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, optval.user, optlen.user, len); 1797 1798 case SO_MARK: 1799 v.val = sk->sk_mark; 1800 break; 1801 1802 case SO_RCVMARK: 1803 v.val = sock_flag(sk, SOCK_RCVMARK); 1804 break; 1805 1806 case SO_RXQ_OVFL: 1807 v.val = sock_flag(sk, SOCK_RXQ_OVFL); 1808 break; 1809 1810 case SO_WIFI_STATUS: 1811 v.val = sock_flag(sk, SOCK_WIFI_STATUS); 1812 break; 1813 1814 case SO_PEEK_OFF: 1815 if (!sock->ops->set_peek_off) 1816 return -EOPNOTSUPP; 1817 1818 v.val = sk->sk_peek_off; 1819 break; 1820 case SO_NOFCS: 1821 v.val = sock_flag(sk, SOCK_NOFCS); 1822 break; 1823 1824 case SO_BINDTODEVICE: 1825 return sock_getbindtodevice(sk, optval, optlen, len); 1826 1827 case SO_GET_FILTER: 1828 len = sk_get_filter(sk, optval, len); 1829 if (len < 0) 1830 return len; 1831 1832 goto lenout; 1833 1834 case SO_LOCK_FILTER: 1835 v.val = sock_flag(sk, SOCK_FILTER_LOCKED); 1836 break; 1837 1838 case SO_BPF_EXTENSIONS: 1839 v.val = bpf_tell_extensions(); 1840 break; 1841 1842 case SO_SELECT_ERR_QUEUE: 1843 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE); 1844 break; 1845 1846 #ifdef CONFIG_NET_RX_BUSY_POLL 1847 case SO_BUSY_POLL: 1848 v.val = sk->sk_ll_usec; 1849 break; 1850 case SO_PREFER_BUSY_POLL: 1851 v.val = READ_ONCE(sk->sk_prefer_busy_poll); 1852 break; 1853 #endif 1854 1855 case SO_MAX_PACING_RATE: 1856 if (sizeof(v.ulval) != sizeof(v.val) && len >= sizeof(v.ulval)) { 1857 lv = sizeof(v.ulval); 1858 v.ulval = sk->sk_max_pacing_rate; 1859 } else { 1860 /* 32bit version */ 1861 v.val = min_t(unsigned long, sk->sk_max_pacing_rate, ~0U); 1862 } 1863 break; 1864 1865 case SO_INCOMING_CPU: 1866 v.val = READ_ONCE(sk->sk_incoming_cpu); 1867 break; 1868 1869 case SO_MEMINFO: 1870 { 1871 u32 meminfo[SK_MEMINFO_VARS]; 1872 1873 sk_get_meminfo(sk, meminfo); 1874 1875 len = min_t(unsigned int, len, sizeof(meminfo)); 1876 if (copy_to_sockptr(optval, &meminfo, len)) 1877 return -EFAULT; 1878 1879 goto lenout; 1880 } 1881 1882 #ifdef CONFIG_NET_RX_BUSY_POLL 1883 case SO_INCOMING_NAPI_ID: 1884 v.val = READ_ONCE(sk->sk_napi_id); 1885 1886 /* aggregate non-NAPI IDs down to 0 */ 1887 if (v.val < MIN_NAPI_ID) 1888 v.val = 0; 1889 1890 break; 1891 #endif 1892 1893 case SO_COOKIE: 1894 lv = sizeof(u64); 1895 if (len < lv) 1896 return -EINVAL; 1897 v.val64 = sock_gen_cookie(sk); 1898 break; 1899 1900 case SO_ZEROCOPY: 1901 v.val = sock_flag(sk, SOCK_ZEROCOPY); 1902 break; 1903 1904 case SO_TXTIME: 1905 lv = sizeof(v.txtime); 1906 v.txtime.clockid = sk->sk_clockid; 1907 v.txtime.flags |= sk->sk_txtime_deadline_mode ? 1908 SOF_TXTIME_DEADLINE_MODE : 0; 1909 v.txtime.flags |= sk->sk_txtime_report_errors ? 1910 SOF_TXTIME_REPORT_ERRORS : 0; 1911 break; 1912 1913 case SO_BINDTOIFINDEX: 1914 v.val = READ_ONCE(sk->sk_bound_dev_if); 1915 break; 1916 1917 case SO_NETNS_COOKIE: 1918 lv = sizeof(u64); 1919 if (len != lv) 1920 return -EINVAL; 1921 v.val64 = sock_net(sk)->net_cookie; 1922 break; 1923 1924 case SO_BUF_LOCK: 1925 v.val = sk->sk_userlocks & SOCK_BUF_LOCK_MASK; 1926 break; 1927 1928 case SO_RESERVE_MEM: 1929 v.val = sk->sk_reserved_mem; 1930 break; 1931 1932 case SO_TXREHASH: 1933 v.val = sk->sk_txrehash; 1934 break; 1935 1936 default: 1937 /* We implement the SO_SNDLOWAT etc to not be settable 1938 * (1003.1g 7). 1939 */ 1940 return -ENOPROTOOPT; 1941 } 1942 1943 if (len > lv) 1944 len = lv; 1945 if (copy_to_sockptr(optval, &v, len)) 1946 return -EFAULT; 1947 lenout: 1948 if (copy_to_sockptr(optlen, &len, sizeof(int))) 1949 return -EFAULT; 1950 return 0; 1951 } 1952 1953 int sock_getsockopt(struct socket *sock, int level, int optname, 1954 char __user *optval, int __user *optlen) 1955 { 1956 return sk_getsockopt(sock->sk, level, optname, 1957 USER_SOCKPTR(optval), 1958 USER_SOCKPTR(optlen)); 1959 } 1960 1961 /* 1962 * Initialize an sk_lock. 1963 * 1964 * (We also register the sk_lock with the lock validator.) 1965 */ 1966 static inline void sock_lock_init(struct sock *sk) 1967 { 1968 if (sk->sk_kern_sock) 1969 sock_lock_init_class_and_name( 1970 sk, 1971 af_family_kern_slock_key_strings[sk->sk_family], 1972 af_family_kern_slock_keys + sk->sk_family, 1973 af_family_kern_key_strings[sk->sk_family], 1974 af_family_kern_keys + sk->sk_family); 1975 else 1976 sock_lock_init_class_and_name( 1977 sk, 1978 af_family_slock_key_strings[sk->sk_family], 1979 af_family_slock_keys + sk->sk_family, 1980 af_family_key_strings[sk->sk_family], 1981 af_family_keys + sk->sk_family); 1982 } 1983 1984 /* 1985 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet, 1986 * even temporarly, because of RCU lookups. sk_node should also be left as is. 1987 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end 1988 */ 1989 static void sock_copy(struct sock *nsk, const struct sock *osk) 1990 { 1991 const struct proto *prot = READ_ONCE(osk->sk_prot); 1992 #ifdef CONFIG_SECURITY_NETWORK 1993 void *sptr = nsk->sk_security; 1994 #endif 1995 1996 /* If we move sk_tx_queue_mapping out of the private section, 1997 * we must check if sk_tx_queue_clear() is called after 1998 * sock_copy() in sk_clone_lock(). 1999 */ 2000 BUILD_BUG_ON(offsetof(struct sock, sk_tx_queue_mapping) < 2001 offsetof(struct sock, sk_dontcopy_begin) || 2002 offsetof(struct sock, sk_tx_queue_mapping) >= 2003 offsetof(struct sock, sk_dontcopy_end)); 2004 2005 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin)); 2006 2007 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end, 2008 prot->obj_size - offsetof(struct sock, sk_dontcopy_end)); 2009 2010 #ifdef CONFIG_SECURITY_NETWORK 2011 nsk->sk_security = sptr; 2012 security_sk_clone(osk, nsk); 2013 #endif 2014 } 2015 2016 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority, 2017 int family) 2018 { 2019 struct sock *sk; 2020 struct kmem_cache *slab; 2021 2022 slab = prot->slab; 2023 if (slab != NULL) { 2024 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO); 2025 if (!sk) 2026 return sk; 2027 if (want_init_on_alloc(priority)) 2028 sk_prot_clear_nulls(sk, prot->obj_size); 2029 } else 2030 sk = kmalloc(prot->obj_size, priority); 2031 2032 if (sk != NULL) { 2033 if (security_sk_alloc(sk, family, priority)) 2034 goto out_free; 2035 2036 if (!try_module_get(prot->owner)) 2037 goto out_free_sec; 2038 } 2039 2040 return sk; 2041 2042 out_free_sec: 2043 security_sk_free(sk); 2044 out_free: 2045 if (slab != NULL) 2046 kmem_cache_free(slab, sk); 2047 else 2048 kfree(sk); 2049 return NULL; 2050 } 2051 2052 static void sk_prot_free(struct proto *prot, struct sock *sk) 2053 { 2054 struct kmem_cache *slab; 2055 struct module *owner; 2056 2057 owner = prot->owner; 2058 slab = prot->slab; 2059 2060 cgroup_sk_free(&sk->sk_cgrp_data); 2061 mem_cgroup_sk_free(sk); 2062 security_sk_free(sk); 2063 if (slab != NULL) 2064 kmem_cache_free(slab, sk); 2065 else 2066 kfree(sk); 2067 module_put(owner); 2068 } 2069 2070 /** 2071 * sk_alloc - All socket objects are allocated here 2072 * @net: the applicable net namespace 2073 * @family: protocol family 2074 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 2075 * @prot: struct proto associated with this new sock instance 2076 * @kern: is this to be a kernel socket? 2077 */ 2078 struct sock *sk_alloc(struct net *net, int family, gfp_t priority, 2079 struct proto *prot, int kern) 2080 { 2081 struct sock *sk; 2082 2083 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family); 2084 if (sk) { 2085 sk->sk_family = family; 2086 /* 2087 * See comment in struct sock definition to understand 2088 * why we need sk_prot_creator -acme 2089 */ 2090 sk->sk_prot = sk->sk_prot_creator = prot; 2091 sk->sk_kern_sock = kern; 2092 sock_lock_init(sk); 2093 sk->sk_net_refcnt = kern ? 0 : 1; 2094 if (likely(sk->sk_net_refcnt)) { 2095 get_net_track(net, &sk->ns_tracker, priority); 2096 sock_inuse_add(net, 1); 2097 } 2098 2099 sock_net_set(sk, net); 2100 refcount_set(&sk->sk_wmem_alloc, 1); 2101 2102 mem_cgroup_sk_alloc(sk); 2103 cgroup_sk_alloc(&sk->sk_cgrp_data); 2104 sock_update_classid(&sk->sk_cgrp_data); 2105 sock_update_netprioidx(&sk->sk_cgrp_data); 2106 sk_tx_queue_clear(sk); 2107 } 2108 2109 return sk; 2110 } 2111 EXPORT_SYMBOL(sk_alloc); 2112 2113 /* Sockets having SOCK_RCU_FREE will call this function after one RCU 2114 * grace period. This is the case for UDP sockets and TCP listeners. 2115 */ 2116 static void __sk_destruct(struct rcu_head *head) 2117 { 2118 struct sock *sk = container_of(head, struct sock, sk_rcu); 2119 struct sk_filter *filter; 2120 2121 if (sk->sk_destruct) 2122 sk->sk_destruct(sk); 2123 2124 filter = rcu_dereference_check(sk->sk_filter, 2125 refcount_read(&sk->sk_wmem_alloc) == 0); 2126 if (filter) { 2127 sk_filter_uncharge(sk, filter); 2128 RCU_INIT_POINTER(sk->sk_filter, NULL); 2129 } 2130 2131 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP); 2132 2133 #ifdef CONFIG_BPF_SYSCALL 2134 bpf_sk_storage_free(sk); 2135 #endif 2136 2137 if (atomic_read(&sk->sk_omem_alloc)) 2138 pr_debug("%s: optmem leakage (%d bytes) detected\n", 2139 __func__, atomic_read(&sk->sk_omem_alloc)); 2140 2141 if (sk->sk_frag.page) { 2142 put_page(sk->sk_frag.page); 2143 sk->sk_frag.page = NULL; 2144 } 2145 2146 /* We do not need to acquire sk->sk_peer_lock, we are the last user. */ 2147 put_cred(sk->sk_peer_cred); 2148 put_pid(sk->sk_peer_pid); 2149 2150 if (likely(sk->sk_net_refcnt)) 2151 put_net_track(sock_net(sk), &sk->ns_tracker); 2152 sk_prot_free(sk->sk_prot_creator, sk); 2153 } 2154 2155 void sk_destruct(struct sock *sk) 2156 { 2157 bool use_call_rcu = sock_flag(sk, SOCK_RCU_FREE); 2158 2159 if (rcu_access_pointer(sk->sk_reuseport_cb)) { 2160 reuseport_detach_sock(sk); 2161 use_call_rcu = true; 2162 } 2163 2164 if (use_call_rcu) 2165 call_rcu(&sk->sk_rcu, __sk_destruct); 2166 else 2167 __sk_destruct(&sk->sk_rcu); 2168 } 2169 2170 static void __sk_free(struct sock *sk) 2171 { 2172 if (likely(sk->sk_net_refcnt)) 2173 sock_inuse_add(sock_net(sk), -1); 2174 2175 if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk))) 2176 sock_diag_broadcast_destroy(sk); 2177 else 2178 sk_destruct(sk); 2179 } 2180 2181 void sk_free(struct sock *sk) 2182 { 2183 /* 2184 * We subtract one from sk_wmem_alloc and can know if 2185 * some packets are still in some tx queue. 2186 * If not null, sock_wfree() will call __sk_free(sk) later 2187 */ 2188 if (refcount_dec_and_test(&sk->sk_wmem_alloc)) 2189 __sk_free(sk); 2190 } 2191 EXPORT_SYMBOL(sk_free); 2192 2193 static void sk_init_common(struct sock *sk) 2194 { 2195 skb_queue_head_init(&sk->sk_receive_queue); 2196 skb_queue_head_init(&sk->sk_write_queue); 2197 skb_queue_head_init(&sk->sk_error_queue); 2198 2199 rwlock_init(&sk->sk_callback_lock); 2200 lockdep_set_class_and_name(&sk->sk_receive_queue.lock, 2201 af_rlock_keys + sk->sk_family, 2202 af_family_rlock_key_strings[sk->sk_family]); 2203 lockdep_set_class_and_name(&sk->sk_write_queue.lock, 2204 af_wlock_keys + sk->sk_family, 2205 af_family_wlock_key_strings[sk->sk_family]); 2206 lockdep_set_class_and_name(&sk->sk_error_queue.lock, 2207 af_elock_keys + sk->sk_family, 2208 af_family_elock_key_strings[sk->sk_family]); 2209 lockdep_set_class_and_name(&sk->sk_callback_lock, 2210 af_callback_keys + sk->sk_family, 2211 af_family_clock_key_strings[sk->sk_family]); 2212 } 2213 2214 /** 2215 * sk_clone_lock - clone a socket, and lock its clone 2216 * @sk: the socket to clone 2217 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 2218 * 2219 * Caller must unlock socket even in error path (bh_unlock_sock(newsk)) 2220 */ 2221 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority) 2222 { 2223 struct proto *prot = READ_ONCE(sk->sk_prot); 2224 struct sk_filter *filter; 2225 bool is_charged = true; 2226 struct sock *newsk; 2227 2228 newsk = sk_prot_alloc(prot, priority, sk->sk_family); 2229 if (!newsk) 2230 goto out; 2231 2232 sock_copy(newsk, sk); 2233 2234 newsk->sk_prot_creator = prot; 2235 2236 /* SANITY */ 2237 if (likely(newsk->sk_net_refcnt)) { 2238 get_net_track(sock_net(newsk), &newsk->ns_tracker, priority); 2239 sock_inuse_add(sock_net(newsk), 1); 2240 } 2241 sk_node_init(&newsk->sk_node); 2242 sock_lock_init(newsk); 2243 bh_lock_sock(newsk); 2244 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL; 2245 newsk->sk_backlog.len = 0; 2246 2247 atomic_set(&newsk->sk_rmem_alloc, 0); 2248 2249 /* sk_wmem_alloc set to one (see sk_free() and sock_wfree()) */ 2250 refcount_set(&newsk->sk_wmem_alloc, 1); 2251 2252 atomic_set(&newsk->sk_omem_alloc, 0); 2253 sk_init_common(newsk); 2254 2255 newsk->sk_dst_cache = NULL; 2256 newsk->sk_dst_pending_confirm = 0; 2257 newsk->sk_wmem_queued = 0; 2258 newsk->sk_forward_alloc = 0; 2259 newsk->sk_reserved_mem = 0; 2260 atomic_set(&newsk->sk_drops, 0); 2261 newsk->sk_send_head = NULL; 2262 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK; 2263 atomic_set(&newsk->sk_zckey, 0); 2264 2265 sock_reset_flag(newsk, SOCK_DONE); 2266 2267 /* sk->sk_memcg will be populated at accept() time */ 2268 newsk->sk_memcg = NULL; 2269 2270 cgroup_sk_clone(&newsk->sk_cgrp_data); 2271 2272 rcu_read_lock(); 2273 filter = rcu_dereference(sk->sk_filter); 2274 if (filter != NULL) 2275 /* though it's an empty new sock, the charging may fail 2276 * if sysctl_optmem_max was changed between creation of 2277 * original socket and cloning 2278 */ 2279 is_charged = sk_filter_charge(newsk, filter); 2280 RCU_INIT_POINTER(newsk->sk_filter, filter); 2281 rcu_read_unlock(); 2282 2283 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) { 2284 /* We need to make sure that we don't uncharge the new 2285 * socket if we couldn't charge it in the first place 2286 * as otherwise we uncharge the parent's filter. 2287 */ 2288 if (!is_charged) 2289 RCU_INIT_POINTER(newsk->sk_filter, NULL); 2290 sk_free_unlock_clone(newsk); 2291 newsk = NULL; 2292 goto out; 2293 } 2294 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL); 2295 2296 if (bpf_sk_storage_clone(sk, newsk)) { 2297 sk_free_unlock_clone(newsk); 2298 newsk = NULL; 2299 goto out; 2300 } 2301 2302 /* Clear sk_user_data if parent had the pointer tagged 2303 * as not suitable for copying when cloning. 2304 */ 2305 if (sk_user_data_is_nocopy(newsk)) 2306 newsk->sk_user_data = NULL; 2307 2308 newsk->sk_err = 0; 2309 newsk->sk_err_soft = 0; 2310 newsk->sk_priority = 0; 2311 newsk->sk_incoming_cpu = raw_smp_processor_id(); 2312 2313 /* Before updating sk_refcnt, we must commit prior changes to memory 2314 * (Documentation/RCU/rculist_nulls.rst for details) 2315 */ 2316 smp_wmb(); 2317 refcount_set(&newsk->sk_refcnt, 2); 2318 2319 /* Increment the counter in the same struct proto as the master 2320 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that 2321 * is the same as sk->sk_prot->socks, as this field was copied 2322 * with memcpy). 2323 * 2324 * This _changes_ the previous behaviour, where 2325 * tcp_create_openreq_child always was incrementing the 2326 * equivalent to tcp_prot->socks (inet_sock_nr), so this have 2327 * to be taken into account in all callers. -acme 2328 */ 2329 sk_refcnt_debug_inc(newsk); 2330 sk_set_socket(newsk, NULL); 2331 sk_tx_queue_clear(newsk); 2332 RCU_INIT_POINTER(newsk->sk_wq, NULL); 2333 2334 if (newsk->sk_prot->sockets_allocated) 2335 sk_sockets_allocated_inc(newsk); 2336 2337 if (sock_needs_netstamp(sk) && newsk->sk_flags & SK_FLAGS_TIMESTAMP) 2338 net_enable_timestamp(); 2339 out: 2340 return newsk; 2341 } 2342 EXPORT_SYMBOL_GPL(sk_clone_lock); 2343 2344 void sk_free_unlock_clone(struct sock *sk) 2345 { 2346 /* It is still raw copy of parent, so invalidate 2347 * destructor and make plain sk_free() */ 2348 sk->sk_destruct = NULL; 2349 bh_unlock_sock(sk); 2350 sk_free(sk); 2351 } 2352 EXPORT_SYMBOL_GPL(sk_free_unlock_clone); 2353 2354 static void sk_trim_gso_size(struct sock *sk) 2355 { 2356 if (sk->sk_gso_max_size <= GSO_LEGACY_MAX_SIZE) 2357 return; 2358 #if IS_ENABLED(CONFIG_IPV6) 2359 if (sk->sk_family == AF_INET6 && 2360 sk_is_tcp(sk) && 2361 !ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr)) 2362 return; 2363 #endif 2364 sk->sk_gso_max_size = GSO_LEGACY_MAX_SIZE; 2365 } 2366 2367 void sk_setup_caps(struct sock *sk, struct dst_entry *dst) 2368 { 2369 u32 max_segs = 1; 2370 2371 sk_dst_set(sk, dst); 2372 sk->sk_route_caps = dst->dev->features; 2373 if (sk_is_tcp(sk)) 2374 sk->sk_route_caps |= NETIF_F_GSO; 2375 if (sk->sk_route_caps & NETIF_F_GSO) 2376 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE; 2377 if (unlikely(sk->sk_gso_disabled)) 2378 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 2379 if (sk_can_gso(sk)) { 2380 if (dst->header_len && !xfrm_dst_offload_ok(dst)) { 2381 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 2382 } else { 2383 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM; 2384 /* pairs with the WRITE_ONCE() in netif_set_gso_max_size() */ 2385 sk->sk_gso_max_size = READ_ONCE(dst->dev->gso_max_size); 2386 sk_trim_gso_size(sk); 2387 sk->sk_gso_max_size -= (MAX_TCP_HEADER + 1); 2388 /* pairs with the WRITE_ONCE() in netif_set_gso_max_segs() */ 2389 max_segs = max_t(u32, READ_ONCE(dst->dev->gso_max_segs), 1); 2390 } 2391 } 2392 sk->sk_gso_max_segs = max_segs; 2393 } 2394 EXPORT_SYMBOL_GPL(sk_setup_caps); 2395 2396 /* 2397 * Simple resource managers for sockets. 2398 */ 2399 2400 2401 /* 2402 * Write buffer destructor automatically called from kfree_skb. 2403 */ 2404 void sock_wfree(struct sk_buff *skb) 2405 { 2406 struct sock *sk = skb->sk; 2407 unsigned int len = skb->truesize; 2408 bool free; 2409 2410 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) { 2411 if (sock_flag(sk, SOCK_RCU_FREE) && 2412 sk->sk_write_space == sock_def_write_space) { 2413 rcu_read_lock(); 2414 free = refcount_sub_and_test(len, &sk->sk_wmem_alloc); 2415 sock_def_write_space_wfree(sk); 2416 rcu_read_unlock(); 2417 if (unlikely(free)) 2418 __sk_free(sk); 2419 return; 2420 } 2421 2422 /* 2423 * Keep a reference on sk_wmem_alloc, this will be released 2424 * after sk_write_space() call 2425 */ 2426 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc)); 2427 sk->sk_write_space(sk); 2428 len = 1; 2429 } 2430 /* 2431 * if sk_wmem_alloc reaches 0, we must finish what sk_free() 2432 * could not do because of in-flight packets 2433 */ 2434 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc)) 2435 __sk_free(sk); 2436 } 2437 EXPORT_SYMBOL(sock_wfree); 2438 2439 /* This variant of sock_wfree() is used by TCP, 2440 * since it sets SOCK_USE_WRITE_QUEUE. 2441 */ 2442 void __sock_wfree(struct sk_buff *skb) 2443 { 2444 struct sock *sk = skb->sk; 2445 2446 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc)) 2447 __sk_free(sk); 2448 } 2449 2450 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk) 2451 { 2452 skb_orphan(skb); 2453 skb->sk = sk; 2454 #ifdef CONFIG_INET 2455 if (unlikely(!sk_fullsock(sk))) { 2456 skb->destructor = sock_edemux; 2457 sock_hold(sk); 2458 return; 2459 } 2460 #endif 2461 skb->destructor = sock_wfree; 2462 skb_set_hash_from_sk(skb, sk); 2463 /* 2464 * We used to take a refcount on sk, but following operation 2465 * is enough to guarantee sk_free() wont free this sock until 2466 * all in-flight packets are completed 2467 */ 2468 refcount_add(skb->truesize, &sk->sk_wmem_alloc); 2469 } 2470 EXPORT_SYMBOL(skb_set_owner_w); 2471 2472 static bool can_skb_orphan_partial(const struct sk_buff *skb) 2473 { 2474 #ifdef CONFIG_TLS_DEVICE 2475 /* Drivers depend on in-order delivery for crypto offload, 2476 * partial orphan breaks out-of-order-OK logic. 2477 */ 2478 if (skb->decrypted) 2479 return false; 2480 #endif 2481 return (skb->destructor == sock_wfree || 2482 (IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree)); 2483 } 2484 2485 /* This helper is used by netem, as it can hold packets in its 2486 * delay queue. We want to allow the owner socket to send more 2487 * packets, as if they were already TX completed by a typical driver. 2488 * But we also want to keep skb->sk set because some packet schedulers 2489 * rely on it (sch_fq for example). 2490 */ 2491 void skb_orphan_partial(struct sk_buff *skb) 2492 { 2493 if (skb_is_tcp_pure_ack(skb)) 2494 return; 2495 2496 if (can_skb_orphan_partial(skb) && skb_set_owner_sk_safe(skb, skb->sk)) 2497 return; 2498 2499 skb_orphan(skb); 2500 } 2501 EXPORT_SYMBOL(skb_orphan_partial); 2502 2503 /* 2504 * Read buffer destructor automatically called from kfree_skb. 2505 */ 2506 void sock_rfree(struct sk_buff *skb) 2507 { 2508 struct sock *sk = skb->sk; 2509 unsigned int len = skb->truesize; 2510 2511 atomic_sub(len, &sk->sk_rmem_alloc); 2512 sk_mem_uncharge(sk, len); 2513 } 2514 EXPORT_SYMBOL(sock_rfree); 2515 2516 /* 2517 * Buffer destructor for skbs that are not used directly in read or write 2518 * path, e.g. for error handler skbs. Automatically called from kfree_skb. 2519 */ 2520 void sock_efree(struct sk_buff *skb) 2521 { 2522 sock_put(skb->sk); 2523 } 2524 EXPORT_SYMBOL(sock_efree); 2525 2526 /* Buffer destructor for prefetch/receive path where reference count may 2527 * not be held, e.g. for listen sockets. 2528 */ 2529 #ifdef CONFIG_INET 2530 void sock_pfree(struct sk_buff *skb) 2531 { 2532 if (sk_is_refcounted(skb->sk)) 2533 sock_gen_put(skb->sk); 2534 } 2535 EXPORT_SYMBOL(sock_pfree); 2536 #endif /* CONFIG_INET */ 2537 2538 kuid_t sock_i_uid(struct sock *sk) 2539 { 2540 kuid_t uid; 2541 2542 read_lock_bh(&sk->sk_callback_lock); 2543 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID; 2544 read_unlock_bh(&sk->sk_callback_lock); 2545 return uid; 2546 } 2547 EXPORT_SYMBOL(sock_i_uid); 2548 2549 unsigned long sock_i_ino(struct sock *sk) 2550 { 2551 unsigned long ino; 2552 2553 read_lock_bh(&sk->sk_callback_lock); 2554 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0; 2555 read_unlock_bh(&sk->sk_callback_lock); 2556 return ino; 2557 } 2558 EXPORT_SYMBOL(sock_i_ino); 2559 2560 /* 2561 * Allocate a skb from the socket's send buffer. 2562 */ 2563 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 2564 gfp_t priority) 2565 { 2566 if (force || 2567 refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) { 2568 struct sk_buff *skb = alloc_skb(size, priority); 2569 2570 if (skb) { 2571 skb_set_owner_w(skb, sk); 2572 return skb; 2573 } 2574 } 2575 return NULL; 2576 } 2577 EXPORT_SYMBOL(sock_wmalloc); 2578 2579 static void sock_ofree(struct sk_buff *skb) 2580 { 2581 struct sock *sk = skb->sk; 2582 2583 atomic_sub(skb->truesize, &sk->sk_omem_alloc); 2584 } 2585 2586 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size, 2587 gfp_t priority) 2588 { 2589 struct sk_buff *skb; 2590 2591 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */ 2592 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) > 2593 READ_ONCE(sysctl_optmem_max)) 2594 return NULL; 2595 2596 skb = alloc_skb(size, priority); 2597 if (!skb) 2598 return NULL; 2599 2600 atomic_add(skb->truesize, &sk->sk_omem_alloc); 2601 skb->sk = sk; 2602 skb->destructor = sock_ofree; 2603 return skb; 2604 } 2605 2606 /* 2607 * Allocate a memory block from the socket's option memory buffer. 2608 */ 2609 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority) 2610 { 2611 int optmem_max = READ_ONCE(sysctl_optmem_max); 2612 2613 if ((unsigned int)size <= optmem_max && 2614 atomic_read(&sk->sk_omem_alloc) + size < optmem_max) { 2615 void *mem; 2616 /* First do the add, to avoid the race if kmalloc 2617 * might sleep. 2618 */ 2619 atomic_add(size, &sk->sk_omem_alloc); 2620 mem = kmalloc(size, priority); 2621 if (mem) 2622 return mem; 2623 atomic_sub(size, &sk->sk_omem_alloc); 2624 } 2625 return NULL; 2626 } 2627 EXPORT_SYMBOL(sock_kmalloc); 2628 2629 /* Free an option memory block. Note, we actually want the inline 2630 * here as this allows gcc to detect the nullify and fold away the 2631 * condition entirely. 2632 */ 2633 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size, 2634 const bool nullify) 2635 { 2636 if (WARN_ON_ONCE(!mem)) 2637 return; 2638 if (nullify) 2639 kfree_sensitive(mem); 2640 else 2641 kfree(mem); 2642 atomic_sub(size, &sk->sk_omem_alloc); 2643 } 2644 2645 void sock_kfree_s(struct sock *sk, void *mem, int size) 2646 { 2647 __sock_kfree_s(sk, mem, size, false); 2648 } 2649 EXPORT_SYMBOL(sock_kfree_s); 2650 2651 void sock_kzfree_s(struct sock *sk, void *mem, int size) 2652 { 2653 __sock_kfree_s(sk, mem, size, true); 2654 } 2655 EXPORT_SYMBOL(sock_kzfree_s); 2656 2657 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock. 2658 I think, these locks should be removed for datagram sockets. 2659 */ 2660 static long sock_wait_for_wmem(struct sock *sk, long timeo) 2661 { 2662 DEFINE_WAIT(wait); 2663 2664 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk); 2665 for (;;) { 2666 if (!timeo) 2667 break; 2668 if (signal_pending(current)) 2669 break; 2670 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 2671 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 2672 if (refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) 2673 break; 2674 if (sk->sk_shutdown & SEND_SHUTDOWN) 2675 break; 2676 if (sk->sk_err) 2677 break; 2678 timeo = schedule_timeout(timeo); 2679 } 2680 finish_wait(sk_sleep(sk), &wait); 2681 return timeo; 2682 } 2683 2684 2685 /* 2686 * Generic send/receive buffer handlers 2687 */ 2688 2689 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 2690 unsigned long data_len, int noblock, 2691 int *errcode, int max_page_order) 2692 { 2693 struct sk_buff *skb; 2694 long timeo; 2695 int err; 2696 2697 timeo = sock_sndtimeo(sk, noblock); 2698 for (;;) { 2699 err = sock_error(sk); 2700 if (err != 0) 2701 goto failure; 2702 2703 err = -EPIPE; 2704 if (sk->sk_shutdown & SEND_SHUTDOWN) 2705 goto failure; 2706 2707 if (sk_wmem_alloc_get(sk) < READ_ONCE(sk->sk_sndbuf)) 2708 break; 2709 2710 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk); 2711 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 2712 err = -EAGAIN; 2713 if (!timeo) 2714 goto failure; 2715 if (signal_pending(current)) 2716 goto interrupted; 2717 timeo = sock_wait_for_wmem(sk, timeo); 2718 } 2719 skb = alloc_skb_with_frags(header_len, data_len, max_page_order, 2720 errcode, sk->sk_allocation); 2721 if (skb) 2722 skb_set_owner_w(skb, sk); 2723 return skb; 2724 2725 interrupted: 2726 err = sock_intr_errno(timeo); 2727 failure: 2728 *errcode = err; 2729 return NULL; 2730 } 2731 EXPORT_SYMBOL(sock_alloc_send_pskb); 2732 2733 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg, 2734 struct sockcm_cookie *sockc) 2735 { 2736 u32 tsflags; 2737 2738 switch (cmsg->cmsg_type) { 2739 case SO_MARK: 2740 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) && 2741 !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) 2742 return -EPERM; 2743 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 2744 return -EINVAL; 2745 sockc->mark = *(u32 *)CMSG_DATA(cmsg); 2746 break; 2747 case SO_TIMESTAMPING_OLD: 2748 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 2749 return -EINVAL; 2750 2751 tsflags = *(u32 *)CMSG_DATA(cmsg); 2752 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK) 2753 return -EINVAL; 2754 2755 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK; 2756 sockc->tsflags |= tsflags; 2757 break; 2758 case SCM_TXTIME: 2759 if (!sock_flag(sk, SOCK_TXTIME)) 2760 return -EINVAL; 2761 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64))) 2762 return -EINVAL; 2763 sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg)); 2764 break; 2765 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */ 2766 case SCM_RIGHTS: 2767 case SCM_CREDENTIALS: 2768 break; 2769 default: 2770 return -EINVAL; 2771 } 2772 return 0; 2773 } 2774 EXPORT_SYMBOL(__sock_cmsg_send); 2775 2776 int sock_cmsg_send(struct sock *sk, struct msghdr *msg, 2777 struct sockcm_cookie *sockc) 2778 { 2779 struct cmsghdr *cmsg; 2780 int ret; 2781 2782 for_each_cmsghdr(cmsg, msg) { 2783 if (!CMSG_OK(msg, cmsg)) 2784 return -EINVAL; 2785 if (cmsg->cmsg_level != SOL_SOCKET) 2786 continue; 2787 ret = __sock_cmsg_send(sk, msg, cmsg, sockc); 2788 if (ret) 2789 return ret; 2790 } 2791 return 0; 2792 } 2793 EXPORT_SYMBOL(sock_cmsg_send); 2794 2795 static void sk_enter_memory_pressure(struct sock *sk) 2796 { 2797 if (!sk->sk_prot->enter_memory_pressure) 2798 return; 2799 2800 sk->sk_prot->enter_memory_pressure(sk); 2801 } 2802 2803 static void sk_leave_memory_pressure(struct sock *sk) 2804 { 2805 if (sk->sk_prot->leave_memory_pressure) { 2806 sk->sk_prot->leave_memory_pressure(sk); 2807 } else { 2808 unsigned long *memory_pressure = sk->sk_prot->memory_pressure; 2809 2810 if (memory_pressure && READ_ONCE(*memory_pressure)) 2811 WRITE_ONCE(*memory_pressure, 0); 2812 } 2813 } 2814 2815 DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key); 2816 2817 /** 2818 * skb_page_frag_refill - check that a page_frag contains enough room 2819 * @sz: minimum size of the fragment we want to get 2820 * @pfrag: pointer to page_frag 2821 * @gfp: priority for memory allocation 2822 * 2823 * Note: While this allocator tries to use high order pages, there is 2824 * no guarantee that allocations succeed. Therefore, @sz MUST be 2825 * less or equal than PAGE_SIZE. 2826 */ 2827 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp) 2828 { 2829 if (pfrag->page) { 2830 if (page_ref_count(pfrag->page) == 1) { 2831 pfrag->offset = 0; 2832 return true; 2833 } 2834 if (pfrag->offset + sz <= pfrag->size) 2835 return true; 2836 put_page(pfrag->page); 2837 } 2838 2839 pfrag->offset = 0; 2840 if (SKB_FRAG_PAGE_ORDER && 2841 !static_branch_unlikely(&net_high_order_alloc_disable_key)) { 2842 /* Avoid direct reclaim but allow kswapd to wake */ 2843 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) | 2844 __GFP_COMP | __GFP_NOWARN | 2845 __GFP_NORETRY, 2846 SKB_FRAG_PAGE_ORDER); 2847 if (likely(pfrag->page)) { 2848 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER; 2849 return true; 2850 } 2851 } 2852 pfrag->page = alloc_page(gfp); 2853 if (likely(pfrag->page)) { 2854 pfrag->size = PAGE_SIZE; 2855 return true; 2856 } 2857 return false; 2858 } 2859 EXPORT_SYMBOL(skb_page_frag_refill); 2860 2861 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag) 2862 { 2863 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation))) 2864 return true; 2865 2866 sk_enter_memory_pressure(sk); 2867 sk_stream_moderate_sndbuf(sk); 2868 return false; 2869 } 2870 EXPORT_SYMBOL(sk_page_frag_refill); 2871 2872 void __lock_sock(struct sock *sk) 2873 __releases(&sk->sk_lock.slock) 2874 __acquires(&sk->sk_lock.slock) 2875 { 2876 DEFINE_WAIT(wait); 2877 2878 for (;;) { 2879 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait, 2880 TASK_UNINTERRUPTIBLE); 2881 spin_unlock_bh(&sk->sk_lock.slock); 2882 schedule(); 2883 spin_lock_bh(&sk->sk_lock.slock); 2884 if (!sock_owned_by_user(sk)) 2885 break; 2886 } 2887 finish_wait(&sk->sk_lock.wq, &wait); 2888 } 2889 2890 void __release_sock(struct sock *sk) 2891 __releases(&sk->sk_lock.slock) 2892 __acquires(&sk->sk_lock.slock) 2893 { 2894 struct sk_buff *skb, *next; 2895 2896 while ((skb = sk->sk_backlog.head) != NULL) { 2897 sk->sk_backlog.head = sk->sk_backlog.tail = NULL; 2898 2899 spin_unlock_bh(&sk->sk_lock.slock); 2900 2901 do { 2902 next = skb->next; 2903 prefetch(next); 2904 DEBUG_NET_WARN_ON_ONCE(skb_dst_is_noref(skb)); 2905 skb_mark_not_on_list(skb); 2906 sk_backlog_rcv(sk, skb); 2907 2908 cond_resched(); 2909 2910 skb = next; 2911 } while (skb != NULL); 2912 2913 spin_lock_bh(&sk->sk_lock.slock); 2914 } 2915 2916 /* 2917 * Doing the zeroing here guarantee we can not loop forever 2918 * while a wild producer attempts to flood us. 2919 */ 2920 sk->sk_backlog.len = 0; 2921 } 2922 2923 void __sk_flush_backlog(struct sock *sk) 2924 { 2925 spin_lock_bh(&sk->sk_lock.slock); 2926 __release_sock(sk); 2927 spin_unlock_bh(&sk->sk_lock.slock); 2928 } 2929 EXPORT_SYMBOL_GPL(__sk_flush_backlog); 2930 2931 /** 2932 * sk_wait_data - wait for data to arrive at sk_receive_queue 2933 * @sk: sock to wait on 2934 * @timeo: for how long 2935 * @skb: last skb seen on sk_receive_queue 2936 * 2937 * Now socket state including sk->sk_err is changed only under lock, 2938 * hence we may omit checks after joining wait queue. 2939 * We check receive queue before schedule() only as optimization; 2940 * it is very likely that release_sock() added new data. 2941 */ 2942 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb) 2943 { 2944 DEFINE_WAIT_FUNC(wait, woken_wake_function); 2945 int rc; 2946 2947 add_wait_queue(sk_sleep(sk), &wait); 2948 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk); 2949 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait); 2950 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk); 2951 remove_wait_queue(sk_sleep(sk), &wait); 2952 return rc; 2953 } 2954 EXPORT_SYMBOL(sk_wait_data); 2955 2956 /** 2957 * __sk_mem_raise_allocated - increase memory_allocated 2958 * @sk: socket 2959 * @size: memory size to allocate 2960 * @amt: pages to allocate 2961 * @kind: allocation type 2962 * 2963 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc 2964 */ 2965 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind) 2966 { 2967 bool memcg_charge = mem_cgroup_sockets_enabled && sk->sk_memcg; 2968 struct proto *prot = sk->sk_prot; 2969 bool charged = true; 2970 long allocated; 2971 2972 sk_memory_allocated_add(sk, amt); 2973 allocated = sk_memory_allocated(sk); 2974 if (memcg_charge && 2975 !(charged = mem_cgroup_charge_skmem(sk->sk_memcg, amt, 2976 gfp_memcg_charge()))) 2977 goto suppress_allocation; 2978 2979 /* Under limit. */ 2980 if (allocated <= sk_prot_mem_limits(sk, 0)) { 2981 sk_leave_memory_pressure(sk); 2982 return 1; 2983 } 2984 2985 /* Under pressure. */ 2986 if (allocated > sk_prot_mem_limits(sk, 1)) 2987 sk_enter_memory_pressure(sk); 2988 2989 /* Over hard limit. */ 2990 if (allocated > sk_prot_mem_limits(sk, 2)) 2991 goto suppress_allocation; 2992 2993 /* guarantee minimum buffer size under pressure */ 2994 if (kind == SK_MEM_RECV) { 2995 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot)) 2996 return 1; 2997 2998 } else { /* SK_MEM_SEND */ 2999 int wmem0 = sk_get_wmem0(sk, prot); 3000 3001 if (sk->sk_type == SOCK_STREAM) { 3002 if (sk->sk_wmem_queued < wmem0) 3003 return 1; 3004 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) { 3005 return 1; 3006 } 3007 } 3008 3009 if (sk_has_memory_pressure(sk)) { 3010 u64 alloc; 3011 3012 if (!sk_under_memory_pressure(sk)) 3013 return 1; 3014 alloc = sk_sockets_allocated_read_positive(sk); 3015 if (sk_prot_mem_limits(sk, 2) > alloc * 3016 sk_mem_pages(sk->sk_wmem_queued + 3017 atomic_read(&sk->sk_rmem_alloc) + 3018 sk->sk_forward_alloc)) 3019 return 1; 3020 } 3021 3022 suppress_allocation: 3023 3024 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) { 3025 sk_stream_moderate_sndbuf(sk); 3026 3027 /* Fail only if socket is _under_ its sndbuf. 3028 * In this case we cannot block, so that we have to fail. 3029 */ 3030 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf) { 3031 /* Force charge with __GFP_NOFAIL */ 3032 if (memcg_charge && !charged) { 3033 mem_cgroup_charge_skmem(sk->sk_memcg, amt, 3034 gfp_memcg_charge() | __GFP_NOFAIL); 3035 } 3036 return 1; 3037 } 3038 } 3039 3040 if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged)) 3041 trace_sock_exceed_buf_limit(sk, prot, allocated, kind); 3042 3043 sk_memory_allocated_sub(sk, amt); 3044 3045 if (memcg_charge && charged) 3046 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt); 3047 3048 return 0; 3049 } 3050 3051 /** 3052 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated 3053 * @sk: socket 3054 * @size: memory size to allocate 3055 * @kind: allocation type 3056 * 3057 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means 3058 * rmem allocation. This function assumes that protocols which have 3059 * memory_pressure use sk_wmem_queued as write buffer accounting. 3060 */ 3061 int __sk_mem_schedule(struct sock *sk, int size, int kind) 3062 { 3063 int ret, amt = sk_mem_pages(size); 3064 3065 sk->sk_forward_alloc += amt << PAGE_SHIFT; 3066 ret = __sk_mem_raise_allocated(sk, size, amt, kind); 3067 if (!ret) 3068 sk->sk_forward_alloc -= amt << PAGE_SHIFT; 3069 return ret; 3070 } 3071 EXPORT_SYMBOL(__sk_mem_schedule); 3072 3073 /** 3074 * __sk_mem_reduce_allocated - reclaim memory_allocated 3075 * @sk: socket 3076 * @amount: number of quanta 3077 * 3078 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc 3079 */ 3080 void __sk_mem_reduce_allocated(struct sock *sk, int amount) 3081 { 3082 sk_memory_allocated_sub(sk, amount); 3083 3084 if (mem_cgroup_sockets_enabled && sk->sk_memcg) 3085 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount); 3086 3087 if (sk_under_memory_pressure(sk) && 3088 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0))) 3089 sk_leave_memory_pressure(sk); 3090 } 3091 3092 /** 3093 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated 3094 * @sk: socket 3095 * @amount: number of bytes (rounded down to a PAGE_SIZE multiple) 3096 */ 3097 void __sk_mem_reclaim(struct sock *sk, int amount) 3098 { 3099 amount >>= PAGE_SHIFT; 3100 sk->sk_forward_alloc -= amount << PAGE_SHIFT; 3101 __sk_mem_reduce_allocated(sk, amount); 3102 } 3103 EXPORT_SYMBOL(__sk_mem_reclaim); 3104 3105 int sk_set_peek_off(struct sock *sk, int val) 3106 { 3107 sk->sk_peek_off = val; 3108 return 0; 3109 } 3110 EXPORT_SYMBOL_GPL(sk_set_peek_off); 3111 3112 /* 3113 * Set of default routines for initialising struct proto_ops when 3114 * the protocol does not support a particular function. In certain 3115 * cases where it makes no sense for a protocol to have a "do nothing" 3116 * function, some default processing is provided. 3117 */ 3118 3119 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len) 3120 { 3121 return -EOPNOTSUPP; 3122 } 3123 EXPORT_SYMBOL(sock_no_bind); 3124 3125 int sock_no_connect(struct socket *sock, struct sockaddr *saddr, 3126 int len, int flags) 3127 { 3128 return -EOPNOTSUPP; 3129 } 3130 EXPORT_SYMBOL(sock_no_connect); 3131 3132 int sock_no_socketpair(struct socket *sock1, struct socket *sock2) 3133 { 3134 return -EOPNOTSUPP; 3135 } 3136 EXPORT_SYMBOL(sock_no_socketpair); 3137 3138 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags, 3139 bool kern) 3140 { 3141 return -EOPNOTSUPP; 3142 } 3143 EXPORT_SYMBOL(sock_no_accept); 3144 3145 int sock_no_getname(struct socket *sock, struct sockaddr *saddr, 3146 int peer) 3147 { 3148 return -EOPNOTSUPP; 3149 } 3150 EXPORT_SYMBOL(sock_no_getname); 3151 3152 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 3153 { 3154 return -EOPNOTSUPP; 3155 } 3156 EXPORT_SYMBOL(sock_no_ioctl); 3157 3158 int sock_no_listen(struct socket *sock, int backlog) 3159 { 3160 return -EOPNOTSUPP; 3161 } 3162 EXPORT_SYMBOL(sock_no_listen); 3163 3164 int sock_no_shutdown(struct socket *sock, int how) 3165 { 3166 return -EOPNOTSUPP; 3167 } 3168 EXPORT_SYMBOL(sock_no_shutdown); 3169 3170 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len) 3171 { 3172 return -EOPNOTSUPP; 3173 } 3174 EXPORT_SYMBOL(sock_no_sendmsg); 3175 3176 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len) 3177 { 3178 return -EOPNOTSUPP; 3179 } 3180 EXPORT_SYMBOL(sock_no_sendmsg_locked); 3181 3182 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len, 3183 int flags) 3184 { 3185 return -EOPNOTSUPP; 3186 } 3187 EXPORT_SYMBOL(sock_no_recvmsg); 3188 3189 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma) 3190 { 3191 /* Mirror missing mmap method error code */ 3192 return -ENODEV; 3193 } 3194 EXPORT_SYMBOL(sock_no_mmap); 3195 3196 /* 3197 * When a file is received (via SCM_RIGHTS, etc), we must bump the 3198 * various sock-based usage counts. 3199 */ 3200 void __receive_sock(struct file *file) 3201 { 3202 struct socket *sock; 3203 3204 sock = sock_from_file(file); 3205 if (sock) { 3206 sock_update_netprioidx(&sock->sk->sk_cgrp_data); 3207 sock_update_classid(&sock->sk->sk_cgrp_data); 3208 } 3209 } 3210 3211 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags) 3212 { 3213 ssize_t res; 3214 struct msghdr msg = {.msg_flags = flags}; 3215 struct kvec iov; 3216 char *kaddr = kmap(page); 3217 iov.iov_base = kaddr + offset; 3218 iov.iov_len = size; 3219 res = kernel_sendmsg(sock, &msg, &iov, 1, size); 3220 kunmap(page); 3221 return res; 3222 } 3223 EXPORT_SYMBOL(sock_no_sendpage); 3224 3225 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page, 3226 int offset, size_t size, int flags) 3227 { 3228 ssize_t res; 3229 struct msghdr msg = {.msg_flags = flags}; 3230 struct kvec iov; 3231 char *kaddr = kmap(page); 3232 3233 iov.iov_base = kaddr + offset; 3234 iov.iov_len = size; 3235 res = kernel_sendmsg_locked(sk, &msg, &iov, 1, size); 3236 kunmap(page); 3237 return res; 3238 } 3239 EXPORT_SYMBOL(sock_no_sendpage_locked); 3240 3241 /* 3242 * Default Socket Callbacks 3243 */ 3244 3245 static void sock_def_wakeup(struct sock *sk) 3246 { 3247 struct socket_wq *wq; 3248 3249 rcu_read_lock(); 3250 wq = rcu_dereference(sk->sk_wq); 3251 if (skwq_has_sleeper(wq)) 3252 wake_up_interruptible_all(&wq->wait); 3253 rcu_read_unlock(); 3254 } 3255 3256 static void sock_def_error_report(struct sock *sk) 3257 { 3258 struct socket_wq *wq; 3259 3260 rcu_read_lock(); 3261 wq = rcu_dereference(sk->sk_wq); 3262 if (skwq_has_sleeper(wq)) 3263 wake_up_interruptible_poll(&wq->wait, EPOLLERR); 3264 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR); 3265 rcu_read_unlock(); 3266 } 3267 3268 void sock_def_readable(struct sock *sk) 3269 { 3270 struct socket_wq *wq; 3271 3272 rcu_read_lock(); 3273 wq = rcu_dereference(sk->sk_wq); 3274 if (skwq_has_sleeper(wq)) 3275 wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI | 3276 EPOLLRDNORM | EPOLLRDBAND); 3277 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN); 3278 rcu_read_unlock(); 3279 } 3280 3281 static void sock_def_write_space(struct sock *sk) 3282 { 3283 struct socket_wq *wq; 3284 3285 rcu_read_lock(); 3286 3287 /* Do not wake up a writer until he can make "significant" 3288 * progress. --DaveM 3289 */ 3290 if (sock_writeable(sk)) { 3291 wq = rcu_dereference(sk->sk_wq); 3292 if (skwq_has_sleeper(wq)) 3293 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT | 3294 EPOLLWRNORM | EPOLLWRBAND); 3295 3296 /* Should agree with poll, otherwise some programs break */ 3297 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT); 3298 } 3299 3300 rcu_read_unlock(); 3301 } 3302 3303 /* An optimised version of sock_def_write_space(), should only be called 3304 * for SOCK_RCU_FREE sockets under RCU read section and after putting 3305 * ->sk_wmem_alloc. 3306 */ 3307 static void sock_def_write_space_wfree(struct sock *sk) 3308 { 3309 /* Do not wake up a writer until he can make "significant" 3310 * progress. --DaveM 3311 */ 3312 if (sock_writeable(sk)) { 3313 struct socket_wq *wq = rcu_dereference(sk->sk_wq); 3314 3315 /* rely on refcount_sub from sock_wfree() */ 3316 smp_mb__after_atomic(); 3317 if (wq && waitqueue_active(&wq->wait)) 3318 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT | 3319 EPOLLWRNORM | EPOLLWRBAND); 3320 3321 /* Should agree with poll, otherwise some programs break */ 3322 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT); 3323 } 3324 } 3325 3326 static void sock_def_destruct(struct sock *sk) 3327 { 3328 } 3329 3330 void sk_send_sigurg(struct sock *sk) 3331 { 3332 if (sk->sk_socket && sk->sk_socket->file) 3333 if (send_sigurg(&sk->sk_socket->file->f_owner)) 3334 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI); 3335 } 3336 EXPORT_SYMBOL(sk_send_sigurg); 3337 3338 void sk_reset_timer(struct sock *sk, struct timer_list* timer, 3339 unsigned long expires) 3340 { 3341 if (!mod_timer(timer, expires)) 3342 sock_hold(sk); 3343 } 3344 EXPORT_SYMBOL(sk_reset_timer); 3345 3346 void sk_stop_timer(struct sock *sk, struct timer_list* timer) 3347 { 3348 if (del_timer(timer)) 3349 __sock_put(sk); 3350 } 3351 EXPORT_SYMBOL(sk_stop_timer); 3352 3353 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer) 3354 { 3355 if (del_timer_sync(timer)) 3356 __sock_put(sk); 3357 } 3358 EXPORT_SYMBOL(sk_stop_timer_sync); 3359 3360 void sock_init_data(struct socket *sock, struct sock *sk) 3361 { 3362 sk_init_common(sk); 3363 sk->sk_send_head = NULL; 3364 3365 timer_setup(&sk->sk_timer, NULL, 0); 3366 3367 sk->sk_allocation = GFP_KERNEL; 3368 sk->sk_rcvbuf = READ_ONCE(sysctl_rmem_default); 3369 sk->sk_sndbuf = READ_ONCE(sysctl_wmem_default); 3370 sk->sk_state = TCP_CLOSE; 3371 sk_set_socket(sk, sock); 3372 3373 sock_set_flag(sk, SOCK_ZAPPED); 3374 3375 if (sock) { 3376 sk->sk_type = sock->type; 3377 RCU_INIT_POINTER(sk->sk_wq, &sock->wq); 3378 sock->sk = sk; 3379 sk->sk_uid = SOCK_INODE(sock)->i_uid; 3380 } else { 3381 RCU_INIT_POINTER(sk->sk_wq, NULL); 3382 sk->sk_uid = make_kuid(sock_net(sk)->user_ns, 0); 3383 } 3384 3385 rwlock_init(&sk->sk_callback_lock); 3386 if (sk->sk_kern_sock) 3387 lockdep_set_class_and_name( 3388 &sk->sk_callback_lock, 3389 af_kern_callback_keys + sk->sk_family, 3390 af_family_kern_clock_key_strings[sk->sk_family]); 3391 else 3392 lockdep_set_class_and_name( 3393 &sk->sk_callback_lock, 3394 af_callback_keys + sk->sk_family, 3395 af_family_clock_key_strings[sk->sk_family]); 3396 3397 sk->sk_state_change = sock_def_wakeup; 3398 sk->sk_data_ready = sock_def_readable; 3399 sk->sk_write_space = sock_def_write_space; 3400 sk->sk_error_report = sock_def_error_report; 3401 sk->sk_destruct = sock_def_destruct; 3402 3403 sk->sk_frag.page = NULL; 3404 sk->sk_frag.offset = 0; 3405 sk->sk_peek_off = -1; 3406 3407 sk->sk_peer_pid = NULL; 3408 sk->sk_peer_cred = NULL; 3409 spin_lock_init(&sk->sk_peer_lock); 3410 3411 sk->sk_write_pending = 0; 3412 sk->sk_rcvlowat = 1; 3413 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; 3414 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT; 3415 3416 sk->sk_stamp = SK_DEFAULT_STAMP; 3417 #if BITS_PER_LONG==32 3418 seqlock_init(&sk->sk_stamp_seq); 3419 #endif 3420 atomic_set(&sk->sk_zckey, 0); 3421 3422 #ifdef CONFIG_NET_RX_BUSY_POLL 3423 sk->sk_napi_id = 0; 3424 sk->sk_ll_usec = READ_ONCE(sysctl_net_busy_read); 3425 #endif 3426 3427 sk->sk_max_pacing_rate = ~0UL; 3428 sk->sk_pacing_rate = ~0UL; 3429 WRITE_ONCE(sk->sk_pacing_shift, 10); 3430 sk->sk_incoming_cpu = -1; 3431 sk->sk_txrehash = SOCK_TXREHASH_DEFAULT; 3432 3433 sk_rx_queue_clear(sk); 3434 /* 3435 * Before updating sk_refcnt, we must commit prior changes to memory 3436 * (Documentation/RCU/rculist_nulls.rst for details) 3437 */ 3438 smp_wmb(); 3439 refcount_set(&sk->sk_refcnt, 1); 3440 atomic_set(&sk->sk_drops, 0); 3441 } 3442 EXPORT_SYMBOL(sock_init_data); 3443 3444 void lock_sock_nested(struct sock *sk, int subclass) 3445 { 3446 /* The sk_lock has mutex_lock() semantics here. */ 3447 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_); 3448 3449 might_sleep(); 3450 spin_lock_bh(&sk->sk_lock.slock); 3451 if (sock_owned_by_user_nocheck(sk)) 3452 __lock_sock(sk); 3453 sk->sk_lock.owned = 1; 3454 spin_unlock_bh(&sk->sk_lock.slock); 3455 } 3456 EXPORT_SYMBOL(lock_sock_nested); 3457 3458 void release_sock(struct sock *sk) 3459 { 3460 spin_lock_bh(&sk->sk_lock.slock); 3461 if (sk->sk_backlog.tail) 3462 __release_sock(sk); 3463 3464 /* Warning : release_cb() might need to release sk ownership, 3465 * ie call sock_release_ownership(sk) before us. 3466 */ 3467 if (sk->sk_prot->release_cb) 3468 sk->sk_prot->release_cb(sk); 3469 3470 sock_release_ownership(sk); 3471 if (waitqueue_active(&sk->sk_lock.wq)) 3472 wake_up(&sk->sk_lock.wq); 3473 spin_unlock_bh(&sk->sk_lock.slock); 3474 } 3475 EXPORT_SYMBOL(release_sock); 3476 3477 bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock) 3478 { 3479 might_sleep(); 3480 spin_lock_bh(&sk->sk_lock.slock); 3481 3482 if (!sock_owned_by_user_nocheck(sk)) { 3483 /* 3484 * Fast path return with bottom halves disabled and 3485 * sock::sk_lock.slock held. 3486 * 3487 * The 'mutex' is not contended and holding 3488 * sock::sk_lock.slock prevents all other lockers to 3489 * proceed so the corresponding unlock_sock_fast() can 3490 * avoid the slow path of release_sock() completely and 3491 * just release slock. 3492 * 3493 * From a semantical POV this is equivalent to 'acquiring' 3494 * the 'mutex', hence the corresponding lockdep 3495 * mutex_release() has to happen in the fast path of 3496 * unlock_sock_fast(). 3497 */ 3498 return false; 3499 } 3500 3501 __lock_sock(sk); 3502 sk->sk_lock.owned = 1; 3503 __acquire(&sk->sk_lock.slock); 3504 spin_unlock_bh(&sk->sk_lock.slock); 3505 return true; 3506 } 3507 EXPORT_SYMBOL(__lock_sock_fast); 3508 3509 int sock_gettstamp(struct socket *sock, void __user *userstamp, 3510 bool timeval, bool time32) 3511 { 3512 struct sock *sk = sock->sk; 3513 struct timespec64 ts; 3514 3515 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 3516 ts = ktime_to_timespec64(sock_read_timestamp(sk)); 3517 if (ts.tv_sec == -1) 3518 return -ENOENT; 3519 if (ts.tv_sec == 0) { 3520 ktime_t kt = ktime_get_real(); 3521 sock_write_timestamp(sk, kt); 3522 ts = ktime_to_timespec64(kt); 3523 } 3524 3525 if (timeval) 3526 ts.tv_nsec /= 1000; 3527 3528 #ifdef CONFIG_COMPAT_32BIT_TIME 3529 if (time32) 3530 return put_old_timespec32(&ts, userstamp); 3531 #endif 3532 #ifdef CONFIG_SPARC64 3533 /* beware of padding in sparc64 timeval */ 3534 if (timeval && !in_compat_syscall()) { 3535 struct __kernel_old_timeval __user tv = { 3536 .tv_sec = ts.tv_sec, 3537 .tv_usec = ts.tv_nsec, 3538 }; 3539 if (copy_to_user(userstamp, &tv, sizeof(tv))) 3540 return -EFAULT; 3541 return 0; 3542 } 3543 #endif 3544 return put_timespec64(&ts, userstamp); 3545 } 3546 EXPORT_SYMBOL(sock_gettstamp); 3547 3548 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag) 3549 { 3550 if (!sock_flag(sk, flag)) { 3551 unsigned long previous_flags = sk->sk_flags; 3552 3553 sock_set_flag(sk, flag); 3554 /* 3555 * we just set one of the two flags which require net 3556 * time stamping, but time stamping might have been on 3557 * already because of the other one 3558 */ 3559 if (sock_needs_netstamp(sk) && 3560 !(previous_flags & SK_FLAGS_TIMESTAMP)) 3561 net_enable_timestamp(); 3562 } 3563 } 3564 3565 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, 3566 int level, int type) 3567 { 3568 struct sock_exterr_skb *serr; 3569 struct sk_buff *skb; 3570 int copied, err; 3571 3572 err = -EAGAIN; 3573 skb = sock_dequeue_err_skb(sk); 3574 if (skb == NULL) 3575 goto out; 3576 3577 copied = skb->len; 3578 if (copied > len) { 3579 msg->msg_flags |= MSG_TRUNC; 3580 copied = len; 3581 } 3582 err = skb_copy_datagram_msg(skb, 0, msg, copied); 3583 if (err) 3584 goto out_free_skb; 3585 3586 sock_recv_timestamp(msg, sk, skb); 3587 3588 serr = SKB_EXT_ERR(skb); 3589 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee); 3590 3591 msg->msg_flags |= MSG_ERRQUEUE; 3592 err = copied; 3593 3594 out_free_skb: 3595 kfree_skb(skb); 3596 out: 3597 return err; 3598 } 3599 EXPORT_SYMBOL(sock_recv_errqueue); 3600 3601 /* 3602 * Get a socket option on an socket. 3603 * 3604 * FIX: POSIX 1003.1g is very ambiguous here. It states that 3605 * asynchronous errors should be reported by getsockopt. We assume 3606 * this means if you specify SO_ERROR (otherwise whats the point of it). 3607 */ 3608 int sock_common_getsockopt(struct socket *sock, int level, int optname, 3609 char __user *optval, int __user *optlen) 3610 { 3611 struct sock *sk = sock->sk; 3612 3613 /* IPV6_ADDRFORM can change sk->sk_prot under us. */ 3614 return READ_ONCE(sk->sk_prot)->getsockopt(sk, level, optname, optval, optlen); 3615 } 3616 EXPORT_SYMBOL(sock_common_getsockopt); 3617 3618 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 3619 int flags) 3620 { 3621 struct sock *sk = sock->sk; 3622 int addr_len = 0; 3623 int err; 3624 3625 err = sk->sk_prot->recvmsg(sk, msg, size, flags, &addr_len); 3626 if (err >= 0) 3627 msg->msg_namelen = addr_len; 3628 return err; 3629 } 3630 EXPORT_SYMBOL(sock_common_recvmsg); 3631 3632 /* 3633 * Set socket options on an inet socket. 3634 */ 3635 int sock_common_setsockopt(struct socket *sock, int level, int optname, 3636 sockptr_t optval, unsigned int optlen) 3637 { 3638 struct sock *sk = sock->sk; 3639 3640 /* IPV6_ADDRFORM can change sk->sk_prot under us. */ 3641 return READ_ONCE(sk->sk_prot)->setsockopt(sk, level, optname, optval, optlen); 3642 } 3643 EXPORT_SYMBOL(sock_common_setsockopt); 3644 3645 void sk_common_release(struct sock *sk) 3646 { 3647 if (sk->sk_prot->destroy) 3648 sk->sk_prot->destroy(sk); 3649 3650 /* 3651 * Observation: when sk_common_release is called, processes have 3652 * no access to socket. But net still has. 3653 * Step one, detach it from networking: 3654 * 3655 * A. Remove from hash tables. 3656 */ 3657 3658 sk->sk_prot->unhash(sk); 3659 3660 /* 3661 * In this point socket cannot receive new packets, but it is possible 3662 * that some packets are in flight because some CPU runs receiver and 3663 * did hash table lookup before we unhashed socket. They will achieve 3664 * receive queue and will be purged by socket destructor. 3665 * 3666 * Also we still have packets pending on receive queue and probably, 3667 * our own packets waiting in device queues. sock_destroy will drain 3668 * receive queue, but transmitted packets will delay socket destruction 3669 * until the last reference will be released. 3670 */ 3671 3672 sock_orphan(sk); 3673 3674 xfrm_sk_free_policy(sk); 3675 3676 sk_refcnt_debug_release(sk); 3677 3678 sock_put(sk); 3679 } 3680 EXPORT_SYMBOL(sk_common_release); 3681 3682 void sk_get_meminfo(const struct sock *sk, u32 *mem) 3683 { 3684 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS); 3685 3686 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk); 3687 mem[SK_MEMINFO_RCVBUF] = READ_ONCE(sk->sk_rcvbuf); 3688 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk); 3689 mem[SK_MEMINFO_SNDBUF] = READ_ONCE(sk->sk_sndbuf); 3690 mem[SK_MEMINFO_FWD_ALLOC] = sk->sk_forward_alloc; 3691 mem[SK_MEMINFO_WMEM_QUEUED] = READ_ONCE(sk->sk_wmem_queued); 3692 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc); 3693 mem[SK_MEMINFO_BACKLOG] = READ_ONCE(sk->sk_backlog.len); 3694 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops); 3695 } 3696 3697 #ifdef CONFIG_PROC_FS 3698 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR); 3699 3700 int sock_prot_inuse_get(struct net *net, struct proto *prot) 3701 { 3702 int cpu, idx = prot->inuse_idx; 3703 int res = 0; 3704 3705 for_each_possible_cpu(cpu) 3706 res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx]; 3707 3708 return res >= 0 ? res : 0; 3709 } 3710 EXPORT_SYMBOL_GPL(sock_prot_inuse_get); 3711 3712 int sock_inuse_get(struct net *net) 3713 { 3714 int cpu, res = 0; 3715 3716 for_each_possible_cpu(cpu) 3717 res += per_cpu_ptr(net->core.prot_inuse, cpu)->all; 3718 3719 return res; 3720 } 3721 3722 EXPORT_SYMBOL_GPL(sock_inuse_get); 3723 3724 static int __net_init sock_inuse_init_net(struct net *net) 3725 { 3726 net->core.prot_inuse = alloc_percpu(struct prot_inuse); 3727 if (net->core.prot_inuse == NULL) 3728 return -ENOMEM; 3729 return 0; 3730 } 3731 3732 static void __net_exit sock_inuse_exit_net(struct net *net) 3733 { 3734 free_percpu(net->core.prot_inuse); 3735 } 3736 3737 static struct pernet_operations net_inuse_ops = { 3738 .init = sock_inuse_init_net, 3739 .exit = sock_inuse_exit_net, 3740 }; 3741 3742 static __init int net_inuse_init(void) 3743 { 3744 if (register_pernet_subsys(&net_inuse_ops)) 3745 panic("Cannot initialize net inuse counters"); 3746 3747 return 0; 3748 } 3749 3750 core_initcall(net_inuse_init); 3751 3752 static int assign_proto_idx(struct proto *prot) 3753 { 3754 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR); 3755 3756 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) { 3757 pr_err("PROTO_INUSE_NR exhausted\n"); 3758 return -ENOSPC; 3759 } 3760 3761 set_bit(prot->inuse_idx, proto_inuse_idx); 3762 return 0; 3763 } 3764 3765 static void release_proto_idx(struct proto *prot) 3766 { 3767 if (prot->inuse_idx != PROTO_INUSE_NR - 1) 3768 clear_bit(prot->inuse_idx, proto_inuse_idx); 3769 } 3770 #else 3771 static inline int assign_proto_idx(struct proto *prot) 3772 { 3773 return 0; 3774 } 3775 3776 static inline void release_proto_idx(struct proto *prot) 3777 { 3778 } 3779 3780 #endif 3781 3782 static void tw_prot_cleanup(struct timewait_sock_ops *twsk_prot) 3783 { 3784 if (!twsk_prot) 3785 return; 3786 kfree(twsk_prot->twsk_slab_name); 3787 twsk_prot->twsk_slab_name = NULL; 3788 kmem_cache_destroy(twsk_prot->twsk_slab); 3789 twsk_prot->twsk_slab = NULL; 3790 } 3791 3792 static int tw_prot_init(const struct proto *prot) 3793 { 3794 struct timewait_sock_ops *twsk_prot = prot->twsk_prot; 3795 3796 if (!twsk_prot) 3797 return 0; 3798 3799 twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", 3800 prot->name); 3801 if (!twsk_prot->twsk_slab_name) 3802 return -ENOMEM; 3803 3804 twsk_prot->twsk_slab = 3805 kmem_cache_create(twsk_prot->twsk_slab_name, 3806 twsk_prot->twsk_obj_size, 0, 3807 SLAB_ACCOUNT | prot->slab_flags, 3808 NULL); 3809 if (!twsk_prot->twsk_slab) { 3810 pr_crit("%s: Can't create timewait sock SLAB cache!\n", 3811 prot->name); 3812 return -ENOMEM; 3813 } 3814 3815 return 0; 3816 } 3817 3818 static void req_prot_cleanup(struct request_sock_ops *rsk_prot) 3819 { 3820 if (!rsk_prot) 3821 return; 3822 kfree(rsk_prot->slab_name); 3823 rsk_prot->slab_name = NULL; 3824 kmem_cache_destroy(rsk_prot->slab); 3825 rsk_prot->slab = NULL; 3826 } 3827 3828 static int req_prot_init(const struct proto *prot) 3829 { 3830 struct request_sock_ops *rsk_prot = prot->rsk_prot; 3831 3832 if (!rsk_prot) 3833 return 0; 3834 3835 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", 3836 prot->name); 3837 if (!rsk_prot->slab_name) 3838 return -ENOMEM; 3839 3840 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name, 3841 rsk_prot->obj_size, 0, 3842 SLAB_ACCOUNT | prot->slab_flags, 3843 NULL); 3844 3845 if (!rsk_prot->slab) { 3846 pr_crit("%s: Can't create request sock SLAB cache!\n", 3847 prot->name); 3848 return -ENOMEM; 3849 } 3850 return 0; 3851 } 3852 3853 int proto_register(struct proto *prot, int alloc_slab) 3854 { 3855 int ret = -ENOBUFS; 3856 3857 if (prot->memory_allocated && !prot->sysctl_mem) { 3858 pr_err("%s: missing sysctl_mem\n", prot->name); 3859 return -EINVAL; 3860 } 3861 if (prot->memory_allocated && !prot->per_cpu_fw_alloc) { 3862 pr_err("%s: missing per_cpu_fw_alloc\n", prot->name); 3863 return -EINVAL; 3864 } 3865 if (alloc_slab) { 3866 prot->slab = kmem_cache_create_usercopy(prot->name, 3867 prot->obj_size, 0, 3868 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT | 3869 prot->slab_flags, 3870 prot->useroffset, prot->usersize, 3871 NULL); 3872 3873 if (prot->slab == NULL) { 3874 pr_crit("%s: Can't create sock SLAB cache!\n", 3875 prot->name); 3876 goto out; 3877 } 3878 3879 if (req_prot_init(prot)) 3880 goto out_free_request_sock_slab; 3881 3882 if (tw_prot_init(prot)) 3883 goto out_free_timewait_sock_slab; 3884 } 3885 3886 mutex_lock(&proto_list_mutex); 3887 ret = assign_proto_idx(prot); 3888 if (ret) { 3889 mutex_unlock(&proto_list_mutex); 3890 goto out_free_timewait_sock_slab; 3891 } 3892 list_add(&prot->node, &proto_list); 3893 mutex_unlock(&proto_list_mutex); 3894 return ret; 3895 3896 out_free_timewait_sock_slab: 3897 if (alloc_slab) 3898 tw_prot_cleanup(prot->twsk_prot); 3899 out_free_request_sock_slab: 3900 if (alloc_slab) { 3901 req_prot_cleanup(prot->rsk_prot); 3902 3903 kmem_cache_destroy(prot->slab); 3904 prot->slab = NULL; 3905 } 3906 out: 3907 return ret; 3908 } 3909 EXPORT_SYMBOL(proto_register); 3910 3911 void proto_unregister(struct proto *prot) 3912 { 3913 mutex_lock(&proto_list_mutex); 3914 release_proto_idx(prot); 3915 list_del(&prot->node); 3916 mutex_unlock(&proto_list_mutex); 3917 3918 kmem_cache_destroy(prot->slab); 3919 prot->slab = NULL; 3920 3921 req_prot_cleanup(prot->rsk_prot); 3922 tw_prot_cleanup(prot->twsk_prot); 3923 } 3924 EXPORT_SYMBOL(proto_unregister); 3925 3926 int sock_load_diag_module(int family, int protocol) 3927 { 3928 if (!protocol) { 3929 if (!sock_is_registered(family)) 3930 return -ENOENT; 3931 3932 return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK, 3933 NETLINK_SOCK_DIAG, family); 3934 } 3935 3936 #ifdef CONFIG_INET 3937 if (family == AF_INET && 3938 protocol != IPPROTO_RAW && 3939 protocol < MAX_INET_PROTOS && 3940 !rcu_access_pointer(inet_protos[protocol])) 3941 return -ENOENT; 3942 #endif 3943 3944 return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK, 3945 NETLINK_SOCK_DIAG, family, protocol); 3946 } 3947 EXPORT_SYMBOL(sock_load_diag_module); 3948 3949 #ifdef CONFIG_PROC_FS 3950 static void *proto_seq_start(struct seq_file *seq, loff_t *pos) 3951 __acquires(proto_list_mutex) 3952 { 3953 mutex_lock(&proto_list_mutex); 3954 return seq_list_start_head(&proto_list, *pos); 3955 } 3956 3957 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos) 3958 { 3959 return seq_list_next(v, &proto_list, pos); 3960 } 3961 3962 static void proto_seq_stop(struct seq_file *seq, void *v) 3963 __releases(proto_list_mutex) 3964 { 3965 mutex_unlock(&proto_list_mutex); 3966 } 3967 3968 static char proto_method_implemented(const void *method) 3969 { 3970 return method == NULL ? 'n' : 'y'; 3971 } 3972 static long sock_prot_memory_allocated(struct proto *proto) 3973 { 3974 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L; 3975 } 3976 3977 static const char *sock_prot_memory_pressure(struct proto *proto) 3978 { 3979 return proto->memory_pressure != NULL ? 3980 proto_memory_pressure(proto) ? "yes" : "no" : "NI"; 3981 } 3982 3983 static void proto_seq_printf(struct seq_file *seq, struct proto *proto) 3984 { 3985 3986 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s " 3987 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n", 3988 proto->name, 3989 proto->obj_size, 3990 sock_prot_inuse_get(seq_file_net(seq), proto), 3991 sock_prot_memory_allocated(proto), 3992 sock_prot_memory_pressure(proto), 3993 proto->max_header, 3994 proto->slab == NULL ? "no" : "yes", 3995 module_name(proto->owner), 3996 proto_method_implemented(proto->close), 3997 proto_method_implemented(proto->connect), 3998 proto_method_implemented(proto->disconnect), 3999 proto_method_implemented(proto->accept), 4000 proto_method_implemented(proto->ioctl), 4001 proto_method_implemented(proto->init), 4002 proto_method_implemented(proto->destroy), 4003 proto_method_implemented(proto->shutdown), 4004 proto_method_implemented(proto->setsockopt), 4005 proto_method_implemented(proto->getsockopt), 4006 proto_method_implemented(proto->sendmsg), 4007 proto_method_implemented(proto->recvmsg), 4008 proto_method_implemented(proto->sendpage), 4009 proto_method_implemented(proto->bind), 4010 proto_method_implemented(proto->backlog_rcv), 4011 proto_method_implemented(proto->hash), 4012 proto_method_implemented(proto->unhash), 4013 proto_method_implemented(proto->get_port), 4014 proto_method_implemented(proto->enter_memory_pressure)); 4015 } 4016 4017 static int proto_seq_show(struct seq_file *seq, void *v) 4018 { 4019 if (v == &proto_list) 4020 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s", 4021 "protocol", 4022 "size", 4023 "sockets", 4024 "memory", 4025 "press", 4026 "maxhdr", 4027 "slab", 4028 "module", 4029 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n"); 4030 else 4031 proto_seq_printf(seq, list_entry(v, struct proto, node)); 4032 return 0; 4033 } 4034 4035 static const struct seq_operations proto_seq_ops = { 4036 .start = proto_seq_start, 4037 .next = proto_seq_next, 4038 .stop = proto_seq_stop, 4039 .show = proto_seq_show, 4040 }; 4041 4042 static __net_init int proto_init_net(struct net *net) 4043 { 4044 if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops, 4045 sizeof(struct seq_net_private))) 4046 return -ENOMEM; 4047 4048 return 0; 4049 } 4050 4051 static __net_exit void proto_exit_net(struct net *net) 4052 { 4053 remove_proc_entry("protocols", net->proc_net); 4054 } 4055 4056 4057 static __net_initdata struct pernet_operations proto_net_ops = { 4058 .init = proto_init_net, 4059 .exit = proto_exit_net, 4060 }; 4061 4062 static int __init proto_init(void) 4063 { 4064 return register_pernet_subsys(&proto_net_ops); 4065 } 4066 4067 subsys_initcall(proto_init); 4068 4069 #endif /* PROC_FS */ 4070 4071 #ifdef CONFIG_NET_RX_BUSY_POLL 4072 bool sk_busy_loop_end(void *p, unsigned long start_time) 4073 { 4074 struct sock *sk = p; 4075 4076 return !skb_queue_empty_lockless(&sk->sk_receive_queue) || 4077 sk_busy_loop_timeout(sk, start_time); 4078 } 4079 EXPORT_SYMBOL(sk_busy_loop_end); 4080 #endif /* CONFIG_NET_RX_BUSY_POLL */ 4081 4082 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len) 4083 { 4084 if (!sk->sk_prot->bind_add) 4085 return -EOPNOTSUPP; 4086 return sk->sk_prot->bind_add(sk, addr, addr_len); 4087 } 4088 EXPORT_SYMBOL(sock_bind_add); 4089