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/udp.h> 111#include <linux/init.h> 112#include <linux/highmem.h> 113#include <linux/user_namespace.h> 114#include <linux/static_key.h> 115#include <linux/memcontrol.h> 116#include <linux/prefetch.h> 117#include <linux/compat.h> 118#include <linux/mroute.h> 119#include <linux/mroute6.h> 120#include <linux/icmpv6.h> 121 122#include <linux/uaccess.h> 123 124#include <linux/netdevice.h> 125#include <net/protocol.h> 126#include <linux/skbuff.h> 127#include <net/net_namespace.h> 128#include <net/request_sock.h> 129#include <net/sock.h> 130#include <linux/net_tstamp.h> 131#include <net/xfrm.h> 132#include <linux/ipsec.h> 133#include <net/cls_cgroup.h> 134#include <net/netprio_cgroup.h> 135#include <linux/sock_diag.h> 136 137#include <linux/filter.h> 138#include <net/sock_reuseport.h> 139#include <net/bpf_sk_storage.h> 140 141#include <trace/events/sock.h> 142 143#include <net/tcp.h> 144#include <net/busy_poll.h> 145#include <net/phonet/phonet.h> 146 147#include <linux/ethtool.h> 148 149#include "dev.h" 150 151static DEFINE_MUTEX(proto_list_mutex); 152static LIST_HEAD(proto_list); 153 154static void sock_def_write_space_wfree(struct sock *sk); 155static void sock_def_write_space(struct sock *sk); 156 157/** 158 * sk_ns_capable - General socket capability test 159 * @sk: Socket to use a capability on or through 160 * @user_ns: The user namespace of the capability to use 161 * @cap: The capability to use 162 * 163 * Test to see if the opener of the socket had when the socket was 164 * created and the current process has the capability @cap in the user 165 * namespace @user_ns. 166 */ 167bool sk_ns_capable(const struct sock *sk, 168 struct user_namespace *user_ns, int cap) 169{ 170 return file_ns_capable(sk->sk_socket->file, user_ns, cap) && 171 ns_capable(user_ns, cap); 172} 173EXPORT_SYMBOL(sk_ns_capable); 174 175/** 176 * sk_capable - Socket global capability test 177 * @sk: Socket to use a capability on or through 178 * @cap: The global capability to use 179 * 180 * Test to see if the opener of the socket had when the socket was 181 * created and the current process has the capability @cap in all user 182 * namespaces. 183 */ 184bool sk_capable(const struct sock *sk, int cap) 185{ 186 return sk_ns_capable(sk, &init_user_ns, cap); 187} 188EXPORT_SYMBOL(sk_capable); 189 190/** 191 * sk_net_capable - Network namespace socket capability test 192 * @sk: Socket to use a capability on or through 193 * @cap: The capability to use 194 * 195 * Test to see if the opener of the socket had when the socket was created 196 * and the current process has the capability @cap over the network namespace 197 * the socket is a member of. 198 */ 199bool sk_net_capable(const struct sock *sk, int cap) 200{ 201 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap); 202} 203EXPORT_SYMBOL(sk_net_capable); 204 205/* 206 * Each address family might have different locking rules, so we have 207 * one slock key per address family and separate keys for internal and 208 * userspace sockets. 209 */ 210static struct lock_class_key af_family_keys[AF_MAX]; 211static struct lock_class_key af_family_kern_keys[AF_MAX]; 212static struct lock_class_key af_family_slock_keys[AF_MAX]; 213static struct lock_class_key af_family_kern_slock_keys[AF_MAX]; 214 215/* 216 * Make lock validator output more readable. (we pre-construct these 217 * strings build-time, so that runtime initialization of socket 218 * locks is fast): 219 */ 220 221#define _sock_locks(x) \ 222 x "AF_UNSPEC", x "AF_UNIX" , x "AF_INET" , \ 223 x "AF_AX25" , x "AF_IPX" , x "AF_APPLETALK", \ 224 x "AF_NETROM", x "AF_BRIDGE" , x "AF_ATMPVC" , \ 225 x "AF_X25" , x "AF_INET6" , x "AF_ROSE" , \ 226 x "AF_DECnet", x "AF_NETBEUI" , x "AF_SECURITY" , \ 227 x "AF_KEY" , x "AF_NETLINK" , x "AF_PACKET" , \ 228 x "AF_ASH" , x "AF_ECONET" , x "AF_ATMSVC" , \ 229 x "AF_RDS" , x "AF_SNA" , x "AF_IRDA" , \ 230 x "AF_PPPOX" , x "AF_WANPIPE" , x "AF_LLC" , \ 231 x "27" , x "28" , x "AF_CAN" , \ 232 x "AF_TIPC" , x "AF_BLUETOOTH", x "IUCV" , \ 233 x "AF_RXRPC" , x "AF_ISDN" , x "AF_PHONET" , \ 234 x "AF_IEEE802154", x "AF_CAIF" , x "AF_ALG" , \ 235 x "AF_NFC" , x "AF_VSOCK" , x "AF_KCM" , \ 236 x "AF_QIPCRTR", x "AF_SMC" , x "AF_XDP" , \ 237 x "AF_MCTP" , \ 238 x "AF_MAX" 239 240static const char *const af_family_key_strings[AF_MAX+1] = { 241 _sock_locks("sk_lock-") 242}; 243static const char *const af_family_slock_key_strings[AF_MAX+1] = { 244 _sock_locks("slock-") 245}; 246static const char *const af_family_clock_key_strings[AF_MAX+1] = { 247 _sock_locks("clock-") 248}; 249 250static const char *const af_family_kern_key_strings[AF_MAX+1] = { 251 _sock_locks("k-sk_lock-") 252}; 253static const char *const af_family_kern_slock_key_strings[AF_MAX+1] = { 254 _sock_locks("k-slock-") 255}; 256static const char *const af_family_kern_clock_key_strings[AF_MAX+1] = { 257 _sock_locks("k-clock-") 258}; 259static const char *const af_family_rlock_key_strings[AF_MAX+1] = { 260 _sock_locks("rlock-") 261}; 262static const char *const af_family_wlock_key_strings[AF_MAX+1] = { 263 _sock_locks("wlock-") 264}; 265static const char *const af_family_elock_key_strings[AF_MAX+1] = { 266 _sock_locks("elock-") 267}; 268 269/* 270 * sk_callback_lock and sk queues locking rules are per-address-family, 271 * so split the lock classes by using a per-AF key: 272 */ 273static struct lock_class_key af_callback_keys[AF_MAX]; 274static struct lock_class_key af_rlock_keys[AF_MAX]; 275static struct lock_class_key af_wlock_keys[AF_MAX]; 276static struct lock_class_key af_elock_keys[AF_MAX]; 277static struct lock_class_key af_kern_callback_keys[AF_MAX]; 278 279/* Run time adjustable parameters. */ 280__u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX; 281EXPORT_SYMBOL(sysctl_wmem_max); 282__u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX; 283EXPORT_SYMBOL(sysctl_rmem_max); 284__u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX; 285__u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX; 286 287/* Maximal space eaten by iovec or ancillary data plus some space */ 288int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512); 289EXPORT_SYMBOL(sysctl_optmem_max); 290 291int sysctl_tstamp_allow_data __read_mostly = 1; 292 293DEFINE_STATIC_KEY_FALSE(memalloc_socks_key); 294EXPORT_SYMBOL_GPL(memalloc_socks_key); 295 296/** 297 * sk_set_memalloc - sets %SOCK_MEMALLOC 298 * @sk: socket to set it on 299 * 300 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves. 301 * It's the responsibility of the admin to adjust min_free_kbytes 302 * to meet the requirements 303 */ 304void sk_set_memalloc(struct sock *sk) 305{ 306 sock_set_flag(sk, SOCK_MEMALLOC); 307 sk->sk_allocation |= __GFP_MEMALLOC; 308 static_branch_inc(&memalloc_socks_key); 309} 310EXPORT_SYMBOL_GPL(sk_set_memalloc); 311 312void sk_clear_memalloc(struct sock *sk) 313{ 314 sock_reset_flag(sk, SOCK_MEMALLOC); 315 sk->sk_allocation &= ~__GFP_MEMALLOC; 316 static_branch_dec(&memalloc_socks_key); 317 318 /* 319 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward 320 * progress of swapping. SOCK_MEMALLOC may be cleared while 321 * it has rmem allocations due to the last swapfile being deactivated 322 * but there is a risk that the socket is unusable due to exceeding 323 * the rmem limits. Reclaim the reserves and obey rmem limits again. 324 */ 325 sk_mem_reclaim(sk); 326} 327EXPORT_SYMBOL_GPL(sk_clear_memalloc); 328 329int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb) 330{ 331 int ret; 332 unsigned int noreclaim_flag; 333 334 /* these should have been dropped before queueing */ 335 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC)); 336 337 noreclaim_flag = memalloc_noreclaim_save(); 338 ret = INDIRECT_CALL_INET(sk->sk_backlog_rcv, 339 tcp_v6_do_rcv, 340 tcp_v4_do_rcv, 341 sk, skb); 342 memalloc_noreclaim_restore(noreclaim_flag); 343 344 return ret; 345} 346EXPORT_SYMBOL(__sk_backlog_rcv); 347 348void sk_error_report(struct sock *sk) 349{ 350 sk->sk_error_report(sk); 351 352 switch (sk->sk_family) { 353 case AF_INET: 354 fallthrough; 355 case AF_INET6: 356 trace_inet_sk_error_report(sk); 357 break; 358 default: 359 break; 360 } 361} 362EXPORT_SYMBOL(sk_error_report); 363 364int sock_get_timeout(long timeo, void *optval, bool old_timeval) 365{ 366 struct __kernel_sock_timeval tv; 367 368 if (timeo == MAX_SCHEDULE_TIMEOUT) { 369 tv.tv_sec = 0; 370 tv.tv_usec = 0; 371 } else { 372 tv.tv_sec = timeo / HZ; 373 tv.tv_usec = ((timeo % HZ) * USEC_PER_SEC) / HZ; 374 } 375 376 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) { 377 struct old_timeval32 tv32 = { tv.tv_sec, tv.tv_usec }; 378 *(struct old_timeval32 *)optval = tv32; 379 return sizeof(tv32); 380 } 381 382 if (old_timeval) { 383 struct __kernel_old_timeval old_tv; 384 old_tv.tv_sec = tv.tv_sec; 385 old_tv.tv_usec = tv.tv_usec; 386 *(struct __kernel_old_timeval *)optval = old_tv; 387 return sizeof(old_tv); 388 } 389 390 *(struct __kernel_sock_timeval *)optval = tv; 391 return sizeof(tv); 392} 393EXPORT_SYMBOL(sock_get_timeout); 394 395int sock_copy_user_timeval(struct __kernel_sock_timeval *tv, 396 sockptr_t optval, int optlen, bool old_timeval) 397{ 398 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) { 399 struct old_timeval32 tv32; 400 401 if (optlen < sizeof(tv32)) 402 return -EINVAL; 403 404 if (copy_from_sockptr(&tv32, optval, sizeof(tv32))) 405 return -EFAULT; 406 tv->tv_sec = tv32.tv_sec; 407 tv->tv_usec = tv32.tv_usec; 408 } else if (old_timeval) { 409 struct __kernel_old_timeval old_tv; 410 411 if (optlen < sizeof(old_tv)) 412 return -EINVAL; 413 if (copy_from_sockptr(&old_tv, optval, sizeof(old_tv))) 414 return -EFAULT; 415 tv->tv_sec = old_tv.tv_sec; 416 tv->tv_usec = old_tv.tv_usec; 417 } else { 418 if (optlen < sizeof(*tv)) 419 return -EINVAL; 420 if (copy_from_sockptr(tv, optval, sizeof(*tv))) 421 return -EFAULT; 422 } 423 424 return 0; 425} 426EXPORT_SYMBOL(sock_copy_user_timeval); 427 428static int sock_set_timeout(long *timeo_p, sockptr_t optval, int optlen, 429 bool old_timeval) 430{ 431 struct __kernel_sock_timeval tv; 432 int err = sock_copy_user_timeval(&tv, optval, optlen, old_timeval); 433 long val; 434 435 if (err) 436 return err; 437 438 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC) 439 return -EDOM; 440 441 if (tv.tv_sec < 0) { 442 static int warned __read_mostly; 443 444 WRITE_ONCE(*timeo_p, 0); 445 if (warned < 10 && net_ratelimit()) { 446 warned++; 447 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n", 448 __func__, current->comm, task_pid_nr(current)); 449 } 450 return 0; 451 } 452 val = MAX_SCHEDULE_TIMEOUT; 453 if ((tv.tv_sec || tv.tv_usec) && 454 (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1))) 455 val = tv.tv_sec * HZ + DIV_ROUND_UP((unsigned long)tv.tv_usec, 456 USEC_PER_SEC / HZ); 457 WRITE_ONCE(*timeo_p, val); 458 return 0; 459} 460 461static bool sock_needs_netstamp(const struct sock *sk) 462{ 463 switch (sk->sk_family) { 464 case AF_UNSPEC: 465 case AF_UNIX: 466 return false; 467 default: 468 return true; 469 } 470} 471 472static void sock_disable_timestamp(struct sock *sk, unsigned long flags) 473{ 474 if (sk->sk_flags & flags) { 475 sk->sk_flags &= ~flags; 476 if (sock_needs_netstamp(sk) && 477 !(sk->sk_flags & SK_FLAGS_TIMESTAMP)) 478 net_disable_timestamp(); 479 } 480} 481 482 483int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 484{ 485 unsigned long flags; 486 struct sk_buff_head *list = &sk->sk_receive_queue; 487 488 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) { 489 atomic_inc(&sk->sk_drops); 490 trace_sock_rcvqueue_full(sk, skb); 491 return -ENOMEM; 492 } 493 494 if (!sk_rmem_schedule(sk, skb, skb->truesize)) { 495 atomic_inc(&sk->sk_drops); 496 return -ENOBUFS; 497 } 498 499 skb->dev = NULL; 500 skb_set_owner_r(skb, sk); 501 502 /* we escape from rcu protected region, make sure we dont leak 503 * a norefcounted dst 504 */ 505 skb_dst_force(skb); 506 507 spin_lock_irqsave(&list->lock, flags); 508 sock_skb_set_dropcount(sk, skb); 509 __skb_queue_tail(list, skb); 510 spin_unlock_irqrestore(&list->lock, flags); 511 512 if (!sock_flag(sk, SOCK_DEAD)) 513 sk->sk_data_ready(sk); 514 return 0; 515} 516EXPORT_SYMBOL(__sock_queue_rcv_skb); 517 518int sock_queue_rcv_skb_reason(struct sock *sk, struct sk_buff *skb, 519 enum skb_drop_reason *reason) 520{ 521 enum skb_drop_reason drop_reason; 522 int err; 523 524 err = sk_filter(sk, skb); 525 if (err) { 526 drop_reason = SKB_DROP_REASON_SOCKET_FILTER; 527 goto out; 528 } 529 err = __sock_queue_rcv_skb(sk, skb); 530 switch (err) { 531 case -ENOMEM: 532 drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF; 533 break; 534 case -ENOBUFS: 535 drop_reason = SKB_DROP_REASON_PROTO_MEM; 536 break; 537 default: 538 drop_reason = SKB_NOT_DROPPED_YET; 539 break; 540 } 541out: 542 if (reason) 543 *reason = drop_reason; 544 return err; 545} 546EXPORT_SYMBOL(sock_queue_rcv_skb_reason); 547 548int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, 549 const int nested, unsigned int trim_cap, bool refcounted) 550{ 551 int rc = NET_RX_SUCCESS; 552 553 if (sk_filter_trim_cap(sk, skb, trim_cap)) 554 goto discard_and_relse; 555 556 skb->dev = NULL; 557 558 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) { 559 atomic_inc(&sk->sk_drops); 560 goto discard_and_relse; 561 } 562 if (nested) 563 bh_lock_sock_nested(sk); 564 else 565 bh_lock_sock(sk); 566 if (!sock_owned_by_user(sk)) { 567 /* 568 * trylock + unlock semantics: 569 */ 570 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_); 571 572 rc = sk_backlog_rcv(sk, skb); 573 574 mutex_release(&sk->sk_lock.dep_map, _RET_IP_); 575 } else if (sk_add_backlog(sk, skb, READ_ONCE(sk->sk_rcvbuf))) { 576 bh_unlock_sock(sk); 577 atomic_inc(&sk->sk_drops); 578 goto discard_and_relse; 579 } 580 581 bh_unlock_sock(sk); 582out: 583 if (refcounted) 584 sock_put(sk); 585 return rc; 586discard_and_relse: 587 kfree_skb(skb); 588 goto out; 589} 590EXPORT_SYMBOL(__sk_receive_skb); 591 592INDIRECT_CALLABLE_DECLARE(struct dst_entry *ip6_dst_check(struct dst_entry *, 593 u32)); 594INDIRECT_CALLABLE_DECLARE(struct dst_entry *ipv4_dst_check(struct dst_entry *, 595 u32)); 596struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie) 597{ 598 struct dst_entry *dst = __sk_dst_get(sk); 599 600 if (dst && dst->obsolete && 601 INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check, 602 dst, cookie) == NULL) { 603 sk_tx_queue_clear(sk); 604 WRITE_ONCE(sk->sk_dst_pending_confirm, 0); 605 RCU_INIT_POINTER(sk->sk_dst_cache, NULL); 606 dst_release(dst); 607 return NULL; 608 } 609 610 return dst; 611} 612EXPORT_SYMBOL(__sk_dst_check); 613 614struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie) 615{ 616 struct dst_entry *dst = sk_dst_get(sk); 617 618 if (dst && dst->obsolete && 619 INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check, 620 dst, cookie) == NULL) { 621 sk_dst_reset(sk); 622 dst_release(dst); 623 return NULL; 624 } 625 626 return dst; 627} 628EXPORT_SYMBOL(sk_dst_check); 629 630static int sock_bindtoindex_locked(struct sock *sk, int ifindex) 631{ 632 int ret = -ENOPROTOOPT; 633#ifdef CONFIG_NETDEVICES 634 struct net *net = sock_net(sk); 635 636 /* Sorry... */ 637 ret = -EPERM; 638 if (sk->sk_bound_dev_if && !ns_capable(net->user_ns, CAP_NET_RAW)) 639 goto out; 640 641 ret = -EINVAL; 642 if (ifindex < 0) 643 goto out; 644 645 /* Paired with all READ_ONCE() done locklessly. */ 646 WRITE_ONCE(sk->sk_bound_dev_if, ifindex); 647 648 if (sk->sk_prot->rehash) 649 sk->sk_prot->rehash(sk); 650 sk_dst_reset(sk); 651 652 ret = 0; 653 654out: 655#endif 656 657 return ret; 658} 659 660int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk) 661{ 662 int ret; 663 664 if (lock_sk) 665 lock_sock(sk); 666 ret = sock_bindtoindex_locked(sk, ifindex); 667 if (lock_sk) 668 release_sock(sk); 669 670 return ret; 671} 672EXPORT_SYMBOL(sock_bindtoindex); 673 674static int sock_setbindtodevice(struct sock *sk, sockptr_t optval, int optlen) 675{ 676 int ret = -ENOPROTOOPT; 677#ifdef CONFIG_NETDEVICES 678 struct net *net = sock_net(sk); 679 char devname[IFNAMSIZ]; 680 int index; 681 682 ret = -EINVAL; 683 if (optlen < 0) 684 goto out; 685 686 /* Bind this socket to a particular device like "eth0", 687 * as specified in the passed interface name. If the 688 * name is "" or the option length is zero the socket 689 * is not bound. 690 */ 691 if (optlen > IFNAMSIZ - 1) 692 optlen = IFNAMSIZ - 1; 693 memset(devname, 0, sizeof(devname)); 694 695 ret = -EFAULT; 696 if (copy_from_sockptr(devname, optval, optlen)) 697 goto out; 698 699 index = 0; 700 if (devname[0] != '\0') { 701 struct net_device *dev; 702 703 rcu_read_lock(); 704 dev = dev_get_by_name_rcu(net, devname); 705 if (dev) 706 index = dev->ifindex; 707 rcu_read_unlock(); 708 ret = -ENODEV; 709 if (!dev) 710 goto out; 711 } 712 713 sockopt_lock_sock(sk); 714 ret = sock_bindtoindex_locked(sk, index); 715 sockopt_release_sock(sk); 716out: 717#endif 718 719 return ret; 720} 721 722static int sock_getbindtodevice(struct sock *sk, sockptr_t optval, 723 sockptr_t optlen, int len) 724{ 725 int ret = -ENOPROTOOPT; 726#ifdef CONFIG_NETDEVICES 727 int bound_dev_if = READ_ONCE(sk->sk_bound_dev_if); 728 struct net *net = sock_net(sk); 729 char devname[IFNAMSIZ]; 730 731 if (bound_dev_if == 0) { 732 len = 0; 733 goto zero; 734 } 735 736 ret = -EINVAL; 737 if (len < IFNAMSIZ) 738 goto out; 739 740 ret = netdev_get_name(net, devname, bound_dev_if); 741 if (ret) 742 goto out; 743 744 len = strlen(devname) + 1; 745 746 ret = -EFAULT; 747 if (copy_to_sockptr(optval, devname, len)) 748 goto out; 749 750zero: 751 ret = -EFAULT; 752 if (copy_to_sockptr(optlen, &len, sizeof(int))) 753 goto out; 754 755 ret = 0; 756 757out: 758#endif 759 760 return ret; 761} 762 763bool sk_mc_loop(struct sock *sk) 764{ 765 if (dev_recursion_level()) 766 return false; 767 if (!sk) 768 return true; 769 /* IPV6_ADDRFORM can change sk->sk_family under us. */ 770 switch (READ_ONCE(sk->sk_family)) { 771 case AF_INET: 772 return inet_test_bit(MC_LOOP, sk); 773#if IS_ENABLED(CONFIG_IPV6) 774 case AF_INET6: 775 return inet6_sk(sk)->mc_loop; 776#endif 777 } 778 WARN_ON_ONCE(1); 779 return true; 780} 781EXPORT_SYMBOL(sk_mc_loop); 782 783void sock_set_reuseaddr(struct sock *sk) 784{ 785 lock_sock(sk); 786 sk->sk_reuse = SK_CAN_REUSE; 787 release_sock(sk); 788} 789EXPORT_SYMBOL(sock_set_reuseaddr); 790 791void sock_set_reuseport(struct sock *sk) 792{ 793 lock_sock(sk); 794 sk->sk_reuseport = true; 795 release_sock(sk); 796} 797EXPORT_SYMBOL(sock_set_reuseport); 798 799void sock_no_linger(struct sock *sk) 800{ 801 lock_sock(sk); 802 WRITE_ONCE(sk->sk_lingertime, 0); 803 sock_set_flag(sk, SOCK_LINGER); 804 release_sock(sk); 805} 806EXPORT_SYMBOL(sock_no_linger); 807 808void sock_set_priority(struct sock *sk, u32 priority) 809{ 810 lock_sock(sk); 811 WRITE_ONCE(sk->sk_priority, priority); 812 release_sock(sk); 813} 814EXPORT_SYMBOL(sock_set_priority); 815 816void sock_set_sndtimeo(struct sock *sk, s64 secs) 817{ 818 lock_sock(sk); 819 if (secs && secs < MAX_SCHEDULE_TIMEOUT / HZ - 1) 820 WRITE_ONCE(sk->sk_sndtimeo, secs * HZ); 821 else 822 WRITE_ONCE(sk->sk_sndtimeo, MAX_SCHEDULE_TIMEOUT); 823 release_sock(sk); 824} 825EXPORT_SYMBOL(sock_set_sndtimeo); 826 827static void __sock_set_timestamps(struct sock *sk, bool val, bool new, bool ns) 828{ 829 if (val) { 830 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, new); 831 sock_valbool_flag(sk, SOCK_RCVTSTAMPNS, ns); 832 sock_set_flag(sk, SOCK_RCVTSTAMP); 833 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 834 } else { 835 sock_reset_flag(sk, SOCK_RCVTSTAMP); 836 sock_reset_flag(sk, SOCK_RCVTSTAMPNS); 837 } 838} 839 840void sock_enable_timestamps(struct sock *sk) 841{ 842 lock_sock(sk); 843 __sock_set_timestamps(sk, true, false, true); 844 release_sock(sk); 845} 846EXPORT_SYMBOL(sock_enable_timestamps); 847 848void sock_set_timestamp(struct sock *sk, int optname, bool valbool) 849{ 850 switch (optname) { 851 case SO_TIMESTAMP_OLD: 852 __sock_set_timestamps(sk, valbool, false, false); 853 break; 854 case SO_TIMESTAMP_NEW: 855 __sock_set_timestamps(sk, valbool, true, false); 856 break; 857 case SO_TIMESTAMPNS_OLD: 858 __sock_set_timestamps(sk, valbool, false, true); 859 break; 860 case SO_TIMESTAMPNS_NEW: 861 __sock_set_timestamps(sk, valbool, true, true); 862 break; 863 } 864} 865 866static int sock_timestamping_bind_phc(struct sock *sk, int phc_index) 867{ 868 struct net *net = sock_net(sk); 869 struct net_device *dev = NULL; 870 bool match = false; 871 int *vclock_index; 872 int i, num; 873 874 if (sk->sk_bound_dev_if) 875 dev = dev_get_by_index(net, sk->sk_bound_dev_if); 876 877 if (!dev) { 878 pr_err("%s: sock not bind to device\n", __func__); 879 return -EOPNOTSUPP; 880 } 881 882 num = ethtool_get_phc_vclocks(dev, &vclock_index); 883 dev_put(dev); 884 885 for (i = 0; i < num; i++) { 886 if (*(vclock_index + i) == phc_index) { 887 match = true; 888 break; 889 } 890 } 891 892 if (num > 0) 893 kfree(vclock_index); 894 895 if (!match) 896 return -EINVAL; 897 898 WRITE_ONCE(sk->sk_bind_phc, phc_index); 899 900 return 0; 901} 902 903int sock_set_timestamping(struct sock *sk, int optname, 904 struct so_timestamping timestamping) 905{ 906 int val = timestamping.flags; 907 int ret; 908 909 if (val & ~SOF_TIMESTAMPING_MASK) 910 return -EINVAL; 911 912 if (val & SOF_TIMESTAMPING_OPT_ID_TCP && 913 !(val & SOF_TIMESTAMPING_OPT_ID)) 914 return -EINVAL; 915 916 if (val & SOF_TIMESTAMPING_OPT_ID && 917 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) { 918 if (sk_is_tcp(sk)) { 919 if ((1 << sk->sk_state) & 920 (TCPF_CLOSE | TCPF_LISTEN)) 921 return -EINVAL; 922 if (val & SOF_TIMESTAMPING_OPT_ID_TCP) 923 atomic_set(&sk->sk_tskey, tcp_sk(sk)->write_seq); 924 else 925 atomic_set(&sk->sk_tskey, tcp_sk(sk)->snd_una); 926 } else { 927 atomic_set(&sk->sk_tskey, 0); 928 } 929 } 930 931 if (val & SOF_TIMESTAMPING_OPT_STATS && 932 !(val & SOF_TIMESTAMPING_OPT_TSONLY)) 933 return -EINVAL; 934 935 if (val & SOF_TIMESTAMPING_BIND_PHC) { 936 ret = sock_timestamping_bind_phc(sk, timestamping.bind_phc); 937 if (ret) 938 return ret; 939 } 940 941 WRITE_ONCE(sk->sk_tsflags, val); 942 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, optname == SO_TIMESTAMPING_NEW); 943 944 if (val & SOF_TIMESTAMPING_RX_SOFTWARE) 945 sock_enable_timestamp(sk, 946 SOCK_TIMESTAMPING_RX_SOFTWARE); 947 else 948 sock_disable_timestamp(sk, 949 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE)); 950 return 0; 951} 952 953void sock_set_keepalive(struct sock *sk) 954{ 955 lock_sock(sk); 956 if (sk->sk_prot->keepalive) 957 sk->sk_prot->keepalive(sk, true); 958 sock_valbool_flag(sk, SOCK_KEEPOPEN, true); 959 release_sock(sk); 960} 961EXPORT_SYMBOL(sock_set_keepalive); 962 963static void __sock_set_rcvbuf(struct sock *sk, int val) 964{ 965 /* Ensure val * 2 fits into an int, to prevent max_t() from treating it 966 * as a negative value. 967 */ 968 val = min_t(int, val, INT_MAX / 2); 969 sk->sk_userlocks |= SOCK_RCVBUF_LOCK; 970 971 /* We double it on the way in to account for "struct sk_buff" etc. 972 * overhead. Applications assume that the SO_RCVBUF setting they make 973 * will allow that much actual data to be received on that socket. 974 * 975 * Applications are unaware that "struct sk_buff" and other overheads 976 * allocate from the receive buffer during socket buffer allocation. 977 * 978 * And after considering the possible alternatives, returning the value 979 * we actually used in getsockopt is the most desirable behavior. 980 */ 981 WRITE_ONCE(sk->sk_rcvbuf, max_t(int, val * 2, SOCK_MIN_RCVBUF)); 982} 983 984void sock_set_rcvbuf(struct sock *sk, int val) 985{ 986 lock_sock(sk); 987 __sock_set_rcvbuf(sk, val); 988 release_sock(sk); 989} 990EXPORT_SYMBOL(sock_set_rcvbuf); 991 992static void __sock_set_mark(struct sock *sk, u32 val) 993{ 994 if (val != sk->sk_mark) { 995 WRITE_ONCE(sk->sk_mark, val); 996 sk_dst_reset(sk); 997 } 998} 999 1000void sock_set_mark(struct sock *sk, u32 val) 1001{ 1002 lock_sock(sk); 1003 __sock_set_mark(sk, val); 1004 release_sock(sk); 1005} 1006EXPORT_SYMBOL(sock_set_mark); 1007 1008static void sock_release_reserved_memory(struct sock *sk, int bytes) 1009{ 1010 /* Round down bytes to multiple of pages */ 1011 bytes = round_down(bytes, PAGE_SIZE); 1012 1013 WARN_ON(bytes > sk->sk_reserved_mem); 1014 WRITE_ONCE(sk->sk_reserved_mem, sk->sk_reserved_mem - bytes); 1015 sk_mem_reclaim(sk); 1016} 1017 1018static int sock_reserve_memory(struct sock *sk, int bytes) 1019{ 1020 long allocated; 1021 bool charged; 1022 int pages; 1023 1024 if (!mem_cgroup_sockets_enabled || !sk->sk_memcg || !sk_has_account(sk)) 1025 return -EOPNOTSUPP; 1026 1027 if (!bytes) 1028 return 0; 1029 1030 pages = sk_mem_pages(bytes); 1031 1032 /* pre-charge to memcg */ 1033 charged = mem_cgroup_charge_skmem(sk->sk_memcg, pages, 1034 GFP_KERNEL | __GFP_RETRY_MAYFAIL); 1035 if (!charged) 1036 return -ENOMEM; 1037 1038 /* pre-charge to forward_alloc */ 1039 sk_memory_allocated_add(sk, pages); 1040 allocated = sk_memory_allocated(sk); 1041 /* If the system goes into memory pressure with this 1042 * precharge, give up and return error. 1043 */ 1044 if (allocated > sk_prot_mem_limits(sk, 1)) { 1045 sk_memory_allocated_sub(sk, pages); 1046 mem_cgroup_uncharge_skmem(sk->sk_memcg, pages); 1047 return -ENOMEM; 1048 } 1049 sk_forward_alloc_add(sk, pages << PAGE_SHIFT); 1050 1051 WRITE_ONCE(sk->sk_reserved_mem, 1052 sk->sk_reserved_mem + (pages << PAGE_SHIFT)); 1053 1054 return 0; 1055} 1056 1057void sockopt_lock_sock(struct sock *sk) 1058{ 1059 /* When current->bpf_ctx is set, the setsockopt is called from 1060 * a bpf prog. bpf has ensured the sk lock has been 1061 * acquired before calling setsockopt(). 1062 */ 1063 if (has_current_bpf_ctx()) 1064 return; 1065 1066 lock_sock(sk); 1067} 1068EXPORT_SYMBOL(sockopt_lock_sock); 1069 1070void sockopt_release_sock(struct sock *sk) 1071{ 1072 if (has_current_bpf_ctx()) 1073 return; 1074 1075 release_sock(sk); 1076} 1077EXPORT_SYMBOL(sockopt_release_sock); 1078 1079bool sockopt_ns_capable(struct user_namespace *ns, int cap) 1080{ 1081 return has_current_bpf_ctx() || ns_capable(ns, cap); 1082} 1083EXPORT_SYMBOL(sockopt_ns_capable); 1084 1085bool sockopt_capable(int cap) 1086{ 1087 return has_current_bpf_ctx() || capable(cap); 1088} 1089EXPORT_SYMBOL(sockopt_capable); 1090 1091/* 1092 * This is meant for all protocols to use and covers goings on 1093 * at the socket level. Everything here is generic. 1094 */ 1095 1096int sk_setsockopt(struct sock *sk, int level, int optname, 1097 sockptr_t optval, unsigned int optlen) 1098{ 1099 struct so_timestamping timestamping; 1100 struct socket *sock = sk->sk_socket; 1101 struct sock_txtime sk_txtime; 1102 int val; 1103 int valbool; 1104 struct linger ling; 1105 int ret = 0; 1106 1107 /* 1108 * Options without arguments 1109 */ 1110 1111 if (optname == SO_BINDTODEVICE) 1112 return sock_setbindtodevice(sk, optval, optlen); 1113 1114 if (optlen < sizeof(int)) 1115 return -EINVAL; 1116 1117 if (copy_from_sockptr(&val, optval, sizeof(val))) 1118 return -EFAULT; 1119 1120 valbool = val ? 1 : 0; 1121 1122 sockopt_lock_sock(sk); 1123 1124 switch (optname) { 1125 case SO_DEBUG: 1126 if (val && !sockopt_capable(CAP_NET_ADMIN)) 1127 ret = -EACCES; 1128 else 1129 sock_valbool_flag(sk, SOCK_DBG, valbool); 1130 break; 1131 case SO_REUSEADDR: 1132 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE); 1133 break; 1134 case SO_REUSEPORT: 1135 sk->sk_reuseport = valbool; 1136 break; 1137 case SO_TYPE: 1138 case SO_PROTOCOL: 1139 case SO_DOMAIN: 1140 case SO_ERROR: 1141 ret = -ENOPROTOOPT; 1142 break; 1143 case SO_DONTROUTE: 1144 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool); 1145 sk_dst_reset(sk); 1146 break; 1147 case SO_BROADCAST: 1148 sock_valbool_flag(sk, SOCK_BROADCAST, valbool); 1149 break; 1150 case SO_SNDBUF: 1151 /* Don't error on this BSD doesn't and if you think 1152 * about it this is right. Otherwise apps have to 1153 * play 'guess the biggest size' games. RCVBUF/SNDBUF 1154 * are treated in BSD as hints 1155 */ 1156 val = min_t(u32, val, READ_ONCE(sysctl_wmem_max)); 1157set_sndbuf: 1158 /* Ensure val * 2 fits into an int, to prevent max_t() 1159 * from treating it as a negative value. 1160 */ 1161 val = min_t(int, val, INT_MAX / 2); 1162 sk->sk_userlocks |= SOCK_SNDBUF_LOCK; 1163 WRITE_ONCE(sk->sk_sndbuf, 1164 max_t(int, val * 2, SOCK_MIN_SNDBUF)); 1165 /* Wake up sending tasks if we upped the value. */ 1166 sk->sk_write_space(sk); 1167 break; 1168 1169 case SO_SNDBUFFORCE: 1170 if (!sockopt_capable(CAP_NET_ADMIN)) { 1171 ret = -EPERM; 1172 break; 1173 } 1174 1175 /* No negative values (to prevent underflow, as val will be 1176 * multiplied by 2). 1177 */ 1178 if (val < 0) 1179 val = 0; 1180 goto set_sndbuf; 1181 1182 case SO_RCVBUF: 1183 /* Don't error on this BSD doesn't and if you think 1184 * about it this is right. Otherwise apps have to 1185 * play 'guess the biggest size' games. RCVBUF/SNDBUF 1186 * are treated in BSD as hints 1187 */ 1188 __sock_set_rcvbuf(sk, min_t(u32, val, READ_ONCE(sysctl_rmem_max))); 1189 break; 1190 1191 case SO_RCVBUFFORCE: 1192 if (!sockopt_capable(CAP_NET_ADMIN)) { 1193 ret = -EPERM; 1194 break; 1195 } 1196 1197 /* No negative values (to prevent underflow, as val will be 1198 * multiplied by 2). 1199 */ 1200 __sock_set_rcvbuf(sk, max(val, 0)); 1201 break; 1202 1203 case SO_KEEPALIVE: 1204 if (sk->sk_prot->keepalive) 1205 sk->sk_prot->keepalive(sk, valbool); 1206 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool); 1207 break; 1208 1209 case SO_OOBINLINE: 1210 sock_valbool_flag(sk, SOCK_URGINLINE, valbool); 1211 break; 1212 1213 case SO_NO_CHECK: 1214 sk->sk_no_check_tx = valbool; 1215 break; 1216 1217 case SO_PRIORITY: 1218 if ((val >= 0 && val <= 6) || 1219 sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) || 1220 sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) 1221 WRITE_ONCE(sk->sk_priority, val); 1222 else 1223 ret = -EPERM; 1224 break; 1225 1226 case SO_LINGER: 1227 if (optlen < sizeof(ling)) { 1228 ret = -EINVAL; /* 1003.1g */ 1229 break; 1230 } 1231 if (copy_from_sockptr(&ling, optval, sizeof(ling))) { 1232 ret = -EFAULT; 1233 break; 1234 } 1235 if (!ling.l_onoff) { 1236 sock_reset_flag(sk, SOCK_LINGER); 1237 } else { 1238 unsigned long t_sec = ling.l_linger; 1239 1240 if (t_sec >= MAX_SCHEDULE_TIMEOUT / HZ) 1241 WRITE_ONCE(sk->sk_lingertime, MAX_SCHEDULE_TIMEOUT); 1242 else 1243 WRITE_ONCE(sk->sk_lingertime, t_sec * HZ); 1244 sock_set_flag(sk, SOCK_LINGER); 1245 } 1246 break; 1247 1248 case SO_BSDCOMPAT: 1249 break; 1250 1251 case SO_PASSCRED: 1252 assign_bit(SOCK_PASSCRED, &sock->flags, valbool); 1253 break; 1254 1255 case SO_PASSPIDFD: 1256 assign_bit(SOCK_PASSPIDFD, &sock->flags, valbool); 1257 break; 1258 1259 case SO_TIMESTAMP_OLD: 1260 case SO_TIMESTAMP_NEW: 1261 case SO_TIMESTAMPNS_OLD: 1262 case SO_TIMESTAMPNS_NEW: 1263 sock_set_timestamp(sk, optname, valbool); 1264 break; 1265 1266 case SO_TIMESTAMPING_NEW: 1267 case SO_TIMESTAMPING_OLD: 1268 if (optlen == sizeof(timestamping)) { 1269 if (copy_from_sockptr(×tamping, optval, 1270 sizeof(timestamping))) { 1271 ret = -EFAULT; 1272 break; 1273 } 1274 } else { 1275 memset(×tamping, 0, sizeof(timestamping)); 1276 timestamping.flags = val; 1277 } 1278 ret = sock_set_timestamping(sk, optname, timestamping); 1279 break; 1280 1281 case SO_RCVLOWAT: 1282 { 1283 int (*set_rcvlowat)(struct sock *sk, int val) = NULL; 1284 1285 if (val < 0) 1286 val = INT_MAX; 1287 if (sock) 1288 set_rcvlowat = READ_ONCE(sock->ops)->set_rcvlowat; 1289 if (set_rcvlowat) 1290 ret = set_rcvlowat(sk, val); 1291 else 1292 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1); 1293 break; 1294 } 1295 case SO_RCVTIMEO_OLD: 1296 case SO_RCVTIMEO_NEW: 1297 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, 1298 optlen, optname == SO_RCVTIMEO_OLD); 1299 break; 1300 1301 case SO_SNDTIMEO_OLD: 1302 case SO_SNDTIMEO_NEW: 1303 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, 1304 optlen, optname == SO_SNDTIMEO_OLD); 1305 break; 1306 1307 case SO_ATTACH_FILTER: { 1308 struct sock_fprog fprog; 1309 1310 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen); 1311 if (!ret) 1312 ret = sk_attach_filter(&fprog, sk); 1313 break; 1314 } 1315 case SO_ATTACH_BPF: 1316 ret = -EINVAL; 1317 if (optlen == sizeof(u32)) { 1318 u32 ufd; 1319 1320 ret = -EFAULT; 1321 if (copy_from_sockptr(&ufd, optval, sizeof(ufd))) 1322 break; 1323 1324 ret = sk_attach_bpf(ufd, sk); 1325 } 1326 break; 1327 1328 case SO_ATTACH_REUSEPORT_CBPF: { 1329 struct sock_fprog fprog; 1330 1331 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen); 1332 if (!ret) 1333 ret = sk_reuseport_attach_filter(&fprog, sk); 1334 break; 1335 } 1336 case SO_ATTACH_REUSEPORT_EBPF: 1337 ret = -EINVAL; 1338 if (optlen == sizeof(u32)) { 1339 u32 ufd; 1340 1341 ret = -EFAULT; 1342 if (copy_from_sockptr(&ufd, optval, sizeof(ufd))) 1343 break; 1344 1345 ret = sk_reuseport_attach_bpf(ufd, sk); 1346 } 1347 break; 1348 1349 case SO_DETACH_REUSEPORT_BPF: 1350 ret = reuseport_detach_prog(sk); 1351 break; 1352 1353 case SO_DETACH_FILTER: 1354 ret = sk_detach_filter(sk); 1355 break; 1356 1357 case SO_LOCK_FILTER: 1358 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool) 1359 ret = -EPERM; 1360 else 1361 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool); 1362 break; 1363 1364 case SO_PASSSEC: 1365 assign_bit(SOCK_PASSSEC, &sock->flags, valbool); 1366 break; 1367 case SO_MARK: 1368 if (!sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) && 1369 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) { 1370 ret = -EPERM; 1371 break; 1372 } 1373 1374 __sock_set_mark(sk, val); 1375 break; 1376 case SO_RCVMARK: 1377 sock_valbool_flag(sk, SOCK_RCVMARK, valbool); 1378 break; 1379 1380 case SO_RXQ_OVFL: 1381 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool); 1382 break; 1383 1384 case SO_WIFI_STATUS: 1385 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool); 1386 break; 1387 1388 case SO_PEEK_OFF: 1389 { 1390 int (*set_peek_off)(struct sock *sk, int val); 1391 1392 set_peek_off = READ_ONCE(sock->ops)->set_peek_off; 1393 if (set_peek_off) 1394 ret = set_peek_off(sk, val); 1395 else 1396 ret = -EOPNOTSUPP; 1397 break; 1398 } 1399 1400 case SO_NOFCS: 1401 sock_valbool_flag(sk, SOCK_NOFCS, valbool); 1402 break; 1403 1404 case SO_SELECT_ERR_QUEUE: 1405 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool); 1406 break; 1407 1408#ifdef CONFIG_NET_RX_BUSY_POLL 1409 case SO_BUSY_POLL: 1410 if (val < 0) 1411 ret = -EINVAL; 1412 else 1413 WRITE_ONCE(sk->sk_ll_usec, val); 1414 break; 1415 case SO_PREFER_BUSY_POLL: 1416 if (valbool && !sockopt_capable(CAP_NET_ADMIN)) 1417 ret = -EPERM; 1418 else 1419 WRITE_ONCE(sk->sk_prefer_busy_poll, valbool); 1420 break; 1421 case SO_BUSY_POLL_BUDGET: 1422 if (val > READ_ONCE(sk->sk_busy_poll_budget) && !sockopt_capable(CAP_NET_ADMIN)) { 1423 ret = -EPERM; 1424 } else { 1425 if (val < 0 || val > U16_MAX) 1426 ret = -EINVAL; 1427 else 1428 WRITE_ONCE(sk->sk_busy_poll_budget, val); 1429 } 1430 break; 1431#endif 1432 1433 case SO_MAX_PACING_RATE: 1434 { 1435 unsigned long ulval = (val == ~0U) ? ~0UL : (unsigned int)val; 1436 1437 if (sizeof(ulval) != sizeof(val) && 1438 optlen >= sizeof(ulval) && 1439 copy_from_sockptr(&ulval, optval, sizeof(ulval))) { 1440 ret = -EFAULT; 1441 break; 1442 } 1443 if (ulval != ~0UL) 1444 cmpxchg(&sk->sk_pacing_status, 1445 SK_PACING_NONE, 1446 SK_PACING_NEEDED); 1447 /* Pairs with READ_ONCE() from sk_getsockopt() */ 1448 WRITE_ONCE(sk->sk_max_pacing_rate, ulval); 1449 sk->sk_pacing_rate = min(sk->sk_pacing_rate, ulval); 1450 break; 1451 } 1452 case SO_INCOMING_CPU: 1453 reuseport_update_incoming_cpu(sk, val); 1454 break; 1455 1456 case SO_CNX_ADVICE: 1457 if (val == 1) 1458 dst_negative_advice(sk); 1459 break; 1460 1461 case SO_ZEROCOPY: 1462 if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) { 1463 if (!(sk_is_tcp(sk) || 1464 (sk->sk_type == SOCK_DGRAM && 1465 sk->sk_protocol == IPPROTO_UDP))) 1466 ret = -EOPNOTSUPP; 1467 } else if (sk->sk_family != PF_RDS) { 1468 ret = -EOPNOTSUPP; 1469 } 1470 if (!ret) { 1471 if (val < 0 || val > 1) 1472 ret = -EINVAL; 1473 else 1474 sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool); 1475 } 1476 break; 1477 1478 case SO_TXTIME: 1479 if (optlen != sizeof(struct sock_txtime)) { 1480 ret = -EINVAL; 1481 break; 1482 } else if (copy_from_sockptr(&sk_txtime, optval, 1483 sizeof(struct sock_txtime))) { 1484 ret = -EFAULT; 1485 break; 1486 } else if (sk_txtime.flags & ~SOF_TXTIME_FLAGS_MASK) { 1487 ret = -EINVAL; 1488 break; 1489 } 1490 /* CLOCK_MONOTONIC is only used by sch_fq, and this packet 1491 * scheduler has enough safe guards. 1492 */ 1493 if (sk_txtime.clockid != CLOCK_MONOTONIC && 1494 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) { 1495 ret = -EPERM; 1496 break; 1497 } 1498 sock_valbool_flag(sk, SOCK_TXTIME, true); 1499 sk->sk_clockid = sk_txtime.clockid; 1500 sk->sk_txtime_deadline_mode = 1501 !!(sk_txtime.flags & SOF_TXTIME_DEADLINE_MODE); 1502 sk->sk_txtime_report_errors = 1503 !!(sk_txtime.flags & SOF_TXTIME_REPORT_ERRORS); 1504 break; 1505 1506 case SO_BINDTOIFINDEX: 1507 ret = sock_bindtoindex_locked(sk, val); 1508 break; 1509 1510 case SO_BUF_LOCK: 1511 if (val & ~SOCK_BUF_LOCK_MASK) { 1512 ret = -EINVAL; 1513 break; 1514 } 1515 sk->sk_userlocks = val | (sk->sk_userlocks & 1516 ~SOCK_BUF_LOCK_MASK); 1517 break; 1518 1519 case SO_RESERVE_MEM: 1520 { 1521 int delta; 1522 1523 if (val < 0) { 1524 ret = -EINVAL; 1525 break; 1526 } 1527 1528 delta = val - sk->sk_reserved_mem; 1529 if (delta < 0) 1530 sock_release_reserved_memory(sk, -delta); 1531 else 1532 ret = sock_reserve_memory(sk, delta); 1533 break; 1534 } 1535 1536 case SO_TXREHASH: 1537 if (val < -1 || val > 1) { 1538 ret = -EINVAL; 1539 break; 1540 } 1541 if ((u8)val == SOCK_TXREHASH_DEFAULT) 1542 val = READ_ONCE(sock_net(sk)->core.sysctl_txrehash); 1543 /* Paired with READ_ONCE() in tcp_rtx_synack() 1544 * and sk_getsockopt(). 1545 */ 1546 WRITE_ONCE(sk->sk_txrehash, (u8)val); 1547 break; 1548 1549 default: 1550 ret = -ENOPROTOOPT; 1551 break; 1552 } 1553 sockopt_release_sock(sk); 1554 return ret; 1555} 1556 1557int sock_setsockopt(struct socket *sock, int level, int optname, 1558 sockptr_t optval, unsigned int optlen) 1559{ 1560 return sk_setsockopt(sock->sk, level, optname, 1561 optval, optlen); 1562} 1563EXPORT_SYMBOL(sock_setsockopt); 1564 1565static const struct cred *sk_get_peer_cred(struct sock *sk) 1566{ 1567 const struct cred *cred; 1568 1569 spin_lock(&sk->sk_peer_lock); 1570 cred = get_cred(sk->sk_peer_cred); 1571 spin_unlock(&sk->sk_peer_lock); 1572 1573 return cred; 1574} 1575 1576static void cred_to_ucred(struct pid *pid, const struct cred *cred, 1577 struct ucred *ucred) 1578{ 1579 ucred->pid = pid_vnr(pid); 1580 ucred->uid = ucred->gid = -1; 1581 if (cred) { 1582 struct user_namespace *current_ns = current_user_ns(); 1583 1584 ucred->uid = from_kuid_munged(current_ns, cred->euid); 1585 ucred->gid = from_kgid_munged(current_ns, cred->egid); 1586 } 1587} 1588 1589static int groups_to_user(sockptr_t dst, const struct group_info *src) 1590{ 1591 struct user_namespace *user_ns = current_user_ns(); 1592 int i; 1593 1594 for (i = 0; i < src->ngroups; i++) { 1595 gid_t gid = from_kgid_munged(user_ns, src->gid[i]); 1596 1597 if (copy_to_sockptr_offset(dst, i * sizeof(gid), &gid, sizeof(gid))) 1598 return -EFAULT; 1599 } 1600 1601 return 0; 1602} 1603 1604int sk_getsockopt(struct sock *sk, int level, int optname, 1605 sockptr_t optval, sockptr_t optlen) 1606{ 1607 struct socket *sock = sk->sk_socket; 1608 1609 union { 1610 int val; 1611 u64 val64; 1612 unsigned long ulval; 1613 struct linger ling; 1614 struct old_timeval32 tm32; 1615 struct __kernel_old_timeval tm; 1616 struct __kernel_sock_timeval stm; 1617 struct sock_txtime txtime; 1618 struct so_timestamping timestamping; 1619 } v; 1620 1621 int lv = sizeof(int); 1622 int len; 1623 1624 if (copy_from_sockptr(&len, optlen, sizeof(int))) 1625 return -EFAULT; 1626 if (len < 0) 1627 return -EINVAL; 1628 1629 memset(&v, 0, sizeof(v)); 1630 1631 switch (optname) { 1632 case SO_DEBUG: 1633 v.val = sock_flag(sk, SOCK_DBG); 1634 break; 1635 1636 case SO_DONTROUTE: 1637 v.val = sock_flag(sk, SOCK_LOCALROUTE); 1638 break; 1639 1640 case SO_BROADCAST: 1641 v.val = sock_flag(sk, SOCK_BROADCAST); 1642 break; 1643 1644 case SO_SNDBUF: 1645 v.val = READ_ONCE(sk->sk_sndbuf); 1646 break; 1647 1648 case SO_RCVBUF: 1649 v.val = READ_ONCE(sk->sk_rcvbuf); 1650 break; 1651 1652 case SO_REUSEADDR: 1653 v.val = sk->sk_reuse; 1654 break; 1655 1656 case SO_REUSEPORT: 1657 v.val = sk->sk_reuseport; 1658 break; 1659 1660 case SO_KEEPALIVE: 1661 v.val = sock_flag(sk, SOCK_KEEPOPEN); 1662 break; 1663 1664 case SO_TYPE: 1665 v.val = sk->sk_type; 1666 break; 1667 1668 case SO_PROTOCOL: 1669 v.val = sk->sk_protocol; 1670 break; 1671 1672 case SO_DOMAIN: 1673 v.val = sk->sk_family; 1674 break; 1675 1676 case SO_ERROR: 1677 v.val = -sock_error(sk); 1678 if (v.val == 0) 1679 v.val = xchg(&sk->sk_err_soft, 0); 1680 break; 1681 1682 case SO_OOBINLINE: 1683 v.val = sock_flag(sk, SOCK_URGINLINE); 1684 break; 1685 1686 case SO_NO_CHECK: 1687 v.val = sk->sk_no_check_tx; 1688 break; 1689 1690 case SO_PRIORITY: 1691 v.val = READ_ONCE(sk->sk_priority); 1692 break; 1693 1694 case SO_LINGER: 1695 lv = sizeof(v.ling); 1696 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER); 1697 v.ling.l_linger = READ_ONCE(sk->sk_lingertime) / HZ; 1698 break; 1699 1700 case SO_BSDCOMPAT: 1701 break; 1702 1703 case SO_TIMESTAMP_OLD: 1704 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && 1705 !sock_flag(sk, SOCK_TSTAMP_NEW) && 1706 !sock_flag(sk, SOCK_RCVTSTAMPNS); 1707 break; 1708 1709 case SO_TIMESTAMPNS_OLD: 1710 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && !sock_flag(sk, SOCK_TSTAMP_NEW); 1711 break; 1712 1713 case SO_TIMESTAMP_NEW: 1714 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && sock_flag(sk, SOCK_TSTAMP_NEW); 1715 break; 1716 1717 case SO_TIMESTAMPNS_NEW: 1718 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && sock_flag(sk, SOCK_TSTAMP_NEW); 1719 break; 1720 1721 case SO_TIMESTAMPING_OLD: 1722 case SO_TIMESTAMPING_NEW: 1723 lv = sizeof(v.timestamping); 1724 /* For the later-added case SO_TIMESTAMPING_NEW: Be strict about only 1725 * returning the flags when they were set through the same option. 1726 * Don't change the beviour for the old case SO_TIMESTAMPING_OLD. 1727 */ 1728 if (optname == SO_TIMESTAMPING_OLD || sock_flag(sk, SOCK_TSTAMP_NEW)) { 1729 v.timestamping.flags = READ_ONCE(sk->sk_tsflags); 1730 v.timestamping.bind_phc = READ_ONCE(sk->sk_bind_phc); 1731 } 1732 break; 1733 1734 case SO_RCVTIMEO_OLD: 1735 case SO_RCVTIMEO_NEW: 1736 lv = sock_get_timeout(READ_ONCE(sk->sk_rcvtimeo), &v, 1737 SO_RCVTIMEO_OLD == optname); 1738 break; 1739 1740 case SO_SNDTIMEO_OLD: 1741 case SO_SNDTIMEO_NEW: 1742 lv = sock_get_timeout(READ_ONCE(sk->sk_sndtimeo), &v, 1743 SO_SNDTIMEO_OLD == optname); 1744 break; 1745 1746 case SO_RCVLOWAT: 1747 v.val = READ_ONCE(sk->sk_rcvlowat); 1748 break; 1749 1750 case SO_SNDLOWAT: 1751 v.val = 1; 1752 break; 1753 1754 case SO_PASSCRED: 1755 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags); 1756 break; 1757 1758 case SO_PASSPIDFD: 1759 v.val = !!test_bit(SOCK_PASSPIDFD, &sock->flags); 1760 break; 1761 1762 case SO_PEERCRED: 1763 { 1764 struct ucred peercred; 1765 if (len > sizeof(peercred)) 1766 len = sizeof(peercred); 1767 1768 spin_lock(&sk->sk_peer_lock); 1769 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred); 1770 spin_unlock(&sk->sk_peer_lock); 1771 1772 if (copy_to_sockptr(optval, &peercred, len)) 1773 return -EFAULT; 1774 goto lenout; 1775 } 1776 1777 case SO_PEERPIDFD: 1778 { 1779 struct pid *peer_pid; 1780 struct file *pidfd_file = NULL; 1781 int pidfd; 1782 1783 if (len > sizeof(pidfd)) 1784 len = sizeof(pidfd); 1785 1786 spin_lock(&sk->sk_peer_lock); 1787 peer_pid = get_pid(sk->sk_peer_pid); 1788 spin_unlock(&sk->sk_peer_lock); 1789 1790 if (!peer_pid) 1791 return -ENODATA; 1792 1793 pidfd = pidfd_prepare(peer_pid, 0, &pidfd_file); 1794 put_pid(peer_pid); 1795 if (pidfd < 0) 1796 return pidfd; 1797 1798 if (copy_to_sockptr(optval, &pidfd, len) || 1799 copy_to_sockptr(optlen, &len, sizeof(int))) { 1800 put_unused_fd(pidfd); 1801 fput(pidfd_file); 1802 1803 return -EFAULT; 1804 } 1805 1806 fd_install(pidfd, pidfd_file); 1807 return 0; 1808 } 1809 1810 case SO_PEERGROUPS: 1811 { 1812 const struct cred *cred; 1813 int ret, n; 1814 1815 cred = sk_get_peer_cred(sk); 1816 if (!cred) 1817 return -ENODATA; 1818 1819 n = cred->group_info->ngroups; 1820 if (len < n * sizeof(gid_t)) { 1821 len = n * sizeof(gid_t); 1822 put_cred(cred); 1823 return copy_to_sockptr(optlen, &len, sizeof(int)) ? -EFAULT : -ERANGE; 1824 } 1825 len = n * sizeof(gid_t); 1826 1827 ret = groups_to_user(optval, cred->group_info); 1828 put_cred(cred); 1829 if (ret) 1830 return ret; 1831 goto lenout; 1832 } 1833 1834 case SO_PEERNAME: 1835 { 1836 struct sockaddr_storage address; 1837 1838 lv = READ_ONCE(sock->ops)->getname(sock, (struct sockaddr *)&address, 2); 1839 if (lv < 0) 1840 return -ENOTCONN; 1841 if (lv < len) 1842 return -EINVAL; 1843 if (copy_to_sockptr(optval, &address, len)) 1844 return -EFAULT; 1845 goto lenout; 1846 } 1847 1848 /* Dubious BSD thing... Probably nobody even uses it, but 1849 * the UNIX standard wants it for whatever reason... -DaveM 1850 */ 1851 case SO_ACCEPTCONN: 1852 v.val = sk->sk_state == TCP_LISTEN; 1853 break; 1854 1855 case SO_PASSSEC: 1856 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags); 1857 break; 1858 1859 case SO_PEERSEC: 1860 return security_socket_getpeersec_stream(sock, 1861 optval, optlen, len); 1862 1863 case SO_MARK: 1864 v.val = READ_ONCE(sk->sk_mark); 1865 break; 1866 1867 case SO_RCVMARK: 1868 v.val = sock_flag(sk, SOCK_RCVMARK); 1869 break; 1870 1871 case SO_RXQ_OVFL: 1872 v.val = sock_flag(sk, SOCK_RXQ_OVFL); 1873 break; 1874 1875 case SO_WIFI_STATUS: 1876 v.val = sock_flag(sk, SOCK_WIFI_STATUS); 1877 break; 1878 1879 case SO_PEEK_OFF: 1880 if (!READ_ONCE(sock->ops)->set_peek_off) 1881 return -EOPNOTSUPP; 1882 1883 v.val = READ_ONCE(sk->sk_peek_off); 1884 break; 1885 case SO_NOFCS: 1886 v.val = sock_flag(sk, SOCK_NOFCS); 1887 break; 1888 1889 case SO_BINDTODEVICE: 1890 return sock_getbindtodevice(sk, optval, optlen, len); 1891 1892 case SO_GET_FILTER: 1893 len = sk_get_filter(sk, optval, len); 1894 if (len < 0) 1895 return len; 1896 1897 goto lenout; 1898 1899 case SO_LOCK_FILTER: 1900 v.val = sock_flag(sk, SOCK_FILTER_LOCKED); 1901 break; 1902 1903 case SO_BPF_EXTENSIONS: 1904 v.val = bpf_tell_extensions(); 1905 break; 1906 1907 case SO_SELECT_ERR_QUEUE: 1908 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE); 1909 break; 1910 1911#ifdef CONFIG_NET_RX_BUSY_POLL 1912 case SO_BUSY_POLL: 1913 v.val = READ_ONCE(sk->sk_ll_usec); 1914 break; 1915 case SO_PREFER_BUSY_POLL: 1916 v.val = READ_ONCE(sk->sk_prefer_busy_poll); 1917 break; 1918#endif 1919 1920 case SO_MAX_PACING_RATE: 1921 /* The READ_ONCE() pair with the WRITE_ONCE() in sk_setsockopt() */ 1922 if (sizeof(v.ulval) != sizeof(v.val) && len >= sizeof(v.ulval)) { 1923 lv = sizeof(v.ulval); 1924 v.ulval = READ_ONCE(sk->sk_max_pacing_rate); 1925 } else { 1926 /* 32bit version */ 1927 v.val = min_t(unsigned long, ~0U, 1928 READ_ONCE(sk->sk_max_pacing_rate)); 1929 } 1930 break; 1931 1932 case SO_INCOMING_CPU: 1933 v.val = READ_ONCE(sk->sk_incoming_cpu); 1934 break; 1935 1936 case SO_MEMINFO: 1937 { 1938 u32 meminfo[SK_MEMINFO_VARS]; 1939 1940 sk_get_meminfo(sk, meminfo); 1941 1942 len = min_t(unsigned int, len, sizeof(meminfo)); 1943 if (copy_to_sockptr(optval, &meminfo, len)) 1944 return -EFAULT; 1945 1946 goto lenout; 1947 } 1948 1949#ifdef CONFIG_NET_RX_BUSY_POLL 1950 case SO_INCOMING_NAPI_ID: 1951 v.val = READ_ONCE(sk->sk_napi_id); 1952 1953 /* aggregate non-NAPI IDs down to 0 */ 1954 if (v.val < MIN_NAPI_ID) 1955 v.val = 0; 1956 1957 break; 1958#endif 1959 1960 case SO_COOKIE: 1961 lv = sizeof(u64); 1962 if (len < lv) 1963 return -EINVAL; 1964 v.val64 = sock_gen_cookie(sk); 1965 break; 1966 1967 case SO_ZEROCOPY: 1968 v.val = sock_flag(sk, SOCK_ZEROCOPY); 1969 break; 1970 1971 case SO_TXTIME: 1972 lv = sizeof(v.txtime); 1973 v.txtime.clockid = sk->sk_clockid; 1974 v.txtime.flags |= sk->sk_txtime_deadline_mode ? 1975 SOF_TXTIME_DEADLINE_MODE : 0; 1976 v.txtime.flags |= sk->sk_txtime_report_errors ? 1977 SOF_TXTIME_REPORT_ERRORS : 0; 1978 break; 1979 1980 case SO_BINDTOIFINDEX: 1981 v.val = READ_ONCE(sk->sk_bound_dev_if); 1982 break; 1983 1984 case SO_NETNS_COOKIE: 1985 lv = sizeof(u64); 1986 if (len != lv) 1987 return -EINVAL; 1988 v.val64 = sock_net(sk)->net_cookie; 1989 break; 1990 1991 case SO_BUF_LOCK: 1992 v.val = sk->sk_userlocks & SOCK_BUF_LOCK_MASK; 1993 break; 1994 1995 case SO_RESERVE_MEM: 1996 v.val = READ_ONCE(sk->sk_reserved_mem); 1997 break; 1998 1999 case SO_TXREHASH: 2000 /* Paired with WRITE_ONCE() in sk_setsockopt() */ 2001 v.val = READ_ONCE(sk->sk_txrehash); 2002 break; 2003 2004 default: 2005 /* We implement the SO_SNDLOWAT etc to not be settable 2006 * (1003.1g 7). 2007 */ 2008 return -ENOPROTOOPT; 2009 } 2010 2011 if (len > lv) 2012 len = lv; 2013 if (copy_to_sockptr(optval, &v, len)) 2014 return -EFAULT; 2015lenout: 2016 if (copy_to_sockptr(optlen, &len, sizeof(int))) 2017 return -EFAULT; 2018 return 0; 2019} 2020 2021int sock_getsockopt(struct socket *sock, int level, int optname, 2022 char __user *optval, int __user *optlen) 2023{ 2024 return sk_getsockopt(sock->sk, level, optname, 2025 USER_SOCKPTR(optval), 2026 USER_SOCKPTR(optlen)); 2027} 2028 2029/* 2030 * Initialize an sk_lock. 2031 * 2032 * (We also register the sk_lock with the lock validator.) 2033 */ 2034static inline void sock_lock_init(struct sock *sk) 2035{ 2036 if (sk->sk_kern_sock) 2037 sock_lock_init_class_and_name( 2038 sk, 2039 af_family_kern_slock_key_strings[sk->sk_family], 2040 af_family_kern_slock_keys + sk->sk_family, 2041 af_family_kern_key_strings[sk->sk_family], 2042 af_family_kern_keys + sk->sk_family); 2043 else 2044 sock_lock_init_class_and_name( 2045 sk, 2046 af_family_slock_key_strings[sk->sk_family], 2047 af_family_slock_keys + sk->sk_family, 2048 af_family_key_strings[sk->sk_family], 2049 af_family_keys + sk->sk_family); 2050} 2051 2052/* 2053 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet, 2054 * even temporarly, because of RCU lookups. sk_node should also be left as is. 2055 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end 2056 */ 2057static void sock_copy(struct sock *nsk, const struct sock *osk) 2058{ 2059 const struct proto *prot = READ_ONCE(osk->sk_prot); 2060#ifdef CONFIG_SECURITY_NETWORK 2061 void *sptr = nsk->sk_security; 2062#endif 2063 2064 /* If we move sk_tx_queue_mapping out of the private section, 2065 * we must check if sk_tx_queue_clear() is called after 2066 * sock_copy() in sk_clone_lock(). 2067 */ 2068 BUILD_BUG_ON(offsetof(struct sock, sk_tx_queue_mapping) < 2069 offsetof(struct sock, sk_dontcopy_begin) || 2070 offsetof(struct sock, sk_tx_queue_mapping) >= 2071 offsetof(struct sock, sk_dontcopy_end)); 2072 2073 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin)); 2074 2075 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end, 2076 prot->obj_size - offsetof(struct sock, sk_dontcopy_end)); 2077 2078#ifdef CONFIG_SECURITY_NETWORK 2079 nsk->sk_security = sptr; 2080 security_sk_clone(osk, nsk); 2081#endif 2082} 2083 2084static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority, 2085 int family) 2086{ 2087 struct sock *sk; 2088 struct kmem_cache *slab; 2089 2090 slab = prot->slab; 2091 if (slab != NULL) { 2092 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO); 2093 if (!sk) 2094 return sk; 2095 if (want_init_on_alloc(priority)) 2096 sk_prot_clear_nulls(sk, prot->obj_size); 2097 } else 2098 sk = kmalloc(prot->obj_size, priority); 2099 2100 if (sk != NULL) { 2101 if (security_sk_alloc(sk, family, priority)) 2102 goto out_free; 2103 2104 if (!try_module_get(prot->owner)) 2105 goto out_free_sec; 2106 } 2107 2108 return sk; 2109 2110out_free_sec: 2111 security_sk_free(sk); 2112out_free: 2113 if (slab != NULL) 2114 kmem_cache_free(slab, sk); 2115 else 2116 kfree(sk); 2117 return NULL; 2118} 2119 2120static void sk_prot_free(struct proto *prot, struct sock *sk) 2121{ 2122 struct kmem_cache *slab; 2123 struct module *owner; 2124 2125 owner = prot->owner; 2126 slab = prot->slab; 2127 2128 cgroup_sk_free(&sk->sk_cgrp_data); 2129 mem_cgroup_sk_free(sk); 2130 security_sk_free(sk); 2131 if (slab != NULL) 2132 kmem_cache_free(slab, sk); 2133 else 2134 kfree(sk); 2135 module_put(owner); 2136} 2137 2138/** 2139 * sk_alloc - All socket objects are allocated here 2140 * @net: the applicable net namespace 2141 * @family: protocol family 2142 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 2143 * @prot: struct proto associated with this new sock instance 2144 * @kern: is this to be a kernel socket? 2145 */ 2146struct sock *sk_alloc(struct net *net, int family, gfp_t priority, 2147 struct proto *prot, int kern) 2148{ 2149 struct sock *sk; 2150 2151 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family); 2152 if (sk) { 2153 sk->sk_family = family; 2154 /* 2155 * See comment in struct sock definition to understand 2156 * why we need sk_prot_creator -acme 2157 */ 2158 sk->sk_prot = sk->sk_prot_creator = prot; 2159 sk->sk_kern_sock = kern; 2160 sock_lock_init(sk); 2161 sk->sk_net_refcnt = kern ? 0 : 1; 2162 if (likely(sk->sk_net_refcnt)) { 2163 get_net_track(net, &sk->ns_tracker, priority); 2164 sock_inuse_add(net, 1); 2165 } else { 2166 __netns_tracker_alloc(net, &sk->ns_tracker, 2167 false, priority); 2168 } 2169 2170 sock_net_set(sk, net); 2171 refcount_set(&sk->sk_wmem_alloc, 1); 2172 2173 mem_cgroup_sk_alloc(sk); 2174 cgroup_sk_alloc(&sk->sk_cgrp_data); 2175 sock_update_classid(&sk->sk_cgrp_data); 2176 sock_update_netprioidx(&sk->sk_cgrp_data); 2177 sk_tx_queue_clear(sk); 2178 } 2179 2180 return sk; 2181} 2182EXPORT_SYMBOL(sk_alloc); 2183 2184/* Sockets having SOCK_RCU_FREE will call this function after one RCU 2185 * grace period. This is the case for UDP sockets and TCP listeners. 2186 */ 2187static void __sk_destruct(struct rcu_head *head) 2188{ 2189 struct sock *sk = container_of(head, struct sock, sk_rcu); 2190 struct sk_filter *filter; 2191 2192 if (sk->sk_destruct) 2193 sk->sk_destruct(sk); 2194 2195 filter = rcu_dereference_check(sk->sk_filter, 2196 refcount_read(&sk->sk_wmem_alloc) == 0); 2197 if (filter) { 2198 sk_filter_uncharge(sk, filter); 2199 RCU_INIT_POINTER(sk->sk_filter, NULL); 2200 } 2201 2202 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP); 2203 2204#ifdef CONFIG_BPF_SYSCALL 2205 bpf_sk_storage_free(sk); 2206#endif 2207 2208 if (atomic_read(&sk->sk_omem_alloc)) 2209 pr_debug("%s: optmem leakage (%d bytes) detected\n", 2210 __func__, atomic_read(&sk->sk_omem_alloc)); 2211 2212 if (sk->sk_frag.page) { 2213 put_page(sk->sk_frag.page); 2214 sk->sk_frag.page = NULL; 2215 } 2216 2217 /* We do not need to acquire sk->sk_peer_lock, we are the last user. */ 2218 put_cred(sk->sk_peer_cred); 2219 put_pid(sk->sk_peer_pid); 2220 2221 if (likely(sk->sk_net_refcnt)) 2222 put_net_track(sock_net(sk), &sk->ns_tracker); 2223 else 2224 __netns_tracker_free(sock_net(sk), &sk->ns_tracker, false); 2225 2226 sk_prot_free(sk->sk_prot_creator, sk); 2227} 2228 2229void sk_destruct(struct sock *sk) 2230{ 2231 bool use_call_rcu = sock_flag(sk, SOCK_RCU_FREE); 2232 2233 if (rcu_access_pointer(sk->sk_reuseport_cb)) { 2234 reuseport_detach_sock(sk); 2235 use_call_rcu = true; 2236 } 2237 2238 if (use_call_rcu) 2239 call_rcu(&sk->sk_rcu, __sk_destruct); 2240 else 2241 __sk_destruct(&sk->sk_rcu); 2242} 2243 2244static void __sk_free(struct sock *sk) 2245{ 2246 if (likely(sk->sk_net_refcnt)) 2247 sock_inuse_add(sock_net(sk), -1); 2248 2249 if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk))) 2250 sock_diag_broadcast_destroy(sk); 2251 else 2252 sk_destruct(sk); 2253} 2254 2255void sk_free(struct sock *sk) 2256{ 2257 /* 2258 * We subtract one from sk_wmem_alloc and can know if 2259 * some packets are still in some tx queue. 2260 * If not null, sock_wfree() will call __sk_free(sk) later 2261 */ 2262 if (refcount_dec_and_test(&sk->sk_wmem_alloc)) 2263 __sk_free(sk); 2264} 2265EXPORT_SYMBOL(sk_free); 2266 2267static void sk_init_common(struct sock *sk) 2268{ 2269 skb_queue_head_init(&sk->sk_receive_queue); 2270 skb_queue_head_init(&sk->sk_write_queue); 2271 skb_queue_head_init(&sk->sk_error_queue); 2272 2273 rwlock_init(&sk->sk_callback_lock); 2274 lockdep_set_class_and_name(&sk->sk_receive_queue.lock, 2275 af_rlock_keys + sk->sk_family, 2276 af_family_rlock_key_strings[sk->sk_family]); 2277 lockdep_set_class_and_name(&sk->sk_write_queue.lock, 2278 af_wlock_keys + sk->sk_family, 2279 af_family_wlock_key_strings[sk->sk_family]); 2280 lockdep_set_class_and_name(&sk->sk_error_queue.lock, 2281 af_elock_keys + sk->sk_family, 2282 af_family_elock_key_strings[sk->sk_family]); 2283 lockdep_set_class_and_name(&sk->sk_callback_lock, 2284 af_callback_keys + sk->sk_family, 2285 af_family_clock_key_strings[sk->sk_family]); 2286} 2287 2288/** 2289 * sk_clone_lock - clone a socket, and lock its clone 2290 * @sk: the socket to clone 2291 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 2292 * 2293 * Caller must unlock socket even in error path (bh_unlock_sock(newsk)) 2294 */ 2295struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority) 2296{ 2297 struct proto *prot = READ_ONCE(sk->sk_prot); 2298 struct sk_filter *filter; 2299 bool is_charged = true; 2300 struct sock *newsk; 2301 2302 newsk = sk_prot_alloc(prot, priority, sk->sk_family); 2303 if (!newsk) 2304 goto out; 2305 2306 sock_copy(newsk, sk); 2307 2308 newsk->sk_prot_creator = prot; 2309 2310 /* SANITY */ 2311 if (likely(newsk->sk_net_refcnt)) { 2312 get_net_track(sock_net(newsk), &newsk->ns_tracker, priority); 2313 sock_inuse_add(sock_net(newsk), 1); 2314 } else { 2315 /* Kernel sockets are not elevating the struct net refcount. 2316 * Instead, use a tracker to more easily detect if a layer 2317 * is not properly dismantling its kernel sockets at netns 2318 * destroy time. 2319 */ 2320 __netns_tracker_alloc(sock_net(newsk), &newsk->ns_tracker, 2321 false, priority); 2322 } 2323 sk_node_init(&newsk->sk_node); 2324 sock_lock_init(newsk); 2325 bh_lock_sock(newsk); 2326 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL; 2327 newsk->sk_backlog.len = 0; 2328 2329 atomic_set(&newsk->sk_rmem_alloc, 0); 2330 2331 /* sk_wmem_alloc set to one (see sk_free() and sock_wfree()) */ 2332 refcount_set(&newsk->sk_wmem_alloc, 1); 2333 2334 atomic_set(&newsk->sk_omem_alloc, 0); 2335 sk_init_common(newsk); 2336 2337 newsk->sk_dst_cache = NULL; 2338 newsk->sk_dst_pending_confirm = 0; 2339 newsk->sk_wmem_queued = 0; 2340 newsk->sk_forward_alloc = 0; 2341 newsk->sk_reserved_mem = 0; 2342 atomic_set(&newsk->sk_drops, 0); 2343 newsk->sk_send_head = NULL; 2344 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK; 2345 atomic_set(&newsk->sk_zckey, 0); 2346 2347 sock_reset_flag(newsk, SOCK_DONE); 2348 2349 /* sk->sk_memcg will be populated at accept() time */ 2350 newsk->sk_memcg = NULL; 2351 2352 cgroup_sk_clone(&newsk->sk_cgrp_data); 2353 2354 rcu_read_lock(); 2355 filter = rcu_dereference(sk->sk_filter); 2356 if (filter != NULL) 2357 /* though it's an empty new sock, the charging may fail 2358 * if sysctl_optmem_max was changed between creation of 2359 * original socket and cloning 2360 */ 2361 is_charged = sk_filter_charge(newsk, filter); 2362 RCU_INIT_POINTER(newsk->sk_filter, filter); 2363 rcu_read_unlock(); 2364 2365 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) { 2366 /* We need to make sure that we don't uncharge the new 2367 * socket if we couldn't charge it in the first place 2368 * as otherwise we uncharge the parent's filter. 2369 */ 2370 if (!is_charged) 2371 RCU_INIT_POINTER(newsk->sk_filter, NULL); 2372 sk_free_unlock_clone(newsk); 2373 newsk = NULL; 2374 goto out; 2375 } 2376 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL); 2377 2378 if (bpf_sk_storage_clone(sk, newsk)) { 2379 sk_free_unlock_clone(newsk); 2380 newsk = NULL; 2381 goto out; 2382 } 2383 2384 /* Clear sk_user_data if parent had the pointer tagged 2385 * as not suitable for copying when cloning. 2386 */ 2387 if (sk_user_data_is_nocopy(newsk)) 2388 newsk->sk_user_data = NULL; 2389 2390 newsk->sk_err = 0; 2391 newsk->sk_err_soft = 0; 2392 newsk->sk_priority = 0; 2393 newsk->sk_incoming_cpu = raw_smp_processor_id(); 2394 2395 /* Before updating sk_refcnt, we must commit prior changes to memory 2396 * (Documentation/RCU/rculist_nulls.rst for details) 2397 */ 2398 smp_wmb(); 2399 refcount_set(&newsk->sk_refcnt, 2); 2400 2401 sk_set_socket(newsk, NULL); 2402 sk_tx_queue_clear(newsk); 2403 RCU_INIT_POINTER(newsk->sk_wq, NULL); 2404 2405 if (newsk->sk_prot->sockets_allocated) 2406 sk_sockets_allocated_inc(newsk); 2407 2408 if (sock_needs_netstamp(sk) && newsk->sk_flags & SK_FLAGS_TIMESTAMP) 2409 net_enable_timestamp(); 2410out: 2411 return newsk; 2412} 2413EXPORT_SYMBOL_GPL(sk_clone_lock); 2414 2415void sk_free_unlock_clone(struct sock *sk) 2416{ 2417 /* It is still raw copy of parent, so invalidate 2418 * destructor and make plain sk_free() */ 2419 sk->sk_destruct = NULL; 2420 bh_unlock_sock(sk); 2421 sk_free(sk); 2422} 2423EXPORT_SYMBOL_GPL(sk_free_unlock_clone); 2424 2425static u32 sk_dst_gso_max_size(struct sock *sk, struct dst_entry *dst) 2426{ 2427 bool is_ipv6 = false; 2428 u32 max_size; 2429 2430#if IS_ENABLED(CONFIG_IPV6) 2431 is_ipv6 = (sk->sk_family == AF_INET6 && 2432 !ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr)); 2433#endif 2434 /* pairs with the WRITE_ONCE() in netif_set_gso(_ipv4)_max_size() */ 2435 max_size = is_ipv6 ? READ_ONCE(dst->dev->gso_max_size) : 2436 READ_ONCE(dst->dev->gso_ipv4_max_size); 2437 if (max_size > GSO_LEGACY_MAX_SIZE && !sk_is_tcp(sk)) 2438 max_size = GSO_LEGACY_MAX_SIZE; 2439 2440 return max_size - (MAX_TCP_HEADER + 1); 2441} 2442 2443void sk_setup_caps(struct sock *sk, struct dst_entry *dst) 2444{ 2445 u32 max_segs = 1; 2446 2447 sk->sk_route_caps = dst->dev->features; 2448 if (sk_is_tcp(sk)) 2449 sk->sk_route_caps |= NETIF_F_GSO; 2450 if (sk->sk_route_caps & NETIF_F_GSO) 2451 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE; 2452 if (unlikely(sk->sk_gso_disabled)) 2453 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 2454 if (sk_can_gso(sk)) { 2455 if (dst->header_len && !xfrm_dst_offload_ok(dst)) { 2456 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 2457 } else { 2458 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM; 2459 sk->sk_gso_max_size = sk_dst_gso_max_size(sk, dst); 2460 /* pairs with the WRITE_ONCE() in netif_set_gso_max_segs() */ 2461 max_segs = max_t(u32, READ_ONCE(dst->dev->gso_max_segs), 1); 2462 } 2463 } 2464 sk->sk_gso_max_segs = max_segs; 2465 sk_dst_set(sk, dst); 2466} 2467EXPORT_SYMBOL_GPL(sk_setup_caps); 2468 2469/* 2470 * Simple resource managers for sockets. 2471 */ 2472 2473 2474/* 2475 * Write buffer destructor automatically called from kfree_skb. 2476 */ 2477void sock_wfree(struct sk_buff *skb) 2478{ 2479 struct sock *sk = skb->sk; 2480 unsigned int len = skb->truesize; 2481 bool free; 2482 2483 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) { 2484 if (sock_flag(sk, SOCK_RCU_FREE) && 2485 sk->sk_write_space == sock_def_write_space) { 2486 rcu_read_lock(); 2487 free = refcount_sub_and_test(len, &sk->sk_wmem_alloc); 2488 sock_def_write_space_wfree(sk); 2489 rcu_read_unlock(); 2490 if (unlikely(free)) 2491 __sk_free(sk); 2492 return; 2493 } 2494 2495 /* 2496 * Keep a reference on sk_wmem_alloc, this will be released 2497 * after sk_write_space() call 2498 */ 2499 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc)); 2500 sk->sk_write_space(sk); 2501 len = 1; 2502 } 2503 /* 2504 * if sk_wmem_alloc reaches 0, we must finish what sk_free() 2505 * could not do because of in-flight packets 2506 */ 2507 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc)) 2508 __sk_free(sk); 2509} 2510EXPORT_SYMBOL(sock_wfree); 2511 2512/* This variant of sock_wfree() is used by TCP, 2513 * since it sets SOCK_USE_WRITE_QUEUE. 2514 */ 2515void __sock_wfree(struct sk_buff *skb) 2516{ 2517 struct sock *sk = skb->sk; 2518 2519 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc)) 2520 __sk_free(sk); 2521} 2522 2523void skb_set_owner_w(struct sk_buff *skb, struct sock *sk) 2524{ 2525 skb_orphan(skb); 2526 skb->sk = sk; 2527#ifdef CONFIG_INET 2528 if (unlikely(!sk_fullsock(sk))) { 2529 skb->destructor = sock_edemux; 2530 sock_hold(sk); 2531 return; 2532 } 2533#endif 2534 skb->destructor = sock_wfree; 2535 skb_set_hash_from_sk(skb, sk); 2536 /* 2537 * We used to take a refcount on sk, but following operation 2538 * is enough to guarantee sk_free() wont free this sock until 2539 * all in-flight packets are completed 2540 */ 2541 refcount_add(skb->truesize, &sk->sk_wmem_alloc); 2542} 2543EXPORT_SYMBOL(skb_set_owner_w); 2544 2545static bool can_skb_orphan_partial(const struct sk_buff *skb) 2546{ 2547#ifdef CONFIG_TLS_DEVICE 2548 /* Drivers depend on in-order delivery for crypto offload, 2549 * partial orphan breaks out-of-order-OK logic. 2550 */ 2551 if (skb->decrypted) 2552 return false; 2553#endif 2554 return (skb->destructor == sock_wfree || 2555 (IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree)); 2556} 2557 2558/* This helper is used by netem, as it can hold packets in its 2559 * delay queue. We want to allow the owner socket to send more 2560 * packets, as if they were already TX completed by a typical driver. 2561 * But we also want to keep skb->sk set because some packet schedulers 2562 * rely on it (sch_fq for example). 2563 */ 2564void skb_orphan_partial(struct sk_buff *skb) 2565{ 2566 if (skb_is_tcp_pure_ack(skb)) 2567 return; 2568 2569 if (can_skb_orphan_partial(skb) && skb_set_owner_sk_safe(skb, skb->sk)) 2570 return; 2571 2572 skb_orphan(skb); 2573} 2574EXPORT_SYMBOL(skb_orphan_partial); 2575 2576/* 2577 * Read buffer destructor automatically called from kfree_skb. 2578 */ 2579void sock_rfree(struct sk_buff *skb) 2580{ 2581 struct sock *sk = skb->sk; 2582 unsigned int len = skb->truesize; 2583 2584 atomic_sub(len, &sk->sk_rmem_alloc); 2585 sk_mem_uncharge(sk, len); 2586} 2587EXPORT_SYMBOL(sock_rfree); 2588 2589/* 2590 * Buffer destructor for skbs that are not used directly in read or write 2591 * path, e.g. for error handler skbs. Automatically called from kfree_skb. 2592 */ 2593void sock_efree(struct sk_buff *skb) 2594{ 2595 sock_put(skb->sk); 2596} 2597EXPORT_SYMBOL(sock_efree); 2598 2599/* Buffer destructor for prefetch/receive path where reference count may 2600 * not be held, e.g. for listen sockets. 2601 */ 2602#ifdef CONFIG_INET 2603void sock_pfree(struct sk_buff *skb) 2604{ 2605 if (sk_is_refcounted(skb->sk)) 2606 sock_gen_put(skb->sk); 2607} 2608EXPORT_SYMBOL(sock_pfree); 2609#endif /* CONFIG_INET */ 2610 2611kuid_t sock_i_uid(struct sock *sk) 2612{ 2613 kuid_t uid; 2614 2615 read_lock_bh(&sk->sk_callback_lock); 2616 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID; 2617 read_unlock_bh(&sk->sk_callback_lock); 2618 return uid; 2619} 2620EXPORT_SYMBOL(sock_i_uid); 2621 2622unsigned long __sock_i_ino(struct sock *sk) 2623{ 2624 unsigned long ino; 2625 2626 read_lock(&sk->sk_callback_lock); 2627 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0; 2628 read_unlock(&sk->sk_callback_lock); 2629 return ino; 2630} 2631EXPORT_SYMBOL(__sock_i_ino); 2632 2633unsigned long sock_i_ino(struct sock *sk) 2634{ 2635 unsigned long ino; 2636 2637 local_bh_disable(); 2638 ino = __sock_i_ino(sk); 2639 local_bh_enable(); 2640 return ino; 2641} 2642EXPORT_SYMBOL(sock_i_ino); 2643 2644/* 2645 * Allocate a skb from the socket's send buffer. 2646 */ 2647struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 2648 gfp_t priority) 2649{ 2650 if (force || 2651 refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) { 2652 struct sk_buff *skb = alloc_skb(size, priority); 2653 2654 if (skb) { 2655 skb_set_owner_w(skb, sk); 2656 return skb; 2657 } 2658 } 2659 return NULL; 2660} 2661EXPORT_SYMBOL(sock_wmalloc); 2662 2663static void sock_ofree(struct sk_buff *skb) 2664{ 2665 struct sock *sk = skb->sk; 2666 2667 atomic_sub(skb->truesize, &sk->sk_omem_alloc); 2668} 2669 2670struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size, 2671 gfp_t priority) 2672{ 2673 struct sk_buff *skb; 2674 2675 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */ 2676 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) > 2677 READ_ONCE(sysctl_optmem_max)) 2678 return NULL; 2679 2680 skb = alloc_skb(size, priority); 2681 if (!skb) 2682 return NULL; 2683 2684 atomic_add(skb->truesize, &sk->sk_omem_alloc); 2685 skb->sk = sk; 2686 skb->destructor = sock_ofree; 2687 return skb; 2688} 2689 2690/* 2691 * Allocate a memory block from the socket's option memory buffer. 2692 */ 2693void *sock_kmalloc(struct sock *sk, int size, gfp_t priority) 2694{ 2695 int optmem_max = READ_ONCE(sysctl_optmem_max); 2696 2697 if ((unsigned int)size <= optmem_max && 2698 atomic_read(&sk->sk_omem_alloc) + size < optmem_max) { 2699 void *mem; 2700 /* First do the add, to avoid the race if kmalloc 2701 * might sleep. 2702 */ 2703 atomic_add(size, &sk->sk_omem_alloc); 2704 mem = kmalloc(size, priority); 2705 if (mem) 2706 return mem; 2707 atomic_sub(size, &sk->sk_omem_alloc); 2708 } 2709 return NULL; 2710} 2711EXPORT_SYMBOL(sock_kmalloc); 2712 2713/* Free an option memory block. Note, we actually want the inline 2714 * here as this allows gcc to detect the nullify and fold away the 2715 * condition entirely. 2716 */ 2717static inline void __sock_kfree_s(struct sock *sk, void *mem, int size, 2718 const bool nullify) 2719{ 2720 if (WARN_ON_ONCE(!mem)) 2721 return; 2722 if (nullify) 2723 kfree_sensitive(mem); 2724 else 2725 kfree(mem); 2726 atomic_sub(size, &sk->sk_omem_alloc); 2727} 2728 2729void sock_kfree_s(struct sock *sk, void *mem, int size) 2730{ 2731 __sock_kfree_s(sk, mem, size, false); 2732} 2733EXPORT_SYMBOL(sock_kfree_s); 2734 2735void sock_kzfree_s(struct sock *sk, void *mem, int size) 2736{ 2737 __sock_kfree_s(sk, mem, size, true); 2738} 2739EXPORT_SYMBOL(sock_kzfree_s); 2740 2741/* It is almost wait_for_tcp_memory minus release_sock/lock_sock. 2742 I think, these locks should be removed for datagram sockets. 2743 */ 2744static long sock_wait_for_wmem(struct sock *sk, long timeo) 2745{ 2746 DEFINE_WAIT(wait); 2747 2748 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk); 2749 for (;;) { 2750 if (!timeo) 2751 break; 2752 if (signal_pending(current)) 2753 break; 2754 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 2755 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 2756 if (refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) 2757 break; 2758 if (READ_ONCE(sk->sk_shutdown) & SEND_SHUTDOWN) 2759 break; 2760 if (READ_ONCE(sk->sk_err)) 2761 break; 2762 timeo = schedule_timeout(timeo); 2763 } 2764 finish_wait(sk_sleep(sk), &wait); 2765 return timeo; 2766} 2767 2768 2769/* 2770 * Generic send/receive buffer handlers 2771 */ 2772 2773struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 2774 unsigned long data_len, int noblock, 2775 int *errcode, int max_page_order) 2776{ 2777 struct sk_buff *skb; 2778 long timeo; 2779 int err; 2780 2781 timeo = sock_sndtimeo(sk, noblock); 2782 for (;;) { 2783 err = sock_error(sk); 2784 if (err != 0) 2785 goto failure; 2786 2787 err = -EPIPE; 2788 if (READ_ONCE(sk->sk_shutdown) & SEND_SHUTDOWN) 2789 goto failure; 2790 2791 if (sk_wmem_alloc_get(sk) < READ_ONCE(sk->sk_sndbuf)) 2792 break; 2793 2794 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk); 2795 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 2796 err = -EAGAIN; 2797 if (!timeo) 2798 goto failure; 2799 if (signal_pending(current)) 2800 goto interrupted; 2801 timeo = sock_wait_for_wmem(sk, timeo); 2802 } 2803 skb = alloc_skb_with_frags(header_len, data_len, max_page_order, 2804 errcode, sk->sk_allocation); 2805 if (skb) 2806 skb_set_owner_w(skb, sk); 2807 return skb; 2808 2809interrupted: 2810 err = sock_intr_errno(timeo); 2811failure: 2812 *errcode = err; 2813 return NULL; 2814} 2815EXPORT_SYMBOL(sock_alloc_send_pskb); 2816 2817int __sock_cmsg_send(struct sock *sk, struct cmsghdr *cmsg, 2818 struct sockcm_cookie *sockc) 2819{ 2820 u32 tsflags; 2821 2822 switch (cmsg->cmsg_type) { 2823 case SO_MARK: 2824 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) && 2825 !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) 2826 return -EPERM; 2827 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 2828 return -EINVAL; 2829 sockc->mark = *(u32 *)CMSG_DATA(cmsg); 2830 break; 2831 case SO_TIMESTAMPING_OLD: 2832 case SO_TIMESTAMPING_NEW: 2833 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 2834 return -EINVAL; 2835 2836 tsflags = *(u32 *)CMSG_DATA(cmsg); 2837 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK) 2838 return -EINVAL; 2839 2840 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK; 2841 sockc->tsflags |= tsflags; 2842 break; 2843 case SCM_TXTIME: 2844 if (!sock_flag(sk, SOCK_TXTIME)) 2845 return -EINVAL; 2846 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64))) 2847 return -EINVAL; 2848 sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg)); 2849 break; 2850 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */ 2851 case SCM_RIGHTS: 2852 case SCM_CREDENTIALS: 2853 break; 2854 default: 2855 return -EINVAL; 2856 } 2857 return 0; 2858} 2859EXPORT_SYMBOL(__sock_cmsg_send); 2860 2861int sock_cmsg_send(struct sock *sk, struct msghdr *msg, 2862 struct sockcm_cookie *sockc) 2863{ 2864 struct cmsghdr *cmsg; 2865 int ret; 2866 2867 for_each_cmsghdr(cmsg, msg) { 2868 if (!CMSG_OK(msg, cmsg)) 2869 return -EINVAL; 2870 if (cmsg->cmsg_level != SOL_SOCKET) 2871 continue; 2872 ret = __sock_cmsg_send(sk, cmsg, sockc); 2873 if (ret) 2874 return ret; 2875 } 2876 return 0; 2877} 2878EXPORT_SYMBOL(sock_cmsg_send); 2879 2880static void sk_enter_memory_pressure(struct sock *sk) 2881{ 2882 if (!sk->sk_prot->enter_memory_pressure) 2883 return; 2884 2885 sk->sk_prot->enter_memory_pressure(sk); 2886} 2887 2888static void sk_leave_memory_pressure(struct sock *sk) 2889{ 2890 if (sk->sk_prot->leave_memory_pressure) { 2891 INDIRECT_CALL_INET_1(sk->sk_prot->leave_memory_pressure, 2892 tcp_leave_memory_pressure, sk); 2893 } else { 2894 unsigned long *memory_pressure = sk->sk_prot->memory_pressure; 2895 2896 if (memory_pressure && READ_ONCE(*memory_pressure)) 2897 WRITE_ONCE(*memory_pressure, 0); 2898 } 2899} 2900 2901DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key); 2902 2903/** 2904 * skb_page_frag_refill - check that a page_frag contains enough room 2905 * @sz: minimum size of the fragment we want to get 2906 * @pfrag: pointer to page_frag 2907 * @gfp: priority for memory allocation 2908 * 2909 * Note: While this allocator tries to use high order pages, there is 2910 * no guarantee that allocations succeed. Therefore, @sz MUST be 2911 * less or equal than PAGE_SIZE. 2912 */ 2913bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp) 2914{ 2915 if (pfrag->page) { 2916 if (page_ref_count(pfrag->page) == 1) { 2917 pfrag->offset = 0; 2918 return true; 2919 } 2920 if (pfrag->offset + sz <= pfrag->size) 2921 return true; 2922 put_page(pfrag->page); 2923 } 2924 2925 pfrag->offset = 0; 2926 if (SKB_FRAG_PAGE_ORDER && 2927 !static_branch_unlikely(&net_high_order_alloc_disable_key)) { 2928 /* Avoid direct reclaim but allow kswapd to wake */ 2929 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) | 2930 __GFP_COMP | __GFP_NOWARN | 2931 __GFP_NORETRY, 2932 SKB_FRAG_PAGE_ORDER); 2933 if (likely(pfrag->page)) { 2934 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER; 2935 return true; 2936 } 2937 } 2938 pfrag->page = alloc_page(gfp); 2939 if (likely(pfrag->page)) { 2940 pfrag->size = PAGE_SIZE; 2941 return true; 2942 } 2943 return false; 2944} 2945EXPORT_SYMBOL(skb_page_frag_refill); 2946 2947bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag) 2948{ 2949 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation))) 2950 return true; 2951 2952 sk_enter_memory_pressure(sk); 2953 sk_stream_moderate_sndbuf(sk); 2954 return false; 2955} 2956EXPORT_SYMBOL(sk_page_frag_refill); 2957 2958void __lock_sock(struct sock *sk) 2959 __releases(&sk->sk_lock.slock) 2960 __acquires(&sk->sk_lock.slock) 2961{ 2962 DEFINE_WAIT(wait); 2963 2964 for (;;) { 2965 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait, 2966 TASK_UNINTERRUPTIBLE); 2967 spin_unlock_bh(&sk->sk_lock.slock); 2968 schedule(); 2969 spin_lock_bh(&sk->sk_lock.slock); 2970 if (!sock_owned_by_user(sk)) 2971 break; 2972 } 2973 finish_wait(&sk->sk_lock.wq, &wait); 2974} 2975 2976void __release_sock(struct sock *sk) 2977 __releases(&sk->sk_lock.slock) 2978 __acquires(&sk->sk_lock.slock) 2979{ 2980 struct sk_buff *skb, *next; 2981 2982 while ((skb = sk->sk_backlog.head) != NULL) { 2983 sk->sk_backlog.head = sk->sk_backlog.tail = NULL; 2984 2985 spin_unlock_bh(&sk->sk_lock.slock); 2986 2987 do { 2988 next = skb->next; 2989 prefetch(next); 2990 DEBUG_NET_WARN_ON_ONCE(skb_dst_is_noref(skb)); 2991 skb_mark_not_on_list(skb); 2992 sk_backlog_rcv(sk, skb); 2993 2994 cond_resched(); 2995 2996 skb = next; 2997 } while (skb != NULL); 2998 2999 spin_lock_bh(&sk->sk_lock.slock); 3000 } 3001 3002 /* 3003 * Doing the zeroing here guarantee we can not loop forever 3004 * while a wild producer attempts to flood us. 3005 */ 3006 sk->sk_backlog.len = 0; 3007} 3008 3009void __sk_flush_backlog(struct sock *sk) 3010{ 3011 spin_lock_bh(&sk->sk_lock.slock); 3012 __release_sock(sk); 3013 spin_unlock_bh(&sk->sk_lock.slock); 3014} 3015EXPORT_SYMBOL_GPL(__sk_flush_backlog); 3016 3017/** 3018 * sk_wait_data - wait for data to arrive at sk_receive_queue 3019 * @sk: sock to wait on 3020 * @timeo: for how long 3021 * @skb: last skb seen on sk_receive_queue 3022 * 3023 * Now socket state including sk->sk_err is changed only under lock, 3024 * hence we may omit checks after joining wait queue. 3025 * We check receive queue before schedule() only as optimization; 3026 * it is very likely that release_sock() added new data. 3027 */ 3028int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb) 3029{ 3030 DEFINE_WAIT_FUNC(wait, woken_wake_function); 3031 int rc; 3032 3033 add_wait_queue(sk_sleep(sk), &wait); 3034 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk); 3035 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait); 3036 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk); 3037 remove_wait_queue(sk_sleep(sk), &wait); 3038 return rc; 3039} 3040EXPORT_SYMBOL(sk_wait_data); 3041 3042/** 3043 * __sk_mem_raise_allocated - increase memory_allocated 3044 * @sk: socket 3045 * @size: memory size to allocate 3046 * @amt: pages to allocate 3047 * @kind: allocation type 3048 * 3049 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc 3050 */ 3051int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind) 3052{ 3053 bool memcg_charge = mem_cgroup_sockets_enabled && sk->sk_memcg; 3054 struct proto *prot = sk->sk_prot; 3055 bool charged = true; 3056 long allocated; 3057 3058 sk_memory_allocated_add(sk, amt); 3059 allocated = sk_memory_allocated(sk); 3060 if (memcg_charge && 3061 !(charged = mem_cgroup_charge_skmem(sk->sk_memcg, amt, 3062 gfp_memcg_charge()))) 3063 goto suppress_allocation; 3064 3065 /* Under limit. */ 3066 if (allocated <= sk_prot_mem_limits(sk, 0)) { 3067 sk_leave_memory_pressure(sk); 3068 return 1; 3069 } 3070 3071 /* Under pressure. */ 3072 if (allocated > sk_prot_mem_limits(sk, 1)) 3073 sk_enter_memory_pressure(sk); 3074 3075 /* Over hard limit. */ 3076 if (allocated > sk_prot_mem_limits(sk, 2)) 3077 goto suppress_allocation; 3078 3079 /* guarantee minimum buffer size under pressure */ 3080 if (kind == SK_MEM_RECV) { 3081 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot)) 3082 return 1; 3083 3084 } else { /* SK_MEM_SEND */ 3085 int wmem0 = sk_get_wmem0(sk, prot); 3086 3087 if (sk->sk_type == SOCK_STREAM) { 3088 if (sk->sk_wmem_queued < wmem0) 3089 return 1; 3090 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) { 3091 return 1; 3092 } 3093 } 3094 3095 if (sk_has_memory_pressure(sk)) { 3096 u64 alloc; 3097 3098 if (!sk_under_memory_pressure(sk)) 3099 return 1; 3100 alloc = sk_sockets_allocated_read_positive(sk); 3101 if (sk_prot_mem_limits(sk, 2) > alloc * 3102 sk_mem_pages(sk->sk_wmem_queued + 3103 atomic_read(&sk->sk_rmem_alloc) + 3104 sk->sk_forward_alloc)) 3105 return 1; 3106 } 3107 3108suppress_allocation: 3109 3110 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) { 3111 sk_stream_moderate_sndbuf(sk); 3112 3113 /* Fail only if socket is _under_ its sndbuf. 3114 * In this case we cannot block, so that we have to fail. 3115 */ 3116 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf) { 3117 /* Force charge with __GFP_NOFAIL */ 3118 if (memcg_charge && !charged) { 3119 mem_cgroup_charge_skmem(sk->sk_memcg, amt, 3120 gfp_memcg_charge() | __GFP_NOFAIL); 3121 } 3122 return 1; 3123 } 3124 } 3125 3126 if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged)) 3127 trace_sock_exceed_buf_limit(sk, prot, allocated, kind); 3128 3129 sk_memory_allocated_sub(sk, amt); 3130 3131 if (memcg_charge && charged) 3132 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt); 3133 3134 return 0; 3135} 3136 3137/** 3138 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated 3139 * @sk: socket 3140 * @size: memory size to allocate 3141 * @kind: allocation type 3142 * 3143 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means 3144 * rmem allocation. This function assumes that protocols which have 3145 * memory_pressure use sk_wmem_queued as write buffer accounting. 3146 */ 3147int __sk_mem_schedule(struct sock *sk, int size, int kind) 3148{ 3149 int ret, amt = sk_mem_pages(size); 3150 3151 sk_forward_alloc_add(sk, amt << PAGE_SHIFT); 3152 ret = __sk_mem_raise_allocated(sk, size, amt, kind); 3153 if (!ret) 3154 sk_forward_alloc_add(sk, -(amt << PAGE_SHIFT)); 3155 return ret; 3156} 3157EXPORT_SYMBOL(__sk_mem_schedule); 3158 3159/** 3160 * __sk_mem_reduce_allocated - reclaim memory_allocated 3161 * @sk: socket 3162 * @amount: number of quanta 3163 * 3164 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc 3165 */ 3166void __sk_mem_reduce_allocated(struct sock *sk, int amount) 3167{ 3168 sk_memory_allocated_sub(sk, amount); 3169 3170 if (mem_cgroup_sockets_enabled && sk->sk_memcg) 3171 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount); 3172 3173 if (sk_under_global_memory_pressure(sk) && 3174 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0))) 3175 sk_leave_memory_pressure(sk); 3176} 3177 3178/** 3179 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated 3180 * @sk: socket 3181 * @amount: number of bytes (rounded down to a PAGE_SIZE multiple) 3182 */ 3183void __sk_mem_reclaim(struct sock *sk, int amount) 3184{ 3185 amount >>= PAGE_SHIFT; 3186 sk_forward_alloc_add(sk, -(amount << PAGE_SHIFT)); 3187 __sk_mem_reduce_allocated(sk, amount); 3188} 3189EXPORT_SYMBOL(__sk_mem_reclaim); 3190 3191int sk_set_peek_off(struct sock *sk, int val) 3192{ 3193 WRITE_ONCE(sk->sk_peek_off, val); 3194 return 0; 3195} 3196EXPORT_SYMBOL_GPL(sk_set_peek_off); 3197 3198/* 3199 * Set of default routines for initialising struct proto_ops when 3200 * the protocol does not support a particular function. In certain 3201 * cases where it makes no sense for a protocol to have a "do nothing" 3202 * function, some default processing is provided. 3203 */ 3204 3205int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len) 3206{ 3207 return -EOPNOTSUPP; 3208} 3209EXPORT_SYMBOL(sock_no_bind); 3210 3211int sock_no_connect(struct socket *sock, struct sockaddr *saddr, 3212 int len, int flags) 3213{ 3214 return -EOPNOTSUPP; 3215} 3216EXPORT_SYMBOL(sock_no_connect); 3217 3218int sock_no_socketpair(struct socket *sock1, struct socket *sock2) 3219{ 3220 return -EOPNOTSUPP; 3221} 3222EXPORT_SYMBOL(sock_no_socketpair); 3223 3224int sock_no_accept(struct socket *sock, struct socket *newsock, int flags, 3225 bool kern) 3226{ 3227 return -EOPNOTSUPP; 3228} 3229EXPORT_SYMBOL(sock_no_accept); 3230 3231int sock_no_getname(struct socket *sock, struct sockaddr *saddr, 3232 int peer) 3233{ 3234 return -EOPNOTSUPP; 3235} 3236EXPORT_SYMBOL(sock_no_getname); 3237 3238int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 3239{ 3240 return -EOPNOTSUPP; 3241} 3242EXPORT_SYMBOL(sock_no_ioctl); 3243 3244int sock_no_listen(struct socket *sock, int backlog) 3245{ 3246 return -EOPNOTSUPP; 3247} 3248EXPORT_SYMBOL(sock_no_listen); 3249 3250int sock_no_shutdown(struct socket *sock, int how) 3251{ 3252 return -EOPNOTSUPP; 3253} 3254EXPORT_SYMBOL(sock_no_shutdown); 3255 3256int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len) 3257{ 3258 return -EOPNOTSUPP; 3259} 3260EXPORT_SYMBOL(sock_no_sendmsg); 3261 3262int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len) 3263{ 3264 return -EOPNOTSUPP; 3265} 3266EXPORT_SYMBOL(sock_no_sendmsg_locked); 3267 3268int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len, 3269 int flags) 3270{ 3271 return -EOPNOTSUPP; 3272} 3273EXPORT_SYMBOL(sock_no_recvmsg); 3274 3275int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma) 3276{ 3277 /* Mirror missing mmap method error code */ 3278 return -ENODEV; 3279} 3280EXPORT_SYMBOL(sock_no_mmap); 3281 3282/* 3283 * When a file is received (via SCM_RIGHTS, etc), we must bump the 3284 * various sock-based usage counts. 3285 */ 3286void __receive_sock(struct file *file) 3287{ 3288 struct socket *sock; 3289 3290 sock = sock_from_file(file); 3291 if (sock) { 3292 sock_update_netprioidx(&sock->sk->sk_cgrp_data); 3293 sock_update_classid(&sock->sk->sk_cgrp_data); 3294 } 3295} 3296 3297/* 3298 * Default Socket Callbacks 3299 */ 3300 3301static void sock_def_wakeup(struct sock *sk) 3302{ 3303 struct socket_wq *wq; 3304 3305 rcu_read_lock(); 3306 wq = rcu_dereference(sk->sk_wq); 3307 if (skwq_has_sleeper(wq)) 3308 wake_up_interruptible_all(&wq->wait); 3309 rcu_read_unlock(); 3310} 3311 3312static void sock_def_error_report(struct sock *sk) 3313{ 3314 struct socket_wq *wq; 3315 3316 rcu_read_lock(); 3317 wq = rcu_dereference(sk->sk_wq); 3318 if (skwq_has_sleeper(wq)) 3319 wake_up_interruptible_poll(&wq->wait, EPOLLERR); 3320 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR); 3321 rcu_read_unlock(); 3322} 3323 3324void sock_def_readable(struct sock *sk) 3325{ 3326 struct socket_wq *wq; 3327 3328 trace_sk_data_ready(sk); 3329 3330 rcu_read_lock(); 3331 wq = rcu_dereference(sk->sk_wq); 3332 if (skwq_has_sleeper(wq)) 3333 wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI | 3334 EPOLLRDNORM | EPOLLRDBAND); 3335 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN); 3336 rcu_read_unlock(); 3337} 3338 3339static void sock_def_write_space(struct sock *sk) 3340{ 3341 struct socket_wq *wq; 3342 3343 rcu_read_lock(); 3344 3345 /* Do not wake up a writer until he can make "significant" 3346 * progress. --DaveM 3347 */ 3348 if (sock_writeable(sk)) { 3349 wq = rcu_dereference(sk->sk_wq); 3350 if (skwq_has_sleeper(wq)) 3351 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT | 3352 EPOLLWRNORM | EPOLLWRBAND); 3353 3354 /* Should agree with poll, otherwise some programs break */ 3355 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT); 3356 } 3357 3358 rcu_read_unlock(); 3359} 3360 3361/* An optimised version of sock_def_write_space(), should only be called 3362 * for SOCK_RCU_FREE sockets under RCU read section and after putting 3363 * ->sk_wmem_alloc. 3364 */ 3365static void sock_def_write_space_wfree(struct sock *sk) 3366{ 3367 /* Do not wake up a writer until he can make "significant" 3368 * progress. --DaveM 3369 */ 3370 if (sock_writeable(sk)) { 3371 struct socket_wq *wq = rcu_dereference(sk->sk_wq); 3372 3373 /* rely on refcount_sub from sock_wfree() */ 3374 smp_mb__after_atomic(); 3375 if (wq && waitqueue_active(&wq->wait)) 3376 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT | 3377 EPOLLWRNORM | EPOLLWRBAND); 3378 3379 /* Should agree with poll, otherwise some programs break */ 3380 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT); 3381 } 3382} 3383 3384static void sock_def_destruct(struct sock *sk) 3385{ 3386} 3387 3388void sk_send_sigurg(struct sock *sk) 3389{ 3390 if (sk->sk_socket && sk->sk_socket->file) 3391 if (send_sigurg(&sk->sk_socket->file->f_owner)) 3392 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI); 3393} 3394EXPORT_SYMBOL(sk_send_sigurg); 3395 3396void sk_reset_timer(struct sock *sk, struct timer_list* timer, 3397 unsigned long expires) 3398{ 3399 if (!mod_timer(timer, expires)) 3400 sock_hold(sk); 3401} 3402EXPORT_SYMBOL(sk_reset_timer); 3403 3404void sk_stop_timer(struct sock *sk, struct timer_list* timer) 3405{ 3406 if (del_timer(timer)) 3407 __sock_put(sk); 3408} 3409EXPORT_SYMBOL(sk_stop_timer); 3410 3411void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer) 3412{ 3413 if (del_timer_sync(timer)) 3414 __sock_put(sk); 3415} 3416EXPORT_SYMBOL(sk_stop_timer_sync); 3417 3418void sock_init_data_uid(struct socket *sock, struct sock *sk, kuid_t uid) 3419{ 3420 sk_init_common(sk); 3421 sk->sk_send_head = NULL; 3422 3423 timer_setup(&sk->sk_timer, NULL, 0); 3424 3425 sk->sk_allocation = GFP_KERNEL; 3426 sk->sk_rcvbuf = READ_ONCE(sysctl_rmem_default); 3427 sk->sk_sndbuf = READ_ONCE(sysctl_wmem_default); 3428 sk->sk_state = TCP_CLOSE; 3429 sk->sk_use_task_frag = true; 3430 sk_set_socket(sk, sock); 3431 3432 sock_set_flag(sk, SOCK_ZAPPED); 3433 3434 if (sock) { 3435 sk->sk_type = sock->type; 3436 RCU_INIT_POINTER(sk->sk_wq, &sock->wq); 3437 sock->sk = sk; 3438 } else { 3439 RCU_INIT_POINTER(sk->sk_wq, NULL); 3440 } 3441 sk->sk_uid = uid; 3442 3443 rwlock_init(&sk->sk_callback_lock); 3444 if (sk->sk_kern_sock) 3445 lockdep_set_class_and_name( 3446 &sk->sk_callback_lock, 3447 af_kern_callback_keys + sk->sk_family, 3448 af_family_kern_clock_key_strings[sk->sk_family]); 3449 else 3450 lockdep_set_class_and_name( 3451 &sk->sk_callback_lock, 3452 af_callback_keys + sk->sk_family, 3453 af_family_clock_key_strings[sk->sk_family]); 3454 3455 sk->sk_state_change = sock_def_wakeup; 3456 sk->sk_data_ready = sock_def_readable; 3457 sk->sk_write_space = sock_def_write_space; 3458 sk->sk_error_report = sock_def_error_report; 3459 sk->sk_destruct = sock_def_destruct; 3460 3461 sk->sk_frag.page = NULL; 3462 sk->sk_frag.offset = 0; 3463 sk->sk_peek_off = -1; 3464 3465 sk->sk_peer_pid = NULL; 3466 sk->sk_peer_cred = NULL; 3467 spin_lock_init(&sk->sk_peer_lock); 3468 3469 sk->sk_write_pending = 0; 3470 sk->sk_rcvlowat = 1; 3471 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; 3472 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT; 3473 3474 sk->sk_stamp = SK_DEFAULT_STAMP; 3475#if BITS_PER_LONG==32 3476 seqlock_init(&sk->sk_stamp_seq); 3477#endif 3478 atomic_set(&sk->sk_zckey, 0); 3479 3480#ifdef CONFIG_NET_RX_BUSY_POLL 3481 sk->sk_napi_id = 0; 3482 sk->sk_ll_usec = READ_ONCE(sysctl_net_busy_read); 3483#endif 3484 3485 sk->sk_max_pacing_rate = ~0UL; 3486 sk->sk_pacing_rate = ~0UL; 3487 WRITE_ONCE(sk->sk_pacing_shift, 10); 3488 sk->sk_incoming_cpu = -1; 3489 3490 sk_rx_queue_clear(sk); 3491 /* 3492 * Before updating sk_refcnt, we must commit prior changes to memory 3493 * (Documentation/RCU/rculist_nulls.rst for details) 3494 */ 3495 smp_wmb(); 3496 refcount_set(&sk->sk_refcnt, 1); 3497 atomic_set(&sk->sk_drops, 0); 3498} 3499EXPORT_SYMBOL(sock_init_data_uid); 3500 3501void sock_init_data(struct socket *sock, struct sock *sk) 3502{ 3503 kuid_t uid = sock ? 3504 SOCK_INODE(sock)->i_uid : 3505 make_kuid(sock_net(sk)->user_ns, 0); 3506 3507 sock_init_data_uid(sock, sk, uid); 3508} 3509EXPORT_SYMBOL(sock_init_data); 3510 3511void lock_sock_nested(struct sock *sk, int subclass) 3512{ 3513 /* The sk_lock has mutex_lock() semantics here. */ 3514 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_); 3515 3516 might_sleep(); 3517 spin_lock_bh(&sk->sk_lock.slock); 3518 if (sock_owned_by_user_nocheck(sk)) 3519 __lock_sock(sk); 3520 sk->sk_lock.owned = 1; 3521 spin_unlock_bh(&sk->sk_lock.slock); 3522} 3523EXPORT_SYMBOL(lock_sock_nested); 3524 3525void release_sock(struct sock *sk) 3526{ 3527 spin_lock_bh(&sk->sk_lock.slock); 3528 if (sk->sk_backlog.tail) 3529 __release_sock(sk); 3530 3531 /* Warning : release_cb() might need to release sk ownership, 3532 * ie call sock_release_ownership(sk) before us. 3533 */ 3534 if (sk->sk_prot->release_cb) 3535 sk->sk_prot->release_cb(sk); 3536 3537 sock_release_ownership(sk); 3538 if (waitqueue_active(&sk->sk_lock.wq)) 3539 wake_up(&sk->sk_lock.wq); 3540 spin_unlock_bh(&sk->sk_lock.slock); 3541} 3542EXPORT_SYMBOL(release_sock); 3543 3544bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock) 3545{ 3546 might_sleep(); 3547 spin_lock_bh(&sk->sk_lock.slock); 3548 3549 if (!sock_owned_by_user_nocheck(sk)) { 3550 /* 3551 * Fast path return with bottom halves disabled and 3552 * sock::sk_lock.slock held. 3553 * 3554 * The 'mutex' is not contended and holding 3555 * sock::sk_lock.slock prevents all other lockers to 3556 * proceed so the corresponding unlock_sock_fast() can 3557 * avoid the slow path of release_sock() completely and 3558 * just release slock. 3559 * 3560 * From a semantical POV this is equivalent to 'acquiring' 3561 * the 'mutex', hence the corresponding lockdep 3562 * mutex_release() has to happen in the fast path of 3563 * unlock_sock_fast(). 3564 */ 3565 return false; 3566 } 3567 3568 __lock_sock(sk); 3569 sk->sk_lock.owned = 1; 3570 __acquire(&sk->sk_lock.slock); 3571 spin_unlock_bh(&sk->sk_lock.slock); 3572 return true; 3573} 3574EXPORT_SYMBOL(__lock_sock_fast); 3575 3576int sock_gettstamp(struct socket *sock, void __user *userstamp, 3577 bool timeval, bool time32) 3578{ 3579 struct sock *sk = sock->sk; 3580 struct timespec64 ts; 3581 3582 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 3583 ts = ktime_to_timespec64(sock_read_timestamp(sk)); 3584 if (ts.tv_sec == -1) 3585 return -ENOENT; 3586 if (ts.tv_sec == 0) { 3587 ktime_t kt = ktime_get_real(); 3588 sock_write_timestamp(sk, kt); 3589 ts = ktime_to_timespec64(kt); 3590 } 3591 3592 if (timeval) 3593 ts.tv_nsec /= 1000; 3594 3595#ifdef CONFIG_COMPAT_32BIT_TIME 3596 if (time32) 3597 return put_old_timespec32(&ts, userstamp); 3598#endif 3599#ifdef CONFIG_SPARC64 3600 /* beware of padding in sparc64 timeval */ 3601 if (timeval && !in_compat_syscall()) { 3602 struct __kernel_old_timeval __user tv = { 3603 .tv_sec = ts.tv_sec, 3604 .tv_usec = ts.tv_nsec, 3605 }; 3606 if (copy_to_user(userstamp, &tv, sizeof(tv))) 3607 return -EFAULT; 3608 return 0; 3609 } 3610#endif 3611 return put_timespec64(&ts, userstamp); 3612} 3613EXPORT_SYMBOL(sock_gettstamp); 3614 3615void sock_enable_timestamp(struct sock *sk, enum sock_flags flag) 3616{ 3617 if (!sock_flag(sk, flag)) { 3618 unsigned long previous_flags = sk->sk_flags; 3619 3620 sock_set_flag(sk, flag); 3621 /* 3622 * we just set one of the two flags which require net 3623 * time stamping, but time stamping might have been on 3624 * already because of the other one 3625 */ 3626 if (sock_needs_netstamp(sk) && 3627 !(previous_flags & SK_FLAGS_TIMESTAMP)) 3628 net_enable_timestamp(); 3629 } 3630} 3631 3632int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, 3633 int level, int type) 3634{ 3635 struct sock_exterr_skb *serr; 3636 struct sk_buff *skb; 3637 int copied, err; 3638 3639 err = -EAGAIN; 3640 skb = sock_dequeue_err_skb(sk); 3641 if (skb == NULL) 3642 goto out; 3643 3644 copied = skb->len; 3645 if (copied > len) { 3646 msg->msg_flags |= MSG_TRUNC; 3647 copied = len; 3648 } 3649 err = skb_copy_datagram_msg(skb, 0, msg, copied); 3650 if (err) 3651 goto out_free_skb; 3652 3653 sock_recv_timestamp(msg, sk, skb); 3654 3655 serr = SKB_EXT_ERR(skb); 3656 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee); 3657 3658 msg->msg_flags |= MSG_ERRQUEUE; 3659 err = copied; 3660 3661out_free_skb: 3662 kfree_skb(skb); 3663out: 3664 return err; 3665} 3666EXPORT_SYMBOL(sock_recv_errqueue); 3667 3668/* 3669 * Get a socket option on an socket. 3670 * 3671 * FIX: POSIX 1003.1g is very ambiguous here. It states that 3672 * asynchronous errors should be reported by getsockopt. We assume 3673 * this means if you specify SO_ERROR (otherwise whats the point of it). 3674 */ 3675int sock_common_getsockopt(struct socket *sock, int level, int optname, 3676 char __user *optval, int __user *optlen) 3677{ 3678 struct sock *sk = sock->sk; 3679 3680 /* IPV6_ADDRFORM can change sk->sk_prot under us. */ 3681 return READ_ONCE(sk->sk_prot)->getsockopt(sk, level, optname, optval, optlen); 3682} 3683EXPORT_SYMBOL(sock_common_getsockopt); 3684 3685int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 3686 int flags) 3687{ 3688 struct sock *sk = sock->sk; 3689 int addr_len = 0; 3690 int err; 3691 3692 err = sk->sk_prot->recvmsg(sk, msg, size, flags, &addr_len); 3693 if (err >= 0) 3694 msg->msg_namelen = addr_len; 3695 return err; 3696} 3697EXPORT_SYMBOL(sock_common_recvmsg); 3698 3699/* 3700 * Set socket options on an inet socket. 3701 */ 3702int sock_common_setsockopt(struct socket *sock, int level, int optname, 3703 sockptr_t optval, unsigned int optlen) 3704{ 3705 struct sock *sk = sock->sk; 3706 3707 /* IPV6_ADDRFORM can change sk->sk_prot under us. */ 3708 return READ_ONCE(sk->sk_prot)->setsockopt(sk, level, optname, optval, optlen); 3709} 3710EXPORT_SYMBOL(sock_common_setsockopt); 3711 3712void sk_common_release(struct sock *sk) 3713{ 3714 if (sk->sk_prot->destroy) 3715 sk->sk_prot->destroy(sk); 3716 3717 /* 3718 * Observation: when sk_common_release is called, processes have 3719 * no access to socket. But net still has. 3720 * Step one, detach it from networking: 3721 * 3722 * A. Remove from hash tables. 3723 */ 3724 3725 sk->sk_prot->unhash(sk); 3726 3727 /* 3728 * In this point socket cannot receive new packets, but it is possible 3729 * that some packets are in flight because some CPU runs receiver and 3730 * did hash table lookup before we unhashed socket. They will achieve 3731 * receive queue and will be purged by socket destructor. 3732 * 3733 * Also we still have packets pending on receive queue and probably, 3734 * our own packets waiting in device queues. sock_destroy will drain 3735 * receive queue, but transmitted packets will delay socket destruction 3736 * until the last reference will be released. 3737 */ 3738 3739 sock_orphan(sk); 3740 3741 xfrm_sk_free_policy(sk); 3742 3743 sock_put(sk); 3744} 3745EXPORT_SYMBOL(sk_common_release); 3746 3747void sk_get_meminfo(const struct sock *sk, u32 *mem) 3748{ 3749 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS); 3750 3751 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk); 3752 mem[SK_MEMINFO_RCVBUF] = READ_ONCE(sk->sk_rcvbuf); 3753 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk); 3754 mem[SK_MEMINFO_SNDBUF] = READ_ONCE(sk->sk_sndbuf); 3755 mem[SK_MEMINFO_FWD_ALLOC] = sk_forward_alloc_get(sk); 3756 mem[SK_MEMINFO_WMEM_QUEUED] = READ_ONCE(sk->sk_wmem_queued); 3757 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc); 3758 mem[SK_MEMINFO_BACKLOG] = READ_ONCE(sk->sk_backlog.len); 3759 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops); 3760} 3761 3762#ifdef CONFIG_PROC_FS 3763static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR); 3764 3765int sock_prot_inuse_get(struct net *net, struct proto *prot) 3766{ 3767 int cpu, idx = prot->inuse_idx; 3768 int res = 0; 3769 3770 for_each_possible_cpu(cpu) 3771 res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx]; 3772 3773 return res >= 0 ? res : 0; 3774} 3775EXPORT_SYMBOL_GPL(sock_prot_inuse_get); 3776 3777int sock_inuse_get(struct net *net) 3778{ 3779 int cpu, res = 0; 3780 3781 for_each_possible_cpu(cpu) 3782 res += per_cpu_ptr(net->core.prot_inuse, cpu)->all; 3783 3784 return res; 3785} 3786 3787EXPORT_SYMBOL_GPL(sock_inuse_get); 3788 3789static int __net_init sock_inuse_init_net(struct net *net) 3790{ 3791 net->core.prot_inuse = alloc_percpu(struct prot_inuse); 3792 if (net->core.prot_inuse == NULL) 3793 return -ENOMEM; 3794 return 0; 3795} 3796 3797static void __net_exit sock_inuse_exit_net(struct net *net) 3798{ 3799 free_percpu(net->core.prot_inuse); 3800} 3801 3802static struct pernet_operations net_inuse_ops = { 3803 .init = sock_inuse_init_net, 3804 .exit = sock_inuse_exit_net, 3805}; 3806 3807static __init int net_inuse_init(void) 3808{ 3809 if (register_pernet_subsys(&net_inuse_ops)) 3810 panic("Cannot initialize net inuse counters"); 3811 3812 return 0; 3813} 3814 3815core_initcall(net_inuse_init); 3816 3817static int assign_proto_idx(struct proto *prot) 3818{ 3819 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR); 3820 3821 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) { 3822 pr_err("PROTO_INUSE_NR exhausted\n"); 3823 return -ENOSPC; 3824 } 3825 3826 set_bit(prot->inuse_idx, proto_inuse_idx); 3827 return 0; 3828} 3829 3830static void release_proto_idx(struct proto *prot) 3831{ 3832 if (prot->inuse_idx != PROTO_INUSE_NR - 1) 3833 clear_bit(prot->inuse_idx, proto_inuse_idx); 3834} 3835#else 3836static inline int assign_proto_idx(struct proto *prot) 3837{ 3838 return 0; 3839} 3840 3841static inline void release_proto_idx(struct proto *prot) 3842{ 3843} 3844 3845#endif 3846 3847static void tw_prot_cleanup(struct timewait_sock_ops *twsk_prot) 3848{ 3849 if (!twsk_prot) 3850 return; 3851 kfree(twsk_prot->twsk_slab_name); 3852 twsk_prot->twsk_slab_name = NULL; 3853 kmem_cache_destroy(twsk_prot->twsk_slab); 3854 twsk_prot->twsk_slab = NULL; 3855} 3856 3857static int tw_prot_init(const struct proto *prot) 3858{ 3859 struct timewait_sock_ops *twsk_prot = prot->twsk_prot; 3860 3861 if (!twsk_prot) 3862 return 0; 3863 3864 twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", 3865 prot->name); 3866 if (!twsk_prot->twsk_slab_name) 3867 return -ENOMEM; 3868 3869 twsk_prot->twsk_slab = 3870 kmem_cache_create(twsk_prot->twsk_slab_name, 3871 twsk_prot->twsk_obj_size, 0, 3872 SLAB_ACCOUNT | prot->slab_flags, 3873 NULL); 3874 if (!twsk_prot->twsk_slab) { 3875 pr_crit("%s: Can't create timewait sock SLAB cache!\n", 3876 prot->name); 3877 return -ENOMEM; 3878 } 3879 3880 return 0; 3881} 3882 3883static void req_prot_cleanup(struct request_sock_ops *rsk_prot) 3884{ 3885 if (!rsk_prot) 3886 return; 3887 kfree(rsk_prot->slab_name); 3888 rsk_prot->slab_name = NULL; 3889 kmem_cache_destroy(rsk_prot->slab); 3890 rsk_prot->slab = NULL; 3891} 3892 3893static int req_prot_init(const struct proto *prot) 3894{ 3895 struct request_sock_ops *rsk_prot = prot->rsk_prot; 3896 3897 if (!rsk_prot) 3898 return 0; 3899 3900 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", 3901 prot->name); 3902 if (!rsk_prot->slab_name) 3903 return -ENOMEM; 3904 3905 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name, 3906 rsk_prot->obj_size, 0, 3907 SLAB_ACCOUNT | prot->slab_flags, 3908 NULL); 3909 3910 if (!rsk_prot->slab) { 3911 pr_crit("%s: Can't create request sock SLAB cache!\n", 3912 prot->name); 3913 return -ENOMEM; 3914 } 3915 return 0; 3916} 3917 3918int proto_register(struct proto *prot, int alloc_slab) 3919{ 3920 int ret = -ENOBUFS; 3921 3922 if (prot->memory_allocated && !prot->sysctl_mem) { 3923 pr_err("%s: missing sysctl_mem\n", prot->name); 3924 return -EINVAL; 3925 } 3926 if (prot->memory_allocated && !prot->per_cpu_fw_alloc) { 3927 pr_err("%s: missing per_cpu_fw_alloc\n", prot->name); 3928 return -EINVAL; 3929 } 3930 if (alloc_slab) { 3931 prot->slab = kmem_cache_create_usercopy(prot->name, 3932 prot->obj_size, 0, 3933 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT | 3934 prot->slab_flags, 3935 prot->useroffset, prot->usersize, 3936 NULL); 3937 3938 if (prot->slab == NULL) { 3939 pr_crit("%s: Can't create sock SLAB cache!\n", 3940 prot->name); 3941 goto out; 3942 } 3943 3944 if (req_prot_init(prot)) 3945 goto out_free_request_sock_slab; 3946 3947 if (tw_prot_init(prot)) 3948 goto out_free_timewait_sock_slab; 3949 } 3950 3951 mutex_lock(&proto_list_mutex); 3952 ret = assign_proto_idx(prot); 3953 if (ret) { 3954 mutex_unlock(&proto_list_mutex); 3955 goto out_free_timewait_sock_slab; 3956 } 3957 list_add(&prot->node, &proto_list); 3958 mutex_unlock(&proto_list_mutex); 3959 return ret; 3960 3961out_free_timewait_sock_slab: 3962 if (alloc_slab) 3963 tw_prot_cleanup(prot->twsk_prot); 3964out_free_request_sock_slab: 3965 if (alloc_slab) { 3966 req_prot_cleanup(prot->rsk_prot); 3967 3968 kmem_cache_destroy(prot->slab); 3969 prot->slab = NULL; 3970 } 3971out: 3972 return ret; 3973} 3974EXPORT_SYMBOL(proto_register); 3975 3976void proto_unregister(struct proto *prot) 3977{ 3978 mutex_lock(&proto_list_mutex); 3979 release_proto_idx(prot); 3980 list_del(&prot->node); 3981 mutex_unlock(&proto_list_mutex); 3982 3983 kmem_cache_destroy(prot->slab); 3984 prot->slab = NULL; 3985 3986 req_prot_cleanup(prot->rsk_prot); 3987 tw_prot_cleanup(prot->twsk_prot); 3988} 3989EXPORT_SYMBOL(proto_unregister); 3990 3991int sock_load_diag_module(int family, int protocol) 3992{ 3993 if (!protocol) { 3994 if (!sock_is_registered(family)) 3995 return -ENOENT; 3996 3997 return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK, 3998 NETLINK_SOCK_DIAG, family); 3999 } 4000 4001#ifdef CONFIG_INET 4002 if (family == AF_INET && 4003 protocol != IPPROTO_RAW && 4004 protocol < MAX_INET_PROTOS && 4005 !rcu_access_pointer(inet_protos[protocol])) 4006 return -ENOENT; 4007#endif 4008 4009 return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK, 4010 NETLINK_SOCK_DIAG, family, protocol); 4011} 4012EXPORT_SYMBOL(sock_load_diag_module); 4013 4014#ifdef CONFIG_PROC_FS 4015static void *proto_seq_start(struct seq_file *seq, loff_t *pos) 4016 __acquires(proto_list_mutex) 4017{ 4018 mutex_lock(&proto_list_mutex); 4019 return seq_list_start_head(&proto_list, *pos); 4020} 4021 4022static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos) 4023{ 4024 return seq_list_next(v, &proto_list, pos); 4025} 4026 4027static void proto_seq_stop(struct seq_file *seq, void *v) 4028 __releases(proto_list_mutex) 4029{ 4030 mutex_unlock(&proto_list_mutex); 4031} 4032 4033static char proto_method_implemented(const void *method) 4034{ 4035 return method == NULL ? 'n' : 'y'; 4036} 4037static long sock_prot_memory_allocated(struct proto *proto) 4038{ 4039 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L; 4040} 4041 4042static const char *sock_prot_memory_pressure(struct proto *proto) 4043{ 4044 return proto->memory_pressure != NULL ? 4045 proto_memory_pressure(proto) ? "yes" : "no" : "NI"; 4046} 4047 4048static void proto_seq_printf(struct seq_file *seq, struct proto *proto) 4049{ 4050 4051 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s " 4052 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n", 4053 proto->name, 4054 proto->obj_size, 4055 sock_prot_inuse_get(seq_file_net(seq), proto), 4056 sock_prot_memory_allocated(proto), 4057 sock_prot_memory_pressure(proto), 4058 proto->max_header, 4059 proto->slab == NULL ? "no" : "yes", 4060 module_name(proto->owner), 4061 proto_method_implemented(proto->close), 4062 proto_method_implemented(proto->connect), 4063 proto_method_implemented(proto->disconnect), 4064 proto_method_implemented(proto->accept), 4065 proto_method_implemented(proto->ioctl), 4066 proto_method_implemented(proto->init), 4067 proto_method_implemented(proto->destroy), 4068 proto_method_implemented(proto->shutdown), 4069 proto_method_implemented(proto->setsockopt), 4070 proto_method_implemented(proto->getsockopt), 4071 proto_method_implemented(proto->sendmsg), 4072 proto_method_implemented(proto->recvmsg), 4073 proto_method_implemented(proto->bind), 4074 proto_method_implemented(proto->backlog_rcv), 4075 proto_method_implemented(proto->hash), 4076 proto_method_implemented(proto->unhash), 4077 proto_method_implemented(proto->get_port), 4078 proto_method_implemented(proto->enter_memory_pressure)); 4079} 4080 4081static int proto_seq_show(struct seq_file *seq, void *v) 4082{ 4083 if (v == &proto_list) 4084 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s", 4085 "protocol", 4086 "size", 4087 "sockets", 4088 "memory", 4089 "press", 4090 "maxhdr", 4091 "slab", 4092 "module", 4093 "cl co di ac io in de sh ss gs se re bi br ha uh gp em\n"); 4094 else 4095 proto_seq_printf(seq, list_entry(v, struct proto, node)); 4096 return 0; 4097} 4098 4099static const struct seq_operations proto_seq_ops = { 4100 .start = proto_seq_start, 4101 .next = proto_seq_next, 4102 .stop = proto_seq_stop, 4103 .show = proto_seq_show, 4104}; 4105 4106static __net_init int proto_init_net(struct net *net) 4107{ 4108 if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops, 4109 sizeof(struct seq_net_private))) 4110 return -ENOMEM; 4111 4112 return 0; 4113} 4114 4115static __net_exit void proto_exit_net(struct net *net) 4116{ 4117 remove_proc_entry("protocols", net->proc_net); 4118} 4119 4120 4121static __net_initdata struct pernet_operations proto_net_ops = { 4122 .init = proto_init_net, 4123 .exit = proto_exit_net, 4124}; 4125 4126static int __init proto_init(void) 4127{ 4128 return register_pernet_subsys(&proto_net_ops); 4129} 4130 4131subsys_initcall(proto_init); 4132 4133#endif /* PROC_FS */ 4134 4135#ifdef CONFIG_NET_RX_BUSY_POLL 4136bool sk_busy_loop_end(void *p, unsigned long start_time) 4137{ 4138 struct sock *sk = p; 4139 4140 if (!skb_queue_empty_lockless(&sk->sk_receive_queue)) 4141 return true; 4142 4143 if (sk_is_udp(sk) && 4144 !skb_queue_empty_lockless(&udp_sk(sk)->reader_queue)) 4145 return true; 4146 4147 return sk_busy_loop_timeout(sk, start_time); 4148} 4149EXPORT_SYMBOL(sk_busy_loop_end); 4150#endif /* CONFIG_NET_RX_BUSY_POLL */ 4151 4152int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len) 4153{ 4154 if (!sk->sk_prot->bind_add) 4155 return -EOPNOTSUPP; 4156 return sk->sk_prot->bind_add(sk, addr, addr_len); 4157} 4158EXPORT_SYMBOL(sock_bind_add); 4159 4160/* Copy 'size' bytes from userspace and return `size` back to userspace */ 4161int sock_ioctl_inout(struct sock *sk, unsigned int cmd, 4162 void __user *arg, void *karg, size_t size) 4163{ 4164 int ret; 4165 4166 if (copy_from_user(karg, arg, size)) 4167 return -EFAULT; 4168 4169 ret = READ_ONCE(sk->sk_prot)->ioctl(sk, cmd, karg); 4170 if (ret) 4171 return ret; 4172 4173 if (copy_to_user(arg, karg, size)) 4174 return -EFAULT; 4175 4176 return 0; 4177} 4178EXPORT_SYMBOL(sock_ioctl_inout); 4179 4180/* This is the most common ioctl prep function, where the result (4 bytes) is 4181 * copied back to userspace if the ioctl() returns successfully. No input is 4182 * copied from userspace as input argument. 4183 */ 4184static int sock_ioctl_out(struct sock *sk, unsigned int cmd, void __user *arg) 4185{ 4186 int ret, karg = 0; 4187 4188 ret = READ_ONCE(sk->sk_prot)->ioctl(sk, cmd, &karg); 4189 if (ret) 4190 return ret; 4191 4192 return put_user(karg, (int __user *)arg); 4193} 4194 4195/* A wrapper around sock ioctls, which copies the data from userspace 4196 * (depending on the protocol/ioctl), and copies back the result to userspace. 4197 * The main motivation for this function is to pass kernel memory to the 4198 * protocol ioctl callbacks, instead of userspace memory. 4199 */ 4200int sk_ioctl(struct sock *sk, unsigned int cmd, void __user *arg) 4201{ 4202 int rc = 1; 4203 4204 if (sk->sk_type == SOCK_RAW && sk->sk_family == AF_INET) 4205 rc = ipmr_sk_ioctl(sk, cmd, arg); 4206 else if (sk->sk_type == SOCK_RAW && sk->sk_family == AF_INET6) 4207 rc = ip6mr_sk_ioctl(sk, cmd, arg); 4208 else if (sk_is_phonet(sk)) 4209 rc = phonet_sk_ioctl(sk, cmd, arg); 4210 4211 /* If ioctl was processed, returns its value */ 4212 if (rc <= 0) 4213 return rc; 4214 4215 /* Otherwise call the default handler */ 4216 return sock_ioctl_out(sk, cmd, arg); 4217} 4218EXPORT_SYMBOL(sk_ioctl); 4219