1629 v.val = 1; 1630 break; 1631 1632 case SO_PASSCRED: 1633 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags); 1634 break; 1635 1636 case SO_PEERCRED: 1637 { 1638 struct ucred peercred; 1639 if (len > sizeof(peercred)) 1640 len = sizeof(peercred); 1641 1642 spin_lock(&sk->sk_peer_lock); 1643 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred); 1644 spin_unlock(&sk->sk_peer_lock); 1645 1646 if (copy_to_user(optval, &peercred, len)) 1647 return -EFAULT; 1648 goto lenout; 1649 } 1650 1651 case SO_PEERGROUPS: 1652 { 1653 const struct cred *cred; 1654 int ret, n; 1655 1656 cred = sk_get_peer_cred(sk); 1657 if (!cred) 1658 return -ENODATA; 1659 1660 n = cred->group_info->ngroups; 1661 if (len < n * sizeof(gid_t)) { 1662 len = n * sizeof(gid_t); 1663 put_cred(cred); 1664 return put_user(len, optlen) ? -EFAULT : -ERANGE; 1665 } 1666 len = n * sizeof(gid_t); 1667 1668 ret = groups_to_user((gid_t __user *)optval, cred->group_info); 1669 put_cred(cred); 1670 if (ret) 1671 return ret; 1672 goto lenout; 1673 } 1674 1675 case SO_PEERNAME: 1676 { 1677 char address[128]; 1678 1679 lv = sock->ops->getname(sock, (struct sockaddr *)address, 2); 1680 if (lv < 0) 1681 return -ENOTCONN; 1682 if (lv < len) 1683 return -EINVAL; 1684 if (copy_to_user(optval, address, len)) 1685 return -EFAULT; 1686 goto lenout; 1687 } 1688 1689 /* Dubious BSD thing... Probably nobody even uses it, but 1690 * the UNIX standard wants it for whatever reason... -DaveM 1691 */ 1692 case SO_ACCEPTCONN: 1693 v.val = sk->sk_state == TCP_LISTEN; 1694 break; 1695 1696 case SO_PASSSEC: 1697 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags); 1698 break; 1699 1700 case SO_PEERSEC: 1701 return security_socket_getpeersec_stream(sock, optval, optlen, len); 1702 1703 case SO_MARK: 1704 v.val = sk->sk_mark; 1705 break; 1706 1707 case SO_RXQ_OVFL: 1708 v.val = sock_flag(sk, SOCK_RXQ_OVFL); 1709 break; 1710 1711 case SO_WIFI_STATUS: 1712 v.val = sock_flag(sk, SOCK_WIFI_STATUS); 1713 break; 1714 1715 case SO_PEEK_OFF: 1716 if (!sock->ops->set_peek_off) 1717 return -EOPNOTSUPP; 1718 1719 v.val = sk->sk_peek_off; 1720 break; 1721 case SO_NOFCS: 1722 v.val = sock_flag(sk, SOCK_NOFCS); 1723 break; 1724 1725 case SO_BINDTODEVICE: 1726 return sock_getbindtodevice(sk, optval, optlen, len); 1727 1728 case SO_GET_FILTER: 1729 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len); 1730 if (len < 0) 1731 return len; 1732 1733 goto lenout; 1734 1735 case SO_LOCK_FILTER: 1736 v.val = sock_flag(sk, SOCK_FILTER_LOCKED); 1737 break; 1738 1739 case SO_BPF_EXTENSIONS: 1740 v.val = bpf_tell_extensions(); 1741 break; 1742 1743 case SO_SELECT_ERR_QUEUE: 1744 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE); 1745 break; 1746 1747 #ifdef CONFIG_NET_RX_BUSY_POLL 1748 case SO_BUSY_POLL: 1749 v.val = sk->sk_ll_usec; 1750 break; 1751 case SO_PREFER_BUSY_POLL: 1752 v.val = READ_ONCE(sk->sk_prefer_busy_poll); 1753 break; 1754 #endif 1755 1756 case SO_MAX_PACING_RATE: 1757 if (sizeof(v.ulval) != sizeof(v.val) && len >= sizeof(v.ulval)) { 1758 lv = sizeof(v.ulval); 1759 v.ulval = sk->sk_max_pacing_rate; 1760 } else { 1761 /* 32bit version */ 1762 v.val = min_t(unsigned long, sk->sk_max_pacing_rate, ~0U); 1763 } 1764 break; 1765 1766 case SO_INCOMING_CPU: 1767 v.val = READ_ONCE(sk->sk_incoming_cpu); 1768 break; 1769 1770 case SO_MEMINFO: 1771 { 1772 u32 meminfo[SK_MEMINFO_VARS]; 1773 1774 sk_get_meminfo(sk, meminfo); 1775 1776 len = min_t(unsigned int, len, sizeof(meminfo)); 1777 if (copy_to_user(optval, &meminfo, len)) 1778 return -EFAULT; 1779 1780 goto lenout; 1781 } 1782 1783 #ifdef CONFIG_NET_RX_BUSY_POLL 1784 case SO_INCOMING_NAPI_ID: 1785 v.val = READ_ONCE(sk->sk_napi_id); 1786 1787 /* aggregate non-NAPI IDs down to 0 */ 1788 if (v.val < MIN_NAPI_ID) 1789 v.val = 0; 1790 1791 break; 1792 #endif 1793 1794 case SO_COOKIE: 1795 lv = sizeof(u64); 1796 if (len < lv) 1797 return -EINVAL; 1798 v.val64 = sock_gen_cookie(sk); 1799 break; 1800 1801 case SO_ZEROCOPY: 1802 v.val = sock_flag(sk, SOCK_ZEROCOPY); 1803 break; 1804 1805 case SO_TXTIME: 1806 lv = sizeof(v.txtime); 1807 v.txtime.clockid = sk->sk_clockid; 1808 v.txtime.flags |= sk->sk_txtime_deadline_mode ? 1809 SOF_TXTIME_DEADLINE_MODE : 0; 1810 v.txtime.flags |= sk->sk_txtime_report_errors ? 1811 SOF_TXTIME_REPORT_ERRORS : 0; 1812 break; 1813 1814 case SO_BINDTOIFINDEX: 1815 v.val = sk->sk_bound_dev_if; 1816 break; 1817 1818 case SO_NETNS_COOKIE: 1819 lv = sizeof(u64); 1820 if (len != lv) 1821 return -EINVAL; 1822 v.val64 = sock_net(sk)->net_cookie; 1823 break; 1824 1825 case SO_BUF_LOCK: 1826 v.val = sk->sk_userlocks & SOCK_BUF_LOCK_MASK; 1827 break; 1828 1829 case SO_RESERVE_MEM: 1830 v.val = sk->sk_reserved_mem; 1831 break; 1832 1833 default: 1834 /* We implement the SO_SNDLOWAT etc to not be settable 1835 * (1003.1g 7). 1836 */ 1837 return -ENOPROTOOPT; 1838 } 1839 1840 if (len > lv) 1841 len = lv; 1842 if (copy_to_user(optval, &v, len)) 1843 return -EFAULT; 1844 lenout: 1845 if (put_user(len, optlen)) 1846 return -EFAULT; 1847 return 0; 1848 } 1849 1850 /* 1851 * Initialize an sk_lock. 1852 * 1853 * (We also register the sk_lock with the lock validator.) 1854 */ 1855 static inline void sock_lock_init(struct sock *sk) 1856 { 1857 if (sk->sk_kern_sock) 1858 sock_lock_init_class_and_name( 1859 sk, 1860 af_family_kern_slock_key_strings[sk->sk_family], 1861 af_family_kern_slock_keys + sk->sk_family, 1862 af_family_kern_key_strings[sk->sk_family], 1863 af_family_kern_keys + sk->sk_family); 1864 else 1865 sock_lock_init_class_and_name( 1866 sk, 1867 af_family_slock_key_strings[sk->sk_family], 1868 af_family_slock_keys + sk->sk_family, 1869 af_family_key_strings[sk->sk_family], 1870 af_family_keys + sk->sk_family); 1871 } 1872 1873 /* 1874 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet, 1875 * even temporarly, because of RCU lookups. sk_node should also be left as is. 1876 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end 1877 */ 1878 static void sock_copy(struct sock *nsk, const struct sock *osk) 1879 { 1880 const struct proto *prot = READ_ONCE(osk->sk_prot); 1881 #ifdef CONFIG_SECURITY_NETWORK 1882 void *sptr = nsk->sk_security; 1883 #endif 1884 1885 /* If we move sk_tx_queue_mapping out of the private section, 1886 * we must check if sk_tx_queue_clear() is called after 1887 * sock_copy() in sk_clone_lock(). 1888 */ 1889 BUILD_BUG_ON(offsetof(struct sock, sk_tx_queue_mapping) < 1890 offsetof(struct sock, sk_dontcopy_begin) || 1891 offsetof(struct sock, sk_tx_queue_mapping) >= 1892 offsetof(struct sock, sk_dontcopy_end)); 1893 1894 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin)); 1895 1896 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end, 1897 prot->obj_size - offsetof(struct sock, sk_dontcopy_end)); 1898 1899 #ifdef CONFIG_SECURITY_NETWORK 1900 nsk->sk_security = sptr; 1901 security_sk_clone(osk, nsk); 1902 #endif 1903 } 1904 1905 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority, 1906 int family) 1907 { 1908 struct sock *sk; 1909 struct kmem_cache *slab; 1910 1911 slab = prot->slab; 1912 if (slab != NULL) { 1913 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO); 1914 if (!sk) 1915 return sk; 1916 if (want_init_on_alloc(priority)) 1917 sk_prot_clear_nulls(sk, prot->obj_size); 1918 } else 1919 sk = kmalloc(prot->obj_size, priority); 1920 1921 if (sk != NULL) { 1922 if (security_sk_alloc(sk, family, priority)) 1923 goto out_free; 1924 1925 if (!try_module_get(prot->owner)) 1926 goto out_free_sec; 1927 } 1928 1929 return sk; 1930 1931 out_free_sec: 1932 security_sk_free(sk); 1933 out_free: 1934 if (slab != NULL) 1935 kmem_cache_free(slab, sk); 1936 else 1937 kfree(sk); 1938 return NULL; 1939 } 1940 1941 static void sk_prot_free(struct proto *prot, struct sock *sk) 1942 { 1943 struct kmem_cache *slab; 1944 struct module *owner; 1945 1946 owner = prot->owner; 1947 slab = prot->slab; 1948 1949 cgroup_sk_free(&sk->sk_cgrp_data); 1950 mem_cgroup_sk_free(sk); 1951 security_sk_free(sk); 1952 if (slab != NULL) 1953 kmem_cache_free(slab, sk); 1954 else 1955 kfree(sk); 1956 module_put(owner); 1957 } 1958 1959 /** 1960 * sk_alloc - All socket objects are allocated here 1961 * @net: the applicable net namespace 1962 * @family: protocol family 1963 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 1964 * @prot: struct proto associated with this new sock instance 1965 * @kern: is this to be a kernel socket? 1966 */ 1967 struct sock *sk_alloc(struct net *net, int family, gfp_t priority, 1968 struct proto *prot, int kern) 1969 { 1970 struct sock *sk; 1971 1972 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family); 1973 if (sk) { 1974 sk->sk_family = family; 1975 /* 1976 * See comment in struct sock definition to understand 1977 * why we need sk_prot_creator -acme 1978 */ 1979 sk->sk_prot = sk->sk_prot_creator = prot; 1980 sk->sk_kern_sock = kern; 1981 sock_lock_init(sk); 1982 sk->sk_net_refcnt = kern ? 0 : 1; 1983 if (likely(sk->sk_net_refcnt)) { 1984 get_net(net); 1985 sock_inuse_add(net, 1); 1986 } 1987 1988 sock_net_set(sk, net); 1989 refcount_set(&sk->sk_wmem_alloc, 1); 1990 1991 mem_cgroup_sk_alloc(sk); 1992 cgroup_sk_alloc(&sk->sk_cgrp_data); 1993 sock_update_classid(&sk->sk_cgrp_data); 1994 sock_update_netprioidx(&sk->sk_cgrp_data); 1995 sk_tx_queue_clear(sk); 1996 } 1997 1998 return sk; 1999 } 2000 EXPORT_SYMBOL(sk_alloc); 2001 2002 /* Sockets having SOCK_RCU_FREE will call this function after one RCU 2003 * grace period. This is the case for UDP sockets and TCP listeners. 2004 */ 2005 static void __sk_destruct(struct rcu_head *head) 2006 { 2007 struct sock *sk = container_of(head, struct sock, sk_rcu); 2008 struct sk_filter *filter; 2009 2010 if (sk->sk_destruct) 2011 sk->sk_destruct(sk); 2012 2013 filter = rcu_dereference_check(sk->sk_filter, 2014 refcount_read(&sk->sk_wmem_alloc) == 0); 2015 if (filter) { 2016 sk_filter_uncharge(sk, filter); 2017 RCU_INIT_POINTER(sk->sk_filter, NULL); 2018 } 2019 2020 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP); 2021 2022 #ifdef CONFIG_BPF_SYSCALL 2023 bpf_sk_storage_free(sk); 2024 #endif 2025 2026 if (atomic_read(&sk->sk_omem_alloc)) 2027 pr_debug("%s: optmem leakage (%d bytes) detected\n", 2028 __func__, atomic_read(&sk->sk_omem_alloc)); 2029 2030 if (sk->sk_frag.page) { 2031 put_page(sk->sk_frag.page); 2032 sk->sk_frag.page = NULL; 2033 } 2034 2035 /* We do not need to acquire sk->sk_peer_lock, we are the last user. */ 2036 put_cred(sk->sk_peer_cred); 2037 put_pid(sk->sk_peer_pid); 2038 2039 if (likely(sk->sk_net_refcnt)) 2040 put_net(sock_net(sk)); 2041 sk_prot_free(sk->sk_prot_creator, sk); 2042 } 2043 2044 void sk_destruct(struct sock *sk) 2045 { 2046 bool use_call_rcu = sock_flag(sk, SOCK_RCU_FREE); 2047 2048 if (rcu_access_pointer(sk->sk_reuseport_cb)) { 2049 reuseport_detach_sock(sk); 2050 use_call_rcu = true; 2051 } 2052 2053 if (use_call_rcu) 2054 call_rcu(&sk->sk_rcu, __sk_destruct); 2055 else 2056 __sk_destruct(&sk->sk_rcu); 2057 } 2058 2059 static void __sk_free(struct sock *sk) 2060 { 2061 if (likely(sk->sk_net_refcnt)) 2062 sock_inuse_add(sock_net(sk), -1); 2063 2064 if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk))) 2065 sock_diag_broadcast_destroy(sk); 2066 else 2067 sk_destruct(sk); 2068 } 2069 2070 void sk_free(struct sock *sk) 2071 { 2072 /* 2073 * We subtract one from sk_wmem_alloc and can know if 2074 * some packets are still in some tx queue. 2075 * If not null, sock_wfree() will call __sk_free(sk) later 2076 */ 2077 if (refcount_dec_and_test(&sk->sk_wmem_alloc)) 2078 __sk_free(sk); 2079 } 2080 EXPORT_SYMBOL(sk_free); 2081 2082 static void sk_init_common(struct sock *sk) 2083 { 2084 skb_queue_head_init(&sk->sk_receive_queue); 2085 skb_queue_head_init(&sk->sk_write_queue); 2086 skb_queue_head_init(&sk->sk_error_queue); 2087 2088 rwlock_init(&sk->sk_callback_lock); 2089 lockdep_set_class_and_name(&sk->sk_receive_queue.lock, 2090 af_rlock_keys + sk->sk_family, 2091 af_family_rlock_key_strings[sk->sk_family]); 2092 lockdep_set_class_and_name(&sk->sk_write_queue.lock, 2093 af_wlock_keys + sk->sk_family, 2094 af_family_wlock_key_strings[sk->sk_family]); 2095 lockdep_set_class_and_name(&sk->sk_error_queue.lock, 2096 af_elock_keys + sk->sk_family, 2097 af_family_elock_key_strings[sk->sk_family]); 2098 lockdep_set_class_and_name(&sk->sk_callback_lock, 2099 af_callback_keys + sk->sk_family, 2100 af_family_clock_key_strings[sk->sk_family]); 2101 } 2102 2103 /** 2104 * sk_clone_lock - clone a socket, and lock its clone 2105 * @sk: the socket to clone 2106 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 2107 * 2108 * Caller must unlock socket even in error path (bh_unlock_sock(newsk)) 2109 */ 2110 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority) 2111 { 2112 struct proto *prot = READ_ONCE(sk->sk_prot); 2113 struct sk_filter *filter; 2114 bool is_charged = true; 2115 struct sock *newsk; 2116 2117 newsk = sk_prot_alloc(prot, priority, sk->sk_family); 2118 if (!newsk) 2119 goto out; 2120 2121 sock_copy(newsk, sk); 2122 2123 newsk->sk_prot_creator = prot; 2124 2125 /* SANITY */ 2126 if (likely(newsk->sk_net_refcnt)) { 2127 get_net(sock_net(newsk)); 2128 sock_inuse_add(sock_net(newsk), 1); 2129 } 2130 sk_node_init(&newsk->sk_node); 2131 sock_lock_init(newsk); 2132 bh_lock_sock(newsk); 2133 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL; 2134 newsk->sk_backlog.len = 0; 2135 2136 atomic_set(&newsk->sk_rmem_alloc, 0); 2137 2138 /* sk_wmem_alloc set to one (see sk_free() and sock_wfree()) */ 2139 refcount_set(&newsk->sk_wmem_alloc, 1); 2140 2141 atomic_set(&newsk->sk_omem_alloc, 0); 2142 sk_init_common(newsk); 2143 2144 newsk->sk_dst_cache = NULL; 2145 newsk->sk_dst_pending_confirm = 0; 2146 newsk->sk_wmem_queued = 0; 2147 newsk->sk_forward_alloc = 0; 2148 newsk->sk_reserved_mem = 0; 2149 atomic_set(&newsk->sk_drops, 0); 2150 newsk->sk_send_head = NULL; 2151 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK; 2152 atomic_set(&newsk->sk_zckey, 0); 2153 2154 sock_reset_flag(newsk, SOCK_DONE); 2155 2156 /* sk->sk_memcg will be populated at accept() time */ 2157 newsk->sk_memcg = NULL; 2158 2159 cgroup_sk_clone(&newsk->sk_cgrp_data); 2160 2161 rcu_read_lock(); 2162 filter = rcu_dereference(sk->sk_filter); 2163 if (filter != NULL) 2164 /* though it's an empty new sock, the charging may fail 2165 * if sysctl_optmem_max was changed between creation of 2166 * original socket and cloning 2167 */ 2168 is_charged = sk_filter_charge(newsk, filter); 2169 RCU_INIT_POINTER(newsk->sk_filter, filter); 2170 rcu_read_unlock(); 2171 2172 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) { 2173 /* We need to make sure that we don't uncharge the new 2174 * socket if we couldn't charge it in the first place 2175 * as otherwise we uncharge the parent's filter. 2176 */ 2177 if (!is_charged) 2178 RCU_INIT_POINTER(newsk->sk_filter, NULL); 2179 sk_free_unlock_clone(newsk); 2180 newsk = NULL; 2181 goto out; 2182 } 2183 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL); 2184 2185 if (bpf_sk_storage_clone(sk, newsk)) { 2186 sk_free_unlock_clone(newsk); 2187 newsk = NULL; 2188 goto out; 2189 } 2190 2191 /* Clear sk_user_data if parent had the pointer tagged 2192 * as not suitable for copying when cloning. 2193 */ 2194 if (sk_user_data_is_nocopy(newsk)) 2195 newsk->sk_user_data = NULL; 2196 2197 newsk->sk_err = 0; 2198 newsk->sk_err_soft = 0; 2199 newsk->sk_priority = 0; 2200 newsk->sk_incoming_cpu = raw_smp_processor_id(); 2201 2202 /* Before updating sk_refcnt, we must commit prior changes to memory 2203 * (Documentation/RCU/rculist_nulls.rst for details) 2204 */ 2205 smp_wmb(); 2206 refcount_set(&newsk->sk_refcnt, 2); 2207 2208 /* Increment the counter in the same struct proto as the master 2209 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that 2210 * is the same as sk->sk_prot->socks, as this field was copied 2211 * with memcpy). 2212 * 2213 * This _changes_ the previous behaviour, where 2214 * tcp_create_openreq_child always was incrementing the 2215 * equivalent to tcp_prot->socks (inet_sock_nr), so this have 2216 * to be taken into account in all callers. -acme 2217 */ 2218 sk_refcnt_debug_inc(newsk); 2219 sk_set_socket(newsk, NULL); 2220 sk_tx_queue_clear(newsk); 2221 RCU_INIT_POINTER(newsk->sk_wq, NULL); 2222 2223 if (newsk->sk_prot->sockets_allocated) 2224 sk_sockets_allocated_inc(newsk); 2225 2226 if (sock_needs_netstamp(sk) && newsk->sk_flags & SK_FLAGS_TIMESTAMP) 2227 net_enable_timestamp(); 2228 out: 2229 return newsk; 2230 } 2231 EXPORT_SYMBOL_GPL(sk_clone_lock); 2232 2233 void sk_free_unlock_clone(struct sock *sk) 2234 { 2235 /* It is still raw copy of parent, so invalidate 2236 * destructor and make plain sk_free() */ 2237 sk->sk_destruct = NULL; 2238 bh_unlock_sock(sk); 2239 sk_free(sk); 2240 } 2241 EXPORT_SYMBOL_GPL(sk_free_unlock_clone); 2242 2243 void sk_setup_caps(struct sock *sk, struct dst_entry *dst) 2244 { 2245 u32 max_segs = 1; 2246 2247 sk_dst_set(sk, dst); 2248 sk->sk_route_caps = dst->dev->features; 2249 if (sk_is_tcp(sk)) 2250 sk->sk_route_caps |= NETIF_F_GSO; 2251 if (sk->sk_route_caps & NETIF_F_GSO) 2252 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE; 2253 if (unlikely(sk->sk_gso_disabled)) 2254 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 2255 if (sk_can_gso(sk)) { 2256 if (dst->header_len && !xfrm_dst_offload_ok(dst)) { 2257 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 2258 } else { 2259 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM; 2260 sk->sk_gso_max_size = dst->dev->gso_max_size; 2261 max_segs = max_t(u32, dst->dev->gso_max_segs, 1); 2262 } 2263 } 2264 sk->sk_gso_max_segs = max_segs; 2265 } 2266 EXPORT_SYMBOL_GPL(sk_setup_caps); 2267 2268 /* 2269 * Simple resource managers for sockets. 2270 */ 2271 2272 2273 /* 2274 * Write buffer destructor automatically called from kfree_skb. 2275 */ 2276 void sock_wfree(struct sk_buff *skb) 2277 { 2278 struct sock *sk = skb->sk; 2279 unsigned int len = skb->truesize; 2280 2281 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) { 2282 /* 2283 * Keep a reference on sk_wmem_alloc, this will be released 2284 * after sk_write_space() call 2285 */ 2286 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc)); 2287 sk->sk_write_space(sk); 2288 len = 1; 2289 } 2290 /* 2291 * if sk_wmem_alloc reaches 0, we must finish what sk_free() 2292 * could not do because of in-flight packets 2293 */ 2294 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc)) 2295 __sk_free(sk); 2296 } 2297 EXPORT_SYMBOL(sock_wfree); 2298 2299 /* This variant of sock_wfree() is used by TCP, 2300 * since it sets SOCK_USE_WRITE_QUEUE. 2301 */ 2302 void __sock_wfree(struct sk_buff *skb) 2303 { 2304 struct sock *sk = skb->sk; 2305 2306 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc)) 2307 __sk_free(sk); 2308 } 2309 2310 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk) 2311 { 2312 skb_orphan(skb); 2313 skb->sk = sk; 2314 #ifdef CONFIG_INET 2315 if (unlikely(!sk_fullsock(sk))) { 2316 skb->destructor = sock_edemux; 2317 sock_hold(sk); 2318 return; 2319 } 2320 #endif 2321 skb->destructor = sock_wfree; 2322 skb_set_hash_from_sk(skb, sk); 2323 /* 2324 * We used to take a refcount on sk, but following operation 2325 * is enough to guarantee sk_free() wont free this sock until 2326 * all in-flight packets are completed 2327 */ 2328 refcount_add(skb->truesize, &sk->sk_wmem_alloc); 2329 } 2330 EXPORT_SYMBOL(skb_set_owner_w); 2331 2332 static bool can_skb_orphan_partial(const struct sk_buff *skb) 2333 { 2334 #ifdef CONFIG_TLS_DEVICE 2335 /* Drivers depend on in-order delivery for crypto offload, 2336 * partial orphan breaks out-of-order-OK logic. 2337 */ 2338 if (skb->decrypted) 2339 return false; 2340 #endif 2341 return (skb->destructor == sock_wfree || 2342 (IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree)); 2343 } 2344 2345 /* This helper is used by netem, as it can hold packets in its 2346 * delay queue. We want to allow the owner socket to send more 2347 * packets, as if they were already TX completed by a typical driver. 2348 * But we also want to keep skb->sk set because some packet schedulers 2349 * rely on it (sch_fq for example). 2350 */ 2351 void skb_orphan_partial(struct sk_buff *skb) 2352 { 2353 if (skb_is_tcp_pure_ack(skb)) 2354 return; 2355 2356 if (can_skb_orphan_partial(skb) && skb_set_owner_sk_safe(skb, skb->sk)) 2357 return; 2358 2359 skb_orphan(skb); 2360 } 2361 EXPORT_SYMBOL(skb_orphan_partial); 2362 2363 /* 2364 * Read buffer destructor automatically called from kfree_skb. 2365 */ 2366 void sock_rfree(struct sk_buff *skb) 2367 { 2368 struct sock *sk = skb->sk; 2369 unsigned int len = skb->truesize; 2370 2371 atomic_sub(len, &sk->sk_rmem_alloc); 2372 sk_mem_uncharge(sk, len); 2373 } 2374 EXPORT_SYMBOL(sock_rfree); 2375 2376 /* 2377 * Buffer destructor for skbs that are not used directly in read or write 2378 * path, e.g. for error handler skbs. Automatically called from kfree_skb. 2379 */ 2380 void sock_efree(struct sk_buff *skb) 2381 { 2382 sock_put(skb->sk); 2383 } 2384 EXPORT_SYMBOL(sock_efree); 2385 2386 /* Buffer destructor for prefetch/receive path where reference count may 2387 * not be held, e.g. for listen sockets. 2388 */ 2389 #ifdef CONFIG_INET 2390 void sock_pfree(struct sk_buff *skb) 2391 { 2392 if (sk_is_refcounted(skb->sk)) 2393 sock_gen_put(skb->sk); 2394 } 2395 EXPORT_SYMBOL(sock_pfree); 2396 #endif /* CONFIG_INET */ 2397 2398 kuid_t sock_i_uid(struct sock *sk) 2399 { 2400 kuid_t uid; 2401 2402 read_lock_bh(&sk->sk_callback_lock); 2403 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID; 2404 read_unlock_bh(&sk->sk_callback_lock); 2405 return uid; 2406 } 2407 EXPORT_SYMBOL(sock_i_uid); 2408 2409 unsigned long sock_i_ino(struct sock *sk) 2410 { 2411 unsigned long ino; 2412 2413 read_lock_bh(&sk->sk_callback_lock); 2414 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0; 2415 read_unlock_bh(&sk->sk_callback_lock); 2416 return ino; 2417 } 2418 EXPORT_SYMBOL(sock_i_ino); 2419 2420 /* 2421 * Allocate a skb from the socket's send buffer. 2422 */ 2423 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 2424 gfp_t priority) 2425 { 2426 if (force || 2427 refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) { 2428 struct sk_buff *skb = alloc_skb(size, priority); 2429 2430 if (skb) { 2431 skb_set_owner_w(skb, sk); 2432 return skb; 2433 } 2434 } 2435 return NULL; 2436 } 2437 EXPORT_SYMBOL(sock_wmalloc); 2438 2439 static void sock_ofree(struct sk_buff *skb) 2440 { 2441 struct sock *sk = skb->sk; 2442 2443 atomic_sub(skb->truesize, &sk->sk_omem_alloc); 2444 } 2445 2446 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size, 2447 gfp_t priority) 2448 { 2449 struct sk_buff *skb; 2450 2451 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */ 2452 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) > 2453 sysctl_optmem_max) 2454 return NULL; 2455 2456 skb = alloc_skb(size, priority); 2457 if (!skb) 2458 return NULL; 2459 2460 atomic_add(skb->truesize, &sk->sk_omem_alloc); 2461 skb->sk = sk; 2462 skb->destructor = sock_ofree; 2463 return skb; 2464 } 2465 2466 /* 2467 * Allocate a memory block from the socket's option memory buffer. 2468 */ 2469 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority) 2470 { 2471 if ((unsigned int)size <= sysctl_optmem_max && 2472 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) { 2473 void *mem; 2474 /* First do the add, to avoid the race if kmalloc 2475 * might sleep. 2476 */ 2477 atomic_add(size, &sk->sk_omem_alloc); 2478 mem = kmalloc(size, priority); 2479 if (mem) 2480 return mem; 2481 atomic_sub(size, &sk->sk_omem_alloc); 2482 } 2483 return NULL; 2484 } 2485 EXPORT_SYMBOL(sock_kmalloc); 2486 2487 /* Free an option memory block. Note, we actually want the inline 2488 * here as this allows gcc to detect the nullify and fold away the 2489 * condition entirely. 2490 */ 2491 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size, 2492 const bool nullify) 2493 { 2494 if (WARN_ON_ONCE(!mem)) 2495 return; 2496 if (nullify) 2497 kfree_sensitive(mem); 2498 else 2499 kfree(mem); 2500 atomic_sub(size, &sk->sk_omem_alloc); 2501 } 2502 2503 void sock_kfree_s(struct sock *sk, void *mem, int size) 2504 { 2505 __sock_kfree_s(sk, mem, size, false); 2506 } 2507 EXPORT_SYMBOL(sock_kfree_s); 2508 2509 void sock_kzfree_s(struct sock *sk, void *mem, int size) 2510 { 2511 __sock_kfree_s(sk, mem, size, true); 2512 } 2513 EXPORT_SYMBOL(sock_kzfree_s); 2514 2515 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock. 2516 I think, these locks should be removed for datagram sockets. 2517 */ 2518 static long sock_wait_for_wmem(struct sock *sk, long timeo) 2519 { 2520 DEFINE_WAIT(wait); 2521 2522 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk); 2523 for (;;) { 2524 if (!timeo) 2525 break; 2526 if (signal_pending(current)) 2527 break; 2528 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 2529 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 2530 if (refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) 2531 break; 2532 if (sk->sk_shutdown & SEND_SHUTDOWN) 2533 break; 2534 if (sk->sk_err) 2535 break; 2536 timeo = schedule_timeout(timeo); 2537 } 2538 finish_wait(sk_sleep(sk), &wait); 2539 return timeo; 2540 } 2541 2542 2543 /* 2544 * Generic send/receive buffer handlers 2545 */ 2546 2547 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 2548 unsigned long data_len, int noblock, 2549 int *errcode, int max_page_order) 2550 { 2551 struct sk_buff *skb; 2552 long timeo; 2553 int err; 2554 2555 timeo = sock_sndtimeo(sk, noblock); 2556 for (;;) { 2557 err = sock_error(sk); 2558 if (err != 0) 2559 goto failure; 2560 2561 err = -EPIPE; 2562 if (sk->sk_shutdown & SEND_SHUTDOWN) 2563 goto failure; 2564 2565 if (sk_wmem_alloc_get(sk) < READ_ONCE(sk->sk_sndbuf)) 2566 break; 2567 2568 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk); 2569 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 2570 err = -EAGAIN; 2571 if (!timeo) 2572 goto failure; 2573 if (signal_pending(current)) 2574 goto interrupted; 2575 timeo = sock_wait_for_wmem(sk, timeo); 2576 } 2577 skb = alloc_skb_with_frags(header_len, data_len, max_page_order, 2578 errcode, sk->sk_allocation); 2579 if (skb) 2580 skb_set_owner_w(skb, sk); 2581 return skb; 2582 2583 interrupted: 2584 err = sock_intr_errno(timeo); 2585 failure: 2586 *errcode = err; 2587 return NULL; 2588 } 2589 EXPORT_SYMBOL(sock_alloc_send_pskb); 2590 2591 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size, 2592 int noblock, int *errcode) 2593 { 2594 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0); 2595 } 2596 EXPORT_SYMBOL(sock_alloc_send_skb); 2597 2598 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg, 2599 struct sockcm_cookie *sockc) 2600 { 2601 u32 tsflags; 2602 2603 switch (cmsg->cmsg_type) { 2604 case SO_MARK: 2605 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) 2606 return -EPERM; 2607 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 2608 return -EINVAL; 2609 sockc->mark = *(u32 *)CMSG_DATA(cmsg); 2610 break; 2611 case SO_TIMESTAMPING_OLD: 2612 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 2613 return -EINVAL; 2614 2615 tsflags = *(u32 *)CMSG_DATA(cmsg); 2616 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK) 2617 return -EINVAL; 2618 2619 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK; 2620 sockc->tsflags |= tsflags; 2621 break; 2622 case SCM_TXTIME: 2623 if (!sock_flag(sk, SOCK_TXTIME)) 2624 return -EINVAL; 2625 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64))) 2626 return -EINVAL; 2627 sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg)); 2628 break; 2629 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */ 2630 case SCM_RIGHTS: 2631 case SCM_CREDENTIALS: 2632 break; 2633 default: 2634 return -EINVAL; 2635 } 2636 return 0; 2637 } 2638 EXPORT_SYMBOL(__sock_cmsg_send); 2639 2640 int sock_cmsg_send(struct sock *sk, struct msghdr *msg, 2641 struct sockcm_cookie *sockc) 2642 { 2643 struct cmsghdr *cmsg; 2644 int ret; 2645 2646 for_each_cmsghdr(cmsg, msg) { 2647 if (!CMSG_OK(msg, cmsg)) 2648 return -EINVAL; 2649 if (cmsg->cmsg_level != SOL_SOCKET) 2650 continue; 2651 ret = __sock_cmsg_send(sk, msg, cmsg, sockc); 2652 if (ret) 2653 return ret; 2654 } 2655 return 0; 2656 } 2657 EXPORT_SYMBOL(sock_cmsg_send); 2658 2659 static void sk_enter_memory_pressure(struct sock *sk) 2660 { 2661 if (!sk->sk_prot->enter_memory_pressure) 2662 return; 2663 2664 sk->sk_prot->enter_memory_pressure(sk); 2665 } 2666 2667 static void sk_leave_memory_pressure(struct sock *sk) 2668 { 2669 if (sk->sk_prot->leave_memory_pressure) { 2670 sk->sk_prot->leave_memory_pressure(sk); 2671 } else { 2672 unsigned long *memory_pressure = sk->sk_prot->memory_pressure; 2673 2674 if (memory_pressure && READ_ONCE(*memory_pressure)) 2675 WRITE_ONCE(*memory_pressure, 0); 2676 } 2677 } 2678 2679 DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key); 2680 2681 /** 2682 * skb_page_frag_refill - check that a page_frag contains enough room 2683 * @sz: minimum size of the fragment we want to get 2684 * @pfrag: pointer to page_frag 2685 * @gfp: priority for memory allocation 2686 * 2687 * Note: While this allocator tries to use high order pages, there is 2688 * no guarantee that allocations succeed. Therefore, @sz MUST be 2689 * less or equal than PAGE_SIZE. 2690 */ 2691 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp) 2692 { 2693 if (pfrag->page) { 2694 if (page_ref_count(pfrag->page) == 1) { 2695 pfrag->offset = 0; 2696 return true; 2697 } 2698 if (pfrag->offset + sz <= pfrag->size) 2699 return true; 2700 put_page(pfrag->page); 2701 } 2702 2703 pfrag->offset = 0; 2704 if (SKB_FRAG_PAGE_ORDER && 2705 !static_branch_unlikely(&net_high_order_alloc_disable_key)) { 2706 /* Avoid direct reclaim but allow kswapd to wake */ 2707 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) | 2708 __GFP_COMP | __GFP_NOWARN | 2709 __GFP_NORETRY, 2710 SKB_FRAG_PAGE_ORDER); 2711 if (likely(pfrag->page)) { 2712 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER; 2713 return true; 2714 } 2715 } 2716 pfrag->page = alloc_page(gfp); 2717 if (likely(pfrag->page)) { 2718 pfrag->size = PAGE_SIZE; 2719 return true; 2720 } 2721 return false; 2722 } 2723 EXPORT_SYMBOL(skb_page_frag_refill); 2724 2725 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag) 2726 { 2727 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation))) 2728 return true; 2729 2730 sk_enter_memory_pressure(sk); 2731 sk_stream_moderate_sndbuf(sk); 2732 return false; 2733 } 2734 EXPORT_SYMBOL(sk_page_frag_refill); 2735 2736 void __lock_sock(struct sock *sk) 2737 __releases(&sk->sk_lock.slock) 2738 __acquires(&sk->sk_lock.slock) 2739 { 2740 DEFINE_WAIT(wait); 2741 2742 for (;;) { 2743 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait, 2744 TASK_UNINTERRUPTIBLE); 2745 spin_unlock_bh(&sk->sk_lock.slock); 2746 schedule(); 2747 spin_lock_bh(&sk->sk_lock.slock); 2748 if (!sock_owned_by_user(sk)) 2749 break; 2750 } 2751 finish_wait(&sk->sk_lock.wq, &wait); 2752 } 2753 2754 void __release_sock(struct sock *sk) 2755 __releases(&sk->sk_lock.slock) 2756 __acquires(&sk->sk_lock.slock) 2757 { 2758 struct sk_buff *skb, *next; 2759 2760 while ((skb = sk->sk_backlog.head) != NULL) { 2761 sk->sk_backlog.head = sk->sk_backlog.tail = NULL; 2762 2763 spin_unlock_bh(&sk->sk_lock.slock); 2764 2765 do { 2766 next = skb->next; 2767 prefetch(next); 2768 WARN_ON_ONCE(skb_dst_is_noref(skb)); 2769 skb_mark_not_on_list(skb); 2770 sk_backlog_rcv(sk, skb); 2771 2772 cond_resched(); 2773 2774 skb = next; 2775 } while (skb != NULL); 2776 2777 spin_lock_bh(&sk->sk_lock.slock); 2778 } 2779 2780 /* 2781 * Doing the zeroing here guarantee we can not loop forever 2782 * while a wild producer attempts to flood us. 2783 */ 2784 sk->sk_backlog.len = 0; 2785 } 2786 2787 void __sk_flush_backlog(struct sock *sk) 2788 { 2789 spin_lock_bh(&sk->sk_lock.slock); 2790 __release_sock(sk); 2791 spin_unlock_bh(&sk->sk_lock.slock); 2792 } 2793 2794 /** 2795 * sk_wait_data - wait for data to arrive at sk_receive_queue 2796 * @sk: sock to wait on 2797 * @timeo: for how long 2798 * @skb: last skb seen on sk_receive_queue 2799 * 2800 * Now socket state including sk->sk_err is changed only under lock, 2801 * hence we may omit checks after joining wait queue. 2802 * We check receive queue before schedule() only as optimization; 2803 * it is very likely that release_sock() added new data. 2804 */ 2805 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb) 2806 { 2807 DEFINE_WAIT_FUNC(wait, woken_wake_function); 2808 int rc; 2809 2810 add_wait_queue(sk_sleep(sk), &wait); 2811 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk); 2812 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait); 2813 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk); 2814 remove_wait_queue(sk_sleep(sk), &wait); 2815 return rc; 2816 } 2817 EXPORT_SYMBOL(sk_wait_data); 2818 2819 /** 2820 * __sk_mem_raise_allocated - increase memory_allocated 2821 * @sk: socket 2822 * @size: memory size to allocate 2823 * @amt: pages to allocate 2824 * @kind: allocation type 2825 * 2826 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc 2827 */ 2828 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind) 2829 { 2830 struct proto *prot = sk->sk_prot; 2831 long allocated = sk_memory_allocated_add(sk, amt); 2832 bool memcg_charge = mem_cgroup_sockets_enabled && sk->sk_memcg; 2833 bool charged = true; 2834 2835 if (memcg_charge && 2836 !(charged = mem_cgroup_charge_skmem(sk->sk_memcg, amt, 2837 gfp_memcg_charge()))) 2838 goto suppress_allocation; 2839 2840 /* Under limit. */ 2841 if (allocated <= sk_prot_mem_limits(sk, 0)) { 2842 sk_leave_memory_pressure(sk); 2843 return 1; 2844 } 2845 2846 /* Under pressure. */ 2847 if (allocated > sk_prot_mem_limits(sk, 1)) 2848 sk_enter_memory_pressure(sk); 2849 2850 /* Over hard limit. */ 2851 if (allocated > sk_prot_mem_limits(sk, 2)) 2852 goto suppress_allocation; 2853 2854 /* guarantee minimum buffer size under pressure */ 2855 if (kind == SK_MEM_RECV) { 2856 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot)) 2857 return 1; 2858 2859 } else { /* SK_MEM_SEND */ 2860 int wmem0 = sk_get_wmem0(sk, prot); 2861 2862 if (sk->sk_type == SOCK_STREAM) { 2863 if (sk->sk_wmem_queued < wmem0) 2864 return 1; 2865 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) { 2866 return 1; 2867 } 2868 } 2869 2870 if (sk_has_memory_pressure(sk)) { 2871 u64 alloc; 2872 2873 if (!sk_under_memory_pressure(sk)) 2874 return 1; 2875 alloc = sk_sockets_allocated_read_positive(sk); 2876 if (sk_prot_mem_limits(sk, 2) > alloc * 2877 sk_mem_pages(sk->sk_wmem_queued + 2878 atomic_read(&sk->sk_rmem_alloc) + 2879 sk->sk_forward_alloc)) 2880 return 1; 2881 } 2882 2883 suppress_allocation: 2884 2885 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) { 2886 sk_stream_moderate_sndbuf(sk); 2887 2888 /* Fail only if socket is _under_ its sndbuf. 2889 * In this case we cannot block, so that we have to fail. 2890 */ 2891 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf) { 2892 /* Force charge with __GFP_NOFAIL */ 2893 if (memcg_charge && !charged) { 2894 mem_cgroup_charge_skmem(sk->sk_memcg, amt, 2895 gfp_memcg_charge() | __GFP_NOFAIL); 2896 } 2897 return 1; 2898 } 2899 } 2900 2901 if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged)) 2902 trace_sock_exceed_buf_limit(sk, prot, allocated, kind); 2903 2904 sk_memory_allocated_sub(sk, amt); 2905 2906 if (memcg_charge && charged) 2907 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt); 2908 2909 return 0; 2910 } 2911 EXPORT_SYMBOL(__sk_mem_raise_allocated); 2912 2913 /** 2914 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated 2915 * @sk: socket 2916 * @size: memory size to allocate 2917 * @kind: allocation type 2918 * 2919 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means 2920 * rmem allocation. This function assumes that protocols which have 2921 * memory_pressure use sk_wmem_queued as write buffer accounting. 2922 */ 2923 int __sk_mem_schedule(struct sock *sk, int size, int kind) 2924 { 2925 int ret, amt = sk_mem_pages(size); 2926 2927 sk->sk_forward_alloc += amt << SK_MEM_QUANTUM_SHIFT; 2928 ret = __sk_mem_raise_allocated(sk, size, amt, kind); 2929 if (!ret) 2930 sk->sk_forward_alloc -= amt << SK_MEM_QUANTUM_SHIFT; 2931 return ret; 2932 } 2933 EXPORT_SYMBOL(__sk_mem_schedule); 2934 2935 /** 2936 * __sk_mem_reduce_allocated - reclaim memory_allocated 2937 * @sk: socket 2938 * @amount: number of quanta 2939 * 2940 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc 2941 */ 2942 void __sk_mem_reduce_allocated(struct sock *sk, int amount) 2943 { 2944 sk_memory_allocated_sub(sk, amount); 2945 2946 if (mem_cgroup_sockets_enabled && sk->sk_memcg) 2947 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount); 2948 2949 if (sk_under_memory_pressure(sk) && 2950 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0))) 2951 sk_leave_memory_pressure(sk); 2952 } 2953 EXPORT_SYMBOL(__sk_mem_reduce_allocated); 2954 2955 /** 2956 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated 2957 * @sk: socket 2958 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple) 2959 */ 2960 void __sk_mem_reclaim(struct sock *sk, int amount) 2961 { 2962 amount >>= SK_MEM_QUANTUM_SHIFT; 2963 sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT; 2964 __sk_mem_reduce_allocated(sk, amount); 2965 } 2966 EXPORT_SYMBOL(__sk_mem_reclaim); 2967 2968 int sk_set_peek_off(struct sock *sk, int val) 2969 { 2970 sk->sk_peek_off = val; 2971 return 0; 2972 } 2973 EXPORT_SYMBOL_GPL(sk_set_peek_off); 2974 2975 /* 2976 * Set of default routines for initialising struct proto_ops when 2977 * the protocol does not support a particular function. In certain 2978 * cases where it makes no sense for a protocol to have a "do nothing" 2979 * function, some default processing is provided. 2980 */ 2981 2982 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len) 2983 { 2984 return -EOPNOTSUPP; 2985 } 2986 EXPORT_SYMBOL(sock_no_bind); 2987 2988 int sock_no_connect(struct socket *sock, struct sockaddr *saddr, 2989 int len, int flags) 2990 { 2991 return -EOPNOTSUPP; 2992 } 2993 EXPORT_SYMBOL(sock_no_connect); 2994 2995 int sock_no_socketpair(struct socket *sock1, struct socket *sock2) 2996 { 2997 return -EOPNOTSUPP; 2998 } 2999 EXPORT_SYMBOL(sock_no_socketpair); 3000 3001 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags, 3002 bool kern) 3003 { 3004 return -EOPNOTSUPP; 3005 } 3006 EXPORT_SYMBOL(sock_no_accept); 3007 3008 int sock_no_getname(struct socket *sock, struct sockaddr *saddr, 3009 int peer) 3010 { 3011 return -EOPNOTSUPP; 3012 } 3013 EXPORT_SYMBOL(sock_no_getname); 3014 3015 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 3016 { 3017 return -EOPNOTSUPP; 3018 } 3019 EXPORT_SYMBOL(sock_no_ioctl); 3020 3021 int sock_no_listen(struct socket *sock, int backlog) 3022 { 3023 return -EOPNOTSUPP; 3024 } 3025 EXPORT_SYMBOL(sock_no_listen); 3026 3027 int sock_no_shutdown(struct socket *sock, int how) 3028 { 3029 return -EOPNOTSUPP; 3030 } 3031 EXPORT_SYMBOL(sock_no_shutdown); 3032 3033 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len) 3034 { 3035 return -EOPNOTSUPP; 3036 } 3037 EXPORT_SYMBOL(sock_no_sendmsg); 3038 3039 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len) 3040 { 3041 return -EOPNOTSUPP; 3042 } 3043 EXPORT_SYMBOL(sock_no_sendmsg_locked); 3044 3045 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len, 3046 int flags) 3047 { 3048 return -EOPNOTSUPP; 3049 } 3050 EXPORT_SYMBOL(sock_no_recvmsg); 3051 3052 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma) 3053 { 3054 /* Mirror missing mmap method error code */ 3055 return -ENODEV; 3056 } 3057 EXPORT_SYMBOL(sock_no_mmap); 3058 3059 /* 3060 * When a file is received (via SCM_RIGHTS, etc), we must bump the 3061 * various sock-based usage counts. 3062 */ 3063 void __receive_sock(struct file *file) 3064 { 3065 struct socket *sock; 3066 3067 sock = sock_from_file(file); 3068 if (sock) { 3069 sock_update_netprioidx(&sock->sk->sk_cgrp_data); 3070 sock_update_classid(&sock->sk->sk_cgrp_data); 3071 } 3072 } 3073 3074 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags) 3075 { 3076 ssize_t res; 3077 struct msghdr msg = {.msg_flags = flags}; 3078 struct kvec iov; 3079 char *kaddr = kmap(page); 3080 iov.iov_base = kaddr + offset; 3081 iov.iov_len = size; 3082 res = kernel_sendmsg(sock, &msg, &iov, 1, size); 3083 kunmap(page); 3084 return res; 3085 } 3086 EXPORT_SYMBOL(sock_no_sendpage); 3087 3088 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page, 3089 int offset, size_t size, int flags) 3090 { 3091 ssize_t res; 3092 struct msghdr msg = {.msg_flags = flags}; 3093 struct kvec iov; 3094 char *kaddr = kmap(page); 3095 3096 iov.iov_base = kaddr + offset; 3097 iov.iov_len = size; 3098 res = kernel_sendmsg_locked(sk, &msg, &iov, 1, size); 3099 kunmap(page); 3100 return res; 3101 } 3102 EXPORT_SYMBOL(sock_no_sendpage_locked); 3103 3104 /* 3105 * Default Socket Callbacks 3106 */ 3107 3108 static void sock_def_wakeup(struct sock *sk) 3109 { 3110 struct socket_wq *wq; 3111 3112 rcu_read_lock(); 3113 wq = rcu_dereference(sk->sk_wq); 3114 if (skwq_has_sleeper(wq)) 3115 wake_up_interruptible_all(&wq->wait); 3116 rcu_read_unlock(); 3117 } 3118 3119 static void sock_def_error_report(struct sock *sk) 3120 { 3121 struct socket_wq *wq; 3122 3123 rcu_read_lock(); 3124 wq = rcu_dereference(sk->sk_wq); 3125 if (skwq_has_sleeper(wq)) 3126 wake_up_interruptible_poll(&wq->wait, EPOLLERR); 3127 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR); 3128 rcu_read_unlock(); 3129 } 3130 3131 void sock_def_readable(struct sock *sk) 3132 { 3133 struct socket_wq *wq; 3134 3135 rcu_read_lock(); 3136 wq = rcu_dereference(sk->sk_wq); 3137 if (skwq_has_sleeper(wq)) 3138 wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI | 3139 EPOLLRDNORM | EPOLLRDBAND); 3140 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN); 3141 rcu_read_unlock(); 3142 } 3143 3144 static void sock_def_write_space(struct sock *sk) 3145 { 3146 struct socket_wq *wq; 3147 3148 rcu_read_lock(); 3149 3150 /* Do not wake up a writer until he can make "significant" 3151 * progress. --DaveM 3152 */ 3153 if ((refcount_read(&sk->sk_wmem_alloc) << 1) <= READ_ONCE(sk->sk_sndbuf)) { 3154 wq = rcu_dereference(sk->sk_wq); 3155 if (skwq_has_sleeper(wq)) 3156 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT | 3157 EPOLLWRNORM | EPOLLWRBAND); 3158 3159 /* Should agree with poll, otherwise some programs break */ 3160 if (sock_writeable(sk)) 3161 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT); 3162 } 3163 3164 rcu_read_unlock(); 3165 } 3166 3167 static void sock_def_destruct(struct sock *sk) 3168 { 3169 } 3170 3171 void sk_send_sigurg(struct sock *sk) 3172 { 3173 if (sk->sk_socket && sk->sk_socket->file) 3174 if (send_sigurg(&sk->sk_socket->file->f_owner)) 3175 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI); 3176 } 3177 EXPORT_SYMBOL(sk_send_sigurg); 3178 3179 void sk_reset_timer(struct sock *sk, struct timer_list* timer, 3180 unsigned long expires) 3181 { 3182 if (!mod_timer(timer, expires)) 3183 sock_hold(sk); 3184 } 3185 EXPORT_SYMBOL(sk_reset_timer); 3186 3187 void sk_stop_timer(struct sock *sk, struct timer_list* timer) 3188 { 3189 if (del_timer(timer)) 3190 __sock_put(sk); 3191 } 3192 EXPORT_SYMBOL(sk_stop_timer); 3193 3194 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer) 3195 { 3196 if (del_timer_sync(timer)) 3197 __sock_put(sk); 3198 } 3199 EXPORT_SYMBOL(sk_stop_timer_sync); 3200 3201 void sock_init_data(struct socket *sock, struct sock *sk) 3202 { 3203 sk_init_common(sk); 3204 sk->sk_send_head = NULL; 3205 3206 timer_setup(&sk->sk_timer, NULL, 0); 3207 3208 sk->sk_allocation = GFP_KERNEL; 3209 sk->sk_rcvbuf = sysctl_rmem_default; 3210 sk->sk_sndbuf = sysctl_wmem_default; 3211 sk->sk_state = TCP_CLOSE; 3212 sk_set_socket(sk, sock); 3213 3214 sock_set_flag(sk, SOCK_ZAPPED); 3215 3216 if (sock) { 3217 sk->sk_type = sock->type; 3218 RCU_INIT_POINTER(sk->sk_wq, &sock->wq); 3219 sock->sk = sk; 3220 sk->sk_uid = SOCK_INODE(sock)->i_uid; 3221 } else { 3222 RCU_INIT_POINTER(sk->sk_wq, NULL); 3223 sk->sk_uid = make_kuid(sock_net(sk)->user_ns, 0); 3224 } 3225 3226 rwlock_init(&sk->sk_callback_lock); 3227 if (sk->sk_kern_sock) 3228 lockdep_set_class_and_name( 3229 &sk->sk_callback_lock, 3230 af_kern_callback_keys + sk->sk_family, 3231 af_family_kern_clock_key_strings[sk->sk_family]); 3232 else 3233 lockdep_set_class_and_name( 3234 &sk->sk_callback_lock, 3235 af_callback_keys + sk->sk_family, 3236 af_family_clock_key_strings[sk->sk_family]); 3237 3238 sk->sk_state_change = sock_def_wakeup; 3239 sk->sk_data_ready = sock_def_readable; 3240 sk->sk_write_space = sock_def_write_space; 3241 sk->sk_error_report = sock_def_error_report; 3242 sk->sk_destruct = sock_def_destruct; 3243 3244 sk->sk_frag.page = NULL; 3245 sk->sk_frag.offset = 0; 3246 sk->sk_peek_off = -1; 3247 3248 sk->sk_peer_pid = NULL; 3249 sk->sk_peer_cred = NULL; 3250 spin_lock_init(&sk->sk_peer_lock); 3251 3252 sk->sk_write_pending = 0; 3253 sk->sk_rcvlowat = 1; 3254 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; 3255 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT; 3256 3257 sk->sk_stamp = SK_DEFAULT_STAMP; 3258 #if BITS_PER_LONG==32 3259 seqlock_init(&sk->sk_stamp_seq); 3260 #endif 3261 atomic_set(&sk->sk_zckey, 0); 3262 3263 #ifdef CONFIG_NET_RX_BUSY_POLL 3264 sk->sk_napi_id = 0; 3265 sk->sk_ll_usec = sysctl_net_busy_read; 3266 #endif 3267 3268 sk->sk_max_pacing_rate = ~0UL; 3269 sk->sk_pacing_rate = ~0UL; 3270 WRITE_ONCE(sk->sk_pacing_shift, 10); 3271 sk->sk_incoming_cpu = -1; 3272 3273 sk_rx_queue_clear(sk); 3274 /* 3275 * Before updating sk_refcnt, we must commit prior changes to memory 3276 * (Documentation/RCU/rculist_nulls.rst for details) 3277 */ 3278 smp_wmb(); 3279 refcount_set(&sk->sk_refcnt, 1); 3280 atomic_set(&sk->sk_drops, 0); 3281 } 3282 EXPORT_SYMBOL(sock_init_data); 3283 3284 void lock_sock_nested(struct sock *sk, int subclass) 3285 { 3286 /* The sk_lock has mutex_lock() semantics here. */ 3287 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_); 3288 3289 might_sleep(); 3290 spin_lock_bh(&sk->sk_lock.slock); 3291 if (sk->sk_lock.owned) 3292 __lock_sock(sk); 3293 sk->sk_lock.owned = 1; 3294 spin_unlock_bh(&sk->sk_lock.slock); 3295 } 3296 EXPORT_SYMBOL(lock_sock_nested); 3297 3298 void release_sock(struct sock *sk) 3299 { 3300 spin_lock_bh(&sk->sk_lock.slock); 3301 if (sk->sk_backlog.tail) 3302 __release_sock(sk); 3303 3304 /* Warning : release_cb() might need to release sk ownership, 3305 * ie call sock_release_ownership(sk) before us. 3306 */ 3307 if (sk->sk_prot->release_cb) 3308 sk->sk_prot->release_cb(sk); 3309 3310 sock_release_ownership(sk); 3311 if (waitqueue_active(&sk->sk_lock.wq)) 3312 wake_up(&sk->sk_lock.wq); 3313 spin_unlock_bh(&sk->sk_lock.slock); 3314 } 3315 EXPORT_SYMBOL(release_sock); 3316 3317 bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock) 3318 { 3319 might_sleep(); 3320 spin_lock_bh(&sk->sk_lock.slock); 3321 3322 if (!sk->sk_lock.owned) { 3323 /* 3324 * Fast path return with bottom halves disabled and 3325 * sock::sk_lock.slock held. 3326 * 3327 * The 'mutex' is not contended and holding 3328 * sock::sk_lock.slock prevents all other lockers to 3329 * proceed so the corresponding unlock_sock_fast() can 3330 * avoid the slow path of release_sock() completely and 3331 * just release slock. 3332 * 3333 * From a semantical POV this is equivalent to 'acquiring' 3334 * the 'mutex', hence the corresponding lockdep 3335 * mutex_release() has to happen in the fast path of 3336 * unlock_sock_fast(). 3337 */ 3338 return false; 3339 } 3340 3341 __lock_sock(sk); 3342 sk->sk_lock.owned = 1; 3343 __acquire(&sk->sk_lock.slock); 3344 spin_unlock_bh(&sk->sk_lock.slock); 3345 return true; 3346 } 3347 EXPORT_SYMBOL(__lock_sock_fast); 3348 3349 int sock_gettstamp(struct socket *sock, void __user *userstamp, 3350 bool timeval, bool time32) 3351 { 3352 struct sock *sk = sock->sk; 3353 struct timespec64 ts; 3354 3355 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 3356 ts = ktime_to_timespec64(sock_read_timestamp(sk)); 3357 if (ts.tv_sec == -1) 3358 return -ENOENT; 3359 if (ts.tv_sec == 0) { 3360 ktime_t kt = ktime_get_real(); 3361 sock_write_timestamp(sk, kt); 3362 ts = ktime_to_timespec64(kt); 3363 } 3364 3365 if (timeval) 3366 ts.tv_nsec /= 1000; 3367 3368 #ifdef CONFIG_COMPAT_32BIT_TIME 3369 if (time32) 3370 return put_old_timespec32(&ts, userstamp); 3371 #endif 3372 #ifdef CONFIG_SPARC64 3373 /* beware of padding in sparc64 timeval */ 3374 if (timeval && !in_compat_syscall()) { 3375 struct __kernel_old_timeval __user tv = { 3376 .tv_sec = ts.tv_sec, 3377 .tv_usec = ts.tv_nsec, 3378 }; 3379 if (copy_to_user(userstamp, &tv, sizeof(tv))) 3380 return -EFAULT; 3381 return 0; 3382 } 3383 #endif 3384 return put_timespec64(&ts, userstamp); 3385 } 3386 EXPORT_SYMBOL(sock_gettstamp); 3387 3388 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag) 3389 { 3390 if (!sock_flag(sk, flag)) { 3391 unsigned long previous_flags = sk->sk_flags; 3392 3393 sock_set_flag(sk, flag); 3394 /* 3395 * we just set one of the two flags which require net 3396 * time stamping, but time stamping might have been on 3397 * already because of the other one 3398 */ 3399 if (sock_needs_netstamp(sk) && 3400 !(previous_flags & SK_FLAGS_TIMESTAMP)) 3401 net_enable_timestamp(); 3402 } 3403 } 3404 3405 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, 3406 int level, int type) 3407 { 3408 struct sock_exterr_skb *serr; 3409 struct sk_buff *skb; 3410 int copied, err; 3411 3412 err = -EAGAIN; 3413 skb = sock_dequeue_err_skb(sk); 3414 if (skb == NULL) 3415 goto out; 3416 3417 copied = skb->len; 3418 if (copied > len) { 3419 msg->msg_flags |= MSG_TRUNC; 3420 copied = len; 3421 } 3422 err = skb_copy_datagram_msg(skb, 0, msg, copied); 3423 if (err) 3424 goto out_free_skb; 3425 3426 sock_recv_timestamp(msg, sk, skb); 3427 3428 serr = SKB_EXT_ERR(skb); 3429 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee); 3430 3431 msg->msg_flags |= MSG_ERRQUEUE; 3432 err = copied; 3433 3434 out_free_skb: 3435 kfree_skb(skb); 3436 out: 3437 return err; 3438 } 3439 EXPORT_SYMBOL(sock_recv_errqueue); 3440 3441 /* 3442 * Get a socket option on an socket. 3443 * 3444 * FIX: POSIX 1003.1g is very ambiguous here. It states that 3445 * asynchronous errors should be reported by getsockopt. We assume 3446 * this means if you specify SO_ERROR (otherwise whats the point of it). 3447 */ 3448 int sock_common_getsockopt(struct socket *sock, int level, int optname, 3449 char __user *optval, int __user *optlen) 3450 { 3451 struct sock *sk = sock->sk; 3452 3453 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen); 3454 } 3455 EXPORT_SYMBOL(sock_common_getsockopt); 3456 3457 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 3458 int flags) 3459 { 3460 struct sock *sk = sock->sk; 3461 int addr_len = 0; 3462 int err; 3463 3464 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT, 3465 flags & ~MSG_DONTWAIT, &addr_len); 3466 if (err >= 0) 3467 msg->msg_namelen = addr_len; 3468 return err; 3469 } 3470 EXPORT_SYMBOL(sock_common_recvmsg); 3471 3472 /* 3473 * Set socket options on an inet socket. 3474 */ 3475 int sock_common_setsockopt(struct socket *sock, int level, int optname, 3476 sockptr_t optval, unsigned int optlen) 3477 { 3478 struct sock *sk = sock->sk; 3479 3480 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen); 3481 } 3482 EXPORT_SYMBOL(sock_common_setsockopt); 3483 3484 void sk_common_release(struct sock *sk) 3485 { 3486 if (sk->sk_prot->destroy) 3487 sk->sk_prot->destroy(sk); 3488 3489 /* 3490 * Observation: when sk_common_release is called, processes have 3491 * no access to socket. But net still has. 3492 * Step one, detach it from networking: 3493 * 3494 * A. Remove from hash tables. 3495 */ 3496 3497 sk->sk_prot->unhash(sk); 3498 3499 /* 3500 * In this point socket cannot receive new packets, but it is possible 3501 * that some packets are in flight because some CPU runs receiver and 3502 * did hash table lookup before we unhashed socket. They will achieve 3503 * receive queue and will be purged by socket destructor. 3504 * 3505 * Also we still have packets pending on receive queue and probably, 3506 * our own packets waiting in device queues. sock_destroy will drain 3507 * receive queue, but transmitted packets will delay socket destruction 3508 * until the last reference will be released. 3509 */ 3510 3511 sock_orphan(sk); 3512 3513 xfrm_sk_free_policy(sk); 3514 3515 sk_refcnt_debug_release(sk); 3516 3517 sock_put(sk); 3518 } 3519 EXPORT_SYMBOL(sk_common_release); 3520 3521 void sk_get_meminfo(const struct sock *sk, u32 *mem) 3522 { 3523 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS); 3524 3525 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk); 3526 mem[SK_MEMINFO_RCVBUF] = READ_ONCE(sk->sk_rcvbuf); 3527 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk); 3528 mem[SK_MEMINFO_SNDBUF] = READ_ONCE(sk->sk_sndbuf); 3529 mem[SK_MEMINFO_FWD_ALLOC] = sk->sk_forward_alloc; 3530 mem[SK_MEMINFO_WMEM_QUEUED] = READ_ONCE(sk->sk_wmem_queued); 3531 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc); 3532 mem[SK_MEMINFO_BACKLOG] = READ_ONCE(sk->sk_backlog.len); 3533 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops); 3534 } 3535 3536 #ifdef CONFIG_PROC_FS 3537 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR); 3538 3539 int sock_prot_inuse_get(struct net *net, struct proto *prot) 3540 { 3541 int cpu, idx = prot->inuse_idx; 3542 int res = 0; 3543 3544 for_each_possible_cpu(cpu) 3545 res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx]; 3546 3547 return res >= 0 ? res : 0; 3548 } 3549 EXPORT_SYMBOL_GPL(sock_prot_inuse_get); 3550 3551 int sock_inuse_get(struct net *net) 3552 { 3553 int cpu, res = 0; 3554 3555 for_each_possible_cpu(cpu) 3556 res += per_cpu_ptr(net->core.prot_inuse, cpu)->all; 3557 3558 return res; 3559 } 3560 3561 EXPORT_SYMBOL_GPL(sock_inuse_get); 3562 3563 static int __net_init sock_inuse_init_net(struct net *net) 3564 { 3565 net->core.prot_inuse = alloc_percpu(struct prot_inuse); 3566 if (net->core.prot_inuse == NULL) 3567 return -ENOMEM; 3568 return 0; 3569 } 3570 3571 static void __net_exit sock_inuse_exit_net(struct net *net) 3572 { 3573 free_percpu(net->core.prot_inuse); 3574 } 3575 3576 static struct pernet_operations net_inuse_ops = { 3577 .init = sock_inuse_init_net, 3578 .exit = sock_inuse_exit_net, 3579 }; 3580 3581 static __init int net_inuse_init(void) 3582 { 3583 if (register_pernet_subsys(&net_inuse_ops)) 3584 panic("Cannot initialize net inuse counters"); 3585 3586 return 0; 3587 } 3588 3589 core_initcall(net_inuse_init); 3590 3591 static int assign_proto_idx(struct proto *prot) 3592 { 3593 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR); 3594 3595 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) { 3596 pr_err("PROTO_INUSE_NR exhausted\n"); 3597 return -ENOSPC; 3598 } 3599 3600 set_bit(prot->inuse_idx, proto_inuse_idx); 3601 return 0; 3602 } 3603 3604 static void release_proto_idx(struct proto *prot) 3605 { 3606 if (prot->inuse_idx != PROTO_INUSE_NR - 1) 3607 clear_bit(prot->inuse_idx, proto_inuse_idx); 3608 } 3609 #else 3610 static inline int assign_proto_idx(struct proto *prot) 3611 { 3612 return 0; 3613 } 3614 3615 static inline void release_proto_idx(struct proto *prot) 3616 { 3617 } 3618 3619 #endif 3620 3621 static void tw_prot_cleanup(struct timewait_sock_ops *twsk_prot) 3622 { 3623 if (!twsk_prot) 3624 return; 3625 kfree(twsk_prot->twsk_slab_name); 3626 twsk_prot->twsk_slab_name = NULL; 3627 kmem_cache_destroy(twsk_prot->twsk_slab); 3628 twsk_prot->twsk_slab = NULL; 3629 } 3630 3631 static int tw_prot_init(const struct proto *prot) 3632 { 3633 struct timewait_sock_ops *twsk_prot = prot->twsk_prot; 3634 3635 if (!twsk_prot) 3636 return 0; 3637 3638 twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", 3639 prot->name); 3640 if (!twsk_prot->twsk_slab_name) 3641 return -ENOMEM; 3642 3643 twsk_prot->twsk_slab = 3644 kmem_cache_create(twsk_prot->twsk_slab_name, 3645 twsk_prot->twsk_obj_size, 0, 3646 SLAB_ACCOUNT | prot->slab_flags, 3647 NULL); 3648 if (!twsk_prot->twsk_slab) { 3649 pr_crit("%s: Can't create timewait sock SLAB cache!\n", 3650 prot->name); 3651 return -ENOMEM; 3652 } 3653 3654 return 0; 3655 } 3656 3657 static void req_prot_cleanup(struct request_sock_ops *rsk_prot) 3658 { 3659 if (!rsk_prot) 3660 return; 3661 kfree(rsk_prot->slab_name); 3662 rsk_prot->slab_name = NULL; 3663 kmem_cache_destroy(rsk_prot->slab); 3664 rsk_prot->slab = NULL; 3665 } 3666 3667 static int req_prot_init(const struct proto *prot) 3668 { 3669 struct request_sock_ops *rsk_prot = prot->rsk_prot; 3670 3671 if (!rsk_prot) 3672 return 0; 3673 3674 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", 3675 prot->name); 3676 if (!rsk_prot->slab_name) 3677 return -ENOMEM; 3678 3679 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name, 3680 rsk_prot->obj_size, 0, 3681 SLAB_ACCOUNT | prot->slab_flags, 3682 NULL); 3683 3684 if (!rsk_prot->slab) { 3685 pr_crit("%s: Can't create request sock SLAB cache!\n", 3686 prot->name); 3687 return -ENOMEM; 3688 } 3689 return 0; 3690 } 3691 3692 int proto_register(struct proto *prot, int alloc_slab) 3693 { 3694 int ret = -ENOBUFS; 3695 3696 if (alloc_slab) { 3697 prot->slab = kmem_cache_create_usercopy(prot->name, 3698 prot->obj_size, 0, 3699 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT | 3700 prot->slab_flags, 3701 prot->useroffset, prot->usersize, 3702 NULL); 3703 3704 if (prot->slab == NULL) { 3705 pr_crit("%s: Can't create sock SLAB cache!\n", 3706 prot->name); 3707 goto out; 3708 } 3709 3710 if (req_prot_init(prot)) 3711 goto out_free_request_sock_slab; 3712 3713 if (tw_prot_init(prot)) 3714 goto out_free_timewait_sock_slab; 3715 } 3716 3717 mutex_lock(&proto_list_mutex); 3718 ret = assign_proto_idx(prot); 3719 if (ret) { 3720 mutex_unlock(&proto_list_mutex); 3721 goto out_free_timewait_sock_slab; 3722 } 3723 list_add(&prot->node, &proto_list); 3724 mutex_unlock(&proto_list_mutex); 3725 return ret; 3726 3727 out_free_timewait_sock_slab: 3728 if (alloc_slab) 3729 tw_prot_cleanup(prot->twsk_prot); 3730 out_free_request_sock_slab: 3731 if (alloc_slab) { 3732 req_prot_cleanup(prot->rsk_prot); 3733 3734 kmem_cache_destroy(prot->slab); 3735 prot->slab = NULL; 3736 } 3737 out: 3738 return ret; 3739 } 3740 EXPORT_SYMBOL(proto_register); 3741 3742 void proto_unregister(struct proto *prot) 3743 { 3744 mutex_lock(&proto_list_mutex); 3745 release_proto_idx(prot); 3746 list_del(&prot->node); 3747 mutex_unlock(&proto_list_mutex); 3748 3749 kmem_cache_destroy(prot->slab); 3750 prot->slab = NULL; 3751 3752 req_prot_cleanup(prot->rsk_prot); 3753 tw_prot_cleanup(prot->twsk_prot); 3754 } 3755 EXPORT_SYMBOL(proto_unregister); 3756 3757 int sock_load_diag_module(int family, int protocol) 3758 { 3759 if (!protocol) { 3760 if (!sock_is_registered(family)) 3761 return -ENOENT; 3762 3763 return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK, 3764 NETLINK_SOCK_DIAG, family); 3765 } 3766 3767 #ifdef CONFIG_INET 3768 if (family == AF_INET && 3769 protocol != IPPROTO_RAW && 3770 protocol < MAX_INET_PROTOS && 3771 !rcu_access_pointer(inet_protos[protocol])) 3772 return -ENOENT; 3773 #endif 3774 3775 return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK, 3776 NETLINK_SOCK_DIAG, family, protocol); 3777 } 3778 EXPORT_SYMBOL(sock_load_diag_module); 3779 3780 #ifdef CONFIG_PROC_FS 3781 static void *proto_seq_start(struct seq_file *seq, loff_t *pos) 3782 __acquires(proto_list_mutex) 3783 { 3784 mutex_lock(&proto_list_mutex); 3785 return seq_list_start_head(&proto_list, *pos); 3786 } 3787 3788 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos) 3789 { 3790 return seq_list_next(v, &proto_list, pos); 3791 } 3792 3793 static void proto_seq_stop(struct seq_file *seq, void *v) 3794 __releases(proto_list_mutex) 3795 { 3796 mutex_unlock(&proto_list_mutex); 3797 } 3798 3799 static char proto_method_implemented(const void *method) 3800 { 3801 return method == NULL ? 'n' : 'y'; 3802 } 3803 static long sock_prot_memory_allocated(struct proto *proto) 3804 { 3805 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L; 3806 } 3807 3808 static const char *sock_prot_memory_pressure(struct proto *proto) 3809 { 3810 return proto->memory_pressure != NULL ? 3811 proto_memory_pressure(proto) ? "yes" : "no" : "NI"; 3812 } 3813 3814 static void proto_seq_printf(struct seq_file *seq, struct proto *proto) 3815 { 3816 3817 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s " 3818 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n", 3819 proto->name, 3820 proto->obj_size, 3821 sock_prot_inuse_get(seq_file_net(seq), proto), 3822 sock_prot_memory_allocated(proto), 3823 sock_prot_memory_pressure(proto), 3824 proto->max_header, 3825 proto->slab == NULL ? "no" : "yes", 3826 module_name(proto->owner), 3827 proto_method_implemented(proto->close), 3828 proto_method_implemented(proto->connect), 3829 proto_method_implemented(proto->disconnect), 3830 proto_method_implemented(proto->accept), 3831 proto_method_implemented(proto->ioctl), 3832 proto_method_implemented(proto->init), 3833 proto_method_implemented(proto->destroy), 3834 proto_method_implemented(proto->shutdown), 3835 proto_method_implemented(proto->setsockopt), 3836 proto_method_implemented(proto->getsockopt), 3837 proto_method_implemented(proto->sendmsg), 3838 proto_method_implemented(proto->recvmsg), 3839 proto_method_implemented(proto->sendpage), 3840 proto_method_implemented(proto->bind), 3841 proto_method_implemented(proto->backlog_rcv), 3842 proto_method_implemented(proto->hash), 3843 proto_method_implemented(proto->unhash), 3844 proto_method_implemented(proto->get_port), 3845 proto_method_implemented(proto->enter_memory_pressure)); 3846 } 3847 3848 static int proto_seq_show(struct seq_file *seq, void *v) 3849 { 3850 if (v == &proto_list) 3851 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s", 3852 "protocol", 3853 "size", 3854 "sockets", 3855 "memory", 3856 "press", 3857 "maxhdr", 3858 "slab", 3859 "module", 3860 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n"); 3861 else 3862 proto_seq_printf(seq, list_entry(v, struct proto, node)); 3863 return 0; 3864 } 3865 3866 static const struct seq_operations proto_seq_ops = { 3867 .start = proto_seq_start, 3868 .next = proto_seq_next, 3869 .stop = proto_seq_stop, 3870 .show = proto_seq_show, 3871 }; 3872 3873 static __net_init int proto_init_net(struct net *net) 3874 { 3875 if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops, 3876 sizeof(struct seq_net_private))) 3877 return -ENOMEM; 3878 3879 return 0; 3880 } 3881 3882 static __net_exit void proto_exit_net(struct net *net) 3883 { 3884 remove_proc_entry("protocols", net->proc_net); 3885 } 3886 3887 3888 static __net_initdata struct pernet_operations proto_net_ops = { 3889 .init = proto_init_net, 3890 .exit = proto_exit_net, 3891 }; 3892 3893 static int __init proto_init(void) 3894 { 3895 return register_pernet_subsys(&proto_net_ops); 3896 } 3897 3898 subsys_initcall(proto_init); 3899 3900 #endif /* PROC_FS */ 3901 3902 #ifdef CONFIG_NET_RX_BUSY_POLL 3903 bool sk_busy_loop_end(void *p, unsigned long start_time) 3904 { 3905 struct sock *sk = p; 3906 3907 return !skb_queue_empty_lockless(&sk->sk_receive_queue) || 3908 sk_busy_loop_timeout(sk, start_time); 3909 } 3910 EXPORT_SYMBOL(sk_busy_loop_end); 3911 #endif /* CONFIG_NET_RX_BUSY_POLL */ 3912 3913 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len) 3914 { 3915 if (!sk->sk_prot->bind_add) 3916 return -EOPNOTSUPP; 3917 return sk->sk_prot->bind_add(sk, addr, addr_len); 3918 } 3919 EXPORT_SYMBOL(sock_bind_add); 3920