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_RAW) || 1139 ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) 1140 sk->sk_priority = val; 1141 else 1142 ret = -EPERM; 1143 break; 1144 1145 case SO_LINGER: 1146 if (optlen < sizeof(ling)) { 1147 ret = -EINVAL; /* 1003.1g */ 1148 break; 1149 } 1150 if (copy_from_sockptr(&ling, optval, sizeof(ling))) { 1151 ret = -EFAULT; 1152 break; 1153 } 1154 if (!ling.l_onoff) 1155 sock_reset_flag(sk, SOCK_LINGER); 1156 else { 1157 #if (BITS_PER_LONG == 32) 1158 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ) 1159 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT; 1160 else 1161 #endif 1162 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ; 1163 sock_set_flag(sk, SOCK_LINGER); 1164 } 1165 break; 1166 1167 case SO_BSDCOMPAT: 1168 break; 1169 1170 case SO_PASSCRED: 1171 if (valbool) 1172 set_bit(SOCK_PASSCRED, &sock->flags); 1173 else 1174 clear_bit(SOCK_PASSCRED, &sock->flags); 1175 break; 1176 1177 case SO_TIMESTAMP_OLD: 1178 case SO_TIMESTAMP_NEW: 1179 case SO_TIMESTAMPNS_OLD: 1180 case SO_TIMESTAMPNS_NEW: 1181 sock_set_timestamp(sk, optname, valbool); 1182 break; 1183 1184 case SO_TIMESTAMPING_NEW: 1185 case SO_TIMESTAMPING_OLD: 1186 if (optlen == sizeof(timestamping)) { 1187 if (copy_from_sockptr(×tamping, optval, 1188 sizeof(timestamping))) { 1189 ret = -EFAULT; 1190 break; 1191 } 1192 } else { 1193 memset(×tamping, 0, sizeof(timestamping)); 1194 timestamping.flags = val; 1195 } 1196 ret = sock_set_timestamping(sk, optname, timestamping); 1197 break; 1198 1199 case SO_RCVLOWAT: 1200 if (val < 0) 1201 val = INT_MAX; 1202 if (sock->ops->set_rcvlowat) 1203 ret = sock->ops->set_rcvlowat(sk, val); 1204 else 1205 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1); 1206 break; 1207 1208 case SO_RCVTIMEO_OLD: 1209 case SO_RCVTIMEO_NEW: 1210 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, 1211 optlen, optname == SO_RCVTIMEO_OLD); 1212 break; 1213 1214 case SO_SNDTIMEO_OLD: 1215 case SO_SNDTIMEO_NEW: 1216 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, 1217 optlen, optname == SO_SNDTIMEO_OLD); 1218 break; 1219 1220 case SO_ATTACH_FILTER: { 1221 struct sock_fprog fprog; 1222 1223 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen); 1224 if (!ret) 1225 ret = sk_attach_filter(&fprog, sk); 1226 break; 1227 } 1228 case SO_ATTACH_BPF: 1229 ret = -EINVAL; 1230 if (optlen == sizeof(u32)) { 1231 u32 ufd; 1232 1233 ret = -EFAULT; 1234 if (copy_from_sockptr(&ufd, optval, sizeof(ufd))) 1235 break; 1236 1237 ret = sk_attach_bpf(ufd, sk); 1238 } 1239 break; 1240 1241 case SO_ATTACH_REUSEPORT_CBPF: { 1242 struct sock_fprog fprog; 1243 1244 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen); 1245 if (!ret) 1246 ret = sk_reuseport_attach_filter(&fprog, sk); 1247 break; 1248 } 1249 case SO_ATTACH_REUSEPORT_EBPF: 1250 ret = -EINVAL; 1251 if (optlen == sizeof(u32)) { 1252 u32 ufd; 1253 1254 ret = -EFAULT; 1255 if (copy_from_sockptr(&ufd, optval, sizeof(ufd))) 1256 break; 1257 1258 ret = sk_reuseport_attach_bpf(ufd, sk); 1259 } 1260 break; 1261 1262 case SO_DETACH_REUSEPORT_BPF: 1263 ret = reuseport_detach_prog(sk); 1264 break; 1265 1266 case SO_DETACH_FILTER: 1267 ret = sk_detach_filter(sk); 1268 break; 1269 1270 case SO_LOCK_FILTER: 1271 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool) 1272 ret = -EPERM; 1273 else 1274 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool); 1275 break; 1276 1277 case SO_PASSSEC: 1278 if (valbool) 1279 set_bit(SOCK_PASSSEC, &sock->flags); 1280 else 1281 clear_bit(SOCK_PASSSEC, &sock->flags); 1282 break; 1283 case SO_MARK: 1284 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) && 1285 !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) { 1286 ret = -EPERM; 1287 break; 1288 } 1289 1290 __sock_set_mark(sk, val); 1291 break; 1292 1293 case SO_RXQ_OVFL: 1294 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool); 1295 break; 1296 1297 case SO_WIFI_STATUS: 1298 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool); 1299 break; 1300 1301 case SO_PEEK_OFF: 1302 if (sock->ops->set_peek_off) 1303 ret = sock->ops->set_peek_off(sk, val); 1304 else 1305 ret = -EOPNOTSUPP; 1306 break; 1307 1308 case SO_NOFCS: 1309 sock_valbool_flag(sk, SOCK_NOFCS, valbool); 1310 break; 1311 1312 case SO_SELECT_ERR_QUEUE: 1313 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool); 1314 break; 1315 1316 #ifdef CONFIG_NET_RX_BUSY_POLL 1317 case SO_BUSY_POLL: 1318 /* allow unprivileged users to decrease the value */ 1319 if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN)) 1320 ret = -EPERM; 1321 else { 1322 if (val < 0) 1323 ret = -EINVAL; 1324 else 1325 WRITE_ONCE(sk->sk_ll_usec, val); 1326 } 1327 break; 1328 case SO_PREFER_BUSY_POLL: 1329 if (valbool && !capable(CAP_NET_ADMIN)) 1330 ret = -EPERM; 1331 else 1332 WRITE_ONCE(sk->sk_prefer_busy_poll, valbool); 1333 break; 1334 case SO_BUSY_POLL_BUDGET: 1335 if (val > READ_ONCE(sk->sk_busy_poll_budget) && !capable(CAP_NET_ADMIN)) { 1336 ret = -EPERM; 1337 } else { 1338 if (val < 0 || val > U16_MAX) 1339 ret = -EINVAL; 1340 else 1341 WRITE_ONCE(sk->sk_busy_poll_budget, val); 1342 } 1343 break; 1344 #endif 1345 1346 case SO_MAX_PACING_RATE: 1347 { 1348 unsigned long ulval = (val == ~0U) ? ~0UL : (unsigned int)val; 1349 1350 if (sizeof(ulval) != sizeof(val) && 1351 optlen >= sizeof(ulval) && 1352 copy_from_sockptr(&ulval, optval, sizeof(ulval))) { 1353 ret = -EFAULT; 1354 break; 1355 } 1356 if (ulval != ~0UL) 1357 cmpxchg(&sk->sk_pacing_status, 1358 SK_PACING_NONE, 1359 SK_PACING_NEEDED); 1360 sk->sk_max_pacing_rate = ulval; 1361 sk->sk_pacing_rate = min(sk->sk_pacing_rate, ulval); 1362 break; 1363 } 1364 case SO_INCOMING_CPU: 1365 WRITE_ONCE(sk->sk_incoming_cpu, val); 1366 break; 1367 1368 case SO_CNX_ADVICE: 1369 if (val == 1) 1370 dst_negative_advice(sk); 1371 break; 1372 1373 case SO_ZEROCOPY: 1374 if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) { 1375 if (!(sk_is_tcp(sk) || 1376 (sk->sk_type == SOCK_DGRAM && 1377 sk->sk_protocol == IPPROTO_UDP))) 1378 ret = -ENOTSUPP; 1379 } else if (sk->sk_family != PF_RDS) { 1380 ret = -ENOTSUPP; 1381 } 1382 if (!ret) { 1383 if (val < 0 || val > 1) 1384 ret = -EINVAL; 1385 else 1386 sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool); 1387 } 1388 break; 1389 1390 case SO_TXTIME: 1391 if (optlen != sizeof(struct sock_txtime)) { 1392 ret = -EINVAL; 1393 break; 1394 } else if (copy_from_sockptr(&sk_txtime, optval, 1395 sizeof(struct sock_txtime))) { 1396 ret = -EFAULT; 1397 break; 1398 } else if (sk_txtime.flags & ~SOF_TXTIME_FLAGS_MASK) { 1399 ret = -EINVAL; 1400 break; 1401 } 1402 /* CLOCK_MONOTONIC is only used by sch_fq, and this packet 1403 * scheduler has enough safe guards. 1404 */ 1405 if (sk_txtime.clockid != CLOCK_MONOTONIC && 1406 !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) { 1407 ret = -EPERM; 1408 break; 1409 } 1410 sock_valbool_flag(sk, SOCK_TXTIME, true); 1411 sk->sk_clockid = sk_txtime.clockid; 1412 sk->sk_txtime_deadline_mode = 1413 !!(sk_txtime.flags & SOF_TXTIME_DEADLINE_MODE); 1414 sk->sk_txtime_report_errors = 1415 !!(sk_txtime.flags & SOF_TXTIME_REPORT_ERRORS); 1416 break; 1417 1418 case SO_BINDTOIFINDEX: 1419 ret = sock_bindtoindex_locked(sk, val); 1420 break; 1421 1422 case SO_BUF_LOCK: 1423 if (val & ~SOCK_BUF_LOCK_MASK) { 1424 ret = -EINVAL; 1425 break; 1426 } 1427 sk->sk_userlocks = val | (sk->sk_userlocks & 1428 ~SOCK_BUF_LOCK_MASK); 1429 break; 1430 1431 case SO_RESERVE_MEM: 1432 { 1433 int delta; 1434 1435 if (val < 0) { 1436 ret = -EINVAL; 1437 break; 1438 } 1439 1440 delta = val - sk->sk_reserved_mem; 1441 if (delta < 0) 1442 sock_release_reserved_memory(sk, -delta); 1443 else 1444 ret = sock_reserve_memory(sk, delta); 1445 break; 1446 } 1447 1448 default: 1449 ret = -ENOPROTOOPT; 1450 break; 1451 } 1452 release_sock(sk); 1453 return ret; 1454 } 1455 EXPORT_SYMBOL(sock_setsockopt); 1456 1457 static const struct cred *sk_get_peer_cred(struct sock *sk) 1458 { 1459 const struct cred *cred; 1460 1461 spin_lock(&sk->sk_peer_lock); 1462 cred = get_cred(sk->sk_peer_cred); 1463 spin_unlock(&sk->sk_peer_lock); 1464 1465 return cred; 1466 } 1467 1468 static void cred_to_ucred(struct pid *pid, const struct cred *cred, 1469 struct ucred *ucred) 1470 { 1471 ucred->pid = pid_vnr(pid); 1472 ucred->uid = ucred->gid = -1; 1473 if (cred) { 1474 struct user_namespace *current_ns = current_user_ns(); 1475 1476 ucred->uid = from_kuid_munged(current_ns, cred->euid); 1477 ucred->gid = from_kgid_munged(current_ns, cred->egid); 1478 } 1479 } 1480 1481 static int groups_to_user(gid_t __user *dst, const struct group_info *src) 1482 { 1483 struct user_namespace *user_ns = current_user_ns(); 1484 int i; 1485 1486 for (i = 0; i < src->ngroups; i++) 1487 if (put_user(from_kgid_munged(user_ns, src->gid[i]), dst + i)) 1488 return -EFAULT; 1489 1490 return 0; 1491 } 1492 1493 int sock_getsockopt(struct socket *sock, int level, int optname, 1494 char __user *optval, int __user *optlen) 1495 { 1496 struct sock *sk = sock->sk; 1497 1498 union { 1499 int val; 1500 u64 val64; 1501 unsigned long ulval; 1502 struct linger ling; 1503 struct old_timeval32 tm32; 1504 struct __kernel_old_timeval tm; 1505 struct __kernel_sock_timeval stm; 1506 struct sock_txtime txtime; 1507 struct so_timestamping timestamping; 1508 } v; 1509 1510 int lv = sizeof(int); 1511 int len; 1512 1513 if (get_user(len, optlen)) 1514 return -EFAULT; 1515 if (len < 0) 1516 return -EINVAL; 1517 1518 memset(&v, 0, sizeof(v)); 1519 1520 switch (optname) { 1521 case SO_DEBUG: 1522 v.val = sock_flag(sk, SOCK_DBG); 1523 break; 1524 1525 case SO_DONTROUTE: 1526 v.val = sock_flag(sk, SOCK_LOCALROUTE); 1527 break; 1528 1529 case SO_BROADCAST: 1530 v.val = sock_flag(sk, SOCK_BROADCAST); 1531 break; 1532 1533 case SO_SNDBUF: 1534 v.val = sk->sk_sndbuf; 1535 break; 1536 1537 case SO_RCVBUF: 1538 v.val = sk->sk_rcvbuf; 1539 break; 1540 1541 case SO_REUSEADDR: 1542 v.val = sk->sk_reuse; 1543 break; 1544 1545 case SO_REUSEPORT: 1546 v.val = sk->sk_reuseport; 1547 break; 1548 1549 case SO_KEEPALIVE: 1550 v.val = sock_flag(sk, SOCK_KEEPOPEN); 1551 break; 1552 1553 case SO_TYPE: 1554 v.val = sk->sk_type; 1555 break; 1556 1557 case SO_PROTOCOL: 1558 v.val = sk->sk_protocol; 1559 break; 1560 1561 case SO_DOMAIN: 1562 v.val = sk->sk_family; 1563 break; 1564 1565 case SO_ERROR: 1566 v.val = -sock_error(sk); 1567 if (v.val == 0) 1568 v.val = xchg(&sk->sk_err_soft, 0); 1569 break; 1570 1571 case SO_OOBINLINE: 1572 v.val = sock_flag(sk, SOCK_URGINLINE); 1573 break; 1574 1575 case SO_NO_CHECK: 1576 v.val = sk->sk_no_check_tx; 1577 break; 1578 1579 case SO_PRIORITY: 1580 v.val = sk->sk_priority; 1581 break; 1582 1583 case SO_LINGER: 1584 lv = sizeof(v.ling); 1585 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER); 1586 v.ling.l_linger = sk->sk_lingertime / HZ; 1587 break; 1588 1589 case SO_BSDCOMPAT: 1590 break; 1591 1592 case SO_TIMESTAMP_OLD: 1593 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && 1594 !sock_flag(sk, SOCK_TSTAMP_NEW) && 1595 !sock_flag(sk, SOCK_RCVTSTAMPNS); 1596 break; 1597 1598 case SO_TIMESTAMPNS_OLD: 1599 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && !sock_flag(sk, SOCK_TSTAMP_NEW); 1600 break; 1601 1602 case SO_TIMESTAMP_NEW: 1603 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && sock_flag(sk, SOCK_TSTAMP_NEW); 1604 break; 1605 1606 case SO_TIMESTAMPNS_NEW: 1607 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && sock_flag(sk, SOCK_TSTAMP_NEW); 1608 break; 1609 1610 case SO_TIMESTAMPING_OLD: 1611 lv = sizeof(v.timestamping); 1612 v.timestamping.flags = sk->sk_tsflags; 1613 v.timestamping.bind_phc = sk->sk_bind_phc; 1614 break; 1615 1616 case SO_RCVTIMEO_OLD: 1617 case SO_RCVTIMEO_NEW: 1618 lv = sock_get_timeout(sk->sk_rcvtimeo, &v, SO_RCVTIMEO_OLD == optname); 1619 break; 1620 1621 case SO_SNDTIMEO_OLD: 1622 case SO_SNDTIMEO_NEW: 1623 lv = sock_get_timeout(sk->sk_sndtimeo, &v, SO_SNDTIMEO_OLD == optname); 1624 break; 1625 1626 case SO_RCVLOWAT: 1627 v.val = sk->sk_rcvlowat; 1628 break; 1629 1630 case SO_SNDLOWAT: 1631 v.val = 1; 1632 break; 1633 1634 case SO_PASSCRED: 1635 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags); 1636 break; 1637 1638 case SO_PEERCRED: 1639 { 1640 struct ucred peercred; 1641 if (len > sizeof(peercred)) 1642 len = sizeof(peercred); 1643 1644 spin_lock(&sk->sk_peer_lock); 1645 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred); 1646 spin_unlock(&sk->sk_peer_lock); 1647 1648 if (copy_to_user(optval, &peercred, len)) 1649 return -EFAULT; 1650 goto lenout; 1651 } 1652 1653 case SO_PEERGROUPS: 1654 { 1655 const struct cred *cred; 1656 int ret, n; 1657 1658 cred = sk_get_peer_cred(sk); 1659 if (!cred) 1660 return -ENODATA; 1661 1662 n = cred->group_info->ngroups; 1663 if (len < n * sizeof(gid_t)) { 1664 len = n * sizeof(gid_t); 1665 put_cred(cred); 1666 return put_user(len, optlen) ? -EFAULT : -ERANGE; 1667 } 1668 len = n * sizeof(gid_t); 1669 1670 ret = groups_to_user((gid_t __user *)optval, cred->group_info); 1671 put_cred(cred); 1672 if (ret) 1673 return ret; 1674 goto lenout; 1675 } 1676 1677 case SO_PEERNAME: 1678 { 1679 char address[128]; 1680 1681 lv = sock->ops->getname(sock, (struct sockaddr *)address, 2); 1682 if (lv < 0) 1683 return -ENOTCONN; 1684 if (lv < len) 1685 return -EINVAL; 1686 if (copy_to_user(optval, address, len)) 1687 return -EFAULT; 1688 goto lenout; 1689 } 1690 1691 /* Dubious BSD thing... Probably nobody even uses it, but 1692 * the UNIX standard wants it for whatever reason... -DaveM 1693 */ 1694 case SO_ACCEPTCONN: 1695 v.val = sk->sk_state == TCP_LISTEN; 1696 break; 1697 1698 case SO_PASSSEC: 1699 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags); 1700 break; 1701 1702 case SO_PEERSEC: 1703 return security_socket_getpeersec_stream(sock, optval, optlen, len); 1704 1705 case SO_MARK: 1706 v.val = sk->sk_mark; 1707 break; 1708 1709 case SO_RXQ_OVFL: 1710 v.val = sock_flag(sk, SOCK_RXQ_OVFL); 1711 break; 1712 1713 case SO_WIFI_STATUS: 1714 v.val = sock_flag(sk, SOCK_WIFI_STATUS); 1715 break; 1716 1717 case SO_PEEK_OFF: 1718 if (!sock->ops->set_peek_off) 1719 return -EOPNOTSUPP; 1720 1721 v.val = sk->sk_peek_off; 1722 break; 1723 case SO_NOFCS: 1724 v.val = sock_flag(sk, SOCK_NOFCS); 1725 break; 1726 1727 case SO_BINDTODEVICE: 1728 return sock_getbindtodevice(sk, optval, optlen, len); 1729 1730 case SO_GET_FILTER: 1731 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len); 1732 if (len < 0) 1733 return len; 1734 1735 goto lenout; 1736 1737 case SO_LOCK_FILTER: 1738 v.val = sock_flag(sk, SOCK_FILTER_LOCKED); 1739 break; 1740 1741 case SO_BPF_EXTENSIONS: 1742 v.val = bpf_tell_extensions(); 1743 break; 1744 1745 case SO_SELECT_ERR_QUEUE: 1746 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE); 1747 break; 1748 1749 #ifdef CONFIG_NET_RX_BUSY_POLL 1750 case SO_BUSY_POLL: 1751 v.val = sk->sk_ll_usec; 1752 break; 1753 case SO_PREFER_BUSY_POLL: 1754 v.val = READ_ONCE(sk->sk_prefer_busy_poll); 1755 break; 1756 #endif 1757 1758 case SO_MAX_PACING_RATE: 1759 if (sizeof(v.ulval) != sizeof(v.val) && len >= sizeof(v.ulval)) { 1760 lv = sizeof(v.ulval); 1761 v.ulval = sk->sk_max_pacing_rate; 1762 } else { 1763 /* 32bit version */ 1764 v.val = min_t(unsigned long, sk->sk_max_pacing_rate, ~0U); 1765 } 1766 break; 1767 1768 case SO_INCOMING_CPU: 1769 v.val = READ_ONCE(sk->sk_incoming_cpu); 1770 break; 1771 1772 case SO_MEMINFO: 1773 { 1774 u32 meminfo[SK_MEMINFO_VARS]; 1775 1776 sk_get_meminfo(sk, meminfo); 1777 1778 len = min_t(unsigned int, len, sizeof(meminfo)); 1779 if (copy_to_user(optval, &meminfo, len)) 1780 return -EFAULT; 1781 1782 goto lenout; 1783 } 1784 1785 #ifdef CONFIG_NET_RX_BUSY_POLL 1786 case SO_INCOMING_NAPI_ID: 1787 v.val = READ_ONCE(sk->sk_napi_id); 1788 1789 /* aggregate non-NAPI IDs down to 0 */ 1790 if (v.val < MIN_NAPI_ID) 1791 v.val = 0; 1792 1793 break; 1794 #endif 1795 1796 case SO_COOKIE: 1797 lv = sizeof(u64); 1798 if (len < lv) 1799 return -EINVAL; 1800 v.val64 = sock_gen_cookie(sk); 1801 break; 1802 1803 case SO_ZEROCOPY: 1804 v.val = sock_flag(sk, SOCK_ZEROCOPY); 1805 break; 1806 1807 case SO_TXTIME: 1808 lv = sizeof(v.txtime); 1809 v.txtime.clockid = sk->sk_clockid; 1810 v.txtime.flags |= sk->sk_txtime_deadline_mode ? 1811 SOF_TXTIME_DEADLINE_MODE : 0; 1812 v.txtime.flags |= sk->sk_txtime_report_errors ? 1813 SOF_TXTIME_REPORT_ERRORS : 0; 1814 break; 1815 1816 case SO_BINDTOIFINDEX: 1817 v.val = sk->sk_bound_dev_if; 1818 break; 1819 1820 case SO_NETNS_COOKIE: 1821 lv = sizeof(u64); 1822 if (len != lv) 1823 return -EINVAL; 1824 v.val64 = sock_net(sk)->net_cookie; 1825 break; 1826 1827 case SO_BUF_LOCK: 1828 v.val = sk->sk_userlocks & SOCK_BUF_LOCK_MASK; 1829 break; 1830 1831 case SO_RESERVE_MEM: 1832 v.val = sk->sk_reserved_mem; 1833 break; 1834 1835 default: 1836 /* We implement the SO_SNDLOWAT etc to not be settable 1837 * (1003.1g 7). 1838 */ 1839 return -ENOPROTOOPT; 1840 } 1841 1842 if (len > lv) 1843 len = lv; 1844 if (copy_to_user(optval, &v, len)) 1845 return -EFAULT; 1846 lenout: 1847 if (put_user(len, optlen)) 1848 return -EFAULT; 1849 return 0; 1850 } 1851 1852 /* 1853 * Initialize an sk_lock. 1854 * 1855 * (We also register the sk_lock with the lock validator.) 1856 */ 1857 static inline void sock_lock_init(struct sock *sk) 1858 { 1859 if (sk->sk_kern_sock) 1860 sock_lock_init_class_and_name( 1861 sk, 1862 af_family_kern_slock_key_strings[sk->sk_family], 1863 af_family_kern_slock_keys + sk->sk_family, 1864 af_family_kern_key_strings[sk->sk_family], 1865 af_family_kern_keys + sk->sk_family); 1866 else 1867 sock_lock_init_class_and_name( 1868 sk, 1869 af_family_slock_key_strings[sk->sk_family], 1870 af_family_slock_keys + sk->sk_family, 1871 af_family_key_strings[sk->sk_family], 1872 af_family_keys + sk->sk_family); 1873 } 1874 1875 /* 1876 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet, 1877 * even temporarly, because of RCU lookups. sk_node should also be left as is. 1878 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end 1879 */ 1880 static void sock_copy(struct sock *nsk, const struct sock *osk) 1881 { 1882 const struct proto *prot = READ_ONCE(osk->sk_prot); 1883 #ifdef CONFIG_SECURITY_NETWORK 1884 void *sptr = nsk->sk_security; 1885 #endif 1886 1887 /* If we move sk_tx_queue_mapping out of the private section, 1888 * we must check if sk_tx_queue_clear() is called after 1889 * sock_copy() in sk_clone_lock(). 1890 */ 1891 BUILD_BUG_ON(offsetof(struct sock, sk_tx_queue_mapping) < 1892 offsetof(struct sock, sk_dontcopy_begin) || 1893 offsetof(struct sock, sk_tx_queue_mapping) >= 1894 offsetof(struct sock, sk_dontcopy_end)); 1895 1896 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin)); 1897 1898 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end, 1899 prot->obj_size - offsetof(struct sock, sk_dontcopy_end)); 1900 1901 #ifdef CONFIG_SECURITY_NETWORK 1902 nsk->sk_security = sptr; 1903 security_sk_clone(osk, nsk); 1904 #endif 1905 } 1906 1907 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority, 1908 int family) 1909 { 1910 struct sock *sk; 1911 struct kmem_cache *slab; 1912 1913 slab = prot->slab; 1914 if (slab != NULL) { 1915 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO); 1916 if (!sk) 1917 return sk; 1918 if (want_init_on_alloc(priority)) 1919 sk_prot_clear_nulls(sk, prot->obj_size); 1920 } else 1921 sk = kmalloc(prot->obj_size, priority); 1922 1923 if (sk != NULL) { 1924 if (security_sk_alloc(sk, family, priority)) 1925 goto out_free; 1926 1927 if (!try_module_get(prot->owner)) 1928 goto out_free_sec; 1929 } 1930 1931 return sk; 1932 1933 out_free_sec: 1934 security_sk_free(sk); 1935 out_free: 1936 if (slab != NULL) 1937 kmem_cache_free(slab, sk); 1938 else 1939 kfree(sk); 1940 return NULL; 1941 } 1942 1943 static void sk_prot_free(struct proto *prot, struct sock *sk) 1944 { 1945 struct kmem_cache *slab; 1946 struct module *owner; 1947 1948 owner = prot->owner; 1949 slab = prot->slab; 1950 1951 cgroup_sk_free(&sk->sk_cgrp_data); 1952 mem_cgroup_sk_free(sk); 1953 security_sk_free(sk); 1954 if (slab != NULL) 1955 kmem_cache_free(slab, sk); 1956 else 1957 kfree(sk); 1958 module_put(owner); 1959 } 1960 1961 /** 1962 * sk_alloc - All socket objects are allocated here 1963 * @net: the applicable net namespace 1964 * @family: protocol family 1965 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 1966 * @prot: struct proto associated with this new sock instance 1967 * @kern: is this to be a kernel socket? 1968 */ 1969 struct sock *sk_alloc(struct net *net, int family, gfp_t priority, 1970 struct proto *prot, int kern) 1971 { 1972 struct sock *sk; 1973 1974 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family); 1975 if (sk) { 1976 sk->sk_family = family; 1977 /* 1978 * See comment in struct sock definition to understand 1979 * why we need sk_prot_creator -acme 1980 */ 1981 sk->sk_prot = sk->sk_prot_creator = prot; 1982 sk->sk_kern_sock = kern; 1983 sock_lock_init(sk); 1984 sk->sk_net_refcnt = kern ? 0 : 1; 1985 if (likely(sk->sk_net_refcnt)) { 1986 get_net(net); 1987 sock_inuse_add(net, 1); 1988 } 1989 1990 sock_net_set(sk, net); 1991 refcount_set(&sk->sk_wmem_alloc, 1); 1992 1993 mem_cgroup_sk_alloc(sk); 1994 cgroup_sk_alloc(&sk->sk_cgrp_data); 1995 sock_update_classid(&sk->sk_cgrp_data); 1996 sock_update_netprioidx(&sk->sk_cgrp_data); 1997 sk_tx_queue_clear(sk); 1998 } 1999 2000 return sk; 2001 } 2002 EXPORT_SYMBOL(sk_alloc); 2003 2004 /* Sockets having SOCK_RCU_FREE will call this function after one RCU 2005 * grace period. This is the case for UDP sockets and TCP listeners. 2006 */ 2007 static void __sk_destruct(struct rcu_head *head) 2008 { 2009 struct sock *sk = container_of(head, struct sock, sk_rcu); 2010 struct sk_filter *filter; 2011 2012 if (sk->sk_destruct) 2013 sk->sk_destruct(sk); 2014 2015 filter = rcu_dereference_check(sk->sk_filter, 2016 refcount_read(&sk->sk_wmem_alloc) == 0); 2017 if (filter) { 2018 sk_filter_uncharge(sk, filter); 2019 RCU_INIT_POINTER(sk->sk_filter, NULL); 2020 } 2021 2022 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP); 2023 2024 #ifdef CONFIG_BPF_SYSCALL 2025 bpf_sk_storage_free(sk); 2026 #endif 2027 2028 if (atomic_read(&sk->sk_omem_alloc)) 2029 pr_debug("%s: optmem leakage (%d bytes) detected\n", 2030 __func__, atomic_read(&sk->sk_omem_alloc)); 2031 2032 if (sk->sk_frag.page) { 2033 put_page(sk->sk_frag.page); 2034 sk->sk_frag.page = NULL; 2035 } 2036 2037 /* We do not need to acquire sk->sk_peer_lock, we are the last user. */ 2038 put_cred(sk->sk_peer_cred); 2039 put_pid(sk->sk_peer_pid); 2040 2041 if (likely(sk->sk_net_refcnt)) 2042 put_net(sock_net(sk)); 2043 sk_prot_free(sk->sk_prot_creator, sk); 2044 } 2045 2046 void sk_destruct(struct sock *sk) 2047 { 2048 bool use_call_rcu = sock_flag(sk, SOCK_RCU_FREE); 2049 2050 if (rcu_access_pointer(sk->sk_reuseport_cb)) { 2051 reuseport_detach_sock(sk); 2052 use_call_rcu = true; 2053 } 2054 2055 if (use_call_rcu) 2056 call_rcu(&sk->sk_rcu, __sk_destruct); 2057 else 2058 __sk_destruct(&sk->sk_rcu); 2059 } 2060 2061 static void __sk_free(struct sock *sk) 2062 { 2063 if (likely(sk->sk_net_refcnt)) 2064 sock_inuse_add(sock_net(sk), -1); 2065 2066 if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk))) 2067 sock_diag_broadcast_destroy(sk); 2068 else 2069 sk_destruct(sk); 2070 } 2071 2072 void sk_free(struct sock *sk) 2073 { 2074 /* 2075 * We subtract one from sk_wmem_alloc and can know if 2076 * some packets are still in some tx queue. 2077 * If not null, sock_wfree() will call __sk_free(sk) later 2078 */ 2079 if (refcount_dec_and_test(&sk->sk_wmem_alloc)) 2080 __sk_free(sk); 2081 } 2082 EXPORT_SYMBOL(sk_free); 2083 2084 static void sk_init_common(struct sock *sk) 2085 { 2086 skb_queue_head_init(&sk->sk_receive_queue); 2087 skb_queue_head_init(&sk->sk_write_queue); 2088 skb_queue_head_init(&sk->sk_error_queue); 2089 2090 rwlock_init(&sk->sk_callback_lock); 2091 lockdep_set_class_and_name(&sk->sk_receive_queue.lock, 2092 af_rlock_keys + sk->sk_family, 2093 af_family_rlock_key_strings[sk->sk_family]); 2094 lockdep_set_class_and_name(&sk->sk_write_queue.lock, 2095 af_wlock_keys + sk->sk_family, 2096 af_family_wlock_key_strings[sk->sk_family]); 2097 lockdep_set_class_and_name(&sk->sk_error_queue.lock, 2098 af_elock_keys + sk->sk_family, 2099 af_family_elock_key_strings[sk->sk_family]); 2100 lockdep_set_class_and_name(&sk->sk_callback_lock, 2101 af_callback_keys + sk->sk_family, 2102 af_family_clock_key_strings[sk->sk_family]); 2103 } 2104 2105 /** 2106 * sk_clone_lock - clone a socket, and lock its clone 2107 * @sk: the socket to clone 2108 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 2109 * 2110 * Caller must unlock socket even in error path (bh_unlock_sock(newsk)) 2111 */ 2112 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority) 2113 { 2114 struct proto *prot = READ_ONCE(sk->sk_prot); 2115 struct sk_filter *filter; 2116 bool is_charged = true; 2117 struct sock *newsk; 2118 2119 newsk = sk_prot_alloc(prot, priority, sk->sk_family); 2120 if (!newsk) 2121 goto out; 2122 2123 sock_copy(newsk, sk); 2124 2125 newsk->sk_prot_creator = prot; 2126 2127 /* SANITY */ 2128 if (likely(newsk->sk_net_refcnt)) { 2129 get_net(sock_net(newsk)); 2130 sock_inuse_add(sock_net(newsk), 1); 2131 } 2132 sk_node_init(&newsk->sk_node); 2133 sock_lock_init(newsk); 2134 bh_lock_sock(newsk); 2135 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL; 2136 newsk->sk_backlog.len = 0; 2137 2138 atomic_set(&newsk->sk_rmem_alloc, 0); 2139 2140 /* sk_wmem_alloc set to one (see sk_free() and sock_wfree()) */ 2141 refcount_set(&newsk->sk_wmem_alloc, 1); 2142 2143 atomic_set(&newsk->sk_omem_alloc, 0); 2144 sk_init_common(newsk); 2145 2146 newsk->sk_dst_cache = NULL; 2147 newsk->sk_dst_pending_confirm = 0; 2148 newsk->sk_wmem_queued = 0; 2149 newsk->sk_forward_alloc = 0; 2150 newsk->sk_reserved_mem = 0; 2151 atomic_set(&newsk->sk_drops, 0); 2152 newsk->sk_send_head = NULL; 2153 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK; 2154 atomic_set(&newsk->sk_zckey, 0); 2155 2156 sock_reset_flag(newsk, SOCK_DONE); 2157 2158 /* sk->sk_memcg will be populated at accept() time */ 2159 newsk->sk_memcg = NULL; 2160 2161 cgroup_sk_clone(&newsk->sk_cgrp_data); 2162 2163 rcu_read_lock(); 2164 filter = rcu_dereference(sk->sk_filter); 2165 if (filter != NULL) 2166 /* though it's an empty new sock, the charging may fail 2167 * if sysctl_optmem_max was changed between creation of 2168 * original socket and cloning 2169 */ 2170 is_charged = sk_filter_charge(newsk, filter); 2171 RCU_INIT_POINTER(newsk->sk_filter, filter); 2172 rcu_read_unlock(); 2173 2174 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) { 2175 /* We need to make sure that we don't uncharge the new 2176 * socket if we couldn't charge it in the first place 2177 * as otherwise we uncharge the parent's filter. 2178 */ 2179 if (!is_charged) 2180 RCU_INIT_POINTER(newsk->sk_filter, NULL); 2181 sk_free_unlock_clone(newsk); 2182 newsk = NULL; 2183 goto out; 2184 } 2185 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL); 2186 2187 if (bpf_sk_storage_clone(sk, newsk)) { 2188 sk_free_unlock_clone(newsk); 2189 newsk = NULL; 2190 goto out; 2191 } 2192 2193 /* Clear sk_user_data if parent had the pointer tagged 2194 * as not suitable for copying when cloning. 2195 */ 2196 if (sk_user_data_is_nocopy(newsk)) 2197 newsk->sk_user_data = NULL; 2198 2199 newsk->sk_err = 0; 2200 newsk->sk_err_soft = 0; 2201 newsk->sk_priority = 0; 2202 newsk->sk_incoming_cpu = raw_smp_processor_id(); 2203 2204 /* Before updating sk_refcnt, we must commit prior changes to memory 2205 * (Documentation/RCU/rculist_nulls.rst for details) 2206 */ 2207 smp_wmb(); 2208 refcount_set(&newsk->sk_refcnt, 2); 2209 2210 /* Increment the counter in the same struct proto as the master 2211 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that 2212 * is the same as sk->sk_prot->socks, as this field was copied 2213 * with memcpy). 2214 * 2215 * This _changes_ the previous behaviour, where 2216 * tcp_create_openreq_child always was incrementing the 2217 * equivalent to tcp_prot->socks (inet_sock_nr), so this have 2218 * to be taken into account in all callers. -acme 2219 */ 2220 sk_refcnt_debug_inc(newsk); 2221 sk_set_socket(newsk, NULL); 2222 sk_tx_queue_clear(newsk); 2223 RCU_INIT_POINTER(newsk->sk_wq, NULL); 2224 2225 if (newsk->sk_prot->sockets_allocated) 2226 sk_sockets_allocated_inc(newsk); 2227 2228 if (sock_needs_netstamp(sk) && newsk->sk_flags & SK_FLAGS_TIMESTAMP) 2229 net_enable_timestamp(); 2230 out: 2231 return newsk; 2232 } 2233 EXPORT_SYMBOL_GPL(sk_clone_lock); 2234 2235 void sk_free_unlock_clone(struct sock *sk) 2236 { 2237 /* It is still raw copy of parent, so invalidate 2238 * destructor and make plain sk_free() */ 2239 sk->sk_destruct = NULL; 2240 bh_unlock_sock(sk); 2241 sk_free(sk); 2242 } 2243 EXPORT_SYMBOL_GPL(sk_free_unlock_clone); 2244 2245 void sk_setup_caps(struct sock *sk, struct dst_entry *dst) 2246 { 2247 u32 max_segs = 1; 2248 2249 sk_dst_set(sk, dst); 2250 sk->sk_route_caps = dst->dev->features; 2251 if (sk_is_tcp(sk)) 2252 sk->sk_route_caps |= NETIF_F_GSO; 2253 if (sk->sk_route_caps & NETIF_F_GSO) 2254 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE; 2255 if (unlikely(sk->sk_gso_disabled)) 2256 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 2257 if (sk_can_gso(sk)) { 2258 if (dst->header_len && !xfrm_dst_offload_ok(dst)) { 2259 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 2260 } else { 2261 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM; 2262 /* pairs with the WRITE_ONCE() in netif_set_gso_max_size() */ 2263 sk->sk_gso_max_size = READ_ONCE(dst->dev->gso_max_size); 2264 /* pairs with the WRITE_ONCE() in netif_set_gso_max_segs() */ 2265 max_segs = max_t(u32, READ_ONCE(dst->dev->gso_max_segs), 1); 2266 } 2267 } 2268 sk->sk_gso_max_segs = max_segs; 2269 } 2270 EXPORT_SYMBOL_GPL(sk_setup_caps); 2271 2272 /* 2273 * Simple resource managers for sockets. 2274 */ 2275 2276 2277 /* 2278 * Write buffer destructor automatically called from kfree_skb. 2279 */ 2280 void sock_wfree(struct sk_buff *skb) 2281 { 2282 struct sock *sk = skb->sk; 2283 unsigned int len = skb->truesize; 2284 2285 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) { 2286 /* 2287 * Keep a reference on sk_wmem_alloc, this will be released 2288 * after sk_write_space() call 2289 */ 2290 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc)); 2291 sk->sk_write_space(sk); 2292 len = 1; 2293 } 2294 /* 2295 * if sk_wmem_alloc reaches 0, we must finish what sk_free() 2296 * could not do because of in-flight packets 2297 */ 2298 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc)) 2299 __sk_free(sk); 2300 } 2301 EXPORT_SYMBOL(sock_wfree); 2302 2303 /* This variant of sock_wfree() is used by TCP, 2304 * since it sets SOCK_USE_WRITE_QUEUE. 2305 */ 2306 void __sock_wfree(struct sk_buff *skb) 2307 { 2308 struct sock *sk = skb->sk; 2309 2310 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc)) 2311 __sk_free(sk); 2312 } 2313 2314 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk) 2315 { 2316 skb_orphan(skb); 2317 skb->sk = sk; 2318 #ifdef CONFIG_INET 2319 if (unlikely(!sk_fullsock(sk))) { 2320 skb->destructor = sock_edemux; 2321 sock_hold(sk); 2322 return; 2323 } 2324 #endif 2325 skb->destructor = sock_wfree; 2326 skb_set_hash_from_sk(skb, sk); 2327 /* 2328 * We used to take a refcount on sk, but following operation 2329 * is enough to guarantee sk_free() wont free this sock until 2330 * all in-flight packets are completed 2331 */ 2332 refcount_add(skb->truesize, &sk->sk_wmem_alloc); 2333 } 2334 EXPORT_SYMBOL(skb_set_owner_w); 2335 2336 static bool can_skb_orphan_partial(const struct sk_buff *skb) 2337 { 2338 #ifdef CONFIG_TLS_DEVICE 2339 /* Drivers depend on in-order delivery for crypto offload, 2340 * partial orphan breaks out-of-order-OK logic. 2341 */ 2342 if (skb->decrypted) 2343 return false; 2344 #endif 2345 return (skb->destructor == sock_wfree || 2346 (IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree)); 2347 } 2348 2349 /* This helper is used by netem, as it can hold packets in its 2350 * delay queue. We want to allow the owner socket to send more 2351 * packets, as if they were already TX completed by a typical driver. 2352 * But we also want to keep skb->sk set because some packet schedulers 2353 * rely on it (sch_fq for example). 2354 */ 2355 void skb_orphan_partial(struct sk_buff *skb) 2356 { 2357 if (skb_is_tcp_pure_ack(skb)) 2358 return; 2359 2360 if (can_skb_orphan_partial(skb) && skb_set_owner_sk_safe(skb, skb->sk)) 2361 return; 2362 2363 skb_orphan(skb); 2364 } 2365 EXPORT_SYMBOL(skb_orphan_partial); 2366 2367 /* 2368 * Read buffer destructor automatically called from kfree_skb. 2369 */ 2370 void sock_rfree(struct sk_buff *skb) 2371 { 2372 struct sock *sk = skb->sk; 2373 unsigned int len = skb->truesize; 2374 2375 atomic_sub(len, &sk->sk_rmem_alloc); 2376 sk_mem_uncharge(sk, len); 2377 } 2378 EXPORT_SYMBOL(sock_rfree); 2379 2380 /* 2381 * Buffer destructor for skbs that are not used directly in read or write 2382 * path, e.g. for error handler skbs. Automatically called from kfree_skb. 2383 */ 2384 void sock_efree(struct sk_buff *skb) 2385 { 2386 sock_put(skb->sk); 2387 } 2388 EXPORT_SYMBOL(sock_efree); 2389 2390 /* Buffer destructor for prefetch/receive path where reference count may 2391 * not be held, e.g. for listen sockets. 2392 */ 2393 #ifdef CONFIG_INET 2394 void sock_pfree(struct sk_buff *skb) 2395 { 2396 if (sk_is_refcounted(skb->sk)) 2397 sock_gen_put(skb->sk); 2398 } 2399 EXPORT_SYMBOL(sock_pfree); 2400 #endif /* CONFIG_INET */ 2401 2402 kuid_t sock_i_uid(struct sock *sk) 2403 { 2404 kuid_t uid; 2405 2406 read_lock_bh(&sk->sk_callback_lock); 2407 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID; 2408 read_unlock_bh(&sk->sk_callback_lock); 2409 return uid; 2410 } 2411 EXPORT_SYMBOL(sock_i_uid); 2412 2413 unsigned long sock_i_ino(struct sock *sk) 2414 { 2415 unsigned long ino; 2416 2417 read_lock_bh(&sk->sk_callback_lock); 2418 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0; 2419 read_unlock_bh(&sk->sk_callback_lock); 2420 return ino; 2421 } 2422 EXPORT_SYMBOL(sock_i_ino); 2423 2424 /* 2425 * Allocate a skb from the socket's send buffer. 2426 */ 2427 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 2428 gfp_t priority) 2429 { 2430 if (force || 2431 refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) { 2432 struct sk_buff *skb = alloc_skb(size, priority); 2433 2434 if (skb) { 2435 skb_set_owner_w(skb, sk); 2436 return skb; 2437 } 2438 } 2439 return NULL; 2440 } 2441 EXPORT_SYMBOL(sock_wmalloc); 2442 2443 static void sock_ofree(struct sk_buff *skb) 2444 { 2445 struct sock *sk = skb->sk; 2446 2447 atomic_sub(skb->truesize, &sk->sk_omem_alloc); 2448 } 2449 2450 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size, 2451 gfp_t priority) 2452 { 2453 struct sk_buff *skb; 2454 2455 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */ 2456 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) > 2457 sysctl_optmem_max) 2458 return NULL; 2459 2460 skb = alloc_skb(size, priority); 2461 if (!skb) 2462 return NULL; 2463 2464 atomic_add(skb->truesize, &sk->sk_omem_alloc); 2465 skb->sk = sk; 2466 skb->destructor = sock_ofree; 2467 return skb; 2468 } 2469 2470 /* 2471 * Allocate a memory block from the socket's option memory buffer. 2472 */ 2473 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority) 2474 { 2475 if ((unsigned int)size <= sysctl_optmem_max && 2476 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) { 2477 void *mem; 2478 /* First do the add, to avoid the race if kmalloc 2479 * might sleep. 2480 */ 2481 atomic_add(size, &sk->sk_omem_alloc); 2482 mem = kmalloc(size, priority); 2483 if (mem) 2484 return mem; 2485 atomic_sub(size, &sk->sk_omem_alloc); 2486 } 2487 return NULL; 2488 } 2489 EXPORT_SYMBOL(sock_kmalloc); 2490 2491 /* Free an option memory block. Note, we actually want the inline 2492 * here as this allows gcc to detect the nullify and fold away the 2493 * condition entirely. 2494 */ 2495 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size, 2496 const bool nullify) 2497 { 2498 if (WARN_ON_ONCE(!mem)) 2499 return; 2500 if (nullify) 2501 kfree_sensitive(mem); 2502 else 2503 kfree(mem); 2504 atomic_sub(size, &sk->sk_omem_alloc); 2505 } 2506 2507 void sock_kfree_s(struct sock *sk, void *mem, int size) 2508 { 2509 __sock_kfree_s(sk, mem, size, false); 2510 } 2511 EXPORT_SYMBOL(sock_kfree_s); 2512 2513 void sock_kzfree_s(struct sock *sk, void *mem, int size) 2514 { 2515 __sock_kfree_s(sk, mem, size, true); 2516 } 2517 EXPORT_SYMBOL(sock_kzfree_s); 2518 2519 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock. 2520 I think, these locks should be removed for datagram sockets. 2521 */ 2522 static long sock_wait_for_wmem(struct sock *sk, long timeo) 2523 { 2524 DEFINE_WAIT(wait); 2525 2526 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk); 2527 for (;;) { 2528 if (!timeo) 2529 break; 2530 if (signal_pending(current)) 2531 break; 2532 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 2533 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 2534 if (refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) 2535 break; 2536 if (sk->sk_shutdown & SEND_SHUTDOWN) 2537 break; 2538 if (sk->sk_err) 2539 break; 2540 timeo = schedule_timeout(timeo); 2541 } 2542 finish_wait(sk_sleep(sk), &wait); 2543 return timeo; 2544 } 2545 2546 2547 /* 2548 * Generic send/receive buffer handlers 2549 */ 2550 2551 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 2552 unsigned long data_len, int noblock, 2553 int *errcode, int max_page_order) 2554 { 2555 struct sk_buff *skb; 2556 long timeo; 2557 int err; 2558 2559 timeo = sock_sndtimeo(sk, noblock); 2560 for (;;) { 2561 err = sock_error(sk); 2562 if (err != 0) 2563 goto failure; 2564 2565 err = -EPIPE; 2566 if (sk->sk_shutdown & SEND_SHUTDOWN) 2567 goto failure; 2568 2569 if (sk_wmem_alloc_get(sk) < READ_ONCE(sk->sk_sndbuf)) 2570 break; 2571 2572 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk); 2573 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 2574 err = -EAGAIN; 2575 if (!timeo) 2576 goto failure; 2577 if (signal_pending(current)) 2578 goto interrupted; 2579 timeo = sock_wait_for_wmem(sk, timeo); 2580 } 2581 skb = alloc_skb_with_frags(header_len, data_len, max_page_order, 2582 errcode, sk->sk_allocation); 2583 if (skb) 2584 skb_set_owner_w(skb, sk); 2585 return skb; 2586 2587 interrupted: 2588 err = sock_intr_errno(timeo); 2589 failure: 2590 *errcode = err; 2591 return NULL; 2592 } 2593 EXPORT_SYMBOL(sock_alloc_send_pskb); 2594 2595 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size, 2596 int noblock, int *errcode) 2597 { 2598 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0); 2599 } 2600 EXPORT_SYMBOL(sock_alloc_send_skb); 2601 2602 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg, 2603 struct sockcm_cookie *sockc) 2604 { 2605 u32 tsflags; 2606 2607 switch (cmsg->cmsg_type) { 2608 case SO_MARK: 2609 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) 2610 return -EPERM; 2611 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 2612 return -EINVAL; 2613 sockc->mark = *(u32 *)CMSG_DATA(cmsg); 2614 break; 2615 case SO_TIMESTAMPING_OLD: 2616 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 2617 return -EINVAL; 2618 2619 tsflags = *(u32 *)CMSG_DATA(cmsg); 2620 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK) 2621 return -EINVAL; 2622 2623 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK; 2624 sockc->tsflags |= tsflags; 2625 break; 2626 case SCM_TXTIME: 2627 if (!sock_flag(sk, SOCK_TXTIME)) 2628 return -EINVAL; 2629 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64))) 2630 return -EINVAL; 2631 sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg)); 2632 break; 2633 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */ 2634 case SCM_RIGHTS: 2635 case SCM_CREDENTIALS: 2636 break; 2637 default: 2638 return -EINVAL; 2639 } 2640 return 0; 2641 } 2642 EXPORT_SYMBOL(__sock_cmsg_send); 2643 2644 int sock_cmsg_send(struct sock *sk, struct msghdr *msg, 2645 struct sockcm_cookie *sockc) 2646 { 2647 struct cmsghdr *cmsg; 2648 int ret; 2649 2650 for_each_cmsghdr(cmsg, msg) { 2651 if (!CMSG_OK(msg, cmsg)) 2652 return -EINVAL; 2653 if (cmsg->cmsg_level != SOL_SOCKET) 2654 continue; 2655 ret = __sock_cmsg_send(sk, msg, cmsg, sockc); 2656 if (ret) 2657 return ret; 2658 } 2659 return 0; 2660 } 2661 EXPORT_SYMBOL(sock_cmsg_send); 2662 2663 static void sk_enter_memory_pressure(struct sock *sk) 2664 { 2665 if (!sk->sk_prot->enter_memory_pressure) 2666 return; 2667 2668 sk->sk_prot->enter_memory_pressure(sk); 2669 } 2670 2671 static void sk_leave_memory_pressure(struct sock *sk) 2672 { 2673 if (sk->sk_prot->leave_memory_pressure) { 2674 sk->sk_prot->leave_memory_pressure(sk); 2675 } else { 2676 unsigned long *memory_pressure = sk->sk_prot->memory_pressure; 2677 2678 if (memory_pressure && READ_ONCE(*memory_pressure)) 2679 WRITE_ONCE(*memory_pressure, 0); 2680 } 2681 } 2682 2683 DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key); 2684 2685 /** 2686 * skb_page_frag_refill - check that a page_frag contains enough room 2687 * @sz: minimum size of the fragment we want to get 2688 * @pfrag: pointer to page_frag 2689 * @gfp: priority for memory allocation 2690 * 2691 * Note: While this allocator tries to use high order pages, there is 2692 * no guarantee that allocations succeed. Therefore, @sz MUST be 2693 * less or equal than PAGE_SIZE. 2694 */ 2695 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp) 2696 { 2697 if (pfrag->page) { 2698 if (page_ref_count(pfrag->page) == 1) { 2699 pfrag->offset = 0; 2700 return true; 2701 } 2702 if (pfrag->offset + sz <= pfrag->size) 2703 return true; 2704 put_page(pfrag->page); 2705 } 2706 2707 pfrag->offset = 0; 2708 if (SKB_FRAG_PAGE_ORDER && 2709 !static_branch_unlikely(&net_high_order_alloc_disable_key)) { 2710 /* Avoid direct reclaim but allow kswapd to wake */ 2711 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) | 2712 __GFP_COMP | __GFP_NOWARN | 2713 __GFP_NORETRY, 2714 SKB_FRAG_PAGE_ORDER); 2715 if (likely(pfrag->page)) { 2716 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER; 2717 return true; 2718 } 2719 } 2720 pfrag->page = alloc_page(gfp); 2721 if (likely(pfrag->page)) { 2722 pfrag->size = PAGE_SIZE; 2723 return true; 2724 } 2725 return false; 2726 } 2727 EXPORT_SYMBOL(skb_page_frag_refill); 2728 2729 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag) 2730 { 2731 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation))) 2732 return true; 2733 2734 sk_enter_memory_pressure(sk); 2735 sk_stream_moderate_sndbuf(sk); 2736 return false; 2737 } 2738 EXPORT_SYMBOL(sk_page_frag_refill); 2739 2740 void __lock_sock(struct sock *sk) 2741 __releases(&sk->sk_lock.slock) 2742 __acquires(&sk->sk_lock.slock) 2743 { 2744 DEFINE_WAIT(wait); 2745 2746 for (;;) { 2747 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait, 2748 TASK_UNINTERRUPTIBLE); 2749 spin_unlock_bh(&sk->sk_lock.slock); 2750 schedule(); 2751 spin_lock_bh(&sk->sk_lock.slock); 2752 if (!sock_owned_by_user(sk)) 2753 break; 2754 } 2755 finish_wait(&sk->sk_lock.wq, &wait); 2756 } 2757 2758 void __release_sock(struct sock *sk) 2759 __releases(&sk->sk_lock.slock) 2760 __acquires(&sk->sk_lock.slock) 2761 { 2762 struct sk_buff *skb, *next; 2763 2764 while ((skb = sk->sk_backlog.head) != NULL) { 2765 sk->sk_backlog.head = sk->sk_backlog.tail = NULL; 2766 2767 spin_unlock_bh(&sk->sk_lock.slock); 2768 2769 do { 2770 next = skb->next; 2771 prefetch(next); 2772 WARN_ON_ONCE(skb_dst_is_noref(skb)); 2773 skb_mark_not_on_list(skb); 2774 sk_backlog_rcv(sk, skb); 2775 2776 cond_resched(); 2777 2778 skb = next; 2779 } while (skb != NULL); 2780 2781 spin_lock_bh(&sk->sk_lock.slock); 2782 } 2783 2784 /* 2785 * Doing the zeroing here guarantee we can not loop forever 2786 * while a wild producer attempts to flood us. 2787 */ 2788 sk->sk_backlog.len = 0; 2789 } 2790 2791 void __sk_flush_backlog(struct sock *sk) 2792 { 2793 spin_lock_bh(&sk->sk_lock.slock); 2794 __release_sock(sk); 2795 spin_unlock_bh(&sk->sk_lock.slock); 2796 } 2797 2798 /** 2799 * sk_wait_data - wait for data to arrive at sk_receive_queue 2800 * @sk: sock to wait on 2801 * @timeo: for how long 2802 * @skb: last skb seen on sk_receive_queue 2803 * 2804 * Now socket state including sk->sk_err is changed only under lock, 2805 * hence we may omit checks after joining wait queue. 2806 * We check receive queue before schedule() only as optimization; 2807 * it is very likely that release_sock() added new data. 2808 */ 2809 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb) 2810 { 2811 DEFINE_WAIT_FUNC(wait, woken_wake_function); 2812 int rc; 2813 2814 add_wait_queue(sk_sleep(sk), &wait); 2815 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk); 2816 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait); 2817 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk); 2818 remove_wait_queue(sk_sleep(sk), &wait); 2819 return rc; 2820 } 2821 EXPORT_SYMBOL(sk_wait_data); 2822 2823 /** 2824 * __sk_mem_raise_allocated - increase memory_allocated 2825 * @sk: socket 2826 * @size: memory size to allocate 2827 * @amt: pages to allocate 2828 * @kind: allocation type 2829 * 2830 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc 2831 */ 2832 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind) 2833 { 2834 struct proto *prot = sk->sk_prot; 2835 long allocated = sk_memory_allocated_add(sk, amt); 2836 bool memcg_charge = mem_cgroup_sockets_enabled && sk->sk_memcg; 2837 bool charged = true; 2838 2839 if (memcg_charge && 2840 !(charged = mem_cgroup_charge_skmem(sk->sk_memcg, amt, 2841 gfp_memcg_charge()))) 2842 goto suppress_allocation; 2843 2844 /* Under limit. */ 2845 if (allocated <= sk_prot_mem_limits(sk, 0)) { 2846 sk_leave_memory_pressure(sk); 2847 return 1; 2848 } 2849 2850 /* Under pressure. */ 2851 if (allocated > sk_prot_mem_limits(sk, 1)) 2852 sk_enter_memory_pressure(sk); 2853 2854 /* Over hard limit. */ 2855 if (allocated > sk_prot_mem_limits(sk, 2)) 2856 goto suppress_allocation; 2857 2858 /* guarantee minimum buffer size under pressure */ 2859 if (kind == SK_MEM_RECV) { 2860 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot)) 2861 return 1; 2862 2863 } else { /* SK_MEM_SEND */ 2864 int wmem0 = sk_get_wmem0(sk, prot); 2865 2866 if (sk->sk_type == SOCK_STREAM) { 2867 if (sk->sk_wmem_queued < wmem0) 2868 return 1; 2869 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) { 2870 return 1; 2871 } 2872 } 2873 2874 if (sk_has_memory_pressure(sk)) { 2875 u64 alloc; 2876 2877 if (!sk_under_memory_pressure(sk)) 2878 return 1; 2879 alloc = sk_sockets_allocated_read_positive(sk); 2880 if (sk_prot_mem_limits(sk, 2) > alloc * 2881 sk_mem_pages(sk->sk_wmem_queued + 2882 atomic_read(&sk->sk_rmem_alloc) + 2883 sk->sk_forward_alloc)) 2884 return 1; 2885 } 2886 2887 suppress_allocation: 2888 2889 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) { 2890 sk_stream_moderate_sndbuf(sk); 2891 2892 /* Fail only if socket is _under_ its sndbuf. 2893 * In this case we cannot block, so that we have to fail. 2894 */ 2895 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf) { 2896 /* Force charge with __GFP_NOFAIL */ 2897 if (memcg_charge && !charged) { 2898 mem_cgroup_charge_skmem(sk->sk_memcg, amt, 2899 gfp_memcg_charge() | __GFP_NOFAIL); 2900 } 2901 return 1; 2902 } 2903 } 2904 2905 if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged)) 2906 trace_sock_exceed_buf_limit(sk, prot, allocated, kind); 2907 2908 sk_memory_allocated_sub(sk, amt); 2909 2910 if (memcg_charge && charged) 2911 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt); 2912 2913 return 0; 2914 } 2915 EXPORT_SYMBOL(__sk_mem_raise_allocated); 2916 2917 /** 2918 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated 2919 * @sk: socket 2920 * @size: memory size to allocate 2921 * @kind: allocation type 2922 * 2923 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means 2924 * rmem allocation. This function assumes that protocols which have 2925 * memory_pressure use sk_wmem_queued as write buffer accounting. 2926 */ 2927 int __sk_mem_schedule(struct sock *sk, int size, int kind) 2928 { 2929 int ret, amt = sk_mem_pages(size); 2930 2931 sk->sk_forward_alloc += amt << SK_MEM_QUANTUM_SHIFT; 2932 ret = __sk_mem_raise_allocated(sk, size, amt, kind); 2933 if (!ret) 2934 sk->sk_forward_alloc -= amt << SK_MEM_QUANTUM_SHIFT; 2935 return ret; 2936 } 2937 EXPORT_SYMBOL(__sk_mem_schedule); 2938 2939 /** 2940 * __sk_mem_reduce_allocated - reclaim memory_allocated 2941 * @sk: socket 2942 * @amount: number of quanta 2943 * 2944 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc 2945 */ 2946 void __sk_mem_reduce_allocated(struct sock *sk, int amount) 2947 { 2948 sk_memory_allocated_sub(sk, amount); 2949 2950 if (mem_cgroup_sockets_enabled && sk->sk_memcg) 2951 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount); 2952 2953 if (sk_under_memory_pressure(sk) && 2954 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0))) 2955 sk_leave_memory_pressure(sk); 2956 } 2957 EXPORT_SYMBOL(__sk_mem_reduce_allocated); 2958 2959 /** 2960 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated 2961 * @sk: socket 2962 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple) 2963 */ 2964 void __sk_mem_reclaim(struct sock *sk, int amount) 2965 { 2966 amount >>= SK_MEM_QUANTUM_SHIFT; 2967 sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT; 2968 __sk_mem_reduce_allocated(sk, amount); 2969 } 2970 EXPORT_SYMBOL(__sk_mem_reclaim); 2971 2972 int sk_set_peek_off(struct sock *sk, int val) 2973 { 2974 sk->sk_peek_off = val; 2975 return 0; 2976 } 2977 EXPORT_SYMBOL_GPL(sk_set_peek_off); 2978 2979 /* 2980 * Set of default routines for initialising struct proto_ops when 2981 * the protocol does not support a particular function. In certain 2982 * cases where it makes no sense for a protocol to have a "do nothing" 2983 * function, some default processing is provided. 2984 */ 2985 2986 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len) 2987 { 2988 return -EOPNOTSUPP; 2989 } 2990 EXPORT_SYMBOL(sock_no_bind); 2991 2992 int sock_no_connect(struct socket *sock, struct sockaddr *saddr, 2993 int len, int flags) 2994 { 2995 return -EOPNOTSUPP; 2996 } 2997 EXPORT_SYMBOL(sock_no_connect); 2998 2999 int sock_no_socketpair(struct socket *sock1, struct socket *sock2) 3000 { 3001 return -EOPNOTSUPP; 3002 } 3003 EXPORT_SYMBOL(sock_no_socketpair); 3004 3005 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags, 3006 bool kern) 3007 { 3008 return -EOPNOTSUPP; 3009 } 3010 EXPORT_SYMBOL(sock_no_accept); 3011 3012 int sock_no_getname(struct socket *sock, struct sockaddr *saddr, 3013 int peer) 3014 { 3015 return -EOPNOTSUPP; 3016 } 3017 EXPORT_SYMBOL(sock_no_getname); 3018 3019 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 3020 { 3021 return -EOPNOTSUPP; 3022 } 3023 EXPORT_SYMBOL(sock_no_ioctl); 3024 3025 int sock_no_listen(struct socket *sock, int backlog) 3026 { 3027 return -EOPNOTSUPP; 3028 } 3029 EXPORT_SYMBOL(sock_no_listen); 3030 3031 int sock_no_shutdown(struct socket *sock, int how) 3032 { 3033 return -EOPNOTSUPP; 3034 } 3035 EXPORT_SYMBOL(sock_no_shutdown); 3036 3037 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len) 3038 { 3039 return -EOPNOTSUPP; 3040 } 3041 EXPORT_SYMBOL(sock_no_sendmsg); 3042 3043 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len) 3044 { 3045 return -EOPNOTSUPP; 3046 } 3047 EXPORT_SYMBOL(sock_no_sendmsg_locked); 3048 3049 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len, 3050 int flags) 3051 { 3052 return -EOPNOTSUPP; 3053 } 3054 EXPORT_SYMBOL(sock_no_recvmsg); 3055 3056 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma) 3057 { 3058 /* Mirror missing mmap method error code */ 3059 return -ENODEV; 3060 } 3061 EXPORT_SYMBOL(sock_no_mmap); 3062 3063 /* 3064 * When a file is received (via SCM_RIGHTS, etc), we must bump the 3065 * various sock-based usage counts. 3066 */ 3067 void __receive_sock(struct file *file) 3068 { 3069 struct socket *sock; 3070 3071 sock = sock_from_file(file); 3072 if (sock) { 3073 sock_update_netprioidx(&sock->sk->sk_cgrp_data); 3074 sock_update_classid(&sock->sk->sk_cgrp_data); 3075 } 3076 } 3077 3078 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags) 3079 { 3080 ssize_t res; 3081 struct msghdr msg = {.msg_flags = flags}; 3082 struct kvec iov; 3083 char *kaddr = kmap(page); 3084 iov.iov_base = kaddr + offset; 3085 iov.iov_len = size; 3086 res = kernel_sendmsg(sock, &msg, &iov, 1, size); 3087 kunmap(page); 3088 return res; 3089 } 3090 EXPORT_SYMBOL(sock_no_sendpage); 3091 3092 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page, 3093 int offset, size_t size, int flags) 3094 { 3095 ssize_t res; 3096 struct msghdr msg = {.msg_flags = flags}; 3097 struct kvec iov; 3098 char *kaddr = kmap(page); 3099 3100 iov.iov_base = kaddr + offset; 3101 iov.iov_len = size; 3102 res = kernel_sendmsg_locked(sk, &msg, &iov, 1, size); 3103 kunmap(page); 3104 return res; 3105 } 3106 EXPORT_SYMBOL(sock_no_sendpage_locked); 3107 3108 /* 3109 * Default Socket Callbacks 3110 */ 3111 3112 static void sock_def_wakeup(struct sock *sk) 3113 { 3114 struct socket_wq *wq; 3115 3116 rcu_read_lock(); 3117 wq = rcu_dereference(sk->sk_wq); 3118 if (skwq_has_sleeper(wq)) 3119 wake_up_interruptible_all(&wq->wait); 3120 rcu_read_unlock(); 3121 } 3122 3123 static void sock_def_error_report(struct sock *sk) 3124 { 3125 struct socket_wq *wq; 3126 3127 rcu_read_lock(); 3128 wq = rcu_dereference(sk->sk_wq); 3129 if (skwq_has_sleeper(wq)) 3130 wake_up_interruptible_poll(&wq->wait, EPOLLERR); 3131 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR); 3132 rcu_read_unlock(); 3133 } 3134 3135 void sock_def_readable(struct sock *sk) 3136 { 3137 struct socket_wq *wq; 3138 3139 rcu_read_lock(); 3140 wq = rcu_dereference(sk->sk_wq); 3141 if (skwq_has_sleeper(wq)) 3142 wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI | 3143 EPOLLRDNORM | EPOLLRDBAND); 3144 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN); 3145 rcu_read_unlock(); 3146 } 3147 3148 static void sock_def_write_space(struct sock *sk) 3149 { 3150 struct socket_wq *wq; 3151 3152 rcu_read_lock(); 3153 3154 /* Do not wake up a writer until he can make "significant" 3155 * progress. --DaveM 3156 */ 3157 if ((refcount_read(&sk->sk_wmem_alloc) << 1) <= READ_ONCE(sk->sk_sndbuf)) { 3158 wq = rcu_dereference(sk->sk_wq); 3159 if (skwq_has_sleeper(wq)) 3160 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT | 3161 EPOLLWRNORM | EPOLLWRBAND); 3162 3163 /* Should agree with poll, otherwise some programs break */ 3164 if (sock_writeable(sk)) 3165 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT); 3166 } 3167 3168 rcu_read_unlock(); 3169 } 3170 3171 static void sock_def_destruct(struct sock *sk) 3172 { 3173 } 3174 3175 void sk_send_sigurg(struct sock *sk) 3176 { 3177 if (sk->sk_socket && sk->sk_socket->file) 3178 if (send_sigurg(&sk->sk_socket->file->f_owner)) 3179 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI); 3180 } 3181 EXPORT_SYMBOL(sk_send_sigurg); 3182 3183 void sk_reset_timer(struct sock *sk, struct timer_list* timer, 3184 unsigned long expires) 3185 { 3186 if (!mod_timer(timer, expires)) 3187 sock_hold(sk); 3188 } 3189 EXPORT_SYMBOL(sk_reset_timer); 3190 3191 void sk_stop_timer(struct sock *sk, struct timer_list* timer) 3192 { 3193 if (del_timer(timer)) 3194 __sock_put(sk); 3195 } 3196 EXPORT_SYMBOL(sk_stop_timer); 3197 3198 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer) 3199 { 3200 if (del_timer_sync(timer)) 3201 __sock_put(sk); 3202 } 3203 EXPORT_SYMBOL(sk_stop_timer_sync); 3204 3205 void sock_init_data(struct socket *sock, struct sock *sk) 3206 { 3207 sk_init_common(sk); 3208 sk->sk_send_head = NULL; 3209 3210 timer_setup(&sk->sk_timer, NULL, 0); 3211 3212 sk->sk_allocation = GFP_KERNEL; 3213 sk->sk_rcvbuf = sysctl_rmem_default; 3214 sk->sk_sndbuf = sysctl_wmem_default; 3215 sk->sk_state = TCP_CLOSE; 3216 sk_set_socket(sk, sock); 3217 3218 sock_set_flag(sk, SOCK_ZAPPED); 3219 3220 if (sock) { 3221 sk->sk_type = sock->type; 3222 RCU_INIT_POINTER(sk->sk_wq, &sock->wq); 3223 sock->sk = sk; 3224 sk->sk_uid = SOCK_INODE(sock)->i_uid; 3225 } else { 3226 RCU_INIT_POINTER(sk->sk_wq, NULL); 3227 sk->sk_uid = make_kuid(sock_net(sk)->user_ns, 0); 3228 } 3229 3230 rwlock_init(&sk->sk_callback_lock); 3231 if (sk->sk_kern_sock) 3232 lockdep_set_class_and_name( 3233 &sk->sk_callback_lock, 3234 af_kern_callback_keys + sk->sk_family, 3235 af_family_kern_clock_key_strings[sk->sk_family]); 3236 else 3237 lockdep_set_class_and_name( 3238 &sk->sk_callback_lock, 3239 af_callback_keys + sk->sk_family, 3240 af_family_clock_key_strings[sk->sk_family]); 3241 3242 sk->sk_state_change = sock_def_wakeup; 3243 sk->sk_data_ready = sock_def_readable; 3244 sk->sk_write_space = sock_def_write_space; 3245 sk->sk_error_report = sock_def_error_report; 3246 sk->sk_destruct = sock_def_destruct; 3247 3248 sk->sk_frag.page = NULL; 3249 sk->sk_frag.offset = 0; 3250 sk->sk_peek_off = -1; 3251 3252 sk->sk_peer_pid = NULL; 3253 sk->sk_peer_cred = NULL; 3254 spin_lock_init(&sk->sk_peer_lock); 3255 3256 sk->sk_write_pending = 0; 3257 sk->sk_rcvlowat = 1; 3258 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; 3259 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT; 3260 3261 sk->sk_stamp = SK_DEFAULT_STAMP; 3262 #if BITS_PER_LONG==32 3263 seqlock_init(&sk->sk_stamp_seq); 3264 #endif 3265 atomic_set(&sk->sk_zckey, 0); 3266 3267 #ifdef CONFIG_NET_RX_BUSY_POLL 3268 sk->sk_napi_id = 0; 3269 sk->sk_ll_usec = sysctl_net_busy_read; 3270 #endif 3271 3272 sk->sk_max_pacing_rate = ~0UL; 3273 sk->sk_pacing_rate = ~0UL; 3274 WRITE_ONCE(sk->sk_pacing_shift, 10); 3275 sk->sk_incoming_cpu = -1; 3276 3277 sk_rx_queue_clear(sk); 3278 /* 3279 * Before updating sk_refcnt, we must commit prior changes to memory 3280 * (Documentation/RCU/rculist_nulls.rst for details) 3281 */ 3282 smp_wmb(); 3283 refcount_set(&sk->sk_refcnt, 1); 3284 atomic_set(&sk->sk_drops, 0); 3285 } 3286 EXPORT_SYMBOL(sock_init_data); 3287 3288 void lock_sock_nested(struct sock *sk, int subclass) 3289 { 3290 /* The sk_lock has mutex_lock() semantics here. */ 3291 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_); 3292 3293 might_sleep(); 3294 spin_lock_bh(&sk->sk_lock.slock); 3295 if (sk->sk_lock.owned) 3296 __lock_sock(sk); 3297 sk->sk_lock.owned = 1; 3298 spin_unlock_bh(&sk->sk_lock.slock); 3299 } 3300 EXPORT_SYMBOL(lock_sock_nested); 3301 3302 void release_sock(struct sock *sk) 3303 { 3304 spin_lock_bh(&sk->sk_lock.slock); 3305 if (sk->sk_backlog.tail) 3306 __release_sock(sk); 3307 3308 /* Warning : release_cb() might need to release sk ownership, 3309 * ie call sock_release_ownership(sk) before us. 3310 */ 3311 if (sk->sk_prot->release_cb) 3312 sk->sk_prot->release_cb(sk); 3313 3314 sock_release_ownership(sk); 3315 if (waitqueue_active(&sk->sk_lock.wq)) 3316 wake_up(&sk->sk_lock.wq); 3317 spin_unlock_bh(&sk->sk_lock.slock); 3318 } 3319 EXPORT_SYMBOL(release_sock); 3320 3321 bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock) 3322 { 3323 might_sleep(); 3324 spin_lock_bh(&sk->sk_lock.slock); 3325 3326 if (!sk->sk_lock.owned) { 3327 /* 3328 * Fast path return with bottom halves disabled and 3329 * sock::sk_lock.slock held. 3330 * 3331 * The 'mutex' is not contended and holding 3332 * sock::sk_lock.slock prevents all other lockers to 3333 * proceed so the corresponding unlock_sock_fast() can 3334 * avoid the slow path of release_sock() completely and 3335 * just release slock. 3336 * 3337 * From a semantical POV this is equivalent to 'acquiring' 3338 * the 'mutex', hence the corresponding lockdep 3339 * mutex_release() has to happen in the fast path of 3340 * unlock_sock_fast(). 3341 */ 3342 return false; 3343 } 3344 3345 __lock_sock(sk); 3346 sk->sk_lock.owned = 1; 3347 __acquire(&sk->sk_lock.slock); 3348 spin_unlock_bh(&sk->sk_lock.slock); 3349 return true; 3350 } 3351 EXPORT_SYMBOL(__lock_sock_fast); 3352 3353 int sock_gettstamp(struct socket *sock, void __user *userstamp, 3354 bool timeval, bool time32) 3355 { 3356 struct sock *sk = sock->sk; 3357 struct timespec64 ts; 3358 3359 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 3360 ts = ktime_to_timespec64(sock_read_timestamp(sk)); 3361 if (ts.tv_sec == -1) 3362 return -ENOENT; 3363 if (ts.tv_sec == 0) { 3364 ktime_t kt = ktime_get_real(); 3365 sock_write_timestamp(sk, kt); 3366 ts = ktime_to_timespec64(kt); 3367 } 3368 3369 if (timeval) 3370 ts.tv_nsec /= 1000; 3371 3372 #ifdef CONFIG_COMPAT_32BIT_TIME 3373 if (time32) 3374 return put_old_timespec32(&ts, userstamp); 3375 #endif 3376 #ifdef CONFIG_SPARC64 3377 /* beware of padding in sparc64 timeval */ 3378 if (timeval && !in_compat_syscall()) { 3379 struct __kernel_old_timeval __user tv = { 3380 .tv_sec = ts.tv_sec, 3381 .tv_usec = ts.tv_nsec, 3382 }; 3383 if (copy_to_user(userstamp, &tv, sizeof(tv))) 3384 return -EFAULT; 3385 return 0; 3386 } 3387 #endif 3388 return put_timespec64(&ts, userstamp); 3389 } 3390 EXPORT_SYMBOL(sock_gettstamp); 3391 3392 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag) 3393 { 3394 if (!sock_flag(sk, flag)) { 3395 unsigned long previous_flags = sk->sk_flags; 3396 3397 sock_set_flag(sk, flag); 3398 /* 3399 * we just set one of the two flags which require net 3400 * time stamping, but time stamping might have been on 3401 * already because of the other one 3402 */ 3403 if (sock_needs_netstamp(sk) && 3404 !(previous_flags & SK_FLAGS_TIMESTAMP)) 3405 net_enable_timestamp(); 3406 } 3407 } 3408 3409 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, 3410 int level, int type) 3411 { 3412 struct sock_exterr_skb *serr; 3413 struct sk_buff *skb; 3414 int copied, err; 3415 3416 err = -EAGAIN; 3417 skb = sock_dequeue_err_skb(sk); 3418 if (skb == NULL) 3419 goto out; 3420 3421 copied = skb->len; 3422 if (copied > len) { 3423 msg->msg_flags |= MSG_TRUNC; 3424 copied = len; 3425 } 3426 err = skb_copy_datagram_msg(skb, 0, msg, copied); 3427 if (err) 3428 goto out_free_skb; 3429 3430 sock_recv_timestamp(msg, sk, skb); 3431 3432 serr = SKB_EXT_ERR(skb); 3433 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee); 3434 3435 msg->msg_flags |= MSG_ERRQUEUE; 3436 err = copied; 3437 3438 out_free_skb: 3439 kfree_skb(skb); 3440 out: 3441 return err; 3442 } 3443 EXPORT_SYMBOL(sock_recv_errqueue); 3444 3445 /* 3446 * Get a socket option on an socket. 3447 * 3448 * FIX: POSIX 1003.1g is very ambiguous here. It states that 3449 * asynchronous errors should be reported by getsockopt. We assume 3450 * this means if you specify SO_ERROR (otherwise whats the point of it). 3451 */ 3452 int sock_common_getsockopt(struct socket *sock, int level, int optname, 3453 char __user *optval, int __user *optlen) 3454 { 3455 struct sock *sk = sock->sk; 3456 3457 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen); 3458 } 3459 EXPORT_SYMBOL(sock_common_getsockopt); 3460 3461 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 3462 int flags) 3463 { 3464 struct sock *sk = sock->sk; 3465 int addr_len = 0; 3466 int err; 3467 3468 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT, 3469 flags & ~MSG_DONTWAIT, &addr_len); 3470 if (err >= 0) 3471 msg->msg_namelen = addr_len; 3472 return err; 3473 } 3474 EXPORT_SYMBOL(sock_common_recvmsg); 3475 3476 /* 3477 * Set socket options on an inet socket. 3478 */ 3479 int sock_common_setsockopt(struct socket *sock, int level, int optname, 3480 sockptr_t optval, unsigned int optlen) 3481 { 3482 struct sock *sk = sock->sk; 3483 3484 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen); 3485 } 3486 EXPORT_SYMBOL(sock_common_setsockopt); 3487 3488 void sk_common_release(struct sock *sk) 3489 { 3490 if (sk->sk_prot->destroy) 3491 sk->sk_prot->destroy(sk); 3492 3493 /* 3494 * Observation: when sk_common_release is called, processes have 3495 * no access to socket. But net still has. 3496 * Step one, detach it from networking: 3497 * 3498 * A. Remove from hash tables. 3499 */ 3500 3501 sk->sk_prot->unhash(sk); 3502 3503 /* 3504 * In this point socket cannot receive new packets, but it is possible 3505 * that some packets are in flight because some CPU runs receiver and 3506 * did hash table lookup before we unhashed socket. They will achieve 3507 * receive queue and will be purged by socket destructor. 3508 * 3509 * Also we still have packets pending on receive queue and probably, 3510 * our own packets waiting in device queues. sock_destroy will drain 3511 * receive queue, but transmitted packets will delay socket destruction 3512 * until the last reference will be released. 3513 */ 3514 3515 sock_orphan(sk); 3516 3517 xfrm_sk_free_policy(sk); 3518 3519 sk_refcnt_debug_release(sk); 3520 3521 sock_put(sk); 3522 } 3523 EXPORT_SYMBOL(sk_common_release); 3524 3525 void sk_get_meminfo(const struct sock *sk, u32 *mem) 3526 { 3527 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS); 3528 3529 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk); 3530 mem[SK_MEMINFO_RCVBUF] = READ_ONCE(sk->sk_rcvbuf); 3531 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk); 3532 mem[SK_MEMINFO_SNDBUF] = READ_ONCE(sk->sk_sndbuf); 3533 mem[SK_MEMINFO_FWD_ALLOC] = sk->sk_forward_alloc; 3534 mem[SK_MEMINFO_WMEM_QUEUED] = READ_ONCE(sk->sk_wmem_queued); 3535 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc); 3536 mem[SK_MEMINFO_BACKLOG] = READ_ONCE(sk->sk_backlog.len); 3537 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops); 3538 } 3539 3540 #ifdef CONFIG_PROC_FS 3541 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR); 3542 3543 int sock_prot_inuse_get(struct net *net, struct proto *prot) 3544 { 3545 int cpu, idx = prot->inuse_idx; 3546 int res = 0; 3547 3548 for_each_possible_cpu(cpu) 3549 res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx]; 3550 3551 return res >= 0 ? res : 0; 3552 } 3553 EXPORT_SYMBOL_GPL(sock_prot_inuse_get); 3554 3555 int sock_inuse_get(struct net *net) 3556 { 3557 int cpu, res = 0; 3558 3559 for_each_possible_cpu(cpu) 3560 res += per_cpu_ptr(net->core.prot_inuse, cpu)->all; 3561 3562 return res; 3563 } 3564 3565 EXPORT_SYMBOL_GPL(sock_inuse_get); 3566 3567 static int __net_init sock_inuse_init_net(struct net *net) 3568 { 3569 net->core.prot_inuse = alloc_percpu(struct prot_inuse); 3570 if (net->core.prot_inuse == NULL) 3571 return -ENOMEM; 3572 return 0; 3573 } 3574 3575 static void __net_exit sock_inuse_exit_net(struct net *net) 3576 { 3577 free_percpu(net->core.prot_inuse); 3578 } 3579 3580 static struct pernet_operations net_inuse_ops = { 3581 .init = sock_inuse_init_net, 3582 .exit = sock_inuse_exit_net, 3583 }; 3584 3585 static __init int net_inuse_init(void) 3586 { 3587 if (register_pernet_subsys(&net_inuse_ops)) 3588 panic("Cannot initialize net inuse counters"); 3589 3590 return 0; 3591 } 3592 3593 core_initcall(net_inuse_init); 3594 3595 static int assign_proto_idx(struct proto *prot) 3596 { 3597 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR); 3598 3599 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) { 3600 pr_err("PROTO_INUSE_NR exhausted\n"); 3601 return -ENOSPC; 3602 } 3603 3604 set_bit(prot->inuse_idx, proto_inuse_idx); 3605 return 0; 3606 } 3607 3608 static void release_proto_idx(struct proto *prot) 3609 { 3610 if (prot->inuse_idx != PROTO_INUSE_NR - 1) 3611 clear_bit(prot->inuse_idx, proto_inuse_idx); 3612 } 3613 #else 3614 static inline int assign_proto_idx(struct proto *prot) 3615 { 3616 return 0; 3617 } 3618 3619 static inline void release_proto_idx(struct proto *prot) 3620 { 3621 } 3622 3623 #endif 3624 3625 static void tw_prot_cleanup(struct timewait_sock_ops *twsk_prot) 3626 { 3627 if (!twsk_prot) 3628 return; 3629 kfree(twsk_prot->twsk_slab_name); 3630 twsk_prot->twsk_slab_name = NULL; 3631 kmem_cache_destroy(twsk_prot->twsk_slab); 3632 twsk_prot->twsk_slab = NULL; 3633 } 3634 3635 static int tw_prot_init(const struct proto *prot) 3636 { 3637 struct timewait_sock_ops *twsk_prot = prot->twsk_prot; 3638 3639 if (!twsk_prot) 3640 return 0; 3641 3642 twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", 3643 prot->name); 3644 if (!twsk_prot->twsk_slab_name) 3645 return -ENOMEM; 3646 3647 twsk_prot->twsk_slab = 3648 kmem_cache_create(twsk_prot->twsk_slab_name, 3649 twsk_prot->twsk_obj_size, 0, 3650 SLAB_ACCOUNT | prot->slab_flags, 3651 NULL); 3652 if (!twsk_prot->twsk_slab) { 3653 pr_crit("%s: Can't create timewait sock SLAB cache!\n", 3654 prot->name); 3655 return -ENOMEM; 3656 } 3657 3658 return 0; 3659 } 3660 3661 static void req_prot_cleanup(struct request_sock_ops *rsk_prot) 3662 { 3663 if (!rsk_prot) 3664 return; 3665 kfree(rsk_prot->slab_name); 3666 rsk_prot->slab_name = NULL; 3667 kmem_cache_destroy(rsk_prot->slab); 3668 rsk_prot->slab = NULL; 3669 } 3670 3671 static int req_prot_init(const struct proto *prot) 3672 { 3673 struct request_sock_ops *rsk_prot = prot->rsk_prot; 3674 3675 if (!rsk_prot) 3676 return 0; 3677 3678 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", 3679 prot->name); 3680 if (!rsk_prot->slab_name) 3681 return -ENOMEM; 3682 3683 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name, 3684 rsk_prot->obj_size, 0, 3685 SLAB_ACCOUNT | prot->slab_flags, 3686 NULL); 3687 3688 if (!rsk_prot->slab) { 3689 pr_crit("%s: Can't create request sock SLAB cache!\n", 3690 prot->name); 3691 return -ENOMEM; 3692 } 3693 return 0; 3694 } 3695 3696 int proto_register(struct proto *prot, int alloc_slab) 3697 { 3698 int ret = -ENOBUFS; 3699 3700 if (alloc_slab) { 3701 prot->slab = kmem_cache_create_usercopy(prot->name, 3702 prot->obj_size, 0, 3703 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT | 3704 prot->slab_flags, 3705 prot->useroffset, prot->usersize, 3706 NULL); 3707 3708 if (prot->slab == NULL) { 3709 pr_crit("%s: Can't create sock SLAB cache!\n", 3710 prot->name); 3711 goto out; 3712 } 3713 3714 if (req_prot_init(prot)) 3715 goto out_free_request_sock_slab; 3716 3717 if (tw_prot_init(prot)) 3718 goto out_free_timewait_sock_slab; 3719 } 3720 3721 mutex_lock(&proto_list_mutex); 3722 ret = assign_proto_idx(prot); 3723 if (ret) { 3724 mutex_unlock(&proto_list_mutex); 3725 goto out_free_timewait_sock_slab; 3726 } 3727 list_add(&prot->node, &proto_list); 3728 mutex_unlock(&proto_list_mutex); 3729 return ret; 3730 3731 out_free_timewait_sock_slab: 3732 if (alloc_slab) 3733 tw_prot_cleanup(prot->twsk_prot); 3734 out_free_request_sock_slab: 3735 if (alloc_slab) { 3736 req_prot_cleanup(prot->rsk_prot); 3737 3738 kmem_cache_destroy(prot->slab); 3739 prot->slab = NULL; 3740 } 3741 out: 3742 return ret; 3743 } 3744 EXPORT_SYMBOL(proto_register); 3745 3746 void proto_unregister(struct proto *prot) 3747 { 3748 mutex_lock(&proto_list_mutex); 3749 release_proto_idx(prot); 3750 list_del(&prot->node); 3751 mutex_unlock(&proto_list_mutex); 3752 3753 kmem_cache_destroy(prot->slab); 3754 prot->slab = NULL; 3755 3756 req_prot_cleanup(prot->rsk_prot); 3757 tw_prot_cleanup(prot->twsk_prot); 3758 } 3759 EXPORT_SYMBOL(proto_unregister); 3760 3761 int sock_load_diag_module(int family, int protocol) 3762 { 3763 if (!protocol) { 3764 if (!sock_is_registered(family)) 3765 return -ENOENT; 3766 3767 return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK, 3768 NETLINK_SOCK_DIAG, family); 3769 } 3770 3771 #ifdef CONFIG_INET 3772 if (family == AF_INET && 3773 protocol != IPPROTO_RAW && 3774 protocol < MAX_INET_PROTOS && 3775 !rcu_access_pointer(inet_protos[protocol])) 3776 return -ENOENT; 3777 #endif 3778 3779 return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK, 3780 NETLINK_SOCK_DIAG, family, protocol); 3781 } 3782 EXPORT_SYMBOL(sock_load_diag_module); 3783 3784 #ifdef CONFIG_PROC_FS 3785 static void *proto_seq_start(struct seq_file *seq, loff_t *pos) 3786 __acquires(proto_list_mutex) 3787 { 3788 mutex_lock(&proto_list_mutex); 3789 return seq_list_start_head(&proto_list, *pos); 3790 } 3791 3792 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos) 3793 { 3794 return seq_list_next(v, &proto_list, pos); 3795 } 3796 3797 static void proto_seq_stop(struct seq_file *seq, void *v) 3798 __releases(proto_list_mutex) 3799 { 3800 mutex_unlock(&proto_list_mutex); 3801 } 3802 3803 static char proto_method_implemented(const void *method) 3804 { 3805 return method == NULL ? 'n' : 'y'; 3806 } 3807 static long sock_prot_memory_allocated(struct proto *proto) 3808 { 3809 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L; 3810 } 3811 3812 static const char *sock_prot_memory_pressure(struct proto *proto) 3813 { 3814 return proto->memory_pressure != NULL ? 3815 proto_memory_pressure(proto) ? "yes" : "no" : "NI"; 3816 } 3817 3818 static void proto_seq_printf(struct seq_file *seq, struct proto *proto) 3819 { 3820 3821 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s " 3822 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n", 3823 proto->name, 3824 proto->obj_size, 3825 sock_prot_inuse_get(seq_file_net(seq), proto), 3826 sock_prot_memory_allocated(proto), 3827 sock_prot_memory_pressure(proto), 3828 proto->max_header, 3829 proto->slab == NULL ? "no" : "yes", 3830 module_name(proto->owner), 3831 proto_method_implemented(proto->close), 3832 proto_method_implemented(proto->connect), 3833 proto_method_implemented(proto->disconnect), 3834 proto_method_implemented(proto->accept), 3835 proto_method_implemented(proto->ioctl), 3836 proto_method_implemented(proto->init), 3837 proto_method_implemented(proto->destroy), 3838 proto_method_implemented(proto->shutdown), 3839 proto_method_implemented(proto->setsockopt), 3840 proto_method_implemented(proto->getsockopt), 3841 proto_method_implemented(proto->sendmsg), 3842 proto_method_implemented(proto->recvmsg), 3843 proto_method_implemented(proto->sendpage), 3844 proto_method_implemented(proto->bind), 3845 proto_method_implemented(proto->backlog_rcv), 3846 proto_method_implemented(proto->hash), 3847 proto_method_implemented(proto->unhash), 3848 proto_method_implemented(proto->get_port), 3849 proto_method_implemented(proto->enter_memory_pressure)); 3850 } 3851 3852 static int proto_seq_show(struct seq_file *seq, void *v) 3853 { 3854 if (v == &proto_list) 3855 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s", 3856 "protocol", 3857 "size", 3858 "sockets", 3859 "memory", 3860 "press", 3861 "maxhdr", 3862 "slab", 3863 "module", 3864 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n"); 3865 else 3866 proto_seq_printf(seq, list_entry(v, struct proto, node)); 3867 return 0; 3868 } 3869 3870 static const struct seq_operations proto_seq_ops = { 3871 .start = proto_seq_start, 3872 .next = proto_seq_next, 3873 .stop = proto_seq_stop, 3874 .show = proto_seq_show, 3875 }; 3876 3877 static __net_init int proto_init_net(struct net *net) 3878 { 3879 if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops, 3880 sizeof(struct seq_net_private))) 3881 return -ENOMEM; 3882 3883 return 0; 3884 } 3885 3886 static __net_exit void proto_exit_net(struct net *net) 3887 { 3888 remove_proc_entry("protocols", net->proc_net); 3889 } 3890 3891 3892 static __net_initdata struct pernet_operations proto_net_ops = { 3893 .init = proto_init_net, 3894 .exit = proto_exit_net, 3895 }; 3896 3897 static int __init proto_init(void) 3898 { 3899 return register_pernet_subsys(&proto_net_ops); 3900 } 3901 3902 subsys_initcall(proto_init); 3903 3904 #endif /* PROC_FS */ 3905 3906 #ifdef CONFIG_NET_RX_BUSY_POLL 3907 bool sk_busy_loop_end(void *p, unsigned long start_time) 3908 { 3909 struct sock *sk = p; 3910 3911 return !skb_queue_empty_lockless(&sk->sk_receive_queue) || 3912 sk_busy_loop_timeout(sk, start_time); 3913 } 3914 EXPORT_SYMBOL(sk_busy_loop_end); 3915 #endif /* CONFIG_NET_RX_BUSY_POLL */ 3916 3917 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len) 3918 { 3919 if (!sk->sk_prot->bind_add) 3920 return -EOPNOTSUPP; 3921 return sk->sk_prot->bind_add(sk, addr, addr_len); 3922 } 3923 EXPORT_SYMBOL(sock_bind_add); 3924