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 static void sock_inuse_add(struct net *net, int val); 148 149 /** 150 * sk_ns_capable - General socket capability test 151 * @sk: Socket to use a capability on or through 152 * @user_ns: The user namespace of the capability to use 153 * @cap: The capability to use 154 * 155 * Test to see if the opener of the socket had when the socket was 156 * created and the current process has the capability @cap in the user 157 * namespace @user_ns. 158 */ 159 bool sk_ns_capable(const struct sock *sk, 160 struct user_namespace *user_ns, int cap) 161 { 162 return file_ns_capable(sk->sk_socket->file, user_ns, cap) && 163 ns_capable(user_ns, cap); 164 } 165 EXPORT_SYMBOL(sk_ns_capable); 166 167 /** 168 * sk_capable - Socket global capability test 169 * @sk: Socket to use a capability on or through 170 * @cap: The global capability to use 171 * 172 * Test to see if the opener of the socket had when the socket was 173 * created and the current process has the capability @cap in all user 174 * namespaces. 175 */ 176 bool sk_capable(const struct sock *sk, int cap) 177 { 178 return sk_ns_capable(sk, &init_user_ns, cap); 179 } 180 EXPORT_SYMBOL(sk_capable); 181 182 /** 183 * sk_net_capable - Network namespace socket capability test 184 * @sk: Socket to use a capability on or through 185 * @cap: The capability to use 186 * 187 * Test to see if the opener of the socket had when the socket was created 188 * and the current process has the capability @cap over the network namespace 189 * the socket is a member of. 190 */ 191 bool sk_net_capable(const struct sock *sk, int cap) 192 { 193 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap); 194 } 195 EXPORT_SYMBOL(sk_net_capable); 196 197 /* 198 * Each address family might have different locking rules, so we have 199 * one slock key per address family and separate keys for internal and 200 * userspace sockets. 201 */ 202 static struct lock_class_key af_family_keys[AF_MAX]; 203 static struct lock_class_key af_family_kern_keys[AF_MAX]; 204 static struct lock_class_key af_family_slock_keys[AF_MAX]; 205 static struct lock_class_key af_family_kern_slock_keys[AF_MAX]; 206 207 /* 208 * Make lock validator output more readable. (we pre-construct these 209 * strings build-time, so that runtime initialization of socket 210 * locks is fast): 211 */ 212 213 #define _sock_locks(x) \ 214 x "AF_UNSPEC", x "AF_UNIX" , x "AF_INET" , \ 215 x "AF_AX25" , x "AF_IPX" , x "AF_APPLETALK", \ 216 x "AF_NETROM", x "AF_BRIDGE" , x "AF_ATMPVC" , \ 217 x "AF_X25" , x "AF_INET6" , x "AF_ROSE" , \ 218 x "AF_DECnet", x "AF_NETBEUI" , x "AF_SECURITY" , \ 219 x "AF_KEY" , x "AF_NETLINK" , x "AF_PACKET" , \ 220 x "AF_ASH" , x "AF_ECONET" , x "AF_ATMSVC" , \ 221 x "AF_RDS" , x "AF_SNA" , x "AF_IRDA" , \ 222 x "AF_PPPOX" , x "AF_WANPIPE" , x "AF_LLC" , \ 223 x "27" , x "28" , x "AF_CAN" , \ 224 x "AF_TIPC" , x "AF_BLUETOOTH", x "IUCV" , \ 225 x "AF_RXRPC" , x "AF_ISDN" , x "AF_PHONET" , \ 226 x "AF_IEEE802154", x "AF_CAIF" , x "AF_ALG" , \ 227 x "AF_NFC" , x "AF_VSOCK" , x "AF_KCM" , \ 228 x "AF_QIPCRTR", x "AF_SMC" , x "AF_XDP" , \ 229 x "AF_MCTP" , \ 230 x "AF_MAX" 231 232 static const char *const af_family_key_strings[AF_MAX+1] = { 233 _sock_locks("sk_lock-") 234 }; 235 static const char *const af_family_slock_key_strings[AF_MAX+1] = { 236 _sock_locks("slock-") 237 }; 238 static const char *const af_family_clock_key_strings[AF_MAX+1] = { 239 _sock_locks("clock-") 240 }; 241 242 static const char *const af_family_kern_key_strings[AF_MAX+1] = { 243 _sock_locks("k-sk_lock-") 244 }; 245 static const char *const af_family_kern_slock_key_strings[AF_MAX+1] = { 246 _sock_locks("k-slock-") 247 }; 248 static const char *const af_family_kern_clock_key_strings[AF_MAX+1] = { 249 _sock_locks("k-clock-") 250 }; 251 static const char *const af_family_rlock_key_strings[AF_MAX+1] = { 252 _sock_locks("rlock-") 253 }; 254 static const char *const af_family_wlock_key_strings[AF_MAX+1] = { 255 _sock_locks("wlock-") 256 }; 257 static const char *const af_family_elock_key_strings[AF_MAX+1] = { 258 _sock_locks("elock-") 259 }; 260 261 /* 262 * sk_callback_lock and sk queues locking rules are per-address-family, 263 * so split the lock classes by using a per-AF key: 264 */ 265 static struct lock_class_key af_callback_keys[AF_MAX]; 266 static struct lock_class_key af_rlock_keys[AF_MAX]; 267 static struct lock_class_key af_wlock_keys[AF_MAX]; 268 static struct lock_class_key af_elock_keys[AF_MAX]; 269 static struct lock_class_key af_kern_callback_keys[AF_MAX]; 270 271 /* Run time adjustable parameters. */ 272 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX; 273 EXPORT_SYMBOL(sysctl_wmem_max); 274 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX; 275 EXPORT_SYMBOL(sysctl_rmem_max); 276 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX; 277 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX; 278 279 /* Maximal space eaten by iovec or ancillary data plus some space */ 280 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512); 281 EXPORT_SYMBOL(sysctl_optmem_max); 282 283 int sysctl_tstamp_allow_data __read_mostly = 1; 284 285 DEFINE_STATIC_KEY_FALSE(memalloc_socks_key); 286 EXPORT_SYMBOL_GPL(memalloc_socks_key); 287 288 /** 289 * sk_set_memalloc - sets %SOCK_MEMALLOC 290 * @sk: socket to set it on 291 * 292 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves. 293 * It's the responsibility of the admin to adjust min_free_kbytes 294 * to meet the requirements 295 */ 296 void sk_set_memalloc(struct sock *sk) 297 { 298 sock_set_flag(sk, SOCK_MEMALLOC); 299 sk->sk_allocation |= __GFP_MEMALLOC; 300 static_branch_inc(&memalloc_socks_key); 301 } 302 EXPORT_SYMBOL_GPL(sk_set_memalloc); 303 304 void sk_clear_memalloc(struct sock *sk) 305 { 306 sock_reset_flag(sk, SOCK_MEMALLOC); 307 sk->sk_allocation &= ~__GFP_MEMALLOC; 308 static_branch_dec(&memalloc_socks_key); 309 310 /* 311 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward 312 * progress of swapping. SOCK_MEMALLOC may be cleared while 313 * it has rmem allocations due to the last swapfile being deactivated 314 * but there is a risk that the socket is unusable due to exceeding 315 * the rmem limits. Reclaim the reserves and obey rmem limits again. 316 */ 317 sk_mem_reclaim(sk); 318 } 319 EXPORT_SYMBOL_GPL(sk_clear_memalloc); 320 321 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb) 322 { 323 int ret; 324 unsigned int noreclaim_flag; 325 326 /* these should have been dropped before queueing */ 327 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC)); 328 329 noreclaim_flag = memalloc_noreclaim_save(); 330 ret = sk->sk_backlog_rcv(sk, skb); 331 memalloc_noreclaim_restore(noreclaim_flag); 332 333 return ret; 334 } 335 EXPORT_SYMBOL(__sk_backlog_rcv); 336 337 void sk_error_report(struct sock *sk) 338 { 339 sk->sk_error_report(sk); 340 341 switch (sk->sk_family) { 342 case AF_INET: 343 fallthrough; 344 case AF_INET6: 345 trace_inet_sk_error_report(sk); 346 break; 347 default: 348 break; 349 } 350 } 351 EXPORT_SYMBOL(sk_error_report); 352 353 static int sock_get_timeout(long timeo, void *optval, bool old_timeval) 354 { 355 struct __kernel_sock_timeval tv; 356 357 if (timeo == MAX_SCHEDULE_TIMEOUT) { 358 tv.tv_sec = 0; 359 tv.tv_usec = 0; 360 } else { 361 tv.tv_sec = timeo / HZ; 362 tv.tv_usec = ((timeo % HZ) * USEC_PER_SEC) / HZ; 363 } 364 365 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) { 366 struct old_timeval32 tv32 = { tv.tv_sec, tv.tv_usec }; 367 *(struct old_timeval32 *)optval = tv32; 368 return sizeof(tv32); 369 } 370 371 if (old_timeval) { 372 struct __kernel_old_timeval old_tv; 373 old_tv.tv_sec = tv.tv_sec; 374 old_tv.tv_usec = tv.tv_usec; 375 *(struct __kernel_old_timeval *)optval = old_tv; 376 return sizeof(old_tv); 377 } 378 379 *(struct __kernel_sock_timeval *)optval = tv; 380 return sizeof(tv); 381 } 382 383 static int sock_set_timeout(long *timeo_p, sockptr_t optval, int optlen, 384 bool old_timeval) 385 { 386 struct __kernel_sock_timeval tv; 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 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC) 414 return -EDOM; 415 416 if (tv.tv_sec < 0) { 417 static int warned __read_mostly; 418 419 *timeo_p = 0; 420 if (warned < 10 && net_ratelimit()) { 421 warned++; 422 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n", 423 __func__, current->comm, task_pid_nr(current)); 424 } 425 return 0; 426 } 427 *timeo_p = MAX_SCHEDULE_TIMEOUT; 428 if (tv.tv_sec == 0 && tv.tv_usec == 0) 429 return 0; 430 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1)) 431 *timeo_p = tv.tv_sec * HZ + DIV_ROUND_UP((unsigned long)tv.tv_usec, USEC_PER_SEC / HZ); 432 return 0; 433 } 434 435 static bool sock_needs_netstamp(const struct sock *sk) 436 { 437 switch (sk->sk_family) { 438 case AF_UNSPEC: 439 case AF_UNIX: 440 return false; 441 default: 442 return true; 443 } 444 } 445 446 static void sock_disable_timestamp(struct sock *sk, unsigned long flags) 447 { 448 if (sk->sk_flags & flags) { 449 sk->sk_flags &= ~flags; 450 if (sock_needs_netstamp(sk) && 451 !(sk->sk_flags & SK_FLAGS_TIMESTAMP)) 452 net_disable_timestamp(); 453 } 454 } 455 456 457 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 458 { 459 unsigned long flags; 460 struct sk_buff_head *list = &sk->sk_receive_queue; 461 462 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) { 463 atomic_inc(&sk->sk_drops); 464 trace_sock_rcvqueue_full(sk, skb); 465 return -ENOMEM; 466 } 467 468 if (!sk_rmem_schedule(sk, skb, skb->truesize)) { 469 atomic_inc(&sk->sk_drops); 470 return -ENOBUFS; 471 } 472 473 skb->dev = NULL; 474 skb_set_owner_r(skb, sk); 475 476 /* we escape from rcu protected region, make sure we dont leak 477 * a norefcounted dst 478 */ 479 skb_dst_force(skb); 480 481 spin_lock_irqsave(&list->lock, flags); 482 sock_skb_set_dropcount(sk, skb); 483 __skb_queue_tail(list, skb); 484 spin_unlock_irqrestore(&list->lock, flags); 485 486 if (!sock_flag(sk, SOCK_DEAD)) 487 sk->sk_data_ready(sk); 488 return 0; 489 } 490 EXPORT_SYMBOL(__sock_queue_rcv_skb); 491 492 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 493 { 494 int err; 495 496 err = sk_filter(sk, skb); 497 if (err) 498 return err; 499 500 return __sock_queue_rcv_skb(sk, skb); 501 } 502 EXPORT_SYMBOL(sock_queue_rcv_skb); 503 504 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, 505 const int nested, unsigned int trim_cap, bool refcounted) 506 { 507 int rc = NET_RX_SUCCESS; 508 509 if (sk_filter_trim_cap(sk, skb, trim_cap)) 510 goto discard_and_relse; 511 512 skb->dev = NULL; 513 514 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) { 515 atomic_inc(&sk->sk_drops); 516 goto discard_and_relse; 517 } 518 if (nested) 519 bh_lock_sock_nested(sk); 520 else 521 bh_lock_sock(sk); 522 if (!sock_owned_by_user(sk)) { 523 /* 524 * trylock + unlock semantics: 525 */ 526 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_); 527 528 rc = sk_backlog_rcv(sk, skb); 529 530 mutex_release(&sk->sk_lock.dep_map, _RET_IP_); 531 } else if (sk_add_backlog(sk, skb, READ_ONCE(sk->sk_rcvbuf))) { 532 bh_unlock_sock(sk); 533 atomic_inc(&sk->sk_drops); 534 goto discard_and_relse; 535 } 536 537 bh_unlock_sock(sk); 538 out: 539 if (refcounted) 540 sock_put(sk); 541 return rc; 542 discard_and_relse: 543 kfree_skb(skb); 544 goto out; 545 } 546 EXPORT_SYMBOL(__sk_receive_skb); 547 548 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ip6_dst_check(struct dst_entry *, 549 u32)); 550 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ipv4_dst_check(struct dst_entry *, 551 u32)); 552 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie) 553 { 554 struct dst_entry *dst = __sk_dst_get(sk); 555 556 if (dst && dst->obsolete && 557 INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check, 558 dst, cookie) == NULL) { 559 sk_tx_queue_clear(sk); 560 sk->sk_dst_pending_confirm = 0; 561 RCU_INIT_POINTER(sk->sk_dst_cache, NULL); 562 dst_release(dst); 563 return NULL; 564 } 565 566 return dst; 567 } 568 EXPORT_SYMBOL(__sk_dst_check); 569 570 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie) 571 { 572 struct dst_entry *dst = sk_dst_get(sk); 573 574 if (dst && dst->obsolete && 575 INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check, 576 dst, cookie) == NULL) { 577 sk_dst_reset(sk); 578 dst_release(dst); 579 return NULL; 580 } 581 582 return dst; 583 } 584 EXPORT_SYMBOL(sk_dst_check); 585 586 static int sock_bindtoindex_locked(struct sock *sk, int ifindex) 587 { 588 int ret = -ENOPROTOOPT; 589 #ifdef CONFIG_NETDEVICES 590 struct net *net = sock_net(sk); 591 592 /* Sorry... */ 593 ret = -EPERM; 594 if (sk->sk_bound_dev_if && !ns_capable(net->user_ns, CAP_NET_RAW)) 595 goto out; 596 597 ret = -EINVAL; 598 if (ifindex < 0) 599 goto out; 600 601 sk->sk_bound_dev_if = ifindex; 602 if (sk->sk_prot->rehash) 603 sk->sk_prot->rehash(sk); 604 sk_dst_reset(sk); 605 606 ret = 0; 607 608 out: 609 #endif 610 611 return ret; 612 } 613 614 int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk) 615 { 616 int ret; 617 618 if (lock_sk) 619 lock_sock(sk); 620 ret = sock_bindtoindex_locked(sk, ifindex); 621 if (lock_sk) 622 release_sock(sk); 623 624 return ret; 625 } 626 EXPORT_SYMBOL(sock_bindtoindex); 627 628 static int sock_setbindtodevice(struct sock *sk, sockptr_t optval, int optlen) 629 { 630 int ret = -ENOPROTOOPT; 631 #ifdef CONFIG_NETDEVICES 632 struct net *net = sock_net(sk); 633 char devname[IFNAMSIZ]; 634 int index; 635 636 ret = -EINVAL; 637 if (optlen < 0) 638 goto out; 639 640 /* Bind this socket to a particular device like "eth0", 641 * as specified in the passed interface name. If the 642 * name is "" or the option length is zero the socket 643 * is not bound. 644 */ 645 if (optlen > IFNAMSIZ - 1) 646 optlen = IFNAMSIZ - 1; 647 memset(devname, 0, sizeof(devname)); 648 649 ret = -EFAULT; 650 if (copy_from_sockptr(devname, optval, optlen)) 651 goto out; 652 653 index = 0; 654 if (devname[0] != '\0') { 655 struct net_device *dev; 656 657 rcu_read_lock(); 658 dev = dev_get_by_name_rcu(net, devname); 659 if (dev) 660 index = dev->ifindex; 661 rcu_read_unlock(); 662 ret = -ENODEV; 663 if (!dev) 664 goto out; 665 } 666 667 return sock_bindtoindex(sk, index, true); 668 out: 669 #endif 670 671 return ret; 672 } 673 674 static int sock_getbindtodevice(struct sock *sk, char __user *optval, 675 int __user *optlen, int len) 676 { 677 int ret = -ENOPROTOOPT; 678 #ifdef CONFIG_NETDEVICES 679 struct net *net = sock_net(sk); 680 char devname[IFNAMSIZ]; 681 682 if (sk->sk_bound_dev_if == 0) { 683 len = 0; 684 goto zero; 685 } 686 687 ret = -EINVAL; 688 if (len < IFNAMSIZ) 689 goto out; 690 691 ret = netdev_get_name(net, devname, sk->sk_bound_dev_if); 692 if (ret) 693 goto out; 694 695 len = strlen(devname) + 1; 696 697 ret = -EFAULT; 698 if (copy_to_user(optval, devname, len)) 699 goto out; 700 701 zero: 702 ret = -EFAULT; 703 if (put_user(len, optlen)) 704 goto out; 705 706 ret = 0; 707 708 out: 709 #endif 710 711 return ret; 712 } 713 714 bool sk_mc_loop(struct sock *sk) 715 { 716 if (dev_recursion_level()) 717 return false; 718 if (!sk) 719 return true; 720 switch (sk->sk_family) { 721 case AF_INET: 722 return inet_sk(sk)->mc_loop; 723 #if IS_ENABLED(CONFIG_IPV6) 724 case AF_INET6: 725 return inet6_sk(sk)->mc_loop; 726 #endif 727 } 728 WARN_ON_ONCE(1); 729 return true; 730 } 731 EXPORT_SYMBOL(sk_mc_loop); 732 733 void sock_set_reuseaddr(struct sock *sk) 734 { 735 lock_sock(sk); 736 sk->sk_reuse = SK_CAN_REUSE; 737 release_sock(sk); 738 } 739 EXPORT_SYMBOL(sock_set_reuseaddr); 740 741 void sock_set_reuseport(struct sock *sk) 742 { 743 lock_sock(sk); 744 sk->sk_reuseport = true; 745 release_sock(sk); 746 } 747 EXPORT_SYMBOL(sock_set_reuseport); 748 749 void sock_no_linger(struct sock *sk) 750 { 751 lock_sock(sk); 752 sk->sk_lingertime = 0; 753 sock_set_flag(sk, SOCK_LINGER); 754 release_sock(sk); 755 } 756 EXPORT_SYMBOL(sock_no_linger); 757 758 void sock_set_priority(struct sock *sk, u32 priority) 759 { 760 lock_sock(sk); 761 sk->sk_priority = priority; 762 release_sock(sk); 763 } 764 EXPORT_SYMBOL(sock_set_priority); 765 766 void sock_set_sndtimeo(struct sock *sk, s64 secs) 767 { 768 lock_sock(sk); 769 if (secs && secs < MAX_SCHEDULE_TIMEOUT / HZ - 1) 770 sk->sk_sndtimeo = secs * HZ; 771 else 772 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT; 773 release_sock(sk); 774 } 775 EXPORT_SYMBOL(sock_set_sndtimeo); 776 777 static void __sock_set_timestamps(struct sock *sk, bool val, bool new, bool ns) 778 { 779 if (val) { 780 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, new); 781 sock_valbool_flag(sk, SOCK_RCVTSTAMPNS, ns); 782 sock_set_flag(sk, SOCK_RCVTSTAMP); 783 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 784 } else { 785 sock_reset_flag(sk, SOCK_RCVTSTAMP); 786 sock_reset_flag(sk, SOCK_RCVTSTAMPNS); 787 } 788 } 789 790 void sock_enable_timestamps(struct sock *sk) 791 { 792 lock_sock(sk); 793 __sock_set_timestamps(sk, true, false, true); 794 release_sock(sk); 795 } 796 EXPORT_SYMBOL(sock_enable_timestamps); 797 798 void sock_set_timestamp(struct sock *sk, int optname, bool valbool) 799 { 800 switch (optname) { 801 case SO_TIMESTAMP_OLD: 802 __sock_set_timestamps(sk, valbool, false, false); 803 break; 804 case SO_TIMESTAMP_NEW: 805 __sock_set_timestamps(sk, valbool, true, false); 806 break; 807 case SO_TIMESTAMPNS_OLD: 808 __sock_set_timestamps(sk, valbool, false, true); 809 break; 810 case SO_TIMESTAMPNS_NEW: 811 __sock_set_timestamps(sk, valbool, true, true); 812 break; 813 } 814 } 815 816 static int sock_timestamping_bind_phc(struct sock *sk, int phc_index) 817 { 818 struct net *net = sock_net(sk); 819 struct net_device *dev = NULL; 820 bool match = false; 821 int *vclock_index; 822 int i, num; 823 824 if (sk->sk_bound_dev_if) 825 dev = dev_get_by_index(net, sk->sk_bound_dev_if); 826 827 if (!dev) { 828 pr_err("%s: sock not bind to device\n", __func__); 829 return -EOPNOTSUPP; 830 } 831 832 num = ethtool_get_phc_vclocks(dev, &vclock_index); 833 for (i = 0; i < num; i++) { 834 if (*(vclock_index + i) == phc_index) { 835 match = true; 836 break; 837 } 838 } 839 840 if (num > 0) 841 kfree(vclock_index); 842 843 if (!match) 844 return -EINVAL; 845 846 sk->sk_bind_phc = phc_index; 847 848 return 0; 849 } 850 851 int sock_set_timestamping(struct sock *sk, int optname, 852 struct so_timestamping timestamping) 853 { 854 int val = timestamping.flags; 855 int ret; 856 857 if (val & ~SOF_TIMESTAMPING_MASK) 858 return -EINVAL; 859 860 if (val & SOF_TIMESTAMPING_OPT_ID && 861 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) { 862 if (sk->sk_protocol == IPPROTO_TCP && 863 sk->sk_type == SOCK_STREAM) { 864 if ((1 << sk->sk_state) & 865 (TCPF_CLOSE | TCPF_LISTEN)) 866 return -EINVAL; 867 sk->sk_tskey = tcp_sk(sk)->snd_una; 868 } else { 869 sk->sk_tskey = 0; 870 } 871 } 872 873 if (val & SOF_TIMESTAMPING_OPT_STATS && 874 !(val & SOF_TIMESTAMPING_OPT_TSONLY)) 875 return -EINVAL; 876 877 if (val & SOF_TIMESTAMPING_BIND_PHC) { 878 ret = sock_timestamping_bind_phc(sk, timestamping.bind_phc); 879 if (ret) 880 return ret; 881 } 882 883 sk->sk_tsflags = val; 884 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, optname == SO_TIMESTAMPING_NEW); 885 886 if (val & SOF_TIMESTAMPING_RX_SOFTWARE) 887 sock_enable_timestamp(sk, 888 SOCK_TIMESTAMPING_RX_SOFTWARE); 889 else 890 sock_disable_timestamp(sk, 891 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE)); 892 return 0; 893 } 894 895 void sock_set_keepalive(struct sock *sk) 896 { 897 lock_sock(sk); 898 if (sk->sk_prot->keepalive) 899 sk->sk_prot->keepalive(sk, true); 900 sock_valbool_flag(sk, SOCK_KEEPOPEN, true); 901 release_sock(sk); 902 } 903 EXPORT_SYMBOL(sock_set_keepalive); 904 905 static void __sock_set_rcvbuf(struct sock *sk, int val) 906 { 907 /* Ensure val * 2 fits into an int, to prevent max_t() from treating it 908 * as a negative value. 909 */ 910 val = min_t(int, val, INT_MAX / 2); 911 sk->sk_userlocks |= SOCK_RCVBUF_LOCK; 912 913 /* We double it on the way in to account for "struct sk_buff" etc. 914 * overhead. Applications assume that the SO_RCVBUF setting they make 915 * will allow that much actual data to be received on that socket. 916 * 917 * Applications are unaware that "struct sk_buff" and other overheads 918 * allocate from the receive buffer during socket buffer allocation. 919 * 920 * And after considering the possible alternatives, returning the value 921 * we actually used in getsockopt is the most desirable behavior. 922 */ 923 WRITE_ONCE(sk->sk_rcvbuf, max_t(int, val * 2, SOCK_MIN_RCVBUF)); 924 } 925 926 void sock_set_rcvbuf(struct sock *sk, int val) 927 { 928 lock_sock(sk); 929 __sock_set_rcvbuf(sk, val); 930 release_sock(sk); 931 } 932 EXPORT_SYMBOL(sock_set_rcvbuf); 933 934 static void __sock_set_mark(struct sock *sk, u32 val) 935 { 936 if (val != sk->sk_mark) { 937 sk->sk_mark = val; 938 sk_dst_reset(sk); 939 } 940 } 941 942 void sock_set_mark(struct sock *sk, u32 val) 943 { 944 lock_sock(sk); 945 __sock_set_mark(sk, val); 946 release_sock(sk); 947 } 948 EXPORT_SYMBOL(sock_set_mark); 949 950 /* 951 * This is meant for all protocols to use and covers goings on 952 * at the socket level. Everything here is generic. 953 */ 954 955 int sock_setsockopt(struct socket *sock, int level, int optname, 956 sockptr_t optval, unsigned int optlen) 957 { 958 struct so_timestamping timestamping; 959 struct sock_txtime sk_txtime; 960 struct sock *sk = sock->sk; 961 int val; 962 int valbool; 963 struct linger ling; 964 int ret = 0; 965 966 /* 967 * Options without arguments 968 */ 969 970 if (optname == SO_BINDTODEVICE) 971 return sock_setbindtodevice(sk, optval, optlen); 972 973 if (optlen < sizeof(int)) 974 return -EINVAL; 975 976 if (copy_from_sockptr(&val, optval, sizeof(val))) 977 return -EFAULT; 978 979 valbool = val ? 1 : 0; 980 981 lock_sock(sk); 982 983 switch (optname) { 984 case SO_DEBUG: 985 if (val && !capable(CAP_NET_ADMIN)) 986 ret = -EACCES; 987 else 988 sock_valbool_flag(sk, SOCK_DBG, valbool); 989 break; 990 case SO_REUSEADDR: 991 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE); 992 break; 993 case SO_REUSEPORT: 994 sk->sk_reuseport = valbool; 995 break; 996 case SO_TYPE: 997 case SO_PROTOCOL: 998 case SO_DOMAIN: 999 case SO_ERROR: 1000 ret = -ENOPROTOOPT; 1001 break; 1002 case SO_DONTROUTE: 1003 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool); 1004 sk_dst_reset(sk); 1005 break; 1006 case SO_BROADCAST: 1007 sock_valbool_flag(sk, SOCK_BROADCAST, valbool); 1008 break; 1009 case SO_SNDBUF: 1010 /* Don't error on this BSD doesn't and if you think 1011 * about it this is right. Otherwise apps have to 1012 * play 'guess the biggest size' games. RCVBUF/SNDBUF 1013 * are treated in BSD as hints 1014 */ 1015 val = min_t(u32, val, sysctl_wmem_max); 1016 set_sndbuf: 1017 /* Ensure val * 2 fits into an int, to prevent max_t() 1018 * from treating it as a negative value. 1019 */ 1020 val = min_t(int, val, INT_MAX / 2); 1021 sk->sk_userlocks |= SOCK_SNDBUF_LOCK; 1022 WRITE_ONCE(sk->sk_sndbuf, 1023 max_t(int, val * 2, SOCK_MIN_SNDBUF)); 1024 /* Wake up sending tasks if we upped the value. */ 1025 sk->sk_write_space(sk); 1026 break; 1027 1028 case SO_SNDBUFFORCE: 1029 if (!capable(CAP_NET_ADMIN)) { 1030 ret = -EPERM; 1031 break; 1032 } 1033 1034 /* No negative values (to prevent underflow, as val will be 1035 * multiplied by 2). 1036 */ 1037 if (val < 0) 1038 val = 0; 1039 goto set_sndbuf; 1040 1041 case SO_RCVBUF: 1042 /* Don't error on this BSD doesn't and if you think 1043 * about it this is right. Otherwise apps have to 1044 * play 'guess the biggest size' games. RCVBUF/SNDBUF 1045 * are treated in BSD as hints 1046 */ 1047 __sock_set_rcvbuf(sk, min_t(u32, val, sysctl_rmem_max)); 1048 break; 1049 1050 case SO_RCVBUFFORCE: 1051 if (!capable(CAP_NET_ADMIN)) { 1052 ret = -EPERM; 1053 break; 1054 } 1055 1056 /* No negative values (to prevent underflow, as val will be 1057 * multiplied by 2). 1058 */ 1059 __sock_set_rcvbuf(sk, max(val, 0)); 1060 break; 1061 1062 case SO_KEEPALIVE: 1063 if (sk->sk_prot->keepalive) 1064 sk->sk_prot->keepalive(sk, valbool); 1065 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool); 1066 break; 1067 1068 case SO_OOBINLINE: 1069 sock_valbool_flag(sk, SOCK_URGINLINE, valbool); 1070 break; 1071 1072 case SO_NO_CHECK: 1073 sk->sk_no_check_tx = valbool; 1074 break; 1075 1076 case SO_PRIORITY: 1077 if ((val >= 0 && val <= 6) || 1078 ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) 1079 sk->sk_priority = val; 1080 else 1081 ret = -EPERM; 1082 break; 1083 1084 case SO_LINGER: 1085 if (optlen < sizeof(ling)) { 1086 ret = -EINVAL; /* 1003.1g */ 1087 break; 1088 } 1089 if (copy_from_sockptr(&ling, optval, sizeof(ling))) { 1090 ret = -EFAULT; 1091 break; 1092 } 1093 if (!ling.l_onoff) 1094 sock_reset_flag(sk, SOCK_LINGER); 1095 else { 1096 #if (BITS_PER_LONG == 32) 1097 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ) 1098 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT; 1099 else 1100 #endif 1101 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ; 1102 sock_set_flag(sk, SOCK_LINGER); 1103 } 1104 break; 1105 1106 case SO_BSDCOMPAT: 1107 break; 1108 1109 case SO_PASSCRED: 1110 if (valbool) 1111 set_bit(SOCK_PASSCRED, &sock->flags); 1112 else 1113 clear_bit(SOCK_PASSCRED, &sock->flags); 1114 break; 1115 1116 case SO_TIMESTAMP_OLD: 1117 case SO_TIMESTAMP_NEW: 1118 case SO_TIMESTAMPNS_OLD: 1119 case SO_TIMESTAMPNS_NEW: 1120 sock_set_timestamp(sk, optname, valbool); 1121 break; 1122 1123 case SO_TIMESTAMPING_NEW: 1124 case SO_TIMESTAMPING_OLD: 1125 if (optlen == sizeof(timestamping)) { 1126 if (copy_from_sockptr(×tamping, optval, 1127 sizeof(timestamping))) { 1128 ret = -EFAULT; 1129 break; 1130 } 1131 } else { 1132 memset(×tamping, 0, sizeof(timestamping)); 1133 timestamping.flags = val; 1134 } 1135 ret = sock_set_timestamping(sk, optname, timestamping); 1136 break; 1137 1138 case SO_RCVLOWAT: 1139 if (val < 0) 1140 val = INT_MAX; 1141 if (sock->ops->set_rcvlowat) 1142 ret = sock->ops->set_rcvlowat(sk, val); 1143 else 1144 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1); 1145 break; 1146 1147 case SO_RCVTIMEO_OLD: 1148 case SO_RCVTIMEO_NEW: 1149 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, 1150 optlen, optname == SO_RCVTIMEO_OLD); 1151 break; 1152 1153 case SO_SNDTIMEO_OLD: 1154 case SO_SNDTIMEO_NEW: 1155 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, 1156 optlen, optname == SO_SNDTIMEO_OLD); 1157 break; 1158 1159 case SO_ATTACH_FILTER: { 1160 struct sock_fprog fprog; 1161 1162 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen); 1163 if (!ret) 1164 ret = sk_attach_filter(&fprog, sk); 1165 break; 1166 } 1167 case SO_ATTACH_BPF: 1168 ret = -EINVAL; 1169 if (optlen == sizeof(u32)) { 1170 u32 ufd; 1171 1172 ret = -EFAULT; 1173 if (copy_from_sockptr(&ufd, optval, sizeof(ufd))) 1174 break; 1175 1176 ret = sk_attach_bpf(ufd, sk); 1177 } 1178 break; 1179 1180 case SO_ATTACH_REUSEPORT_CBPF: { 1181 struct sock_fprog fprog; 1182 1183 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen); 1184 if (!ret) 1185 ret = sk_reuseport_attach_filter(&fprog, sk); 1186 break; 1187 } 1188 case SO_ATTACH_REUSEPORT_EBPF: 1189 ret = -EINVAL; 1190 if (optlen == sizeof(u32)) { 1191 u32 ufd; 1192 1193 ret = -EFAULT; 1194 if (copy_from_sockptr(&ufd, optval, sizeof(ufd))) 1195 break; 1196 1197 ret = sk_reuseport_attach_bpf(ufd, sk); 1198 } 1199 break; 1200 1201 case SO_DETACH_REUSEPORT_BPF: 1202 ret = reuseport_detach_prog(sk); 1203 break; 1204 1205 case SO_DETACH_FILTER: 1206 ret = sk_detach_filter(sk); 1207 break; 1208 1209 case SO_LOCK_FILTER: 1210 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool) 1211 ret = -EPERM; 1212 else 1213 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool); 1214 break; 1215 1216 case SO_PASSSEC: 1217 if (valbool) 1218 set_bit(SOCK_PASSSEC, &sock->flags); 1219 else 1220 clear_bit(SOCK_PASSSEC, &sock->flags); 1221 break; 1222 case SO_MARK: 1223 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) { 1224 ret = -EPERM; 1225 break; 1226 } 1227 1228 __sock_set_mark(sk, val); 1229 break; 1230 1231 case SO_RXQ_OVFL: 1232 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool); 1233 break; 1234 1235 case SO_WIFI_STATUS: 1236 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool); 1237 break; 1238 1239 case SO_PEEK_OFF: 1240 if (sock->ops->set_peek_off) 1241 ret = sock->ops->set_peek_off(sk, val); 1242 else 1243 ret = -EOPNOTSUPP; 1244 break; 1245 1246 case SO_NOFCS: 1247 sock_valbool_flag(sk, SOCK_NOFCS, valbool); 1248 break; 1249 1250 case SO_SELECT_ERR_QUEUE: 1251 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool); 1252 break; 1253 1254 #ifdef CONFIG_NET_RX_BUSY_POLL 1255 case SO_BUSY_POLL: 1256 /* allow unprivileged users to decrease the value */ 1257 if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN)) 1258 ret = -EPERM; 1259 else { 1260 if (val < 0) 1261 ret = -EINVAL; 1262 else 1263 WRITE_ONCE(sk->sk_ll_usec, val); 1264 } 1265 break; 1266 case SO_PREFER_BUSY_POLL: 1267 if (valbool && !capable(CAP_NET_ADMIN)) 1268 ret = -EPERM; 1269 else 1270 WRITE_ONCE(sk->sk_prefer_busy_poll, valbool); 1271 break; 1272 case SO_BUSY_POLL_BUDGET: 1273 if (val > READ_ONCE(sk->sk_busy_poll_budget) && !capable(CAP_NET_ADMIN)) { 1274 ret = -EPERM; 1275 } else { 1276 if (val < 0 || val > U16_MAX) 1277 ret = -EINVAL; 1278 else 1279 WRITE_ONCE(sk->sk_busy_poll_budget, val); 1280 } 1281 break; 1282 #endif 1283 1284 case SO_MAX_PACING_RATE: 1285 { 1286 unsigned long ulval = (val == ~0U) ? ~0UL : (unsigned int)val; 1287 1288 if (sizeof(ulval) != sizeof(val) && 1289 optlen >= sizeof(ulval) && 1290 copy_from_sockptr(&ulval, optval, sizeof(ulval))) { 1291 ret = -EFAULT; 1292 break; 1293 } 1294 if (ulval != ~0UL) 1295 cmpxchg(&sk->sk_pacing_status, 1296 SK_PACING_NONE, 1297 SK_PACING_NEEDED); 1298 sk->sk_max_pacing_rate = ulval; 1299 sk->sk_pacing_rate = min(sk->sk_pacing_rate, ulval); 1300 break; 1301 } 1302 case SO_INCOMING_CPU: 1303 WRITE_ONCE(sk->sk_incoming_cpu, val); 1304 break; 1305 1306 case SO_CNX_ADVICE: 1307 if (val == 1) 1308 dst_negative_advice(sk); 1309 break; 1310 1311 case SO_ZEROCOPY: 1312 if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) { 1313 if (!((sk->sk_type == SOCK_STREAM && 1314 sk->sk_protocol == IPPROTO_TCP) || 1315 (sk->sk_type == SOCK_DGRAM && 1316 sk->sk_protocol == IPPROTO_UDP))) 1317 ret = -ENOTSUPP; 1318 } else if (sk->sk_family != PF_RDS) { 1319 ret = -ENOTSUPP; 1320 } 1321 if (!ret) { 1322 if (val < 0 || val > 1) 1323 ret = -EINVAL; 1324 else 1325 sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool); 1326 } 1327 break; 1328 1329 case SO_TXTIME: 1330 if (optlen != sizeof(struct sock_txtime)) { 1331 ret = -EINVAL; 1332 break; 1333 } else if (copy_from_sockptr(&sk_txtime, optval, 1334 sizeof(struct sock_txtime))) { 1335 ret = -EFAULT; 1336 break; 1337 } else if (sk_txtime.flags & ~SOF_TXTIME_FLAGS_MASK) { 1338 ret = -EINVAL; 1339 break; 1340 } 1341 /* CLOCK_MONOTONIC is only used by sch_fq, and this packet 1342 * scheduler has enough safe guards. 1343 */ 1344 if (sk_txtime.clockid != CLOCK_MONOTONIC && 1345 !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) { 1346 ret = -EPERM; 1347 break; 1348 } 1349 sock_valbool_flag(sk, SOCK_TXTIME, true); 1350 sk->sk_clockid = sk_txtime.clockid; 1351 sk->sk_txtime_deadline_mode = 1352 !!(sk_txtime.flags & SOF_TXTIME_DEADLINE_MODE); 1353 sk->sk_txtime_report_errors = 1354 !!(sk_txtime.flags & SOF_TXTIME_REPORT_ERRORS); 1355 break; 1356 1357 case SO_BINDTOIFINDEX: 1358 ret = sock_bindtoindex_locked(sk, val); 1359 break; 1360 1361 case SO_BUF_LOCK: 1362 if (val & ~SOCK_BUF_LOCK_MASK) { 1363 ret = -EINVAL; 1364 break; 1365 } 1366 sk->sk_userlocks = val | (sk->sk_userlocks & 1367 ~SOCK_BUF_LOCK_MASK); 1368 break; 1369 1370 default: 1371 ret = -ENOPROTOOPT; 1372 break; 1373 } 1374 release_sock(sk); 1375 return ret; 1376 } 1377 EXPORT_SYMBOL(sock_setsockopt); 1378 1379 static const struct cred *sk_get_peer_cred(struct sock *sk) 1380 { 1381 const struct cred *cred; 1382 1383 spin_lock(&sk->sk_peer_lock); 1384 cred = get_cred(sk->sk_peer_cred); 1385 spin_unlock(&sk->sk_peer_lock); 1386 1387 return cred; 1388 } 1389 1390 static void cred_to_ucred(struct pid *pid, const struct cred *cred, 1391 struct ucred *ucred) 1392 { 1393 ucred->pid = pid_vnr(pid); 1394 ucred->uid = ucred->gid = -1; 1395 if (cred) { 1396 struct user_namespace *current_ns = current_user_ns(); 1397 1398 ucred->uid = from_kuid_munged(current_ns, cred->euid); 1399 ucred->gid = from_kgid_munged(current_ns, cred->egid); 1400 } 1401 } 1402 1403 static int groups_to_user(gid_t __user *dst, const struct group_info *src) 1404 { 1405 struct user_namespace *user_ns = current_user_ns(); 1406 int i; 1407 1408 for (i = 0; i < src->ngroups; i++) 1409 if (put_user(from_kgid_munged(user_ns, src->gid[i]), dst + i)) 1410 return -EFAULT; 1411 1412 return 0; 1413 } 1414 1415 int sock_getsockopt(struct socket *sock, int level, int optname, 1416 char __user *optval, int __user *optlen) 1417 { 1418 struct sock *sk = sock->sk; 1419 1420 union { 1421 int val; 1422 u64 val64; 1423 unsigned long ulval; 1424 struct linger ling; 1425 struct old_timeval32 tm32; 1426 struct __kernel_old_timeval tm; 1427 struct __kernel_sock_timeval stm; 1428 struct sock_txtime txtime; 1429 struct so_timestamping timestamping; 1430 } v; 1431 1432 int lv = sizeof(int); 1433 int len; 1434 1435 if (get_user(len, optlen)) 1436 return -EFAULT; 1437 if (len < 0) 1438 return -EINVAL; 1439 1440 memset(&v, 0, sizeof(v)); 1441 1442 switch (optname) { 1443 case SO_DEBUG: 1444 v.val = sock_flag(sk, SOCK_DBG); 1445 break; 1446 1447 case SO_DONTROUTE: 1448 v.val = sock_flag(sk, SOCK_LOCALROUTE); 1449 break; 1450 1451 case SO_BROADCAST: 1452 v.val = sock_flag(sk, SOCK_BROADCAST); 1453 break; 1454 1455 case SO_SNDBUF: 1456 v.val = sk->sk_sndbuf; 1457 break; 1458 1459 case SO_RCVBUF: 1460 v.val = sk->sk_rcvbuf; 1461 break; 1462 1463 case SO_REUSEADDR: 1464 v.val = sk->sk_reuse; 1465 break; 1466 1467 case SO_REUSEPORT: 1468 v.val = sk->sk_reuseport; 1469 break; 1470 1471 case SO_KEEPALIVE: 1472 v.val = sock_flag(sk, SOCK_KEEPOPEN); 1473 break; 1474 1475 case SO_TYPE: 1476 v.val = sk->sk_type; 1477 break; 1478 1479 case SO_PROTOCOL: 1480 v.val = sk->sk_protocol; 1481 break; 1482 1483 case SO_DOMAIN: 1484 v.val = sk->sk_family; 1485 break; 1486 1487 case SO_ERROR: 1488 v.val = -sock_error(sk); 1489 if (v.val == 0) 1490 v.val = xchg(&sk->sk_err_soft, 0); 1491 break; 1492 1493 case SO_OOBINLINE: 1494 v.val = sock_flag(sk, SOCK_URGINLINE); 1495 break; 1496 1497 case SO_NO_CHECK: 1498 v.val = sk->sk_no_check_tx; 1499 break; 1500 1501 case SO_PRIORITY: 1502 v.val = sk->sk_priority; 1503 break; 1504 1505 case SO_LINGER: 1506 lv = sizeof(v.ling); 1507 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER); 1508 v.ling.l_linger = sk->sk_lingertime / HZ; 1509 break; 1510 1511 case SO_BSDCOMPAT: 1512 break; 1513 1514 case SO_TIMESTAMP_OLD: 1515 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && 1516 !sock_flag(sk, SOCK_TSTAMP_NEW) && 1517 !sock_flag(sk, SOCK_RCVTSTAMPNS); 1518 break; 1519 1520 case SO_TIMESTAMPNS_OLD: 1521 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && !sock_flag(sk, SOCK_TSTAMP_NEW); 1522 break; 1523 1524 case SO_TIMESTAMP_NEW: 1525 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && sock_flag(sk, SOCK_TSTAMP_NEW); 1526 break; 1527 1528 case SO_TIMESTAMPNS_NEW: 1529 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && sock_flag(sk, SOCK_TSTAMP_NEW); 1530 break; 1531 1532 case SO_TIMESTAMPING_OLD: 1533 lv = sizeof(v.timestamping); 1534 v.timestamping.flags = sk->sk_tsflags; 1535 v.timestamping.bind_phc = sk->sk_bind_phc; 1536 break; 1537 1538 case SO_RCVTIMEO_OLD: 1539 case SO_RCVTIMEO_NEW: 1540 lv = sock_get_timeout(sk->sk_rcvtimeo, &v, SO_RCVTIMEO_OLD == optname); 1541 break; 1542 1543 case SO_SNDTIMEO_OLD: 1544 case SO_SNDTIMEO_NEW: 1545 lv = sock_get_timeout(sk->sk_sndtimeo, &v, SO_SNDTIMEO_OLD == optname); 1546 break; 1547 1548 case SO_RCVLOWAT: 1549 v.val = sk->sk_rcvlowat; 1550 break; 1551 1552 case SO_SNDLOWAT: 1553 v.val = 1; 1554 break; 1555 1556 case SO_PASSCRED: 1557 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags); 1558 break; 1559 1560 case SO_PEERCRED: 1561 { 1562 struct ucred peercred; 1563 if (len > sizeof(peercred)) 1564 len = sizeof(peercred); 1565 1566 spin_lock(&sk->sk_peer_lock); 1567 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred); 1568 spin_unlock(&sk->sk_peer_lock); 1569 1570 if (copy_to_user(optval, &peercred, len)) 1571 return -EFAULT; 1572 goto lenout; 1573 } 1574 1575 case SO_PEERGROUPS: 1576 { 1577 const struct cred *cred; 1578 int ret, n; 1579 1580 cred = sk_get_peer_cred(sk); 1581 if (!cred) 1582 return -ENODATA; 1583 1584 n = cred->group_info->ngroups; 1585 if (len < n * sizeof(gid_t)) { 1586 len = n * sizeof(gid_t); 1587 put_cred(cred); 1588 return put_user(len, optlen) ? -EFAULT : -ERANGE; 1589 } 1590 len = n * sizeof(gid_t); 1591 1592 ret = groups_to_user((gid_t __user *)optval, cred->group_info); 1593 put_cred(cred); 1594 if (ret) 1595 return ret; 1596 goto lenout; 1597 } 1598 1599 case SO_PEERNAME: 1600 { 1601 char address[128]; 1602 1603 lv = sock->ops->getname(sock, (struct sockaddr *)address, 2); 1604 if (lv < 0) 1605 return -ENOTCONN; 1606 if (lv < len) 1607 return -EINVAL; 1608 if (copy_to_user(optval, address, len)) 1609 return -EFAULT; 1610 goto lenout; 1611 } 1612 1613 /* Dubious BSD thing... Probably nobody even uses it, but 1614 * the UNIX standard wants it for whatever reason... -DaveM 1615 */ 1616 case SO_ACCEPTCONN: 1617 v.val = sk->sk_state == TCP_LISTEN; 1618 break; 1619 1620 case SO_PASSSEC: 1621 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags); 1622 break; 1623 1624 case SO_PEERSEC: 1625 return security_socket_getpeersec_stream(sock, optval, optlen, len); 1626 1627 case SO_MARK: 1628 v.val = sk->sk_mark; 1629 break; 1630 1631 case SO_RXQ_OVFL: 1632 v.val = sock_flag(sk, SOCK_RXQ_OVFL); 1633 break; 1634 1635 case SO_WIFI_STATUS: 1636 v.val = sock_flag(sk, SOCK_WIFI_STATUS); 1637 break; 1638 1639 case SO_PEEK_OFF: 1640 if (!sock->ops->set_peek_off) 1641 return -EOPNOTSUPP; 1642 1643 v.val = sk->sk_peek_off; 1644 break; 1645 case SO_NOFCS: 1646 v.val = sock_flag(sk, SOCK_NOFCS); 1647 break; 1648 1649 case SO_BINDTODEVICE: 1650 return sock_getbindtodevice(sk, optval, optlen, len); 1651 1652 case SO_GET_FILTER: 1653 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len); 1654 if (len < 0) 1655 return len; 1656 1657 goto lenout; 1658 1659 case SO_LOCK_FILTER: 1660 v.val = sock_flag(sk, SOCK_FILTER_LOCKED); 1661 break; 1662 1663 case SO_BPF_EXTENSIONS: 1664 v.val = bpf_tell_extensions(); 1665 break; 1666 1667 case SO_SELECT_ERR_QUEUE: 1668 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE); 1669 break; 1670 1671 #ifdef CONFIG_NET_RX_BUSY_POLL 1672 case SO_BUSY_POLL: 1673 v.val = sk->sk_ll_usec; 1674 break; 1675 case SO_PREFER_BUSY_POLL: 1676 v.val = READ_ONCE(sk->sk_prefer_busy_poll); 1677 break; 1678 #endif 1679 1680 case SO_MAX_PACING_RATE: 1681 if (sizeof(v.ulval) != sizeof(v.val) && len >= sizeof(v.ulval)) { 1682 lv = sizeof(v.ulval); 1683 v.ulval = sk->sk_max_pacing_rate; 1684 } else { 1685 /* 32bit version */ 1686 v.val = min_t(unsigned long, sk->sk_max_pacing_rate, ~0U); 1687 } 1688 break; 1689 1690 case SO_INCOMING_CPU: 1691 v.val = READ_ONCE(sk->sk_incoming_cpu); 1692 break; 1693 1694 case SO_MEMINFO: 1695 { 1696 u32 meminfo[SK_MEMINFO_VARS]; 1697 1698 sk_get_meminfo(sk, meminfo); 1699 1700 len = min_t(unsigned int, len, sizeof(meminfo)); 1701 if (copy_to_user(optval, &meminfo, len)) 1702 return -EFAULT; 1703 1704 goto lenout; 1705 } 1706 1707 #ifdef CONFIG_NET_RX_BUSY_POLL 1708 case SO_INCOMING_NAPI_ID: 1709 v.val = READ_ONCE(sk->sk_napi_id); 1710 1711 /* aggregate non-NAPI IDs down to 0 */ 1712 if (v.val < MIN_NAPI_ID) 1713 v.val = 0; 1714 1715 break; 1716 #endif 1717 1718 case SO_COOKIE: 1719 lv = sizeof(u64); 1720 if (len < lv) 1721 return -EINVAL; 1722 v.val64 = sock_gen_cookie(sk); 1723 break; 1724 1725 case SO_ZEROCOPY: 1726 v.val = sock_flag(sk, SOCK_ZEROCOPY); 1727 break; 1728 1729 case SO_TXTIME: 1730 lv = sizeof(v.txtime); 1731 v.txtime.clockid = sk->sk_clockid; 1732 v.txtime.flags |= sk->sk_txtime_deadline_mode ? 1733 SOF_TXTIME_DEADLINE_MODE : 0; 1734 v.txtime.flags |= sk->sk_txtime_report_errors ? 1735 SOF_TXTIME_REPORT_ERRORS : 0; 1736 break; 1737 1738 case SO_BINDTOIFINDEX: 1739 v.val = sk->sk_bound_dev_if; 1740 break; 1741 1742 case SO_NETNS_COOKIE: 1743 lv = sizeof(u64); 1744 if (len != lv) 1745 return -EINVAL; 1746 v.val64 = sock_net(sk)->net_cookie; 1747 break; 1748 1749 case SO_BUF_LOCK: 1750 v.val = sk->sk_userlocks & SOCK_BUF_LOCK_MASK; 1751 break; 1752 1753 default: 1754 /* We implement the SO_SNDLOWAT etc to not be settable 1755 * (1003.1g 7). 1756 */ 1757 return -ENOPROTOOPT; 1758 } 1759 1760 if (len > lv) 1761 len = lv; 1762 if (copy_to_user(optval, &v, len)) 1763 return -EFAULT; 1764 lenout: 1765 if (put_user(len, optlen)) 1766 return -EFAULT; 1767 return 0; 1768 } 1769 1770 /* 1771 * Initialize an sk_lock. 1772 * 1773 * (We also register the sk_lock with the lock validator.) 1774 */ 1775 static inline void sock_lock_init(struct sock *sk) 1776 { 1777 if (sk->sk_kern_sock) 1778 sock_lock_init_class_and_name( 1779 sk, 1780 af_family_kern_slock_key_strings[sk->sk_family], 1781 af_family_kern_slock_keys + sk->sk_family, 1782 af_family_kern_key_strings[sk->sk_family], 1783 af_family_kern_keys + sk->sk_family); 1784 else 1785 sock_lock_init_class_and_name( 1786 sk, 1787 af_family_slock_key_strings[sk->sk_family], 1788 af_family_slock_keys + sk->sk_family, 1789 af_family_key_strings[sk->sk_family], 1790 af_family_keys + sk->sk_family); 1791 } 1792 1793 /* 1794 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet, 1795 * even temporarly, because of RCU lookups. sk_node should also be left as is. 1796 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end 1797 */ 1798 static void sock_copy(struct sock *nsk, const struct sock *osk) 1799 { 1800 const struct proto *prot = READ_ONCE(osk->sk_prot); 1801 #ifdef CONFIG_SECURITY_NETWORK 1802 void *sptr = nsk->sk_security; 1803 #endif 1804 1805 /* If we move sk_tx_queue_mapping out of the private section, 1806 * we must check if sk_tx_queue_clear() is called after 1807 * sock_copy() in sk_clone_lock(). 1808 */ 1809 BUILD_BUG_ON(offsetof(struct sock, sk_tx_queue_mapping) < 1810 offsetof(struct sock, sk_dontcopy_begin) || 1811 offsetof(struct sock, sk_tx_queue_mapping) >= 1812 offsetof(struct sock, sk_dontcopy_end)); 1813 1814 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin)); 1815 1816 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end, 1817 prot->obj_size - offsetof(struct sock, sk_dontcopy_end)); 1818 1819 #ifdef CONFIG_SECURITY_NETWORK 1820 nsk->sk_security = sptr; 1821 security_sk_clone(osk, nsk); 1822 #endif 1823 } 1824 1825 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority, 1826 int family) 1827 { 1828 struct sock *sk; 1829 struct kmem_cache *slab; 1830 1831 slab = prot->slab; 1832 if (slab != NULL) { 1833 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO); 1834 if (!sk) 1835 return sk; 1836 if (want_init_on_alloc(priority)) 1837 sk_prot_clear_nulls(sk, prot->obj_size); 1838 } else 1839 sk = kmalloc(prot->obj_size, priority); 1840 1841 if (sk != NULL) { 1842 if (security_sk_alloc(sk, family, priority)) 1843 goto out_free; 1844 1845 if (!try_module_get(prot->owner)) 1846 goto out_free_sec; 1847 } 1848 1849 return sk; 1850 1851 out_free_sec: 1852 security_sk_free(sk); 1853 out_free: 1854 if (slab != NULL) 1855 kmem_cache_free(slab, sk); 1856 else 1857 kfree(sk); 1858 return NULL; 1859 } 1860 1861 static void sk_prot_free(struct proto *prot, struct sock *sk) 1862 { 1863 struct kmem_cache *slab; 1864 struct module *owner; 1865 1866 owner = prot->owner; 1867 slab = prot->slab; 1868 1869 cgroup_sk_free(&sk->sk_cgrp_data); 1870 mem_cgroup_sk_free(sk); 1871 security_sk_free(sk); 1872 if (slab != NULL) 1873 kmem_cache_free(slab, sk); 1874 else 1875 kfree(sk); 1876 module_put(owner); 1877 } 1878 1879 /** 1880 * sk_alloc - All socket objects are allocated here 1881 * @net: the applicable net namespace 1882 * @family: protocol family 1883 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 1884 * @prot: struct proto associated with this new sock instance 1885 * @kern: is this to be a kernel socket? 1886 */ 1887 struct sock *sk_alloc(struct net *net, int family, gfp_t priority, 1888 struct proto *prot, int kern) 1889 { 1890 struct sock *sk; 1891 1892 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family); 1893 if (sk) { 1894 sk->sk_family = family; 1895 /* 1896 * See comment in struct sock definition to understand 1897 * why we need sk_prot_creator -acme 1898 */ 1899 sk->sk_prot = sk->sk_prot_creator = prot; 1900 sk->sk_kern_sock = kern; 1901 sock_lock_init(sk); 1902 sk->sk_net_refcnt = kern ? 0 : 1; 1903 if (likely(sk->sk_net_refcnt)) { 1904 get_net(net); 1905 sock_inuse_add(net, 1); 1906 } 1907 1908 sock_net_set(sk, net); 1909 refcount_set(&sk->sk_wmem_alloc, 1); 1910 1911 mem_cgroup_sk_alloc(sk); 1912 cgroup_sk_alloc(&sk->sk_cgrp_data); 1913 sock_update_classid(&sk->sk_cgrp_data); 1914 sock_update_netprioidx(&sk->sk_cgrp_data); 1915 sk_tx_queue_clear(sk); 1916 } 1917 1918 return sk; 1919 } 1920 EXPORT_SYMBOL(sk_alloc); 1921 1922 /* Sockets having SOCK_RCU_FREE will call this function after one RCU 1923 * grace period. This is the case for UDP sockets and TCP listeners. 1924 */ 1925 static void __sk_destruct(struct rcu_head *head) 1926 { 1927 struct sock *sk = container_of(head, struct sock, sk_rcu); 1928 struct sk_filter *filter; 1929 1930 if (sk->sk_destruct) 1931 sk->sk_destruct(sk); 1932 1933 filter = rcu_dereference_check(sk->sk_filter, 1934 refcount_read(&sk->sk_wmem_alloc) == 0); 1935 if (filter) { 1936 sk_filter_uncharge(sk, filter); 1937 RCU_INIT_POINTER(sk->sk_filter, NULL); 1938 } 1939 1940 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP); 1941 1942 #ifdef CONFIG_BPF_SYSCALL 1943 bpf_sk_storage_free(sk); 1944 #endif 1945 1946 if (atomic_read(&sk->sk_omem_alloc)) 1947 pr_debug("%s: optmem leakage (%d bytes) detected\n", 1948 __func__, atomic_read(&sk->sk_omem_alloc)); 1949 1950 if (sk->sk_frag.page) { 1951 put_page(sk->sk_frag.page); 1952 sk->sk_frag.page = NULL; 1953 } 1954 1955 /* We do not need to acquire sk->sk_peer_lock, we are the last user. */ 1956 put_cred(sk->sk_peer_cred); 1957 put_pid(sk->sk_peer_pid); 1958 1959 if (likely(sk->sk_net_refcnt)) 1960 put_net(sock_net(sk)); 1961 sk_prot_free(sk->sk_prot_creator, sk); 1962 } 1963 1964 void sk_destruct(struct sock *sk) 1965 { 1966 bool use_call_rcu = sock_flag(sk, SOCK_RCU_FREE); 1967 1968 if (rcu_access_pointer(sk->sk_reuseport_cb)) { 1969 reuseport_detach_sock(sk); 1970 use_call_rcu = true; 1971 } 1972 1973 if (use_call_rcu) 1974 call_rcu(&sk->sk_rcu, __sk_destruct); 1975 else 1976 __sk_destruct(&sk->sk_rcu); 1977 } 1978 1979 static void __sk_free(struct sock *sk) 1980 { 1981 if (likely(sk->sk_net_refcnt)) 1982 sock_inuse_add(sock_net(sk), -1); 1983 1984 if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk))) 1985 sock_diag_broadcast_destroy(sk); 1986 else 1987 sk_destruct(sk); 1988 } 1989 1990 void sk_free(struct sock *sk) 1991 { 1992 /* 1993 * We subtract one from sk_wmem_alloc and can know if 1994 * some packets are still in some tx queue. 1995 * If not null, sock_wfree() will call __sk_free(sk) later 1996 */ 1997 if (refcount_dec_and_test(&sk->sk_wmem_alloc)) 1998 __sk_free(sk); 1999 } 2000 EXPORT_SYMBOL(sk_free); 2001 2002 static void sk_init_common(struct sock *sk) 2003 { 2004 skb_queue_head_init(&sk->sk_receive_queue); 2005 skb_queue_head_init(&sk->sk_write_queue); 2006 skb_queue_head_init(&sk->sk_error_queue); 2007 2008 rwlock_init(&sk->sk_callback_lock); 2009 lockdep_set_class_and_name(&sk->sk_receive_queue.lock, 2010 af_rlock_keys + sk->sk_family, 2011 af_family_rlock_key_strings[sk->sk_family]); 2012 lockdep_set_class_and_name(&sk->sk_write_queue.lock, 2013 af_wlock_keys + sk->sk_family, 2014 af_family_wlock_key_strings[sk->sk_family]); 2015 lockdep_set_class_and_name(&sk->sk_error_queue.lock, 2016 af_elock_keys + sk->sk_family, 2017 af_family_elock_key_strings[sk->sk_family]); 2018 lockdep_set_class_and_name(&sk->sk_callback_lock, 2019 af_callback_keys + sk->sk_family, 2020 af_family_clock_key_strings[sk->sk_family]); 2021 } 2022 2023 /** 2024 * sk_clone_lock - clone a socket, and lock its clone 2025 * @sk: the socket to clone 2026 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 2027 * 2028 * Caller must unlock socket even in error path (bh_unlock_sock(newsk)) 2029 */ 2030 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority) 2031 { 2032 struct proto *prot = READ_ONCE(sk->sk_prot); 2033 struct sk_filter *filter; 2034 bool is_charged = true; 2035 struct sock *newsk; 2036 2037 newsk = sk_prot_alloc(prot, priority, sk->sk_family); 2038 if (!newsk) 2039 goto out; 2040 2041 sock_copy(newsk, sk); 2042 2043 newsk->sk_prot_creator = prot; 2044 2045 /* SANITY */ 2046 if (likely(newsk->sk_net_refcnt)) 2047 get_net(sock_net(newsk)); 2048 sk_node_init(&newsk->sk_node); 2049 sock_lock_init(newsk); 2050 bh_lock_sock(newsk); 2051 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL; 2052 newsk->sk_backlog.len = 0; 2053 2054 atomic_set(&newsk->sk_rmem_alloc, 0); 2055 2056 /* sk_wmem_alloc set to one (see sk_free() and sock_wfree()) */ 2057 refcount_set(&newsk->sk_wmem_alloc, 1); 2058 2059 atomic_set(&newsk->sk_omem_alloc, 0); 2060 sk_init_common(newsk); 2061 2062 newsk->sk_dst_cache = NULL; 2063 newsk->sk_dst_pending_confirm = 0; 2064 newsk->sk_wmem_queued = 0; 2065 newsk->sk_forward_alloc = 0; 2066 atomic_set(&newsk->sk_drops, 0); 2067 newsk->sk_send_head = NULL; 2068 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK; 2069 atomic_set(&newsk->sk_zckey, 0); 2070 2071 sock_reset_flag(newsk, SOCK_DONE); 2072 2073 /* sk->sk_memcg will be populated at accept() time */ 2074 newsk->sk_memcg = NULL; 2075 2076 cgroup_sk_clone(&newsk->sk_cgrp_data); 2077 2078 rcu_read_lock(); 2079 filter = rcu_dereference(sk->sk_filter); 2080 if (filter != NULL) 2081 /* though it's an empty new sock, the charging may fail 2082 * if sysctl_optmem_max was changed between creation of 2083 * original socket and cloning 2084 */ 2085 is_charged = sk_filter_charge(newsk, filter); 2086 RCU_INIT_POINTER(newsk->sk_filter, filter); 2087 rcu_read_unlock(); 2088 2089 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) { 2090 /* We need to make sure that we don't uncharge the new 2091 * socket if we couldn't charge it in the first place 2092 * as otherwise we uncharge the parent's filter. 2093 */ 2094 if (!is_charged) 2095 RCU_INIT_POINTER(newsk->sk_filter, NULL); 2096 sk_free_unlock_clone(newsk); 2097 newsk = NULL; 2098 goto out; 2099 } 2100 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL); 2101 2102 if (bpf_sk_storage_clone(sk, newsk)) { 2103 sk_free_unlock_clone(newsk); 2104 newsk = NULL; 2105 goto out; 2106 } 2107 2108 /* Clear sk_user_data if parent had the pointer tagged 2109 * as not suitable for copying when cloning. 2110 */ 2111 if (sk_user_data_is_nocopy(newsk)) 2112 newsk->sk_user_data = NULL; 2113 2114 newsk->sk_err = 0; 2115 newsk->sk_err_soft = 0; 2116 newsk->sk_priority = 0; 2117 newsk->sk_incoming_cpu = raw_smp_processor_id(); 2118 if (likely(newsk->sk_net_refcnt)) 2119 sock_inuse_add(sock_net(newsk), 1); 2120 2121 /* Before updating sk_refcnt, we must commit prior changes to memory 2122 * (Documentation/RCU/rculist_nulls.rst for details) 2123 */ 2124 smp_wmb(); 2125 refcount_set(&newsk->sk_refcnt, 2); 2126 2127 /* Increment the counter in the same struct proto as the master 2128 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that 2129 * is the same as sk->sk_prot->socks, as this field was copied 2130 * with memcpy). 2131 * 2132 * This _changes_ the previous behaviour, where 2133 * tcp_create_openreq_child always was incrementing the 2134 * equivalent to tcp_prot->socks (inet_sock_nr), so this have 2135 * to be taken into account in all callers. -acme 2136 */ 2137 sk_refcnt_debug_inc(newsk); 2138 sk_set_socket(newsk, NULL); 2139 sk_tx_queue_clear(newsk); 2140 RCU_INIT_POINTER(newsk->sk_wq, NULL); 2141 2142 if (newsk->sk_prot->sockets_allocated) 2143 sk_sockets_allocated_inc(newsk); 2144 2145 if (sock_needs_netstamp(sk) && newsk->sk_flags & SK_FLAGS_TIMESTAMP) 2146 net_enable_timestamp(); 2147 out: 2148 return newsk; 2149 } 2150 EXPORT_SYMBOL_GPL(sk_clone_lock); 2151 2152 void sk_free_unlock_clone(struct sock *sk) 2153 { 2154 /* It is still raw copy of parent, so invalidate 2155 * destructor and make plain sk_free() */ 2156 sk->sk_destruct = NULL; 2157 bh_unlock_sock(sk); 2158 sk_free(sk); 2159 } 2160 EXPORT_SYMBOL_GPL(sk_free_unlock_clone); 2161 2162 void sk_setup_caps(struct sock *sk, struct dst_entry *dst) 2163 { 2164 u32 max_segs = 1; 2165 2166 sk_dst_set(sk, dst); 2167 sk->sk_route_caps = dst->dev->features | sk->sk_route_forced_caps; 2168 if (sk->sk_route_caps & NETIF_F_GSO) 2169 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE; 2170 sk->sk_route_caps &= ~sk->sk_route_nocaps; 2171 if (sk_can_gso(sk)) { 2172 if (dst->header_len && !xfrm_dst_offload_ok(dst)) { 2173 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 2174 } else { 2175 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM; 2176 sk->sk_gso_max_size = dst->dev->gso_max_size; 2177 max_segs = max_t(u32, dst->dev->gso_max_segs, 1); 2178 } 2179 } 2180 sk->sk_gso_max_segs = max_segs; 2181 } 2182 EXPORT_SYMBOL_GPL(sk_setup_caps); 2183 2184 /* 2185 * Simple resource managers for sockets. 2186 */ 2187 2188 2189 /* 2190 * Write buffer destructor automatically called from kfree_skb. 2191 */ 2192 void sock_wfree(struct sk_buff *skb) 2193 { 2194 struct sock *sk = skb->sk; 2195 unsigned int len = skb->truesize; 2196 2197 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) { 2198 /* 2199 * Keep a reference on sk_wmem_alloc, this will be released 2200 * after sk_write_space() call 2201 */ 2202 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc)); 2203 sk->sk_write_space(sk); 2204 len = 1; 2205 } 2206 /* 2207 * if sk_wmem_alloc reaches 0, we must finish what sk_free() 2208 * could not do because of in-flight packets 2209 */ 2210 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc)) 2211 __sk_free(sk); 2212 } 2213 EXPORT_SYMBOL(sock_wfree); 2214 2215 /* This variant of sock_wfree() is used by TCP, 2216 * since it sets SOCK_USE_WRITE_QUEUE. 2217 */ 2218 void __sock_wfree(struct sk_buff *skb) 2219 { 2220 struct sock *sk = skb->sk; 2221 2222 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc)) 2223 __sk_free(sk); 2224 } 2225 2226 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk) 2227 { 2228 skb_orphan(skb); 2229 skb->sk = sk; 2230 #ifdef CONFIG_INET 2231 if (unlikely(!sk_fullsock(sk))) { 2232 skb->destructor = sock_edemux; 2233 sock_hold(sk); 2234 return; 2235 } 2236 #endif 2237 skb->destructor = sock_wfree; 2238 skb_set_hash_from_sk(skb, sk); 2239 /* 2240 * We used to take a refcount on sk, but following operation 2241 * is enough to guarantee sk_free() wont free this sock until 2242 * all in-flight packets are completed 2243 */ 2244 refcount_add(skb->truesize, &sk->sk_wmem_alloc); 2245 } 2246 EXPORT_SYMBOL(skb_set_owner_w); 2247 2248 static bool can_skb_orphan_partial(const struct sk_buff *skb) 2249 { 2250 #ifdef CONFIG_TLS_DEVICE 2251 /* Drivers depend on in-order delivery for crypto offload, 2252 * partial orphan breaks out-of-order-OK logic. 2253 */ 2254 if (skb->decrypted) 2255 return false; 2256 #endif 2257 return (skb->destructor == sock_wfree || 2258 (IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree)); 2259 } 2260 2261 /* This helper is used by netem, as it can hold packets in its 2262 * delay queue. We want to allow the owner socket to send more 2263 * packets, as if they were already TX completed by a typical driver. 2264 * But we also want to keep skb->sk set because some packet schedulers 2265 * rely on it (sch_fq for example). 2266 */ 2267 void skb_orphan_partial(struct sk_buff *skb) 2268 { 2269 if (skb_is_tcp_pure_ack(skb)) 2270 return; 2271 2272 if (can_skb_orphan_partial(skb) && skb_set_owner_sk_safe(skb, skb->sk)) 2273 return; 2274 2275 skb_orphan(skb); 2276 } 2277 EXPORT_SYMBOL(skb_orphan_partial); 2278 2279 /* 2280 * Read buffer destructor automatically called from kfree_skb. 2281 */ 2282 void sock_rfree(struct sk_buff *skb) 2283 { 2284 struct sock *sk = skb->sk; 2285 unsigned int len = skb->truesize; 2286 2287 atomic_sub(len, &sk->sk_rmem_alloc); 2288 sk_mem_uncharge(sk, len); 2289 } 2290 EXPORT_SYMBOL(sock_rfree); 2291 2292 /* 2293 * Buffer destructor for skbs that are not used directly in read or write 2294 * path, e.g. for error handler skbs. Automatically called from kfree_skb. 2295 */ 2296 void sock_efree(struct sk_buff *skb) 2297 { 2298 sock_put(skb->sk); 2299 } 2300 EXPORT_SYMBOL(sock_efree); 2301 2302 /* Buffer destructor for prefetch/receive path where reference count may 2303 * not be held, e.g. for listen sockets. 2304 */ 2305 #ifdef CONFIG_INET 2306 void sock_pfree(struct sk_buff *skb) 2307 { 2308 if (sk_is_refcounted(skb->sk)) 2309 sock_gen_put(skb->sk); 2310 } 2311 EXPORT_SYMBOL(sock_pfree); 2312 #endif /* CONFIG_INET */ 2313 2314 kuid_t sock_i_uid(struct sock *sk) 2315 { 2316 kuid_t uid; 2317 2318 read_lock_bh(&sk->sk_callback_lock); 2319 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID; 2320 read_unlock_bh(&sk->sk_callback_lock); 2321 return uid; 2322 } 2323 EXPORT_SYMBOL(sock_i_uid); 2324 2325 unsigned long sock_i_ino(struct sock *sk) 2326 { 2327 unsigned long ino; 2328 2329 read_lock_bh(&sk->sk_callback_lock); 2330 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0; 2331 read_unlock_bh(&sk->sk_callback_lock); 2332 return ino; 2333 } 2334 EXPORT_SYMBOL(sock_i_ino); 2335 2336 /* 2337 * Allocate a skb from the socket's send buffer. 2338 */ 2339 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 2340 gfp_t priority) 2341 { 2342 if (force || 2343 refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) { 2344 struct sk_buff *skb = alloc_skb(size, priority); 2345 2346 if (skb) { 2347 skb_set_owner_w(skb, sk); 2348 return skb; 2349 } 2350 } 2351 return NULL; 2352 } 2353 EXPORT_SYMBOL(sock_wmalloc); 2354 2355 static void sock_ofree(struct sk_buff *skb) 2356 { 2357 struct sock *sk = skb->sk; 2358 2359 atomic_sub(skb->truesize, &sk->sk_omem_alloc); 2360 } 2361 2362 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size, 2363 gfp_t priority) 2364 { 2365 struct sk_buff *skb; 2366 2367 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */ 2368 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) > 2369 sysctl_optmem_max) 2370 return NULL; 2371 2372 skb = alloc_skb(size, priority); 2373 if (!skb) 2374 return NULL; 2375 2376 atomic_add(skb->truesize, &sk->sk_omem_alloc); 2377 skb->sk = sk; 2378 skb->destructor = sock_ofree; 2379 return skb; 2380 } 2381 2382 /* 2383 * Allocate a memory block from the socket's option memory buffer. 2384 */ 2385 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority) 2386 { 2387 if ((unsigned int)size <= sysctl_optmem_max && 2388 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) { 2389 void *mem; 2390 /* First do the add, to avoid the race if kmalloc 2391 * might sleep. 2392 */ 2393 atomic_add(size, &sk->sk_omem_alloc); 2394 mem = kmalloc(size, priority); 2395 if (mem) 2396 return mem; 2397 atomic_sub(size, &sk->sk_omem_alloc); 2398 } 2399 return NULL; 2400 } 2401 EXPORT_SYMBOL(sock_kmalloc); 2402 2403 /* Free an option memory block. Note, we actually want the inline 2404 * here as this allows gcc to detect the nullify and fold away the 2405 * condition entirely. 2406 */ 2407 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size, 2408 const bool nullify) 2409 { 2410 if (WARN_ON_ONCE(!mem)) 2411 return; 2412 if (nullify) 2413 kfree_sensitive(mem); 2414 else 2415 kfree(mem); 2416 atomic_sub(size, &sk->sk_omem_alloc); 2417 } 2418 2419 void sock_kfree_s(struct sock *sk, void *mem, int size) 2420 { 2421 __sock_kfree_s(sk, mem, size, false); 2422 } 2423 EXPORT_SYMBOL(sock_kfree_s); 2424 2425 void sock_kzfree_s(struct sock *sk, void *mem, int size) 2426 { 2427 __sock_kfree_s(sk, mem, size, true); 2428 } 2429 EXPORT_SYMBOL(sock_kzfree_s); 2430 2431 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock. 2432 I think, these locks should be removed for datagram sockets. 2433 */ 2434 static long sock_wait_for_wmem(struct sock *sk, long timeo) 2435 { 2436 DEFINE_WAIT(wait); 2437 2438 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk); 2439 for (;;) { 2440 if (!timeo) 2441 break; 2442 if (signal_pending(current)) 2443 break; 2444 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 2445 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 2446 if (refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) 2447 break; 2448 if (sk->sk_shutdown & SEND_SHUTDOWN) 2449 break; 2450 if (sk->sk_err) 2451 break; 2452 timeo = schedule_timeout(timeo); 2453 } 2454 finish_wait(sk_sleep(sk), &wait); 2455 return timeo; 2456 } 2457 2458 2459 /* 2460 * Generic send/receive buffer handlers 2461 */ 2462 2463 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 2464 unsigned long data_len, int noblock, 2465 int *errcode, int max_page_order) 2466 { 2467 struct sk_buff *skb; 2468 long timeo; 2469 int err; 2470 2471 timeo = sock_sndtimeo(sk, noblock); 2472 for (;;) { 2473 err = sock_error(sk); 2474 if (err != 0) 2475 goto failure; 2476 2477 err = -EPIPE; 2478 if (sk->sk_shutdown & SEND_SHUTDOWN) 2479 goto failure; 2480 2481 if (sk_wmem_alloc_get(sk) < READ_ONCE(sk->sk_sndbuf)) 2482 break; 2483 2484 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk); 2485 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 2486 err = -EAGAIN; 2487 if (!timeo) 2488 goto failure; 2489 if (signal_pending(current)) 2490 goto interrupted; 2491 timeo = sock_wait_for_wmem(sk, timeo); 2492 } 2493 skb = alloc_skb_with_frags(header_len, data_len, max_page_order, 2494 errcode, sk->sk_allocation); 2495 if (skb) 2496 skb_set_owner_w(skb, sk); 2497 return skb; 2498 2499 interrupted: 2500 err = sock_intr_errno(timeo); 2501 failure: 2502 *errcode = err; 2503 return NULL; 2504 } 2505 EXPORT_SYMBOL(sock_alloc_send_pskb); 2506 2507 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size, 2508 int noblock, int *errcode) 2509 { 2510 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0); 2511 } 2512 EXPORT_SYMBOL(sock_alloc_send_skb); 2513 2514 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg, 2515 struct sockcm_cookie *sockc) 2516 { 2517 u32 tsflags; 2518 2519 switch (cmsg->cmsg_type) { 2520 case SO_MARK: 2521 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) 2522 return -EPERM; 2523 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 2524 return -EINVAL; 2525 sockc->mark = *(u32 *)CMSG_DATA(cmsg); 2526 break; 2527 case SO_TIMESTAMPING_OLD: 2528 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 2529 return -EINVAL; 2530 2531 tsflags = *(u32 *)CMSG_DATA(cmsg); 2532 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK) 2533 return -EINVAL; 2534 2535 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK; 2536 sockc->tsflags |= tsflags; 2537 break; 2538 case SCM_TXTIME: 2539 if (!sock_flag(sk, SOCK_TXTIME)) 2540 return -EINVAL; 2541 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64))) 2542 return -EINVAL; 2543 sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg)); 2544 break; 2545 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */ 2546 case SCM_RIGHTS: 2547 case SCM_CREDENTIALS: 2548 break; 2549 default: 2550 return -EINVAL; 2551 } 2552 return 0; 2553 } 2554 EXPORT_SYMBOL(__sock_cmsg_send); 2555 2556 int sock_cmsg_send(struct sock *sk, struct msghdr *msg, 2557 struct sockcm_cookie *sockc) 2558 { 2559 struct cmsghdr *cmsg; 2560 int ret; 2561 2562 for_each_cmsghdr(cmsg, msg) { 2563 if (!CMSG_OK(msg, cmsg)) 2564 return -EINVAL; 2565 if (cmsg->cmsg_level != SOL_SOCKET) 2566 continue; 2567 ret = __sock_cmsg_send(sk, msg, cmsg, sockc); 2568 if (ret) 2569 return ret; 2570 } 2571 return 0; 2572 } 2573 EXPORT_SYMBOL(sock_cmsg_send); 2574 2575 static void sk_enter_memory_pressure(struct sock *sk) 2576 { 2577 if (!sk->sk_prot->enter_memory_pressure) 2578 return; 2579 2580 sk->sk_prot->enter_memory_pressure(sk); 2581 } 2582 2583 static void sk_leave_memory_pressure(struct sock *sk) 2584 { 2585 if (sk->sk_prot->leave_memory_pressure) { 2586 sk->sk_prot->leave_memory_pressure(sk); 2587 } else { 2588 unsigned long *memory_pressure = sk->sk_prot->memory_pressure; 2589 2590 if (memory_pressure && READ_ONCE(*memory_pressure)) 2591 WRITE_ONCE(*memory_pressure, 0); 2592 } 2593 } 2594 2595 DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key); 2596 2597 /** 2598 * skb_page_frag_refill - check that a page_frag contains enough room 2599 * @sz: minimum size of the fragment we want to get 2600 * @pfrag: pointer to page_frag 2601 * @gfp: priority for memory allocation 2602 * 2603 * Note: While this allocator tries to use high order pages, there is 2604 * no guarantee that allocations succeed. Therefore, @sz MUST be 2605 * less or equal than PAGE_SIZE. 2606 */ 2607 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp) 2608 { 2609 if (pfrag->page) { 2610 if (page_ref_count(pfrag->page) == 1) { 2611 pfrag->offset = 0; 2612 return true; 2613 } 2614 if (pfrag->offset + sz <= pfrag->size) 2615 return true; 2616 put_page(pfrag->page); 2617 } 2618 2619 pfrag->offset = 0; 2620 if (SKB_FRAG_PAGE_ORDER && 2621 !static_branch_unlikely(&net_high_order_alloc_disable_key)) { 2622 /* Avoid direct reclaim but allow kswapd to wake */ 2623 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) | 2624 __GFP_COMP | __GFP_NOWARN | 2625 __GFP_NORETRY, 2626 SKB_FRAG_PAGE_ORDER); 2627 if (likely(pfrag->page)) { 2628 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER; 2629 return true; 2630 } 2631 } 2632 pfrag->page = alloc_page(gfp); 2633 if (likely(pfrag->page)) { 2634 pfrag->size = PAGE_SIZE; 2635 return true; 2636 } 2637 return false; 2638 } 2639 EXPORT_SYMBOL(skb_page_frag_refill); 2640 2641 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag) 2642 { 2643 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation))) 2644 return true; 2645 2646 sk_enter_memory_pressure(sk); 2647 sk_stream_moderate_sndbuf(sk); 2648 return false; 2649 } 2650 EXPORT_SYMBOL(sk_page_frag_refill); 2651 2652 void __lock_sock(struct sock *sk) 2653 __releases(&sk->sk_lock.slock) 2654 __acquires(&sk->sk_lock.slock) 2655 { 2656 DEFINE_WAIT(wait); 2657 2658 for (;;) { 2659 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait, 2660 TASK_UNINTERRUPTIBLE); 2661 spin_unlock_bh(&sk->sk_lock.slock); 2662 schedule(); 2663 spin_lock_bh(&sk->sk_lock.slock); 2664 if (!sock_owned_by_user(sk)) 2665 break; 2666 } 2667 finish_wait(&sk->sk_lock.wq, &wait); 2668 } 2669 2670 void __release_sock(struct sock *sk) 2671 __releases(&sk->sk_lock.slock) 2672 __acquires(&sk->sk_lock.slock) 2673 { 2674 struct sk_buff *skb, *next; 2675 2676 while ((skb = sk->sk_backlog.head) != NULL) { 2677 sk->sk_backlog.head = sk->sk_backlog.tail = NULL; 2678 2679 spin_unlock_bh(&sk->sk_lock.slock); 2680 2681 do { 2682 next = skb->next; 2683 prefetch(next); 2684 WARN_ON_ONCE(skb_dst_is_noref(skb)); 2685 skb_mark_not_on_list(skb); 2686 sk_backlog_rcv(sk, skb); 2687 2688 cond_resched(); 2689 2690 skb = next; 2691 } while (skb != NULL); 2692 2693 spin_lock_bh(&sk->sk_lock.slock); 2694 } 2695 2696 /* 2697 * Doing the zeroing here guarantee we can not loop forever 2698 * while a wild producer attempts to flood us. 2699 */ 2700 sk->sk_backlog.len = 0; 2701 } 2702 2703 void __sk_flush_backlog(struct sock *sk) 2704 { 2705 spin_lock_bh(&sk->sk_lock.slock); 2706 __release_sock(sk); 2707 spin_unlock_bh(&sk->sk_lock.slock); 2708 } 2709 2710 /** 2711 * sk_wait_data - wait for data to arrive at sk_receive_queue 2712 * @sk: sock to wait on 2713 * @timeo: for how long 2714 * @skb: last skb seen on sk_receive_queue 2715 * 2716 * Now socket state including sk->sk_err is changed only under lock, 2717 * hence we may omit checks after joining wait queue. 2718 * We check receive queue before schedule() only as optimization; 2719 * it is very likely that release_sock() added new data. 2720 */ 2721 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb) 2722 { 2723 DEFINE_WAIT_FUNC(wait, woken_wake_function); 2724 int rc; 2725 2726 add_wait_queue(sk_sleep(sk), &wait); 2727 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk); 2728 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait); 2729 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk); 2730 remove_wait_queue(sk_sleep(sk), &wait); 2731 return rc; 2732 } 2733 EXPORT_SYMBOL(sk_wait_data); 2734 2735 /** 2736 * __sk_mem_raise_allocated - increase memory_allocated 2737 * @sk: socket 2738 * @size: memory size to allocate 2739 * @amt: pages to allocate 2740 * @kind: allocation type 2741 * 2742 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc 2743 */ 2744 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind) 2745 { 2746 struct proto *prot = sk->sk_prot; 2747 long allocated = sk_memory_allocated_add(sk, amt); 2748 bool memcg_charge = mem_cgroup_sockets_enabled && sk->sk_memcg; 2749 bool charged = true; 2750 2751 if (memcg_charge && 2752 !(charged = mem_cgroup_charge_skmem(sk->sk_memcg, amt, 2753 gfp_memcg_charge()))) 2754 goto suppress_allocation; 2755 2756 /* Under limit. */ 2757 if (allocated <= sk_prot_mem_limits(sk, 0)) { 2758 sk_leave_memory_pressure(sk); 2759 return 1; 2760 } 2761 2762 /* Under pressure. */ 2763 if (allocated > sk_prot_mem_limits(sk, 1)) 2764 sk_enter_memory_pressure(sk); 2765 2766 /* Over hard limit. */ 2767 if (allocated > sk_prot_mem_limits(sk, 2)) 2768 goto suppress_allocation; 2769 2770 /* guarantee minimum buffer size under pressure */ 2771 if (kind == SK_MEM_RECV) { 2772 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot)) 2773 return 1; 2774 2775 } else { /* SK_MEM_SEND */ 2776 int wmem0 = sk_get_wmem0(sk, prot); 2777 2778 if (sk->sk_type == SOCK_STREAM) { 2779 if (sk->sk_wmem_queued < wmem0) 2780 return 1; 2781 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) { 2782 return 1; 2783 } 2784 } 2785 2786 if (sk_has_memory_pressure(sk)) { 2787 u64 alloc; 2788 2789 if (!sk_under_memory_pressure(sk)) 2790 return 1; 2791 alloc = sk_sockets_allocated_read_positive(sk); 2792 if (sk_prot_mem_limits(sk, 2) > alloc * 2793 sk_mem_pages(sk->sk_wmem_queued + 2794 atomic_read(&sk->sk_rmem_alloc) + 2795 sk->sk_forward_alloc)) 2796 return 1; 2797 } 2798 2799 suppress_allocation: 2800 2801 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) { 2802 sk_stream_moderate_sndbuf(sk); 2803 2804 /* Fail only if socket is _under_ its sndbuf. 2805 * In this case we cannot block, so that we have to fail. 2806 */ 2807 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf) { 2808 /* Force charge with __GFP_NOFAIL */ 2809 if (memcg_charge && !charged) { 2810 mem_cgroup_charge_skmem(sk->sk_memcg, amt, 2811 gfp_memcg_charge() | __GFP_NOFAIL); 2812 } 2813 return 1; 2814 } 2815 } 2816 2817 if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged)) 2818 trace_sock_exceed_buf_limit(sk, prot, allocated, kind); 2819 2820 sk_memory_allocated_sub(sk, amt); 2821 2822 if (memcg_charge && charged) 2823 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt); 2824 2825 return 0; 2826 } 2827 EXPORT_SYMBOL(__sk_mem_raise_allocated); 2828 2829 /** 2830 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated 2831 * @sk: socket 2832 * @size: memory size to allocate 2833 * @kind: allocation type 2834 * 2835 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means 2836 * rmem allocation. This function assumes that protocols which have 2837 * memory_pressure use sk_wmem_queued as write buffer accounting. 2838 */ 2839 int __sk_mem_schedule(struct sock *sk, int size, int kind) 2840 { 2841 int ret, amt = sk_mem_pages(size); 2842 2843 sk->sk_forward_alloc += amt << SK_MEM_QUANTUM_SHIFT; 2844 ret = __sk_mem_raise_allocated(sk, size, amt, kind); 2845 if (!ret) 2846 sk->sk_forward_alloc -= amt << SK_MEM_QUANTUM_SHIFT; 2847 return ret; 2848 } 2849 EXPORT_SYMBOL(__sk_mem_schedule); 2850 2851 /** 2852 * __sk_mem_reduce_allocated - reclaim memory_allocated 2853 * @sk: socket 2854 * @amount: number of quanta 2855 * 2856 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc 2857 */ 2858 void __sk_mem_reduce_allocated(struct sock *sk, int amount) 2859 { 2860 sk_memory_allocated_sub(sk, amount); 2861 2862 if (mem_cgroup_sockets_enabled && sk->sk_memcg) 2863 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount); 2864 2865 if (sk_under_memory_pressure(sk) && 2866 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0))) 2867 sk_leave_memory_pressure(sk); 2868 } 2869 EXPORT_SYMBOL(__sk_mem_reduce_allocated); 2870 2871 /** 2872 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated 2873 * @sk: socket 2874 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple) 2875 */ 2876 void __sk_mem_reclaim(struct sock *sk, int amount) 2877 { 2878 amount >>= SK_MEM_QUANTUM_SHIFT; 2879 sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT; 2880 __sk_mem_reduce_allocated(sk, amount); 2881 } 2882 EXPORT_SYMBOL(__sk_mem_reclaim); 2883 2884 int sk_set_peek_off(struct sock *sk, int val) 2885 { 2886 sk->sk_peek_off = val; 2887 return 0; 2888 } 2889 EXPORT_SYMBOL_GPL(sk_set_peek_off); 2890 2891 /* 2892 * Set of default routines for initialising struct proto_ops when 2893 * the protocol does not support a particular function. In certain 2894 * cases where it makes no sense for a protocol to have a "do nothing" 2895 * function, some default processing is provided. 2896 */ 2897 2898 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len) 2899 { 2900 return -EOPNOTSUPP; 2901 } 2902 EXPORT_SYMBOL(sock_no_bind); 2903 2904 int sock_no_connect(struct socket *sock, struct sockaddr *saddr, 2905 int len, int flags) 2906 { 2907 return -EOPNOTSUPP; 2908 } 2909 EXPORT_SYMBOL(sock_no_connect); 2910 2911 int sock_no_socketpair(struct socket *sock1, struct socket *sock2) 2912 { 2913 return -EOPNOTSUPP; 2914 } 2915 EXPORT_SYMBOL(sock_no_socketpair); 2916 2917 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags, 2918 bool kern) 2919 { 2920 return -EOPNOTSUPP; 2921 } 2922 EXPORT_SYMBOL(sock_no_accept); 2923 2924 int sock_no_getname(struct socket *sock, struct sockaddr *saddr, 2925 int peer) 2926 { 2927 return -EOPNOTSUPP; 2928 } 2929 EXPORT_SYMBOL(sock_no_getname); 2930 2931 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 2932 { 2933 return -EOPNOTSUPP; 2934 } 2935 EXPORT_SYMBOL(sock_no_ioctl); 2936 2937 int sock_no_listen(struct socket *sock, int backlog) 2938 { 2939 return -EOPNOTSUPP; 2940 } 2941 EXPORT_SYMBOL(sock_no_listen); 2942 2943 int sock_no_shutdown(struct socket *sock, int how) 2944 { 2945 return -EOPNOTSUPP; 2946 } 2947 EXPORT_SYMBOL(sock_no_shutdown); 2948 2949 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len) 2950 { 2951 return -EOPNOTSUPP; 2952 } 2953 EXPORT_SYMBOL(sock_no_sendmsg); 2954 2955 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len) 2956 { 2957 return -EOPNOTSUPP; 2958 } 2959 EXPORT_SYMBOL(sock_no_sendmsg_locked); 2960 2961 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len, 2962 int flags) 2963 { 2964 return -EOPNOTSUPP; 2965 } 2966 EXPORT_SYMBOL(sock_no_recvmsg); 2967 2968 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma) 2969 { 2970 /* Mirror missing mmap method error code */ 2971 return -ENODEV; 2972 } 2973 EXPORT_SYMBOL(sock_no_mmap); 2974 2975 /* 2976 * When a file is received (via SCM_RIGHTS, etc), we must bump the 2977 * various sock-based usage counts. 2978 */ 2979 void __receive_sock(struct file *file) 2980 { 2981 struct socket *sock; 2982 2983 sock = sock_from_file(file); 2984 if (sock) { 2985 sock_update_netprioidx(&sock->sk->sk_cgrp_data); 2986 sock_update_classid(&sock->sk->sk_cgrp_data); 2987 } 2988 } 2989 2990 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags) 2991 { 2992 ssize_t res; 2993 struct msghdr msg = {.msg_flags = flags}; 2994 struct kvec iov; 2995 char *kaddr = kmap(page); 2996 iov.iov_base = kaddr + offset; 2997 iov.iov_len = size; 2998 res = kernel_sendmsg(sock, &msg, &iov, 1, size); 2999 kunmap(page); 3000 return res; 3001 } 3002 EXPORT_SYMBOL(sock_no_sendpage); 3003 3004 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page, 3005 int offset, size_t size, int flags) 3006 { 3007 ssize_t res; 3008 struct msghdr msg = {.msg_flags = flags}; 3009 struct kvec iov; 3010 char *kaddr = kmap(page); 3011 3012 iov.iov_base = kaddr + offset; 3013 iov.iov_len = size; 3014 res = kernel_sendmsg_locked(sk, &msg, &iov, 1, size); 3015 kunmap(page); 3016 return res; 3017 } 3018 EXPORT_SYMBOL(sock_no_sendpage_locked); 3019 3020 /* 3021 * Default Socket Callbacks 3022 */ 3023 3024 static void sock_def_wakeup(struct sock *sk) 3025 { 3026 struct socket_wq *wq; 3027 3028 rcu_read_lock(); 3029 wq = rcu_dereference(sk->sk_wq); 3030 if (skwq_has_sleeper(wq)) 3031 wake_up_interruptible_all(&wq->wait); 3032 rcu_read_unlock(); 3033 } 3034 3035 static void sock_def_error_report(struct sock *sk) 3036 { 3037 struct socket_wq *wq; 3038 3039 rcu_read_lock(); 3040 wq = rcu_dereference(sk->sk_wq); 3041 if (skwq_has_sleeper(wq)) 3042 wake_up_interruptible_poll(&wq->wait, EPOLLERR); 3043 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR); 3044 rcu_read_unlock(); 3045 } 3046 3047 void sock_def_readable(struct sock *sk) 3048 { 3049 struct socket_wq *wq; 3050 3051 rcu_read_lock(); 3052 wq = rcu_dereference(sk->sk_wq); 3053 if (skwq_has_sleeper(wq)) 3054 wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI | 3055 EPOLLRDNORM | EPOLLRDBAND); 3056 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN); 3057 rcu_read_unlock(); 3058 } 3059 3060 static void sock_def_write_space(struct sock *sk) 3061 { 3062 struct socket_wq *wq; 3063 3064 rcu_read_lock(); 3065 3066 /* Do not wake up a writer until he can make "significant" 3067 * progress. --DaveM 3068 */ 3069 if ((refcount_read(&sk->sk_wmem_alloc) << 1) <= READ_ONCE(sk->sk_sndbuf)) { 3070 wq = rcu_dereference(sk->sk_wq); 3071 if (skwq_has_sleeper(wq)) 3072 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT | 3073 EPOLLWRNORM | EPOLLWRBAND); 3074 3075 /* Should agree with poll, otherwise some programs break */ 3076 if (sock_writeable(sk)) 3077 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT); 3078 } 3079 3080 rcu_read_unlock(); 3081 } 3082 3083 static void sock_def_destruct(struct sock *sk) 3084 { 3085 } 3086 3087 void sk_send_sigurg(struct sock *sk) 3088 { 3089 if (sk->sk_socket && sk->sk_socket->file) 3090 if (send_sigurg(&sk->sk_socket->file->f_owner)) 3091 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI); 3092 } 3093 EXPORT_SYMBOL(sk_send_sigurg); 3094 3095 void sk_reset_timer(struct sock *sk, struct timer_list* timer, 3096 unsigned long expires) 3097 { 3098 if (!mod_timer(timer, expires)) 3099 sock_hold(sk); 3100 } 3101 EXPORT_SYMBOL(sk_reset_timer); 3102 3103 void sk_stop_timer(struct sock *sk, struct timer_list* timer) 3104 { 3105 if (del_timer(timer)) 3106 __sock_put(sk); 3107 } 3108 EXPORT_SYMBOL(sk_stop_timer); 3109 3110 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer) 3111 { 3112 if (del_timer_sync(timer)) 3113 __sock_put(sk); 3114 } 3115 EXPORT_SYMBOL(sk_stop_timer_sync); 3116 3117 void sock_init_data(struct socket *sock, struct sock *sk) 3118 { 3119 sk_init_common(sk); 3120 sk->sk_send_head = NULL; 3121 3122 timer_setup(&sk->sk_timer, NULL, 0); 3123 3124 sk->sk_allocation = GFP_KERNEL; 3125 sk->sk_rcvbuf = sysctl_rmem_default; 3126 sk->sk_sndbuf = sysctl_wmem_default; 3127 sk->sk_state = TCP_CLOSE; 3128 sk_set_socket(sk, sock); 3129 3130 sock_set_flag(sk, SOCK_ZAPPED); 3131 3132 if (sock) { 3133 sk->sk_type = sock->type; 3134 RCU_INIT_POINTER(sk->sk_wq, &sock->wq); 3135 sock->sk = sk; 3136 sk->sk_uid = SOCK_INODE(sock)->i_uid; 3137 } else { 3138 RCU_INIT_POINTER(sk->sk_wq, NULL); 3139 sk->sk_uid = make_kuid(sock_net(sk)->user_ns, 0); 3140 } 3141 3142 rwlock_init(&sk->sk_callback_lock); 3143 if (sk->sk_kern_sock) 3144 lockdep_set_class_and_name( 3145 &sk->sk_callback_lock, 3146 af_kern_callback_keys + sk->sk_family, 3147 af_family_kern_clock_key_strings[sk->sk_family]); 3148 else 3149 lockdep_set_class_and_name( 3150 &sk->sk_callback_lock, 3151 af_callback_keys + sk->sk_family, 3152 af_family_clock_key_strings[sk->sk_family]); 3153 3154 sk->sk_state_change = sock_def_wakeup; 3155 sk->sk_data_ready = sock_def_readable; 3156 sk->sk_write_space = sock_def_write_space; 3157 sk->sk_error_report = sock_def_error_report; 3158 sk->sk_destruct = sock_def_destruct; 3159 3160 sk->sk_frag.page = NULL; 3161 sk->sk_frag.offset = 0; 3162 sk->sk_peek_off = -1; 3163 3164 sk->sk_peer_pid = NULL; 3165 sk->sk_peer_cred = NULL; 3166 spin_lock_init(&sk->sk_peer_lock); 3167 3168 sk->sk_write_pending = 0; 3169 sk->sk_rcvlowat = 1; 3170 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; 3171 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT; 3172 3173 sk->sk_stamp = SK_DEFAULT_STAMP; 3174 #if BITS_PER_LONG==32 3175 seqlock_init(&sk->sk_stamp_seq); 3176 #endif 3177 atomic_set(&sk->sk_zckey, 0); 3178 3179 #ifdef CONFIG_NET_RX_BUSY_POLL 3180 sk->sk_napi_id = 0; 3181 sk->sk_ll_usec = sysctl_net_busy_read; 3182 #endif 3183 3184 sk->sk_max_pacing_rate = ~0UL; 3185 sk->sk_pacing_rate = ~0UL; 3186 WRITE_ONCE(sk->sk_pacing_shift, 10); 3187 sk->sk_incoming_cpu = -1; 3188 3189 sk_rx_queue_clear(sk); 3190 /* 3191 * Before updating sk_refcnt, we must commit prior changes to memory 3192 * (Documentation/RCU/rculist_nulls.rst for details) 3193 */ 3194 smp_wmb(); 3195 refcount_set(&sk->sk_refcnt, 1); 3196 atomic_set(&sk->sk_drops, 0); 3197 } 3198 EXPORT_SYMBOL(sock_init_data); 3199 3200 void lock_sock_nested(struct sock *sk, int subclass) 3201 { 3202 /* The sk_lock has mutex_lock() semantics here. */ 3203 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_); 3204 3205 might_sleep(); 3206 spin_lock_bh(&sk->sk_lock.slock); 3207 if (sk->sk_lock.owned) 3208 __lock_sock(sk); 3209 sk->sk_lock.owned = 1; 3210 spin_unlock_bh(&sk->sk_lock.slock); 3211 } 3212 EXPORT_SYMBOL(lock_sock_nested); 3213 3214 void release_sock(struct sock *sk) 3215 { 3216 spin_lock_bh(&sk->sk_lock.slock); 3217 if (sk->sk_backlog.tail) 3218 __release_sock(sk); 3219 3220 /* Warning : release_cb() might need to release sk ownership, 3221 * ie call sock_release_ownership(sk) before us. 3222 */ 3223 if (sk->sk_prot->release_cb) 3224 sk->sk_prot->release_cb(sk); 3225 3226 sock_release_ownership(sk); 3227 if (waitqueue_active(&sk->sk_lock.wq)) 3228 wake_up(&sk->sk_lock.wq); 3229 spin_unlock_bh(&sk->sk_lock.slock); 3230 } 3231 EXPORT_SYMBOL(release_sock); 3232 3233 bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock) 3234 { 3235 might_sleep(); 3236 spin_lock_bh(&sk->sk_lock.slock); 3237 3238 if (!sk->sk_lock.owned) { 3239 /* 3240 * Fast path return with bottom halves disabled and 3241 * sock::sk_lock.slock held. 3242 * 3243 * The 'mutex' is not contended and holding 3244 * sock::sk_lock.slock prevents all other lockers to 3245 * proceed so the corresponding unlock_sock_fast() can 3246 * avoid the slow path of release_sock() completely and 3247 * just release slock. 3248 * 3249 * From a semantical POV this is equivalent to 'acquiring' 3250 * the 'mutex', hence the corresponding lockdep 3251 * mutex_release() has to happen in the fast path of 3252 * unlock_sock_fast(). 3253 */ 3254 return false; 3255 } 3256 3257 __lock_sock(sk); 3258 sk->sk_lock.owned = 1; 3259 __acquire(&sk->sk_lock.slock); 3260 spin_unlock_bh(&sk->sk_lock.slock); 3261 return true; 3262 } 3263 EXPORT_SYMBOL(__lock_sock_fast); 3264 3265 int sock_gettstamp(struct socket *sock, void __user *userstamp, 3266 bool timeval, bool time32) 3267 { 3268 struct sock *sk = sock->sk; 3269 struct timespec64 ts; 3270 3271 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 3272 ts = ktime_to_timespec64(sock_read_timestamp(sk)); 3273 if (ts.tv_sec == -1) 3274 return -ENOENT; 3275 if (ts.tv_sec == 0) { 3276 ktime_t kt = ktime_get_real(); 3277 sock_write_timestamp(sk, kt); 3278 ts = ktime_to_timespec64(kt); 3279 } 3280 3281 if (timeval) 3282 ts.tv_nsec /= 1000; 3283 3284 #ifdef CONFIG_COMPAT_32BIT_TIME 3285 if (time32) 3286 return put_old_timespec32(&ts, userstamp); 3287 #endif 3288 #ifdef CONFIG_SPARC64 3289 /* beware of padding in sparc64 timeval */ 3290 if (timeval && !in_compat_syscall()) { 3291 struct __kernel_old_timeval __user tv = { 3292 .tv_sec = ts.tv_sec, 3293 .tv_usec = ts.tv_nsec, 3294 }; 3295 if (copy_to_user(userstamp, &tv, sizeof(tv))) 3296 return -EFAULT; 3297 return 0; 3298 } 3299 #endif 3300 return put_timespec64(&ts, userstamp); 3301 } 3302 EXPORT_SYMBOL(sock_gettstamp); 3303 3304 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag) 3305 { 3306 if (!sock_flag(sk, flag)) { 3307 unsigned long previous_flags = sk->sk_flags; 3308 3309 sock_set_flag(sk, flag); 3310 /* 3311 * we just set one of the two flags which require net 3312 * time stamping, but time stamping might have been on 3313 * already because of the other one 3314 */ 3315 if (sock_needs_netstamp(sk) && 3316 !(previous_flags & SK_FLAGS_TIMESTAMP)) 3317 net_enable_timestamp(); 3318 } 3319 } 3320 3321 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, 3322 int level, int type) 3323 { 3324 struct sock_exterr_skb *serr; 3325 struct sk_buff *skb; 3326 int copied, err; 3327 3328 err = -EAGAIN; 3329 skb = sock_dequeue_err_skb(sk); 3330 if (skb == NULL) 3331 goto out; 3332 3333 copied = skb->len; 3334 if (copied > len) { 3335 msg->msg_flags |= MSG_TRUNC; 3336 copied = len; 3337 } 3338 err = skb_copy_datagram_msg(skb, 0, msg, copied); 3339 if (err) 3340 goto out_free_skb; 3341 3342 sock_recv_timestamp(msg, sk, skb); 3343 3344 serr = SKB_EXT_ERR(skb); 3345 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee); 3346 3347 msg->msg_flags |= MSG_ERRQUEUE; 3348 err = copied; 3349 3350 out_free_skb: 3351 kfree_skb(skb); 3352 out: 3353 return err; 3354 } 3355 EXPORT_SYMBOL(sock_recv_errqueue); 3356 3357 /* 3358 * Get a socket option on an socket. 3359 * 3360 * FIX: POSIX 1003.1g is very ambiguous here. It states that 3361 * asynchronous errors should be reported by getsockopt. We assume 3362 * this means if you specify SO_ERROR (otherwise whats the point of it). 3363 */ 3364 int sock_common_getsockopt(struct socket *sock, int level, int optname, 3365 char __user *optval, int __user *optlen) 3366 { 3367 struct sock *sk = sock->sk; 3368 3369 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen); 3370 } 3371 EXPORT_SYMBOL(sock_common_getsockopt); 3372 3373 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 3374 int flags) 3375 { 3376 struct sock *sk = sock->sk; 3377 int addr_len = 0; 3378 int err; 3379 3380 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT, 3381 flags & ~MSG_DONTWAIT, &addr_len); 3382 if (err >= 0) 3383 msg->msg_namelen = addr_len; 3384 return err; 3385 } 3386 EXPORT_SYMBOL(sock_common_recvmsg); 3387 3388 /* 3389 * Set socket options on an inet socket. 3390 */ 3391 int sock_common_setsockopt(struct socket *sock, int level, int optname, 3392 sockptr_t optval, unsigned int optlen) 3393 { 3394 struct sock *sk = sock->sk; 3395 3396 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen); 3397 } 3398 EXPORT_SYMBOL(sock_common_setsockopt); 3399 3400 void sk_common_release(struct sock *sk) 3401 { 3402 if (sk->sk_prot->destroy) 3403 sk->sk_prot->destroy(sk); 3404 3405 /* 3406 * Observation: when sk_common_release is called, processes have 3407 * no access to socket. But net still has. 3408 * Step one, detach it from networking: 3409 * 3410 * A. Remove from hash tables. 3411 */ 3412 3413 sk->sk_prot->unhash(sk); 3414 3415 /* 3416 * In this point socket cannot receive new packets, but it is possible 3417 * that some packets are in flight because some CPU runs receiver and 3418 * did hash table lookup before we unhashed socket. They will achieve 3419 * receive queue and will be purged by socket destructor. 3420 * 3421 * Also we still have packets pending on receive queue and probably, 3422 * our own packets waiting in device queues. sock_destroy will drain 3423 * receive queue, but transmitted packets will delay socket destruction 3424 * until the last reference will be released. 3425 */ 3426 3427 sock_orphan(sk); 3428 3429 xfrm_sk_free_policy(sk); 3430 3431 sk_refcnt_debug_release(sk); 3432 3433 sock_put(sk); 3434 } 3435 EXPORT_SYMBOL(sk_common_release); 3436 3437 void sk_get_meminfo(const struct sock *sk, u32 *mem) 3438 { 3439 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS); 3440 3441 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk); 3442 mem[SK_MEMINFO_RCVBUF] = READ_ONCE(sk->sk_rcvbuf); 3443 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk); 3444 mem[SK_MEMINFO_SNDBUF] = READ_ONCE(sk->sk_sndbuf); 3445 mem[SK_MEMINFO_FWD_ALLOC] = sk->sk_forward_alloc; 3446 mem[SK_MEMINFO_WMEM_QUEUED] = READ_ONCE(sk->sk_wmem_queued); 3447 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc); 3448 mem[SK_MEMINFO_BACKLOG] = READ_ONCE(sk->sk_backlog.len); 3449 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops); 3450 } 3451 3452 #ifdef CONFIG_PROC_FS 3453 #define PROTO_INUSE_NR 64 /* should be enough for the first time */ 3454 struct prot_inuse { 3455 int val[PROTO_INUSE_NR]; 3456 }; 3457 3458 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR); 3459 3460 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val) 3461 { 3462 __this_cpu_add(net->core.prot_inuse->val[prot->inuse_idx], val); 3463 } 3464 EXPORT_SYMBOL_GPL(sock_prot_inuse_add); 3465 3466 int sock_prot_inuse_get(struct net *net, struct proto *prot) 3467 { 3468 int cpu, idx = prot->inuse_idx; 3469 int res = 0; 3470 3471 for_each_possible_cpu(cpu) 3472 res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx]; 3473 3474 return res >= 0 ? res : 0; 3475 } 3476 EXPORT_SYMBOL_GPL(sock_prot_inuse_get); 3477 3478 static void sock_inuse_add(struct net *net, int val) 3479 { 3480 this_cpu_add(*net->core.sock_inuse, val); 3481 } 3482 3483 int sock_inuse_get(struct net *net) 3484 { 3485 int cpu, res = 0; 3486 3487 for_each_possible_cpu(cpu) 3488 res += *per_cpu_ptr(net->core.sock_inuse, cpu); 3489 3490 return res; 3491 } 3492 3493 EXPORT_SYMBOL_GPL(sock_inuse_get); 3494 3495 static int __net_init sock_inuse_init_net(struct net *net) 3496 { 3497 net->core.prot_inuse = alloc_percpu(struct prot_inuse); 3498 if (net->core.prot_inuse == NULL) 3499 return -ENOMEM; 3500 3501 net->core.sock_inuse = alloc_percpu(int); 3502 if (net->core.sock_inuse == NULL) 3503 goto out; 3504 3505 return 0; 3506 3507 out: 3508 free_percpu(net->core.prot_inuse); 3509 return -ENOMEM; 3510 } 3511 3512 static void __net_exit sock_inuse_exit_net(struct net *net) 3513 { 3514 free_percpu(net->core.prot_inuse); 3515 free_percpu(net->core.sock_inuse); 3516 } 3517 3518 static struct pernet_operations net_inuse_ops = { 3519 .init = sock_inuse_init_net, 3520 .exit = sock_inuse_exit_net, 3521 }; 3522 3523 static __init int net_inuse_init(void) 3524 { 3525 if (register_pernet_subsys(&net_inuse_ops)) 3526 panic("Cannot initialize net inuse counters"); 3527 3528 return 0; 3529 } 3530 3531 core_initcall(net_inuse_init); 3532 3533 static int assign_proto_idx(struct proto *prot) 3534 { 3535 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR); 3536 3537 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) { 3538 pr_err("PROTO_INUSE_NR exhausted\n"); 3539 return -ENOSPC; 3540 } 3541 3542 set_bit(prot->inuse_idx, proto_inuse_idx); 3543 return 0; 3544 } 3545 3546 static void release_proto_idx(struct proto *prot) 3547 { 3548 if (prot->inuse_idx != PROTO_INUSE_NR - 1) 3549 clear_bit(prot->inuse_idx, proto_inuse_idx); 3550 } 3551 #else 3552 static inline int assign_proto_idx(struct proto *prot) 3553 { 3554 return 0; 3555 } 3556 3557 static inline void release_proto_idx(struct proto *prot) 3558 { 3559 } 3560 3561 static void sock_inuse_add(struct net *net, int val) 3562 { 3563 } 3564 #endif 3565 3566 static void tw_prot_cleanup(struct timewait_sock_ops *twsk_prot) 3567 { 3568 if (!twsk_prot) 3569 return; 3570 kfree(twsk_prot->twsk_slab_name); 3571 twsk_prot->twsk_slab_name = NULL; 3572 kmem_cache_destroy(twsk_prot->twsk_slab); 3573 twsk_prot->twsk_slab = NULL; 3574 } 3575 3576 static int tw_prot_init(const struct proto *prot) 3577 { 3578 struct timewait_sock_ops *twsk_prot = prot->twsk_prot; 3579 3580 if (!twsk_prot) 3581 return 0; 3582 3583 twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", 3584 prot->name); 3585 if (!twsk_prot->twsk_slab_name) 3586 return -ENOMEM; 3587 3588 twsk_prot->twsk_slab = 3589 kmem_cache_create(twsk_prot->twsk_slab_name, 3590 twsk_prot->twsk_obj_size, 0, 3591 SLAB_ACCOUNT | prot->slab_flags, 3592 NULL); 3593 if (!twsk_prot->twsk_slab) { 3594 pr_crit("%s: Can't create timewait sock SLAB cache!\n", 3595 prot->name); 3596 return -ENOMEM; 3597 } 3598 3599 return 0; 3600 } 3601 3602 static void req_prot_cleanup(struct request_sock_ops *rsk_prot) 3603 { 3604 if (!rsk_prot) 3605 return; 3606 kfree(rsk_prot->slab_name); 3607 rsk_prot->slab_name = NULL; 3608 kmem_cache_destroy(rsk_prot->slab); 3609 rsk_prot->slab = NULL; 3610 } 3611 3612 static int req_prot_init(const struct proto *prot) 3613 { 3614 struct request_sock_ops *rsk_prot = prot->rsk_prot; 3615 3616 if (!rsk_prot) 3617 return 0; 3618 3619 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", 3620 prot->name); 3621 if (!rsk_prot->slab_name) 3622 return -ENOMEM; 3623 3624 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name, 3625 rsk_prot->obj_size, 0, 3626 SLAB_ACCOUNT | prot->slab_flags, 3627 NULL); 3628 3629 if (!rsk_prot->slab) { 3630 pr_crit("%s: Can't create request sock SLAB cache!\n", 3631 prot->name); 3632 return -ENOMEM; 3633 } 3634 return 0; 3635 } 3636 3637 int proto_register(struct proto *prot, int alloc_slab) 3638 { 3639 int ret = -ENOBUFS; 3640 3641 if (alloc_slab) { 3642 prot->slab = kmem_cache_create_usercopy(prot->name, 3643 prot->obj_size, 0, 3644 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT | 3645 prot->slab_flags, 3646 prot->useroffset, prot->usersize, 3647 NULL); 3648 3649 if (prot->slab == NULL) { 3650 pr_crit("%s: Can't create sock SLAB cache!\n", 3651 prot->name); 3652 goto out; 3653 } 3654 3655 if (req_prot_init(prot)) 3656 goto out_free_request_sock_slab; 3657 3658 if (tw_prot_init(prot)) 3659 goto out_free_timewait_sock_slab; 3660 } 3661 3662 mutex_lock(&proto_list_mutex); 3663 ret = assign_proto_idx(prot); 3664 if (ret) { 3665 mutex_unlock(&proto_list_mutex); 3666 goto out_free_timewait_sock_slab; 3667 } 3668 list_add(&prot->node, &proto_list); 3669 mutex_unlock(&proto_list_mutex); 3670 return ret; 3671 3672 out_free_timewait_sock_slab: 3673 if (alloc_slab) 3674 tw_prot_cleanup(prot->twsk_prot); 3675 out_free_request_sock_slab: 3676 if (alloc_slab) { 3677 req_prot_cleanup(prot->rsk_prot); 3678 3679 kmem_cache_destroy(prot->slab); 3680 prot->slab = NULL; 3681 } 3682 out: 3683 return ret; 3684 } 3685 EXPORT_SYMBOL(proto_register); 3686 3687 void proto_unregister(struct proto *prot) 3688 { 3689 mutex_lock(&proto_list_mutex); 3690 release_proto_idx(prot); 3691 list_del(&prot->node); 3692 mutex_unlock(&proto_list_mutex); 3693 3694 kmem_cache_destroy(prot->slab); 3695 prot->slab = NULL; 3696 3697 req_prot_cleanup(prot->rsk_prot); 3698 tw_prot_cleanup(prot->twsk_prot); 3699 } 3700 EXPORT_SYMBOL(proto_unregister); 3701 3702 int sock_load_diag_module(int family, int protocol) 3703 { 3704 if (!protocol) { 3705 if (!sock_is_registered(family)) 3706 return -ENOENT; 3707 3708 return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK, 3709 NETLINK_SOCK_DIAG, family); 3710 } 3711 3712 #ifdef CONFIG_INET 3713 if (family == AF_INET && 3714 protocol != IPPROTO_RAW && 3715 protocol < MAX_INET_PROTOS && 3716 !rcu_access_pointer(inet_protos[protocol])) 3717 return -ENOENT; 3718 #endif 3719 3720 return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK, 3721 NETLINK_SOCK_DIAG, family, protocol); 3722 } 3723 EXPORT_SYMBOL(sock_load_diag_module); 3724 3725 #ifdef CONFIG_PROC_FS 3726 static void *proto_seq_start(struct seq_file *seq, loff_t *pos) 3727 __acquires(proto_list_mutex) 3728 { 3729 mutex_lock(&proto_list_mutex); 3730 return seq_list_start_head(&proto_list, *pos); 3731 } 3732 3733 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos) 3734 { 3735 return seq_list_next(v, &proto_list, pos); 3736 } 3737 3738 static void proto_seq_stop(struct seq_file *seq, void *v) 3739 __releases(proto_list_mutex) 3740 { 3741 mutex_unlock(&proto_list_mutex); 3742 } 3743 3744 static char proto_method_implemented(const void *method) 3745 { 3746 return method == NULL ? 'n' : 'y'; 3747 } 3748 static long sock_prot_memory_allocated(struct proto *proto) 3749 { 3750 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L; 3751 } 3752 3753 static const char *sock_prot_memory_pressure(struct proto *proto) 3754 { 3755 return proto->memory_pressure != NULL ? 3756 proto_memory_pressure(proto) ? "yes" : "no" : "NI"; 3757 } 3758 3759 static void proto_seq_printf(struct seq_file *seq, struct proto *proto) 3760 { 3761 3762 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s " 3763 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n", 3764 proto->name, 3765 proto->obj_size, 3766 sock_prot_inuse_get(seq_file_net(seq), proto), 3767 sock_prot_memory_allocated(proto), 3768 sock_prot_memory_pressure(proto), 3769 proto->max_header, 3770 proto->slab == NULL ? "no" : "yes", 3771 module_name(proto->owner), 3772 proto_method_implemented(proto->close), 3773 proto_method_implemented(proto->connect), 3774 proto_method_implemented(proto->disconnect), 3775 proto_method_implemented(proto->accept), 3776 proto_method_implemented(proto->ioctl), 3777 proto_method_implemented(proto->init), 3778 proto_method_implemented(proto->destroy), 3779 proto_method_implemented(proto->shutdown), 3780 proto_method_implemented(proto->setsockopt), 3781 proto_method_implemented(proto->getsockopt), 3782 proto_method_implemented(proto->sendmsg), 3783 proto_method_implemented(proto->recvmsg), 3784 proto_method_implemented(proto->sendpage), 3785 proto_method_implemented(proto->bind), 3786 proto_method_implemented(proto->backlog_rcv), 3787 proto_method_implemented(proto->hash), 3788 proto_method_implemented(proto->unhash), 3789 proto_method_implemented(proto->get_port), 3790 proto_method_implemented(proto->enter_memory_pressure)); 3791 } 3792 3793 static int proto_seq_show(struct seq_file *seq, void *v) 3794 { 3795 if (v == &proto_list) 3796 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s", 3797 "protocol", 3798 "size", 3799 "sockets", 3800 "memory", 3801 "press", 3802 "maxhdr", 3803 "slab", 3804 "module", 3805 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n"); 3806 else 3807 proto_seq_printf(seq, list_entry(v, struct proto, node)); 3808 return 0; 3809 } 3810 3811 static const struct seq_operations proto_seq_ops = { 3812 .start = proto_seq_start, 3813 .next = proto_seq_next, 3814 .stop = proto_seq_stop, 3815 .show = proto_seq_show, 3816 }; 3817 3818 static __net_init int proto_init_net(struct net *net) 3819 { 3820 if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops, 3821 sizeof(struct seq_net_private))) 3822 return -ENOMEM; 3823 3824 return 0; 3825 } 3826 3827 static __net_exit void proto_exit_net(struct net *net) 3828 { 3829 remove_proc_entry("protocols", net->proc_net); 3830 } 3831 3832 3833 static __net_initdata struct pernet_operations proto_net_ops = { 3834 .init = proto_init_net, 3835 .exit = proto_exit_net, 3836 }; 3837 3838 static int __init proto_init(void) 3839 { 3840 return register_pernet_subsys(&proto_net_ops); 3841 } 3842 3843 subsys_initcall(proto_init); 3844 3845 #endif /* PROC_FS */ 3846 3847 #ifdef CONFIG_NET_RX_BUSY_POLL 3848 bool sk_busy_loop_end(void *p, unsigned long start_time) 3849 { 3850 struct sock *sk = p; 3851 3852 return !skb_queue_empty_lockless(&sk->sk_receive_queue) || 3853 sk_busy_loop_timeout(sk, start_time); 3854 } 3855 EXPORT_SYMBOL(sk_busy_loop_end); 3856 #endif /* CONFIG_NET_RX_BUSY_POLL */ 3857 3858 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len) 3859 { 3860 if (!sk->sk_prot->bind_add) 3861 return -EOPNOTSUPP; 3862 return sk->sk_prot->bind_add(sk, addr, addr_len); 3863 } 3864 EXPORT_SYMBOL(sock_bind_add); 3865