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 1380 static void cred_to_ucred(struct pid *pid, const struct cred *cred, 1381 struct ucred *ucred) 1382 { 1383 ucred->pid = pid_vnr(pid); 1384 ucred->uid = ucred->gid = -1; 1385 if (cred) { 1386 struct user_namespace *current_ns = current_user_ns(); 1387 1388 ucred->uid = from_kuid_munged(current_ns, cred->euid); 1389 ucred->gid = from_kgid_munged(current_ns, cred->egid); 1390 } 1391 } 1392 1393 static int groups_to_user(gid_t __user *dst, const struct group_info *src) 1394 { 1395 struct user_namespace *user_ns = current_user_ns(); 1396 int i; 1397 1398 for (i = 0; i < src->ngroups; i++) 1399 if (put_user(from_kgid_munged(user_ns, src->gid[i]), dst + i)) 1400 return -EFAULT; 1401 1402 return 0; 1403 } 1404 1405 int sock_getsockopt(struct socket *sock, int level, int optname, 1406 char __user *optval, int __user *optlen) 1407 { 1408 struct sock *sk = sock->sk; 1409 1410 union { 1411 int val; 1412 u64 val64; 1413 unsigned long ulval; 1414 struct linger ling; 1415 struct old_timeval32 tm32; 1416 struct __kernel_old_timeval tm; 1417 struct __kernel_sock_timeval stm; 1418 struct sock_txtime txtime; 1419 struct so_timestamping timestamping; 1420 } v; 1421 1422 int lv = sizeof(int); 1423 int len; 1424 1425 if (get_user(len, optlen)) 1426 return -EFAULT; 1427 if (len < 0) 1428 return -EINVAL; 1429 1430 memset(&v, 0, sizeof(v)); 1431 1432 switch (optname) { 1433 case SO_DEBUG: 1434 v.val = sock_flag(sk, SOCK_DBG); 1435 break; 1436 1437 case SO_DONTROUTE: 1438 v.val = sock_flag(sk, SOCK_LOCALROUTE); 1439 break; 1440 1441 case SO_BROADCAST: 1442 v.val = sock_flag(sk, SOCK_BROADCAST); 1443 break; 1444 1445 case SO_SNDBUF: 1446 v.val = sk->sk_sndbuf; 1447 break; 1448 1449 case SO_RCVBUF: 1450 v.val = sk->sk_rcvbuf; 1451 break; 1452 1453 case SO_REUSEADDR: 1454 v.val = sk->sk_reuse; 1455 break; 1456 1457 case SO_REUSEPORT: 1458 v.val = sk->sk_reuseport; 1459 break; 1460 1461 case SO_KEEPALIVE: 1462 v.val = sock_flag(sk, SOCK_KEEPOPEN); 1463 break; 1464 1465 case SO_TYPE: 1466 v.val = sk->sk_type; 1467 break; 1468 1469 case SO_PROTOCOL: 1470 v.val = sk->sk_protocol; 1471 break; 1472 1473 case SO_DOMAIN: 1474 v.val = sk->sk_family; 1475 break; 1476 1477 case SO_ERROR: 1478 v.val = -sock_error(sk); 1479 if (v.val == 0) 1480 v.val = xchg(&sk->sk_err_soft, 0); 1481 break; 1482 1483 case SO_OOBINLINE: 1484 v.val = sock_flag(sk, SOCK_URGINLINE); 1485 break; 1486 1487 case SO_NO_CHECK: 1488 v.val = sk->sk_no_check_tx; 1489 break; 1490 1491 case SO_PRIORITY: 1492 v.val = sk->sk_priority; 1493 break; 1494 1495 case SO_LINGER: 1496 lv = sizeof(v.ling); 1497 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER); 1498 v.ling.l_linger = sk->sk_lingertime / HZ; 1499 break; 1500 1501 case SO_BSDCOMPAT: 1502 break; 1503 1504 case SO_TIMESTAMP_OLD: 1505 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && 1506 !sock_flag(sk, SOCK_TSTAMP_NEW) && 1507 !sock_flag(sk, SOCK_RCVTSTAMPNS); 1508 break; 1509 1510 case SO_TIMESTAMPNS_OLD: 1511 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && !sock_flag(sk, SOCK_TSTAMP_NEW); 1512 break; 1513 1514 case SO_TIMESTAMP_NEW: 1515 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && sock_flag(sk, SOCK_TSTAMP_NEW); 1516 break; 1517 1518 case SO_TIMESTAMPNS_NEW: 1519 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && sock_flag(sk, SOCK_TSTAMP_NEW); 1520 break; 1521 1522 case SO_TIMESTAMPING_OLD: 1523 lv = sizeof(v.timestamping); 1524 v.timestamping.flags = sk->sk_tsflags; 1525 v.timestamping.bind_phc = sk->sk_bind_phc; 1526 break; 1527 1528 case SO_RCVTIMEO_OLD: 1529 case SO_RCVTIMEO_NEW: 1530 lv = sock_get_timeout(sk->sk_rcvtimeo, &v, SO_RCVTIMEO_OLD == optname); 1531 break; 1532 1533 case SO_SNDTIMEO_OLD: 1534 case SO_SNDTIMEO_NEW: 1535 lv = sock_get_timeout(sk->sk_sndtimeo, &v, SO_SNDTIMEO_OLD == optname); 1536 break; 1537 1538 case SO_RCVLOWAT: 1539 v.val = sk->sk_rcvlowat; 1540 break; 1541 1542 case SO_SNDLOWAT: 1543 v.val = 1; 1544 break; 1545 1546 case SO_PASSCRED: 1547 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags); 1548 break; 1549 1550 case SO_PEERCRED: 1551 { 1552 struct ucred peercred; 1553 if (len > sizeof(peercred)) 1554 len = sizeof(peercred); 1555 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred); 1556 if (copy_to_user(optval, &peercred, len)) 1557 return -EFAULT; 1558 goto lenout; 1559 } 1560 1561 case SO_PEERGROUPS: 1562 { 1563 int ret, n; 1564 1565 if (!sk->sk_peer_cred) 1566 return -ENODATA; 1567 1568 n = sk->sk_peer_cred->group_info->ngroups; 1569 if (len < n * sizeof(gid_t)) { 1570 len = n * sizeof(gid_t); 1571 return put_user(len, optlen) ? -EFAULT : -ERANGE; 1572 } 1573 len = n * sizeof(gid_t); 1574 1575 ret = groups_to_user((gid_t __user *)optval, 1576 sk->sk_peer_cred->group_info); 1577 if (ret) 1578 return ret; 1579 goto lenout; 1580 } 1581 1582 case SO_PEERNAME: 1583 { 1584 char address[128]; 1585 1586 lv = sock->ops->getname(sock, (struct sockaddr *)address, 2); 1587 if (lv < 0) 1588 return -ENOTCONN; 1589 if (lv < len) 1590 return -EINVAL; 1591 if (copy_to_user(optval, address, len)) 1592 return -EFAULT; 1593 goto lenout; 1594 } 1595 1596 /* Dubious BSD thing... Probably nobody even uses it, but 1597 * the UNIX standard wants it for whatever reason... -DaveM 1598 */ 1599 case SO_ACCEPTCONN: 1600 v.val = sk->sk_state == TCP_LISTEN; 1601 break; 1602 1603 case SO_PASSSEC: 1604 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags); 1605 break; 1606 1607 case SO_PEERSEC: 1608 return security_socket_getpeersec_stream(sock, optval, optlen, len); 1609 1610 case SO_MARK: 1611 v.val = sk->sk_mark; 1612 break; 1613 1614 case SO_RXQ_OVFL: 1615 v.val = sock_flag(sk, SOCK_RXQ_OVFL); 1616 break; 1617 1618 case SO_WIFI_STATUS: 1619 v.val = sock_flag(sk, SOCK_WIFI_STATUS); 1620 break; 1621 1622 case SO_PEEK_OFF: 1623 if (!sock->ops->set_peek_off) 1624 return -EOPNOTSUPP; 1625 1626 v.val = sk->sk_peek_off; 1627 break; 1628 case SO_NOFCS: 1629 v.val = sock_flag(sk, SOCK_NOFCS); 1630 break; 1631 1632 case SO_BINDTODEVICE: 1633 return sock_getbindtodevice(sk, optval, optlen, len); 1634 1635 case SO_GET_FILTER: 1636 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len); 1637 if (len < 0) 1638 return len; 1639 1640 goto lenout; 1641 1642 case SO_LOCK_FILTER: 1643 v.val = sock_flag(sk, SOCK_FILTER_LOCKED); 1644 break; 1645 1646 case SO_BPF_EXTENSIONS: 1647 v.val = bpf_tell_extensions(); 1648 break; 1649 1650 case SO_SELECT_ERR_QUEUE: 1651 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE); 1652 break; 1653 1654 #ifdef CONFIG_NET_RX_BUSY_POLL 1655 case SO_BUSY_POLL: 1656 v.val = sk->sk_ll_usec; 1657 break; 1658 case SO_PREFER_BUSY_POLL: 1659 v.val = READ_ONCE(sk->sk_prefer_busy_poll); 1660 break; 1661 #endif 1662 1663 case SO_MAX_PACING_RATE: 1664 if (sizeof(v.ulval) != sizeof(v.val) && len >= sizeof(v.ulval)) { 1665 lv = sizeof(v.ulval); 1666 v.ulval = sk->sk_max_pacing_rate; 1667 } else { 1668 /* 32bit version */ 1669 v.val = min_t(unsigned long, sk->sk_max_pacing_rate, ~0U); 1670 } 1671 break; 1672 1673 case SO_INCOMING_CPU: 1674 v.val = READ_ONCE(sk->sk_incoming_cpu); 1675 break; 1676 1677 case SO_MEMINFO: 1678 { 1679 u32 meminfo[SK_MEMINFO_VARS]; 1680 1681 sk_get_meminfo(sk, meminfo); 1682 1683 len = min_t(unsigned int, len, sizeof(meminfo)); 1684 if (copy_to_user(optval, &meminfo, len)) 1685 return -EFAULT; 1686 1687 goto lenout; 1688 } 1689 1690 #ifdef CONFIG_NET_RX_BUSY_POLL 1691 case SO_INCOMING_NAPI_ID: 1692 v.val = READ_ONCE(sk->sk_napi_id); 1693 1694 /* aggregate non-NAPI IDs down to 0 */ 1695 if (v.val < MIN_NAPI_ID) 1696 v.val = 0; 1697 1698 break; 1699 #endif 1700 1701 case SO_COOKIE: 1702 lv = sizeof(u64); 1703 if (len < lv) 1704 return -EINVAL; 1705 v.val64 = sock_gen_cookie(sk); 1706 break; 1707 1708 case SO_ZEROCOPY: 1709 v.val = sock_flag(sk, SOCK_ZEROCOPY); 1710 break; 1711 1712 case SO_TXTIME: 1713 lv = sizeof(v.txtime); 1714 v.txtime.clockid = sk->sk_clockid; 1715 v.txtime.flags |= sk->sk_txtime_deadline_mode ? 1716 SOF_TXTIME_DEADLINE_MODE : 0; 1717 v.txtime.flags |= sk->sk_txtime_report_errors ? 1718 SOF_TXTIME_REPORT_ERRORS : 0; 1719 break; 1720 1721 case SO_BINDTOIFINDEX: 1722 v.val = sk->sk_bound_dev_if; 1723 break; 1724 1725 case SO_NETNS_COOKIE: 1726 lv = sizeof(u64); 1727 if (len != lv) 1728 return -EINVAL; 1729 v.val64 = sock_net(sk)->net_cookie; 1730 break; 1731 1732 case SO_BUF_LOCK: 1733 v.val = sk->sk_userlocks & SOCK_BUF_LOCK_MASK; 1734 break; 1735 1736 default: 1737 /* We implement the SO_SNDLOWAT etc to not be settable 1738 * (1003.1g 7). 1739 */ 1740 return -ENOPROTOOPT; 1741 } 1742 1743 if (len > lv) 1744 len = lv; 1745 if (copy_to_user(optval, &v, len)) 1746 return -EFAULT; 1747 lenout: 1748 if (put_user(len, optlen)) 1749 return -EFAULT; 1750 return 0; 1751 } 1752 1753 /* 1754 * Initialize an sk_lock. 1755 * 1756 * (We also register the sk_lock with the lock validator.) 1757 */ 1758 static inline void sock_lock_init(struct sock *sk) 1759 { 1760 if (sk->sk_kern_sock) 1761 sock_lock_init_class_and_name( 1762 sk, 1763 af_family_kern_slock_key_strings[sk->sk_family], 1764 af_family_kern_slock_keys + sk->sk_family, 1765 af_family_kern_key_strings[sk->sk_family], 1766 af_family_kern_keys + sk->sk_family); 1767 else 1768 sock_lock_init_class_and_name( 1769 sk, 1770 af_family_slock_key_strings[sk->sk_family], 1771 af_family_slock_keys + sk->sk_family, 1772 af_family_key_strings[sk->sk_family], 1773 af_family_keys + sk->sk_family); 1774 } 1775 1776 /* 1777 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet, 1778 * even temporarly, because of RCU lookups. sk_node should also be left as is. 1779 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end 1780 */ 1781 static void sock_copy(struct sock *nsk, const struct sock *osk) 1782 { 1783 const struct proto *prot = READ_ONCE(osk->sk_prot); 1784 #ifdef CONFIG_SECURITY_NETWORK 1785 void *sptr = nsk->sk_security; 1786 #endif 1787 1788 /* If we move sk_tx_queue_mapping out of the private section, 1789 * we must check if sk_tx_queue_clear() is called after 1790 * sock_copy() in sk_clone_lock(). 1791 */ 1792 BUILD_BUG_ON(offsetof(struct sock, sk_tx_queue_mapping) < 1793 offsetof(struct sock, sk_dontcopy_begin) || 1794 offsetof(struct sock, sk_tx_queue_mapping) >= 1795 offsetof(struct sock, sk_dontcopy_end)); 1796 1797 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin)); 1798 1799 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end, 1800 prot->obj_size - offsetof(struct sock, sk_dontcopy_end)); 1801 1802 #ifdef CONFIG_SECURITY_NETWORK 1803 nsk->sk_security = sptr; 1804 security_sk_clone(osk, nsk); 1805 #endif 1806 } 1807 1808 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority, 1809 int family) 1810 { 1811 struct sock *sk; 1812 struct kmem_cache *slab; 1813 1814 slab = prot->slab; 1815 if (slab != NULL) { 1816 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO); 1817 if (!sk) 1818 return sk; 1819 if (want_init_on_alloc(priority)) 1820 sk_prot_clear_nulls(sk, prot->obj_size); 1821 } else 1822 sk = kmalloc(prot->obj_size, priority); 1823 1824 if (sk != NULL) { 1825 if (security_sk_alloc(sk, family, priority)) 1826 goto out_free; 1827 1828 if (!try_module_get(prot->owner)) 1829 goto out_free_sec; 1830 } 1831 1832 return sk; 1833 1834 out_free_sec: 1835 security_sk_free(sk); 1836 out_free: 1837 if (slab != NULL) 1838 kmem_cache_free(slab, sk); 1839 else 1840 kfree(sk); 1841 return NULL; 1842 } 1843 1844 static void sk_prot_free(struct proto *prot, struct sock *sk) 1845 { 1846 struct kmem_cache *slab; 1847 struct module *owner; 1848 1849 owner = prot->owner; 1850 slab = prot->slab; 1851 1852 cgroup_sk_free(&sk->sk_cgrp_data); 1853 mem_cgroup_sk_free(sk); 1854 security_sk_free(sk); 1855 if (slab != NULL) 1856 kmem_cache_free(slab, sk); 1857 else 1858 kfree(sk); 1859 module_put(owner); 1860 } 1861 1862 /** 1863 * sk_alloc - All socket objects are allocated here 1864 * @net: the applicable net namespace 1865 * @family: protocol family 1866 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 1867 * @prot: struct proto associated with this new sock instance 1868 * @kern: is this to be a kernel socket? 1869 */ 1870 struct sock *sk_alloc(struct net *net, int family, gfp_t priority, 1871 struct proto *prot, int kern) 1872 { 1873 struct sock *sk; 1874 1875 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family); 1876 if (sk) { 1877 sk->sk_family = family; 1878 /* 1879 * See comment in struct sock definition to understand 1880 * why we need sk_prot_creator -acme 1881 */ 1882 sk->sk_prot = sk->sk_prot_creator = prot; 1883 sk->sk_kern_sock = kern; 1884 sock_lock_init(sk); 1885 sk->sk_net_refcnt = kern ? 0 : 1; 1886 if (likely(sk->sk_net_refcnt)) { 1887 get_net(net); 1888 sock_inuse_add(net, 1); 1889 } 1890 1891 sock_net_set(sk, net); 1892 refcount_set(&sk->sk_wmem_alloc, 1); 1893 1894 mem_cgroup_sk_alloc(sk); 1895 cgroup_sk_alloc(&sk->sk_cgrp_data); 1896 sock_update_classid(&sk->sk_cgrp_data); 1897 sock_update_netprioidx(&sk->sk_cgrp_data); 1898 sk_tx_queue_clear(sk); 1899 } 1900 1901 return sk; 1902 } 1903 EXPORT_SYMBOL(sk_alloc); 1904 1905 /* Sockets having SOCK_RCU_FREE will call this function after one RCU 1906 * grace period. This is the case for UDP sockets and TCP listeners. 1907 */ 1908 static void __sk_destruct(struct rcu_head *head) 1909 { 1910 struct sock *sk = container_of(head, struct sock, sk_rcu); 1911 struct sk_filter *filter; 1912 1913 if (sk->sk_destruct) 1914 sk->sk_destruct(sk); 1915 1916 filter = rcu_dereference_check(sk->sk_filter, 1917 refcount_read(&sk->sk_wmem_alloc) == 0); 1918 if (filter) { 1919 sk_filter_uncharge(sk, filter); 1920 RCU_INIT_POINTER(sk->sk_filter, NULL); 1921 } 1922 1923 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP); 1924 1925 #ifdef CONFIG_BPF_SYSCALL 1926 bpf_sk_storage_free(sk); 1927 #endif 1928 1929 if (atomic_read(&sk->sk_omem_alloc)) 1930 pr_debug("%s: optmem leakage (%d bytes) detected\n", 1931 __func__, atomic_read(&sk->sk_omem_alloc)); 1932 1933 if (sk->sk_frag.page) { 1934 put_page(sk->sk_frag.page); 1935 sk->sk_frag.page = NULL; 1936 } 1937 1938 if (sk->sk_peer_cred) 1939 put_cred(sk->sk_peer_cred); 1940 put_pid(sk->sk_peer_pid); 1941 if (likely(sk->sk_net_refcnt)) 1942 put_net(sock_net(sk)); 1943 sk_prot_free(sk->sk_prot_creator, sk); 1944 } 1945 1946 void sk_destruct(struct sock *sk) 1947 { 1948 bool use_call_rcu = sock_flag(sk, SOCK_RCU_FREE); 1949 1950 if (rcu_access_pointer(sk->sk_reuseport_cb)) { 1951 reuseport_detach_sock(sk); 1952 use_call_rcu = true; 1953 } 1954 1955 if (use_call_rcu) 1956 call_rcu(&sk->sk_rcu, __sk_destruct); 1957 else 1958 __sk_destruct(&sk->sk_rcu); 1959 } 1960 1961 static void __sk_free(struct sock *sk) 1962 { 1963 if (likely(sk->sk_net_refcnt)) 1964 sock_inuse_add(sock_net(sk), -1); 1965 1966 if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk))) 1967 sock_diag_broadcast_destroy(sk); 1968 else 1969 sk_destruct(sk); 1970 } 1971 1972 void sk_free(struct sock *sk) 1973 { 1974 /* 1975 * We subtract one from sk_wmem_alloc and can know if 1976 * some packets are still in some tx queue. 1977 * If not null, sock_wfree() will call __sk_free(sk) later 1978 */ 1979 if (refcount_dec_and_test(&sk->sk_wmem_alloc)) 1980 __sk_free(sk); 1981 } 1982 EXPORT_SYMBOL(sk_free); 1983 1984 static void sk_init_common(struct sock *sk) 1985 { 1986 skb_queue_head_init(&sk->sk_receive_queue); 1987 skb_queue_head_init(&sk->sk_write_queue); 1988 skb_queue_head_init(&sk->sk_error_queue); 1989 1990 rwlock_init(&sk->sk_callback_lock); 1991 lockdep_set_class_and_name(&sk->sk_receive_queue.lock, 1992 af_rlock_keys + sk->sk_family, 1993 af_family_rlock_key_strings[sk->sk_family]); 1994 lockdep_set_class_and_name(&sk->sk_write_queue.lock, 1995 af_wlock_keys + sk->sk_family, 1996 af_family_wlock_key_strings[sk->sk_family]); 1997 lockdep_set_class_and_name(&sk->sk_error_queue.lock, 1998 af_elock_keys + sk->sk_family, 1999 af_family_elock_key_strings[sk->sk_family]); 2000 lockdep_set_class_and_name(&sk->sk_callback_lock, 2001 af_callback_keys + sk->sk_family, 2002 af_family_clock_key_strings[sk->sk_family]); 2003 } 2004 2005 /** 2006 * sk_clone_lock - clone a socket, and lock its clone 2007 * @sk: the socket to clone 2008 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 2009 * 2010 * Caller must unlock socket even in error path (bh_unlock_sock(newsk)) 2011 */ 2012 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority) 2013 { 2014 struct proto *prot = READ_ONCE(sk->sk_prot); 2015 struct sk_filter *filter; 2016 bool is_charged = true; 2017 struct sock *newsk; 2018 2019 newsk = sk_prot_alloc(prot, priority, sk->sk_family); 2020 if (!newsk) 2021 goto out; 2022 2023 sock_copy(newsk, sk); 2024 2025 newsk->sk_prot_creator = prot; 2026 2027 /* SANITY */ 2028 if (likely(newsk->sk_net_refcnt)) 2029 get_net(sock_net(newsk)); 2030 sk_node_init(&newsk->sk_node); 2031 sock_lock_init(newsk); 2032 bh_lock_sock(newsk); 2033 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL; 2034 newsk->sk_backlog.len = 0; 2035 2036 atomic_set(&newsk->sk_rmem_alloc, 0); 2037 2038 /* sk_wmem_alloc set to one (see sk_free() and sock_wfree()) */ 2039 refcount_set(&newsk->sk_wmem_alloc, 1); 2040 2041 atomic_set(&newsk->sk_omem_alloc, 0); 2042 sk_init_common(newsk); 2043 2044 newsk->sk_dst_cache = NULL; 2045 newsk->sk_dst_pending_confirm = 0; 2046 newsk->sk_wmem_queued = 0; 2047 newsk->sk_forward_alloc = 0; 2048 atomic_set(&newsk->sk_drops, 0); 2049 newsk->sk_send_head = NULL; 2050 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK; 2051 atomic_set(&newsk->sk_zckey, 0); 2052 2053 sock_reset_flag(newsk, SOCK_DONE); 2054 2055 /* sk->sk_memcg will be populated at accept() time */ 2056 newsk->sk_memcg = NULL; 2057 2058 cgroup_sk_clone(&newsk->sk_cgrp_data); 2059 2060 rcu_read_lock(); 2061 filter = rcu_dereference(sk->sk_filter); 2062 if (filter != NULL) 2063 /* though it's an empty new sock, the charging may fail 2064 * if sysctl_optmem_max was changed between creation of 2065 * original socket and cloning 2066 */ 2067 is_charged = sk_filter_charge(newsk, filter); 2068 RCU_INIT_POINTER(newsk->sk_filter, filter); 2069 rcu_read_unlock(); 2070 2071 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) { 2072 /* We need to make sure that we don't uncharge the new 2073 * socket if we couldn't charge it in the first place 2074 * as otherwise we uncharge the parent's filter. 2075 */ 2076 if (!is_charged) 2077 RCU_INIT_POINTER(newsk->sk_filter, NULL); 2078 sk_free_unlock_clone(newsk); 2079 newsk = NULL; 2080 goto out; 2081 } 2082 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL); 2083 2084 if (bpf_sk_storage_clone(sk, newsk)) { 2085 sk_free_unlock_clone(newsk); 2086 newsk = NULL; 2087 goto out; 2088 } 2089 2090 /* Clear sk_user_data if parent had the pointer tagged 2091 * as not suitable for copying when cloning. 2092 */ 2093 if (sk_user_data_is_nocopy(newsk)) 2094 newsk->sk_user_data = NULL; 2095 2096 newsk->sk_err = 0; 2097 newsk->sk_err_soft = 0; 2098 newsk->sk_priority = 0; 2099 newsk->sk_incoming_cpu = raw_smp_processor_id(); 2100 if (likely(newsk->sk_net_refcnt)) 2101 sock_inuse_add(sock_net(newsk), 1); 2102 2103 /* Before updating sk_refcnt, we must commit prior changes to memory 2104 * (Documentation/RCU/rculist_nulls.rst for details) 2105 */ 2106 smp_wmb(); 2107 refcount_set(&newsk->sk_refcnt, 2); 2108 2109 /* Increment the counter in the same struct proto as the master 2110 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that 2111 * is the same as sk->sk_prot->socks, as this field was copied 2112 * with memcpy). 2113 * 2114 * This _changes_ the previous behaviour, where 2115 * tcp_create_openreq_child always was incrementing the 2116 * equivalent to tcp_prot->socks (inet_sock_nr), so this have 2117 * to be taken into account in all callers. -acme 2118 */ 2119 sk_refcnt_debug_inc(newsk); 2120 sk_set_socket(newsk, NULL); 2121 sk_tx_queue_clear(newsk); 2122 RCU_INIT_POINTER(newsk->sk_wq, NULL); 2123 2124 if (newsk->sk_prot->sockets_allocated) 2125 sk_sockets_allocated_inc(newsk); 2126 2127 if (sock_needs_netstamp(sk) && newsk->sk_flags & SK_FLAGS_TIMESTAMP) 2128 net_enable_timestamp(); 2129 out: 2130 return newsk; 2131 } 2132 EXPORT_SYMBOL_GPL(sk_clone_lock); 2133 2134 void sk_free_unlock_clone(struct sock *sk) 2135 { 2136 /* It is still raw copy of parent, so invalidate 2137 * destructor and make plain sk_free() */ 2138 sk->sk_destruct = NULL; 2139 bh_unlock_sock(sk); 2140 sk_free(sk); 2141 } 2142 EXPORT_SYMBOL_GPL(sk_free_unlock_clone); 2143 2144 void sk_setup_caps(struct sock *sk, struct dst_entry *dst) 2145 { 2146 u32 max_segs = 1; 2147 2148 sk_dst_set(sk, dst); 2149 sk->sk_route_caps = dst->dev->features | sk->sk_route_forced_caps; 2150 if (sk->sk_route_caps & NETIF_F_GSO) 2151 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE; 2152 sk->sk_route_caps &= ~sk->sk_route_nocaps; 2153 if (sk_can_gso(sk)) { 2154 if (dst->header_len && !xfrm_dst_offload_ok(dst)) { 2155 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 2156 } else { 2157 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM; 2158 sk->sk_gso_max_size = dst->dev->gso_max_size; 2159 max_segs = max_t(u32, dst->dev->gso_max_segs, 1); 2160 } 2161 } 2162 sk->sk_gso_max_segs = max_segs; 2163 } 2164 EXPORT_SYMBOL_GPL(sk_setup_caps); 2165 2166 /* 2167 * Simple resource managers for sockets. 2168 */ 2169 2170 2171 /* 2172 * Write buffer destructor automatically called from kfree_skb. 2173 */ 2174 void sock_wfree(struct sk_buff *skb) 2175 { 2176 struct sock *sk = skb->sk; 2177 unsigned int len = skb->truesize; 2178 2179 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) { 2180 /* 2181 * Keep a reference on sk_wmem_alloc, this will be released 2182 * after sk_write_space() call 2183 */ 2184 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc)); 2185 sk->sk_write_space(sk); 2186 len = 1; 2187 } 2188 /* 2189 * if sk_wmem_alloc reaches 0, we must finish what sk_free() 2190 * could not do because of in-flight packets 2191 */ 2192 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc)) 2193 __sk_free(sk); 2194 } 2195 EXPORT_SYMBOL(sock_wfree); 2196 2197 /* This variant of sock_wfree() is used by TCP, 2198 * since it sets SOCK_USE_WRITE_QUEUE. 2199 */ 2200 void __sock_wfree(struct sk_buff *skb) 2201 { 2202 struct sock *sk = skb->sk; 2203 2204 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc)) 2205 __sk_free(sk); 2206 } 2207 2208 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk) 2209 { 2210 skb_orphan(skb); 2211 skb->sk = sk; 2212 #ifdef CONFIG_INET 2213 if (unlikely(!sk_fullsock(sk))) { 2214 skb->destructor = sock_edemux; 2215 sock_hold(sk); 2216 return; 2217 } 2218 #endif 2219 skb->destructor = sock_wfree; 2220 skb_set_hash_from_sk(skb, sk); 2221 /* 2222 * We used to take a refcount on sk, but following operation 2223 * is enough to guarantee sk_free() wont free this sock until 2224 * all in-flight packets are completed 2225 */ 2226 refcount_add(skb->truesize, &sk->sk_wmem_alloc); 2227 } 2228 EXPORT_SYMBOL(skb_set_owner_w); 2229 2230 static bool can_skb_orphan_partial(const struct sk_buff *skb) 2231 { 2232 #ifdef CONFIG_TLS_DEVICE 2233 /* Drivers depend on in-order delivery for crypto offload, 2234 * partial orphan breaks out-of-order-OK logic. 2235 */ 2236 if (skb->decrypted) 2237 return false; 2238 #endif 2239 return (skb->destructor == sock_wfree || 2240 (IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree)); 2241 } 2242 2243 /* This helper is used by netem, as it can hold packets in its 2244 * delay queue. We want to allow the owner socket to send more 2245 * packets, as if they were already TX completed by a typical driver. 2246 * But we also want to keep skb->sk set because some packet schedulers 2247 * rely on it (sch_fq for example). 2248 */ 2249 void skb_orphan_partial(struct sk_buff *skb) 2250 { 2251 if (skb_is_tcp_pure_ack(skb)) 2252 return; 2253 2254 if (can_skb_orphan_partial(skb) && skb_set_owner_sk_safe(skb, skb->sk)) 2255 return; 2256 2257 skb_orphan(skb); 2258 } 2259 EXPORT_SYMBOL(skb_orphan_partial); 2260 2261 /* 2262 * Read buffer destructor automatically called from kfree_skb. 2263 */ 2264 void sock_rfree(struct sk_buff *skb) 2265 { 2266 struct sock *sk = skb->sk; 2267 unsigned int len = skb->truesize; 2268 2269 atomic_sub(len, &sk->sk_rmem_alloc); 2270 sk_mem_uncharge(sk, len); 2271 } 2272 EXPORT_SYMBOL(sock_rfree); 2273 2274 /* 2275 * Buffer destructor for skbs that are not used directly in read or write 2276 * path, e.g. for error handler skbs. Automatically called from kfree_skb. 2277 */ 2278 void sock_efree(struct sk_buff *skb) 2279 { 2280 sock_put(skb->sk); 2281 } 2282 EXPORT_SYMBOL(sock_efree); 2283 2284 /* Buffer destructor for prefetch/receive path where reference count may 2285 * not be held, e.g. for listen sockets. 2286 */ 2287 #ifdef CONFIG_INET 2288 void sock_pfree(struct sk_buff *skb) 2289 { 2290 if (sk_is_refcounted(skb->sk)) 2291 sock_gen_put(skb->sk); 2292 } 2293 EXPORT_SYMBOL(sock_pfree); 2294 #endif /* CONFIG_INET */ 2295 2296 kuid_t sock_i_uid(struct sock *sk) 2297 { 2298 kuid_t uid; 2299 2300 read_lock_bh(&sk->sk_callback_lock); 2301 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID; 2302 read_unlock_bh(&sk->sk_callback_lock); 2303 return uid; 2304 } 2305 EXPORT_SYMBOL(sock_i_uid); 2306 2307 unsigned long sock_i_ino(struct sock *sk) 2308 { 2309 unsigned long ino; 2310 2311 read_lock_bh(&sk->sk_callback_lock); 2312 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0; 2313 read_unlock_bh(&sk->sk_callback_lock); 2314 return ino; 2315 } 2316 EXPORT_SYMBOL(sock_i_ino); 2317 2318 /* 2319 * Allocate a skb from the socket's send buffer. 2320 */ 2321 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 2322 gfp_t priority) 2323 { 2324 if (force || 2325 refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) { 2326 struct sk_buff *skb = alloc_skb(size, priority); 2327 2328 if (skb) { 2329 skb_set_owner_w(skb, sk); 2330 return skb; 2331 } 2332 } 2333 return NULL; 2334 } 2335 EXPORT_SYMBOL(sock_wmalloc); 2336 2337 static void sock_ofree(struct sk_buff *skb) 2338 { 2339 struct sock *sk = skb->sk; 2340 2341 atomic_sub(skb->truesize, &sk->sk_omem_alloc); 2342 } 2343 2344 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size, 2345 gfp_t priority) 2346 { 2347 struct sk_buff *skb; 2348 2349 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */ 2350 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) > 2351 sysctl_optmem_max) 2352 return NULL; 2353 2354 skb = alloc_skb(size, priority); 2355 if (!skb) 2356 return NULL; 2357 2358 atomic_add(skb->truesize, &sk->sk_omem_alloc); 2359 skb->sk = sk; 2360 skb->destructor = sock_ofree; 2361 return skb; 2362 } 2363 2364 /* 2365 * Allocate a memory block from the socket's option memory buffer. 2366 */ 2367 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority) 2368 { 2369 if ((unsigned int)size <= sysctl_optmem_max && 2370 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) { 2371 void *mem; 2372 /* First do the add, to avoid the race if kmalloc 2373 * might sleep. 2374 */ 2375 atomic_add(size, &sk->sk_omem_alloc); 2376 mem = kmalloc(size, priority); 2377 if (mem) 2378 return mem; 2379 atomic_sub(size, &sk->sk_omem_alloc); 2380 } 2381 return NULL; 2382 } 2383 EXPORT_SYMBOL(sock_kmalloc); 2384 2385 /* Free an option memory block. Note, we actually want the inline 2386 * here as this allows gcc to detect the nullify and fold away the 2387 * condition entirely. 2388 */ 2389 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size, 2390 const bool nullify) 2391 { 2392 if (WARN_ON_ONCE(!mem)) 2393 return; 2394 if (nullify) 2395 kfree_sensitive(mem); 2396 else 2397 kfree(mem); 2398 atomic_sub(size, &sk->sk_omem_alloc); 2399 } 2400 2401 void sock_kfree_s(struct sock *sk, void *mem, int size) 2402 { 2403 __sock_kfree_s(sk, mem, size, false); 2404 } 2405 EXPORT_SYMBOL(sock_kfree_s); 2406 2407 void sock_kzfree_s(struct sock *sk, void *mem, int size) 2408 { 2409 __sock_kfree_s(sk, mem, size, true); 2410 } 2411 EXPORT_SYMBOL(sock_kzfree_s); 2412 2413 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock. 2414 I think, these locks should be removed for datagram sockets. 2415 */ 2416 static long sock_wait_for_wmem(struct sock *sk, long timeo) 2417 { 2418 DEFINE_WAIT(wait); 2419 2420 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk); 2421 for (;;) { 2422 if (!timeo) 2423 break; 2424 if (signal_pending(current)) 2425 break; 2426 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 2427 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 2428 if (refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) 2429 break; 2430 if (sk->sk_shutdown & SEND_SHUTDOWN) 2431 break; 2432 if (sk->sk_err) 2433 break; 2434 timeo = schedule_timeout(timeo); 2435 } 2436 finish_wait(sk_sleep(sk), &wait); 2437 return timeo; 2438 } 2439 2440 2441 /* 2442 * Generic send/receive buffer handlers 2443 */ 2444 2445 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 2446 unsigned long data_len, int noblock, 2447 int *errcode, int max_page_order) 2448 { 2449 struct sk_buff *skb; 2450 long timeo; 2451 int err; 2452 2453 timeo = sock_sndtimeo(sk, noblock); 2454 for (;;) { 2455 err = sock_error(sk); 2456 if (err != 0) 2457 goto failure; 2458 2459 err = -EPIPE; 2460 if (sk->sk_shutdown & SEND_SHUTDOWN) 2461 goto failure; 2462 2463 if (sk_wmem_alloc_get(sk) < READ_ONCE(sk->sk_sndbuf)) 2464 break; 2465 2466 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk); 2467 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 2468 err = -EAGAIN; 2469 if (!timeo) 2470 goto failure; 2471 if (signal_pending(current)) 2472 goto interrupted; 2473 timeo = sock_wait_for_wmem(sk, timeo); 2474 } 2475 skb = alloc_skb_with_frags(header_len, data_len, max_page_order, 2476 errcode, sk->sk_allocation); 2477 if (skb) 2478 skb_set_owner_w(skb, sk); 2479 return skb; 2480 2481 interrupted: 2482 err = sock_intr_errno(timeo); 2483 failure: 2484 *errcode = err; 2485 return NULL; 2486 } 2487 EXPORT_SYMBOL(sock_alloc_send_pskb); 2488 2489 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size, 2490 int noblock, int *errcode) 2491 { 2492 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0); 2493 } 2494 EXPORT_SYMBOL(sock_alloc_send_skb); 2495 2496 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg, 2497 struct sockcm_cookie *sockc) 2498 { 2499 u32 tsflags; 2500 2501 switch (cmsg->cmsg_type) { 2502 case SO_MARK: 2503 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) 2504 return -EPERM; 2505 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 2506 return -EINVAL; 2507 sockc->mark = *(u32 *)CMSG_DATA(cmsg); 2508 break; 2509 case SO_TIMESTAMPING_OLD: 2510 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 2511 return -EINVAL; 2512 2513 tsflags = *(u32 *)CMSG_DATA(cmsg); 2514 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK) 2515 return -EINVAL; 2516 2517 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK; 2518 sockc->tsflags |= tsflags; 2519 break; 2520 case SCM_TXTIME: 2521 if (!sock_flag(sk, SOCK_TXTIME)) 2522 return -EINVAL; 2523 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64))) 2524 return -EINVAL; 2525 sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg)); 2526 break; 2527 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */ 2528 case SCM_RIGHTS: 2529 case SCM_CREDENTIALS: 2530 break; 2531 default: 2532 return -EINVAL; 2533 } 2534 return 0; 2535 } 2536 EXPORT_SYMBOL(__sock_cmsg_send); 2537 2538 int sock_cmsg_send(struct sock *sk, struct msghdr *msg, 2539 struct sockcm_cookie *sockc) 2540 { 2541 struct cmsghdr *cmsg; 2542 int ret; 2543 2544 for_each_cmsghdr(cmsg, msg) { 2545 if (!CMSG_OK(msg, cmsg)) 2546 return -EINVAL; 2547 if (cmsg->cmsg_level != SOL_SOCKET) 2548 continue; 2549 ret = __sock_cmsg_send(sk, msg, cmsg, sockc); 2550 if (ret) 2551 return ret; 2552 } 2553 return 0; 2554 } 2555 EXPORT_SYMBOL(sock_cmsg_send); 2556 2557 static void sk_enter_memory_pressure(struct sock *sk) 2558 { 2559 if (!sk->sk_prot->enter_memory_pressure) 2560 return; 2561 2562 sk->sk_prot->enter_memory_pressure(sk); 2563 } 2564 2565 static void sk_leave_memory_pressure(struct sock *sk) 2566 { 2567 if (sk->sk_prot->leave_memory_pressure) { 2568 sk->sk_prot->leave_memory_pressure(sk); 2569 } else { 2570 unsigned long *memory_pressure = sk->sk_prot->memory_pressure; 2571 2572 if (memory_pressure && READ_ONCE(*memory_pressure)) 2573 WRITE_ONCE(*memory_pressure, 0); 2574 } 2575 } 2576 2577 #define SKB_FRAG_PAGE_ORDER get_order(32768) 2578 DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key); 2579 2580 /** 2581 * skb_page_frag_refill - check that a page_frag contains enough room 2582 * @sz: minimum size of the fragment we want to get 2583 * @pfrag: pointer to page_frag 2584 * @gfp: priority for memory allocation 2585 * 2586 * Note: While this allocator tries to use high order pages, there is 2587 * no guarantee that allocations succeed. Therefore, @sz MUST be 2588 * less or equal than PAGE_SIZE. 2589 */ 2590 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp) 2591 { 2592 if (pfrag->page) { 2593 if (page_ref_count(pfrag->page) == 1) { 2594 pfrag->offset = 0; 2595 return true; 2596 } 2597 if (pfrag->offset + sz <= pfrag->size) 2598 return true; 2599 put_page(pfrag->page); 2600 } 2601 2602 pfrag->offset = 0; 2603 if (SKB_FRAG_PAGE_ORDER && 2604 !static_branch_unlikely(&net_high_order_alloc_disable_key)) { 2605 /* Avoid direct reclaim but allow kswapd to wake */ 2606 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) | 2607 __GFP_COMP | __GFP_NOWARN | 2608 __GFP_NORETRY, 2609 SKB_FRAG_PAGE_ORDER); 2610 if (likely(pfrag->page)) { 2611 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER; 2612 return true; 2613 } 2614 } 2615 pfrag->page = alloc_page(gfp); 2616 if (likely(pfrag->page)) { 2617 pfrag->size = PAGE_SIZE; 2618 return true; 2619 } 2620 return false; 2621 } 2622 EXPORT_SYMBOL(skb_page_frag_refill); 2623 2624 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag) 2625 { 2626 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation))) 2627 return true; 2628 2629 sk_enter_memory_pressure(sk); 2630 sk_stream_moderate_sndbuf(sk); 2631 return false; 2632 } 2633 EXPORT_SYMBOL(sk_page_frag_refill); 2634 2635 void __lock_sock(struct sock *sk) 2636 __releases(&sk->sk_lock.slock) 2637 __acquires(&sk->sk_lock.slock) 2638 { 2639 DEFINE_WAIT(wait); 2640 2641 for (;;) { 2642 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait, 2643 TASK_UNINTERRUPTIBLE); 2644 spin_unlock_bh(&sk->sk_lock.slock); 2645 schedule(); 2646 spin_lock_bh(&sk->sk_lock.slock); 2647 if (!sock_owned_by_user(sk)) 2648 break; 2649 } 2650 finish_wait(&sk->sk_lock.wq, &wait); 2651 } 2652 2653 void __release_sock(struct sock *sk) 2654 __releases(&sk->sk_lock.slock) 2655 __acquires(&sk->sk_lock.slock) 2656 { 2657 struct sk_buff *skb, *next; 2658 2659 while ((skb = sk->sk_backlog.head) != NULL) { 2660 sk->sk_backlog.head = sk->sk_backlog.tail = NULL; 2661 2662 spin_unlock_bh(&sk->sk_lock.slock); 2663 2664 do { 2665 next = skb->next; 2666 prefetch(next); 2667 WARN_ON_ONCE(skb_dst_is_noref(skb)); 2668 skb_mark_not_on_list(skb); 2669 sk_backlog_rcv(sk, skb); 2670 2671 cond_resched(); 2672 2673 skb = next; 2674 } while (skb != NULL); 2675 2676 spin_lock_bh(&sk->sk_lock.slock); 2677 } 2678 2679 /* 2680 * Doing the zeroing here guarantee we can not loop forever 2681 * while a wild producer attempts to flood us. 2682 */ 2683 sk->sk_backlog.len = 0; 2684 } 2685 2686 void __sk_flush_backlog(struct sock *sk) 2687 { 2688 spin_lock_bh(&sk->sk_lock.slock); 2689 __release_sock(sk); 2690 spin_unlock_bh(&sk->sk_lock.slock); 2691 } 2692 2693 /** 2694 * sk_wait_data - wait for data to arrive at sk_receive_queue 2695 * @sk: sock to wait on 2696 * @timeo: for how long 2697 * @skb: last skb seen on sk_receive_queue 2698 * 2699 * Now socket state including sk->sk_err is changed only under lock, 2700 * hence we may omit checks after joining wait queue. 2701 * We check receive queue before schedule() only as optimization; 2702 * it is very likely that release_sock() added new data. 2703 */ 2704 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb) 2705 { 2706 DEFINE_WAIT_FUNC(wait, woken_wake_function); 2707 int rc; 2708 2709 add_wait_queue(sk_sleep(sk), &wait); 2710 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk); 2711 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait); 2712 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk); 2713 remove_wait_queue(sk_sleep(sk), &wait); 2714 return rc; 2715 } 2716 EXPORT_SYMBOL(sk_wait_data); 2717 2718 /** 2719 * __sk_mem_raise_allocated - increase memory_allocated 2720 * @sk: socket 2721 * @size: memory size to allocate 2722 * @amt: pages to allocate 2723 * @kind: allocation type 2724 * 2725 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc 2726 */ 2727 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind) 2728 { 2729 struct proto *prot = sk->sk_prot; 2730 long allocated = sk_memory_allocated_add(sk, amt); 2731 bool charged = true; 2732 2733 if (mem_cgroup_sockets_enabled && sk->sk_memcg && 2734 !(charged = mem_cgroup_charge_skmem(sk->sk_memcg, amt))) 2735 goto suppress_allocation; 2736 2737 /* Under limit. */ 2738 if (allocated <= sk_prot_mem_limits(sk, 0)) { 2739 sk_leave_memory_pressure(sk); 2740 return 1; 2741 } 2742 2743 /* Under pressure. */ 2744 if (allocated > sk_prot_mem_limits(sk, 1)) 2745 sk_enter_memory_pressure(sk); 2746 2747 /* Over hard limit. */ 2748 if (allocated > sk_prot_mem_limits(sk, 2)) 2749 goto suppress_allocation; 2750 2751 /* guarantee minimum buffer size under pressure */ 2752 if (kind == SK_MEM_RECV) { 2753 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot)) 2754 return 1; 2755 2756 } else { /* SK_MEM_SEND */ 2757 int wmem0 = sk_get_wmem0(sk, prot); 2758 2759 if (sk->sk_type == SOCK_STREAM) { 2760 if (sk->sk_wmem_queued < wmem0) 2761 return 1; 2762 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) { 2763 return 1; 2764 } 2765 } 2766 2767 if (sk_has_memory_pressure(sk)) { 2768 u64 alloc; 2769 2770 if (!sk_under_memory_pressure(sk)) 2771 return 1; 2772 alloc = sk_sockets_allocated_read_positive(sk); 2773 if (sk_prot_mem_limits(sk, 2) > alloc * 2774 sk_mem_pages(sk->sk_wmem_queued + 2775 atomic_read(&sk->sk_rmem_alloc) + 2776 sk->sk_forward_alloc)) 2777 return 1; 2778 } 2779 2780 suppress_allocation: 2781 2782 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) { 2783 sk_stream_moderate_sndbuf(sk); 2784 2785 /* Fail only if socket is _under_ its sndbuf. 2786 * In this case we cannot block, so that we have to fail. 2787 */ 2788 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf) 2789 return 1; 2790 } 2791 2792 if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged)) 2793 trace_sock_exceed_buf_limit(sk, prot, allocated, kind); 2794 2795 sk_memory_allocated_sub(sk, amt); 2796 2797 if (mem_cgroup_sockets_enabled && sk->sk_memcg) 2798 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt); 2799 2800 return 0; 2801 } 2802 EXPORT_SYMBOL(__sk_mem_raise_allocated); 2803 2804 /** 2805 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated 2806 * @sk: socket 2807 * @size: memory size to allocate 2808 * @kind: allocation type 2809 * 2810 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means 2811 * rmem allocation. This function assumes that protocols which have 2812 * memory_pressure use sk_wmem_queued as write buffer accounting. 2813 */ 2814 int __sk_mem_schedule(struct sock *sk, int size, int kind) 2815 { 2816 int ret, amt = sk_mem_pages(size); 2817 2818 sk->sk_forward_alloc += amt << SK_MEM_QUANTUM_SHIFT; 2819 ret = __sk_mem_raise_allocated(sk, size, amt, kind); 2820 if (!ret) 2821 sk->sk_forward_alloc -= amt << SK_MEM_QUANTUM_SHIFT; 2822 return ret; 2823 } 2824 EXPORT_SYMBOL(__sk_mem_schedule); 2825 2826 /** 2827 * __sk_mem_reduce_allocated - reclaim memory_allocated 2828 * @sk: socket 2829 * @amount: number of quanta 2830 * 2831 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc 2832 */ 2833 void __sk_mem_reduce_allocated(struct sock *sk, int amount) 2834 { 2835 sk_memory_allocated_sub(sk, amount); 2836 2837 if (mem_cgroup_sockets_enabled && sk->sk_memcg) 2838 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount); 2839 2840 if (sk_under_memory_pressure(sk) && 2841 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0))) 2842 sk_leave_memory_pressure(sk); 2843 } 2844 EXPORT_SYMBOL(__sk_mem_reduce_allocated); 2845 2846 /** 2847 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated 2848 * @sk: socket 2849 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple) 2850 */ 2851 void __sk_mem_reclaim(struct sock *sk, int amount) 2852 { 2853 amount >>= SK_MEM_QUANTUM_SHIFT; 2854 sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT; 2855 __sk_mem_reduce_allocated(sk, amount); 2856 } 2857 EXPORT_SYMBOL(__sk_mem_reclaim); 2858 2859 int sk_set_peek_off(struct sock *sk, int val) 2860 { 2861 sk->sk_peek_off = val; 2862 return 0; 2863 } 2864 EXPORT_SYMBOL_GPL(sk_set_peek_off); 2865 2866 /* 2867 * Set of default routines for initialising struct proto_ops when 2868 * the protocol does not support a particular function. In certain 2869 * cases where it makes no sense for a protocol to have a "do nothing" 2870 * function, some default processing is provided. 2871 */ 2872 2873 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len) 2874 { 2875 return -EOPNOTSUPP; 2876 } 2877 EXPORT_SYMBOL(sock_no_bind); 2878 2879 int sock_no_connect(struct socket *sock, struct sockaddr *saddr, 2880 int len, int flags) 2881 { 2882 return -EOPNOTSUPP; 2883 } 2884 EXPORT_SYMBOL(sock_no_connect); 2885 2886 int sock_no_socketpair(struct socket *sock1, struct socket *sock2) 2887 { 2888 return -EOPNOTSUPP; 2889 } 2890 EXPORT_SYMBOL(sock_no_socketpair); 2891 2892 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags, 2893 bool kern) 2894 { 2895 return -EOPNOTSUPP; 2896 } 2897 EXPORT_SYMBOL(sock_no_accept); 2898 2899 int sock_no_getname(struct socket *sock, struct sockaddr *saddr, 2900 int peer) 2901 { 2902 return -EOPNOTSUPP; 2903 } 2904 EXPORT_SYMBOL(sock_no_getname); 2905 2906 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 2907 { 2908 return -EOPNOTSUPP; 2909 } 2910 EXPORT_SYMBOL(sock_no_ioctl); 2911 2912 int sock_no_listen(struct socket *sock, int backlog) 2913 { 2914 return -EOPNOTSUPP; 2915 } 2916 EXPORT_SYMBOL(sock_no_listen); 2917 2918 int sock_no_shutdown(struct socket *sock, int how) 2919 { 2920 return -EOPNOTSUPP; 2921 } 2922 EXPORT_SYMBOL(sock_no_shutdown); 2923 2924 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len) 2925 { 2926 return -EOPNOTSUPP; 2927 } 2928 EXPORT_SYMBOL(sock_no_sendmsg); 2929 2930 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len) 2931 { 2932 return -EOPNOTSUPP; 2933 } 2934 EXPORT_SYMBOL(sock_no_sendmsg_locked); 2935 2936 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len, 2937 int flags) 2938 { 2939 return -EOPNOTSUPP; 2940 } 2941 EXPORT_SYMBOL(sock_no_recvmsg); 2942 2943 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma) 2944 { 2945 /* Mirror missing mmap method error code */ 2946 return -ENODEV; 2947 } 2948 EXPORT_SYMBOL(sock_no_mmap); 2949 2950 /* 2951 * When a file is received (via SCM_RIGHTS, etc), we must bump the 2952 * various sock-based usage counts. 2953 */ 2954 void __receive_sock(struct file *file) 2955 { 2956 struct socket *sock; 2957 2958 sock = sock_from_file(file); 2959 if (sock) { 2960 sock_update_netprioidx(&sock->sk->sk_cgrp_data); 2961 sock_update_classid(&sock->sk->sk_cgrp_data); 2962 } 2963 } 2964 2965 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags) 2966 { 2967 ssize_t res; 2968 struct msghdr msg = {.msg_flags = flags}; 2969 struct kvec iov; 2970 char *kaddr = kmap(page); 2971 iov.iov_base = kaddr + offset; 2972 iov.iov_len = size; 2973 res = kernel_sendmsg(sock, &msg, &iov, 1, size); 2974 kunmap(page); 2975 return res; 2976 } 2977 EXPORT_SYMBOL(sock_no_sendpage); 2978 2979 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page, 2980 int offset, size_t size, int flags) 2981 { 2982 ssize_t res; 2983 struct msghdr msg = {.msg_flags = flags}; 2984 struct kvec iov; 2985 char *kaddr = kmap(page); 2986 2987 iov.iov_base = kaddr + offset; 2988 iov.iov_len = size; 2989 res = kernel_sendmsg_locked(sk, &msg, &iov, 1, size); 2990 kunmap(page); 2991 return res; 2992 } 2993 EXPORT_SYMBOL(sock_no_sendpage_locked); 2994 2995 /* 2996 * Default Socket Callbacks 2997 */ 2998 2999 static void sock_def_wakeup(struct sock *sk) 3000 { 3001 struct socket_wq *wq; 3002 3003 rcu_read_lock(); 3004 wq = rcu_dereference(sk->sk_wq); 3005 if (skwq_has_sleeper(wq)) 3006 wake_up_interruptible_all(&wq->wait); 3007 rcu_read_unlock(); 3008 } 3009 3010 static void sock_def_error_report(struct sock *sk) 3011 { 3012 struct socket_wq *wq; 3013 3014 rcu_read_lock(); 3015 wq = rcu_dereference(sk->sk_wq); 3016 if (skwq_has_sleeper(wq)) 3017 wake_up_interruptible_poll(&wq->wait, EPOLLERR); 3018 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR); 3019 rcu_read_unlock(); 3020 } 3021 3022 void sock_def_readable(struct sock *sk) 3023 { 3024 struct socket_wq *wq; 3025 3026 rcu_read_lock(); 3027 wq = rcu_dereference(sk->sk_wq); 3028 if (skwq_has_sleeper(wq)) 3029 wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI | 3030 EPOLLRDNORM | EPOLLRDBAND); 3031 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN); 3032 rcu_read_unlock(); 3033 } 3034 3035 static void sock_def_write_space(struct sock *sk) 3036 { 3037 struct socket_wq *wq; 3038 3039 rcu_read_lock(); 3040 3041 /* Do not wake up a writer until he can make "significant" 3042 * progress. --DaveM 3043 */ 3044 if ((refcount_read(&sk->sk_wmem_alloc) << 1) <= READ_ONCE(sk->sk_sndbuf)) { 3045 wq = rcu_dereference(sk->sk_wq); 3046 if (skwq_has_sleeper(wq)) 3047 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT | 3048 EPOLLWRNORM | EPOLLWRBAND); 3049 3050 /* Should agree with poll, otherwise some programs break */ 3051 if (sock_writeable(sk)) 3052 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT); 3053 } 3054 3055 rcu_read_unlock(); 3056 } 3057 3058 static void sock_def_destruct(struct sock *sk) 3059 { 3060 } 3061 3062 void sk_send_sigurg(struct sock *sk) 3063 { 3064 if (sk->sk_socket && sk->sk_socket->file) 3065 if (send_sigurg(&sk->sk_socket->file->f_owner)) 3066 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI); 3067 } 3068 EXPORT_SYMBOL(sk_send_sigurg); 3069 3070 void sk_reset_timer(struct sock *sk, struct timer_list* timer, 3071 unsigned long expires) 3072 { 3073 if (!mod_timer(timer, expires)) 3074 sock_hold(sk); 3075 } 3076 EXPORT_SYMBOL(sk_reset_timer); 3077 3078 void sk_stop_timer(struct sock *sk, struct timer_list* timer) 3079 { 3080 if (del_timer(timer)) 3081 __sock_put(sk); 3082 } 3083 EXPORT_SYMBOL(sk_stop_timer); 3084 3085 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer) 3086 { 3087 if (del_timer_sync(timer)) 3088 __sock_put(sk); 3089 } 3090 EXPORT_SYMBOL(sk_stop_timer_sync); 3091 3092 void sock_init_data(struct socket *sock, struct sock *sk) 3093 { 3094 sk_init_common(sk); 3095 sk->sk_send_head = NULL; 3096 3097 timer_setup(&sk->sk_timer, NULL, 0); 3098 3099 sk->sk_allocation = GFP_KERNEL; 3100 sk->sk_rcvbuf = sysctl_rmem_default; 3101 sk->sk_sndbuf = sysctl_wmem_default; 3102 sk->sk_state = TCP_CLOSE; 3103 sk_set_socket(sk, sock); 3104 3105 sock_set_flag(sk, SOCK_ZAPPED); 3106 3107 if (sock) { 3108 sk->sk_type = sock->type; 3109 RCU_INIT_POINTER(sk->sk_wq, &sock->wq); 3110 sock->sk = sk; 3111 sk->sk_uid = SOCK_INODE(sock)->i_uid; 3112 } else { 3113 RCU_INIT_POINTER(sk->sk_wq, NULL); 3114 sk->sk_uid = make_kuid(sock_net(sk)->user_ns, 0); 3115 } 3116 3117 rwlock_init(&sk->sk_callback_lock); 3118 if (sk->sk_kern_sock) 3119 lockdep_set_class_and_name( 3120 &sk->sk_callback_lock, 3121 af_kern_callback_keys + sk->sk_family, 3122 af_family_kern_clock_key_strings[sk->sk_family]); 3123 else 3124 lockdep_set_class_and_name( 3125 &sk->sk_callback_lock, 3126 af_callback_keys + sk->sk_family, 3127 af_family_clock_key_strings[sk->sk_family]); 3128 3129 sk->sk_state_change = sock_def_wakeup; 3130 sk->sk_data_ready = sock_def_readable; 3131 sk->sk_write_space = sock_def_write_space; 3132 sk->sk_error_report = sock_def_error_report; 3133 sk->sk_destruct = sock_def_destruct; 3134 3135 sk->sk_frag.page = NULL; 3136 sk->sk_frag.offset = 0; 3137 sk->sk_peek_off = -1; 3138 3139 sk->sk_peer_pid = NULL; 3140 sk->sk_peer_cred = NULL; 3141 sk->sk_write_pending = 0; 3142 sk->sk_rcvlowat = 1; 3143 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; 3144 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT; 3145 3146 sk->sk_stamp = SK_DEFAULT_STAMP; 3147 #if BITS_PER_LONG==32 3148 seqlock_init(&sk->sk_stamp_seq); 3149 #endif 3150 atomic_set(&sk->sk_zckey, 0); 3151 3152 #ifdef CONFIG_NET_RX_BUSY_POLL 3153 sk->sk_napi_id = 0; 3154 sk->sk_ll_usec = sysctl_net_busy_read; 3155 #endif 3156 3157 sk->sk_max_pacing_rate = ~0UL; 3158 sk->sk_pacing_rate = ~0UL; 3159 WRITE_ONCE(sk->sk_pacing_shift, 10); 3160 sk->sk_incoming_cpu = -1; 3161 3162 sk_rx_queue_clear(sk); 3163 /* 3164 * Before updating sk_refcnt, we must commit prior changes to memory 3165 * (Documentation/RCU/rculist_nulls.rst for details) 3166 */ 3167 smp_wmb(); 3168 refcount_set(&sk->sk_refcnt, 1); 3169 atomic_set(&sk->sk_drops, 0); 3170 } 3171 EXPORT_SYMBOL(sock_init_data); 3172 3173 void lock_sock_nested(struct sock *sk, int subclass) 3174 { 3175 might_sleep(); 3176 spin_lock_bh(&sk->sk_lock.slock); 3177 if (sk->sk_lock.owned) 3178 __lock_sock(sk); 3179 sk->sk_lock.owned = 1; 3180 spin_unlock(&sk->sk_lock.slock); 3181 /* 3182 * The sk_lock has mutex_lock() semantics here: 3183 */ 3184 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_); 3185 local_bh_enable(); 3186 } 3187 EXPORT_SYMBOL(lock_sock_nested); 3188 3189 void release_sock(struct sock *sk) 3190 { 3191 spin_lock_bh(&sk->sk_lock.slock); 3192 if (sk->sk_backlog.tail) 3193 __release_sock(sk); 3194 3195 /* Warning : release_cb() might need to release sk ownership, 3196 * ie call sock_release_ownership(sk) before us. 3197 */ 3198 if (sk->sk_prot->release_cb) 3199 sk->sk_prot->release_cb(sk); 3200 3201 sock_release_ownership(sk); 3202 if (waitqueue_active(&sk->sk_lock.wq)) 3203 wake_up(&sk->sk_lock.wq); 3204 spin_unlock_bh(&sk->sk_lock.slock); 3205 } 3206 EXPORT_SYMBOL(release_sock); 3207 3208 /** 3209 * lock_sock_fast - fast version of lock_sock 3210 * @sk: socket 3211 * 3212 * This version should be used for very small section, where process wont block 3213 * return false if fast path is taken: 3214 * 3215 * sk_lock.slock locked, owned = 0, BH disabled 3216 * 3217 * return true if slow path is taken: 3218 * 3219 * sk_lock.slock unlocked, owned = 1, BH enabled 3220 */ 3221 bool lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock) 3222 { 3223 might_sleep(); 3224 spin_lock_bh(&sk->sk_lock.slock); 3225 3226 if (!sk->sk_lock.owned) 3227 /* 3228 * Note : We must disable BH 3229 */ 3230 return false; 3231 3232 __lock_sock(sk); 3233 sk->sk_lock.owned = 1; 3234 spin_unlock(&sk->sk_lock.slock); 3235 /* 3236 * The sk_lock has mutex_lock() semantics here: 3237 */ 3238 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_); 3239 __acquire(&sk->sk_lock.slock); 3240 local_bh_enable(); 3241 return true; 3242 } 3243 EXPORT_SYMBOL(lock_sock_fast); 3244 3245 int sock_gettstamp(struct socket *sock, void __user *userstamp, 3246 bool timeval, bool time32) 3247 { 3248 struct sock *sk = sock->sk; 3249 struct timespec64 ts; 3250 3251 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 3252 ts = ktime_to_timespec64(sock_read_timestamp(sk)); 3253 if (ts.tv_sec == -1) 3254 return -ENOENT; 3255 if (ts.tv_sec == 0) { 3256 ktime_t kt = ktime_get_real(); 3257 sock_write_timestamp(sk, kt); 3258 ts = ktime_to_timespec64(kt); 3259 } 3260 3261 if (timeval) 3262 ts.tv_nsec /= 1000; 3263 3264 #ifdef CONFIG_COMPAT_32BIT_TIME 3265 if (time32) 3266 return put_old_timespec32(&ts, userstamp); 3267 #endif 3268 #ifdef CONFIG_SPARC64 3269 /* beware of padding in sparc64 timeval */ 3270 if (timeval && !in_compat_syscall()) { 3271 struct __kernel_old_timeval __user tv = { 3272 .tv_sec = ts.tv_sec, 3273 .tv_usec = ts.tv_nsec, 3274 }; 3275 if (copy_to_user(userstamp, &tv, sizeof(tv))) 3276 return -EFAULT; 3277 return 0; 3278 } 3279 #endif 3280 return put_timespec64(&ts, userstamp); 3281 } 3282 EXPORT_SYMBOL(sock_gettstamp); 3283 3284 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag) 3285 { 3286 if (!sock_flag(sk, flag)) { 3287 unsigned long previous_flags = sk->sk_flags; 3288 3289 sock_set_flag(sk, flag); 3290 /* 3291 * we just set one of the two flags which require net 3292 * time stamping, but time stamping might have been on 3293 * already because of the other one 3294 */ 3295 if (sock_needs_netstamp(sk) && 3296 !(previous_flags & SK_FLAGS_TIMESTAMP)) 3297 net_enable_timestamp(); 3298 } 3299 } 3300 3301 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, 3302 int level, int type) 3303 { 3304 struct sock_exterr_skb *serr; 3305 struct sk_buff *skb; 3306 int copied, err; 3307 3308 err = -EAGAIN; 3309 skb = sock_dequeue_err_skb(sk); 3310 if (skb == NULL) 3311 goto out; 3312 3313 copied = skb->len; 3314 if (copied > len) { 3315 msg->msg_flags |= MSG_TRUNC; 3316 copied = len; 3317 } 3318 err = skb_copy_datagram_msg(skb, 0, msg, copied); 3319 if (err) 3320 goto out_free_skb; 3321 3322 sock_recv_timestamp(msg, sk, skb); 3323 3324 serr = SKB_EXT_ERR(skb); 3325 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee); 3326 3327 msg->msg_flags |= MSG_ERRQUEUE; 3328 err = copied; 3329 3330 out_free_skb: 3331 kfree_skb(skb); 3332 out: 3333 return err; 3334 } 3335 EXPORT_SYMBOL(sock_recv_errqueue); 3336 3337 /* 3338 * Get a socket option on an socket. 3339 * 3340 * FIX: POSIX 1003.1g is very ambiguous here. It states that 3341 * asynchronous errors should be reported by getsockopt. We assume 3342 * this means if you specify SO_ERROR (otherwise whats the point of it). 3343 */ 3344 int sock_common_getsockopt(struct socket *sock, int level, int optname, 3345 char __user *optval, int __user *optlen) 3346 { 3347 struct sock *sk = sock->sk; 3348 3349 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen); 3350 } 3351 EXPORT_SYMBOL(sock_common_getsockopt); 3352 3353 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 3354 int flags) 3355 { 3356 struct sock *sk = sock->sk; 3357 int addr_len = 0; 3358 int err; 3359 3360 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT, 3361 flags & ~MSG_DONTWAIT, &addr_len); 3362 if (err >= 0) 3363 msg->msg_namelen = addr_len; 3364 return err; 3365 } 3366 EXPORT_SYMBOL(sock_common_recvmsg); 3367 3368 /* 3369 * Set socket options on an inet socket. 3370 */ 3371 int sock_common_setsockopt(struct socket *sock, int level, int optname, 3372 sockptr_t optval, unsigned int optlen) 3373 { 3374 struct sock *sk = sock->sk; 3375 3376 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen); 3377 } 3378 EXPORT_SYMBOL(sock_common_setsockopt); 3379 3380 void sk_common_release(struct sock *sk) 3381 { 3382 if (sk->sk_prot->destroy) 3383 sk->sk_prot->destroy(sk); 3384 3385 /* 3386 * Observation: when sk_common_release is called, processes have 3387 * no access to socket. But net still has. 3388 * Step one, detach it from networking: 3389 * 3390 * A. Remove from hash tables. 3391 */ 3392 3393 sk->sk_prot->unhash(sk); 3394 3395 /* 3396 * In this point socket cannot receive new packets, but it is possible 3397 * that some packets are in flight because some CPU runs receiver and 3398 * did hash table lookup before we unhashed socket. They will achieve 3399 * receive queue and will be purged by socket destructor. 3400 * 3401 * Also we still have packets pending on receive queue and probably, 3402 * our own packets waiting in device queues. sock_destroy will drain 3403 * receive queue, but transmitted packets will delay socket destruction 3404 * until the last reference will be released. 3405 */ 3406 3407 sock_orphan(sk); 3408 3409 xfrm_sk_free_policy(sk); 3410 3411 sk_refcnt_debug_release(sk); 3412 3413 sock_put(sk); 3414 } 3415 EXPORT_SYMBOL(sk_common_release); 3416 3417 void sk_get_meminfo(const struct sock *sk, u32 *mem) 3418 { 3419 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS); 3420 3421 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk); 3422 mem[SK_MEMINFO_RCVBUF] = READ_ONCE(sk->sk_rcvbuf); 3423 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk); 3424 mem[SK_MEMINFO_SNDBUF] = READ_ONCE(sk->sk_sndbuf); 3425 mem[SK_MEMINFO_FWD_ALLOC] = sk->sk_forward_alloc; 3426 mem[SK_MEMINFO_WMEM_QUEUED] = READ_ONCE(sk->sk_wmem_queued); 3427 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc); 3428 mem[SK_MEMINFO_BACKLOG] = READ_ONCE(sk->sk_backlog.len); 3429 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops); 3430 } 3431 3432 #ifdef CONFIG_PROC_FS 3433 #define PROTO_INUSE_NR 64 /* should be enough for the first time */ 3434 struct prot_inuse { 3435 int val[PROTO_INUSE_NR]; 3436 }; 3437 3438 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR); 3439 3440 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val) 3441 { 3442 __this_cpu_add(net->core.prot_inuse->val[prot->inuse_idx], val); 3443 } 3444 EXPORT_SYMBOL_GPL(sock_prot_inuse_add); 3445 3446 int sock_prot_inuse_get(struct net *net, struct proto *prot) 3447 { 3448 int cpu, idx = prot->inuse_idx; 3449 int res = 0; 3450 3451 for_each_possible_cpu(cpu) 3452 res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx]; 3453 3454 return res >= 0 ? res : 0; 3455 } 3456 EXPORT_SYMBOL_GPL(sock_prot_inuse_get); 3457 3458 static void sock_inuse_add(struct net *net, int val) 3459 { 3460 this_cpu_add(*net->core.sock_inuse, val); 3461 } 3462 3463 int sock_inuse_get(struct net *net) 3464 { 3465 int cpu, res = 0; 3466 3467 for_each_possible_cpu(cpu) 3468 res += *per_cpu_ptr(net->core.sock_inuse, cpu); 3469 3470 return res; 3471 } 3472 3473 EXPORT_SYMBOL_GPL(sock_inuse_get); 3474 3475 static int __net_init sock_inuse_init_net(struct net *net) 3476 { 3477 net->core.prot_inuse = alloc_percpu(struct prot_inuse); 3478 if (net->core.prot_inuse == NULL) 3479 return -ENOMEM; 3480 3481 net->core.sock_inuse = alloc_percpu(int); 3482 if (net->core.sock_inuse == NULL) 3483 goto out; 3484 3485 return 0; 3486 3487 out: 3488 free_percpu(net->core.prot_inuse); 3489 return -ENOMEM; 3490 } 3491 3492 static void __net_exit sock_inuse_exit_net(struct net *net) 3493 { 3494 free_percpu(net->core.prot_inuse); 3495 free_percpu(net->core.sock_inuse); 3496 } 3497 3498 static struct pernet_operations net_inuse_ops = { 3499 .init = sock_inuse_init_net, 3500 .exit = sock_inuse_exit_net, 3501 }; 3502 3503 static __init int net_inuse_init(void) 3504 { 3505 if (register_pernet_subsys(&net_inuse_ops)) 3506 panic("Cannot initialize net inuse counters"); 3507 3508 return 0; 3509 } 3510 3511 core_initcall(net_inuse_init); 3512 3513 static int assign_proto_idx(struct proto *prot) 3514 { 3515 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR); 3516 3517 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) { 3518 pr_err("PROTO_INUSE_NR exhausted\n"); 3519 return -ENOSPC; 3520 } 3521 3522 set_bit(prot->inuse_idx, proto_inuse_idx); 3523 return 0; 3524 } 3525 3526 static void release_proto_idx(struct proto *prot) 3527 { 3528 if (prot->inuse_idx != PROTO_INUSE_NR - 1) 3529 clear_bit(prot->inuse_idx, proto_inuse_idx); 3530 } 3531 #else 3532 static inline int assign_proto_idx(struct proto *prot) 3533 { 3534 return 0; 3535 } 3536 3537 static inline void release_proto_idx(struct proto *prot) 3538 { 3539 } 3540 3541 static void sock_inuse_add(struct net *net, int val) 3542 { 3543 } 3544 #endif 3545 3546 static void tw_prot_cleanup(struct timewait_sock_ops *twsk_prot) 3547 { 3548 if (!twsk_prot) 3549 return; 3550 kfree(twsk_prot->twsk_slab_name); 3551 twsk_prot->twsk_slab_name = NULL; 3552 kmem_cache_destroy(twsk_prot->twsk_slab); 3553 twsk_prot->twsk_slab = NULL; 3554 } 3555 3556 static int tw_prot_init(const struct proto *prot) 3557 { 3558 struct timewait_sock_ops *twsk_prot = prot->twsk_prot; 3559 3560 if (!twsk_prot) 3561 return 0; 3562 3563 twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", 3564 prot->name); 3565 if (!twsk_prot->twsk_slab_name) 3566 return -ENOMEM; 3567 3568 twsk_prot->twsk_slab = 3569 kmem_cache_create(twsk_prot->twsk_slab_name, 3570 twsk_prot->twsk_obj_size, 0, 3571 SLAB_ACCOUNT | prot->slab_flags, 3572 NULL); 3573 if (!twsk_prot->twsk_slab) { 3574 pr_crit("%s: Can't create timewait sock SLAB cache!\n", 3575 prot->name); 3576 return -ENOMEM; 3577 } 3578 3579 return 0; 3580 } 3581 3582 static void req_prot_cleanup(struct request_sock_ops *rsk_prot) 3583 { 3584 if (!rsk_prot) 3585 return; 3586 kfree(rsk_prot->slab_name); 3587 rsk_prot->slab_name = NULL; 3588 kmem_cache_destroy(rsk_prot->slab); 3589 rsk_prot->slab = NULL; 3590 } 3591 3592 static int req_prot_init(const struct proto *prot) 3593 { 3594 struct request_sock_ops *rsk_prot = prot->rsk_prot; 3595 3596 if (!rsk_prot) 3597 return 0; 3598 3599 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", 3600 prot->name); 3601 if (!rsk_prot->slab_name) 3602 return -ENOMEM; 3603 3604 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name, 3605 rsk_prot->obj_size, 0, 3606 SLAB_ACCOUNT | prot->slab_flags, 3607 NULL); 3608 3609 if (!rsk_prot->slab) { 3610 pr_crit("%s: Can't create request sock SLAB cache!\n", 3611 prot->name); 3612 return -ENOMEM; 3613 } 3614 return 0; 3615 } 3616 3617 int proto_register(struct proto *prot, int alloc_slab) 3618 { 3619 int ret = -ENOBUFS; 3620 3621 if (alloc_slab) { 3622 prot->slab = kmem_cache_create_usercopy(prot->name, 3623 prot->obj_size, 0, 3624 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT | 3625 prot->slab_flags, 3626 prot->useroffset, prot->usersize, 3627 NULL); 3628 3629 if (prot->slab == NULL) { 3630 pr_crit("%s: Can't create sock SLAB cache!\n", 3631 prot->name); 3632 goto out; 3633 } 3634 3635 if (req_prot_init(prot)) 3636 goto out_free_request_sock_slab; 3637 3638 if (tw_prot_init(prot)) 3639 goto out_free_timewait_sock_slab; 3640 } 3641 3642 mutex_lock(&proto_list_mutex); 3643 ret = assign_proto_idx(prot); 3644 if (ret) { 3645 mutex_unlock(&proto_list_mutex); 3646 goto out_free_timewait_sock_slab; 3647 } 3648 list_add(&prot->node, &proto_list); 3649 mutex_unlock(&proto_list_mutex); 3650 return ret; 3651 3652 out_free_timewait_sock_slab: 3653 if (alloc_slab) 3654 tw_prot_cleanup(prot->twsk_prot); 3655 out_free_request_sock_slab: 3656 if (alloc_slab) { 3657 req_prot_cleanup(prot->rsk_prot); 3658 3659 kmem_cache_destroy(prot->slab); 3660 prot->slab = NULL; 3661 } 3662 out: 3663 return ret; 3664 } 3665 EXPORT_SYMBOL(proto_register); 3666 3667 void proto_unregister(struct proto *prot) 3668 { 3669 mutex_lock(&proto_list_mutex); 3670 release_proto_idx(prot); 3671 list_del(&prot->node); 3672 mutex_unlock(&proto_list_mutex); 3673 3674 kmem_cache_destroy(prot->slab); 3675 prot->slab = NULL; 3676 3677 req_prot_cleanup(prot->rsk_prot); 3678 tw_prot_cleanup(prot->twsk_prot); 3679 } 3680 EXPORT_SYMBOL(proto_unregister); 3681 3682 int sock_load_diag_module(int family, int protocol) 3683 { 3684 if (!protocol) { 3685 if (!sock_is_registered(family)) 3686 return -ENOENT; 3687 3688 return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK, 3689 NETLINK_SOCK_DIAG, family); 3690 } 3691 3692 #ifdef CONFIG_INET 3693 if (family == AF_INET && 3694 protocol != IPPROTO_RAW && 3695 protocol < MAX_INET_PROTOS && 3696 !rcu_access_pointer(inet_protos[protocol])) 3697 return -ENOENT; 3698 #endif 3699 3700 return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK, 3701 NETLINK_SOCK_DIAG, family, protocol); 3702 } 3703 EXPORT_SYMBOL(sock_load_diag_module); 3704 3705 #ifdef CONFIG_PROC_FS 3706 static void *proto_seq_start(struct seq_file *seq, loff_t *pos) 3707 __acquires(proto_list_mutex) 3708 { 3709 mutex_lock(&proto_list_mutex); 3710 return seq_list_start_head(&proto_list, *pos); 3711 } 3712 3713 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos) 3714 { 3715 return seq_list_next(v, &proto_list, pos); 3716 } 3717 3718 static void proto_seq_stop(struct seq_file *seq, void *v) 3719 __releases(proto_list_mutex) 3720 { 3721 mutex_unlock(&proto_list_mutex); 3722 } 3723 3724 static char proto_method_implemented(const void *method) 3725 { 3726 return method == NULL ? 'n' : 'y'; 3727 } 3728 static long sock_prot_memory_allocated(struct proto *proto) 3729 { 3730 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L; 3731 } 3732 3733 static const char *sock_prot_memory_pressure(struct proto *proto) 3734 { 3735 return proto->memory_pressure != NULL ? 3736 proto_memory_pressure(proto) ? "yes" : "no" : "NI"; 3737 } 3738 3739 static void proto_seq_printf(struct seq_file *seq, struct proto *proto) 3740 { 3741 3742 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s " 3743 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n", 3744 proto->name, 3745 proto->obj_size, 3746 sock_prot_inuse_get(seq_file_net(seq), proto), 3747 sock_prot_memory_allocated(proto), 3748 sock_prot_memory_pressure(proto), 3749 proto->max_header, 3750 proto->slab == NULL ? "no" : "yes", 3751 module_name(proto->owner), 3752 proto_method_implemented(proto->close), 3753 proto_method_implemented(proto->connect), 3754 proto_method_implemented(proto->disconnect), 3755 proto_method_implemented(proto->accept), 3756 proto_method_implemented(proto->ioctl), 3757 proto_method_implemented(proto->init), 3758 proto_method_implemented(proto->destroy), 3759 proto_method_implemented(proto->shutdown), 3760 proto_method_implemented(proto->setsockopt), 3761 proto_method_implemented(proto->getsockopt), 3762 proto_method_implemented(proto->sendmsg), 3763 proto_method_implemented(proto->recvmsg), 3764 proto_method_implemented(proto->sendpage), 3765 proto_method_implemented(proto->bind), 3766 proto_method_implemented(proto->backlog_rcv), 3767 proto_method_implemented(proto->hash), 3768 proto_method_implemented(proto->unhash), 3769 proto_method_implemented(proto->get_port), 3770 proto_method_implemented(proto->enter_memory_pressure)); 3771 } 3772 3773 static int proto_seq_show(struct seq_file *seq, void *v) 3774 { 3775 if (v == &proto_list) 3776 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s", 3777 "protocol", 3778 "size", 3779 "sockets", 3780 "memory", 3781 "press", 3782 "maxhdr", 3783 "slab", 3784 "module", 3785 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n"); 3786 else 3787 proto_seq_printf(seq, list_entry(v, struct proto, node)); 3788 return 0; 3789 } 3790 3791 static const struct seq_operations proto_seq_ops = { 3792 .start = proto_seq_start, 3793 .next = proto_seq_next, 3794 .stop = proto_seq_stop, 3795 .show = proto_seq_show, 3796 }; 3797 3798 static __net_init int proto_init_net(struct net *net) 3799 { 3800 if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops, 3801 sizeof(struct seq_net_private))) 3802 return -ENOMEM; 3803 3804 return 0; 3805 } 3806 3807 static __net_exit void proto_exit_net(struct net *net) 3808 { 3809 remove_proc_entry("protocols", net->proc_net); 3810 } 3811 3812 3813 static __net_initdata struct pernet_operations proto_net_ops = { 3814 .init = proto_init_net, 3815 .exit = proto_exit_net, 3816 }; 3817 3818 static int __init proto_init(void) 3819 { 3820 return register_pernet_subsys(&proto_net_ops); 3821 } 3822 3823 subsys_initcall(proto_init); 3824 3825 #endif /* PROC_FS */ 3826 3827 #ifdef CONFIG_NET_RX_BUSY_POLL 3828 bool sk_busy_loop_end(void *p, unsigned long start_time) 3829 { 3830 struct sock *sk = p; 3831 3832 return !skb_queue_empty_lockless(&sk->sk_receive_queue) || 3833 sk_busy_loop_timeout(sk, start_time); 3834 } 3835 EXPORT_SYMBOL(sk_busy_loop_end); 3836 #endif /* CONFIG_NET_RX_BUSY_POLL */ 3837 3838 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len) 3839 { 3840 if (!sk->sk_prot->bind_add) 3841 return -EOPNOTSUPP; 3842 return sk->sk_prot->bind_add(sk, addr, addr_len); 3843 } 3844 EXPORT_SYMBOL(sock_bind_add); 3845