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_TCP && 905 !(val & SOF_TIMESTAMPING_OPT_ID)) 906 return -EINVAL; 907 908 if (val & SOF_TIMESTAMPING_OPT_ID && 909 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) { 910 if (sk_is_tcp(sk)) { 911 if ((1 << sk->sk_state) & 912 (TCPF_CLOSE | TCPF_LISTEN)) 913 return -EINVAL; 914 if (val & SOF_TIMESTAMPING_OPT_ID_TCP) 915 atomic_set(&sk->sk_tskey, tcp_sk(sk)->write_seq); 916 else 917 atomic_set(&sk->sk_tskey, tcp_sk(sk)->snd_una); 918 } else { 919 atomic_set(&sk->sk_tskey, 0); 920 } 921 } 922 923 if (val & SOF_TIMESTAMPING_OPT_STATS && 924 !(val & SOF_TIMESTAMPING_OPT_TSONLY)) 925 return -EINVAL; 926 927 if (val & SOF_TIMESTAMPING_BIND_PHC) { 928 ret = sock_timestamping_bind_phc(sk, timestamping.bind_phc); 929 if (ret) 930 return ret; 931 } 932 933 sk->sk_tsflags = val; 934 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, optname == SO_TIMESTAMPING_NEW); 935 936 if (val & SOF_TIMESTAMPING_RX_SOFTWARE) 937 sock_enable_timestamp(sk, 938 SOCK_TIMESTAMPING_RX_SOFTWARE); 939 else 940 sock_disable_timestamp(sk, 941 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE)); 942 return 0; 943 } 944 945 void sock_set_keepalive(struct sock *sk) 946 { 947 lock_sock(sk); 948 if (sk->sk_prot->keepalive) 949 sk->sk_prot->keepalive(sk, true); 950 sock_valbool_flag(sk, SOCK_KEEPOPEN, true); 951 release_sock(sk); 952 } 953 EXPORT_SYMBOL(sock_set_keepalive); 954 955 static void __sock_set_rcvbuf(struct sock *sk, int val) 956 { 957 /* Ensure val * 2 fits into an int, to prevent max_t() from treating it 958 * as a negative value. 959 */ 960 val = min_t(int, val, INT_MAX / 2); 961 sk->sk_userlocks |= SOCK_RCVBUF_LOCK; 962 963 /* We double it on the way in to account for "struct sk_buff" etc. 964 * overhead. Applications assume that the SO_RCVBUF setting they make 965 * will allow that much actual data to be received on that socket. 966 * 967 * Applications are unaware that "struct sk_buff" and other overheads 968 * allocate from the receive buffer during socket buffer allocation. 969 * 970 * And after considering the possible alternatives, returning the value 971 * we actually used in getsockopt is the most desirable behavior. 972 */ 973 WRITE_ONCE(sk->sk_rcvbuf, max_t(int, val * 2, SOCK_MIN_RCVBUF)); 974 } 975 976 void sock_set_rcvbuf(struct sock *sk, int val) 977 { 978 lock_sock(sk); 979 __sock_set_rcvbuf(sk, val); 980 release_sock(sk); 981 } 982 EXPORT_SYMBOL(sock_set_rcvbuf); 983 984 static void __sock_set_mark(struct sock *sk, u32 val) 985 { 986 if (val != sk->sk_mark) { 987 sk->sk_mark = val; 988 sk_dst_reset(sk); 989 } 990 } 991 992 void sock_set_mark(struct sock *sk, u32 val) 993 { 994 lock_sock(sk); 995 __sock_set_mark(sk, val); 996 release_sock(sk); 997 } 998 EXPORT_SYMBOL(sock_set_mark); 999 1000 static void sock_release_reserved_memory(struct sock *sk, int bytes) 1001 { 1002 /* Round down bytes to multiple of pages */ 1003 bytes = round_down(bytes, PAGE_SIZE); 1004 1005 WARN_ON(bytes > sk->sk_reserved_mem); 1006 sk->sk_reserved_mem -= bytes; 1007 sk_mem_reclaim(sk); 1008 } 1009 1010 static int sock_reserve_memory(struct sock *sk, int bytes) 1011 { 1012 long allocated; 1013 bool charged; 1014 int pages; 1015 1016 if (!mem_cgroup_sockets_enabled || !sk->sk_memcg || !sk_has_account(sk)) 1017 return -EOPNOTSUPP; 1018 1019 if (!bytes) 1020 return 0; 1021 1022 pages = sk_mem_pages(bytes); 1023 1024 /* pre-charge to memcg */ 1025 charged = mem_cgroup_charge_skmem(sk->sk_memcg, pages, 1026 GFP_KERNEL | __GFP_RETRY_MAYFAIL); 1027 if (!charged) 1028 return -ENOMEM; 1029 1030 /* pre-charge to forward_alloc */ 1031 sk_memory_allocated_add(sk, pages); 1032 allocated = sk_memory_allocated(sk); 1033 /* If the system goes into memory pressure with this 1034 * precharge, give up and return error. 1035 */ 1036 if (allocated > sk_prot_mem_limits(sk, 1)) { 1037 sk_memory_allocated_sub(sk, pages); 1038 mem_cgroup_uncharge_skmem(sk->sk_memcg, pages); 1039 return -ENOMEM; 1040 } 1041 sk->sk_forward_alloc += pages << PAGE_SHIFT; 1042 1043 sk->sk_reserved_mem += pages << PAGE_SHIFT; 1044 1045 return 0; 1046 } 1047 1048 void sockopt_lock_sock(struct sock *sk) 1049 { 1050 /* When current->bpf_ctx is set, the setsockopt is called from 1051 * a bpf prog. bpf has ensured the sk lock has been 1052 * acquired before calling setsockopt(). 1053 */ 1054 if (has_current_bpf_ctx()) 1055 return; 1056 1057 lock_sock(sk); 1058 } 1059 EXPORT_SYMBOL(sockopt_lock_sock); 1060 1061 void sockopt_release_sock(struct sock *sk) 1062 { 1063 if (has_current_bpf_ctx()) 1064 return; 1065 1066 release_sock(sk); 1067 } 1068 EXPORT_SYMBOL(sockopt_release_sock); 1069 1070 bool sockopt_ns_capable(struct user_namespace *ns, int cap) 1071 { 1072 return has_current_bpf_ctx() || ns_capable(ns, cap); 1073 } 1074 EXPORT_SYMBOL(sockopt_ns_capable); 1075 1076 bool sockopt_capable(int cap) 1077 { 1078 return has_current_bpf_ctx() || capable(cap); 1079 } 1080 EXPORT_SYMBOL(sockopt_capable); 1081 1082 /* 1083 * This is meant for all protocols to use and covers goings on 1084 * at the socket level. Everything here is generic. 1085 */ 1086 1087 int sk_setsockopt(struct sock *sk, int level, int optname, 1088 sockptr_t optval, unsigned int optlen) 1089 { 1090 struct so_timestamping timestamping; 1091 struct socket *sock = sk->sk_socket; 1092 struct sock_txtime sk_txtime; 1093 int val; 1094 int valbool; 1095 struct linger ling; 1096 int ret = 0; 1097 1098 /* 1099 * Options without arguments 1100 */ 1101 1102 if (optname == SO_BINDTODEVICE) 1103 return sock_setbindtodevice(sk, optval, optlen); 1104 1105 if (optlen < sizeof(int)) 1106 return -EINVAL; 1107 1108 if (copy_from_sockptr(&val, optval, sizeof(val))) 1109 return -EFAULT; 1110 1111 valbool = val ? 1 : 0; 1112 1113 sockopt_lock_sock(sk); 1114 1115 switch (optname) { 1116 case SO_DEBUG: 1117 if (val && !sockopt_capable(CAP_NET_ADMIN)) 1118 ret = -EACCES; 1119 else 1120 sock_valbool_flag(sk, SOCK_DBG, valbool); 1121 break; 1122 case SO_REUSEADDR: 1123 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE); 1124 break; 1125 case SO_REUSEPORT: 1126 sk->sk_reuseport = valbool; 1127 break; 1128 case SO_TYPE: 1129 case SO_PROTOCOL: 1130 case SO_DOMAIN: 1131 case SO_ERROR: 1132 ret = -ENOPROTOOPT; 1133 break; 1134 case SO_DONTROUTE: 1135 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool); 1136 sk_dst_reset(sk); 1137 break; 1138 case SO_BROADCAST: 1139 sock_valbool_flag(sk, SOCK_BROADCAST, valbool); 1140 break; 1141 case SO_SNDBUF: 1142 /* Don't error on this BSD doesn't and if you think 1143 * about it this is right. Otherwise apps have to 1144 * play 'guess the biggest size' games. RCVBUF/SNDBUF 1145 * are treated in BSD as hints 1146 */ 1147 val = min_t(u32, val, READ_ONCE(sysctl_wmem_max)); 1148 set_sndbuf: 1149 /* Ensure val * 2 fits into an int, to prevent max_t() 1150 * from treating it as a negative value. 1151 */ 1152 val = min_t(int, val, INT_MAX / 2); 1153 sk->sk_userlocks |= SOCK_SNDBUF_LOCK; 1154 WRITE_ONCE(sk->sk_sndbuf, 1155 max_t(int, val * 2, SOCK_MIN_SNDBUF)); 1156 /* Wake up sending tasks if we upped the value. */ 1157 sk->sk_write_space(sk); 1158 break; 1159 1160 case SO_SNDBUFFORCE: 1161 if (!sockopt_capable(CAP_NET_ADMIN)) { 1162 ret = -EPERM; 1163 break; 1164 } 1165 1166 /* No negative values (to prevent underflow, as val will be 1167 * multiplied by 2). 1168 */ 1169 if (val < 0) 1170 val = 0; 1171 goto set_sndbuf; 1172 1173 case SO_RCVBUF: 1174 /* Don't error on this BSD doesn't and if you think 1175 * about it this is right. Otherwise apps have to 1176 * play 'guess the biggest size' games. RCVBUF/SNDBUF 1177 * are treated in BSD as hints 1178 */ 1179 __sock_set_rcvbuf(sk, min_t(u32, val, READ_ONCE(sysctl_rmem_max))); 1180 break; 1181 1182 case SO_RCVBUFFORCE: 1183 if (!sockopt_capable(CAP_NET_ADMIN)) { 1184 ret = -EPERM; 1185 break; 1186 } 1187 1188 /* No negative values (to prevent underflow, as val will be 1189 * multiplied by 2). 1190 */ 1191 __sock_set_rcvbuf(sk, max(val, 0)); 1192 break; 1193 1194 case SO_KEEPALIVE: 1195 if (sk->sk_prot->keepalive) 1196 sk->sk_prot->keepalive(sk, valbool); 1197 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool); 1198 break; 1199 1200 case SO_OOBINLINE: 1201 sock_valbool_flag(sk, SOCK_URGINLINE, valbool); 1202 break; 1203 1204 case SO_NO_CHECK: 1205 sk->sk_no_check_tx = valbool; 1206 break; 1207 1208 case SO_PRIORITY: 1209 if ((val >= 0 && val <= 6) || 1210 sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) || 1211 sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) 1212 sk->sk_priority = val; 1213 else 1214 ret = -EPERM; 1215 break; 1216 1217 case SO_LINGER: 1218 if (optlen < sizeof(ling)) { 1219 ret = -EINVAL; /* 1003.1g */ 1220 break; 1221 } 1222 if (copy_from_sockptr(&ling, optval, sizeof(ling))) { 1223 ret = -EFAULT; 1224 break; 1225 } 1226 if (!ling.l_onoff) 1227 sock_reset_flag(sk, SOCK_LINGER); 1228 else { 1229 #if (BITS_PER_LONG == 32) 1230 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ) 1231 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT; 1232 else 1233 #endif 1234 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ; 1235 sock_set_flag(sk, SOCK_LINGER); 1236 } 1237 break; 1238 1239 case SO_BSDCOMPAT: 1240 break; 1241 1242 case SO_PASSCRED: 1243 if (valbool) 1244 set_bit(SOCK_PASSCRED, &sock->flags); 1245 else 1246 clear_bit(SOCK_PASSCRED, &sock->flags); 1247 break; 1248 1249 case SO_TIMESTAMP_OLD: 1250 case SO_TIMESTAMP_NEW: 1251 case SO_TIMESTAMPNS_OLD: 1252 case SO_TIMESTAMPNS_NEW: 1253 sock_set_timestamp(sk, optname, valbool); 1254 break; 1255 1256 case SO_TIMESTAMPING_NEW: 1257 case SO_TIMESTAMPING_OLD: 1258 if (optlen == sizeof(timestamping)) { 1259 if (copy_from_sockptr(×tamping, optval, 1260 sizeof(timestamping))) { 1261 ret = -EFAULT; 1262 break; 1263 } 1264 } else { 1265 memset(×tamping, 0, sizeof(timestamping)); 1266 timestamping.flags = val; 1267 } 1268 ret = sock_set_timestamping(sk, optname, timestamping); 1269 break; 1270 1271 case SO_RCVLOWAT: 1272 if (val < 0) 1273 val = INT_MAX; 1274 if (sock && sock->ops->set_rcvlowat) 1275 ret = sock->ops->set_rcvlowat(sk, val); 1276 else 1277 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1); 1278 break; 1279 1280 case SO_RCVTIMEO_OLD: 1281 case SO_RCVTIMEO_NEW: 1282 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, 1283 optlen, optname == SO_RCVTIMEO_OLD); 1284 break; 1285 1286 case SO_SNDTIMEO_OLD: 1287 case SO_SNDTIMEO_NEW: 1288 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, 1289 optlen, optname == SO_SNDTIMEO_OLD); 1290 break; 1291 1292 case SO_ATTACH_FILTER: { 1293 struct sock_fprog fprog; 1294 1295 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen); 1296 if (!ret) 1297 ret = sk_attach_filter(&fprog, sk); 1298 break; 1299 } 1300 case SO_ATTACH_BPF: 1301 ret = -EINVAL; 1302 if (optlen == sizeof(u32)) { 1303 u32 ufd; 1304 1305 ret = -EFAULT; 1306 if (copy_from_sockptr(&ufd, optval, sizeof(ufd))) 1307 break; 1308 1309 ret = sk_attach_bpf(ufd, sk); 1310 } 1311 break; 1312 1313 case SO_ATTACH_REUSEPORT_CBPF: { 1314 struct sock_fprog fprog; 1315 1316 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen); 1317 if (!ret) 1318 ret = sk_reuseport_attach_filter(&fprog, sk); 1319 break; 1320 } 1321 case SO_ATTACH_REUSEPORT_EBPF: 1322 ret = -EINVAL; 1323 if (optlen == sizeof(u32)) { 1324 u32 ufd; 1325 1326 ret = -EFAULT; 1327 if (copy_from_sockptr(&ufd, optval, sizeof(ufd))) 1328 break; 1329 1330 ret = sk_reuseport_attach_bpf(ufd, sk); 1331 } 1332 break; 1333 1334 case SO_DETACH_REUSEPORT_BPF: 1335 ret = reuseport_detach_prog(sk); 1336 break; 1337 1338 case SO_DETACH_FILTER: 1339 ret = sk_detach_filter(sk); 1340 break; 1341 1342 case SO_LOCK_FILTER: 1343 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool) 1344 ret = -EPERM; 1345 else 1346 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool); 1347 break; 1348 1349 case SO_PASSSEC: 1350 if (valbool) 1351 set_bit(SOCK_PASSSEC, &sock->flags); 1352 else 1353 clear_bit(SOCK_PASSSEC, &sock->flags); 1354 break; 1355 case SO_MARK: 1356 if (!sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) && 1357 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) { 1358 ret = -EPERM; 1359 break; 1360 } 1361 1362 __sock_set_mark(sk, val); 1363 break; 1364 case SO_RCVMARK: 1365 sock_valbool_flag(sk, SOCK_RCVMARK, valbool); 1366 break; 1367 1368 case SO_RXQ_OVFL: 1369 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool); 1370 break; 1371 1372 case SO_WIFI_STATUS: 1373 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool); 1374 break; 1375 1376 case SO_PEEK_OFF: 1377 if (sock->ops->set_peek_off) 1378 ret = sock->ops->set_peek_off(sk, val); 1379 else 1380 ret = -EOPNOTSUPP; 1381 break; 1382 1383 case SO_NOFCS: 1384 sock_valbool_flag(sk, SOCK_NOFCS, valbool); 1385 break; 1386 1387 case SO_SELECT_ERR_QUEUE: 1388 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool); 1389 break; 1390 1391 #ifdef CONFIG_NET_RX_BUSY_POLL 1392 case SO_BUSY_POLL: 1393 if (val < 0) 1394 ret = -EINVAL; 1395 else 1396 WRITE_ONCE(sk->sk_ll_usec, val); 1397 break; 1398 case SO_PREFER_BUSY_POLL: 1399 if (valbool && !sockopt_capable(CAP_NET_ADMIN)) 1400 ret = -EPERM; 1401 else 1402 WRITE_ONCE(sk->sk_prefer_busy_poll, valbool); 1403 break; 1404 case SO_BUSY_POLL_BUDGET: 1405 if (val > READ_ONCE(sk->sk_busy_poll_budget) && !sockopt_capable(CAP_NET_ADMIN)) { 1406 ret = -EPERM; 1407 } else { 1408 if (val < 0 || val > U16_MAX) 1409 ret = -EINVAL; 1410 else 1411 WRITE_ONCE(sk->sk_busy_poll_budget, val); 1412 } 1413 break; 1414 #endif 1415 1416 case SO_MAX_PACING_RATE: 1417 { 1418 unsigned long ulval = (val == ~0U) ? ~0UL : (unsigned int)val; 1419 1420 if (sizeof(ulval) != sizeof(val) && 1421 optlen >= sizeof(ulval) && 1422 copy_from_sockptr(&ulval, optval, sizeof(ulval))) { 1423 ret = -EFAULT; 1424 break; 1425 } 1426 if (ulval != ~0UL) 1427 cmpxchg(&sk->sk_pacing_status, 1428 SK_PACING_NONE, 1429 SK_PACING_NEEDED); 1430 sk->sk_max_pacing_rate = ulval; 1431 sk->sk_pacing_rate = min(sk->sk_pacing_rate, ulval); 1432 break; 1433 } 1434 case SO_INCOMING_CPU: 1435 reuseport_update_incoming_cpu(sk, val); 1436 break; 1437 1438 case SO_CNX_ADVICE: 1439 if (val == 1) 1440 dst_negative_advice(sk); 1441 break; 1442 1443 case SO_ZEROCOPY: 1444 if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) { 1445 if (!(sk_is_tcp(sk) || 1446 (sk->sk_type == SOCK_DGRAM && 1447 sk->sk_protocol == IPPROTO_UDP))) 1448 ret = -EOPNOTSUPP; 1449 } else if (sk->sk_family != PF_RDS) { 1450 ret = -EOPNOTSUPP; 1451 } 1452 if (!ret) { 1453 if (val < 0 || val > 1) 1454 ret = -EINVAL; 1455 else 1456 sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool); 1457 } 1458 break; 1459 1460 case SO_TXTIME: 1461 if (optlen != sizeof(struct sock_txtime)) { 1462 ret = -EINVAL; 1463 break; 1464 } else if (copy_from_sockptr(&sk_txtime, optval, 1465 sizeof(struct sock_txtime))) { 1466 ret = -EFAULT; 1467 break; 1468 } else if (sk_txtime.flags & ~SOF_TXTIME_FLAGS_MASK) { 1469 ret = -EINVAL; 1470 break; 1471 } 1472 /* CLOCK_MONOTONIC is only used by sch_fq, and this packet 1473 * scheduler has enough safe guards. 1474 */ 1475 if (sk_txtime.clockid != CLOCK_MONOTONIC && 1476 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) { 1477 ret = -EPERM; 1478 break; 1479 } 1480 sock_valbool_flag(sk, SOCK_TXTIME, true); 1481 sk->sk_clockid = sk_txtime.clockid; 1482 sk->sk_txtime_deadline_mode = 1483 !!(sk_txtime.flags & SOF_TXTIME_DEADLINE_MODE); 1484 sk->sk_txtime_report_errors = 1485 !!(sk_txtime.flags & SOF_TXTIME_REPORT_ERRORS); 1486 break; 1487 1488 case SO_BINDTOIFINDEX: 1489 ret = sock_bindtoindex_locked(sk, val); 1490 break; 1491 1492 case SO_BUF_LOCK: 1493 if (val & ~SOCK_BUF_LOCK_MASK) { 1494 ret = -EINVAL; 1495 break; 1496 } 1497 sk->sk_userlocks = val | (sk->sk_userlocks & 1498 ~SOCK_BUF_LOCK_MASK); 1499 break; 1500 1501 case SO_RESERVE_MEM: 1502 { 1503 int delta; 1504 1505 if (val < 0) { 1506 ret = -EINVAL; 1507 break; 1508 } 1509 1510 delta = val - sk->sk_reserved_mem; 1511 if (delta < 0) 1512 sock_release_reserved_memory(sk, -delta); 1513 else 1514 ret = sock_reserve_memory(sk, delta); 1515 break; 1516 } 1517 1518 case SO_TXREHASH: 1519 if (val < -1 || val > 1) { 1520 ret = -EINVAL; 1521 break; 1522 } 1523 if ((u8)val == SOCK_TXREHASH_DEFAULT) 1524 val = READ_ONCE(sock_net(sk)->core.sysctl_txrehash); 1525 /* Paired with READ_ONCE() in tcp_rtx_synack() */ 1526 WRITE_ONCE(sk->sk_txrehash, (u8)val); 1527 break; 1528 1529 default: 1530 ret = -ENOPROTOOPT; 1531 break; 1532 } 1533 sockopt_release_sock(sk); 1534 return ret; 1535 } 1536 1537 int sock_setsockopt(struct socket *sock, int level, int optname, 1538 sockptr_t optval, unsigned int optlen) 1539 { 1540 return sk_setsockopt(sock->sk, level, optname, 1541 optval, optlen); 1542 } 1543 EXPORT_SYMBOL(sock_setsockopt); 1544 1545 static const struct cred *sk_get_peer_cred(struct sock *sk) 1546 { 1547 const struct cred *cred; 1548 1549 spin_lock(&sk->sk_peer_lock); 1550 cred = get_cred(sk->sk_peer_cred); 1551 spin_unlock(&sk->sk_peer_lock); 1552 1553 return cred; 1554 } 1555 1556 static void cred_to_ucred(struct pid *pid, const struct cred *cred, 1557 struct ucred *ucred) 1558 { 1559 ucred->pid = pid_vnr(pid); 1560 ucred->uid = ucred->gid = -1; 1561 if (cred) { 1562 struct user_namespace *current_ns = current_user_ns(); 1563 1564 ucred->uid = from_kuid_munged(current_ns, cred->euid); 1565 ucred->gid = from_kgid_munged(current_ns, cred->egid); 1566 } 1567 } 1568 1569 static int groups_to_user(sockptr_t dst, const struct group_info *src) 1570 { 1571 struct user_namespace *user_ns = current_user_ns(); 1572 int i; 1573 1574 for (i = 0; i < src->ngroups; i++) { 1575 gid_t gid = from_kgid_munged(user_ns, src->gid[i]); 1576 1577 if (copy_to_sockptr_offset(dst, i * sizeof(gid), &gid, sizeof(gid))) 1578 return -EFAULT; 1579 } 1580 1581 return 0; 1582 } 1583 1584 int sk_getsockopt(struct sock *sk, int level, int optname, 1585 sockptr_t optval, sockptr_t optlen) 1586 { 1587 struct socket *sock = sk->sk_socket; 1588 1589 union { 1590 int val; 1591 u64 val64; 1592 unsigned long ulval; 1593 struct linger ling; 1594 struct old_timeval32 tm32; 1595 struct __kernel_old_timeval tm; 1596 struct __kernel_sock_timeval stm; 1597 struct sock_txtime txtime; 1598 struct so_timestamping timestamping; 1599 } v; 1600 1601 int lv = sizeof(int); 1602 int len; 1603 1604 if (copy_from_sockptr(&len, optlen, sizeof(int))) 1605 return -EFAULT; 1606 if (len < 0) 1607 return -EINVAL; 1608 1609 memset(&v, 0, sizeof(v)); 1610 1611 switch (optname) { 1612 case SO_DEBUG: 1613 v.val = sock_flag(sk, SOCK_DBG); 1614 break; 1615 1616 case SO_DONTROUTE: 1617 v.val = sock_flag(sk, SOCK_LOCALROUTE); 1618 break; 1619 1620 case SO_BROADCAST: 1621 v.val = sock_flag(sk, SOCK_BROADCAST); 1622 break; 1623 1624 case SO_SNDBUF: 1625 v.val = sk->sk_sndbuf; 1626 break; 1627 1628 case SO_RCVBUF: 1629 v.val = sk->sk_rcvbuf; 1630 break; 1631 1632 case SO_REUSEADDR: 1633 v.val = sk->sk_reuse; 1634 break; 1635 1636 case SO_REUSEPORT: 1637 v.val = sk->sk_reuseport; 1638 break; 1639 1640 case SO_KEEPALIVE: 1641 v.val = sock_flag(sk, SOCK_KEEPOPEN); 1642 break; 1643 1644 case SO_TYPE: 1645 v.val = sk->sk_type; 1646 break; 1647 1648 case SO_PROTOCOL: 1649 v.val = sk->sk_protocol; 1650 break; 1651 1652 case SO_DOMAIN: 1653 v.val = sk->sk_family; 1654 break; 1655 1656 case SO_ERROR: 1657 v.val = -sock_error(sk); 1658 if (v.val == 0) 1659 v.val = xchg(&sk->sk_err_soft, 0); 1660 break; 1661 1662 case SO_OOBINLINE: 1663 v.val = sock_flag(sk, SOCK_URGINLINE); 1664 break; 1665 1666 case SO_NO_CHECK: 1667 v.val = sk->sk_no_check_tx; 1668 break; 1669 1670 case SO_PRIORITY: 1671 v.val = sk->sk_priority; 1672 break; 1673 1674 case SO_LINGER: 1675 lv = sizeof(v.ling); 1676 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER); 1677 v.ling.l_linger = sk->sk_lingertime / HZ; 1678 break; 1679 1680 case SO_BSDCOMPAT: 1681 break; 1682 1683 case SO_TIMESTAMP_OLD: 1684 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && 1685 !sock_flag(sk, SOCK_TSTAMP_NEW) && 1686 !sock_flag(sk, SOCK_RCVTSTAMPNS); 1687 break; 1688 1689 case SO_TIMESTAMPNS_OLD: 1690 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && !sock_flag(sk, SOCK_TSTAMP_NEW); 1691 break; 1692 1693 case SO_TIMESTAMP_NEW: 1694 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && sock_flag(sk, SOCK_TSTAMP_NEW); 1695 break; 1696 1697 case SO_TIMESTAMPNS_NEW: 1698 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && sock_flag(sk, SOCK_TSTAMP_NEW); 1699 break; 1700 1701 case SO_TIMESTAMPING_OLD: 1702 lv = sizeof(v.timestamping); 1703 v.timestamping.flags = sk->sk_tsflags; 1704 v.timestamping.bind_phc = sk->sk_bind_phc; 1705 break; 1706 1707 case SO_RCVTIMEO_OLD: 1708 case SO_RCVTIMEO_NEW: 1709 lv = sock_get_timeout(sk->sk_rcvtimeo, &v, SO_RCVTIMEO_OLD == optname); 1710 break; 1711 1712 case SO_SNDTIMEO_OLD: 1713 case SO_SNDTIMEO_NEW: 1714 lv = sock_get_timeout(sk->sk_sndtimeo, &v, SO_SNDTIMEO_OLD == optname); 1715 break; 1716 1717 case SO_RCVLOWAT: 1718 v.val = sk->sk_rcvlowat; 1719 break; 1720 1721 case SO_SNDLOWAT: 1722 v.val = 1; 1723 break; 1724 1725 case SO_PASSCRED: 1726 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags); 1727 break; 1728 1729 case SO_PEERCRED: 1730 { 1731 struct ucred peercred; 1732 if (len > sizeof(peercred)) 1733 len = sizeof(peercred); 1734 1735 spin_lock(&sk->sk_peer_lock); 1736 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred); 1737 spin_unlock(&sk->sk_peer_lock); 1738 1739 if (copy_to_sockptr(optval, &peercred, len)) 1740 return -EFAULT; 1741 goto lenout; 1742 } 1743 1744 case SO_PEERGROUPS: 1745 { 1746 const struct cred *cred; 1747 int ret, n; 1748 1749 cred = sk_get_peer_cred(sk); 1750 if (!cred) 1751 return -ENODATA; 1752 1753 n = cred->group_info->ngroups; 1754 if (len < n * sizeof(gid_t)) { 1755 len = n * sizeof(gid_t); 1756 put_cred(cred); 1757 return copy_to_sockptr(optlen, &len, sizeof(int)) ? -EFAULT : -ERANGE; 1758 } 1759 len = n * sizeof(gid_t); 1760 1761 ret = groups_to_user(optval, cred->group_info); 1762 put_cred(cred); 1763 if (ret) 1764 return ret; 1765 goto lenout; 1766 } 1767 1768 case SO_PEERNAME: 1769 { 1770 char address[128]; 1771 1772 lv = sock->ops->getname(sock, (struct sockaddr *)address, 2); 1773 if (lv < 0) 1774 return -ENOTCONN; 1775 if (lv < len) 1776 return -EINVAL; 1777 if (copy_to_sockptr(optval, address, len)) 1778 return -EFAULT; 1779 goto lenout; 1780 } 1781 1782 /* Dubious BSD thing... Probably nobody even uses it, but 1783 * the UNIX standard wants it for whatever reason... -DaveM 1784 */ 1785 case SO_ACCEPTCONN: 1786 v.val = sk->sk_state == TCP_LISTEN; 1787 break; 1788 1789 case SO_PASSSEC: 1790 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags); 1791 break; 1792 1793 case SO_PEERSEC: 1794 return security_socket_getpeersec_stream(sock, 1795 optval, optlen, len); 1796 1797 case SO_MARK: 1798 v.val = sk->sk_mark; 1799 break; 1800 1801 case SO_RCVMARK: 1802 v.val = sock_flag(sk, SOCK_RCVMARK); 1803 break; 1804 1805 case SO_RXQ_OVFL: 1806 v.val = sock_flag(sk, SOCK_RXQ_OVFL); 1807 break; 1808 1809 case SO_WIFI_STATUS: 1810 v.val = sock_flag(sk, SOCK_WIFI_STATUS); 1811 break; 1812 1813 case SO_PEEK_OFF: 1814 if (!sock->ops->set_peek_off) 1815 return -EOPNOTSUPP; 1816 1817 v.val = sk->sk_peek_off; 1818 break; 1819 case SO_NOFCS: 1820 v.val = sock_flag(sk, SOCK_NOFCS); 1821 break; 1822 1823 case SO_BINDTODEVICE: 1824 return sock_getbindtodevice(sk, optval, optlen, len); 1825 1826 case SO_GET_FILTER: 1827 len = sk_get_filter(sk, optval, len); 1828 if (len < 0) 1829 return len; 1830 1831 goto lenout; 1832 1833 case SO_LOCK_FILTER: 1834 v.val = sock_flag(sk, SOCK_FILTER_LOCKED); 1835 break; 1836 1837 case SO_BPF_EXTENSIONS: 1838 v.val = bpf_tell_extensions(); 1839 break; 1840 1841 case SO_SELECT_ERR_QUEUE: 1842 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE); 1843 break; 1844 1845 #ifdef CONFIG_NET_RX_BUSY_POLL 1846 case SO_BUSY_POLL: 1847 v.val = sk->sk_ll_usec; 1848 break; 1849 case SO_PREFER_BUSY_POLL: 1850 v.val = READ_ONCE(sk->sk_prefer_busy_poll); 1851 break; 1852 #endif 1853 1854 case SO_MAX_PACING_RATE: 1855 if (sizeof(v.ulval) != sizeof(v.val) && len >= sizeof(v.ulval)) { 1856 lv = sizeof(v.ulval); 1857 v.ulval = sk->sk_max_pacing_rate; 1858 } else { 1859 /* 32bit version */ 1860 v.val = min_t(unsigned long, sk->sk_max_pacing_rate, ~0U); 1861 } 1862 break; 1863 1864 case SO_INCOMING_CPU: 1865 v.val = READ_ONCE(sk->sk_incoming_cpu); 1866 break; 1867 1868 case SO_MEMINFO: 1869 { 1870 u32 meminfo[SK_MEMINFO_VARS]; 1871 1872 sk_get_meminfo(sk, meminfo); 1873 1874 len = min_t(unsigned int, len, sizeof(meminfo)); 1875 if (copy_to_sockptr(optval, &meminfo, len)) 1876 return -EFAULT; 1877 1878 goto lenout; 1879 } 1880 1881 #ifdef CONFIG_NET_RX_BUSY_POLL 1882 case SO_INCOMING_NAPI_ID: 1883 v.val = READ_ONCE(sk->sk_napi_id); 1884 1885 /* aggregate non-NAPI IDs down to 0 */ 1886 if (v.val < MIN_NAPI_ID) 1887 v.val = 0; 1888 1889 break; 1890 #endif 1891 1892 case SO_COOKIE: 1893 lv = sizeof(u64); 1894 if (len < lv) 1895 return -EINVAL; 1896 v.val64 = sock_gen_cookie(sk); 1897 break; 1898 1899 case SO_ZEROCOPY: 1900 v.val = sock_flag(sk, SOCK_ZEROCOPY); 1901 break; 1902 1903 case SO_TXTIME: 1904 lv = sizeof(v.txtime); 1905 v.txtime.clockid = sk->sk_clockid; 1906 v.txtime.flags |= sk->sk_txtime_deadline_mode ? 1907 SOF_TXTIME_DEADLINE_MODE : 0; 1908 v.txtime.flags |= sk->sk_txtime_report_errors ? 1909 SOF_TXTIME_REPORT_ERRORS : 0; 1910 break; 1911 1912 case SO_BINDTOIFINDEX: 1913 v.val = READ_ONCE(sk->sk_bound_dev_if); 1914 break; 1915 1916 case SO_NETNS_COOKIE: 1917 lv = sizeof(u64); 1918 if (len != lv) 1919 return -EINVAL; 1920 v.val64 = sock_net(sk)->net_cookie; 1921 break; 1922 1923 case SO_BUF_LOCK: 1924 v.val = sk->sk_userlocks & SOCK_BUF_LOCK_MASK; 1925 break; 1926 1927 case SO_RESERVE_MEM: 1928 v.val = sk->sk_reserved_mem; 1929 break; 1930 1931 case SO_TXREHASH: 1932 v.val = sk->sk_txrehash; 1933 break; 1934 1935 default: 1936 /* We implement the SO_SNDLOWAT etc to not be settable 1937 * (1003.1g 7). 1938 */ 1939 return -ENOPROTOOPT; 1940 } 1941 1942 if (len > lv) 1943 len = lv; 1944 if (copy_to_sockptr(optval, &v, len)) 1945 return -EFAULT; 1946 lenout: 1947 if (copy_to_sockptr(optlen, &len, sizeof(int))) 1948 return -EFAULT; 1949 return 0; 1950 } 1951 1952 int sock_getsockopt(struct socket *sock, int level, int optname, 1953 char __user *optval, int __user *optlen) 1954 { 1955 return sk_getsockopt(sock->sk, level, optname, 1956 USER_SOCKPTR(optval), 1957 USER_SOCKPTR(optlen)); 1958 } 1959 1960 /* 1961 * Initialize an sk_lock. 1962 * 1963 * (We also register the sk_lock with the lock validator.) 1964 */ 1965 static inline void sock_lock_init(struct sock *sk) 1966 { 1967 if (sk->sk_kern_sock) 1968 sock_lock_init_class_and_name( 1969 sk, 1970 af_family_kern_slock_key_strings[sk->sk_family], 1971 af_family_kern_slock_keys + sk->sk_family, 1972 af_family_kern_key_strings[sk->sk_family], 1973 af_family_kern_keys + sk->sk_family); 1974 else 1975 sock_lock_init_class_and_name( 1976 sk, 1977 af_family_slock_key_strings[sk->sk_family], 1978 af_family_slock_keys + sk->sk_family, 1979 af_family_key_strings[sk->sk_family], 1980 af_family_keys + sk->sk_family); 1981 } 1982 1983 /* 1984 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet, 1985 * even temporarly, because of RCU lookups. sk_node should also be left as is. 1986 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end 1987 */ 1988 static void sock_copy(struct sock *nsk, const struct sock *osk) 1989 { 1990 const struct proto *prot = READ_ONCE(osk->sk_prot); 1991 #ifdef CONFIG_SECURITY_NETWORK 1992 void *sptr = nsk->sk_security; 1993 #endif 1994 1995 /* If we move sk_tx_queue_mapping out of the private section, 1996 * we must check if sk_tx_queue_clear() is called after 1997 * sock_copy() in sk_clone_lock(). 1998 */ 1999 BUILD_BUG_ON(offsetof(struct sock, sk_tx_queue_mapping) < 2000 offsetof(struct sock, sk_dontcopy_begin) || 2001 offsetof(struct sock, sk_tx_queue_mapping) >= 2002 offsetof(struct sock, sk_dontcopy_end)); 2003 2004 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin)); 2005 2006 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end, 2007 prot->obj_size - offsetof(struct sock, sk_dontcopy_end)); 2008 2009 #ifdef CONFIG_SECURITY_NETWORK 2010 nsk->sk_security = sptr; 2011 security_sk_clone(osk, nsk); 2012 #endif 2013 } 2014 2015 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority, 2016 int family) 2017 { 2018 struct sock *sk; 2019 struct kmem_cache *slab; 2020 2021 slab = prot->slab; 2022 if (slab != NULL) { 2023 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO); 2024 if (!sk) 2025 return sk; 2026 if (want_init_on_alloc(priority)) 2027 sk_prot_clear_nulls(sk, prot->obj_size); 2028 } else 2029 sk = kmalloc(prot->obj_size, priority); 2030 2031 if (sk != NULL) { 2032 if (security_sk_alloc(sk, family, priority)) 2033 goto out_free; 2034 2035 if (!try_module_get(prot->owner)) 2036 goto out_free_sec; 2037 } 2038 2039 return sk; 2040 2041 out_free_sec: 2042 security_sk_free(sk); 2043 out_free: 2044 if (slab != NULL) 2045 kmem_cache_free(slab, sk); 2046 else 2047 kfree(sk); 2048 return NULL; 2049 } 2050 2051 static void sk_prot_free(struct proto *prot, struct sock *sk) 2052 { 2053 struct kmem_cache *slab; 2054 struct module *owner; 2055 2056 owner = prot->owner; 2057 slab = prot->slab; 2058 2059 cgroup_sk_free(&sk->sk_cgrp_data); 2060 mem_cgroup_sk_free(sk); 2061 security_sk_free(sk); 2062 if (slab != NULL) 2063 kmem_cache_free(slab, sk); 2064 else 2065 kfree(sk); 2066 module_put(owner); 2067 } 2068 2069 /** 2070 * sk_alloc - All socket objects are allocated here 2071 * @net: the applicable net namespace 2072 * @family: protocol family 2073 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 2074 * @prot: struct proto associated with this new sock instance 2075 * @kern: is this to be a kernel socket? 2076 */ 2077 struct sock *sk_alloc(struct net *net, int family, gfp_t priority, 2078 struct proto *prot, int kern) 2079 { 2080 struct sock *sk; 2081 2082 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family); 2083 if (sk) { 2084 sk->sk_family = family; 2085 /* 2086 * See comment in struct sock definition to understand 2087 * why we need sk_prot_creator -acme 2088 */ 2089 sk->sk_prot = sk->sk_prot_creator = prot; 2090 sk->sk_kern_sock = kern; 2091 sock_lock_init(sk); 2092 sk->sk_net_refcnt = kern ? 0 : 1; 2093 if (likely(sk->sk_net_refcnt)) { 2094 get_net_track(net, &sk->ns_tracker, priority); 2095 sock_inuse_add(net, 1); 2096 } else { 2097 __netns_tracker_alloc(net, &sk->ns_tracker, 2098 false, priority); 2099 } 2100 2101 sock_net_set(sk, net); 2102 refcount_set(&sk->sk_wmem_alloc, 1); 2103 2104 mem_cgroup_sk_alloc(sk); 2105 cgroup_sk_alloc(&sk->sk_cgrp_data); 2106 sock_update_classid(&sk->sk_cgrp_data); 2107 sock_update_netprioidx(&sk->sk_cgrp_data); 2108 sk_tx_queue_clear(sk); 2109 } 2110 2111 return sk; 2112 } 2113 EXPORT_SYMBOL(sk_alloc); 2114 2115 /* Sockets having SOCK_RCU_FREE will call this function after one RCU 2116 * grace period. This is the case for UDP sockets and TCP listeners. 2117 */ 2118 static void __sk_destruct(struct rcu_head *head) 2119 { 2120 struct sock *sk = container_of(head, struct sock, sk_rcu); 2121 struct sk_filter *filter; 2122 2123 if (sk->sk_destruct) 2124 sk->sk_destruct(sk); 2125 2126 filter = rcu_dereference_check(sk->sk_filter, 2127 refcount_read(&sk->sk_wmem_alloc) == 0); 2128 if (filter) { 2129 sk_filter_uncharge(sk, filter); 2130 RCU_INIT_POINTER(sk->sk_filter, NULL); 2131 } 2132 2133 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP); 2134 2135 #ifdef CONFIG_BPF_SYSCALL 2136 bpf_sk_storage_free(sk); 2137 #endif 2138 2139 if (atomic_read(&sk->sk_omem_alloc)) 2140 pr_debug("%s: optmem leakage (%d bytes) detected\n", 2141 __func__, atomic_read(&sk->sk_omem_alloc)); 2142 2143 if (sk->sk_frag.page) { 2144 put_page(sk->sk_frag.page); 2145 sk->sk_frag.page = NULL; 2146 } 2147 2148 /* We do not need to acquire sk->sk_peer_lock, we are the last user. */ 2149 put_cred(sk->sk_peer_cred); 2150 put_pid(sk->sk_peer_pid); 2151 2152 if (likely(sk->sk_net_refcnt)) 2153 put_net_track(sock_net(sk), &sk->ns_tracker); 2154 else 2155 __netns_tracker_free(sock_net(sk), &sk->ns_tracker, false); 2156 2157 sk_prot_free(sk->sk_prot_creator, sk); 2158 } 2159 2160 void sk_destruct(struct sock *sk) 2161 { 2162 bool use_call_rcu = sock_flag(sk, SOCK_RCU_FREE); 2163 2164 if (rcu_access_pointer(sk->sk_reuseport_cb)) { 2165 reuseport_detach_sock(sk); 2166 use_call_rcu = true; 2167 } 2168 2169 if (use_call_rcu) 2170 call_rcu(&sk->sk_rcu, __sk_destruct); 2171 else 2172 __sk_destruct(&sk->sk_rcu); 2173 } 2174 2175 static void __sk_free(struct sock *sk) 2176 { 2177 if (likely(sk->sk_net_refcnt)) 2178 sock_inuse_add(sock_net(sk), -1); 2179 2180 if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk))) 2181 sock_diag_broadcast_destroy(sk); 2182 else 2183 sk_destruct(sk); 2184 } 2185 2186 void sk_free(struct sock *sk) 2187 { 2188 /* 2189 * We subtract one from sk_wmem_alloc and can know if 2190 * some packets are still in some tx queue. 2191 * If not null, sock_wfree() will call __sk_free(sk) later 2192 */ 2193 if (refcount_dec_and_test(&sk->sk_wmem_alloc)) 2194 __sk_free(sk); 2195 } 2196 EXPORT_SYMBOL(sk_free); 2197 2198 static void sk_init_common(struct sock *sk) 2199 { 2200 skb_queue_head_init(&sk->sk_receive_queue); 2201 skb_queue_head_init(&sk->sk_write_queue); 2202 skb_queue_head_init(&sk->sk_error_queue); 2203 2204 rwlock_init(&sk->sk_callback_lock); 2205 lockdep_set_class_and_name(&sk->sk_receive_queue.lock, 2206 af_rlock_keys + sk->sk_family, 2207 af_family_rlock_key_strings[sk->sk_family]); 2208 lockdep_set_class_and_name(&sk->sk_write_queue.lock, 2209 af_wlock_keys + sk->sk_family, 2210 af_family_wlock_key_strings[sk->sk_family]); 2211 lockdep_set_class_and_name(&sk->sk_error_queue.lock, 2212 af_elock_keys + sk->sk_family, 2213 af_family_elock_key_strings[sk->sk_family]); 2214 lockdep_set_class_and_name(&sk->sk_callback_lock, 2215 af_callback_keys + sk->sk_family, 2216 af_family_clock_key_strings[sk->sk_family]); 2217 } 2218 2219 /** 2220 * sk_clone_lock - clone a socket, and lock its clone 2221 * @sk: the socket to clone 2222 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 2223 * 2224 * Caller must unlock socket even in error path (bh_unlock_sock(newsk)) 2225 */ 2226 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority) 2227 { 2228 struct proto *prot = READ_ONCE(sk->sk_prot); 2229 struct sk_filter *filter; 2230 bool is_charged = true; 2231 struct sock *newsk; 2232 2233 newsk = sk_prot_alloc(prot, priority, sk->sk_family); 2234 if (!newsk) 2235 goto out; 2236 2237 sock_copy(newsk, sk); 2238 2239 newsk->sk_prot_creator = prot; 2240 2241 /* SANITY */ 2242 if (likely(newsk->sk_net_refcnt)) { 2243 get_net_track(sock_net(newsk), &newsk->ns_tracker, priority); 2244 sock_inuse_add(sock_net(newsk), 1); 2245 } else { 2246 /* Kernel sockets are not elevating the struct net refcount. 2247 * Instead, use a tracker to more easily detect if a layer 2248 * is not properly dismantling its kernel sockets at netns 2249 * destroy time. 2250 */ 2251 __netns_tracker_alloc(sock_net(newsk), &newsk->ns_tracker, 2252 false, priority); 2253 } 2254 sk_node_init(&newsk->sk_node); 2255 sock_lock_init(newsk); 2256 bh_lock_sock(newsk); 2257 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL; 2258 newsk->sk_backlog.len = 0; 2259 2260 atomic_set(&newsk->sk_rmem_alloc, 0); 2261 2262 /* sk_wmem_alloc set to one (see sk_free() and sock_wfree()) */ 2263 refcount_set(&newsk->sk_wmem_alloc, 1); 2264 2265 atomic_set(&newsk->sk_omem_alloc, 0); 2266 sk_init_common(newsk); 2267 2268 newsk->sk_dst_cache = NULL; 2269 newsk->sk_dst_pending_confirm = 0; 2270 newsk->sk_wmem_queued = 0; 2271 newsk->sk_forward_alloc = 0; 2272 newsk->sk_reserved_mem = 0; 2273 atomic_set(&newsk->sk_drops, 0); 2274 newsk->sk_send_head = NULL; 2275 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK; 2276 atomic_set(&newsk->sk_zckey, 0); 2277 2278 sock_reset_flag(newsk, SOCK_DONE); 2279 2280 /* sk->sk_memcg will be populated at accept() time */ 2281 newsk->sk_memcg = NULL; 2282 2283 cgroup_sk_clone(&newsk->sk_cgrp_data); 2284 2285 rcu_read_lock(); 2286 filter = rcu_dereference(sk->sk_filter); 2287 if (filter != NULL) 2288 /* though it's an empty new sock, the charging may fail 2289 * if sysctl_optmem_max was changed between creation of 2290 * original socket and cloning 2291 */ 2292 is_charged = sk_filter_charge(newsk, filter); 2293 RCU_INIT_POINTER(newsk->sk_filter, filter); 2294 rcu_read_unlock(); 2295 2296 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) { 2297 /* We need to make sure that we don't uncharge the new 2298 * socket if we couldn't charge it in the first place 2299 * as otherwise we uncharge the parent's filter. 2300 */ 2301 if (!is_charged) 2302 RCU_INIT_POINTER(newsk->sk_filter, NULL); 2303 sk_free_unlock_clone(newsk); 2304 newsk = NULL; 2305 goto out; 2306 } 2307 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL); 2308 2309 if (bpf_sk_storage_clone(sk, newsk)) { 2310 sk_free_unlock_clone(newsk); 2311 newsk = NULL; 2312 goto out; 2313 } 2314 2315 /* Clear sk_user_data if parent had the pointer tagged 2316 * as not suitable for copying when cloning. 2317 */ 2318 if (sk_user_data_is_nocopy(newsk)) 2319 newsk->sk_user_data = NULL; 2320 2321 newsk->sk_err = 0; 2322 newsk->sk_err_soft = 0; 2323 newsk->sk_priority = 0; 2324 newsk->sk_incoming_cpu = raw_smp_processor_id(); 2325 2326 /* Before updating sk_refcnt, we must commit prior changes to memory 2327 * (Documentation/RCU/rculist_nulls.rst for details) 2328 */ 2329 smp_wmb(); 2330 refcount_set(&newsk->sk_refcnt, 2); 2331 2332 sk_set_socket(newsk, NULL); 2333 sk_tx_queue_clear(newsk); 2334 RCU_INIT_POINTER(newsk->sk_wq, NULL); 2335 2336 if (newsk->sk_prot->sockets_allocated) 2337 sk_sockets_allocated_inc(newsk); 2338 2339 if (sock_needs_netstamp(sk) && newsk->sk_flags & SK_FLAGS_TIMESTAMP) 2340 net_enable_timestamp(); 2341 out: 2342 return newsk; 2343 } 2344 EXPORT_SYMBOL_GPL(sk_clone_lock); 2345 2346 void sk_free_unlock_clone(struct sock *sk) 2347 { 2348 /* It is still raw copy of parent, so invalidate 2349 * destructor and make plain sk_free() */ 2350 sk->sk_destruct = NULL; 2351 bh_unlock_sock(sk); 2352 sk_free(sk); 2353 } 2354 EXPORT_SYMBOL_GPL(sk_free_unlock_clone); 2355 2356 static u32 sk_dst_gso_max_size(struct sock *sk, struct dst_entry *dst) 2357 { 2358 bool is_ipv6 = false; 2359 u32 max_size; 2360 2361 #if IS_ENABLED(CONFIG_IPV6) 2362 is_ipv6 = (sk->sk_family == AF_INET6 && 2363 !ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr)); 2364 #endif 2365 /* pairs with the WRITE_ONCE() in netif_set_gso(_ipv4)_max_size() */ 2366 max_size = is_ipv6 ? READ_ONCE(dst->dev->gso_max_size) : 2367 READ_ONCE(dst->dev->gso_ipv4_max_size); 2368 if (max_size > GSO_LEGACY_MAX_SIZE && !sk_is_tcp(sk)) 2369 max_size = GSO_LEGACY_MAX_SIZE; 2370 2371 return max_size - (MAX_TCP_HEADER + 1); 2372 } 2373 2374 void sk_setup_caps(struct sock *sk, struct dst_entry *dst) 2375 { 2376 u32 max_segs = 1; 2377 2378 sk->sk_route_caps = dst->dev->features; 2379 if (sk_is_tcp(sk)) 2380 sk->sk_route_caps |= NETIF_F_GSO; 2381 if (sk->sk_route_caps & NETIF_F_GSO) 2382 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE; 2383 if (unlikely(sk->sk_gso_disabled)) 2384 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 2385 if (sk_can_gso(sk)) { 2386 if (dst->header_len && !xfrm_dst_offload_ok(dst)) { 2387 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 2388 } else { 2389 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM; 2390 sk->sk_gso_max_size = sk_dst_gso_max_size(sk, dst); 2391 /* pairs with the WRITE_ONCE() in netif_set_gso_max_segs() */ 2392 max_segs = max_t(u32, READ_ONCE(dst->dev->gso_max_segs), 1); 2393 } 2394 } 2395 sk->sk_gso_max_segs = max_segs; 2396 sk_dst_set(sk, dst); 2397 } 2398 EXPORT_SYMBOL_GPL(sk_setup_caps); 2399 2400 /* 2401 * Simple resource managers for sockets. 2402 */ 2403 2404 2405 /* 2406 * Write buffer destructor automatically called from kfree_skb. 2407 */ 2408 void sock_wfree(struct sk_buff *skb) 2409 { 2410 struct sock *sk = skb->sk; 2411 unsigned int len = skb->truesize; 2412 bool free; 2413 2414 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) { 2415 if (sock_flag(sk, SOCK_RCU_FREE) && 2416 sk->sk_write_space == sock_def_write_space) { 2417 rcu_read_lock(); 2418 free = refcount_sub_and_test(len, &sk->sk_wmem_alloc); 2419 sock_def_write_space_wfree(sk); 2420 rcu_read_unlock(); 2421 if (unlikely(free)) 2422 __sk_free(sk); 2423 return; 2424 } 2425 2426 /* 2427 * Keep a reference on sk_wmem_alloc, this will be released 2428 * after sk_write_space() call 2429 */ 2430 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc)); 2431 sk->sk_write_space(sk); 2432 len = 1; 2433 } 2434 /* 2435 * if sk_wmem_alloc reaches 0, we must finish what sk_free() 2436 * could not do because of in-flight packets 2437 */ 2438 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc)) 2439 __sk_free(sk); 2440 } 2441 EXPORT_SYMBOL(sock_wfree); 2442 2443 /* This variant of sock_wfree() is used by TCP, 2444 * since it sets SOCK_USE_WRITE_QUEUE. 2445 */ 2446 void __sock_wfree(struct sk_buff *skb) 2447 { 2448 struct sock *sk = skb->sk; 2449 2450 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc)) 2451 __sk_free(sk); 2452 } 2453 2454 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk) 2455 { 2456 skb_orphan(skb); 2457 skb->sk = sk; 2458 #ifdef CONFIG_INET 2459 if (unlikely(!sk_fullsock(sk))) { 2460 skb->destructor = sock_edemux; 2461 sock_hold(sk); 2462 return; 2463 } 2464 #endif 2465 skb->destructor = sock_wfree; 2466 skb_set_hash_from_sk(skb, sk); 2467 /* 2468 * We used to take a refcount on sk, but following operation 2469 * is enough to guarantee sk_free() wont free this sock until 2470 * all in-flight packets are completed 2471 */ 2472 refcount_add(skb->truesize, &sk->sk_wmem_alloc); 2473 } 2474 EXPORT_SYMBOL(skb_set_owner_w); 2475 2476 static bool can_skb_orphan_partial(const struct sk_buff *skb) 2477 { 2478 #ifdef CONFIG_TLS_DEVICE 2479 /* Drivers depend on in-order delivery for crypto offload, 2480 * partial orphan breaks out-of-order-OK logic. 2481 */ 2482 if (skb->decrypted) 2483 return false; 2484 #endif 2485 return (skb->destructor == sock_wfree || 2486 (IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree)); 2487 } 2488 2489 /* This helper is used by netem, as it can hold packets in its 2490 * delay queue. We want to allow the owner socket to send more 2491 * packets, as if they were already TX completed by a typical driver. 2492 * But we also want to keep skb->sk set because some packet schedulers 2493 * rely on it (sch_fq for example). 2494 */ 2495 void skb_orphan_partial(struct sk_buff *skb) 2496 { 2497 if (skb_is_tcp_pure_ack(skb)) 2498 return; 2499 2500 if (can_skb_orphan_partial(skb) && skb_set_owner_sk_safe(skb, skb->sk)) 2501 return; 2502 2503 skb_orphan(skb); 2504 } 2505 EXPORT_SYMBOL(skb_orphan_partial); 2506 2507 /* 2508 * Read buffer destructor automatically called from kfree_skb. 2509 */ 2510 void sock_rfree(struct sk_buff *skb) 2511 { 2512 struct sock *sk = skb->sk; 2513 unsigned int len = skb->truesize; 2514 2515 atomic_sub(len, &sk->sk_rmem_alloc); 2516 sk_mem_uncharge(sk, len); 2517 } 2518 EXPORT_SYMBOL(sock_rfree); 2519 2520 /* 2521 * Buffer destructor for skbs that are not used directly in read or write 2522 * path, e.g. for error handler skbs. Automatically called from kfree_skb. 2523 */ 2524 void sock_efree(struct sk_buff *skb) 2525 { 2526 sock_put(skb->sk); 2527 } 2528 EXPORT_SYMBOL(sock_efree); 2529 2530 /* Buffer destructor for prefetch/receive path where reference count may 2531 * not be held, e.g. for listen sockets. 2532 */ 2533 #ifdef CONFIG_INET 2534 void sock_pfree(struct sk_buff *skb) 2535 { 2536 if (sk_is_refcounted(skb->sk)) 2537 sock_gen_put(skb->sk); 2538 } 2539 EXPORT_SYMBOL(sock_pfree); 2540 #endif /* CONFIG_INET */ 2541 2542 kuid_t sock_i_uid(struct sock *sk) 2543 { 2544 kuid_t uid; 2545 2546 read_lock_bh(&sk->sk_callback_lock); 2547 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID; 2548 read_unlock_bh(&sk->sk_callback_lock); 2549 return uid; 2550 } 2551 EXPORT_SYMBOL(sock_i_uid); 2552 2553 unsigned long sock_i_ino(struct sock *sk) 2554 { 2555 unsigned long ino; 2556 2557 read_lock_bh(&sk->sk_callback_lock); 2558 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0; 2559 read_unlock_bh(&sk->sk_callback_lock); 2560 return ino; 2561 } 2562 EXPORT_SYMBOL(sock_i_ino); 2563 2564 /* 2565 * Allocate a skb from the socket's send buffer. 2566 */ 2567 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 2568 gfp_t priority) 2569 { 2570 if (force || 2571 refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) { 2572 struct sk_buff *skb = alloc_skb(size, priority); 2573 2574 if (skb) { 2575 skb_set_owner_w(skb, sk); 2576 return skb; 2577 } 2578 } 2579 return NULL; 2580 } 2581 EXPORT_SYMBOL(sock_wmalloc); 2582 2583 static void sock_ofree(struct sk_buff *skb) 2584 { 2585 struct sock *sk = skb->sk; 2586 2587 atomic_sub(skb->truesize, &sk->sk_omem_alloc); 2588 } 2589 2590 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size, 2591 gfp_t priority) 2592 { 2593 struct sk_buff *skb; 2594 2595 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */ 2596 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) > 2597 READ_ONCE(sysctl_optmem_max)) 2598 return NULL; 2599 2600 skb = alloc_skb(size, priority); 2601 if (!skb) 2602 return NULL; 2603 2604 atomic_add(skb->truesize, &sk->sk_omem_alloc); 2605 skb->sk = sk; 2606 skb->destructor = sock_ofree; 2607 return skb; 2608 } 2609 2610 /* 2611 * Allocate a memory block from the socket's option memory buffer. 2612 */ 2613 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority) 2614 { 2615 int optmem_max = READ_ONCE(sysctl_optmem_max); 2616 2617 if ((unsigned int)size <= optmem_max && 2618 atomic_read(&sk->sk_omem_alloc) + size < optmem_max) { 2619 void *mem; 2620 /* First do the add, to avoid the race if kmalloc 2621 * might sleep. 2622 */ 2623 atomic_add(size, &sk->sk_omem_alloc); 2624 mem = kmalloc(size, priority); 2625 if (mem) 2626 return mem; 2627 atomic_sub(size, &sk->sk_omem_alloc); 2628 } 2629 return NULL; 2630 } 2631 EXPORT_SYMBOL(sock_kmalloc); 2632 2633 /* Free an option memory block. Note, we actually want the inline 2634 * here as this allows gcc to detect the nullify and fold away the 2635 * condition entirely. 2636 */ 2637 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size, 2638 const bool nullify) 2639 { 2640 if (WARN_ON_ONCE(!mem)) 2641 return; 2642 if (nullify) 2643 kfree_sensitive(mem); 2644 else 2645 kfree(mem); 2646 atomic_sub(size, &sk->sk_omem_alloc); 2647 } 2648 2649 void sock_kfree_s(struct sock *sk, void *mem, int size) 2650 { 2651 __sock_kfree_s(sk, mem, size, false); 2652 } 2653 EXPORT_SYMBOL(sock_kfree_s); 2654 2655 void sock_kzfree_s(struct sock *sk, void *mem, int size) 2656 { 2657 __sock_kfree_s(sk, mem, size, true); 2658 } 2659 EXPORT_SYMBOL(sock_kzfree_s); 2660 2661 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock. 2662 I think, these locks should be removed for datagram sockets. 2663 */ 2664 static long sock_wait_for_wmem(struct sock *sk, long timeo) 2665 { 2666 DEFINE_WAIT(wait); 2667 2668 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk); 2669 for (;;) { 2670 if (!timeo) 2671 break; 2672 if (signal_pending(current)) 2673 break; 2674 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 2675 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 2676 if (refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) 2677 break; 2678 if (sk->sk_shutdown & SEND_SHUTDOWN) 2679 break; 2680 if (sk->sk_err) 2681 break; 2682 timeo = schedule_timeout(timeo); 2683 } 2684 finish_wait(sk_sleep(sk), &wait); 2685 return timeo; 2686 } 2687 2688 2689 /* 2690 * Generic send/receive buffer handlers 2691 */ 2692 2693 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 2694 unsigned long data_len, int noblock, 2695 int *errcode, int max_page_order) 2696 { 2697 struct sk_buff *skb; 2698 long timeo; 2699 int err; 2700 2701 timeo = sock_sndtimeo(sk, noblock); 2702 for (;;) { 2703 err = sock_error(sk); 2704 if (err != 0) 2705 goto failure; 2706 2707 err = -EPIPE; 2708 if (sk->sk_shutdown & SEND_SHUTDOWN) 2709 goto failure; 2710 2711 if (sk_wmem_alloc_get(sk) < READ_ONCE(sk->sk_sndbuf)) 2712 break; 2713 2714 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk); 2715 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 2716 err = -EAGAIN; 2717 if (!timeo) 2718 goto failure; 2719 if (signal_pending(current)) 2720 goto interrupted; 2721 timeo = sock_wait_for_wmem(sk, timeo); 2722 } 2723 skb = alloc_skb_with_frags(header_len, data_len, max_page_order, 2724 errcode, sk->sk_allocation); 2725 if (skb) 2726 skb_set_owner_w(skb, sk); 2727 return skb; 2728 2729 interrupted: 2730 err = sock_intr_errno(timeo); 2731 failure: 2732 *errcode = err; 2733 return NULL; 2734 } 2735 EXPORT_SYMBOL(sock_alloc_send_pskb); 2736 2737 int __sock_cmsg_send(struct sock *sk, struct cmsghdr *cmsg, 2738 struct sockcm_cookie *sockc) 2739 { 2740 u32 tsflags; 2741 2742 switch (cmsg->cmsg_type) { 2743 case SO_MARK: 2744 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) && 2745 !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) 2746 return -EPERM; 2747 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 2748 return -EINVAL; 2749 sockc->mark = *(u32 *)CMSG_DATA(cmsg); 2750 break; 2751 case SO_TIMESTAMPING_OLD: 2752 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 2753 return -EINVAL; 2754 2755 tsflags = *(u32 *)CMSG_DATA(cmsg); 2756 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK) 2757 return -EINVAL; 2758 2759 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK; 2760 sockc->tsflags |= tsflags; 2761 break; 2762 case SCM_TXTIME: 2763 if (!sock_flag(sk, SOCK_TXTIME)) 2764 return -EINVAL; 2765 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64))) 2766 return -EINVAL; 2767 sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg)); 2768 break; 2769 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */ 2770 case SCM_RIGHTS: 2771 case SCM_CREDENTIALS: 2772 break; 2773 default: 2774 return -EINVAL; 2775 } 2776 return 0; 2777 } 2778 EXPORT_SYMBOL(__sock_cmsg_send); 2779 2780 int sock_cmsg_send(struct sock *sk, struct msghdr *msg, 2781 struct sockcm_cookie *sockc) 2782 { 2783 struct cmsghdr *cmsg; 2784 int ret; 2785 2786 for_each_cmsghdr(cmsg, msg) { 2787 if (!CMSG_OK(msg, cmsg)) 2788 return -EINVAL; 2789 if (cmsg->cmsg_level != SOL_SOCKET) 2790 continue; 2791 ret = __sock_cmsg_send(sk, cmsg, sockc); 2792 if (ret) 2793 return ret; 2794 } 2795 return 0; 2796 } 2797 EXPORT_SYMBOL(sock_cmsg_send); 2798 2799 static void sk_enter_memory_pressure(struct sock *sk) 2800 { 2801 if (!sk->sk_prot->enter_memory_pressure) 2802 return; 2803 2804 sk->sk_prot->enter_memory_pressure(sk); 2805 } 2806 2807 static void sk_leave_memory_pressure(struct sock *sk) 2808 { 2809 if (sk->sk_prot->leave_memory_pressure) { 2810 INDIRECT_CALL_INET_1(sk->sk_prot->leave_memory_pressure, 2811 tcp_leave_memory_pressure, sk); 2812 } else { 2813 unsigned long *memory_pressure = sk->sk_prot->memory_pressure; 2814 2815 if (memory_pressure && READ_ONCE(*memory_pressure)) 2816 WRITE_ONCE(*memory_pressure, 0); 2817 } 2818 } 2819 2820 DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key); 2821 2822 /** 2823 * skb_page_frag_refill - check that a page_frag contains enough room 2824 * @sz: minimum size of the fragment we want to get 2825 * @pfrag: pointer to page_frag 2826 * @gfp: priority for memory allocation 2827 * 2828 * Note: While this allocator tries to use high order pages, there is 2829 * no guarantee that allocations succeed. Therefore, @sz MUST be 2830 * less or equal than PAGE_SIZE. 2831 */ 2832 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp) 2833 { 2834 if (pfrag->page) { 2835 if (page_ref_count(pfrag->page) == 1) { 2836 pfrag->offset = 0; 2837 return true; 2838 } 2839 if (pfrag->offset + sz <= pfrag->size) 2840 return true; 2841 put_page(pfrag->page); 2842 } 2843 2844 pfrag->offset = 0; 2845 if (SKB_FRAG_PAGE_ORDER && 2846 !static_branch_unlikely(&net_high_order_alloc_disable_key)) { 2847 /* Avoid direct reclaim but allow kswapd to wake */ 2848 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) | 2849 __GFP_COMP | __GFP_NOWARN | 2850 __GFP_NORETRY, 2851 SKB_FRAG_PAGE_ORDER); 2852 if (likely(pfrag->page)) { 2853 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER; 2854 return true; 2855 } 2856 } 2857 pfrag->page = alloc_page(gfp); 2858 if (likely(pfrag->page)) { 2859 pfrag->size = PAGE_SIZE; 2860 return true; 2861 } 2862 return false; 2863 } 2864 EXPORT_SYMBOL(skb_page_frag_refill); 2865 2866 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag) 2867 { 2868 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation))) 2869 return true; 2870 2871 sk_enter_memory_pressure(sk); 2872 sk_stream_moderate_sndbuf(sk); 2873 return false; 2874 } 2875 EXPORT_SYMBOL(sk_page_frag_refill); 2876 2877 void __lock_sock(struct sock *sk) 2878 __releases(&sk->sk_lock.slock) 2879 __acquires(&sk->sk_lock.slock) 2880 { 2881 DEFINE_WAIT(wait); 2882 2883 for (;;) { 2884 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait, 2885 TASK_UNINTERRUPTIBLE); 2886 spin_unlock_bh(&sk->sk_lock.slock); 2887 schedule(); 2888 spin_lock_bh(&sk->sk_lock.slock); 2889 if (!sock_owned_by_user(sk)) 2890 break; 2891 } 2892 finish_wait(&sk->sk_lock.wq, &wait); 2893 } 2894 2895 void __release_sock(struct sock *sk) 2896 __releases(&sk->sk_lock.slock) 2897 __acquires(&sk->sk_lock.slock) 2898 { 2899 struct sk_buff *skb, *next; 2900 2901 while ((skb = sk->sk_backlog.head) != NULL) { 2902 sk->sk_backlog.head = sk->sk_backlog.tail = NULL; 2903 2904 spin_unlock_bh(&sk->sk_lock.slock); 2905 2906 do { 2907 next = skb->next; 2908 prefetch(next); 2909 DEBUG_NET_WARN_ON_ONCE(skb_dst_is_noref(skb)); 2910 skb_mark_not_on_list(skb); 2911 sk_backlog_rcv(sk, skb); 2912 2913 cond_resched(); 2914 2915 skb = next; 2916 } while (skb != NULL); 2917 2918 spin_lock_bh(&sk->sk_lock.slock); 2919 } 2920 2921 /* 2922 * Doing the zeroing here guarantee we can not loop forever 2923 * while a wild producer attempts to flood us. 2924 */ 2925 sk->sk_backlog.len = 0; 2926 } 2927 2928 void __sk_flush_backlog(struct sock *sk) 2929 { 2930 spin_lock_bh(&sk->sk_lock.slock); 2931 __release_sock(sk); 2932 spin_unlock_bh(&sk->sk_lock.slock); 2933 } 2934 EXPORT_SYMBOL_GPL(__sk_flush_backlog); 2935 2936 /** 2937 * sk_wait_data - wait for data to arrive at sk_receive_queue 2938 * @sk: sock to wait on 2939 * @timeo: for how long 2940 * @skb: last skb seen on sk_receive_queue 2941 * 2942 * Now socket state including sk->sk_err is changed only under lock, 2943 * hence we may omit checks after joining wait queue. 2944 * We check receive queue before schedule() only as optimization; 2945 * it is very likely that release_sock() added new data. 2946 */ 2947 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb) 2948 { 2949 DEFINE_WAIT_FUNC(wait, woken_wake_function); 2950 int rc; 2951 2952 add_wait_queue(sk_sleep(sk), &wait); 2953 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk); 2954 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait); 2955 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk); 2956 remove_wait_queue(sk_sleep(sk), &wait); 2957 return rc; 2958 } 2959 EXPORT_SYMBOL(sk_wait_data); 2960 2961 /** 2962 * __sk_mem_raise_allocated - increase memory_allocated 2963 * @sk: socket 2964 * @size: memory size to allocate 2965 * @amt: pages to allocate 2966 * @kind: allocation type 2967 * 2968 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc 2969 */ 2970 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind) 2971 { 2972 bool memcg_charge = mem_cgroup_sockets_enabled && sk->sk_memcg; 2973 struct proto *prot = sk->sk_prot; 2974 bool charged = true; 2975 long allocated; 2976 2977 sk_memory_allocated_add(sk, amt); 2978 allocated = sk_memory_allocated(sk); 2979 if (memcg_charge && 2980 !(charged = mem_cgroup_charge_skmem(sk->sk_memcg, amt, 2981 gfp_memcg_charge()))) 2982 goto suppress_allocation; 2983 2984 /* Under limit. */ 2985 if (allocated <= sk_prot_mem_limits(sk, 0)) { 2986 sk_leave_memory_pressure(sk); 2987 return 1; 2988 } 2989 2990 /* Under pressure. */ 2991 if (allocated > sk_prot_mem_limits(sk, 1)) 2992 sk_enter_memory_pressure(sk); 2993 2994 /* Over hard limit. */ 2995 if (allocated > sk_prot_mem_limits(sk, 2)) 2996 goto suppress_allocation; 2997 2998 /* guarantee minimum buffer size under pressure */ 2999 if (kind == SK_MEM_RECV) { 3000 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot)) 3001 return 1; 3002 3003 } else { /* SK_MEM_SEND */ 3004 int wmem0 = sk_get_wmem0(sk, prot); 3005 3006 if (sk->sk_type == SOCK_STREAM) { 3007 if (sk->sk_wmem_queued < wmem0) 3008 return 1; 3009 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) { 3010 return 1; 3011 } 3012 } 3013 3014 if (sk_has_memory_pressure(sk)) { 3015 u64 alloc; 3016 3017 if (!sk_under_memory_pressure(sk)) 3018 return 1; 3019 alloc = sk_sockets_allocated_read_positive(sk); 3020 if (sk_prot_mem_limits(sk, 2) > alloc * 3021 sk_mem_pages(sk->sk_wmem_queued + 3022 atomic_read(&sk->sk_rmem_alloc) + 3023 sk->sk_forward_alloc)) 3024 return 1; 3025 } 3026 3027 suppress_allocation: 3028 3029 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) { 3030 sk_stream_moderate_sndbuf(sk); 3031 3032 /* Fail only if socket is _under_ its sndbuf. 3033 * In this case we cannot block, so that we have to fail. 3034 */ 3035 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf) { 3036 /* Force charge with __GFP_NOFAIL */ 3037 if (memcg_charge && !charged) { 3038 mem_cgroup_charge_skmem(sk->sk_memcg, amt, 3039 gfp_memcg_charge() | __GFP_NOFAIL); 3040 } 3041 return 1; 3042 } 3043 } 3044 3045 if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged)) 3046 trace_sock_exceed_buf_limit(sk, prot, allocated, kind); 3047 3048 sk_memory_allocated_sub(sk, amt); 3049 3050 if (memcg_charge && charged) 3051 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt); 3052 3053 return 0; 3054 } 3055 3056 /** 3057 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated 3058 * @sk: socket 3059 * @size: memory size to allocate 3060 * @kind: allocation type 3061 * 3062 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means 3063 * rmem allocation. This function assumes that protocols which have 3064 * memory_pressure use sk_wmem_queued as write buffer accounting. 3065 */ 3066 int __sk_mem_schedule(struct sock *sk, int size, int kind) 3067 { 3068 int ret, amt = sk_mem_pages(size); 3069 3070 sk->sk_forward_alloc += amt << PAGE_SHIFT; 3071 ret = __sk_mem_raise_allocated(sk, size, amt, kind); 3072 if (!ret) 3073 sk->sk_forward_alloc -= amt << PAGE_SHIFT; 3074 return ret; 3075 } 3076 EXPORT_SYMBOL(__sk_mem_schedule); 3077 3078 /** 3079 * __sk_mem_reduce_allocated - reclaim memory_allocated 3080 * @sk: socket 3081 * @amount: number of quanta 3082 * 3083 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc 3084 */ 3085 void __sk_mem_reduce_allocated(struct sock *sk, int amount) 3086 { 3087 sk_memory_allocated_sub(sk, amount); 3088 3089 if (mem_cgroup_sockets_enabled && sk->sk_memcg) 3090 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount); 3091 3092 if (sk_under_memory_pressure(sk) && 3093 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0))) 3094 sk_leave_memory_pressure(sk); 3095 } 3096 3097 /** 3098 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated 3099 * @sk: socket 3100 * @amount: number of bytes (rounded down to a PAGE_SIZE multiple) 3101 */ 3102 void __sk_mem_reclaim(struct sock *sk, int amount) 3103 { 3104 amount >>= PAGE_SHIFT; 3105 sk->sk_forward_alloc -= amount << PAGE_SHIFT; 3106 __sk_mem_reduce_allocated(sk, amount); 3107 } 3108 EXPORT_SYMBOL(__sk_mem_reclaim); 3109 3110 int sk_set_peek_off(struct sock *sk, int val) 3111 { 3112 sk->sk_peek_off = val; 3113 return 0; 3114 } 3115 EXPORT_SYMBOL_GPL(sk_set_peek_off); 3116 3117 /* 3118 * Set of default routines for initialising struct proto_ops when 3119 * the protocol does not support a particular function. In certain 3120 * cases where it makes no sense for a protocol to have a "do nothing" 3121 * function, some default processing is provided. 3122 */ 3123 3124 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len) 3125 { 3126 return -EOPNOTSUPP; 3127 } 3128 EXPORT_SYMBOL(sock_no_bind); 3129 3130 int sock_no_connect(struct socket *sock, struct sockaddr *saddr, 3131 int len, int flags) 3132 { 3133 return -EOPNOTSUPP; 3134 } 3135 EXPORT_SYMBOL(sock_no_connect); 3136 3137 int sock_no_socketpair(struct socket *sock1, struct socket *sock2) 3138 { 3139 return -EOPNOTSUPP; 3140 } 3141 EXPORT_SYMBOL(sock_no_socketpair); 3142 3143 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags, 3144 bool kern) 3145 { 3146 return -EOPNOTSUPP; 3147 } 3148 EXPORT_SYMBOL(sock_no_accept); 3149 3150 int sock_no_getname(struct socket *sock, struct sockaddr *saddr, 3151 int peer) 3152 { 3153 return -EOPNOTSUPP; 3154 } 3155 EXPORT_SYMBOL(sock_no_getname); 3156 3157 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 3158 { 3159 return -EOPNOTSUPP; 3160 } 3161 EXPORT_SYMBOL(sock_no_ioctl); 3162 3163 int sock_no_listen(struct socket *sock, int backlog) 3164 { 3165 return -EOPNOTSUPP; 3166 } 3167 EXPORT_SYMBOL(sock_no_listen); 3168 3169 int sock_no_shutdown(struct socket *sock, int how) 3170 { 3171 return -EOPNOTSUPP; 3172 } 3173 EXPORT_SYMBOL(sock_no_shutdown); 3174 3175 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len) 3176 { 3177 return -EOPNOTSUPP; 3178 } 3179 EXPORT_SYMBOL(sock_no_sendmsg); 3180 3181 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len) 3182 { 3183 return -EOPNOTSUPP; 3184 } 3185 EXPORT_SYMBOL(sock_no_sendmsg_locked); 3186 3187 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len, 3188 int flags) 3189 { 3190 return -EOPNOTSUPP; 3191 } 3192 EXPORT_SYMBOL(sock_no_recvmsg); 3193 3194 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma) 3195 { 3196 /* Mirror missing mmap method error code */ 3197 return -ENODEV; 3198 } 3199 EXPORT_SYMBOL(sock_no_mmap); 3200 3201 /* 3202 * When a file is received (via SCM_RIGHTS, etc), we must bump the 3203 * various sock-based usage counts. 3204 */ 3205 void __receive_sock(struct file *file) 3206 { 3207 struct socket *sock; 3208 3209 sock = sock_from_file(file); 3210 if (sock) { 3211 sock_update_netprioidx(&sock->sk->sk_cgrp_data); 3212 sock_update_classid(&sock->sk->sk_cgrp_data); 3213 } 3214 } 3215 3216 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags) 3217 { 3218 ssize_t res; 3219 struct msghdr msg = {.msg_flags = flags}; 3220 struct kvec iov; 3221 char *kaddr = kmap(page); 3222 iov.iov_base = kaddr + offset; 3223 iov.iov_len = size; 3224 res = kernel_sendmsg(sock, &msg, &iov, 1, size); 3225 kunmap(page); 3226 return res; 3227 } 3228 EXPORT_SYMBOL(sock_no_sendpage); 3229 3230 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page, 3231 int offset, size_t size, int flags) 3232 { 3233 ssize_t res; 3234 struct msghdr msg = {.msg_flags = flags}; 3235 struct kvec iov; 3236 char *kaddr = kmap(page); 3237 3238 iov.iov_base = kaddr + offset; 3239 iov.iov_len = size; 3240 res = kernel_sendmsg_locked(sk, &msg, &iov, 1, size); 3241 kunmap(page); 3242 return res; 3243 } 3244 EXPORT_SYMBOL(sock_no_sendpage_locked); 3245 3246 /* 3247 * Default Socket Callbacks 3248 */ 3249 3250 static void sock_def_wakeup(struct sock *sk) 3251 { 3252 struct socket_wq *wq; 3253 3254 rcu_read_lock(); 3255 wq = rcu_dereference(sk->sk_wq); 3256 if (skwq_has_sleeper(wq)) 3257 wake_up_interruptible_all(&wq->wait); 3258 rcu_read_unlock(); 3259 } 3260 3261 static void sock_def_error_report(struct sock *sk) 3262 { 3263 struct socket_wq *wq; 3264 3265 rcu_read_lock(); 3266 wq = rcu_dereference(sk->sk_wq); 3267 if (skwq_has_sleeper(wq)) 3268 wake_up_interruptible_poll(&wq->wait, EPOLLERR); 3269 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR); 3270 rcu_read_unlock(); 3271 } 3272 3273 void sock_def_readable(struct sock *sk) 3274 { 3275 struct socket_wq *wq; 3276 3277 trace_sk_data_ready(sk); 3278 3279 rcu_read_lock(); 3280 wq = rcu_dereference(sk->sk_wq); 3281 if (skwq_has_sleeper(wq)) 3282 wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI | 3283 EPOLLRDNORM | EPOLLRDBAND); 3284 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN); 3285 rcu_read_unlock(); 3286 } 3287 3288 static void sock_def_write_space(struct sock *sk) 3289 { 3290 struct socket_wq *wq; 3291 3292 rcu_read_lock(); 3293 3294 /* Do not wake up a writer until he can make "significant" 3295 * progress. --DaveM 3296 */ 3297 if (sock_writeable(sk)) { 3298 wq = rcu_dereference(sk->sk_wq); 3299 if (skwq_has_sleeper(wq)) 3300 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT | 3301 EPOLLWRNORM | EPOLLWRBAND); 3302 3303 /* Should agree with poll, otherwise some programs break */ 3304 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT); 3305 } 3306 3307 rcu_read_unlock(); 3308 } 3309 3310 /* An optimised version of sock_def_write_space(), should only be called 3311 * for SOCK_RCU_FREE sockets under RCU read section and after putting 3312 * ->sk_wmem_alloc. 3313 */ 3314 static void sock_def_write_space_wfree(struct sock *sk) 3315 { 3316 /* Do not wake up a writer until he can make "significant" 3317 * progress. --DaveM 3318 */ 3319 if (sock_writeable(sk)) { 3320 struct socket_wq *wq = rcu_dereference(sk->sk_wq); 3321 3322 /* rely on refcount_sub from sock_wfree() */ 3323 smp_mb__after_atomic(); 3324 if (wq && waitqueue_active(&wq->wait)) 3325 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT | 3326 EPOLLWRNORM | EPOLLWRBAND); 3327 3328 /* Should agree with poll, otherwise some programs break */ 3329 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT); 3330 } 3331 } 3332 3333 static void sock_def_destruct(struct sock *sk) 3334 { 3335 } 3336 3337 void sk_send_sigurg(struct sock *sk) 3338 { 3339 if (sk->sk_socket && sk->sk_socket->file) 3340 if (send_sigurg(&sk->sk_socket->file->f_owner)) 3341 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI); 3342 } 3343 EXPORT_SYMBOL(sk_send_sigurg); 3344 3345 void sk_reset_timer(struct sock *sk, struct timer_list* timer, 3346 unsigned long expires) 3347 { 3348 if (!mod_timer(timer, expires)) 3349 sock_hold(sk); 3350 } 3351 EXPORT_SYMBOL(sk_reset_timer); 3352 3353 void sk_stop_timer(struct sock *sk, struct timer_list* timer) 3354 { 3355 if (del_timer(timer)) 3356 __sock_put(sk); 3357 } 3358 EXPORT_SYMBOL(sk_stop_timer); 3359 3360 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer) 3361 { 3362 if (del_timer_sync(timer)) 3363 __sock_put(sk); 3364 } 3365 EXPORT_SYMBOL(sk_stop_timer_sync); 3366 3367 void sock_init_data_uid(struct socket *sock, struct sock *sk, kuid_t uid) 3368 { 3369 sk_init_common(sk); 3370 sk->sk_send_head = NULL; 3371 3372 timer_setup(&sk->sk_timer, NULL, 0); 3373 3374 sk->sk_allocation = GFP_KERNEL; 3375 sk->sk_rcvbuf = READ_ONCE(sysctl_rmem_default); 3376 sk->sk_sndbuf = READ_ONCE(sysctl_wmem_default); 3377 sk->sk_state = TCP_CLOSE; 3378 sk->sk_use_task_frag = true; 3379 sk_set_socket(sk, sock); 3380 3381 sock_set_flag(sk, SOCK_ZAPPED); 3382 3383 if (sock) { 3384 sk->sk_type = sock->type; 3385 RCU_INIT_POINTER(sk->sk_wq, &sock->wq); 3386 sock->sk = sk; 3387 } else { 3388 RCU_INIT_POINTER(sk->sk_wq, NULL); 3389 } 3390 sk->sk_uid = uid; 3391 3392 rwlock_init(&sk->sk_callback_lock); 3393 if (sk->sk_kern_sock) 3394 lockdep_set_class_and_name( 3395 &sk->sk_callback_lock, 3396 af_kern_callback_keys + sk->sk_family, 3397 af_family_kern_clock_key_strings[sk->sk_family]); 3398 else 3399 lockdep_set_class_and_name( 3400 &sk->sk_callback_lock, 3401 af_callback_keys + sk->sk_family, 3402 af_family_clock_key_strings[sk->sk_family]); 3403 3404 sk->sk_state_change = sock_def_wakeup; 3405 sk->sk_data_ready = sock_def_readable; 3406 sk->sk_write_space = sock_def_write_space; 3407 sk->sk_error_report = sock_def_error_report; 3408 sk->sk_destruct = sock_def_destruct; 3409 3410 sk->sk_frag.page = NULL; 3411 sk->sk_frag.offset = 0; 3412 sk->sk_peek_off = -1; 3413 3414 sk->sk_peer_pid = NULL; 3415 sk->sk_peer_cred = NULL; 3416 spin_lock_init(&sk->sk_peer_lock); 3417 3418 sk->sk_write_pending = 0; 3419 sk->sk_rcvlowat = 1; 3420 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; 3421 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT; 3422 3423 sk->sk_stamp = SK_DEFAULT_STAMP; 3424 #if BITS_PER_LONG==32 3425 seqlock_init(&sk->sk_stamp_seq); 3426 #endif 3427 atomic_set(&sk->sk_zckey, 0); 3428 3429 #ifdef CONFIG_NET_RX_BUSY_POLL 3430 sk->sk_napi_id = 0; 3431 sk->sk_ll_usec = READ_ONCE(sysctl_net_busy_read); 3432 #endif 3433 3434 sk->sk_max_pacing_rate = ~0UL; 3435 sk->sk_pacing_rate = ~0UL; 3436 WRITE_ONCE(sk->sk_pacing_shift, 10); 3437 sk->sk_incoming_cpu = -1; 3438 3439 sk_rx_queue_clear(sk); 3440 /* 3441 * Before updating sk_refcnt, we must commit prior changes to memory 3442 * (Documentation/RCU/rculist_nulls.rst for details) 3443 */ 3444 smp_wmb(); 3445 refcount_set(&sk->sk_refcnt, 1); 3446 atomic_set(&sk->sk_drops, 0); 3447 } 3448 EXPORT_SYMBOL(sock_init_data_uid); 3449 3450 void sock_init_data(struct socket *sock, struct sock *sk) 3451 { 3452 kuid_t uid = sock ? 3453 SOCK_INODE(sock)->i_uid : 3454 make_kuid(sock_net(sk)->user_ns, 0); 3455 3456 sock_init_data_uid(sock, sk, uid); 3457 } 3458 EXPORT_SYMBOL(sock_init_data); 3459 3460 void lock_sock_nested(struct sock *sk, int subclass) 3461 { 3462 /* The sk_lock has mutex_lock() semantics here. */ 3463 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_); 3464 3465 might_sleep(); 3466 spin_lock_bh(&sk->sk_lock.slock); 3467 if (sock_owned_by_user_nocheck(sk)) 3468 __lock_sock(sk); 3469 sk->sk_lock.owned = 1; 3470 spin_unlock_bh(&sk->sk_lock.slock); 3471 } 3472 EXPORT_SYMBOL(lock_sock_nested); 3473 3474 void release_sock(struct sock *sk) 3475 { 3476 spin_lock_bh(&sk->sk_lock.slock); 3477 if (sk->sk_backlog.tail) 3478 __release_sock(sk); 3479 3480 /* Warning : release_cb() might need to release sk ownership, 3481 * ie call sock_release_ownership(sk) before us. 3482 */ 3483 if (sk->sk_prot->release_cb) 3484 sk->sk_prot->release_cb(sk); 3485 3486 sock_release_ownership(sk); 3487 if (waitqueue_active(&sk->sk_lock.wq)) 3488 wake_up(&sk->sk_lock.wq); 3489 spin_unlock_bh(&sk->sk_lock.slock); 3490 } 3491 EXPORT_SYMBOL(release_sock); 3492 3493 bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock) 3494 { 3495 might_sleep(); 3496 spin_lock_bh(&sk->sk_lock.slock); 3497 3498 if (!sock_owned_by_user_nocheck(sk)) { 3499 /* 3500 * Fast path return with bottom halves disabled and 3501 * sock::sk_lock.slock held. 3502 * 3503 * The 'mutex' is not contended and holding 3504 * sock::sk_lock.slock prevents all other lockers to 3505 * proceed so the corresponding unlock_sock_fast() can 3506 * avoid the slow path of release_sock() completely and 3507 * just release slock. 3508 * 3509 * From a semantical POV this is equivalent to 'acquiring' 3510 * the 'mutex', hence the corresponding lockdep 3511 * mutex_release() has to happen in the fast path of 3512 * unlock_sock_fast(). 3513 */ 3514 return false; 3515 } 3516 3517 __lock_sock(sk); 3518 sk->sk_lock.owned = 1; 3519 __acquire(&sk->sk_lock.slock); 3520 spin_unlock_bh(&sk->sk_lock.slock); 3521 return true; 3522 } 3523 EXPORT_SYMBOL(__lock_sock_fast); 3524 3525 int sock_gettstamp(struct socket *sock, void __user *userstamp, 3526 bool timeval, bool time32) 3527 { 3528 struct sock *sk = sock->sk; 3529 struct timespec64 ts; 3530 3531 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 3532 ts = ktime_to_timespec64(sock_read_timestamp(sk)); 3533 if (ts.tv_sec == -1) 3534 return -ENOENT; 3535 if (ts.tv_sec == 0) { 3536 ktime_t kt = ktime_get_real(); 3537 sock_write_timestamp(sk, kt); 3538 ts = ktime_to_timespec64(kt); 3539 } 3540 3541 if (timeval) 3542 ts.tv_nsec /= 1000; 3543 3544 #ifdef CONFIG_COMPAT_32BIT_TIME 3545 if (time32) 3546 return put_old_timespec32(&ts, userstamp); 3547 #endif 3548 #ifdef CONFIG_SPARC64 3549 /* beware of padding in sparc64 timeval */ 3550 if (timeval && !in_compat_syscall()) { 3551 struct __kernel_old_timeval __user tv = { 3552 .tv_sec = ts.tv_sec, 3553 .tv_usec = ts.tv_nsec, 3554 }; 3555 if (copy_to_user(userstamp, &tv, sizeof(tv))) 3556 return -EFAULT; 3557 return 0; 3558 } 3559 #endif 3560 return put_timespec64(&ts, userstamp); 3561 } 3562 EXPORT_SYMBOL(sock_gettstamp); 3563 3564 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag) 3565 { 3566 if (!sock_flag(sk, flag)) { 3567 unsigned long previous_flags = sk->sk_flags; 3568 3569 sock_set_flag(sk, flag); 3570 /* 3571 * we just set one of the two flags which require net 3572 * time stamping, but time stamping might have been on 3573 * already because of the other one 3574 */ 3575 if (sock_needs_netstamp(sk) && 3576 !(previous_flags & SK_FLAGS_TIMESTAMP)) 3577 net_enable_timestamp(); 3578 } 3579 } 3580 3581 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, 3582 int level, int type) 3583 { 3584 struct sock_exterr_skb *serr; 3585 struct sk_buff *skb; 3586 int copied, err; 3587 3588 err = -EAGAIN; 3589 skb = sock_dequeue_err_skb(sk); 3590 if (skb == NULL) 3591 goto out; 3592 3593 copied = skb->len; 3594 if (copied > len) { 3595 msg->msg_flags |= MSG_TRUNC; 3596 copied = len; 3597 } 3598 err = skb_copy_datagram_msg(skb, 0, msg, copied); 3599 if (err) 3600 goto out_free_skb; 3601 3602 sock_recv_timestamp(msg, sk, skb); 3603 3604 serr = SKB_EXT_ERR(skb); 3605 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee); 3606 3607 msg->msg_flags |= MSG_ERRQUEUE; 3608 err = copied; 3609 3610 out_free_skb: 3611 kfree_skb(skb); 3612 out: 3613 return err; 3614 } 3615 EXPORT_SYMBOL(sock_recv_errqueue); 3616 3617 /* 3618 * Get a socket option on an socket. 3619 * 3620 * FIX: POSIX 1003.1g is very ambiguous here. It states that 3621 * asynchronous errors should be reported by getsockopt. We assume 3622 * this means if you specify SO_ERROR (otherwise whats the point of it). 3623 */ 3624 int sock_common_getsockopt(struct socket *sock, int level, int optname, 3625 char __user *optval, int __user *optlen) 3626 { 3627 struct sock *sk = sock->sk; 3628 3629 /* IPV6_ADDRFORM can change sk->sk_prot under us. */ 3630 return READ_ONCE(sk->sk_prot)->getsockopt(sk, level, optname, optval, optlen); 3631 } 3632 EXPORT_SYMBOL(sock_common_getsockopt); 3633 3634 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 3635 int flags) 3636 { 3637 struct sock *sk = sock->sk; 3638 int addr_len = 0; 3639 int err; 3640 3641 err = sk->sk_prot->recvmsg(sk, msg, size, flags, &addr_len); 3642 if (err >= 0) 3643 msg->msg_namelen = addr_len; 3644 return err; 3645 } 3646 EXPORT_SYMBOL(sock_common_recvmsg); 3647 3648 /* 3649 * Set socket options on an inet socket. 3650 */ 3651 int sock_common_setsockopt(struct socket *sock, int level, int optname, 3652 sockptr_t optval, unsigned int optlen) 3653 { 3654 struct sock *sk = sock->sk; 3655 3656 /* IPV6_ADDRFORM can change sk->sk_prot under us. */ 3657 return READ_ONCE(sk->sk_prot)->setsockopt(sk, level, optname, optval, optlen); 3658 } 3659 EXPORT_SYMBOL(sock_common_setsockopt); 3660 3661 void sk_common_release(struct sock *sk) 3662 { 3663 if (sk->sk_prot->destroy) 3664 sk->sk_prot->destroy(sk); 3665 3666 /* 3667 * Observation: when sk_common_release is called, processes have 3668 * no access to socket. But net still has. 3669 * Step one, detach it from networking: 3670 * 3671 * A. Remove from hash tables. 3672 */ 3673 3674 sk->sk_prot->unhash(sk); 3675 3676 /* 3677 * In this point socket cannot receive new packets, but it is possible 3678 * that some packets are in flight because some CPU runs receiver and 3679 * did hash table lookup before we unhashed socket. They will achieve 3680 * receive queue and will be purged by socket destructor. 3681 * 3682 * Also we still have packets pending on receive queue and probably, 3683 * our own packets waiting in device queues. sock_destroy will drain 3684 * receive queue, but transmitted packets will delay socket destruction 3685 * until the last reference will be released. 3686 */ 3687 3688 sock_orphan(sk); 3689 3690 xfrm_sk_free_policy(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