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 #include "dev.h" 145 146 static DEFINE_MUTEX(proto_list_mutex); 147 static LIST_HEAD(proto_list); 148 149 static void sock_def_write_space_wfree(struct sock *sk); 150 static void sock_def_write_space(struct sock *sk); 151 152 /** 153 * sk_ns_capable - General socket capability test 154 * @sk: Socket to use a capability on or through 155 * @user_ns: The user namespace of the capability to use 156 * @cap: The capability to use 157 * 158 * Test to see if the opener of the socket had when the socket was 159 * created and the current process has the capability @cap in the user 160 * namespace @user_ns. 161 */ 162 bool sk_ns_capable(const struct sock *sk, 163 struct user_namespace *user_ns, int cap) 164 { 165 return file_ns_capable(sk->sk_socket->file, user_ns, cap) && 166 ns_capable(user_ns, cap); 167 } 168 EXPORT_SYMBOL(sk_ns_capable); 169 170 /** 171 * sk_capable - Socket global capability test 172 * @sk: Socket to use a capability on or through 173 * @cap: The global capability to use 174 * 175 * Test to see if the opener of the socket had when the socket was 176 * created and the current process has the capability @cap in all user 177 * namespaces. 178 */ 179 bool sk_capable(const struct sock *sk, int cap) 180 { 181 return sk_ns_capable(sk, &init_user_ns, cap); 182 } 183 EXPORT_SYMBOL(sk_capable); 184 185 /** 186 * sk_net_capable - Network namespace socket capability test 187 * @sk: Socket to use a capability on or through 188 * @cap: The capability to use 189 * 190 * Test to see if the opener of the socket had when the socket was created 191 * and the current process has the capability @cap over the network namespace 192 * the socket is a member of. 193 */ 194 bool sk_net_capable(const struct sock *sk, int cap) 195 { 196 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap); 197 } 198 EXPORT_SYMBOL(sk_net_capable); 199 200 /* 201 * Each address family might have different locking rules, so we have 202 * one slock key per address family and separate keys for internal and 203 * userspace sockets. 204 */ 205 static struct lock_class_key af_family_keys[AF_MAX]; 206 static struct lock_class_key af_family_kern_keys[AF_MAX]; 207 static struct lock_class_key af_family_slock_keys[AF_MAX]; 208 static struct lock_class_key af_family_kern_slock_keys[AF_MAX]; 209 210 /* 211 * Make lock validator output more readable. (we pre-construct these 212 * strings build-time, so that runtime initialization of socket 213 * locks is fast): 214 */ 215 216 #define _sock_locks(x) \ 217 x "AF_UNSPEC", x "AF_UNIX" , x "AF_INET" , \ 218 x "AF_AX25" , x "AF_IPX" , x "AF_APPLETALK", \ 219 x "AF_NETROM", x "AF_BRIDGE" , x "AF_ATMPVC" , \ 220 x "AF_X25" , x "AF_INET6" , x "AF_ROSE" , \ 221 x "AF_DECnet", x "AF_NETBEUI" , x "AF_SECURITY" , \ 222 x "AF_KEY" , x "AF_NETLINK" , x "AF_PACKET" , \ 223 x "AF_ASH" , x "AF_ECONET" , x "AF_ATMSVC" , \ 224 x "AF_RDS" , x "AF_SNA" , x "AF_IRDA" , \ 225 x "AF_PPPOX" , x "AF_WANPIPE" , x "AF_LLC" , \ 226 x "27" , x "28" , x "AF_CAN" , \ 227 x "AF_TIPC" , x "AF_BLUETOOTH", x "IUCV" , \ 228 x "AF_RXRPC" , x "AF_ISDN" , x "AF_PHONET" , \ 229 x "AF_IEEE802154", x "AF_CAIF" , x "AF_ALG" , \ 230 x "AF_NFC" , x "AF_VSOCK" , x "AF_KCM" , \ 231 x "AF_QIPCRTR", x "AF_SMC" , x "AF_XDP" , \ 232 x "AF_MCTP" , \ 233 x "AF_MAX" 234 235 static const char *const af_family_key_strings[AF_MAX+1] = { 236 _sock_locks("sk_lock-") 237 }; 238 static const char *const af_family_slock_key_strings[AF_MAX+1] = { 239 _sock_locks("slock-") 240 }; 241 static const char *const af_family_clock_key_strings[AF_MAX+1] = { 242 _sock_locks("clock-") 243 }; 244 245 static const char *const af_family_kern_key_strings[AF_MAX+1] = { 246 _sock_locks("k-sk_lock-") 247 }; 248 static const char *const af_family_kern_slock_key_strings[AF_MAX+1] = { 249 _sock_locks("k-slock-") 250 }; 251 static const char *const af_family_kern_clock_key_strings[AF_MAX+1] = { 252 _sock_locks("k-clock-") 253 }; 254 static const char *const af_family_rlock_key_strings[AF_MAX+1] = { 255 _sock_locks("rlock-") 256 }; 257 static const char *const af_family_wlock_key_strings[AF_MAX+1] = { 258 _sock_locks("wlock-") 259 }; 260 static const char *const af_family_elock_key_strings[AF_MAX+1] = { 261 _sock_locks("elock-") 262 }; 263 264 /* 265 * sk_callback_lock and sk queues locking rules are per-address-family, 266 * so split the lock classes by using a per-AF key: 267 */ 268 static struct lock_class_key af_callback_keys[AF_MAX]; 269 static struct lock_class_key af_rlock_keys[AF_MAX]; 270 static struct lock_class_key af_wlock_keys[AF_MAX]; 271 static struct lock_class_key af_elock_keys[AF_MAX]; 272 static struct lock_class_key af_kern_callback_keys[AF_MAX]; 273 274 /* Run time adjustable parameters. */ 275 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX; 276 EXPORT_SYMBOL(sysctl_wmem_max); 277 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX; 278 EXPORT_SYMBOL(sysctl_rmem_max); 279 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX; 280 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX; 281 282 /* Maximal space eaten by iovec or ancillary data plus some space */ 283 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512); 284 EXPORT_SYMBOL(sysctl_optmem_max); 285 286 int sysctl_tstamp_allow_data __read_mostly = 1; 287 288 DEFINE_STATIC_KEY_FALSE(memalloc_socks_key); 289 EXPORT_SYMBOL_GPL(memalloc_socks_key); 290 291 /** 292 * sk_set_memalloc - sets %SOCK_MEMALLOC 293 * @sk: socket to set it on 294 * 295 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves. 296 * It's the responsibility of the admin to adjust min_free_kbytes 297 * to meet the requirements 298 */ 299 void sk_set_memalloc(struct sock *sk) 300 { 301 sock_set_flag(sk, SOCK_MEMALLOC); 302 sk->sk_allocation |= __GFP_MEMALLOC; 303 static_branch_inc(&memalloc_socks_key); 304 } 305 EXPORT_SYMBOL_GPL(sk_set_memalloc); 306 307 void sk_clear_memalloc(struct sock *sk) 308 { 309 sock_reset_flag(sk, SOCK_MEMALLOC); 310 sk->sk_allocation &= ~__GFP_MEMALLOC; 311 static_branch_dec(&memalloc_socks_key); 312 313 /* 314 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward 315 * progress of swapping. SOCK_MEMALLOC may be cleared while 316 * it has rmem allocations due to the last swapfile being deactivated 317 * but there is a risk that the socket is unusable due to exceeding 318 * the rmem limits. Reclaim the reserves and obey rmem limits again. 319 */ 320 sk_mem_reclaim(sk); 321 } 322 EXPORT_SYMBOL_GPL(sk_clear_memalloc); 323 324 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb) 325 { 326 int ret; 327 unsigned int noreclaim_flag; 328 329 /* these should have been dropped before queueing */ 330 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC)); 331 332 noreclaim_flag = memalloc_noreclaim_save(); 333 ret = INDIRECT_CALL_INET(sk->sk_backlog_rcv, 334 tcp_v6_do_rcv, 335 tcp_v4_do_rcv, 336 sk, skb); 337 memalloc_noreclaim_restore(noreclaim_flag); 338 339 return ret; 340 } 341 EXPORT_SYMBOL(__sk_backlog_rcv); 342 343 void sk_error_report(struct sock *sk) 344 { 345 sk->sk_error_report(sk); 346 347 switch (sk->sk_family) { 348 case AF_INET: 349 fallthrough; 350 case AF_INET6: 351 trace_inet_sk_error_report(sk); 352 break; 353 default: 354 break; 355 } 356 } 357 EXPORT_SYMBOL(sk_error_report); 358 359 int sock_get_timeout(long timeo, void *optval, bool old_timeval) 360 { 361 struct __kernel_sock_timeval tv; 362 363 if (timeo == MAX_SCHEDULE_TIMEOUT) { 364 tv.tv_sec = 0; 365 tv.tv_usec = 0; 366 } else { 367 tv.tv_sec = timeo / HZ; 368 tv.tv_usec = ((timeo % HZ) * USEC_PER_SEC) / HZ; 369 } 370 371 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) { 372 struct old_timeval32 tv32 = { tv.tv_sec, tv.tv_usec }; 373 *(struct old_timeval32 *)optval = tv32; 374 return sizeof(tv32); 375 } 376 377 if (old_timeval) { 378 struct __kernel_old_timeval old_tv; 379 old_tv.tv_sec = tv.tv_sec; 380 old_tv.tv_usec = tv.tv_usec; 381 *(struct __kernel_old_timeval *)optval = old_tv; 382 return sizeof(old_tv); 383 } 384 385 *(struct __kernel_sock_timeval *)optval = tv; 386 return sizeof(tv); 387 } 388 EXPORT_SYMBOL(sock_get_timeout); 389 390 int sock_copy_user_timeval(struct __kernel_sock_timeval *tv, 391 sockptr_t optval, int optlen, bool old_timeval) 392 { 393 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) { 394 struct old_timeval32 tv32; 395 396 if (optlen < sizeof(tv32)) 397 return -EINVAL; 398 399 if (copy_from_sockptr(&tv32, optval, sizeof(tv32))) 400 return -EFAULT; 401 tv->tv_sec = tv32.tv_sec; 402 tv->tv_usec = tv32.tv_usec; 403 } else if (old_timeval) { 404 struct __kernel_old_timeval old_tv; 405 406 if (optlen < sizeof(old_tv)) 407 return -EINVAL; 408 if (copy_from_sockptr(&old_tv, optval, sizeof(old_tv))) 409 return -EFAULT; 410 tv->tv_sec = old_tv.tv_sec; 411 tv->tv_usec = old_tv.tv_usec; 412 } else { 413 if (optlen < sizeof(*tv)) 414 return -EINVAL; 415 if (copy_from_sockptr(tv, optval, sizeof(*tv))) 416 return -EFAULT; 417 } 418 419 return 0; 420 } 421 EXPORT_SYMBOL(sock_copy_user_timeval); 422 423 static int sock_set_timeout(long *timeo_p, sockptr_t optval, int optlen, 424 bool old_timeval) 425 { 426 struct __kernel_sock_timeval tv; 427 int err = sock_copy_user_timeval(&tv, optval, optlen, old_timeval); 428 429 if (err) 430 return err; 431 432 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC) 433 return -EDOM; 434 435 if (tv.tv_sec < 0) { 436 static int warned __read_mostly; 437 438 *timeo_p = 0; 439 if (warned < 10 && net_ratelimit()) { 440 warned++; 441 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n", 442 __func__, current->comm, task_pid_nr(current)); 443 } 444 return 0; 445 } 446 *timeo_p = MAX_SCHEDULE_TIMEOUT; 447 if (tv.tv_sec == 0 && tv.tv_usec == 0) 448 return 0; 449 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1)) 450 *timeo_p = tv.tv_sec * HZ + DIV_ROUND_UP((unsigned long)tv.tv_usec, USEC_PER_SEC / HZ); 451 return 0; 452 } 453 454 static bool sock_needs_netstamp(const struct sock *sk) 455 { 456 switch (sk->sk_family) { 457 case AF_UNSPEC: 458 case AF_UNIX: 459 return false; 460 default: 461 return true; 462 } 463 } 464 465 static void sock_disable_timestamp(struct sock *sk, unsigned long flags) 466 { 467 if (sk->sk_flags & flags) { 468 sk->sk_flags &= ~flags; 469 if (sock_needs_netstamp(sk) && 470 !(sk->sk_flags & SK_FLAGS_TIMESTAMP)) 471 net_disable_timestamp(); 472 } 473 } 474 475 476 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 477 { 478 unsigned long flags; 479 struct sk_buff_head *list = &sk->sk_receive_queue; 480 481 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) { 482 atomic_inc(&sk->sk_drops); 483 trace_sock_rcvqueue_full(sk, skb); 484 return -ENOMEM; 485 } 486 487 if (!sk_rmem_schedule(sk, skb, skb->truesize)) { 488 atomic_inc(&sk->sk_drops); 489 return -ENOBUFS; 490 } 491 492 skb->dev = NULL; 493 skb_set_owner_r(skb, sk); 494 495 /* we escape from rcu protected region, make sure we dont leak 496 * a norefcounted dst 497 */ 498 skb_dst_force(skb); 499 500 spin_lock_irqsave(&list->lock, flags); 501 sock_skb_set_dropcount(sk, skb); 502 __skb_queue_tail(list, skb); 503 spin_unlock_irqrestore(&list->lock, flags); 504 505 if (!sock_flag(sk, SOCK_DEAD)) 506 sk->sk_data_ready(sk); 507 return 0; 508 } 509 EXPORT_SYMBOL(__sock_queue_rcv_skb); 510 511 int sock_queue_rcv_skb_reason(struct sock *sk, struct sk_buff *skb, 512 enum skb_drop_reason *reason) 513 { 514 enum skb_drop_reason drop_reason; 515 int err; 516 517 err = sk_filter(sk, skb); 518 if (err) { 519 drop_reason = SKB_DROP_REASON_SOCKET_FILTER; 520 goto out; 521 } 522 err = __sock_queue_rcv_skb(sk, skb); 523 switch (err) { 524 case -ENOMEM: 525 drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF; 526 break; 527 case -ENOBUFS: 528 drop_reason = SKB_DROP_REASON_PROTO_MEM; 529 break; 530 default: 531 drop_reason = SKB_NOT_DROPPED_YET; 532 break; 533 } 534 out: 535 if (reason) 536 *reason = drop_reason; 537 return err; 538 } 539 EXPORT_SYMBOL(sock_queue_rcv_skb_reason); 540 541 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, 542 const int nested, unsigned int trim_cap, bool refcounted) 543 { 544 int rc = NET_RX_SUCCESS; 545 546 if (sk_filter_trim_cap(sk, skb, trim_cap)) 547 goto discard_and_relse; 548 549 skb->dev = NULL; 550 551 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) { 552 atomic_inc(&sk->sk_drops); 553 goto discard_and_relse; 554 } 555 if (nested) 556 bh_lock_sock_nested(sk); 557 else 558 bh_lock_sock(sk); 559 if (!sock_owned_by_user(sk)) { 560 /* 561 * trylock + unlock semantics: 562 */ 563 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_); 564 565 rc = sk_backlog_rcv(sk, skb); 566 567 mutex_release(&sk->sk_lock.dep_map, _RET_IP_); 568 } else if (sk_add_backlog(sk, skb, READ_ONCE(sk->sk_rcvbuf))) { 569 bh_unlock_sock(sk); 570 atomic_inc(&sk->sk_drops); 571 goto discard_and_relse; 572 } 573 574 bh_unlock_sock(sk); 575 out: 576 if (refcounted) 577 sock_put(sk); 578 return rc; 579 discard_and_relse: 580 kfree_skb(skb); 581 goto out; 582 } 583 EXPORT_SYMBOL(__sk_receive_skb); 584 585 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ip6_dst_check(struct dst_entry *, 586 u32)); 587 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ipv4_dst_check(struct dst_entry *, 588 u32)); 589 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie) 590 { 591 struct dst_entry *dst = __sk_dst_get(sk); 592 593 if (dst && dst->obsolete && 594 INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check, 595 dst, cookie) == NULL) { 596 sk_tx_queue_clear(sk); 597 sk->sk_dst_pending_confirm = 0; 598 RCU_INIT_POINTER(sk->sk_dst_cache, NULL); 599 dst_release(dst); 600 return NULL; 601 } 602 603 return dst; 604 } 605 EXPORT_SYMBOL(__sk_dst_check); 606 607 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie) 608 { 609 struct dst_entry *dst = sk_dst_get(sk); 610 611 if (dst && dst->obsolete && 612 INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check, 613 dst, cookie) == NULL) { 614 sk_dst_reset(sk); 615 dst_release(dst); 616 return NULL; 617 } 618 619 return dst; 620 } 621 EXPORT_SYMBOL(sk_dst_check); 622 623 static int sock_bindtoindex_locked(struct sock *sk, int ifindex) 624 { 625 int ret = -ENOPROTOOPT; 626 #ifdef CONFIG_NETDEVICES 627 struct net *net = sock_net(sk); 628 629 /* Sorry... */ 630 ret = -EPERM; 631 if (sk->sk_bound_dev_if && !ns_capable(net->user_ns, CAP_NET_RAW)) 632 goto out; 633 634 ret = -EINVAL; 635 if (ifindex < 0) 636 goto out; 637 638 /* Paired with all READ_ONCE() done locklessly. */ 639 WRITE_ONCE(sk->sk_bound_dev_if, ifindex); 640 641 if (sk->sk_prot->rehash) 642 sk->sk_prot->rehash(sk); 643 sk_dst_reset(sk); 644 645 ret = 0; 646 647 out: 648 #endif 649 650 return ret; 651 } 652 653 int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk) 654 { 655 int ret; 656 657 if (lock_sk) 658 lock_sock(sk); 659 ret = sock_bindtoindex_locked(sk, ifindex); 660 if (lock_sk) 661 release_sock(sk); 662 663 return ret; 664 } 665 EXPORT_SYMBOL(sock_bindtoindex); 666 667 static int sock_setbindtodevice(struct sock *sk, sockptr_t optval, int optlen) 668 { 669 int ret = -ENOPROTOOPT; 670 #ifdef CONFIG_NETDEVICES 671 struct net *net = sock_net(sk); 672 char devname[IFNAMSIZ]; 673 int index; 674 675 ret = -EINVAL; 676 if (optlen < 0) 677 goto out; 678 679 /* Bind this socket to a particular device like "eth0", 680 * as specified in the passed interface name. If the 681 * name is "" or the option length is zero the socket 682 * is not bound. 683 */ 684 if (optlen > IFNAMSIZ - 1) 685 optlen = IFNAMSIZ - 1; 686 memset(devname, 0, sizeof(devname)); 687 688 ret = -EFAULT; 689 if (copy_from_sockptr(devname, optval, optlen)) 690 goto out; 691 692 index = 0; 693 if (devname[0] != '\0') { 694 struct net_device *dev; 695 696 rcu_read_lock(); 697 dev = dev_get_by_name_rcu(net, devname); 698 if (dev) 699 index = dev->ifindex; 700 rcu_read_unlock(); 701 ret = -ENODEV; 702 if (!dev) 703 goto out; 704 } 705 706 sockopt_lock_sock(sk); 707 ret = sock_bindtoindex_locked(sk, index); 708 sockopt_release_sock(sk); 709 out: 710 #endif 711 712 return ret; 713 } 714 715 static int sock_getbindtodevice(struct sock *sk, sockptr_t optval, 716 sockptr_t optlen, int len) 717 { 718 int ret = -ENOPROTOOPT; 719 #ifdef CONFIG_NETDEVICES 720 int bound_dev_if = READ_ONCE(sk->sk_bound_dev_if); 721 struct net *net = sock_net(sk); 722 char devname[IFNAMSIZ]; 723 724 if (bound_dev_if == 0) { 725 len = 0; 726 goto zero; 727 } 728 729 ret = -EINVAL; 730 if (len < IFNAMSIZ) 731 goto out; 732 733 ret = netdev_get_name(net, devname, bound_dev_if); 734 if (ret) 735 goto out; 736 737 len = strlen(devname) + 1; 738 739 ret = -EFAULT; 740 if (copy_to_sockptr(optval, devname, len)) 741 goto out; 742 743 zero: 744 ret = -EFAULT; 745 if (copy_to_sockptr(optlen, &len, sizeof(int))) 746 goto out; 747 748 ret = 0; 749 750 out: 751 #endif 752 753 return ret; 754 } 755 756 bool sk_mc_loop(struct sock *sk) 757 { 758 if (dev_recursion_level()) 759 return false; 760 if (!sk) 761 return true; 762 switch (sk->sk_family) { 763 case AF_INET: 764 return inet_sk(sk)->mc_loop; 765 #if IS_ENABLED(CONFIG_IPV6) 766 case AF_INET6: 767 return inet6_sk(sk)->mc_loop; 768 #endif 769 } 770 WARN_ON_ONCE(1); 771 return true; 772 } 773 EXPORT_SYMBOL(sk_mc_loop); 774 775 void sock_set_reuseaddr(struct sock *sk) 776 { 777 lock_sock(sk); 778 sk->sk_reuse = SK_CAN_REUSE; 779 release_sock(sk); 780 } 781 EXPORT_SYMBOL(sock_set_reuseaddr); 782 783 void sock_set_reuseport(struct sock *sk) 784 { 785 lock_sock(sk); 786 sk->sk_reuseport = true; 787 release_sock(sk); 788 } 789 EXPORT_SYMBOL(sock_set_reuseport); 790 791 void sock_no_linger(struct sock *sk) 792 { 793 lock_sock(sk); 794 sk->sk_lingertime = 0; 795 sock_set_flag(sk, SOCK_LINGER); 796 release_sock(sk); 797 } 798 EXPORT_SYMBOL(sock_no_linger); 799 800 void sock_set_priority(struct sock *sk, u32 priority) 801 { 802 lock_sock(sk); 803 sk->sk_priority = priority; 804 release_sock(sk); 805 } 806 EXPORT_SYMBOL(sock_set_priority); 807 808 void sock_set_sndtimeo(struct sock *sk, s64 secs) 809 { 810 lock_sock(sk); 811 if (secs && secs < MAX_SCHEDULE_TIMEOUT / HZ - 1) 812 sk->sk_sndtimeo = secs * HZ; 813 else 814 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT; 815 release_sock(sk); 816 } 817 EXPORT_SYMBOL(sock_set_sndtimeo); 818 819 static void __sock_set_timestamps(struct sock *sk, bool val, bool new, bool ns) 820 { 821 if (val) { 822 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, new); 823 sock_valbool_flag(sk, SOCK_RCVTSTAMPNS, ns); 824 sock_set_flag(sk, SOCK_RCVTSTAMP); 825 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 826 } else { 827 sock_reset_flag(sk, SOCK_RCVTSTAMP); 828 sock_reset_flag(sk, SOCK_RCVTSTAMPNS); 829 } 830 } 831 832 void sock_enable_timestamps(struct sock *sk) 833 { 834 lock_sock(sk); 835 __sock_set_timestamps(sk, true, false, true); 836 release_sock(sk); 837 } 838 EXPORT_SYMBOL(sock_enable_timestamps); 839 840 void sock_set_timestamp(struct sock *sk, int optname, bool valbool) 841 { 842 switch (optname) { 843 case SO_TIMESTAMP_OLD: 844 __sock_set_timestamps(sk, valbool, false, false); 845 break; 846 case SO_TIMESTAMP_NEW: 847 __sock_set_timestamps(sk, valbool, true, false); 848 break; 849 case SO_TIMESTAMPNS_OLD: 850 __sock_set_timestamps(sk, valbool, false, true); 851 break; 852 case SO_TIMESTAMPNS_NEW: 853 __sock_set_timestamps(sk, valbool, true, true); 854 break; 855 } 856 } 857 858 static int sock_timestamping_bind_phc(struct sock *sk, int phc_index) 859 { 860 struct net *net = sock_net(sk); 861 struct net_device *dev = NULL; 862 bool match = false; 863 int *vclock_index; 864 int i, num; 865 866 if (sk->sk_bound_dev_if) 867 dev = dev_get_by_index(net, sk->sk_bound_dev_if); 868 869 if (!dev) { 870 pr_err("%s: sock not bind to device\n", __func__); 871 return -EOPNOTSUPP; 872 } 873 874 num = ethtool_get_phc_vclocks(dev, &vclock_index); 875 dev_put(dev); 876 877 for (i = 0; i < num; i++) { 878 if (*(vclock_index + i) == phc_index) { 879 match = true; 880 break; 881 } 882 } 883 884 if (num > 0) 885 kfree(vclock_index); 886 887 if (!match) 888 return -EINVAL; 889 890 sk->sk_bind_phc = phc_index; 891 892 return 0; 893 } 894 895 int sock_set_timestamping(struct sock *sk, int optname, 896 struct so_timestamping timestamping) 897 { 898 int val = timestamping.flags; 899 int ret; 900 901 if (val & ~SOF_TIMESTAMPING_MASK) 902 return -EINVAL; 903 904 if (val & SOF_TIMESTAMPING_OPT_ID && 905 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) { 906 if (sk_is_tcp(sk)) { 907 if ((1 << sk->sk_state) & 908 (TCPF_CLOSE | TCPF_LISTEN)) 909 return -EINVAL; 910 atomic_set(&sk->sk_tskey, tcp_sk(sk)->snd_una); 911 } else { 912 atomic_set(&sk->sk_tskey, 0); 913 } 914 } 915 916 if (val & SOF_TIMESTAMPING_OPT_STATS && 917 !(val & SOF_TIMESTAMPING_OPT_TSONLY)) 918 return -EINVAL; 919 920 if (val & SOF_TIMESTAMPING_BIND_PHC) { 921 ret = sock_timestamping_bind_phc(sk, timestamping.bind_phc); 922 if (ret) 923 return ret; 924 } 925 926 sk->sk_tsflags = val; 927 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, optname == SO_TIMESTAMPING_NEW); 928 929 if (val & SOF_TIMESTAMPING_RX_SOFTWARE) 930 sock_enable_timestamp(sk, 931 SOCK_TIMESTAMPING_RX_SOFTWARE); 932 else 933 sock_disable_timestamp(sk, 934 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE)); 935 return 0; 936 } 937 938 void sock_set_keepalive(struct sock *sk) 939 { 940 lock_sock(sk); 941 if (sk->sk_prot->keepalive) 942 sk->sk_prot->keepalive(sk, true); 943 sock_valbool_flag(sk, SOCK_KEEPOPEN, true); 944 release_sock(sk); 945 } 946 EXPORT_SYMBOL(sock_set_keepalive); 947 948 static void __sock_set_rcvbuf(struct sock *sk, int val) 949 { 950 /* Ensure val * 2 fits into an int, to prevent max_t() from treating it 951 * as a negative value. 952 */ 953 val = min_t(int, val, INT_MAX / 2); 954 sk->sk_userlocks |= SOCK_RCVBUF_LOCK; 955 956 /* We double it on the way in to account for "struct sk_buff" etc. 957 * overhead. Applications assume that the SO_RCVBUF setting they make 958 * will allow that much actual data to be received on that socket. 959 * 960 * Applications are unaware that "struct sk_buff" and other overheads 961 * allocate from the receive buffer during socket buffer allocation. 962 * 963 * And after considering the possible alternatives, returning the value 964 * we actually used in getsockopt is the most desirable behavior. 965 */ 966 WRITE_ONCE(sk->sk_rcvbuf, max_t(int, val * 2, SOCK_MIN_RCVBUF)); 967 } 968 969 void sock_set_rcvbuf(struct sock *sk, int val) 970 { 971 lock_sock(sk); 972 __sock_set_rcvbuf(sk, val); 973 release_sock(sk); 974 } 975 EXPORT_SYMBOL(sock_set_rcvbuf); 976 977 static void __sock_set_mark(struct sock *sk, u32 val) 978 { 979 if (val != sk->sk_mark) { 980 sk->sk_mark = val; 981 sk_dst_reset(sk); 982 } 983 } 984 985 void sock_set_mark(struct sock *sk, u32 val) 986 { 987 lock_sock(sk); 988 __sock_set_mark(sk, val); 989 release_sock(sk); 990 } 991 EXPORT_SYMBOL(sock_set_mark); 992 993 static void sock_release_reserved_memory(struct sock *sk, int bytes) 994 { 995 /* Round down bytes to multiple of pages */ 996 bytes = round_down(bytes, PAGE_SIZE); 997 998 WARN_ON(bytes > sk->sk_reserved_mem); 999 sk->sk_reserved_mem -= bytes; 1000 sk_mem_reclaim(sk); 1001 } 1002 1003 static int sock_reserve_memory(struct sock *sk, int bytes) 1004 { 1005 long allocated; 1006 bool charged; 1007 int pages; 1008 1009 if (!mem_cgroup_sockets_enabled || !sk->sk_memcg || !sk_has_account(sk)) 1010 return -EOPNOTSUPP; 1011 1012 if (!bytes) 1013 return 0; 1014 1015 pages = sk_mem_pages(bytes); 1016 1017 /* pre-charge to memcg */ 1018 charged = mem_cgroup_charge_skmem(sk->sk_memcg, pages, 1019 GFP_KERNEL | __GFP_RETRY_MAYFAIL); 1020 if (!charged) 1021 return -ENOMEM; 1022 1023 /* pre-charge to forward_alloc */ 1024 sk_memory_allocated_add(sk, pages); 1025 allocated = sk_memory_allocated(sk); 1026 /* If the system goes into memory pressure with this 1027 * precharge, give up and return error. 1028 */ 1029 if (allocated > sk_prot_mem_limits(sk, 1)) { 1030 sk_memory_allocated_sub(sk, pages); 1031 mem_cgroup_uncharge_skmem(sk->sk_memcg, pages); 1032 return -ENOMEM; 1033 } 1034 sk->sk_forward_alloc += pages << PAGE_SHIFT; 1035 1036 sk->sk_reserved_mem += pages << PAGE_SHIFT; 1037 1038 return 0; 1039 } 1040 1041 void sockopt_lock_sock(struct sock *sk) 1042 { 1043 /* When current->bpf_ctx is set, the setsockopt is called from 1044 * a bpf prog. bpf has ensured the sk lock has been 1045 * acquired before calling setsockopt(). 1046 */ 1047 if (has_current_bpf_ctx()) 1048 return; 1049 1050 lock_sock(sk); 1051 } 1052 EXPORT_SYMBOL(sockopt_lock_sock); 1053 1054 void sockopt_release_sock(struct sock *sk) 1055 { 1056 if (has_current_bpf_ctx()) 1057 return; 1058 1059 release_sock(sk); 1060 } 1061 EXPORT_SYMBOL(sockopt_release_sock); 1062 1063 bool sockopt_ns_capable(struct user_namespace *ns, int cap) 1064 { 1065 return has_current_bpf_ctx() || ns_capable(ns, cap); 1066 } 1067 EXPORT_SYMBOL(sockopt_ns_capable); 1068 1069 bool sockopt_capable(int cap) 1070 { 1071 return has_current_bpf_ctx() || capable(cap); 1072 } 1073 EXPORT_SYMBOL(sockopt_capable); 1074 1075 /* 1076 * This is meant for all protocols to use and covers goings on 1077 * at the socket level. Everything here is generic. 1078 */ 1079 1080 int sk_setsockopt(struct sock *sk, int level, int optname, 1081 sockptr_t optval, unsigned int optlen) 1082 { 1083 struct so_timestamping timestamping; 1084 struct socket *sock = sk->sk_socket; 1085 struct sock_txtime sk_txtime; 1086 int val; 1087 int valbool; 1088 struct linger ling; 1089 int ret = 0; 1090 1091 /* 1092 * Options without arguments 1093 */ 1094 1095 if (optname == SO_BINDTODEVICE) 1096 return sock_setbindtodevice(sk, optval, optlen); 1097 1098 if (optlen < sizeof(int)) 1099 return -EINVAL; 1100 1101 if (copy_from_sockptr(&val, optval, sizeof(val))) 1102 return -EFAULT; 1103 1104 valbool = val ? 1 : 0; 1105 1106 sockopt_lock_sock(sk); 1107 1108 switch (optname) { 1109 case SO_DEBUG: 1110 if (val && !sockopt_capable(CAP_NET_ADMIN)) 1111 ret = -EACCES; 1112 else 1113 sock_valbool_flag(sk, SOCK_DBG, valbool); 1114 break; 1115 case SO_REUSEADDR: 1116 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE); 1117 break; 1118 case SO_REUSEPORT: 1119 sk->sk_reuseport = valbool; 1120 break; 1121 case SO_TYPE: 1122 case SO_PROTOCOL: 1123 case SO_DOMAIN: 1124 case SO_ERROR: 1125 ret = -ENOPROTOOPT; 1126 break; 1127 case SO_DONTROUTE: 1128 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool); 1129 sk_dst_reset(sk); 1130 break; 1131 case SO_BROADCAST: 1132 sock_valbool_flag(sk, SOCK_BROADCAST, valbool); 1133 break; 1134 case SO_SNDBUF: 1135 /* Don't error on this BSD doesn't and if you think 1136 * about it this is right. Otherwise apps have to 1137 * play 'guess the biggest size' games. RCVBUF/SNDBUF 1138 * are treated in BSD as hints 1139 */ 1140 val = min_t(u32, val, READ_ONCE(sysctl_wmem_max)); 1141 set_sndbuf: 1142 /* Ensure val * 2 fits into an int, to prevent max_t() 1143 * from treating it as a negative value. 1144 */ 1145 val = min_t(int, val, INT_MAX / 2); 1146 sk->sk_userlocks |= SOCK_SNDBUF_LOCK; 1147 WRITE_ONCE(sk->sk_sndbuf, 1148 max_t(int, val * 2, SOCK_MIN_SNDBUF)); 1149 /* Wake up sending tasks if we upped the value. */ 1150 sk->sk_write_space(sk); 1151 break; 1152 1153 case SO_SNDBUFFORCE: 1154 if (!sockopt_capable(CAP_NET_ADMIN)) { 1155 ret = -EPERM; 1156 break; 1157 } 1158 1159 /* No negative values (to prevent underflow, as val will be 1160 * multiplied by 2). 1161 */ 1162 if (val < 0) 1163 val = 0; 1164 goto set_sndbuf; 1165 1166 case SO_RCVBUF: 1167 /* Don't error on this BSD doesn't and if you think 1168 * about it this is right. Otherwise apps have to 1169 * play 'guess the biggest size' games. RCVBUF/SNDBUF 1170 * are treated in BSD as hints 1171 */ 1172 __sock_set_rcvbuf(sk, min_t(u32, val, READ_ONCE(sysctl_rmem_max))); 1173 break; 1174 1175 case SO_RCVBUFFORCE: 1176 if (!sockopt_capable(CAP_NET_ADMIN)) { 1177 ret = -EPERM; 1178 break; 1179 } 1180 1181 /* No negative values (to prevent underflow, as val will be 1182 * multiplied by 2). 1183 */ 1184 __sock_set_rcvbuf(sk, max(val, 0)); 1185 break; 1186 1187 case SO_KEEPALIVE: 1188 if (sk->sk_prot->keepalive) 1189 sk->sk_prot->keepalive(sk, valbool); 1190 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool); 1191 break; 1192 1193 case SO_OOBINLINE: 1194 sock_valbool_flag(sk, SOCK_URGINLINE, valbool); 1195 break; 1196 1197 case SO_NO_CHECK: 1198 sk->sk_no_check_tx = valbool; 1199 break; 1200 1201 case SO_PRIORITY: 1202 if ((val >= 0 && val <= 6) || 1203 sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) || 1204 sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) 1205 sk->sk_priority = val; 1206 else 1207 ret = -EPERM; 1208 break; 1209 1210 case SO_LINGER: 1211 if (optlen < sizeof(ling)) { 1212 ret = -EINVAL; /* 1003.1g */ 1213 break; 1214 } 1215 if (copy_from_sockptr(&ling, optval, sizeof(ling))) { 1216 ret = -EFAULT; 1217 break; 1218 } 1219 if (!ling.l_onoff) 1220 sock_reset_flag(sk, SOCK_LINGER); 1221 else { 1222 #if (BITS_PER_LONG == 32) 1223 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ) 1224 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT; 1225 else 1226 #endif 1227 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ; 1228 sock_set_flag(sk, SOCK_LINGER); 1229 } 1230 break; 1231 1232 case SO_BSDCOMPAT: 1233 break; 1234 1235 case SO_PASSCRED: 1236 if (valbool) 1237 set_bit(SOCK_PASSCRED, &sock->flags); 1238 else 1239 clear_bit(SOCK_PASSCRED, &sock->flags); 1240 break; 1241 1242 case SO_TIMESTAMP_OLD: 1243 case SO_TIMESTAMP_NEW: 1244 case SO_TIMESTAMPNS_OLD: 1245 case SO_TIMESTAMPNS_NEW: 1246 sock_set_timestamp(sk, optname, valbool); 1247 break; 1248 1249 case SO_TIMESTAMPING_NEW: 1250 case SO_TIMESTAMPING_OLD: 1251 if (optlen == sizeof(timestamping)) { 1252 if (copy_from_sockptr(×tamping, optval, 1253 sizeof(timestamping))) { 1254 ret = -EFAULT; 1255 break; 1256 } 1257 } else { 1258 memset(×tamping, 0, sizeof(timestamping)); 1259 timestamping.flags = val; 1260 } 1261 ret = sock_set_timestamping(sk, optname, timestamping); 1262 break; 1263 1264 case SO_RCVLOWAT: 1265 if (val < 0) 1266 val = INT_MAX; 1267 if (sock && sock->ops->set_rcvlowat) 1268 ret = sock->ops->set_rcvlowat(sk, val); 1269 else 1270 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1); 1271 break; 1272 1273 case SO_RCVTIMEO_OLD: 1274 case SO_RCVTIMEO_NEW: 1275 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, 1276 optlen, optname == SO_RCVTIMEO_OLD); 1277 break; 1278 1279 case SO_SNDTIMEO_OLD: 1280 case SO_SNDTIMEO_NEW: 1281 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, 1282 optlen, optname == SO_SNDTIMEO_OLD); 1283 break; 1284 1285 case SO_ATTACH_FILTER: { 1286 struct sock_fprog fprog; 1287 1288 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen); 1289 if (!ret) 1290 ret = sk_attach_filter(&fprog, sk); 1291 break; 1292 } 1293 case SO_ATTACH_BPF: 1294 ret = -EINVAL; 1295 if (optlen == sizeof(u32)) { 1296 u32 ufd; 1297 1298 ret = -EFAULT; 1299 if (copy_from_sockptr(&ufd, optval, sizeof(ufd))) 1300 break; 1301 1302 ret = sk_attach_bpf(ufd, sk); 1303 } 1304 break; 1305 1306 case SO_ATTACH_REUSEPORT_CBPF: { 1307 struct sock_fprog fprog; 1308 1309 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen); 1310 if (!ret) 1311 ret = sk_reuseport_attach_filter(&fprog, sk); 1312 break; 1313 } 1314 case SO_ATTACH_REUSEPORT_EBPF: 1315 ret = -EINVAL; 1316 if (optlen == sizeof(u32)) { 1317 u32 ufd; 1318 1319 ret = -EFAULT; 1320 if (copy_from_sockptr(&ufd, optval, sizeof(ufd))) 1321 break; 1322 1323 ret = sk_reuseport_attach_bpf(ufd, sk); 1324 } 1325 break; 1326 1327 case SO_DETACH_REUSEPORT_BPF: 1328 ret = reuseport_detach_prog(sk); 1329 break; 1330 1331 case SO_DETACH_FILTER: 1332 ret = sk_detach_filter(sk); 1333 break; 1334 1335 case SO_LOCK_FILTER: 1336 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool) 1337 ret = -EPERM; 1338 else 1339 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool); 1340 break; 1341 1342 case SO_PASSSEC: 1343 if (valbool) 1344 set_bit(SOCK_PASSSEC, &sock->flags); 1345 else 1346 clear_bit(SOCK_PASSSEC, &sock->flags); 1347 break; 1348 case SO_MARK: 1349 if (!sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) && 1350 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) { 1351 ret = -EPERM; 1352 break; 1353 } 1354 1355 __sock_set_mark(sk, val); 1356 break; 1357 case SO_RCVMARK: 1358 if (!sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) && 1359 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) { 1360 ret = -EPERM; 1361 break; 1362 } 1363 1364 sock_valbool_flag(sk, SOCK_RCVMARK, valbool); 1365 break; 1366 1367 case SO_RXQ_OVFL: 1368 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool); 1369 break; 1370 1371 case SO_WIFI_STATUS: 1372 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool); 1373 break; 1374 1375 case SO_PEEK_OFF: 1376 if (sock->ops->set_peek_off) 1377 ret = sock->ops->set_peek_off(sk, val); 1378 else 1379 ret = -EOPNOTSUPP; 1380 break; 1381 1382 case SO_NOFCS: 1383 sock_valbool_flag(sk, SOCK_NOFCS, valbool); 1384 break; 1385 1386 case SO_SELECT_ERR_QUEUE: 1387 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool); 1388 break; 1389 1390 #ifdef CONFIG_NET_RX_BUSY_POLL 1391 case SO_BUSY_POLL: 1392 /* allow unprivileged users to decrease the value */ 1393 if ((val > sk->sk_ll_usec) && !sockopt_capable(CAP_NET_ADMIN)) 1394 ret = -EPERM; 1395 else { 1396 if (val < 0) 1397 ret = -EINVAL; 1398 else 1399 WRITE_ONCE(sk->sk_ll_usec, val); 1400 } 1401 break; 1402 case SO_PREFER_BUSY_POLL: 1403 if (valbool && !sockopt_capable(CAP_NET_ADMIN)) 1404 ret = -EPERM; 1405 else 1406 WRITE_ONCE(sk->sk_prefer_busy_poll, valbool); 1407 break; 1408 case SO_BUSY_POLL_BUDGET: 1409 if (val > READ_ONCE(sk->sk_busy_poll_budget) && !sockopt_capable(CAP_NET_ADMIN)) { 1410 ret = -EPERM; 1411 } else { 1412 if (val < 0 || val > U16_MAX) 1413 ret = -EINVAL; 1414 else 1415 WRITE_ONCE(sk->sk_busy_poll_budget, val); 1416 } 1417 break; 1418 #endif 1419 1420 case SO_MAX_PACING_RATE: 1421 { 1422 unsigned long ulval = (val == ~0U) ? ~0UL : (unsigned int)val; 1423 1424 if (sizeof(ulval) != sizeof(val) && 1425 optlen >= sizeof(ulval) && 1426 copy_from_sockptr(&ulval, optval, sizeof(ulval))) { 1427 ret = -EFAULT; 1428 break; 1429 } 1430 if (ulval != ~0UL) 1431 cmpxchg(&sk->sk_pacing_status, 1432 SK_PACING_NONE, 1433 SK_PACING_NEEDED); 1434 sk->sk_max_pacing_rate = ulval; 1435 sk->sk_pacing_rate = min(sk->sk_pacing_rate, ulval); 1436 break; 1437 } 1438 case SO_INCOMING_CPU: 1439 WRITE_ONCE(sk->sk_incoming_cpu, val); 1440 break; 1441 1442 case SO_CNX_ADVICE: 1443 if (val == 1) 1444 dst_negative_advice(sk); 1445 break; 1446 1447 case SO_ZEROCOPY: 1448 if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) { 1449 if (!(sk_is_tcp(sk) || 1450 (sk->sk_type == SOCK_DGRAM && 1451 sk->sk_protocol == IPPROTO_UDP))) 1452 ret = -EOPNOTSUPP; 1453 } else if (sk->sk_family != PF_RDS) { 1454 ret = -EOPNOTSUPP; 1455 } 1456 if (!ret) { 1457 if (val < 0 || val > 1) 1458 ret = -EINVAL; 1459 else 1460 sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool); 1461 } 1462 break; 1463 1464 case SO_TXTIME: 1465 if (optlen != sizeof(struct sock_txtime)) { 1466 ret = -EINVAL; 1467 break; 1468 } else if (copy_from_sockptr(&sk_txtime, optval, 1469 sizeof(struct sock_txtime))) { 1470 ret = -EFAULT; 1471 break; 1472 } else if (sk_txtime.flags & ~SOF_TXTIME_FLAGS_MASK) { 1473 ret = -EINVAL; 1474 break; 1475 } 1476 /* CLOCK_MONOTONIC is only used by sch_fq, and this packet 1477 * scheduler has enough safe guards. 1478 */ 1479 if (sk_txtime.clockid != CLOCK_MONOTONIC && 1480 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) { 1481 ret = -EPERM; 1482 break; 1483 } 1484 sock_valbool_flag(sk, SOCK_TXTIME, true); 1485 sk->sk_clockid = sk_txtime.clockid; 1486 sk->sk_txtime_deadline_mode = 1487 !!(sk_txtime.flags & SOF_TXTIME_DEADLINE_MODE); 1488 sk->sk_txtime_report_errors = 1489 !!(sk_txtime.flags & SOF_TXTIME_REPORT_ERRORS); 1490 break; 1491 1492 case SO_BINDTOIFINDEX: 1493 ret = sock_bindtoindex_locked(sk, val); 1494 break; 1495 1496 case SO_BUF_LOCK: 1497 if (val & ~SOCK_BUF_LOCK_MASK) { 1498 ret = -EINVAL; 1499 break; 1500 } 1501 sk->sk_userlocks = val | (sk->sk_userlocks & 1502 ~SOCK_BUF_LOCK_MASK); 1503 break; 1504 1505 case SO_RESERVE_MEM: 1506 { 1507 int delta; 1508 1509 if (val < 0) { 1510 ret = -EINVAL; 1511 break; 1512 } 1513 1514 delta = val - sk->sk_reserved_mem; 1515 if (delta < 0) 1516 sock_release_reserved_memory(sk, -delta); 1517 else 1518 ret = sock_reserve_memory(sk, delta); 1519 break; 1520 } 1521 1522 case SO_TXREHASH: 1523 if (val < -1 || val > 1) { 1524 ret = -EINVAL; 1525 break; 1526 } 1527 /* Paired with READ_ONCE() in tcp_rtx_synack() */ 1528 WRITE_ONCE(sk->sk_txrehash, (u8)val); 1529 break; 1530 1531 default: 1532 ret = -ENOPROTOOPT; 1533 break; 1534 } 1535 sockopt_release_sock(sk); 1536 return ret; 1537 } 1538 1539 int sock_setsockopt(struct socket *sock, int level, int optname, 1540 sockptr_t optval, unsigned int optlen) 1541 { 1542 return sk_setsockopt(sock->sk, level, optname, 1543 optval, optlen); 1544 } 1545 EXPORT_SYMBOL(sock_setsockopt); 1546 1547 static const struct cred *sk_get_peer_cred(struct sock *sk) 1548 { 1549 const struct cred *cred; 1550 1551 spin_lock(&sk->sk_peer_lock); 1552 cred = get_cred(sk->sk_peer_cred); 1553 spin_unlock(&sk->sk_peer_lock); 1554 1555 return cred; 1556 } 1557 1558 static void cred_to_ucred(struct pid *pid, const struct cred *cred, 1559 struct ucred *ucred) 1560 { 1561 ucred->pid = pid_vnr(pid); 1562 ucred->uid = ucred->gid = -1; 1563 if (cred) { 1564 struct user_namespace *current_ns = current_user_ns(); 1565 1566 ucred->uid = from_kuid_munged(current_ns, cred->euid); 1567 ucred->gid = from_kgid_munged(current_ns, cred->egid); 1568 } 1569 } 1570 1571 static int groups_to_user(sockptr_t dst, const struct group_info *src) 1572 { 1573 struct user_namespace *user_ns = current_user_ns(); 1574 int i; 1575 1576 for (i = 0; i < src->ngroups; i++) { 1577 gid_t gid = from_kgid_munged(user_ns, src->gid[i]); 1578 1579 if (copy_to_sockptr_offset(dst, i * sizeof(gid), &gid, sizeof(gid))) 1580 return -EFAULT; 1581 } 1582 1583 return 0; 1584 } 1585 1586 int sk_getsockopt(struct sock *sk, int level, int optname, 1587 sockptr_t optval, sockptr_t optlen) 1588 { 1589 struct socket *sock = sk->sk_socket; 1590 1591 union { 1592 int val; 1593 u64 val64; 1594 unsigned long ulval; 1595 struct linger ling; 1596 struct old_timeval32 tm32; 1597 struct __kernel_old_timeval tm; 1598 struct __kernel_sock_timeval stm; 1599 struct sock_txtime txtime; 1600 struct so_timestamping timestamping; 1601 } v; 1602 1603 int lv = sizeof(int); 1604 int len; 1605 1606 if (copy_from_sockptr(&len, optlen, sizeof(int))) 1607 return -EFAULT; 1608 if (len < 0) 1609 return -EINVAL; 1610 1611 memset(&v, 0, sizeof(v)); 1612 1613 switch (optname) { 1614 case SO_DEBUG: 1615 v.val = sock_flag(sk, SOCK_DBG); 1616 break; 1617 1618 case SO_DONTROUTE: 1619 v.val = sock_flag(sk, SOCK_LOCALROUTE); 1620 break; 1621 1622 case SO_BROADCAST: 1623 v.val = sock_flag(sk, SOCK_BROADCAST); 1624 break; 1625 1626 case SO_SNDBUF: 1627 v.val = sk->sk_sndbuf; 1628 break; 1629 1630 case SO_RCVBUF: 1631 v.val = sk->sk_rcvbuf; 1632 break; 1633 1634 case SO_REUSEADDR: 1635 v.val = sk->sk_reuse; 1636 break; 1637 1638 case SO_REUSEPORT: 1639 v.val = sk->sk_reuseport; 1640 break; 1641 1642 case SO_KEEPALIVE: 1643 v.val = sock_flag(sk, SOCK_KEEPOPEN); 1644 break; 1645 1646 case SO_TYPE: 1647 v.val = sk->sk_type; 1648 break; 1649 1650 case SO_PROTOCOL: 1651 v.val = sk->sk_protocol; 1652 break; 1653 1654 case SO_DOMAIN: 1655 v.val = sk->sk_family; 1656 break; 1657 1658 case SO_ERROR: 1659 v.val = -sock_error(sk); 1660 if (v.val == 0) 1661 v.val = xchg(&sk->sk_err_soft, 0); 1662 break; 1663 1664 case SO_OOBINLINE: 1665 v.val = sock_flag(sk, SOCK_URGINLINE); 1666 break; 1667 1668 case SO_NO_CHECK: 1669 v.val = sk->sk_no_check_tx; 1670 break; 1671 1672 case SO_PRIORITY: 1673 v.val = sk->sk_priority; 1674 break; 1675 1676 case SO_LINGER: 1677 lv = sizeof(v.ling); 1678 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER); 1679 v.ling.l_linger = sk->sk_lingertime / HZ; 1680 break; 1681 1682 case SO_BSDCOMPAT: 1683 break; 1684 1685 case SO_TIMESTAMP_OLD: 1686 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && 1687 !sock_flag(sk, SOCK_TSTAMP_NEW) && 1688 !sock_flag(sk, SOCK_RCVTSTAMPNS); 1689 break; 1690 1691 case SO_TIMESTAMPNS_OLD: 1692 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && !sock_flag(sk, SOCK_TSTAMP_NEW); 1693 break; 1694 1695 case SO_TIMESTAMP_NEW: 1696 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && sock_flag(sk, SOCK_TSTAMP_NEW); 1697 break; 1698 1699 case SO_TIMESTAMPNS_NEW: 1700 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && sock_flag(sk, SOCK_TSTAMP_NEW); 1701 break; 1702 1703 case SO_TIMESTAMPING_OLD: 1704 lv = sizeof(v.timestamping); 1705 v.timestamping.flags = sk->sk_tsflags; 1706 v.timestamping.bind_phc = sk->sk_bind_phc; 1707 break; 1708 1709 case SO_RCVTIMEO_OLD: 1710 case SO_RCVTIMEO_NEW: 1711 lv = sock_get_timeout(sk->sk_rcvtimeo, &v, SO_RCVTIMEO_OLD == optname); 1712 break; 1713 1714 case SO_SNDTIMEO_OLD: 1715 case SO_SNDTIMEO_NEW: 1716 lv = sock_get_timeout(sk->sk_sndtimeo, &v, SO_SNDTIMEO_OLD == optname); 1717 break; 1718 1719 case SO_RCVLOWAT: 1720 v.val = sk->sk_rcvlowat; 1721 break; 1722 1723 case SO_SNDLOWAT: 1724 v.val = 1; 1725 break; 1726 1727 case SO_PASSCRED: 1728 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags); 1729 break; 1730 1731 case SO_PEERCRED: 1732 { 1733 struct ucred peercred; 1734 if (len > sizeof(peercred)) 1735 len = sizeof(peercred); 1736 1737 spin_lock(&sk->sk_peer_lock); 1738 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred); 1739 spin_unlock(&sk->sk_peer_lock); 1740 1741 if (copy_to_sockptr(optval, &peercred, len)) 1742 return -EFAULT; 1743 goto lenout; 1744 } 1745 1746 case SO_PEERGROUPS: 1747 { 1748 const struct cred *cred; 1749 int ret, n; 1750 1751 cred = sk_get_peer_cred(sk); 1752 if (!cred) 1753 return -ENODATA; 1754 1755 n = cred->group_info->ngroups; 1756 if (len < n * sizeof(gid_t)) { 1757 len = n * sizeof(gid_t); 1758 put_cred(cred); 1759 return copy_to_sockptr(optlen, &len, sizeof(int)) ? -EFAULT : -ERANGE; 1760 } 1761 len = n * sizeof(gid_t); 1762 1763 ret = groups_to_user(optval, cred->group_info); 1764 put_cred(cred); 1765 if (ret) 1766 return ret; 1767 goto lenout; 1768 } 1769 1770 case SO_PEERNAME: 1771 { 1772 char address[128]; 1773 1774 lv = sock->ops->getname(sock, (struct sockaddr *)address, 2); 1775 if (lv < 0) 1776 return -ENOTCONN; 1777 if (lv < len) 1778 return -EINVAL; 1779 if (copy_to_sockptr(optval, address, len)) 1780 return -EFAULT; 1781 goto lenout; 1782 } 1783 1784 /* Dubious BSD thing... Probably nobody even uses it, but 1785 * the UNIX standard wants it for whatever reason... -DaveM 1786 */ 1787 case SO_ACCEPTCONN: 1788 v.val = sk->sk_state == TCP_LISTEN; 1789 break; 1790 1791 case SO_PASSSEC: 1792 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags); 1793 break; 1794 1795 case SO_PEERSEC: 1796 return security_socket_getpeersec_stream(sock, 1797 optval, optlen, len); 1798 1799 case SO_MARK: 1800 v.val = sk->sk_mark; 1801 break; 1802 1803 case SO_RCVMARK: 1804 v.val = sock_flag(sk, SOCK_RCVMARK); 1805 break; 1806 1807 case SO_RXQ_OVFL: 1808 v.val = sock_flag(sk, SOCK_RXQ_OVFL); 1809 break; 1810 1811 case SO_WIFI_STATUS: 1812 v.val = sock_flag(sk, SOCK_WIFI_STATUS); 1813 break; 1814 1815 case SO_PEEK_OFF: 1816 if (!sock->ops->set_peek_off) 1817 return -EOPNOTSUPP; 1818 1819 v.val = sk->sk_peek_off; 1820 break; 1821 case SO_NOFCS: 1822 v.val = sock_flag(sk, SOCK_NOFCS); 1823 break; 1824 1825 case SO_BINDTODEVICE: 1826 return sock_getbindtodevice(sk, optval, optlen, len); 1827 1828 case SO_GET_FILTER: 1829 len = sk_get_filter(sk, optval, len); 1830 if (len < 0) 1831 return len; 1832 1833 goto lenout; 1834 1835 case SO_LOCK_FILTER: 1836 v.val = sock_flag(sk, SOCK_FILTER_LOCKED); 1837 break; 1838 1839 case SO_BPF_EXTENSIONS: 1840 v.val = bpf_tell_extensions(); 1841 break; 1842 1843 case SO_SELECT_ERR_QUEUE: 1844 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE); 1845 break; 1846 1847 #ifdef CONFIG_NET_RX_BUSY_POLL 1848 case SO_BUSY_POLL: 1849 v.val = sk->sk_ll_usec; 1850 break; 1851 case SO_PREFER_BUSY_POLL: 1852 v.val = READ_ONCE(sk->sk_prefer_busy_poll); 1853 break; 1854 #endif 1855 1856 case SO_MAX_PACING_RATE: 1857 if (sizeof(v.ulval) != sizeof(v.val) && len >= sizeof(v.ulval)) { 1858 lv = sizeof(v.ulval); 1859 v.ulval = sk->sk_max_pacing_rate; 1860 } else { 1861 /* 32bit version */ 1862 v.val = min_t(unsigned long, sk->sk_max_pacing_rate, ~0U); 1863 } 1864 break; 1865 1866 case SO_INCOMING_CPU: 1867 v.val = READ_ONCE(sk->sk_incoming_cpu); 1868 break; 1869 1870 case SO_MEMINFO: 1871 { 1872 u32 meminfo[SK_MEMINFO_VARS]; 1873 1874 sk_get_meminfo(sk, meminfo); 1875 1876 len = min_t(unsigned int, len, sizeof(meminfo)); 1877 if (copy_to_sockptr(optval, &meminfo, len)) 1878 return -EFAULT; 1879 1880 goto lenout; 1881 } 1882 1883 #ifdef CONFIG_NET_RX_BUSY_POLL 1884 case SO_INCOMING_NAPI_ID: 1885 v.val = READ_ONCE(sk->sk_napi_id); 1886 1887 /* aggregate non-NAPI IDs down to 0 */ 1888 if (v.val < MIN_NAPI_ID) 1889 v.val = 0; 1890 1891 break; 1892 #endif 1893 1894 case SO_COOKIE: 1895 lv = sizeof(u64); 1896 if (len < lv) 1897 return -EINVAL; 1898 v.val64 = sock_gen_cookie(sk); 1899 break; 1900 1901 case SO_ZEROCOPY: 1902 v.val = sock_flag(sk, SOCK_ZEROCOPY); 1903 break; 1904 1905 case SO_TXTIME: 1906 lv = sizeof(v.txtime); 1907 v.txtime.clockid = sk->sk_clockid; 1908 v.txtime.flags |= sk->sk_txtime_deadline_mode ? 1909 SOF_TXTIME_DEADLINE_MODE : 0; 1910 v.txtime.flags |= sk->sk_txtime_report_errors ? 1911 SOF_TXTIME_REPORT_ERRORS : 0; 1912 break; 1913 1914 case SO_BINDTOIFINDEX: 1915 v.val = READ_ONCE(sk->sk_bound_dev_if); 1916 break; 1917 1918 case SO_NETNS_COOKIE: 1919 lv = sizeof(u64); 1920 if (len != lv) 1921 return -EINVAL; 1922 v.val64 = sock_net(sk)->net_cookie; 1923 break; 1924 1925 case SO_BUF_LOCK: 1926 v.val = sk->sk_userlocks & SOCK_BUF_LOCK_MASK; 1927 break; 1928 1929 case SO_RESERVE_MEM: 1930 v.val = sk->sk_reserved_mem; 1931 break; 1932 1933 case SO_TXREHASH: 1934 v.val = sk->sk_txrehash; 1935 break; 1936 1937 default: 1938 /* We implement the SO_SNDLOWAT etc to not be settable 1939 * (1003.1g 7). 1940 */ 1941 return -ENOPROTOOPT; 1942 } 1943 1944 if (len > lv) 1945 len = lv; 1946 if (copy_to_sockptr(optval, &v, len)) 1947 return -EFAULT; 1948 lenout: 1949 if (copy_to_sockptr(optlen, &len, sizeof(int))) 1950 return -EFAULT; 1951 return 0; 1952 } 1953 1954 int sock_getsockopt(struct socket *sock, int level, int optname, 1955 char __user *optval, int __user *optlen) 1956 { 1957 return sk_getsockopt(sock->sk, level, optname, 1958 USER_SOCKPTR(optval), 1959 USER_SOCKPTR(optlen)); 1960 } 1961 1962 /* 1963 * Initialize an sk_lock. 1964 * 1965 * (We also register the sk_lock with the lock validator.) 1966 */ 1967 static inline void sock_lock_init(struct sock *sk) 1968 { 1969 if (sk->sk_kern_sock) 1970 sock_lock_init_class_and_name( 1971 sk, 1972 af_family_kern_slock_key_strings[sk->sk_family], 1973 af_family_kern_slock_keys + sk->sk_family, 1974 af_family_kern_key_strings[sk->sk_family], 1975 af_family_kern_keys + sk->sk_family); 1976 else 1977 sock_lock_init_class_and_name( 1978 sk, 1979 af_family_slock_key_strings[sk->sk_family], 1980 af_family_slock_keys + sk->sk_family, 1981 af_family_key_strings[sk->sk_family], 1982 af_family_keys + sk->sk_family); 1983 } 1984 1985 /* 1986 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet, 1987 * even temporarly, because of RCU lookups. sk_node should also be left as is. 1988 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end 1989 */ 1990 static void sock_copy(struct sock *nsk, const struct sock *osk) 1991 { 1992 const struct proto *prot = READ_ONCE(osk->sk_prot); 1993 #ifdef CONFIG_SECURITY_NETWORK 1994 void *sptr = nsk->sk_security; 1995 #endif 1996 1997 /* If we move sk_tx_queue_mapping out of the private section, 1998 * we must check if sk_tx_queue_clear() is called after 1999 * sock_copy() in sk_clone_lock(). 2000 */ 2001 BUILD_BUG_ON(offsetof(struct sock, sk_tx_queue_mapping) < 2002 offsetof(struct sock, sk_dontcopy_begin) || 2003 offsetof(struct sock, sk_tx_queue_mapping) >= 2004 offsetof(struct sock, sk_dontcopy_end)); 2005 2006 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin)); 2007 2008 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end, 2009 prot->obj_size - offsetof(struct sock, sk_dontcopy_end)); 2010 2011 #ifdef CONFIG_SECURITY_NETWORK 2012 nsk->sk_security = sptr; 2013 security_sk_clone(osk, nsk); 2014 #endif 2015 } 2016 2017 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority, 2018 int family) 2019 { 2020 struct sock *sk; 2021 struct kmem_cache *slab; 2022 2023 slab = prot->slab; 2024 if (slab != NULL) { 2025 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO); 2026 if (!sk) 2027 return sk; 2028 if (want_init_on_alloc(priority)) 2029 sk_prot_clear_nulls(sk, prot->obj_size); 2030 } else 2031 sk = kmalloc(prot->obj_size, priority); 2032 2033 if (sk != NULL) { 2034 if (security_sk_alloc(sk, family, priority)) 2035 goto out_free; 2036 2037 if (!try_module_get(prot->owner)) 2038 goto out_free_sec; 2039 } 2040 2041 return sk; 2042 2043 out_free_sec: 2044 security_sk_free(sk); 2045 out_free: 2046 if (slab != NULL) 2047 kmem_cache_free(slab, sk); 2048 else 2049 kfree(sk); 2050 return NULL; 2051 } 2052 2053 static void sk_prot_free(struct proto *prot, struct sock *sk) 2054 { 2055 struct kmem_cache *slab; 2056 struct module *owner; 2057 2058 owner = prot->owner; 2059 slab = prot->slab; 2060 2061 cgroup_sk_free(&sk->sk_cgrp_data); 2062 mem_cgroup_sk_free(sk); 2063 security_sk_free(sk); 2064 if (slab != NULL) 2065 kmem_cache_free(slab, sk); 2066 else 2067 kfree(sk); 2068 module_put(owner); 2069 } 2070 2071 /** 2072 * sk_alloc - All socket objects are allocated here 2073 * @net: the applicable net namespace 2074 * @family: protocol family 2075 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 2076 * @prot: struct proto associated with this new sock instance 2077 * @kern: is this to be a kernel socket? 2078 */ 2079 struct sock *sk_alloc(struct net *net, int family, gfp_t priority, 2080 struct proto *prot, int kern) 2081 { 2082 struct sock *sk; 2083 2084 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family); 2085 if (sk) { 2086 sk->sk_family = family; 2087 /* 2088 * See comment in struct sock definition to understand 2089 * why we need sk_prot_creator -acme 2090 */ 2091 sk->sk_prot = sk->sk_prot_creator = prot; 2092 sk->sk_kern_sock = kern; 2093 sock_lock_init(sk); 2094 sk->sk_net_refcnt = kern ? 0 : 1; 2095 if (likely(sk->sk_net_refcnt)) { 2096 get_net_track(net, &sk->ns_tracker, priority); 2097 sock_inuse_add(net, 1); 2098 } 2099 2100 sock_net_set(sk, net); 2101 refcount_set(&sk->sk_wmem_alloc, 1); 2102 2103 mem_cgroup_sk_alloc(sk); 2104 cgroup_sk_alloc(&sk->sk_cgrp_data); 2105 sock_update_classid(&sk->sk_cgrp_data); 2106 sock_update_netprioidx(&sk->sk_cgrp_data); 2107 sk_tx_queue_clear(sk); 2108 } 2109 2110 return sk; 2111 } 2112 EXPORT_SYMBOL(sk_alloc); 2113 2114 /* Sockets having SOCK_RCU_FREE will call this function after one RCU 2115 * grace period. This is the case for UDP sockets and TCP listeners. 2116 */ 2117 static void __sk_destruct(struct rcu_head *head) 2118 { 2119 struct sock *sk = container_of(head, struct sock, sk_rcu); 2120 struct sk_filter *filter; 2121 2122 if (sk->sk_destruct) 2123 sk->sk_destruct(sk); 2124 2125 filter = rcu_dereference_check(sk->sk_filter, 2126 refcount_read(&sk->sk_wmem_alloc) == 0); 2127 if (filter) { 2128 sk_filter_uncharge(sk, filter); 2129 RCU_INIT_POINTER(sk->sk_filter, NULL); 2130 } 2131 2132 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP); 2133 2134 #ifdef CONFIG_BPF_SYSCALL 2135 bpf_sk_storage_free(sk); 2136 #endif 2137 2138 if (atomic_read(&sk->sk_omem_alloc)) 2139 pr_debug("%s: optmem leakage (%d bytes) detected\n", 2140 __func__, atomic_read(&sk->sk_omem_alloc)); 2141 2142 if (sk->sk_frag.page) { 2143 put_page(sk->sk_frag.page); 2144 sk->sk_frag.page = NULL; 2145 } 2146 2147 /* We do not need to acquire sk->sk_peer_lock, we are the last user. */ 2148 put_cred(sk->sk_peer_cred); 2149 put_pid(sk->sk_peer_pid); 2150 2151 if (likely(sk->sk_net_refcnt)) 2152 put_net_track(sock_net(sk), &sk->ns_tracker); 2153 sk_prot_free(sk->sk_prot_creator, sk); 2154 } 2155 2156 void sk_destruct(struct sock *sk) 2157 { 2158 bool use_call_rcu = sock_flag(sk, SOCK_RCU_FREE); 2159 2160 if (rcu_access_pointer(sk->sk_reuseport_cb)) { 2161 reuseport_detach_sock(sk); 2162 use_call_rcu = true; 2163 } 2164 2165 if (use_call_rcu) 2166 call_rcu(&sk->sk_rcu, __sk_destruct); 2167 else 2168 __sk_destruct(&sk->sk_rcu); 2169 } 2170 2171 static void __sk_free(struct sock *sk) 2172 { 2173 if (likely(sk->sk_net_refcnt)) 2174 sock_inuse_add(sock_net(sk), -1); 2175 2176 if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk))) 2177 sock_diag_broadcast_destroy(sk); 2178 else 2179 sk_destruct(sk); 2180 } 2181 2182 void sk_free(struct sock *sk) 2183 { 2184 /* 2185 * We subtract one from sk_wmem_alloc and can know if 2186 * some packets are still in some tx queue. 2187 * If not null, sock_wfree() will call __sk_free(sk) later 2188 */ 2189 if (refcount_dec_and_test(&sk->sk_wmem_alloc)) 2190 __sk_free(sk); 2191 } 2192 EXPORT_SYMBOL(sk_free); 2193 2194 static void sk_init_common(struct sock *sk) 2195 { 2196 skb_queue_head_init(&sk->sk_receive_queue); 2197 skb_queue_head_init(&sk->sk_write_queue); 2198 skb_queue_head_init(&sk->sk_error_queue); 2199 2200 rwlock_init(&sk->sk_callback_lock); 2201 lockdep_set_class_and_name(&sk->sk_receive_queue.lock, 2202 af_rlock_keys + sk->sk_family, 2203 af_family_rlock_key_strings[sk->sk_family]); 2204 lockdep_set_class_and_name(&sk->sk_write_queue.lock, 2205 af_wlock_keys + sk->sk_family, 2206 af_family_wlock_key_strings[sk->sk_family]); 2207 lockdep_set_class_and_name(&sk->sk_error_queue.lock, 2208 af_elock_keys + sk->sk_family, 2209 af_family_elock_key_strings[sk->sk_family]); 2210 lockdep_set_class_and_name(&sk->sk_callback_lock, 2211 af_callback_keys + sk->sk_family, 2212 af_family_clock_key_strings[sk->sk_family]); 2213 } 2214 2215 /** 2216 * sk_clone_lock - clone a socket, and lock its clone 2217 * @sk: the socket to clone 2218 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 2219 * 2220 * Caller must unlock socket even in error path (bh_unlock_sock(newsk)) 2221 */ 2222 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority) 2223 { 2224 struct proto *prot = READ_ONCE(sk->sk_prot); 2225 struct sk_filter *filter; 2226 bool is_charged = true; 2227 struct sock *newsk; 2228 2229 newsk = sk_prot_alloc(prot, priority, sk->sk_family); 2230 if (!newsk) 2231 goto out; 2232 2233 sock_copy(newsk, sk); 2234 2235 newsk->sk_prot_creator = prot; 2236 2237 /* SANITY */ 2238 if (likely(newsk->sk_net_refcnt)) { 2239 get_net_track(sock_net(newsk), &newsk->ns_tracker, priority); 2240 sock_inuse_add(sock_net(newsk), 1); 2241 } 2242 sk_node_init(&newsk->sk_node); 2243 sock_lock_init(newsk); 2244 bh_lock_sock(newsk); 2245 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL; 2246 newsk->sk_backlog.len = 0; 2247 2248 atomic_set(&newsk->sk_rmem_alloc, 0); 2249 2250 /* sk_wmem_alloc set to one (see sk_free() and sock_wfree()) */ 2251 refcount_set(&newsk->sk_wmem_alloc, 1); 2252 2253 atomic_set(&newsk->sk_omem_alloc, 0); 2254 sk_init_common(newsk); 2255 2256 newsk->sk_dst_cache = NULL; 2257 newsk->sk_dst_pending_confirm = 0; 2258 newsk->sk_wmem_queued = 0; 2259 newsk->sk_forward_alloc = 0; 2260 newsk->sk_reserved_mem = 0; 2261 atomic_set(&newsk->sk_drops, 0); 2262 newsk->sk_send_head = NULL; 2263 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK; 2264 atomic_set(&newsk->sk_zckey, 0); 2265 2266 sock_reset_flag(newsk, SOCK_DONE); 2267 2268 /* sk->sk_memcg will be populated at accept() time */ 2269 newsk->sk_memcg = NULL; 2270 2271 cgroup_sk_clone(&newsk->sk_cgrp_data); 2272 2273 rcu_read_lock(); 2274 filter = rcu_dereference(sk->sk_filter); 2275 if (filter != NULL) 2276 /* though it's an empty new sock, the charging may fail 2277 * if sysctl_optmem_max was changed between creation of 2278 * original socket and cloning 2279 */ 2280 is_charged = sk_filter_charge(newsk, filter); 2281 RCU_INIT_POINTER(newsk->sk_filter, filter); 2282 rcu_read_unlock(); 2283 2284 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) { 2285 /* We need to make sure that we don't uncharge the new 2286 * socket if we couldn't charge it in the first place 2287 * as otherwise we uncharge the parent's filter. 2288 */ 2289 if (!is_charged) 2290 RCU_INIT_POINTER(newsk->sk_filter, NULL); 2291 sk_free_unlock_clone(newsk); 2292 newsk = NULL; 2293 goto out; 2294 } 2295 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL); 2296 2297 if (bpf_sk_storage_clone(sk, newsk)) { 2298 sk_free_unlock_clone(newsk); 2299 newsk = NULL; 2300 goto out; 2301 } 2302 2303 /* Clear sk_user_data if parent had the pointer tagged 2304 * as not suitable for copying when cloning. 2305 */ 2306 if (sk_user_data_is_nocopy(newsk)) 2307 newsk->sk_user_data = NULL; 2308 2309 newsk->sk_err = 0; 2310 newsk->sk_err_soft = 0; 2311 newsk->sk_priority = 0; 2312 newsk->sk_incoming_cpu = raw_smp_processor_id(); 2313 2314 /* Before updating sk_refcnt, we must commit prior changes to memory 2315 * (Documentation/RCU/rculist_nulls.rst for details) 2316 */ 2317 smp_wmb(); 2318 refcount_set(&newsk->sk_refcnt, 2); 2319 2320 /* Increment the counter in the same struct proto as the master 2321 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that 2322 * is the same as sk->sk_prot->socks, as this field was copied 2323 * with memcpy). 2324 * 2325 * This _changes_ the previous behaviour, where 2326 * tcp_create_openreq_child always was incrementing the 2327 * equivalent to tcp_prot->socks (inet_sock_nr), so this have 2328 * to be taken into account in all callers. -acme 2329 */ 2330 sk_refcnt_debug_inc(newsk); 2331 sk_set_socket(newsk, NULL); 2332 sk_tx_queue_clear(newsk); 2333 RCU_INIT_POINTER(newsk->sk_wq, NULL); 2334 2335 if (newsk->sk_prot->sockets_allocated) 2336 sk_sockets_allocated_inc(newsk); 2337 2338 if (sock_needs_netstamp(sk) && newsk->sk_flags & SK_FLAGS_TIMESTAMP) 2339 net_enable_timestamp(); 2340 out: 2341 return newsk; 2342 } 2343 EXPORT_SYMBOL_GPL(sk_clone_lock); 2344 2345 void sk_free_unlock_clone(struct sock *sk) 2346 { 2347 /* It is still raw copy of parent, so invalidate 2348 * destructor and make plain sk_free() */ 2349 sk->sk_destruct = NULL; 2350 bh_unlock_sock(sk); 2351 sk_free(sk); 2352 } 2353 EXPORT_SYMBOL_GPL(sk_free_unlock_clone); 2354 2355 static void sk_trim_gso_size(struct sock *sk) 2356 { 2357 if (sk->sk_gso_max_size <= GSO_LEGACY_MAX_SIZE) 2358 return; 2359 #if IS_ENABLED(CONFIG_IPV6) 2360 if (sk->sk_family == AF_INET6 && 2361 sk_is_tcp(sk) && 2362 !ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr)) 2363 return; 2364 #endif 2365 sk->sk_gso_max_size = GSO_LEGACY_MAX_SIZE; 2366 } 2367 2368 void sk_setup_caps(struct sock *sk, struct dst_entry *dst) 2369 { 2370 u32 max_segs = 1; 2371 2372 sk_dst_set(sk, dst); 2373 sk->sk_route_caps = dst->dev->features; 2374 if (sk_is_tcp(sk)) 2375 sk->sk_route_caps |= NETIF_F_GSO; 2376 if (sk->sk_route_caps & NETIF_F_GSO) 2377 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE; 2378 if (unlikely(sk->sk_gso_disabled)) 2379 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 2380 if (sk_can_gso(sk)) { 2381 if (dst->header_len && !xfrm_dst_offload_ok(dst)) { 2382 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 2383 } else { 2384 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM; 2385 /* pairs with the WRITE_ONCE() in netif_set_gso_max_size() */ 2386 sk->sk_gso_max_size = READ_ONCE(dst->dev->gso_max_size); 2387 sk_trim_gso_size(sk); 2388 sk->sk_gso_max_size -= (MAX_TCP_HEADER + 1); 2389 /* pairs with the WRITE_ONCE() in netif_set_gso_max_segs() */ 2390 max_segs = max_t(u32, READ_ONCE(dst->dev->gso_max_segs), 1); 2391 } 2392 } 2393 sk->sk_gso_max_segs = max_segs; 2394 } 2395 EXPORT_SYMBOL_GPL(sk_setup_caps); 2396 2397 /* 2398 * Simple resource managers for sockets. 2399 */ 2400 2401 2402 /* 2403 * Write buffer destructor automatically called from kfree_skb. 2404 */ 2405 void sock_wfree(struct sk_buff *skb) 2406 { 2407 struct sock *sk = skb->sk; 2408 unsigned int len = skb->truesize; 2409 bool free; 2410 2411 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) { 2412 if (sock_flag(sk, SOCK_RCU_FREE) && 2413 sk->sk_write_space == sock_def_write_space) { 2414 rcu_read_lock(); 2415 free = refcount_sub_and_test(len, &sk->sk_wmem_alloc); 2416 sock_def_write_space_wfree(sk); 2417 rcu_read_unlock(); 2418 if (unlikely(free)) 2419 __sk_free(sk); 2420 return; 2421 } 2422 2423 /* 2424 * Keep a reference on sk_wmem_alloc, this will be released 2425 * after sk_write_space() call 2426 */ 2427 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc)); 2428 sk->sk_write_space(sk); 2429 len = 1; 2430 } 2431 /* 2432 * if sk_wmem_alloc reaches 0, we must finish what sk_free() 2433 * could not do because of in-flight packets 2434 */ 2435 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc)) 2436 __sk_free(sk); 2437 } 2438 EXPORT_SYMBOL(sock_wfree); 2439 2440 /* This variant of sock_wfree() is used by TCP, 2441 * since it sets SOCK_USE_WRITE_QUEUE. 2442 */ 2443 void __sock_wfree(struct sk_buff *skb) 2444 { 2445 struct sock *sk = skb->sk; 2446 2447 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc)) 2448 __sk_free(sk); 2449 } 2450 2451 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk) 2452 { 2453 skb_orphan(skb); 2454 skb->sk = sk; 2455 #ifdef CONFIG_INET 2456 if (unlikely(!sk_fullsock(sk))) { 2457 skb->destructor = sock_edemux; 2458 sock_hold(sk); 2459 return; 2460 } 2461 #endif 2462 skb->destructor = sock_wfree; 2463 skb_set_hash_from_sk(skb, sk); 2464 /* 2465 * We used to take a refcount on sk, but following operation 2466 * is enough to guarantee sk_free() wont free this sock until 2467 * all in-flight packets are completed 2468 */ 2469 refcount_add(skb->truesize, &sk->sk_wmem_alloc); 2470 } 2471 EXPORT_SYMBOL(skb_set_owner_w); 2472 2473 static bool can_skb_orphan_partial(const struct sk_buff *skb) 2474 { 2475 #ifdef CONFIG_TLS_DEVICE 2476 /* Drivers depend on in-order delivery for crypto offload, 2477 * partial orphan breaks out-of-order-OK logic. 2478 */ 2479 if (skb->decrypted) 2480 return false; 2481 #endif 2482 return (skb->destructor == sock_wfree || 2483 (IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree)); 2484 } 2485 2486 /* This helper is used by netem, as it can hold packets in its 2487 * delay queue. We want to allow the owner socket to send more 2488 * packets, as if they were already TX completed by a typical driver. 2489 * But we also want to keep skb->sk set because some packet schedulers 2490 * rely on it (sch_fq for example). 2491 */ 2492 void skb_orphan_partial(struct sk_buff *skb) 2493 { 2494 if (skb_is_tcp_pure_ack(skb)) 2495 return; 2496 2497 if (can_skb_orphan_partial(skb) && skb_set_owner_sk_safe(skb, skb->sk)) 2498 return; 2499 2500 skb_orphan(skb); 2501 } 2502 EXPORT_SYMBOL(skb_orphan_partial); 2503 2504 /* 2505 * Read buffer destructor automatically called from kfree_skb. 2506 */ 2507 void sock_rfree(struct sk_buff *skb) 2508 { 2509 struct sock *sk = skb->sk; 2510 unsigned int len = skb->truesize; 2511 2512 atomic_sub(len, &sk->sk_rmem_alloc); 2513 sk_mem_uncharge(sk, len); 2514 } 2515 EXPORT_SYMBOL(sock_rfree); 2516 2517 /* 2518 * Buffer destructor for skbs that are not used directly in read or write 2519 * path, e.g. for error handler skbs. Automatically called from kfree_skb. 2520 */ 2521 void sock_efree(struct sk_buff *skb) 2522 { 2523 sock_put(skb->sk); 2524 } 2525 EXPORT_SYMBOL(sock_efree); 2526 2527 /* Buffer destructor for prefetch/receive path where reference count may 2528 * not be held, e.g. for listen sockets. 2529 */ 2530 #ifdef CONFIG_INET 2531 void sock_pfree(struct sk_buff *skb) 2532 { 2533 if (sk_is_refcounted(skb->sk)) 2534 sock_gen_put(skb->sk); 2535 } 2536 EXPORT_SYMBOL(sock_pfree); 2537 #endif /* CONFIG_INET */ 2538 2539 kuid_t sock_i_uid(struct sock *sk) 2540 { 2541 kuid_t uid; 2542 2543 read_lock_bh(&sk->sk_callback_lock); 2544 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID; 2545 read_unlock_bh(&sk->sk_callback_lock); 2546 return uid; 2547 } 2548 EXPORT_SYMBOL(sock_i_uid); 2549 2550 unsigned long sock_i_ino(struct sock *sk) 2551 { 2552 unsigned long ino; 2553 2554 read_lock_bh(&sk->sk_callback_lock); 2555 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0; 2556 read_unlock_bh(&sk->sk_callback_lock); 2557 return ino; 2558 } 2559 EXPORT_SYMBOL(sock_i_ino); 2560 2561 /* 2562 * Allocate a skb from the socket's send buffer. 2563 */ 2564 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 2565 gfp_t priority) 2566 { 2567 if (force || 2568 refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) { 2569 struct sk_buff *skb = alloc_skb(size, priority); 2570 2571 if (skb) { 2572 skb_set_owner_w(skb, sk); 2573 return skb; 2574 } 2575 } 2576 return NULL; 2577 } 2578 EXPORT_SYMBOL(sock_wmalloc); 2579 2580 static void sock_ofree(struct sk_buff *skb) 2581 { 2582 struct sock *sk = skb->sk; 2583 2584 atomic_sub(skb->truesize, &sk->sk_omem_alloc); 2585 } 2586 2587 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size, 2588 gfp_t priority) 2589 { 2590 struct sk_buff *skb; 2591 2592 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */ 2593 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) > 2594 READ_ONCE(sysctl_optmem_max)) 2595 return NULL; 2596 2597 skb = alloc_skb(size, priority); 2598 if (!skb) 2599 return NULL; 2600 2601 atomic_add(skb->truesize, &sk->sk_omem_alloc); 2602 skb->sk = sk; 2603 skb->destructor = sock_ofree; 2604 return skb; 2605 } 2606 2607 /* 2608 * Allocate a memory block from the socket's option memory buffer. 2609 */ 2610 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority) 2611 { 2612 int optmem_max = READ_ONCE(sysctl_optmem_max); 2613 2614 if ((unsigned int)size <= optmem_max && 2615 atomic_read(&sk->sk_omem_alloc) + size < optmem_max) { 2616 void *mem; 2617 /* First do the add, to avoid the race if kmalloc 2618 * might sleep. 2619 */ 2620 atomic_add(size, &sk->sk_omem_alloc); 2621 mem = kmalloc(size, priority); 2622 if (mem) 2623 return mem; 2624 atomic_sub(size, &sk->sk_omem_alloc); 2625 } 2626 return NULL; 2627 } 2628 EXPORT_SYMBOL(sock_kmalloc); 2629 2630 /* Free an option memory block. Note, we actually want the inline 2631 * here as this allows gcc to detect the nullify and fold away the 2632 * condition entirely. 2633 */ 2634 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size, 2635 const bool nullify) 2636 { 2637 if (WARN_ON_ONCE(!mem)) 2638 return; 2639 if (nullify) 2640 kfree_sensitive(mem); 2641 else 2642 kfree(mem); 2643 atomic_sub(size, &sk->sk_omem_alloc); 2644 } 2645 2646 void sock_kfree_s(struct sock *sk, void *mem, int size) 2647 { 2648 __sock_kfree_s(sk, mem, size, false); 2649 } 2650 EXPORT_SYMBOL(sock_kfree_s); 2651 2652 void sock_kzfree_s(struct sock *sk, void *mem, int size) 2653 { 2654 __sock_kfree_s(sk, mem, size, true); 2655 } 2656 EXPORT_SYMBOL(sock_kzfree_s); 2657 2658 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock. 2659 I think, these locks should be removed for datagram sockets. 2660 */ 2661 static long sock_wait_for_wmem(struct sock *sk, long timeo) 2662 { 2663 DEFINE_WAIT(wait); 2664 2665 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk); 2666 for (;;) { 2667 if (!timeo) 2668 break; 2669 if (signal_pending(current)) 2670 break; 2671 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 2672 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 2673 if (refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) 2674 break; 2675 if (sk->sk_shutdown & SEND_SHUTDOWN) 2676 break; 2677 if (sk->sk_err) 2678 break; 2679 timeo = schedule_timeout(timeo); 2680 } 2681 finish_wait(sk_sleep(sk), &wait); 2682 return timeo; 2683 } 2684 2685 2686 /* 2687 * Generic send/receive buffer handlers 2688 */ 2689 2690 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 2691 unsigned long data_len, int noblock, 2692 int *errcode, int max_page_order) 2693 { 2694 struct sk_buff *skb; 2695 long timeo; 2696 int err; 2697 2698 timeo = sock_sndtimeo(sk, noblock); 2699 for (;;) { 2700 err = sock_error(sk); 2701 if (err != 0) 2702 goto failure; 2703 2704 err = -EPIPE; 2705 if (sk->sk_shutdown & SEND_SHUTDOWN) 2706 goto failure; 2707 2708 if (sk_wmem_alloc_get(sk) < READ_ONCE(sk->sk_sndbuf)) 2709 break; 2710 2711 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk); 2712 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 2713 err = -EAGAIN; 2714 if (!timeo) 2715 goto failure; 2716 if (signal_pending(current)) 2717 goto interrupted; 2718 timeo = sock_wait_for_wmem(sk, timeo); 2719 } 2720 skb = alloc_skb_with_frags(header_len, data_len, max_page_order, 2721 errcode, sk->sk_allocation); 2722 if (skb) 2723 skb_set_owner_w(skb, sk); 2724 return skb; 2725 2726 interrupted: 2727 err = sock_intr_errno(timeo); 2728 failure: 2729 *errcode = err; 2730 return NULL; 2731 } 2732 EXPORT_SYMBOL(sock_alloc_send_pskb); 2733 2734 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg, 2735 struct sockcm_cookie *sockc) 2736 { 2737 u32 tsflags; 2738 2739 switch (cmsg->cmsg_type) { 2740 case SO_MARK: 2741 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) && 2742 !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) 2743 return -EPERM; 2744 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 2745 return -EINVAL; 2746 sockc->mark = *(u32 *)CMSG_DATA(cmsg); 2747 break; 2748 case SO_TIMESTAMPING_OLD: 2749 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 2750 return -EINVAL; 2751 2752 tsflags = *(u32 *)CMSG_DATA(cmsg); 2753 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK) 2754 return -EINVAL; 2755 2756 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK; 2757 sockc->tsflags |= tsflags; 2758 break; 2759 case SCM_TXTIME: 2760 if (!sock_flag(sk, SOCK_TXTIME)) 2761 return -EINVAL; 2762 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64))) 2763 return -EINVAL; 2764 sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg)); 2765 break; 2766 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */ 2767 case SCM_RIGHTS: 2768 case SCM_CREDENTIALS: 2769 break; 2770 default: 2771 return -EINVAL; 2772 } 2773 return 0; 2774 } 2775 EXPORT_SYMBOL(__sock_cmsg_send); 2776 2777 int sock_cmsg_send(struct sock *sk, struct msghdr *msg, 2778 struct sockcm_cookie *sockc) 2779 { 2780 struct cmsghdr *cmsg; 2781 int ret; 2782 2783 for_each_cmsghdr(cmsg, msg) { 2784 if (!CMSG_OK(msg, cmsg)) 2785 return -EINVAL; 2786 if (cmsg->cmsg_level != SOL_SOCKET) 2787 continue; 2788 ret = __sock_cmsg_send(sk, msg, cmsg, sockc); 2789 if (ret) 2790 return ret; 2791 } 2792 return 0; 2793 } 2794 EXPORT_SYMBOL(sock_cmsg_send); 2795 2796 static void sk_enter_memory_pressure(struct sock *sk) 2797 { 2798 if (!sk->sk_prot->enter_memory_pressure) 2799 return; 2800 2801 sk->sk_prot->enter_memory_pressure(sk); 2802 } 2803 2804 static void sk_leave_memory_pressure(struct sock *sk) 2805 { 2806 if (sk->sk_prot->leave_memory_pressure) { 2807 sk->sk_prot->leave_memory_pressure(sk); 2808 } else { 2809 unsigned long *memory_pressure = sk->sk_prot->memory_pressure; 2810 2811 if (memory_pressure && READ_ONCE(*memory_pressure)) 2812 WRITE_ONCE(*memory_pressure, 0); 2813 } 2814 } 2815 2816 DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key); 2817 2818 /** 2819 * skb_page_frag_refill - check that a page_frag contains enough room 2820 * @sz: minimum size of the fragment we want to get 2821 * @pfrag: pointer to page_frag 2822 * @gfp: priority for memory allocation 2823 * 2824 * Note: While this allocator tries to use high order pages, there is 2825 * no guarantee that allocations succeed. Therefore, @sz MUST be 2826 * less or equal than PAGE_SIZE. 2827 */ 2828 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp) 2829 { 2830 if (pfrag->page) { 2831 if (page_ref_count(pfrag->page) == 1) { 2832 pfrag->offset = 0; 2833 return true; 2834 } 2835 if (pfrag->offset + sz <= pfrag->size) 2836 return true; 2837 put_page(pfrag->page); 2838 } 2839 2840 pfrag->offset = 0; 2841 if (SKB_FRAG_PAGE_ORDER && 2842 !static_branch_unlikely(&net_high_order_alloc_disable_key)) { 2843 /* Avoid direct reclaim but allow kswapd to wake */ 2844 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) | 2845 __GFP_COMP | __GFP_NOWARN | 2846 __GFP_NORETRY, 2847 SKB_FRAG_PAGE_ORDER); 2848 if (likely(pfrag->page)) { 2849 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER; 2850 return true; 2851 } 2852 } 2853 pfrag->page = alloc_page(gfp); 2854 if (likely(pfrag->page)) { 2855 pfrag->size = PAGE_SIZE; 2856 return true; 2857 } 2858 return false; 2859 } 2860 EXPORT_SYMBOL(skb_page_frag_refill); 2861 2862 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag) 2863 { 2864 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation))) 2865 return true; 2866 2867 sk_enter_memory_pressure(sk); 2868 sk_stream_moderate_sndbuf(sk); 2869 return false; 2870 } 2871 EXPORT_SYMBOL(sk_page_frag_refill); 2872 2873 void __lock_sock(struct sock *sk) 2874 __releases(&sk->sk_lock.slock) 2875 __acquires(&sk->sk_lock.slock) 2876 { 2877 DEFINE_WAIT(wait); 2878 2879 for (;;) { 2880 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait, 2881 TASK_UNINTERRUPTIBLE); 2882 spin_unlock_bh(&sk->sk_lock.slock); 2883 schedule(); 2884 spin_lock_bh(&sk->sk_lock.slock); 2885 if (!sock_owned_by_user(sk)) 2886 break; 2887 } 2888 finish_wait(&sk->sk_lock.wq, &wait); 2889 } 2890 2891 void __release_sock(struct sock *sk) 2892 __releases(&sk->sk_lock.slock) 2893 __acquires(&sk->sk_lock.slock) 2894 { 2895 struct sk_buff *skb, *next; 2896 2897 while ((skb = sk->sk_backlog.head) != NULL) { 2898 sk->sk_backlog.head = sk->sk_backlog.tail = NULL; 2899 2900 spin_unlock_bh(&sk->sk_lock.slock); 2901 2902 do { 2903 next = skb->next; 2904 prefetch(next); 2905 DEBUG_NET_WARN_ON_ONCE(skb_dst_is_noref(skb)); 2906 skb_mark_not_on_list(skb); 2907 sk_backlog_rcv(sk, skb); 2908 2909 cond_resched(); 2910 2911 skb = next; 2912 } while (skb != NULL); 2913 2914 spin_lock_bh(&sk->sk_lock.slock); 2915 } 2916 2917 /* 2918 * Doing the zeroing here guarantee we can not loop forever 2919 * while a wild producer attempts to flood us. 2920 */ 2921 sk->sk_backlog.len = 0; 2922 } 2923 2924 void __sk_flush_backlog(struct sock *sk) 2925 { 2926 spin_lock_bh(&sk->sk_lock.slock); 2927 __release_sock(sk); 2928 spin_unlock_bh(&sk->sk_lock.slock); 2929 } 2930 EXPORT_SYMBOL_GPL(__sk_flush_backlog); 2931 2932 /** 2933 * sk_wait_data - wait for data to arrive at sk_receive_queue 2934 * @sk: sock to wait on 2935 * @timeo: for how long 2936 * @skb: last skb seen on sk_receive_queue 2937 * 2938 * Now socket state including sk->sk_err is changed only under lock, 2939 * hence we may omit checks after joining wait queue. 2940 * We check receive queue before schedule() only as optimization; 2941 * it is very likely that release_sock() added new data. 2942 */ 2943 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb) 2944 { 2945 DEFINE_WAIT_FUNC(wait, woken_wake_function); 2946 int rc; 2947 2948 add_wait_queue(sk_sleep(sk), &wait); 2949 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk); 2950 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait); 2951 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk); 2952 remove_wait_queue(sk_sleep(sk), &wait); 2953 return rc; 2954 } 2955 EXPORT_SYMBOL(sk_wait_data); 2956 2957 /** 2958 * __sk_mem_raise_allocated - increase memory_allocated 2959 * @sk: socket 2960 * @size: memory size to allocate 2961 * @amt: pages to allocate 2962 * @kind: allocation type 2963 * 2964 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc 2965 */ 2966 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind) 2967 { 2968 bool memcg_charge = mem_cgroup_sockets_enabled && sk->sk_memcg; 2969 struct proto *prot = sk->sk_prot; 2970 bool charged = true; 2971 long allocated; 2972 2973 sk_memory_allocated_add(sk, amt); 2974 allocated = sk_memory_allocated(sk); 2975 if (memcg_charge && 2976 !(charged = mem_cgroup_charge_skmem(sk->sk_memcg, amt, 2977 gfp_memcg_charge()))) 2978 goto suppress_allocation; 2979 2980 /* Under limit. */ 2981 if (allocated <= sk_prot_mem_limits(sk, 0)) { 2982 sk_leave_memory_pressure(sk); 2983 return 1; 2984 } 2985 2986 /* Under pressure. */ 2987 if (allocated > sk_prot_mem_limits(sk, 1)) 2988 sk_enter_memory_pressure(sk); 2989 2990 /* Over hard limit. */ 2991 if (allocated > sk_prot_mem_limits(sk, 2)) 2992 goto suppress_allocation; 2993 2994 /* guarantee minimum buffer size under pressure */ 2995 if (kind == SK_MEM_RECV) { 2996 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot)) 2997 return 1; 2998 2999 } else { /* SK_MEM_SEND */ 3000 int wmem0 = sk_get_wmem0(sk, prot); 3001 3002 if (sk->sk_type == SOCK_STREAM) { 3003 if (sk->sk_wmem_queued < wmem0) 3004 return 1; 3005 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) { 3006 return 1; 3007 } 3008 } 3009 3010 if (sk_has_memory_pressure(sk)) { 3011 u64 alloc; 3012 3013 if (!sk_under_memory_pressure(sk)) 3014 return 1; 3015 alloc = sk_sockets_allocated_read_positive(sk); 3016 if (sk_prot_mem_limits(sk, 2) > alloc * 3017 sk_mem_pages(sk->sk_wmem_queued + 3018 atomic_read(&sk->sk_rmem_alloc) + 3019 sk->sk_forward_alloc)) 3020 return 1; 3021 } 3022 3023 suppress_allocation: 3024 3025 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) { 3026 sk_stream_moderate_sndbuf(sk); 3027 3028 /* Fail only if socket is _under_ its sndbuf. 3029 * In this case we cannot block, so that we have to fail. 3030 */ 3031 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf) { 3032 /* Force charge with __GFP_NOFAIL */ 3033 if (memcg_charge && !charged) { 3034 mem_cgroup_charge_skmem(sk->sk_memcg, amt, 3035 gfp_memcg_charge() | __GFP_NOFAIL); 3036 } 3037 return 1; 3038 } 3039 } 3040 3041 if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged)) 3042 trace_sock_exceed_buf_limit(sk, prot, allocated, kind); 3043 3044 sk_memory_allocated_sub(sk, amt); 3045 3046 if (memcg_charge && charged) 3047 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt); 3048 3049 return 0; 3050 } 3051 3052 /** 3053 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated 3054 * @sk: socket 3055 * @size: memory size to allocate 3056 * @kind: allocation type 3057 * 3058 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means 3059 * rmem allocation. This function assumes that protocols which have 3060 * memory_pressure use sk_wmem_queued as write buffer accounting. 3061 */ 3062 int __sk_mem_schedule(struct sock *sk, int size, int kind) 3063 { 3064 int ret, amt = sk_mem_pages(size); 3065 3066 sk->sk_forward_alloc += amt << PAGE_SHIFT; 3067 ret = __sk_mem_raise_allocated(sk, size, amt, kind); 3068 if (!ret) 3069 sk->sk_forward_alloc -= amt << PAGE_SHIFT; 3070 return ret; 3071 } 3072 EXPORT_SYMBOL(__sk_mem_schedule); 3073 3074 /** 3075 * __sk_mem_reduce_allocated - reclaim memory_allocated 3076 * @sk: socket 3077 * @amount: number of quanta 3078 * 3079 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc 3080 */ 3081 void __sk_mem_reduce_allocated(struct sock *sk, int amount) 3082 { 3083 sk_memory_allocated_sub(sk, amount); 3084 3085 if (mem_cgroup_sockets_enabled && sk->sk_memcg) 3086 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount); 3087 3088 if (sk_under_memory_pressure(sk) && 3089 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0))) 3090 sk_leave_memory_pressure(sk); 3091 } 3092 3093 /** 3094 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated 3095 * @sk: socket 3096 * @amount: number of bytes (rounded down to a PAGE_SIZE multiple) 3097 */ 3098 void __sk_mem_reclaim(struct sock *sk, int amount) 3099 { 3100 amount >>= PAGE_SHIFT; 3101 sk->sk_forward_alloc -= amount << PAGE_SHIFT; 3102 __sk_mem_reduce_allocated(sk, amount); 3103 } 3104 EXPORT_SYMBOL(__sk_mem_reclaim); 3105 3106 int sk_set_peek_off(struct sock *sk, int val) 3107 { 3108 sk->sk_peek_off = val; 3109 return 0; 3110 } 3111 EXPORT_SYMBOL_GPL(sk_set_peek_off); 3112 3113 /* 3114 * Set of default routines for initialising struct proto_ops when 3115 * the protocol does not support a particular function. In certain 3116 * cases where it makes no sense for a protocol to have a "do nothing" 3117 * function, some default processing is provided. 3118 */ 3119 3120 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len) 3121 { 3122 return -EOPNOTSUPP; 3123 } 3124 EXPORT_SYMBOL(sock_no_bind); 3125 3126 int sock_no_connect(struct socket *sock, struct sockaddr *saddr, 3127 int len, int flags) 3128 { 3129 return -EOPNOTSUPP; 3130 } 3131 EXPORT_SYMBOL(sock_no_connect); 3132 3133 int sock_no_socketpair(struct socket *sock1, struct socket *sock2) 3134 { 3135 return -EOPNOTSUPP; 3136 } 3137 EXPORT_SYMBOL(sock_no_socketpair); 3138 3139 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags, 3140 bool kern) 3141 { 3142 return -EOPNOTSUPP; 3143 } 3144 EXPORT_SYMBOL(sock_no_accept); 3145 3146 int sock_no_getname(struct socket *sock, struct sockaddr *saddr, 3147 int peer) 3148 { 3149 return -EOPNOTSUPP; 3150 } 3151 EXPORT_SYMBOL(sock_no_getname); 3152 3153 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 3154 { 3155 return -EOPNOTSUPP; 3156 } 3157 EXPORT_SYMBOL(sock_no_ioctl); 3158 3159 int sock_no_listen(struct socket *sock, int backlog) 3160 { 3161 return -EOPNOTSUPP; 3162 } 3163 EXPORT_SYMBOL(sock_no_listen); 3164 3165 int sock_no_shutdown(struct socket *sock, int how) 3166 { 3167 return -EOPNOTSUPP; 3168 } 3169 EXPORT_SYMBOL(sock_no_shutdown); 3170 3171 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len) 3172 { 3173 return -EOPNOTSUPP; 3174 } 3175 EXPORT_SYMBOL(sock_no_sendmsg); 3176 3177 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len) 3178 { 3179 return -EOPNOTSUPP; 3180 } 3181 EXPORT_SYMBOL(sock_no_sendmsg_locked); 3182 3183 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len, 3184 int flags) 3185 { 3186 return -EOPNOTSUPP; 3187 } 3188 EXPORT_SYMBOL(sock_no_recvmsg); 3189 3190 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma) 3191 { 3192 /* Mirror missing mmap method error code */ 3193 return -ENODEV; 3194 } 3195 EXPORT_SYMBOL(sock_no_mmap); 3196 3197 /* 3198 * When a file is received (via SCM_RIGHTS, etc), we must bump the 3199 * various sock-based usage counts. 3200 */ 3201 void __receive_sock(struct file *file) 3202 { 3203 struct socket *sock; 3204 3205 sock = sock_from_file(file); 3206 if (sock) { 3207 sock_update_netprioidx(&sock->sk->sk_cgrp_data); 3208 sock_update_classid(&sock->sk->sk_cgrp_data); 3209 } 3210 } 3211 3212 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags) 3213 { 3214 ssize_t res; 3215 struct msghdr msg = {.msg_flags = flags}; 3216 struct kvec iov; 3217 char *kaddr = kmap(page); 3218 iov.iov_base = kaddr + offset; 3219 iov.iov_len = size; 3220 res = kernel_sendmsg(sock, &msg, &iov, 1, size); 3221 kunmap(page); 3222 return res; 3223 } 3224 EXPORT_SYMBOL(sock_no_sendpage); 3225 3226 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page, 3227 int offset, size_t size, int flags) 3228 { 3229 ssize_t res; 3230 struct msghdr msg = {.msg_flags = flags}; 3231 struct kvec iov; 3232 char *kaddr = kmap(page); 3233 3234 iov.iov_base = kaddr + offset; 3235 iov.iov_len = size; 3236 res = kernel_sendmsg_locked(sk, &msg, &iov, 1, size); 3237 kunmap(page); 3238 return res; 3239 } 3240 EXPORT_SYMBOL(sock_no_sendpage_locked); 3241 3242 /* 3243 * Default Socket Callbacks 3244 */ 3245 3246 static void sock_def_wakeup(struct sock *sk) 3247 { 3248 struct socket_wq *wq; 3249 3250 rcu_read_lock(); 3251 wq = rcu_dereference(sk->sk_wq); 3252 if (skwq_has_sleeper(wq)) 3253 wake_up_interruptible_all(&wq->wait); 3254 rcu_read_unlock(); 3255 } 3256 3257 static void sock_def_error_report(struct sock *sk) 3258 { 3259 struct socket_wq *wq; 3260 3261 rcu_read_lock(); 3262 wq = rcu_dereference(sk->sk_wq); 3263 if (skwq_has_sleeper(wq)) 3264 wake_up_interruptible_poll(&wq->wait, EPOLLERR); 3265 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR); 3266 rcu_read_unlock(); 3267 } 3268 3269 void sock_def_readable(struct sock *sk) 3270 { 3271 struct socket_wq *wq; 3272 3273 rcu_read_lock(); 3274 wq = rcu_dereference(sk->sk_wq); 3275 if (skwq_has_sleeper(wq)) 3276 wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI | 3277 EPOLLRDNORM | EPOLLRDBAND); 3278 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN); 3279 rcu_read_unlock(); 3280 } 3281 3282 static void sock_def_write_space(struct sock *sk) 3283 { 3284 struct socket_wq *wq; 3285 3286 rcu_read_lock(); 3287 3288 /* Do not wake up a writer until he can make "significant" 3289 * progress. --DaveM 3290 */ 3291 if (sock_writeable(sk)) { 3292 wq = rcu_dereference(sk->sk_wq); 3293 if (skwq_has_sleeper(wq)) 3294 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT | 3295 EPOLLWRNORM | EPOLLWRBAND); 3296 3297 /* Should agree with poll, otherwise some programs break */ 3298 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT); 3299 } 3300 3301 rcu_read_unlock(); 3302 } 3303 3304 /* An optimised version of sock_def_write_space(), should only be called 3305 * for SOCK_RCU_FREE sockets under RCU read section and after putting 3306 * ->sk_wmem_alloc. 3307 */ 3308 static void sock_def_write_space_wfree(struct sock *sk) 3309 { 3310 /* Do not wake up a writer until he can make "significant" 3311 * progress. --DaveM 3312 */ 3313 if (sock_writeable(sk)) { 3314 struct socket_wq *wq = rcu_dereference(sk->sk_wq); 3315 3316 /* rely on refcount_sub from sock_wfree() */ 3317 smp_mb__after_atomic(); 3318 if (wq && waitqueue_active(&wq->wait)) 3319 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT | 3320 EPOLLWRNORM | EPOLLWRBAND); 3321 3322 /* Should agree with poll, otherwise some programs break */ 3323 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT); 3324 } 3325 } 3326 3327 static void sock_def_destruct(struct sock *sk) 3328 { 3329 } 3330 3331 void sk_send_sigurg(struct sock *sk) 3332 { 3333 if (sk->sk_socket && sk->sk_socket->file) 3334 if (send_sigurg(&sk->sk_socket->file->f_owner)) 3335 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI); 3336 } 3337 EXPORT_SYMBOL(sk_send_sigurg); 3338 3339 void sk_reset_timer(struct sock *sk, struct timer_list* timer, 3340 unsigned long expires) 3341 { 3342 if (!mod_timer(timer, expires)) 3343 sock_hold(sk); 3344 } 3345 EXPORT_SYMBOL(sk_reset_timer); 3346 3347 void sk_stop_timer(struct sock *sk, struct timer_list* timer) 3348 { 3349 if (del_timer(timer)) 3350 __sock_put(sk); 3351 } 3352 EXPORT_SYMBOL(sk_stop_timer); 3353 3354 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer) 3355 { 3356 if (del_timer_sync(timer)) 3357 __sock_put(sk); 3358 } 3359 EXPORT_SYMBOL(sk_stop_timer_sync); 3360 3361 void sock_init_data(struct socket *sock, struct sock *sk) 3362 { 3363 sk_init_common(sk); 3364 sk->sk_send_head = NULL; 3365 3366 timer_setup(&sk->sk_timer, NULL, 0); 3367 3368 sk->sk_allocation = GFP_KERNEL; 3369 sk->sk_rcvbuf = READ_ONCE(sysctl_rmem_default); 3370 sk->sk_sndbuf = READ_ONCE(sysctl_wmem_default); 3371 sk->sk_state = TCP_CLOSE; 3372 sk_set_socket(sk, sock); 3373 3374 sock_set_flag(sk, SOCK_ZAPPED); 3375 3376 if (sock) { 3377 sk->sk_type = sock->type; 3378 RCU_INIT_POINTER(sk->sk_wq, &sock->wq); 3379 sock->sk = sk; 3380 sk->sk_uid = SOCK_INODE(sock)->i_uid; 3381 } else { 3382 RCU_INIT_POINTER(sk->sk_wq, NULL); 3383 sk->sk_uid = make_kuid(sock_net(sk)->user_ns, 0); 3384 } 3385 3386 rwlock_init(&sk->sk_callback_lock); 3387 if (sk->sk_kern_sock) 3388 lockdep_set_class_and_name( 3389 &sk->sk_callback_lock, 3390 af_kern_callback_keys + sk->sk_family, 3391 af_family_kern_clock_key_strings[sk->sk_family]); 3392 else 3393 lockdep_set_class_and_name( 3394 &sk->sk_callback_lock, 3395 af_callback_keys + sk->sk_family, 3396 af_family_clock_key_strings[sk->sk_family]); 3397 3398 sk->sk_state_change = sock_def_wakeup; 3399 sk->sk_data_ready = sock_def_readable; 3400 sk->sk_write_space = sock_def_write_space; 3401 sk->sk_error_report = sock_def_error_report; 3402 sk->sk_destruct = sock_def_destruct; 3403 3404 sk->sk_frag.page = NULL; 3405 sk->sk_frag.offset = 0; 3406 sk->sk_peek_off = -1; 3407 3408 sk->sk_peer_pid = NULL; 3409 sk->sk_peer_cred = NULL; 3410 spin_lock_init(&sk->sk_peer_lock); 3411 3412 sk->sk_write_pending = 0; 3413 sk->sk_rcvlowat = 1; 3414 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; 3415 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT; 3416 3417 sk->sk_stamp = SK_DEFAULT_STAMP; 3418 #if BITS_PER_LONG==32 3419 seqlock_init(&sk->sk_stamp_seq); 3420 #endif 3421 atomic_set(&sk->sk_zckey, 0); 3422 3423 #ifdef CONFIG_NET_RX_BUSY_POLL 3424 sk->sk_napi_id = 0; 3425 sk->sk_ll_usec = READ_ONCE(sysctl_net_busy_read); 3426 #endif 3427 3428 sk->sk_max_pacing_rate = ~0UL; 3429 sk->sk_pacing_rate = ~0UL; 3430 WRITE_ONCE(sk->sk_pacing_shift, 10); 3431 sk->sk_incoming_cpu = -1; 3432 sk->sk_txrehash = SOCK_TXREHASH_DEFAULT; 3433 3434 sk_rx_queue_clear(sk); 3435 /* 3436 * Before updating sk_refcnt, we must commit prior changes to memory 3437 * (Documentation/RCU/rculist_nulls.rst for details) 3438 */ 3439 smp_wmb(); 3440 refcount_set(&sk->sk_refcnt, 1); 3441 atomic_set(&sk->sk_drops, 0); 3442 } 3443 EXPORT_SYMBOL(sock_init_data); 3444 3445 void lock_sock_nested(struct sock *sk, int subclass) 3446 { 3447 /* The sk_lock has mutex_lock() semantics here. */ 3448 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_); 3449 3450 might_sleep(); 3451 spin_lock_bh(&sk->sk_lock.slock); 3452 if (sock_owned_by_user_nocheck(sk)) 3453 __lock_sock(sk); 3454 sk->sk_lock.owned = 1; 3455 spin_unlock_bh(&sk->sk_lock.slock); 3456 } 3457 EXPORT_SYMBOL(lock_sock_nested); 3458 3459 void release_sock(struct sock *sk) 3460 { 3461 spin_lock_bh(&sk->sk_lock.slock); 3462 if (sk->sk_backlog.tail) 3463 __release_sock(sk); 3464 3465 /* Warning : release_cb() might need to release sk ownership, 3466 * ie call sock_release_ownership(sk) before us. 3467 */ 3468 if (sk->sk_prot->release_cb) 3469 sk->sk_prot->release_cb(sk); 3470 3471 sock_release_ownership(sk); 3472 if (waitqueue_active(&sk->sk_lock.wq)) 3473 wake_up(&sk->sk_lock.wq); 3474 spin_unlock_bh(&sk->sk_lock.slock); 3475 } 3476 EXPORT_SYMBOL(release_sock); 3477 3478 bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock) 3479 { 3480 might_sleep(); 3481 spin_lock_bh(&sk->sk_lock.slock); 3482 3483 if (!sock_owned_by_user_nocheck(sk)) { 3484 /* 3485 * Fast path return with bottom halves disabled and 3486 * sock::sk_lock.slock held. 3487 * 3488 * The 'mutex' is not contended and holding 3489 * sock::sk_lock.slock prevents all other lockers to 3490 * proceed so the corresponding unlock_sock_fast() can 3491 * avoid the slow path of release_sock() completely and 3492 * just release slock. 3493 * 3494 * From a semantical POV this is equivalent to 'acquiring' 3495 * the 'mutex', hence the corresponding lockdep 3496 * mutex_release() has to happen in the fast path of 3497 * unlock_sock_fast(). 3498 */ 3499 return false; 3500 } 3501 3502 __lock_sock(sk); 3503 sk->sk_lock.owned = 1; 3504 __acquire(&sk->sk_lock.slock); 3505 spin_unlock_bh(&sk->sk_lock.slock); 3506 return true; 3507 } 3508 EXPORT_SYMBOL(__lock_sock_fast); 3509 3510 int sock_gettstamp(struct socket *sock, void __user *userstamp, 3511 bool timeval, bool time32) 3512 { 3513 struct sock *sk = sock->sk; 3514 struct timespec64 ts; 3515 3516 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 3517 ts = ktime_to_timespec64(sock_read_timestamp(sk)); 3518 if (ts.tv_sec == -1) 3519 return -ENOENT; 3520 if (ts.tv_sec == 0) { 3521 ktime_t kt = ktime_get_real(); 3522 sock_write_timestamp(sk, kt); 3523 ts = ktime_to_timespec64(kt); 3524 } 3525 3526 if (timeval) 3527 ts.tv_nsec /= 1000; 3528 3529 #ifdef CONFIG_COMPAT_32BIT_TIME 3530 if (time32) 3531 return put_old_timespec32(&ts, userstamp); 3532 #endif 3533 #ifdef CONFIG_SPARC64 3534 /* beware of padding in sparc64 timeval */ 3535 if (timeval && !in_compat_syscall()) { 3536 struct __kernel_old_timeval __user tv = { 3537 .tv_sec = ts.tv_sec, 3538 .tv_usec = ts.tv_nsec, 3539 }; 3540 if (copy_to_user(userstamp, &tv, sizeof(tv))) 3541 return -EFAULT; 3542 return 0; 3543 } 3544 #endif 3545 return put_timespec64(&ts, userstamp); 3546 } 3547 EXPORT_SYMBOL(sock_gettstamp); 3548 3549 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag) 3550 { 3551 if (!sock_flag(sk, flag)) { 3552 unsigned long previous_flags = sk->sk_flags; 3553 3554 sock_set_flag(sk, flag); 3555 /* 3556 * we just set one of the two flags which require net 3557 * time stamping, but time stamping might have been on 3558 * already because of the other one 3559 */ 3560 if (sock_needs_netstamp(sk) && 3561 !(previous_flags & SK_FLAGS_TIMESTAMP)) 3562 net_enable_timestamp(); 3563 } 3564 } 3565 3566 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, 3567 int level, int type) 3568 { 3569 struct sock_exterr_skb *serr; 3570 struct sk_buff *skb; 3571 int copied, err; 3572 3573 err = -EAGAIN; 3574 skb = sock_dequeue_err_skb(sk); 3575 if (skb == NULL) 3576 goto out; 3577 3578 copied = skb->len; 3579 if (copied > len) { 3580 msg->msg_flags |= MSG_TRUNC; 3581 copied = len; 3582 } 3583 err = skb_copy_datagram_msg(skb, 0, msg, copied); 3584 if (err) 3585 goto out_free_skb; 3586 3587 sock_recv_timestamp(msg, sk, skb); 3588 3589 serr = SKB_EXT_ERR(skb); 3590 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee); 3591 3592 msg->msg_flags |= MSG_ERRQUEUE; 3593 err = copied; 3594 3595 out_free_skb: 3596 kfree_skb(skb); 3597 out: 3598 return err; 3599 } 3600 EXPORT_SYMBOL(sock_recv_errqueue); 3601 3602 /* 3603 * Get a socket option on an socket. 3604 * 3605 * FIX: POSIX 1003.1g is very ambiguous here. It states that 3606 * asynchronous errors should be reported by getsockopt. We assume 3607 * this means if you specify SO_ERROR (otherwise whats the point of it). 3608 */ 3609 int sock_common_getsockopt(struct socket *sock, int level, int optname, 3610 char __user *optval, int __user *optlen) 3611 { 3612 struct sock *sk = sock->sk; 3613 3614 /* IPV6_ADDRFORM can change sk->sk_prot under us. */ 3615 return READ_ONCE(sk->sk_prot)->getsockopt(sk, level, optname, optval, optlen); 3616 } 3617 EXPORT_SYMBOL(sock_common_getsockopt); 3618 3619 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 3620 int flags) 3621 { 3622 struct sock *sk = sock->sk; 3623 int addr_len = 0; 3624 int err; 3625 3626 err = sk->sk_prot->recvmsg(sk, msg, size, flags, &addr_len); 3627 if (err >= 0) 3628 msg->msg_namelen = addr_len; 3629 return err; 3630 } 3631 EXPORT_SYMBOL(sock_common_recvmsg); 3632 3633 /* 3634 * Set socket options on an inet socket. 3635 */ 3636 int sock_common_setsockopt(struct socket *sock, int level, int optname, 3637 sockptr_t optval, unsigned int optlen) 3638 { 3639 struct sock *sk = sock->sk; 3640 3641 /* IPV6_ADDRFORM can change sk->sk_prot under us. */ 3642 return READ_ONCE(sk->sk_prot)->setsockopt(sk, level, optname, optval, optlen); 3643 } 3644 EXPORT_SYMBOL(sock_common_setsockopt); 3645 3646 void sk_common_release(struct sock *sk) 3647 { 3648 if (sk->sk_prot->destroy) 3649 sk->sk_prot->destroy(sk); 3650 3651 /* 3652 * Observation: when sk_common_release is called, processes have 3653 * no access to socket. But net still has. 3654 * Step one, detach it from networking: 3655 * 3656 * A. Remove from hash tables. 3657 */ 3658 3659 sk->sk_prot->unhash(sk); 3660 3661 /* 3662 * In this point socket cannot receive new packets, but it is possible 3663 * that some packets are in flight because some CPU runs receiver and 3664 * did hash table lookup before we unhashed socket. They will achieve 3665 * receive queue and will be purged by socket destructor. 3666 * 3667 * Also we still have packets pending on receive queue and probably, 3668 * our own packets waiting in device queues. sock_destroy will drain 3669 * receive queue, but transmitted packets will delay socket destruction 3670 * until the last reference will be released. 3671 */ 3672 3673 sock_orphan(sk); 3674 3675 xfrm_sk_free_policy(sk); 3676 3677 sk_refcnt_debug_release(sk); 3678 3679 sock_put(sk); 3680 } 3681 EXPORT_SYMBOL(sk_common_release); 3682 3683 void sk_get_meminfo(const struct sock *sk, u32 *mem) 3684 { 3685 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS); 3686 3687 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk); 3688 mem[SK_MEMINFO_RCVBUF] = READ_ONCE(sk->sk_rcvbuf); 3689 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk); 3690 mem[SK_MEMINFO_SNDBUF] = READ_ONCE(sk->sk_sndbuf); 3691 mem[SK_MEMINFO_FWD_ALLOC] = sk->sk_forward_alloc; 3692 mem[SK_MEMINFO_WMEM_QUEUED] = READ_ONCE(sk->sk_wmem_queued); 3693 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc); 3694 mem[SK_MEMINFO_BACKLOG] = READ_ONCE(sk->sk_backlog.len); 3695 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops); 3696 } 3697 3698 #ifdef CONFIG_PROC_FS 3699 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR); 3700 3701 int sock_prot_inuse_get(struct net *net, struct proto *prot) 3702 { 3703 int cpu, idx = prot->inuse_idx; 3704 int res = 0; 3705 3706 for_each_possible_cpu(cpu) 3707 res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx]; 3708 3709 return res >= 0 ? res : 0; 3710 } 3711 EXPORT_SYMBOL_GPL(sock_prot_inuse_get); 3712 3713 int sock_inuse_get(struct net *net) 3714 { 3715 int cpu, res = 0; 3716 3717 for_each_possible_cpu(cpu) 3718 res += per_cpu_ptr(net->core.prot_inuse, cpu)->all; 3719 3720 return res; 3721 } 3722 3723 EXPORT_SYMBOL_GPL(sock_inuse_get); 3724 3725 static int __net_init sock_inuse_init_net(struct net *net) 3726 { 3727 net->core.prot_inuse = alloc_percpu(struct prot_inuse); 3728 if (net->core.prot_inuse == NULL) 3729 return -ENOMEM; 3730 return 0; 3731 } 3732 3733 static void __net_exit sock_inuse_exit_net(struct net *net) 3734 { 3735 free_percpu(net->core.prot_inuse); 3736 } 3737 3738 static struct pernet_operations net_inuse_ops = { 3739 .init = sock_inuse_init_net, 3740 .exit = sock_inuse_exit_net, 3741 }; 3742 3743 static __init int net_inuse_init(void) 3744 { 3745 if (register_pernet_subsys(&net_inuse_ops)) 3746 panic("Cannot initialize net inuse counters"); 3747 3748 return 0; 3749 } 3750 3751 core_initcall(net_inuse_init); 3752 3753 static int assign_proto_idx(struct proto *prot) 3754 { 3755 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR); 3756 3757 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) { 3758 pr_err("PROTO_INUSE_NR exhausted\n"); 3759 return -ENOSPC; 3760 } 3761 3762 set_bit(prot->inuse_idx, proto_inuse_idx); 3763 return 0; 3764 } 3765 3766 static void release_proto_idx(struct proto *prot) 3767 { 3768 if (prot->inuse_idx != PROTO_INUSE_NR - 1) 3769 clear_bit(prot->inuse_idx, proto_inuse_idx); 3770 } 3771 #else 3772 static inline int assign_proto_idx(struct proto *prot) 3773 { 3774 return 0; 3775 } 3776 3777 static inline void release_proto_idx(struct proto *prot) 3778 { 3779 } 3780 3781 #endif 3782 3783 static void tw_prot_cleanup(struct timewait_sock_ops *twsk_prot) 3784 { 3785 if (!twsk_prot) 3786 return; 3787 kfree(twsk_prot->twsk_slab_name); 3788 twsk_prot->twsk_slab_name = NULL; 3789 kmem_cache_destroy(twsk_prot->twsk_slab); 3790 twsk_prot->twsk_slab = NULL; 3791 } 3792 3793 static int tw_prot_init(const struct proto *prot) 3794 { 3795 struct timewait_sock_ops *twsk_prot = prot->twsk_prot; 3796 3797 if (!twsk_prot) 3798 return 0; 3799 3800 twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", 3801 prot->name); 3802 if (!twsk_prot->twsk_slab_name) 3803 return -ENOMEM; 3804 3805 twsk_prot->twsk_slab = 3806 kmem_cache_create(twsk_prot->twsk_slab_name, 3807 twsk_prot->twsk_obj_size, 0, 3808 SLAB_ACCOUNT | prot->slab_flags, 3809 NULL); 3810 if (!twsk_prot->twsk_slab) { 3811 pr_crit("%s: Can't create timewait sock SLAB cache!\n", 3812 prot->name); 3813 return -ENOMEM; 3814 } 3815 3816 return 0; 3817 } 3818 3819 static void req_prot_cleanup(struct request_sock_ops *rsk_prot) 3820 { 3821 if (!rsk_prot) 3822 return; 3823 kfree(rsk_prot->slab_name); 3824 rsk_prot->slab_name = NULL; 3825 kmem_cache_destroy(rsk_prot->slab); 3826 rsk_prot->slab = NULL; 3827 } 3828 3829 static int req_prot_init(const struct proto *prot) 3830 { 3831 struct request_sock_ops *rsk_prot = prot->rsk_prot; 3832 3833 if (!rsk_prot) 3834 return 0; 3835 3836 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", 3837 prot->name); 3838 if (!rsk_prot->slab_name) 3839 return -ENOMEM; 3840 3841 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name, 3842 rsk_prot->obj_size, 0, 3843 SLAB_ACCOUNT | prot->slab_flags, 3844 NULL); 3845 3846 if (!rsk_prot->slab) { 3847 pr_crit("%s: Can't create request sock SLAB cache!\n", 3848 prot->name); 3849 return -ENOMEM; 3850 } 3851 return 0; 3852 } 3853 3854 int proto_register(struct proto *prot, int alloc_slab) 3855 { 3856 int ret = -ENOBUFS; 3857 3858 if (prot->memory_allocated && !prot->sysctl_mem) { 3859 pr_err("%s: missing sysctl_mem\n", prot->name); 3860 return -EINVAL; 3861 } 3862 if (prot->memory_allocated && !prot->per_cpu_fw_alloc) { 3863 pr_err("%s: missing per_cpu_fw_alloc\n", prot->name); 3864 return -EINVAL; 3865 } 3866 if (alloc_slab) { 3867 prot->slab = kmem_cache_create_usercopy(prot->name, 3868 prot->obj_size, 0, 3869 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT | 3870 prot->slab_flags, 3871 prot->useroffset, prot->usersize, 3872 NULL); 3873 3874 if (prot->slab == NULL) { 3875 pr_crit("%s: Can't create sock SLAB cache!\n", 3876 prot->name); 3877 goto out; 3878 } 3879 3880 if (req_prot_init(prot)) 3881 goto out_free_request_sock_slab; 3882 3883 if (tw_prot_init(prot)) 3884 goto out_free_timewait_sock_slab; 3885 } 3886 3887 mutex_lock(&proto_list_mutex); 3888 ret = assign_proto_idx(prot); 3889 if (ret) { 3890 mutex_unlock(&proto_list_mutex); 3891 goto out_free_timewait_sock_slab; 3892 } 3893 list_add(&prot->node, &proto_list); 3894 mutex_unlock(&proto_list_mutex); 3895 return ret; 3896 3897 out_free_timewait_sock_slab: 3898 if (alloc_slab) 3899 tw_prot_cleanup(prot->twsk_prot); 3900 out_free_request_sock_slab: 3901 if (alloc_slab) { 3902 req_prot_cleanup(prot->rsk_prot); 3903 3904 kmem_cache_destroy(prot->slab); 3905 prot->slab = NULL; 3906 } 3907 out: 3908 return ret; 3909 } 3910 EXPORT_SYMBOL(proto_register); 3911 3912 void proto_unregister(struct proto *prot) 3913 { 3914 mutex_lock(&proto_list_mutex); 3915 release_proto_idx(prot); 3916 list_del(&prot->node); 3917 mutex_unlock(&proto_list_mutex); 3918 3919 kmem_cache_destroy(prot->slab); 3920 prot->slab = NULL; 3921 3922 req_prot_cleanup(prot->rsk_prot); 3923 tw_prot_cleanup(prot->twsk_prot); 3924 } 3925 EXPORT_SYMBOL(proto_unregister); 3926 3927 int sock_load_diag_module(int family, int protocol) 3928 { 3929 if (!protocol) { 3930 if (!sock_is_registered(family)) 3931 return -ENOENT; 3932 3933 return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK, 3934 NETLINK_SOCK_DIAG, family); 3935 } 3936 3937 #ifdef CONFIG_INET 3938 if (family == AF_INET && 3939 protocol != IPPROTO_RAW && 3940 protocol < MAX_INET_PROTOS && 3941 !rcu_access_pointer(inet_protos[protocol])) 3942 return -ENOENT; 3943 #endif 3944 3945 return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK, 3946 NETLINK_SOCK_DIAG, family, protocol); 3947 } 3948 EXPORT_SYMBOL(sock_load_diag_module); 3949 3950 #ifdef CONFIG_PROC_FS 3951 static void *proto_seq_start(struct seq_file *seq, loff_t *pos) 3952 __acquires(proto_list_mutex) 3953 { 3954 mutex_lock(&proto_list_mutex); 3955 return seq_list_start_head(&proto_list, *pos); 3956 } 3957 3958 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos) 3959 { 3960 return seq_list_next(v, &proto_list, pos); 3961 } 3962 3963 static void proto_seq_stop(struct seq_file *seq, void *v) 3964 __releases(proto_list_mutex) 3965 { 3966 mutex_unlock(&proto_list_mutex); 3967 } 3968 3969 static char proto_method_implemented(const void *method) 3970 { 3971 return method == NULL ? 'n' : 'y'; 3972 } 3973 static long sock_prot_memory_allocated(struct proto *proto) 3974 { 3975 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L; 3976 } 3977 3978 static const char *sock_prot_memory_pressure(struct proto *proto) 3979 { 3980 return proto->memory_pressure != NULL ? 3981 proto_memory_pressure(proto) ? "yes" : "no" : "NI"; 3982 } 3983 3984 static void proto_seq_printf(struct seq_file *seq, struct proto *proto) 3985 { 3986 3987 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s " 3988 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n", 3989 proto->name, 3990 proto->obj_size, 3991 sock_prot_inuse_get(seq_file_net(seq), proto), 3992 sock_prot_memory_allocated(proto), 3993 sock_prot_memory_pressure(proto), 3994 proto->max_header, 3995 proto->slab == NULL ? "no" : "yes", 3996 module_name(proto->owner), 3997 proto_method_implemented(proto->close), 3998 proto_method_implemented(proto->connect), 3999 proto_method_implemented(proto->disconnect), 4000 proto_method_implemented(proto->accept), 4001 proto_method_implemented(proto->ioctl), 4002 proto_method_implemented(proto->init), 4003 proto_method_implemented(proto->destroy), 4004 proto_method_implemented(proto->shutdown), 4005 proto_method_implemented(proto->setsockopt), 4006 proto_method_implemented(proto->getsockopt), 4007 proto_method_implemented(proto->sendmsg), 4008 proto_method_implemented(proto->recvmsg), 4009 proto_method_implemented(proto->sendpage), 4010 proto_method_implemented(proto->bind), 4011 proto_method_implemented(proto->backlog_rcv), 4012 proto_method_implemented(proto->hash), 4013 proto_method_implemented(proto->unhash), 4014 proto_method_implemented(proto->get_port), 4015 proto_method_implemented(proto->enter_memory_pressure)); 4016 } 4017 4018 static int proto_seq_show(struct seq_file *seq, void *v) 4019 { 4020 if (v == &proto_list) 4021 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s", 4022 "protocol", 4023 "size", 4024 "sockets", 4025 "memory", 4026 "press", 4027 "maxhdr", 4028 "slab", 4029 "module", 4030 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n"); 4031 else 4032 proto_seq_printf(seq, list_entry(v, struct proto, node)); 4033 return 0; 4034 } 4035 4036 static const struct seq_operations proto_seq_ops = { 4037 .start = proto_seq_start, 4038 .next = proto_seq_next, 4039 .stop = proto_seq_stop, 4040 .show = proto_seq_show, 4041 }; 4042 4043 static __net_init int proto_init_net(struct net *net) 4044 { 4045 if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops, 4046 sizeof(struct seq_net_private))) 4047 return -ENOMEM; 4048 4049 return 0; 4050 } 4051 4052 static __net_exit void proto_exit_net(struct net *net) 4053 { 4054 remove_proc_entry("protocols", net->proc_net); 4055 } 4056 4057 4058 static __net_initdata struct pernet_operations proto_net_ops = { 4059 .init = proto_init_net, 4060 .exit = proto_exit_net, 4061 }; 4062 4063 static int __init proto_init(void) 4064 { 4065 return register_pernet_subsys(&proto_net_ops); 4066 } 4067 4068 subsys_initcall(proto_init); 4069 4070 #endif /* PROC_FS */ 4071 4072 #ifdef CONFIG_NET_RX_BUSY_POLL 4073 bool sk_busy_loop_end(void *p, unsigned long start_time) 4074 { 4075 struct sock *sk = p; 4076 4077 return !skb_queue_empty_lockless(&sk->sk_receive_queue) || 4078 sk_busy_loop_timeout(sk, start_time); 4079 } 4080 EXPORT_SYMBOL(sk_busy_loop_end); 4081 #endif /* CONFIG_NET_RX_BUSY_POLL */ 4082 4083 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len) 4084 { 4085 if (!sk->sk_prot->bind_add) 4086 return -EOPNOTSUPP; 4087 return sk->sk_prot->bind_add(sk, addr, addr_len); 4088 } 4089 EXPORT_SYMBOL(sock_bind_add); 4090