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 reuseport_update_incoming_cpu(sk, 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 } else { 2098 __netns_tracker_alloc(net, &sk->ns_tracker, 2099 false, priority); 2100 } 2101 2102 sock_net_set(sk, net); 2103 refcount_set(&sk->sk_wmem_alloc, 1); 2104 2105 mem_cgroup_sk_alloc(sk); 2106 cgroup_sk_alloc(&sk->sk_cgrp_data); 2107 sock_update_classid(&sk->sk_cgrp_data); 2108 sock_update_netprioidx(&sk->sk_cgrp_data); 2109 sk_tx_queue_clear(sk); 2110 } 2111 2112 return sk; 2113 } 2114 EXPORT_SYMBOL(sk_alloc); 2115 2116 /* Sockets having SOCK_RCU_FREE will call this function after one RCU 2117 * grace period. This is the case for UDP sockets and TCP listeners. 2118 */ 2119 static void __sk_destruct(struct rcu_head *head) 2120 { 2121 struct sock *sk = container_of(head, struct sock, sk_rcu); 2122 struct sk_filter *filter; 2123 2124 if (sk->sk_destruct) 2125 sk->sk_destruct(sk); 2126 2127 filter = rcu_dereference_check(sk->sk_filter, 2128 refcount_read(&sk->sk_wmem_alloc) == 0); 2129 if (filter) { 2130 sk_filter_uncharge(sk, filter); 2131 RCU_INIT_POINTER(sk->sk_filter, NULL); 2132 } 2133 2134 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP); 2135 2136 #ifdef CONFIG_BPF_SYSCALL 2137 bpf_sk_storage_free(sk); 2138 #endif 2139 2140 if (atomic_read(&sk->sk_omem_alloc)) 2141 pr_debug("%s: optmem leakage (%d bytes) detected\n", 2142 __func__, atomic_read(&sk->sk_omem_alloc)); 2143 2144 if (sk->sk_frag.page) { 2145 put_page(sk->sk_frag.page); 2146 sk->sk_frag.page = NULL; 2147 } 2148 2149 /* We do not need to acquire sk->sk_peer_lock, we are the last user. */ 2150 put_cred(sk->sk_peer_cred); 2151 put_pid(sk->sk_peer_pid); 2152 2153 if (likely(sk->sk_net_refcnt)) 2154 put_net_track(sock_net(sk), &sk->ns_tracker); 2155 else 2156 __netns_tracker_free(sock_net(sk), &sk->ns_tracker, false); 2157 2158 sk_prot_free(sk->sk_prot_creator, sk); 2159 } 2160 2161 void sk_destruct(struct sock *sk) 2162 { 2163 bool use_call_rcu = sock_flag(sk, SOCK_RCU_FREE); 2164 2165 if (rcu_access_pointer(sk->sk_reuseport_cb)) { 2166 reuseport_detach_sock(sk); 2167 use_call_rcu = true; 2168 } 2169 2170 if (use_call_rcu) 2171 call_rcu(&sk->sk_rcu, __sk_destruct); 2172 else 2173 __sk_destruct(&sk->sk_rcu); 2174 } 2175 2176 static void __sk_free(struct sock *sk) 2177 { 2178 if (likely(sk->sk_net_refcnt)) 2179 sock_inuse_add(sock_net(sk), -1); 2180 2181 if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk))) 2182 sock_diag_broadcast_destroy(sk); 2183 else 2184 sk_destruct(sk); 2185 } 2186 2187 void sk_free(struct sock *sk) 2188 { 2189 /* 2190 * We subtract one from sk_wmem_alloc and can know if 2191 * some packets are still in some tx queue. 2192 * If not null, sock_wfree() will call __sk_free(sk) later 2193 */ 2194 if (refcount_dec_and_test(&sk->sk_wmem_alloc)) 2195 __sk_free(sk); 2196 } 2197 EXPORT_SYMBOL(sk_free); 2198 2199 static void sk_init_common(struct sock *sk) 2200 { 2201 skb_queue_head_init(&sk->sk_receive_queue); 2202 skb_queue_head_init(&sk->sk_write_queue); 2203 skb_queue_head_init(&sk->sk_error_queue); 2204 2205 rwlock_init(&sk->sk_callback_lock); 2206 lockdep_set_class_and_name(&sk->sk_receive_queue.lock, 2207 af_rlock_keys + sk->sk_family, 2208 af_family_rlock_key_strings[sk->sk_family]); 2209 lockdep_set_class_and_name(&sk->sk_write_queue.lock, 2210 af_wlock_keys + sk->sk_family, 2211 af_family_wlock_key_strings[sk->sk_family]); 2212 lockdep_set_class_and_name(&sk->sk_error_queue.lock, 2213 af_elock_keys + sk->sk_family, 2214 af_family_elock_key_strings[sk->sk_family]); 2215 lockdep_set_class_and_name(&sk->sk_callback_lock, 2216 af_callback_keys + sk->sk_family, 2217 af_family_clock_key_strings[sk->sk_family]); 2218 } 2219 2220 /** 2221 * sk_clone_lock - clone a socket, and lock its clone 2222 * @sk: the socket to clone 2223 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 2224 * 2225 * Caller must unlock socket even in error path (bh_unlock_sock(newsk)) 2226 */ 2227 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority) 2228 { 2229 struct proto *prot = READ_ONCE(sk->sk_prot); 2230 struct sk_filter *filter; 2231 bool is_charged = true; 2232 struct sock *newsk; 2233 2234 newsk = sk_prot_alloc(prot, priority, sk->sk_family); 2235 if (!newsk) 2236 goto out; 2237 2238 sock_copy(newsk, sk); 2239 2240 newsk->sk_prot_creator = prot; 2241 2242 /* SANITY */ 2243 if (likely(newsk->sk_net_refcnt)) { 2244 get_net_track(sock_net(newsk), &newsk->ns_tracker, priority); 2245 sock_inuse_add(sock_net(newsk), 1); 2246 } else { 2247 /* Kernel sockets are not elevating the struct net refcount. 2248 * Instead, use a tracker to more easily detect if a layer 2249 * is not properly dismantling its kernel sockets at netns 2250 * destroy time. 2251 */ 2252 __netns_tracker_alloc(sock_net(newsk), &newsk->ns_tracker, 2253 false, priority); 2254 } 2255 sk_node_init(&newsk->sk_node); 2256 sock_lock_init(newsk); 2257 bh_lock_sock(newsk); 2258 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL; 2259 newsk->sk_backlog.len = 0; 2260 2261 atomic_set(&newsk->sk_rmem_alloc, 0); 2262 2263 /* sk_wmem_alloc set to one (see sk_free() and sock_wfree()) */ 2264 refcount_set(&newsk->sk_wmem_alloc, 1); 2265 2266 atomic_set(&newsk->sk_omem_alloc, 0); 2267 sk_init_common(newsk); 2268 2269 newsk->sk_dst_cache = NULL; 2270 newsk->sk_dst_pending_confirm = 0; 2271 newsk->sk_wmem_queued = 0; 2272 newsk->sk_forward_alloc = 0; 2273 newsk->sk_reserved_mem = 0; 2274 atomic_set(&newsk->sk_drops, 0); 2275 newsk->sk_send_head = NULL; 2276 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK; 2277 atomic_set(&newsk->sk_zckey, 0); 2278 2279 sock_reset_flag(newsk, SOCK_DONE); 2280 2281 /* sk->sk_memcg will be populated at accept() time */ 2282 newsk->sk_memcg = NULL; 2283 2284 cgroup_sk_clone(&newsk->sk_cgrp_data); 2285 2286 rcu_read_lock(); 2287 filter = rcu_dereference(sk->sk_filter); 2288 if (filter != NULL) 2289 /* though it's an empty new sock, the charging may fail 2290 * if sysctl_optmem_max was changed between creation of 2291 * original socket and cloning 2292 */ 2293 is_charged = sk_filter_charge(newsk, filter); 2294 RCU_INIT_POINTER(newsk->sk_filter, filter); 2295 rcu_read_unlock(); 2296 2297 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) { 2298 /* We need to make sure that we don't uncharge the new 2299 * socket if we couldn't charge it in the first place 2300 * as otherwise we uncharge the parent's filter. 2301 */ 2302 if (!is_charged) 2303 RCU_INIT_POINTER(newsk->sk_filter, NULL); 2304 sk_free_unlock_clone(newsk); 2305 newsk = NULL; 2306 goto out; 2307 } 2308 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL); 2309 2310 if (bpf_sk_storage_clone(sk, newsk)) { 2311 sk_free_unlock_clone(newsk); 2312 newsk = NULL; 2313 goto out; 2314 } 2315 2316 /* Clear sk_user_data if parent had the pointer tagged 2317 * as not suitable for copying when cloning. 2318 */ 2319 if (sk_user_data_is_nocopy(newsk)) 2320 newsk->sk_user_data = NULL; 2321 2322 newsk->sk_err = 0; 2323 newsk->sk_err_soft = 0; 2324 newsk->sk_priority = 0; 2325 newsk->sk_incoming_cpu = raw_smp_processor_id(); 2326 2327 /* Before updating sk_refcnt, we must commit prior changes to memory 2328 * (Documentation/RCU/rculist_nulls.rst for details) 2329 */ 2330 smp_wmb(); 2331 refcount_set(&newsk->sk_refcnt, 2); 2332 2333 /* Increment the counter in the same struct proto as the master 2334 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that 2335 * is the same as sk->sk_prot->socks, as this field was copied 2336 * with memcpy). 2337 * 2338 * This _changes_ the previous behaviour, where 2339 * tcp_create_openreq_child always was incrementing the 2340 * equivalent to tcp_prot->socks (inet_sock_nr), so this have 2341 * to be taken into account in all callers. -acme 2342 */ 2343 sk_refcnt_debug_inc(newsk); 2344 sk_set_socket(newsk, NULL); 2345 sk_tx_queue_clear(newsk); 2346 RCU_INIT_POINTER(newsk->sk_wq, NULL); 2347 2348 if (newsk->sk_prot->sockets_allocated) 2349 sk_sockets_allocated_inc(newsk); 2350 2351 if (sock_needs_netstamp(sk) && newsk->sk_flags & SK_FLAGS_TIMESTAMP) 2352 net_enable_timestamp(); 2353 out: 2354 return newsk; 2355 } 2356 EXPORT_SYMBOL_GPL(sk_clone_lock); 2357 2358 void sk_free_unlock_clone(struct sock *sk) 2359 { 2360 /* It is still raw copy of parent, so invalidate 2361 * destructor and make plain sk_free() */ 2362 sk->sk_destruct = NULL; 2363 bh_unlock_sock(sk); 2364 sk_free(sk); 2365 } 2366 EXPORT_SYMBOL_GPL(sk_free_unlock_clone); 2367 2368 static void sk_trim_gso_size(struct sock *sk) 2369 { 2370 if (sk->sk_gso_max_size <= GSO_LEGACY_MAX_SIZE) 2371 return; 2372 #if IS_ENABLED(CONFIG_IPV6) 2373 if (sk->sk_family == AF_INET6 && 2374 sk_is_tcp(sk) && 2375 !ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr)) 2376 return; 2377 #endif 2378 sk->sk_gso_max_size = GSO_LEGACY_MAX_SIZE; 2379 } 2380 2381 void sk_setup_caps(struct sock *sk, struct dst_entry *dst) 2382 { 2383 u32 max_segs = 1; 2384 2385 sk_dst_set(sk, dst); 2386 sk->sk_route_caps = dst->dev->features; 2387 if (sk_is_tcp(sk)) 2388 sk->sk_route_caps |= NETIF_F_GSO; 2389 if (sk->sk_route_caps & NETIF_F_GSO) 2390 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE; 2391 if (unlikely(sk->sk_gso_disabled)) 2392 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 2393 if (sk_can_gso(sk)) { 2394 if (dst->header_len && !xfrm_dst_offload_ok(dst)) { 2395 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 2396 } else { 2397 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM; 2398 /* pairs with the WRITE_ONCE() in netif_set_gso_max_size() */ 2399 sk->sk_gso_max_size = READ_ONCE(dst->dev->gso_max_size); 2400 sk_trim_gso_size(sk); 2401 sk->sk_gso_max_size -= (MAX_TCP_HEADER + 1); 2402 /* pairs with the WRITE_ONCE() in netif_set_gso_max_segs() */ 2403 max_segs = max_t(u32, READ_ONCE(dst->dev->gso_max_segs), 1); 2404 } 2405 } 2406 sk->sk_gso_max_segs = max_segs; 2407 } 2408 EXPORT_SYMBOL_GPL(sk_setup_caps); 2409 2410 /* 2411 * Simple resource managers for sockets. 2412 */ 2413 2414 2415 /* 2416 * Write buffer destructor automatically called from kfree_skb. 2417 */ 2418 void sock_wfree(struct sk_buff *skb) 2419 { 2420 struct sock *sk = skb->sk; 2421 unsigned int len = skb->truesize; 2422 bool free; 2423 2424 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) { 2425 if (sock_flag(sk, SOCK_RCU_FREE) && 2426 sk->sk_write_space == sock_def_write_space) { 2427 rcu_read_lock(); 2428 free = refcount_sub_and_test(len, &sk->sk_wmem_alloc); 2429 sock_def_write_space_wfree(sk); 2430 rcu_read_unlock(); 2431 if (unlikely(free)) 2432 __sk_free(sk); 2433 return; 2434 } 2435 2436 /* 2437 * Keep a reference on sk_wmem_alloc, this will be released 2438 * after sk_write_space() call 2439 */ 2440 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc)); 2441 sk->sk_write_space(sk); 2442 len = 1; 2443 } 2444 /* 2445 * if sk_wmem_alloc reaches 0, we must finish what sk_free() 2446 * could not do because of in-flight packets 2447 */ 2448 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc)) 2449 __sk_free(sk); 2450 } 2451 EXPORT_SYMBOL(sock_wfree); 2452 2453 /* This variant of sock_wfree() is used by TCP, 2454 * since it sets SOCK_USE_WRITE_QUEUE. 2455 */ 2456 void __sock_wfree(struct sk_buff *skb) 2457 { 2458 struct sock *sk = skb->sk; 2459 2460 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc)) 2461 __sk_free(sk); 2462 } 2463 2464 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk) 2465 { 2466 skb_orphan(skb); 2467 skb->sk = sk; 2468 #ifdef CONFIG_INET 2469 if (unlikely(!sk_fullsock(sk))) { 2470 skb->destructor = sock_edemux; 2471 sock_hold(sk); 2472 return; 2473 } 2474 #endif 2475 skb->destructor = sock_wfree; 2476 skb_set_hash_from_sk(skb, sk); 2477 /* 2478 * We used to take a refcount on sk, but following operation 2479 * is enough to guarantee sk_free() wont free this sock until 2480 * all in-flight packets are completed 2481 */ 2482 refcount_add(skb->truesize, &sk->sk_wmem_alloc); 2483 } 2484 EXPORT_SYMBOL(skb_set_owner_w); 2485 2486 static bool can_skb_orphan_partial(const struct sk_buff *skb) 2487 { 2488 #ifdef CONFIG_TLS_DEVICE 2489 /* Drivers depend on in-order delivery for crypto offload, 2490 * partial orphan breaks out-of-order-OK logic. 2491 */ 2492 if (skb->decrypted) 2493 return false; 2494 #endif 2495 return (skb->destructor == sock_wfree || 2496 (IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree)); 2497 } 2498 2499 /* This helper is used by netem, as it can hold packets in its 2500 * delay queue. We want to allow the owner socket to send more 2501 * packets, as if they were already TX completed by a typical driver. 2502 * But we also want to keep skb->sk set because some packet schedulers 2503 * rely on it (sch_fq for example). 2504 */ 2505 void skb_orphan_partial(struct sk_buff *skb) 2506 { 2507 if (skb_is_tcp_pure_ack(skb)) 2508 return; 2509 2510 if (can_skb_orphan_partial(skb) && skb_set_owner_sk_safe(skb, skb->sk)) 2511 return; 2512 2513 skb_orphan(skb); 2514 } 2515 EXPORT_SYMBOL(skb_orphan_partial); 2516 2517 /* 2518 * Read buffer destructor automatically called from kfree_skb. 2519 */ 2520 void sock_rfree(struct sk_buff *skb) 2521 { 2522 struct sock *sk = skb->sk; 2523 unsigned int len = skb->truesize; 2524 2525 atomic_sub(len, &sk->sk_rmem_alloc); 2526 sk_mem_uncharge(sk, len); 2527 } 2528 EXPORT_SYMBOL(sock_rfree); 2529 2530 /* 2531 * Buffer destructor for skbs that are not used directly in read or write 2532 * path, e.g. for error handler skbs. Automatically called from kfree_skb. 2533 */ 2534 void sock_efree(struct sk_buff *skb) 2535 { 2536 sock_put(skb->sk); 2537 } 2538 EXPORT_SYMBOL(sock_efree); 2539 2540 /* Buffer destructor for prefetch/receive path where reference count may 2541 * not be held, e.g. for listen sockets. 2542 */ 2543 #ifdef CONFIG_INET 2544 void sock_pfree(struct sk_buff *skb) 2545 { 2546 if (sk_is_refcounted(skb->sk)) 2547 sock_gen_put(skb->sk); 2548 } 2549 EXPORT_SYMBOL(sock_pfree); 2550 #endif /* CONFIG_INET */ 2551 2552 kuid_t sock_i_uid(struct sock *sk) 2553 { 2554 kuid_t uid; 2555 2556 read_lock_bh(&sk->sk_callback_lock); 2557 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID; 2558 read_unlock_bh(&sk->sk_callback_lock); 2559 return uid; 2560 } 2561 EXPORT_SYMBOL(sock_i_uid); 2562 2563 unsigned long sock_i_ino(struct sock *sk) 2564 { 2565 unsigned long ino; 2566 2567 read_lock_bh(&sk->sk_callback_lock); 2568 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0; 2569 read_unlock_bh(&sk->sk_callback_lock); 2570 return ino; 2571 } 2572 EXPORT_SYMBOL(sock_i_ino); 2573 2574 /* 2575 * Allocate a skb from the socket's send buffer. 2576 */ 2577 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 2578 gfp_t priority) 2579 { 2580 if (force || 2581 refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) { 2582 struct sk_buff *skb = alloc_skb(size, priority); 2583 2584 if (skb) { 2585 skb_set_owner_w(skb, sk); 2586 return skb; 2587 } 2588 } 2589 return NULL; 2590 } 2591 EXPORT_SYMBOL(sock_wmalloc); 2592 2593 static void sock_ofree(struct sk_buff *skb) 2594 { 2595 struct sock *sk = skb->sk; 2596 2597 atomic_sub(skb->truesize, &sk->sk_omem_alloc); 2598 } 2599 2600 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size, 2601 gfp_t priority) 2602 { 2603 struct sk_buff *skb; 2604 2605 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */ 2606 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) > 2607 READ_ONCE(sysctl_optmem_max)) 2608 return NULL; 2609 2610 skb = alloc_skb(size, priority); 2611 if (!skb) 2612 return NULL; 2613 2614 atomic_add(skb->truesize, &sk->sk_omem_alloc); 2615 skb->sk = sk; 2616 skb->destructor = sock_ofree; 2617 return skb; 2618 } 2619 2620 /* 2621 * Allocate a memory block from the socket's option memory buffer. 2622 */ 2623 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority) 2624 { 2625 int optmem_max = READ_ONCE(sysctl_optmem_max); 2626 2627 if ((unsigned int)size <= optmem_max && 2628 atomic_read(&sk->sk_omem_alloc) + size < optmem_max) { 2629 void *mem; 2630 /* First do the add, to avoid the race if kmalloc 2631 * might sleep. 2632 */ 2633 atomic_add(size, &sk->sk_omem_alloc); 2634 mem = kmalloc(size, priority); 2635 if (mem) 2636 return mem; 2637 atomic_sub(size, &sk->sk_omem_alloc); 2638 } 2639 return NULL; 2640 } 2641 EXPORT_SYMBOL(sock_kmalloc); 2642 2643 /* Free an option memory block. Note, we actually want the inline 2644 * here as this allows gcc to detect the nullify and fold away the 2645 * condition entirely. 2646 */ 2647 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size, 2648 const bool nullify) 2649 { 2650 if (WARN_ON_ONCE(!mem)) 2651 return; 2652 if (nullify) 2653 kfree_sensitive(mem); 2654 else 2655 kfree(mem); 2656 atomic_sub(size, &sk->sk_omem_alloc); 2657 } 2658 2659 void sock_kfree_s(struct sock *sk, void *mem, int size) 2660 { 2661 __sock_kfree_s(sk, mem, size, false); 2662 } 2663 EXPORT_SYMBOL(sock_kfree_s); 2664 2665 void sock_kzfree_s(struct sock *sk, void *mem, int size) 2666 { 2667 __sock_kfree_s(sk, mem, size, true); 2668 } 2669 EXPORT_SYMBOL(sock_kzfree_s); 2670 2671 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock. 2672 I think, these locks should be removed for datagram sockets. 2673 */ 2674 static long sock_wait_for_wmem(struct sock *sk, long timeo) 2675 { 2676 DEFINE_WAIT(wait); 2677 2678 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk); 2679 for (;;) { 2680 if (!timeo) 2681 break; 2682 if (signal_pending(current)) 2683 break; 2684 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 2685 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 2686 if (refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) 2687 break; 2688 if (sk->sk_shutdown & SEND_SHUTDOWN) 2689 break; 2690 if (sk->sk_err) 2691 break; 2692 timeo = schedule_timeout(timeo); 2693 } 2694 finish_wait(sk_sleep(sk), &wait); 2695 return timeo; 2696 } 2697 2698 2699 /* 2700 * Generic send/receive buffer handlers 2701 */ 2702 2703 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 2704 unsigned long data_len, int noblock, 2705 int *errcode, int max_page_order) 2706 { 2707 struct sk_buff *skb; 2708 long timeo; 2709 int err; 2710 2711 timeo = sock_sndtimeo(sk, noblock); 2712 for (;;) { 2713 err = sock_error(sk); 2714 if (err != 0) 2715 goto failure; 2716 2717 err = -EPIPE; 2718 if (sk->sk_shutdown & SEND_SHUTDOWN) 2719 goto failure; 2720 2721 if (sk_wmem_alloc_get(sk) < READ_ONCE(sk->sk_sndbuf)) 2722 break; 2723 2724 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk); 2725 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 2726 err = -EAGAIN; 2727 if (!timeo) 2728 goto failure; 2729 if (signal_pending(current)) 2730 goto interrupted; 2731 timeo = sock_wait_for_wmem(sk, timeo); 2732 } 2733 skb = alloc_skb_with_frags(header_len, data_len, max_page_order, 2734 errcode, sk->sk_allocation); 2735 if (skb) 2736 skb_set_owner_w(skb, sk); 2737 return skb; 2738 2739 interrupted: 2740 err = sock_intr_errno(timeo); 2741 failure: 2742 *errcode = err; 2743 return NULL; 2744 } 2745 EXPORT_SYMBOL(sock_alloc_send_pskb); 2746 2747 int __sock_cmsg_send(struct sock *sk, struct cmsghdr *cmsg, 2748 struct sockcm_cookie *sockc) 2749 { 2750 u32 tsflags; 2751 2752 switch (cmsg->cmsg_type) { 2753 case SO_MARK: 2754 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) && 2755 !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) 2756 return -EPERM; 2757 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 2758 return -EINVAL; 2759 sockc->mark = *(u32 *)CMSG_DATA(cmsg); 2760 break; 2761 case SO_TIMESTAMPING_OLD: 2762 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 2763 return -EINVAL; 2764 2765 tsflags = *(u32 *)CMSG_DATA(cmsg); 2766 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK) 2767 return -EINVAL; 2768 2769 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK; 2770 sockc->tsflags |= tsflags; 2771 break; 2772 case SCM_TXTIME: 2773 if (!sock_flag(sk, SOCK_TXTIME)) 2774 return -EINVAL; 2775 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64))) 2776 return -EINVAL; 2777 sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg)); 2778 break; 2779 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */ 2780 case SCM_RIGHTS: 2781 case SCM_CREDENTIALS: 2782 break; 2783 default: 2784 return -EINVAL; 2785 } 2786 return 0; 2787 } 2788 EXPORT_SYMBOL(__sock_cmsg_send); 2789 2790 int sock_cmsg_send(struct sock *sk, struct msghdr *msg, 2791 struct sockcm_cookie *sockc) 2792 { 2793 struct cmsghdr *cmsg; 2794 int ret; 2795 2796 for_each_cmsghdr(cmsg, msg) { 2797 if (!CMSG_OK(msg, cmsg)) 2798 return -EINVAL; 2799 if (cmsg->cmsg_level != SOL_SOCKET) 2800 continue; 2801 ret = __sock_cmsg_send(sk, cmsg, sockc); 2802 if (ret) 2803 return ret; 2804 } 2805 return 0; 2806 } 2807 EXPORT_SYMBOL(sock_cmsg_send); 2808 2809 static void sk_enter_memory_pressure(struct sock *sk) 2810 { 2811 if (!sk->sk_prot->enter_memory_pressure) 2812 return; 2813 2814 sk->sk_prot->enter_memory_pressure(sk); 2815 } 2816 2817 static void sk_leave_memory_pressure(struct sock *sk) 2818 { 2819 if (sk->sk_prot->leave_memory_pressure) { 2820 sk->sk_prot->leave_memory_pressure(sk); 2821 } else { 2822 unsigned long *memory_pressure = sk->sk_prot->memory_pressure; 2823 2824 if (memory_pressure && READ_ONCE(*memory_pressure)) 2825 WRITE_ONCE(*memory_pressure, 0); 2826 } 2827 } 2828 2829 DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key); 2830 2831 /** 2832 * skb_page_frag_refill - check that a page_frag contains enough room 2833 * @sz: minimum size of the fragment we want to get 2834 * @pfrag: pointer to page_frag 2835 * @gfp: priority for memory allocation 2836 * 2837 * Note: While this allocator tries to use high order pages, there is 2838 * no guarantee that allocations succeed. Therefore, @sz MUST be 2839 * less or equal than PAGE_SIZE. 2840 */ 2841 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp) 2842 { 2843 if (pfrag->page) { 2844 if (page_ref_count(pfrag->page) == 1) { 2845 pfrag->offset = 0; 2846 return true; 2847 } 2848 if (pfrag->offset + sz <= pfrag->size) 2849 return true; 2850 put_page(pfrag->page); 2851 } 2852 2853 pfrag->offset = 0; 2854 if (SKB_FRAG_PAGE_ORDER && 2855 !static_branch_unlikely(&net_high_order_alloc_disable_key)) { 2856 /* Avoid direct reclaim but allow kswapd to wake */ 2857 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) | 2858 __GFP_COMP | __GFP_NOWARN | 2859 __GFP_NORETRY, 2860 SKB_FRAG_PAGE_ORDER); 2861 if (likely(pfrag->page)) { 2862 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER; 2863 return true; 2864 } 2865 } 2866 pfrag->page = alloc_page(gfp); 2867 if (likely(pfrag->page)) { 2868 pfrag->size = PAGE_SIZE; 2869 return true; 2870 } 2871 return false; 2872 } 2873 EXPORT_SYMBOL(skb_page_frag_refill); 2874 2875 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag) 2876 { 2877 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation))) 2878 return true; 2879 2880 sk_enter_memory_pressure(sk); 2881 sk_stream_moderate_sndbuf(sk); 2882 return false; 2883 } 2884 EXPORT_SYMBOL(sk_page_frag_refill); 2885 2886 void __lock_sock(struct sock *sk) 2887 __releases(&sk->sk_lock.slock) 2888 __acquires(&sk->sk_lock.slock) 2889 { 2890 DEFINE_WAIT(wait); 2891 2892 for (;;) { 2893 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait, 2894 TASK_UNINTERRUPTIBLE); 2895 spin_unlock_bh(&sk->sk_lock.slock); 2896 schedule(); 2897 spin_lock_bh(&sk->sk_lock.slock); 2898 if (!sock_owned_by_user(sk)) 2899 break; 2900 } 2901 finish_wait(&sk->sk_lock.wq, &wait); 2902 } 2903 2904 void __release_sock(struct sock *sk) 2905 __releases(&sk->sk_lock.slock) 2906 __acquires(&sk->sk_lock.slock) 2907 { 2908 struct sk_buff *skb, *next; 2909 2910 while ((skb = sk->sk_backlog.head) != NULL) { 2911 sk->sk_backlog.head = sk->sk_backlog.tail = NULL; 2912 2913 spin_unlock_bh(&sk->sk_lock.slock); 2914 2915 do { 2916 next = skb->next; 2917 prefetch(next); 2918 DEBUG_NET_WARN_ON_ONCE(skb_dst_is_noref(skb)); 2919 skb_mark_not_on_list(skb); 2920 sk_backlog_rcv(sk, skb); 2921 2922 cond_resched(); 2923 2924 skb = next; 2925 } while (skb != NULL); 2926 2927 spin_lock_bh(&sk->sk_lock.slock); 2928 } 2929 2930 /* 2931 * Doing the zeroing here guarantee we can not loop forever 2932 * while a wild producer attempts to flood us. 2933 */ 2934 sk->sk_backlog.len = 0; 2935 } 2936 2937 void __sk_flush_backlog(struct sock *sk) 2938 { 2939 spin_lock_bh(&sk->sk_lock.slock); 2940 __release_sock(sk); 2941 spin_unlock_bh(&sk->sk_lock.slock); 2942 } 2943 EXPORT_SYMBOL_GPL(__sk_flush_backlog); 2944 2945 /** 2946 * sk_wait_data - wait for data to arrive at sk_receive_queue 2947 * @sk: sock to wait on 2948 * @timeo: for how long 2949 * @skb: last skb seen on sk_receive_queue 2950 * 2951 * Now socket state including sk->sk_err is changed only under lock, 2952 * hence we may omit checks after joining wait queue. 2953 * We check receive queue before schedule() only as optimization; 2954 * it is very likely that release_sock() added new data. 2955 */ 2956 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb) 2957 { 2958 DEFINE_WAIT_FUNC(wait, woken_wake_function); 2959 int rc; 2960 2961 add_wait_queue(sk_sleep(sk), &wait); 2962 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk); 2963 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait); 2964 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk); 2965 remove_wait_queue(sk_sleep(sk), &wait); 2966 return rc; 2967 } 2968 EXPORT_SYMBOL(sk_wait_data); 2969 2970 /** 2971 * __sk_mem_raise_allocated - increase memory_allocated 2972 * @sk: socket 2973 * @size: memory size to allocate 2974 * @amt: pages to allocate 2975 * @kind: allocation type 2976 * 2977 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc 2978 */ 2979 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind) 2980 { 2981 bool memcg_charge = mem_cgroup_sockets_enabled && sk->sk_memcg; 2982 struct proto *prot = sk->sk_prot; 2983 bool charged = true; 2984 long allocated; 2985 2986 sk_memory_allocated_add(sk, amt); 2987 allocated = sk_memory_allocated(sk); 2988 if (memcg_charge && 2989 !(charged = mem_cgroup_charge_skmem(sk->sk_memcg, amt, 2990 gfp_memcg_charge()))) 2991 goto suppress_allocation; 2992 2993 /* Under limit. */ 2994 if (allocated <= sk_prot_mem_limits(sk, 0)) { 2995 sk_leave_memory_pressure(sk); 2996 return 1; 2997 } 2998 2999 /* Under pressure. */ 3000 if (allocated > sk_prot_mem_limits(sk, 1)) 3001 sk_enter_memory_pressure(sk); 3002 3003 /* Over hard limit. */ 3004 if (allocated > sk_prot_mem_limits(sk, 2)) 3005 goto suppress_allocation; 3006 3007 /* guarantee minimum buffer size under pressure */ 3008 if (kind == SK_MEM_RECV) { 3009 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot)) 3010 return 1; 3011 3012 } else { /* SK_MEM_SEND */ 3013 int wmem0 = sk_get_wmem0(sk, prot); 3014 3015 if (sk->sk_type == SOCK_STREAM) { 3016 if (sk->sk_wmem_queued < wmem0) 3017 return 1; 3018 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) { 3019 return 1; 3020 } 3021 } 3022 3023 if (sk_has_memory_pressure(sk)) { 3024 u64 alloc; 3025 3026 if (!sk_under_memory_pressure(sk)) 3027 return 1; 3028 alloc = sk_sockets_allocated_read_positive(sk); 3029 if (sk_prot_mem_limits(sk, 2) > alloc * 3030 sk_mem_pages(sk->sk_wmem_queued + 3031 atomic_read(&sk->sk_rmem_alloc) + 3032 sk->sk_forward_alloc)) 3033 return 1; 3034 } 3035 3036 suppress_allocation: 3037 3038 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) { 3039 sk_stream_moderate_sndbuf(sk); 3040 3041 /* Fail only if socket is _under_ its sndbuf. 3042 * In this case we cannot block, so that we have to fail. 3043 */ 3044 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf) { 3045 /* Force charge with __GFP_NOFAIL */ 3046 if (memcg_charge && !charged) { 3047 mem_cgroup_charge_skmem(sk->sk_memcg, amt, 3048 gfp_memcg_charge() | __GFP_NOFAIL); 3049 } 3050 return 1; 3051 } 3052 } 3053 3054 if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged)) 3055 trace_sock_exceed_buf_limit(sk, prot, allocated, kind); 3056 3057 sk_memory_allocated_sub(sk, amt); 3058 3059 if (memcg_charge && charged) 3060 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt); 3061 3062 return 0; 3063 } 3064 3065 /** 3066 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated 3067 * @sk: socket 3068 * @size: memory size to allocate 3069 * @kind: allocation type 3070 * 3071 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means 3072 * rmem allocation. This function assumes that protocols which have 3073 * memory_pressure use sk_wmem_queued as write buffer accounting. 3074 */ 3075 int __sk_mem_schedule(struct sock *sk, int size, int kind) 3076 { 3077 int ret, amt = sk_mem_pages(size); 3078 3079 sk->sk_forward_alloc += amt << PAGE_SHIFT; 3080 ret = __sk_mem_raise_allocated(sk, size, amt, kind); 3081 if (!ret) 3082 sk->sk_forward_alloc -= amt << PAGE_SHIFT; 3083 return ret; 3084 } 3085 EXPORT_SYMBOL(__sk_mem_schedule); 3086 3087 /** 3088 * __sk_mem_reduce_allocated - reclaim memory_allocated 3089 * @sk: socket 3090 * @amount: number of quanta 3091 * 3092 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc 3093 */ 3094 void __sk_mem_reduce_allocated(struct sock *sk, int amount) 3095 { 3096 sk_memory_allocated_sub(sk, amount); 3097 3098 if (mem_cgroup_sockets_enabled && sk->sk_memcg) 3099 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount); 3100 3101 if (sk_under_memory_pressure(sk) && 3102 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0))) 3103 sk_leave_memory_pressure(sk); 3104 } 3105 3106 /** 3107 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated 3108 * @sk: socket 3109 * @amount: number of bytes (rounded down to a PAGE_SIZE multiple) 3110 */ 3111 void __sk_mem_reclaim(struct sock *sk, int amount) 3112 { 3113 amount >>= PAGE_SHIFT; 3114 sk->sk_forward_alloc -= amount << PAGE_SHIFT; 3115 __sk_mem_reduce_allocated(sk, amount); 3116 } 3117 EXPORT_SYMBOL(__sk_mem_reclaim); 3118 3119 int sk_set_peek_off(struct sock *sk, int val) 3120 { 3121 sk->sk_peek_off = val; 3122 return 0; 3123 } 3124 EXPORT_SYMBOL_GPL(sk_set_peek_off); 3125 3126 /* 3127 * Set of default routines for initialising struct proto_ops when 3128 * the protocol does not support a particular function. In certain 3129 * cases where it makes no sense for a protocol to have a "do nothing" 3130 * function, some default processing is provided. 3131 */ 3132 3133 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len) 3134 { 3135 return -EOPNOTSUPP; 3136 } 3137 EXPORT_SYMBOL(sock_no_bind); 3138 3139 int sock_no_connect(struct socket *sock, struct sockaddr *saddr, 3140 int len, int flags) 3141 { 3142 return -EOPNOTSUPP; 3143 } 3144 EXPORT_SYMBOL(sock_no_connect); 3145 3146 int sock_no_socketpair(struct socket *sock1, struct socket *sock2) 3147 { 3148 return -EOPNOTSUPP; 3149 } 3150 EXPORT_SYMBOL(sock_no_socketpair); 3151 3152 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags, 3153 bool kern) 3154 { 3155 return -EOPNOTSUPP; 3156 } 3157 EXPORT_SYMBOL(sock_no_accept); 3158 3159 int sock_no_getname(struct socket *sock, struct sockaddr *saddr, 3160 int peer) 3161 { 3162 return -EOPNOTSUPP; 3163 } 3164 EXPORT_SYMBOL(sock_no_getname); 3165 3166 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 3167 { 3168 return -EOPNOTSUPP; 3169 } 3170 EXPORT_SYMBOL(sock_no_ioctl); 3171 3172 int sock_no_listen(struct socket *sock, int backlog) 3173 { 3174 return -EOPNOTSUPP; 3175 } 3176 EXPORT_SYMBOL(sock_no_listen); 3177 3178 int sock_no_shutdown(struct socket *sock, int how) 3179 { 3180 return -EOPNOTSUPP; 3181 } 3182 EXPORT_SYMBOL(sock_no_shutdown); 3183 3184 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len) 3185 { 3186 return -EOPNOTSUPP; 3187 } 3188 EXPORT_SYMBOL(sock_no_sendmsg); 3189 3190 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len) 3191 { 3192 return -EOPNOTSUPP; 3193 } 3194 EXPORT_SYMBOL(sock_no_sendmsg_locked); 3195 3196 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len, 3197 int flags) 3198 { 3199 return -EOPNOTSUPP; 3200 } 3201 EXPORT_SYMBOL(sock_no_recvmsg); 3202 3203 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma) 3204 { 3205 /* Mirror missing mmap method error code */ 3206 return -ENODEV; 3207 } 3208 EXPORT_SYMBOL(sock_no_mmap); 3209 3210 /* 3211 * When a file is received (via SCM_RIGHTS, etc), we must bump the 3212 * various sock-based usage counts. 3213 */ 3214 void __receive_sock(struct file *file) 3215 { 3216 struct socket *sock; 3217 3218 sock = sock_from_file(file); 3219 if (sock) { 3220 sock_update_netprioidx(&sock->sk->sk_cgrp_data); 3221 sock_update_classid(&sock->sk->sk_cgrp_data); 3222 } 3223 } 3224 3225 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags) 3226 { 3227 ssize_t res; 3228 struct msghdr msg = {.msg_flags = flags}; 3229 struct kvec iov; 3230 char *kaddr = kmap(page); 3231 iov.iov_base = kaddr + offset; 3232 iov.iov_len = size; 3233 res = kernel_sendmsg(sock, &msg, &iov, 1, size); 3234 kunmap(page); 3235 return res; 3236 } 3237 EXPORT_SYMBOL(sock_no_sendpage); 3238 3239 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page, 3240 int offset, size_t size, int flags) 3241 { 3242 ssize_t res; 3243 struct msghdr msg = {.msg_flags = flags}; 3244 struct kvec iov; 3245 char *kaddr = kmap(page); 3246 3247 iov.iov_base = kaddr + offset; 3248 iov.iov_len = size; 3249 res = kernel_sendmsg_locked(sk, &msg, &iov, 1, size); 3250 kunmap(page); 3251 return res; 3252 } 3253 EXPORT_SYMBOL(sock_no_sendpage_locked); 3254 3255 /* 3256 * Default Socket Callbacks 3257 */ 3258 3259 static void sock_def_wakeup(struct sock *sk) 3260 { 3261 struct socket_wq *wq; 3262 3263 rcu_read_lock(); 3264 wq = rcu_dereference(sk->sk_wq); 3265 if (skwq_has_sleeper(wq)) 3266 wake_up_interruptible_all(&wq->wait); 3267 rcu_read_unlock(); 3268 } 3269 3270 static void sock_def_error_report(struct sock *sk) 3271 { 3272 struct socket_wq *wq; 3273 3274 rcu_read_lock(); 3275 wq = rcu_dereference(sk->sk_wq); 3276 if (skwq_has_sleeper(wq)) 3277 wake_up_interruptible_poll(&wq->wait, EPOLLERR); 3278 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR); 3279 rcu_read_unlock(); 3280 } 3281 3282 void sock_def_readable(struct sock *sk) 3283 { 3284 struct socket_wq *wq; 3285 3286 rcu_read_lock(); 3287 wq = rcu_dereference(sk->sk_wq); 3288 if (skwq_has_sleeper(wq)) 3289 wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI | 3290 EPOLLRDNORM | EPOLLRDBAND); 3291 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN); 3292 rcu_read_unlock(); 3293 } 3294 3295 static void sock_def_write_space(struct sock *sk) 3296 { 3297 struct socket_wq *wq; 3298 3299 rcu_read_lock(); 3300 3301 /* Do not wake up a writer until he can make "significant" 3302 * progress. --DaveM 3303 */ 3304 if (sock_writeable(sk)) { 3305 wq = rcu_dereference(sk->sk_wq); 3306 if (skwq_has_sleeper(wq)) 3307 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT | 3308 EPOLLWRNORM | EPOLLWRBAND); 3309 3310 /* Should agree with poll, otherwise some programs break */ 3311 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT); 3312 } 3313 3314 rcu_read_unlock(); 3315 } 3316 3317 /* An optimised version of sock_def_write_space(), should only be called 3318 * for SOCK_RCU_FREE sockets under RCU read section and after putting 3319 * ->sk_wmem_alloc. 3320 */ 3321 static void sock_def_write_space_wfree(struct sock *sk) 3322 { 3323 /* Do not wake up a writer until he can make "significant" 3324 * progress. --DaveM 3325 */ 3326 if (sock_writeable(sk)) { 3327 struct socket_wq *wq = rcu_dereference(sk->sk_wq); 3328 3329 /* rely on refcount_sub from sock_wfree() */ 3330 smp_mb__after_atomic(); 3331 if (wq && waitqueue_active(&wq->wait)) 3332 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT | 3333 EPOLLWRNORM | EPOLLWRBAND); 3334 3335 /* Should agree with poll, otherwise some programs break */ 3336 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT); 3337 } 3338 } 3339 3340 static void sock_def_destruct(struct sock *sk) 3341 { 3342 } 3343 3344 void sk_send_sigurg(struct sock *sk) 3345 { 3346 if (sk->sk_socket && sk->sk_socket->file) 3347 if (send_sigurg(&sk->sk_socket->file->f_owner)) 3348 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI); 3349 } 3350 EXPORT_SYMBOL(sk_send_sigurg); 3351 3352 void sk_reset_timer(struct sock *sk, struct timer_list* timer, 3353 unsigned long expires) 3354 { 3355 if (!mod_timer(timer, expires)) 3356 sock_hold(sk); 3357 } 3358 EXPORT_SYMBOL(sk_reset_timer); 3359 3360 void sk_stop_timer(struct sock *sk, struct timer_list* timer) 3361 { 3362 if (del_timer(timer)) 3363 __sock_put(sk); 3364 } 3365 EXPORT_SYMBOL(sk_stop_timer); 3366 3367 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer) 3368 { 3369 if (del_timer_sync(timer)) 3370 __sock_put(sk); 3371 } 3372 EXPORT_SYMBOL(sk_stop_timer_sync); 3373 3374 void sock_init_data(struct socket *sock, struct sock *sk) 3375 { 3376 sk_init_common(sk); 3377 sk->sk_send_head = NULL; 3378 3379 timer_setup(&sk->sk_timer, NULL, 0); 3380 3381 sk->sk_allocation = GFP_KERNEL; 3382 sk->sk_rcvbuf = READ_ONCE(sysctl_rmem_default); 3383 sk->sk_sndbuf = READ_ONCE(sysctl_wmem_default); 3384 sk->sk_state = TCP_CLOSE; 3385 sk_set_socket(sk, sock); 3386 3387 sock_set_flag(sk, SOCK_ZAPPED); 3388 3389 if (sock) { 3390 sk->sk_type = sock->type; 3391 RCU_INIT_POINTER(sk->sk_wq, &sock->wq); 3392 sock->sk = sk; 3393 sk->sk_uid = SOCK_INODE(sock)->i_uid; 3394 } else { 3395 RCU_INIT_POINTER(sk->sk_wq, NULL); 3396 sk->sk_uid = make_kuid(sock_net(sk)->user_ns, 0); 3397 } 3398 3399 rwlock_init(&sk->sk_callback_lock); 3400 if (sk->sk_kern_sock) 3401 lockdep_set_class_and_name( 3402 &sk->sk_callback_lock, 3403 af_kern_callback_keys + sk->sk_family, 3404 af_family_kern_clock_key_strings[sk->sk_family]); 3405 else 3406 lockdep_set_class_and_name( 3407 &sk->sk_callback_lock, 3408 af_callback_keys + sk->sk_family, 3409 af_family_clock_key_strings[sk->sk_family]); 3410 3411 sk->sk_state_change = sock_def_wakeup; 3412 sk->sk_data_ready = sock_def_readable; 3413 sk->sk_write_space = sock_def_write_space; 3414 sk->sk_error_report = sock_def_error_report; 3415 sk->sk_destruct = sock_def_destruct; 3416 3417 sk->sk_frag.page = NULL; 3418 sk->sk_frag.offset = 0; 3419 sk->sk_peek_off = -1; 3420 3421 sk->sk_peer_pid = NULL; 3422 sk->sk_peer_cred = NULL; 3423 spin_lock_init(&sk->sk_peer_lock); 3424 3425 sk->sk_write_pending = 0; 3426 sk->sk_rcvlowat = 1; 3427 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; 3428 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT; 3429 3430 sk->sk_stamp = SK_DEFAULT_STAMP; 3431 #if BITS_PER_LONG==32 3432 seqlock_init(&sk->sk_stamp_seq); 3433 #endif 3434 atomic_set(&sk->sk_zckey, 0); 3435 3436 #ifdef CONFIG_NET_RX_BUSY_POLL 3437 sk->sk_napi_id = 0; 3438 sk->sk_ll_usec = READ_ONCE(sysctl_net_busy_read); 3439 #endif 3440 3441 sk->sk_max_pacing_rate = ~0UL; 3442 sk->sk_pacing_rate = ~0UL; 3443 WRITE_ONCE(sk->sk_pacing_shift, 10); 3444 sk->sk_incoming_cpu = -1; 3445 sk->sk_txrehash = SOCK_TXREHASH_DEFAULT; 3446 3447 sk_rx_queue_clear(sk); 3448 /* 3449 * Before updating sk_refcnt, we must commit prior changes to memory 3450 * (Documentation/RCU/rculist_nulls.rst for details) 3451 */ 3452 smp_wmb(); 3453 refcount_set(&sk->sk_refcnt, 1); 3454 atomic_set(&sk->sk_drops, 0); 3455 } 3456 EXPORT_SYMBOL(sock_init_data); 3457 3458 void lock_sock_nested(struct sock *sk, int subclass) 3459 { 3460 /* The sk_lock has mutex_lock() semantics here. */ 3461 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_); 3462 3463 might_sleep(); 3464 spin_lock_bh(&sk->sk_lock.slock); 3465 if (sock_owned_by_user_nocheck(sk)) 3466 __lock_sock(sk); 3467 sk->sk_lock.owned = 1; 3468 spin_unlock_bh(&sk->sk_lock.slock); 3469 } 3470 EXPORT_SYMBOL(lock_sock_nested); 3471 3472 void release_sock(struct sock *sk) 3473 { 3474 spin_lock_bh(&sk->sk_lock.slock); 3475 if (sk->sk_backlog.tail) 3476 __release_sock(sk); 3477 3478 /* Warning : release_cb() might need to release sk ownership, 3479 * ie call sock_release_ownership(sk) before us. 3480 */ 3481 if (sk->sk_prot->release_cb) 3482 sk->sk_prot->release_cb(sk); 3483 3484 sock_release_ownership(sk); 3485 if (waitqueue_active(&sk->sk_lock.wq)) 3486 wake_up(&sk->sk_lock.wq); 3487 spin_unlock_bh(&sk->sk_lock.slock); 3488 } 3489 EXPORT_SYMBOL(release_sock); 3490 3491 bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock) 3492 { 3493 might_sleep(); 3494 spin_lock_bh(&sk->sk_lock.slock); 3495 3496 if (!sock_owned_by_user_nocheck(sk)) { 3497 /* 3498 * Fast path return with bottom halves disabled and 3499 * sock::sk_lock.slock held. 3500 * 3501 * The 'mutex' is not contended and holding 3502 * sock::sk_lock.slock prevents all other lockers to 3503 * proceed so the corresponding unlock_sock_fast() can 3504 * avoid the slow path of release_sock() completely and 3505 * just release slock. 3506 * 3507 * From a semantical POV this is equivalent to 'acquiring' 3508 * the 'mutex', hence the corresponding lockdep 3509 * mutex_release() has to happen in the fast path of 3510 * unlock_sock_fast(). 3511 */ 3512 return false; 3513 } 3514 3515 __lock_sock(sk); 3516 sk->sk_lock.owned = 1; 3517 __acquire(&sk->sk_lock.slock); 3518 spin_unlock_bh(&sk->sk_lock.slock); 3519 return true; 3520 } 3521 EXPORT_SYMBOL(__lock_sock_fast); 3522 3523 int sock_gettstamp(struct socket *sock, void __user *userstamp, 3524 bool timeval, bool time32) 3525 { 3526 struct sock *sk = sock->sk; 3527 struct timespec64 ts; 3528 3529 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 3530 ts = ktime_to_timespec64(sock_read_timestamp(sk)); 3531 if (ts.tv_sec == -1) 3532 return -ENOENT; 3533 if (ts.tv_sec == 0) { 3534 ktime_t kt = ktime_get_real(); 3535 sock_write_timestamp(sk, kt); 3536 ts = ktime_to_timespec64(kt); 3537 } 3538 3539 if (timeval) 3540 ts.tv_nsec /= 1000; 3541 3542 #ifdef CONFIG_COMPAT_32BIT_TIME 3543 if (time32) 3544 return put_old_timespec32(&ts, userstamp); 3545 #endif 3546 #ifdef CONFIG_SPARC64 3547 /* beware of padding in sparc64 timeval */ 3548 if (timeval && !in_compat_syscall()) { 3549 struct __kernel_old_timeval __user tv = { 3550 .tv_sec = ts.tv_sec, 3551 .tv_usec = ts.tv_nsec, 3552 }; 3553 if (copy_to_user(userstamp, &tv, sizeof(tv))) 3554 return -EFAULT; 3555 return 0; 3556 } 3557 #endif 3558 return put_timespec64(&ts, userstamp); 3559 } 3560 EXPORT_SYMBOL(sock_gettstamp); 3561 3562 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag) 3563 { 3564 if (!sock_flag(sk, flag)) { 3565 unsigned long previous_flags = sk->sk_flags; 3566 3567 sock_set_flag(sk, flag); 3568 /* 3569 * we just set one of the two flags which require net 3570 * time stamping, but time stamping might have been on 3571 * already because of the other one 3572 */ 3573 if (sock_needs_netstamp(sk) && 3574 !(previous_flags & SK_FLAGS_TIMESTAMP)) 3575 net_enable_timestamp(); 3576 } 3577 } 3578 3579 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, 3580 int level, int type) 3581 { 3582 struct sock_exterr_skb *serr; 3583 struct sk_buff *skb; 3584 int copied, err; 3585 3586 err = -EAGAIN; 3587 skb = sock_dequeue_err_skb(sk); 3588 if (skb == NULL) 3589 goto out; 3590 3591 copied = skb->len; 3592 if (copied > len) { 3593 msg->msg_flags |= MSG_TRUNC; 3594 copied = len; 3595 } 3596 err = skb_copy_datagram_msg(skb, 0, msg, copied); 3597 if (err) 3598 goto out_free_skb; 3599 3600 sock_recv_timestamp(msg, sk, skb); 3601 3602 serr = SKB_EXT_ERR(skb); 3603 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee); 3604 3605 msg->msg_flags |= MSG_ERRQUEUE; 3606 err = copied; 3607 3608 out_free_skb: 3609 kfree_skb(skb); 3610 out: 3611 return err; 3612 } 3613 EXPORT_SYMBOL(sock_recv_errqueue); 3614 3615 /* 3616 * Get a socket option on an socket. 3617 * 3618 * FIX: POSIX 1003.1g is very ambiguous here. It states that 3619 * asynchronous errors should be reported by getsockopt. We assume 3620 * this means if you specify SO_ERROR (otherwise whats the point of it). 3621 */ 3622 int sock_common_getsockopt(struct socket *sock, int level, int optname, 3623 char __user *optval, int __user *optlen) 3624 { 3625 struct sock *sk = sock->sk; 3626 3627 /* IPV6_ADDRFORM can change sk->sk_prot under us. */ 3628 return READ_ONCE(sk->sk_prot)->getsockopt(sk, level, optname, optval, optlen); 3629 } 3630 EXPORT_SYMBOL(sock_common_getsockopt); 3631 3632 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 3633 int flags) 3634 { 3635 struct sock *sk = sock->sk; 3636 int addr_len = 0; 3637 int err; 3638 3639 err = sk->sk_prot->recvmsg(sk, msg, size, flags, &addr_len); 3640 if (err >= 0) 3641 msg->msg_namelen = addr_len; 3642 return err; 3643 } 3644 EXPORT_SYMBOL(sock_common_recvmsg); 3645 3646 /* 3647 * Set socket options on an inet socket. 3648 */ 3649 int sock_common_setsockopt(struct socket *sock, int level, int optname, 3650 sockptr_t optval, unsigned int optlen) 3651 { 3652 struct sock *sk = sock->sk; 3653 3654 /* IPV6_ADDRFORM can change sk->sk_prot under us. */ 3655 return READ_ONCE(sk->sk_prot)->setsockopt(sk, level, optname, optval, optlen); 3656 } 3657 EXPORT_SYMBOL(sock_common_setsockopt); 3658 3659 void sk_common_release(struct sock *sk) 3660 { 3661 if (sk->sk_prot->destroy) 3662 sk->sk_prot->destroy(sk); 3663 3664 /* 3665 * Observation: when sk_common_release is called, processes have 3666 * no access to socket. But net still has. 3667 * Step one, detach it from networking: 3668 * 3669 * A. Remove from hash tables. 3670 */ 3671 3672 sk->sk_prot->unhash(sk); 3673 3674 /* 3675 * In this point socket cannot receive new packets, but it is possible 3676 * that some packets are in flight because some CPU runs receiver and 3677 * did hash table lookup before we unhashed socket. They will achieve 3678 * receive queue and will be purged by socket destructor. 3679 * 3680 * Also we still have packets pending on receive queue and probably, 3681 * our own packets waiting in device queues. sock_destroy will drain 3682 * receive queue, but transmitted packets will delay socket destruction 3683 * until the last reference will be released. 3684 */ 3685 3686 sock_orphan(sk); 3687 3688 xfrm_sk_free_policy(sk); 3689 3690 sk_refcnt_debug_release(sk); 3691 3692 sock_put(sk); 3693 } 3694 EXPORT_SYMBOL(sk_common_release); 3695 3696 void sk_get_meminfo(const struct sock *sk, u32 *mem) 3697 { 3698 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS); 3699 3700 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk); 3701 mem[SK_MEMINFO_RCVBUF] = READ_ONCE(sk->sk_rcvbuf); 3702 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk); 3703 mem[SK_MEMINFO_SNDBUF] = READ_ONCE(sk->sk_sndbuf); 3704 mem[SK_MEMINFO_FWD_ALLOC] = sk->sk_forward_alloc; 3705 mem[SK_MEMINFO_WMEM_QUEUED] = READ_ONCE(sk->sk_wmem_queued); 3706 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc); 3707 mem[SK_MEMINFO_BACKLOG] = READ_ONCE(sk->sk_backlog.len); 3708 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops); 3709 } 3710 3711 #ifdef CONFIG_PROC_FS 3712 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR); 3713 3714 int sock_prot_inuse_get(struct net *net, struct proto *prot) 3715 { 3716 int cpu, idx = prot->inuse_idx; 3717 int res = 0; 3718 3719 for_each_possible_cpu(cpu) 3720 res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx]; 3721 3722 return res >= 0 ? res : 0; 3723 } 3724 EXPORT_SYMBOL_GPL(sock_prot_inuse_get); 3725 3726 int sock_inuse_get(struct net *net) 3727 { 3728 int cpu, res = 0; 3729 3730 for_each_possible_cpu(cpu) 3731 res += per_cpu_ptr(net->core.prot_inuse, cpu)->all; 3732 3733 return res; 3734 } 3735 3736 EXPORT_SYMBOL_GPL(sock_inuse_get); 3737 3738 static int __net_init sock_inuse_init_net(struct net *net) 3739 { 3740 net->core.prot_inuse = alloc_percpu(struct prot_inuse); 3741 if (net->core.prot_inuse == NULL) 3742 return -ENOMEM; 3743 return 0; 3744 } 3745 3746 static void __net_exit sock_inuse_exit_net(struct net *net) 3747 { 3748 free_percpu(net->core.prot_inuse); 3749 } 3750 3751 static struct pernet_operations net_inuse_ops = { 3752 .init = sock_inuse_init_net, 3753 .exit = sock_inuse_exit_net, 3754 }; 3755 3756 static __init int net_inuse_init(void) 3757 { 3758 if (register_pernet_subsys(&net_inuse_ops)) 3759 panic("Cannot initialize net inuse counters"); 3760 3761 return 0; 3762 } 3763 3764 core_initcall(net_inuse_init); 3765 3766 static int assign_proto_idx(struct proto *prot) 3767 { 3768 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR); 3769 3770 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) { 3771 pr_err("PROTO_INUSE_NR exhausted\n"); 3772 return -ENOSPC; 3773 } 3774 3775 set_bit(prot->inuse_idx, proto_inuse_idx); 3776 return 0; 3777 } 3778 3779 static void release_proto_idx(struct proto *prot) 3780 { 3781 if (prot->inuse_idx != PROTO_INUSE_NR - 1) 3782 clear_bit(prot->inuse_idx, proto_inuse_idx); 3783 } 3784 #else 3785 static inline int assign_proto_idx(struct proto *prot) 3786 { 3787 return 0; 3788 } 3789 3790 static inline void release_proto_idx(struct proto *prot) 3791 { 3792 } 3793 3794 #endif 3795 3796 static void tw_prot_cleanup(struct timewait_sock_ops *twsk_prot) 3797 { 3798 if (!twsk_prot) 3799 return; 3800 kfree(twsk_prot->twsk_slab_name); 3801 twsk_prot->twsk_slab_name = NULL; 3802 kmem_cache_destroy(twsk_prot->twsk_slab); 3803 twsk_prot->twsk_slab = NULL; 3804 } 3805 3806 static int tw_prot_init(const struct proto *prot) 3807 { 3808 struct timewait_sock_ops *twsk_prot = prot->twsk_prot; 3809 3810 if (!twsk_prot) 3811 return 0; 3812 3813 twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", 3814 prot->name); 3815 if (!twsk_prot->twsk_slab_name) 3816 return -ENOMEM; 3817 3818 twsk_prot->twsk_slab = 3819 kmem_cache_create(twsk_prot->twsk_slab_name, 3820 twsk_prot->twsk_obj_size, 0, 3821 SLAB_ACCOUNT | prot->slab_flags, 3822 NULL); 3823 if (!twsk_prot->twsk_slab) { 3824 pr_crit("%s: Can't create timewait sock SLAB cache!\n", 3825 prot->name); 3826 return -ENOMEM; 3827 } 3828 3829 return 0; 3830 } 3831 3832 static void req_prot_cleanup(struct request_sock_ops *rsk_prot) 3833 { 3834 if (!rsk_prot) 3835 return; 3836 kfree(rsk_prot->slab_name); 3837 rsk_prot->slab_name = NULL; 3838 kmem_cache_destroy(rsk_prot->slab); 3839 rsk_prot->slab = NULL; 3840 } 3841 3842 static int req_prot_init(const struct proto *prot) 3843 { 3844 struct request_sock_ops *rsk_prot = prot->rsk_prot; 3845 3846 if (!rsk_prot) 3847 return 0; 3848 3849 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", 3850 prot->name); 3851 if (!rsk_prot->slab_name) 3852 return -ENOMEM; 3853 3854 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name, 3855 rsk_prot->obj_size, 0, 3856 SLAB_ACCOUNT | prot->slab_flags, 3857 NULL); 3858 3859 if (!rsk_prot->slab) { 3860 pr_crit("%s: Can't create request sock SLAB cache!\n", 3861 prot->name); 3862 return -ENOMEM; 3863 } 3864 return 0; 3865 } 3866 3867 int proto_register(struct proto *prot, int alloc_slab) 3868 { 3869 int ret = -ENOBUFS; 3870 3871 if (prot->memory_allocated && !prot->sysctl_mem) { 3872 pr_err("%s: missing sysctl_mem\n", prot->name); 3873 return -EINVAL; 3874 } 3875 if (prot->memory_allocated && !prot->per_cpu_fw_alloc) { 3876 pr_err("%s: missing per_cpu_fw_alloc\n", prot->name); 3877 return -EINVAL; 3878 } 3879 if (alloc_slab) { 3880 prot->slab = kmem_cache_create_usercopy(prot->name, 3881 prot->obj_size, 0, 3882 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT | 3883 prot->slab_flags, 3884 prot->useroffset, prot->usersize, 3885 NULL); 3886 3887 if (prot->slab == NULL) { 3888 pr_crit("%s: Can't create sock SLAB cache!\n", 3889 prot->name); 3890 goto out; 3891 } 3892 3893 if (req_prot_init(prot)) 3894 goto out_free_request_sock_slab; 3895 3896 if (tw_prot_init(prot)) 3897 goto out_free_timewait_sock_slab; 3898 } 3899 3900 mutex_lock(&proto_list_mutex); 3901 ret = assign_proto_idx(prot); 3902 if (ret) { 3903 mutex_unlock(&proto_list_mutex); 3904 goto out_free_timewait_sock_slab; 3905 } 3906 list_add(&prot->node, &proto_list); 3907 mutex_unlock(&proto_list_mutex); 3908 return ret; 3909 3910 out_free_timewait_sock_slab: 3911 if (alloc_slab) 3912 tw_prot_cleanup(prot->twsk_prot); 3913 out_free_request_sock_slab: 3914 if (alloc_slab) { 3915 req_prot_cleanup(prot->rsk_prot); 3916 3917 kmem_cache_destroy(prot->slab); 3918 prot->slab = NULL; 3919 } 3920 out: 3921 return ret; 3922 } 3923 EXPORT_SYMBOL(proto_register); 3924 3925 void proto_unregister(struct proto *prot) 3926 { 3927 mutex_lock(&proto_list_mutex); 3928 release_proto_idx(prot); 3929 list_del(&prot->node); 3930 mutex_unlock(&proto_list_mutex); 3931 3932 kmem_cache_destroy(prot->slab); 3933 prot->slab = NULL; 3934 3935 req_prot_cleanup(prot->rsk_prot); 3936 tw_prot_cleanup(prot->twsk_prot); 3937 } 3938 EXPORT_SYMBOL(proto_unregister); 3939 3940 int sock_load_diag_module(int family, int protocol) 3941 { 3942 if (!protocol) { 3943 if (!sock_is_registered(family)) 3944 return -ENOENT; 3945 3946 return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK, 3947 NETLINK_SOCK_DIAG, family); 3948 } 3949 3950 #ifdef CONFIG_INET 3951 if (family == AF_INET && 3952 protocol != IPPROTO_RAW && 3953 protocol < MAX_INET_PROTOS && 3954 !rcu_access_pointer(inet_protos[protocol])) 3955 return -ENOENT; 3956 #endif 3957 3958 return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK, 3959 NETLINK_SOCK_DIAG, family, protocol); 3960 } 3961 EXPORT_SYMBOL(sock_load_diag_module); 3962 3963 #ifdef CONFIG_PROC_FS 3964 static void *proto_seq_start(struct seq_file *seq, loff_t *pos) 3965 __acquires(proto_list_mutex) 3966 { 3967 mutex_lock(&proto_list_mutex); 3968 return seq_list_start_head(&proto_list, *pos); 3969 } 3970 3971 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos) 3972 { 3973 return seq_list_next(v, &proto_list, pos); 3974 } 3975 3976 static void proto_seq_stop(struct seq_file *seq, void *v) 3977 __releases(proto_list_mutex) 3978 { 3979 mutex_unlock(&proto_list_mutex); 3980 } 3981 3982 static char proto_method_implemented(const void *method) 3983 { 3984 return method == NULL ? 'n' : 'y'; 3985 } 3986 static long sock_prot_memory_allocated(struct proto *proto) 3987 { 3988 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L; 3989 } 3990 3991 static const char *sock_prot_memory_pressure(struct proto *proto) 3992 { 3993 return proto->memory_pressure != NULL ? 3994 proto_memory_pressure(proto) ? "yes" : "no" : "NI"; 3995 } 3996 3997 static void proto_seq_printf(struct seq_file *seq, struct proto *proto) 3998 { 3999 4000 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s " 4001 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n", 4002 proto->name, 4003 proto->obj_size, 4004 sock_prot_inuse_get(seq_file_net(seq), proto), 4005 sock_prot_memory_allocated(proto), 4006 sock_prot_memory_pressure(proto), 4007 proto->max_header, 4008 proto->slab == NULL ? "no" : "yes", 4009 module_name(proto->owner), 4010 proto_method_implemented(proto->close), 4011 proto_method_implemented(proto->connect), 4012 proto_method_implemented(proto->disconnect), 4013 proto_method_implemented(proto->accept), 4014 proto_method_implemented(proto->ioctl), 4015 proto_method_implemented(proto->init), 4016 proto_method_implemented(proto->destroy), 4017 proto_method_implemented(proto->shutdown), 4018 proto_method_implemented(proto->setsockopt), 4019 proto_method_implemented(proto->getsockopt), 4020 proto_method_implemented(proto->sendmsg), 4021 proto_method_implemented(proto->recvmsg), 4022 proto_method_implemented(proto->sendpage), 4023 proto_method_implemented(proto->bind), 4024 proto_method_implemented(proto->backlog_rcv), 4025 proto_method_implemented(proto->hash), 4026 proto_method_implemented(proto->unhash), 4027 proto_method_implemented(proto->get_port), 4028 proto_method_implemented(proto->enter_memory_pressure)); 4029 } 4030 4031 static int proto_seq_show(struct seq_file *seq, void *v) 4032 { 4033 if (v == &proto_list) 4034 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s", 4035 "protocol", 4036 "size", 4037 "sockets", 4038 "memory", 4039 "press", 4040 "maxhdr", 4041 "slab", 4042 "module", 4043 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n"); 4044 else 4045 proto_seq_printf(seq, list_entry(v, struct proto, node)); 4046 return 0; 4047 } 4048 4049 static const struct seq_operations proto_seq_ops = { 4050 .start = proto_seq_start, 4051 .next = proto_seq_next, 4052 .stop = proto_seq_stop, 4053 .show = proto_seq_show, 4054 }; 4055 4056 static __net_init int proto_init_net(struct net *net) 4057 { 4058 if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops, 4059 sizeof(struct seq_net_private))) 4060 return -ENOMEM; 4061 4062 return 0; 4063 } 4064 4065 static __net_exit void proto_exit_net(struct net *net) 4066 { 4067 remove_proc_entry("protocols", net->proc_net); 4068 } 4069 4070 4071 static __net_initdata struct pernet_operations proto_net_ops = { 4072 .init = proto_init_net, 4073 .exit = proto_exit_net, 4074 }; 4075 4076 static int __init proto_init(void) 4077 { 4078 return register_pernet_subsys(&proto_net_ops); 4079 } 4080 4081 subsys_initcall(proto_init); 4082 4083 #endif /* PROC_FS */ 4084 4085 #ifdef CONFIG_NET_RX_BUSY_POLL 4086 bool sk_busy_loop_end(void *p, unsigned long start_time) 4087 { 4088 struct sock *sk = p; 4089 4090 return !skb_queue_empty_lockless(&sk->sk_receive_queue) || 4091 sk_busy_loop_timeout(sk, start_time); 4092 } 4093 EXPORT_SYMBOL(sk_busy_loop_end); 4094 #endif /* CONFIG_NET_RX_BUSY_POLL */ 4095 4096 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len) 4097 { 4098 if (!sk->sk_prot->bind_add) 4099 return -EOPNOTSUPP; 4100 return sk->sk_prot->bind_add(sk, addr, addr_len); 4101 } 4102 EXPORT_SYMBOL(sock_bind_add); 4103