1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * INET An implementation of the TCP/IP protocol suite for the LINUX 4 * operating system. INET is implemented using the BSD Socket 5 * interface as the means of communication with the user level. 6 * 7 * Generic socket support routines. Memory allocators, socket lock/release 8 * handler for protocols to use and generic option handler. 9 * 10 * Authors: Ross Biro 11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 12 * Florian La Roche, <flla@stud.uni-sb.de> 13 * Alan Cox, <A.Cox@swansea.ac.uk> 14 * 15 * Fixes: 16 * Alan Cox : Numerous verify_area() problems 17 * Alan Cox : Connecting on a connecting socket 18 * now returns an error for tcp. 19 * Alan Cox : sock->protocol is set correctly. 20 * and is not sometimes left as 0. 21 * Alan Cox : connect handles icmp errors on a 22 * connect properly. Unfortunately there 23 * is a restart syscall nasty there. I 24 * can't match BSD without hacking the C 25 * library. Ideas urgently sought! 26 * Alan Cox : Disallow bind() to addresses that are 27 * not ours - especially broadcast ones!! 28 * Alan Cox : Socket 1024 _IS_ ok for users. (fencepost) 29 * Alan Cox : sock_wfree/sock_rfree don't destroy sockets, 30 * instead they leave that for the DESTROY timer. 31 * Alan Cox : Clean up error flag in accept 32 * Alan Cox : TCP ack handling is buggy, the DESTROY timer 33 * was buggy. Put a remove_sock() in the handler 34 * for memory when we hit 0. Also altered the timer 35 * code. The ACK stuff can wait and needs major 36 * TCP layer surgery. 37 * Alan Cox : Fixed TCP ack bug, removed remove sock 38 * and fixed timer/inet_bh race. 39 * Alan Cox : Added zapped flag for TCP 40 * Alan Cox : Move kfree_skb into skbuff.c and tidied up surplus code 41 * Alan Cox : for new sk_buff allocations wmalloc/rmalloc now call alloc_skb 42 * Alan Cox : kfree_s calls now are kfree_skbmem so we can track skb resources 43 * Alan Cox : Supports socket option broadcast now as does udp. Packet and raw need fixing. 44 * Alan Cox : Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so... 45 * Rick Sladkey : Relaxed UDP rules for matching packets. 46 * C.E.Hawkins : IFF_PROMISC/SIOCGHWADDR support 47 * Pauline Middelink : identd support 48 * Alan Cox : Fixed connect() taking signals I think. 49 * Alan Cox : SO_LINGER supported 50 * Alan Cox : Error reporting fixes 51 * Anonymous : inet_create tidied up (sk->reuse setting) 52 * Alan Cox : inet sockets don't set sk->type! 53 * Alan Cox : Split socket option code 54 * Alan Cox : Callbacks 55 * Alan Cox : Nagle flag for Charles & Johannes stuff 56 * Alex : Removed restriction on inet fioctl 57 * Alan Cox : Splitting INET from NET core 58 * Alan Cox : Fixed bogus SO_TYPE handling in getsockopt() 59 * Adam Caldwell : Missing return in SO_DONTROUTE/SO_DEBUG code 60 * Alan Cox : Split IP from generic code 61 * Alan Cox : New kfree_skbmem() 62 * Alan Cox : Make SO_DEBUG superuser only. 63 * Alan Cox : Allow anyone to clear SO_DEBUG 64 * (compatibility fix) 65 * Alan Cox : Added optimistic memory grabbing for AF_UNIX throughput. 66 * Alan Cox : Allocator for a socket is settable. 67 * Alan Cox : SO_ERROR includes soft errors. 68 * Alan Cox : Allow NULL arguments on some SO_ opts 69 * Alan Cox : Generic socket allocation to make hooks 70 * easier (suggested by Craig Metz). 71 * Michael Pall : SO_ERROR returns positive errno again 72 * Steve Whitehouse: Added default destructor to free 73 * protocol private data. 74 * Steve Whitehouse: Added various other default routines 75 * common to several socket families. 76 * Chris Evans : Call suser() check last on F_SETOWN 77 * Jay Schulist : Added SO_ATTACH_FILTER and SO_DETACH_FILTER. 78 * Andi Kleen : Add sock_kmalloc()/sock_kfree_s() 79 * Andi Kleen : Fix write_space callback 80 * Chris Evans : Security fixes - signedness again 81 * Arnaldo C. Melo : cleanups, use skb_queue_purge 82 * 83 * To Fix: 84 */ 85 86 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 87 88 #include <asm/unaligned.h> 89 #include <linux/capability.h> 90 #include <linux/errno.h> 91 #include <linux/errqueue.h> 92 #include <linux/types.h> 93 #include <linux/socket.h> 94 #include <linux/in.h> 95 #include <linux/kernel.h> 96 #include <linux/module.h> 97 #include <linux/proc_fs.h> 98 #include <linux/seq_file.h> 99 #include <linux/sched.h> 100 #include <linux/sched/mm.h> 101 #include <linux/timer.h> 102 #include <linux/string.h> 103 #include <linux/sockios.h> 104 #include <linux/net.h> 105 #include <linux/mm.h> 106 #include <linux/slab.h> 107 #include <linux/interrupt.h> 108 #include <linux/poll.h> 109 #include <linux/tcp.h> 110 #include <linux/init.h> 111 #include <linux/highmem.h> 112 #include <linux/user_namespace.h> 113 #include <linux/static_key.h> 114 #include <linux/memcontrol.h> 115 #include <linux/prefetch.h> 116 #include <linux/compat.h> 117 #include <linux/mroute.h> 118 #include <linux/mroute6.h> 119 #include <linux/icmpv6.h> 120 121 #include <linux/uaccess.h> 122 123 #include <linux/netdevice.h> 124 #include <net/protocol.h> 125 #include <linux/skbuff.h> 126 #include <net/net_namespace.h> 127 #include <net/request_sock.h> 128 #include <net/sock.h> 129 #include <linux/net_tstamp.h> 130 #include <net/xfrm.h> 131 #include <linux/ipsec.h> 132 #include <net/cls_cgroup.h> 133 #include <net/netprio_cgroup.h> 134 #include <linux/sock_diag.h> 135 136 #include <linux/filter.h> 137 #include <net/sock_reuseport.h> 138 #include <net/bpf_sk_storage.h> 139 140 #include <trace/events/sock.h> 141 142 #include <net/tcp.h> 143 #include <net/busy_poll.h> 144 #include <net/phonet/phonet.h> 145 146 #include <linux/ethtool.h> 147 148 #include "dev.h" 149 150 static DEFINE_MUTEX(proto_list_mutex); 151 static LIST_HEAD(proto_list); 152 153 static void sock_def_write_space_wfree(struct sock *sk); 154 static void sock_def_write_space(struct sock *sk); 155 156 /** 157 * sk_ns_capable - General socket capability test 158 * @sk: Socket to use a capability on or through 159 * @user_ns: The user namespace of the capability to use 160 * @cap: The capability to use 161 * 162 * Test to see if the opener of the socket had when the socket was 163 * created and the current process has the capability @cap in the user 164 * namespace @user_ns. 165 */ 166 bool sk_ns_capable(const struct sock *sk, 167 struct user_namespace *user_ns, int cap) 168 { 169 return file_ns_capable(sk->sk_socket->file, user_ns, cap) && 170 ns_capable(user_ns, cap); 171 } 172 EXPORT_SYMBOL(sk_ns_capable); 173 174 /** 175 * sk_capable - Socket global capability test 176 * @sk: Socket to use a capability on or through 177 * @cap: The global capability to use 178 * 179 * Test to see if the opener of the socket had when the socket was 180 * created and the current process has the capability @cap in all user 181 * namespaces. 182 */ 183 bool sk_capable(const struct sock *sk, int cap) 184 { 185 return sk_ns_capable(sk, &init_user_ns, cap); 186 } 187 EXPORT_SYMBOL(sk_capable); 188 189 /** 190 * sk_net_capable - Network namespace socket capability test 191 * @sk: Socket to use a capability on or through 192 * @cap: The capability to use 193 * 194 * Test to see if the opener of the socket had when the socket was created 195 * and the current process has the capability @cap over the network namespace 196 * the socket is a member of. 197 */ 198 bool sk_net_capable(const struct sock *sk, int cap) 199 { 200 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap); 201 } 202 EXPORT_SYMBOL(sk_net_capable); 203 204 /* 205 * Each address family might have different locking rules, so we have 206 * one slock key per address family and separate keys for internal and 207 * userspace sockets. 208 */ 209 static struct lock_class_key af_family_keys[AF_MAX]; 210 static struct lock_class_key af_family_kern_keys[AF_MAX]; 211 static struct lock_class_key af_family_slock_keys[AF_MAX]; 212 static struct lock_class_key af_family_kern_slock_keys[AF_MAX]; 213 214 /* 215 * Make lock validator output more readable. (we pre-construct these 216 * strings build-time, so that runtime initialization of socket 217 * locks is fast): 218 */ 219 220 #define _sock_locks(x) \ 221 x "AF_UNSPEC", x "AF_UNIX" , x "AF_INET" , \ 222 x "AF_AX25" , x "AF_IPX" , x "AF_APPLETALK", \ 223 x "AF_NETROM", x "AF_BRIDGE" , x "AF_ATMPVC" , \ 224 x "AF_X25" , x "AF_INET6" , x "AF_ROSE" , \ 225 x "AF_DECnet", x "AF_NETBEUI" , x "AF_SECURITY" , \ 226 x "AF_KEY" , x "AF_NETLINK" , x "AF_PACKET" , \ 227 x "AF_ASH" , x "AF_ECONET" , x "AF_ATMSVC" , \ 228 x "AF_RDS" , x "AF_SNA" , x "AF_IRDA" , \ 229 x "AF_PPPOX" , x "AF_WANPIPE" , x "AF_LLC" , \ 230 x "27" , x "28" , x "AF_CAN" , \ 231 x "AF_TIPC" , x "AF_BLUETOOTH", x "IUCV" , \ 232 x "AF_RXRPC" , x "AF_ISDN" , x "AF_PHONET" , \ 233 x "AF_IEEE802154", x "AF_CAIF" , x "AF_ALG" , \ 234 x "AF_NFC" , x "AF_VSOCK" , x "AF_KCM" , \ 235 x "AF_QIPCRTR", x "AF_SMC" , x "AF_XDP" , \ 236 x "AF_MCTP" , \ 237 x "AF_MAX" 238 239 static const char *const af_family_key_strings[AF_MAX+1] = { 240 _sock_locks("sk_lock-") 241 }; 242 static const char *const af_family_slock_key_strings[AF_MAX+1] = { 243 _sock_locks("slock-") 244 }; 245 static const char *const af_family_clock_key_strings[AF_MAX+1] = { 246 _sock_locks("clock-") 247 }; 248 249 static const char *const af_family_kern_key_strings[AF_MAX+1] = { 250 _sock_locks("k-sk_lock-") 251 }; 252 static const char *const af_family_kern_slock_key_strings[AF_MAX+1] = { 253 _sock_locks("k-slock-") 254 }; 255 static const char *const af_family_kern_clock_key_strings[AF_MAX+1] = { 256 _sock_locks("k-clock-") 257 }; 258 static const char *const af_family_rlock_key_strings[AF_MAX+1] = { 259 _sock_locks("rlock-") 260 }; 261 static const char *const af_family_wlock_key_strings[AF_MAX+1] = { 262 _sock_locks("wlock-") 263 }; 264 static const char *const af_family_elock_key_strings[AF_MAX+1] = { 265 _sock_locks("elock-") 266 }; 267 268 /* 269 * sk_callback_lock and sk queues locking rules are per-address-family, 270 * so split the lock classes by using a per-AF key: 271 */ 272 static struct lock_class_key af_callback_keys[AF_MAX]; 273 static struct lock_class_key af_rlock_keys[AF_MAX]; 274 static struct lock_class_key af_wlock_keys[AF_MAX]; 275 static struct lock_class_key af_elock_keys[AF_MAX]; 276 static struct lock_class_key af_kern_callback_keys[AF_MAX]; 277 278 /* Run time adjustable parameters. */ 279 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX; 280 EXPORT_SYMBOL(sysctl_wmem_max); 281 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX; 282 EXPORT_SYMBOL(sysctl_rmem_max); 283 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX; 284 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX; 285 286 /* Maximal space eaten by iovec or ancillary data plus some space */ 287 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512); 288 EXPORT_SYMBOL(sysctl_optmem_max); 289 290 int sysctl_tstamp_allow_data __read_mostly = 1; 291 292 DEFINE_STATIC_KEY_FALSE(memalloc_socks_key); 293 EXPORT_SYMBOL_GPL(memalloc_socks_key); 294 295 /** 296 * sk_set_memalloc - sets %SOCK_MEMALLOC 297 * @sk: socket to set it on 298 * 299 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves. 300 * It's the responsibility of the admin to adjust min_free_kbytes 301 * to meet the requirements 302 */ 303 void sk_set_memalloc(struct sock *sk) 304 { 305 sock_set_flag(sk, SOCK_MEMALLOC); 306 sk->sk_allocation |= __GFP_MEMALLOC; 307 static_branch_inc(&memalloc_socks_key); 308 } 309 EXPORT_SYMBOL_GPL(sk_set_memalloc); 310 311 void sk_clear_memalloc(struct sock *sk) 312 { 313 sock_reset_flag(sk, SOCK_MEMALLOC); 314 sk->sk_allocation &= ~__GFP_MEMALLOC; 315 static_branch_dec(&memalloc_socks_key); 316 317 /* 318 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward 319 * progress of swapping. SOCK_MEMALLOC may be cleared while 320 * it has rmem allocations due to the last swapfile being deactivated 321 * but there is a risk that the socket is unusable due to exceeding 322 * the rmem limits. Reclaim the reserves and obey rmem limits again. 323 */ 324 sk_mem_reclaim(sk); 325 } 326 EXPORT_SYMBOL_GPL(sk_clear_memalloc); 327 328 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb) 329 { 330 int ret; 331 unsigned int noreclaim_flag; 332 333 /* these should have been dropped before queueing */ 334 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC)); 335 336 noreclaim_flag = memalloc_noreclaim_save(); 337 ret = INDIRECT_CALL_INET(sk->sk_backlog_rcv, 338 tcp_v6_do_rcv, 339 tcp_v4_do_rcv, 340 sk, skb); 341 memalloc_noreclaim_restore(noreclaim_flag); 342 343 return ret; 344 } 345 EXPORT_SYMBOL(__sk_backlog_rcv); 346 347 void sk_error_report(struct sock *sk) 348 { 349 sk->sk_error_report(sk); 350 351 switch (sk->sk_family) { 352 case AF_INET: 353 fallthrough; 354 case AF_INET6: 355 trace_inet_sk_error_report(sk); 356 break; 357 default: 358 break; 359 } 360 } 361 EXPORT_SYMBOL(sk_error_report); 362 363 int sock_get_timeout(long timeo, void *optval, bool old_timeval) 364 { 365 struct __kernel_sock_timeval tv; 366 367 if (timeo == MAX_SCHEDULE_TIMEOUT) { 368 tv.tv_sec = 0; 369 tv.tv_usec = 0; 370 } else { 371 tv.tv_sec = timeo / HZ; 372 tv.tv_usec = ((timeo % HZ) * USEC_PER_SEC) / HZ; 373 } 374 375 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) { 376 struct old_timeval32 tv32 = { tv.tv_sec, tv.tv_usec }; 377 *(struct old_timeval32 *)optval = tv32; 378 return sizeof(tv32); 379 } 380 381 if (old_timeval) { 382 struct __kernel_old_timeval old_tv; 383 old_tv.tv_sec = tv.tv_sec; 384 old_tv.tv_usec = tv.tv_usec; 385 *(struct __kernel_old_timeval *)optval = old_tv; 386 return sizeof(old_tv); 387 } 388 389 *(struct __kernel_sock_timeval *)optval = tv; 390 return sizeof(tv); 391 } 392 EXPORT_SYMBOL(sock_get_timeout); 393 394 int sock_copy_user_timeval(struct __kernel_sock_timeval *tv, 395 sockptr_t optval, int optlen, bool old_timeval) 396 { 397 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) { 398 struct old_timeval32 tv32; 399 400 if (optlen < sizeof(tv32)) 401 return -EINVAL; 402 403 if (copy_from_sockptr(&tv32, optval, sizeof(tv32))) 404 return -EFAULT; 405 tv->tv_sec = tv32.tv_sec; 406 tv->tv_usec = tv32.tv_usec; 407 } else if (old_timeval) { 408 struct __kernel_old_timeval old_tv; 409 410 if (optlen < sizeof(old_tv)) 411 return -EINVAL; 412 if (copy_from_sockptr(&old_tv, optval, sizeof(old_tv))) 413 return -EFAULT; 414 tv->tv_sec = old_tv.tv_sec; 415 tv->tv_usec = old_tv.tv_usec; 416 } else { 417 if (optlen < sizeof(*tv)) 418 return -EINVAL; 419 if (copy_from_sockptr(tv, optval, sizeof(*tv))) 420 return -EFAULT; 421 } 422 423 return 0; 424 } 425 EXPORT_SYMBOL(sock_copy_user_timeval); 426 427 static int sock_set_timeout(long *timeo_p, sockptr_t optval, int optlen, 428 bool old_timeval) 429 { 430 struct __kernel_sock_timeval tv; 431 int err = sock_copy_user_timeval(&tv, optval, optlen, old_timeval); 432 long val; 433 434 if (err) 435 return err; 436 437 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC) 438 return -EDOM; 439 440 if (tv.tv_sec < 0) { 441 static int warned __read_mostly; 442 443 WRITE_ONCE(*timeo_p, 0); 444 if (warned < 10 && net_ratelimit()) { 445 warned++; 446 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n", 447 __func__, current->comm, task_pid_nr(current)); 448 } 449 return 0; 450 } 451 val = MAX_SCHEDULE_TIMEOUT; 452 if ((tv.tv_sec || tv.tv_usec) && 453 (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1))) 454 val = tv.tv_sec * HZ + DIV_ROUND_UP((unsigned long)tv.tv_usec, 455 USEC_PER_SEC / HZ); 456 WRITE_ONCE(*timeo_p, val); 457 return 0; 458 } 459 460 static bool sock_needs_netstamp(const struct sock *sk) 461 { 462 switch (sk->sk_family) { 463 case AF_UNSPEC: 464 case AF_UNIX: 465 return false; 466 default: 467 return true; 468 } 469 } 470 471 static void sock_disable_timestamp(struct sock *sk, unsigned long flags) 472 { 473 if (sk->sk_flags & flags) { 474 sk->sk_flags &= ~flags; 475 if (sock_needs_netstamp(sk) && 476 !(sk->sk_flags & SK_FLAGS_TIMESTAMP)) 477 net_disable_timestamp(); 478 } 479 } 480 481 482 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 483 { 484 unsigned long flags; 485 struct sk_buff_head *list = &sk->sk_receive_queue; 486 487 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) { 488 atomic_inc(&sk->sk_drops); 489 trace_sock_rcvqueue_full(sk, skb); 490 return -ENOMEM; 491 } 492 493 if (!sk_rmem_schedule(sk, skb, skb->truesize)) { 494 atomic_inc(&sk->sk_drops); 495 return -ENOBUFS; 496 } 497 498 skb->dev = NULL; 499 skb_set_owner_r(skb, sk); 500 501 /* we escape from rcu protected region, make sure we dont leak 502 * a norefcounted dst 503 */ 504 skb_dst_force(skb); 505 506 spin_lock_irqsave(&list->lock, flags); 507 sock_skb_set_dropcount(sk, skb); 508 __skb_queue_tail(list, skb); 509 spin_unlock_irqrestore(&list->lock, flags); 510 511 if (!sock_flag(sk, SOCK_DEAD)) 512 sk->sk_data_ready(sk); 513 return 0; 514 } 515 EXPORT_SYMBOL(__sock_queue_rcv_skb); 516 517 int sock_queue_rcv_skb_reason(struct sock *sk, struct sk_buff *skb, 518 enum skb_drop_reason *reason) 519 { 520 enum skb_drop_reason drop_reason; 521 int err; 522 523 err = sk_filter(sk, skb); 524 if (err) { 525 drop_reason = SKB_DROP_REASON_SOCKET_FILTER; 526 goto out; 527 } 528 err = __sock_queue_rcv_skb(sk, skb); 529 switch (err) { 530 case -ENOMEM: 531 drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF; 532 break; 533 case -ENOBUFS: 534 drop_reason = SKB_DROP_REASON_PROTO_MEM; 535 break; 536 default: 537 drop_reason = SKB_NOT_DROPPED_YET; 538 break; 539 } 540 out: 541 if (reason) 542 *reason = drop_reason; 543 return err; 544 } 545 EXPORT_SYMBOL(sock_queue_rcv_skb_reason); 546 547 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, 548 const int nested, unsigned int trim_cap, bool refcounted) 549 { 550 int rc = NET_RX_SUCCESS; 551 552 if (sk_filter_trim_cap(sk, skb, trim_cap)) 553 goto discard_and_relse; 554 555 skb->dev = NULL; 556 557 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) { 558 atomic_inc(&sk->sk_drops); 559 goto discard_and_relse; 560 } 561 if (nested) 562 bh_lock_sock_nested(sk); 563 else 564 bh_lock_sock(sk); 565 if (!sock_owned_by_user(sk)) { 566 /* 567 * trylock + unlock semantics: 568 */ 569 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_); 570 571 rc = sk_backlog_rcv(sk, skb); 572 573 mutex_release(&sk->sk_lock.dep_map, _RET_IP_); 574 } else if (sk_add_backlog(sk, skb, READ_ONCE(sk->sk_rcvbuf))) { 575 bh_unlock_sock(sk); 576 atomic_inc(&sk->sk_drops); 577 goto discard_and_relse; 578 } 579 580 bh_unlock_sock(sk); 581 out: 582 if (refcounted) 583 sock_put(sk); 584 return rc; 585 discard_and_relse: 586 kfree_skb(skb); 587 goto out; 588 } 589 EXPORT_SYMBOL(__sk_receive_skb); 590 591 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ip6_dst_check(struct dst_entry *, 592 u32)); 593 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ipv4_dst_check(struct dst_entry *, 594 u32)); 595 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie) 596 { 597 struct dst_entry *dst = __sk_dst_get(sk); 598 599 if (dst && dst->obsolete && 600 INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check, 601 dst, cookie) == NULL) { 602 sk_tx_queue_clear(sk); 603 sk->sk_dst_pending_confirm = 0; 604 RCU_INIT_POINTER(sk->sk_dst_cache, NULL); 605 dst_release(dst); 606 return NULL; 607 } 608 609 return dst; 610 } 611 EXPORT_SYMBOL(__sk_dst_check); 612 613 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie) 614 { 615 struct dst_entry *dst = sk_dst_get(sk); 616 617 if (dst && dst->obsolete && 618 INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check, 619 dst, cookie) == NULL) { 620 sk_dst_reset(sk); 621 dst_release(dst); 622 return NULL; 623 } 624 625 return dst; 626 } 627 EXPORT_SYMBOL(sk_dst_check); 628 629 static int sock_bindtoindex_locked(struct sock *sk, int ifindex) 630 { 631 int ret = -ENOPROTOOPT; 632 #ifdef CONFIG_NETDEVICES 633 struct net *net = sock_net(sk); 634 635 /* Sorry... */ 636 ret = -EPERM; 637 if (sk->sk_bound_dev_if && !ns_capable(net->user_ns, CAP_NET_RAW)) 638 goto out; 639 640 ret = -EINVAL; 641 if (ifindex < 0) 642 goto out; 643 644 /* Paired with all READ_ONCE() done locklessly. */ 645 WRITE_ONCE(sk->sk_bound_dev_if, ifindex); 646 647 if (sk->sk_prot->rehash) 648 sk->sk_prot->rehash(sk); 649 sk_dst_reset(sk); 650 651 ret = 0; 652 653 out: 654 #endif 655 656 return ret; 657 } 658 659 int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk) 660 { 661 int ret; 662 663 if (lock_sk) 664 lock_sock(sk); 665 ret = sock_bindtoindex_locked(sk, ifindex); 666 if (lock_sk) 667 release_sock(sk); 668 669 return ret; 670 } 671 EXPORT_SYMBOL(sock_bindtoindex); 672 673 static int sock_setbindtodevice(struct sock *sk, sockptr_t optval, int optlen) 674 { 675 int ret = -ENOPROTOOPT; 676 #ifdef CONFIG_NETDEVICES 677 struct net *net = sock_net(sk); 678 char devname[IFNAMSIZ]; 679 int index; 680 681 ret = -EINVAL; 682 if (optlen < 0) 683 goto out; 684 685 /* Bind this socket to a particular device like "eth0", 686 * as specified in the passed interface name. If the 687 * name is "" or the option length is zero the socket 688 * is not bound. 689 */ 690 if (optlen > IFNAMSIZ - 1) 691 optlen = IFNAMSIZ - 1; 692 memset(devname, 0, sizeof(devname)); 693 694 ret = -EFAULT; 695 if (copy_from_sockptr(devname, optval, optlen)) 696 goto out; 697 698 index = 0; 699 if (devname[0] != '\0') { 700 struct net_device *dev; 701 702 rcu_read_lock(); 703 dev = dev_get_by_name_rcu(net, devname); 704 if (dev) 705 index = dev->ifindex; 706 rcu_read_unlock(); 707 ret = -ENODEV; 708 if (!dev) 709 goto out; 710 } 711 712 sockopt_lock_sock(sk); 713 ret = sock_bindtoindex_locked(sk, index); 714 sockopt_release_sock(sk); 715 out: 716 #endif 717 718 return ret; 719 } 720 721 static int sock_getbindtodevice(struct sock *sk, sockptr_t optval, 722 sockptr_t optlen, int len) 723 { 724 int ret = -ENOPROTOOPT; 725 #ifdef CONFIG_NETDEVICES 726 int bound_dev_if = READ_ONCE(sk->sk_bound_dev_if); 727 struct net *net = sock_net(sk); 728 char devname[IFNAMSIZ]; 729 730 if (bound_dev_if == 0) { 731 len = 0; 732 goto zero; 733 } 734 735 ret = -EINVAL; 736 if (len < IFNAMSIZ) 737 goto out; 738 739 ret = netdev_get_name(net, devname, bound_dev_if); 740 if (ret) 741 goto out; 742 743 len = strlen(devname) + 1; 744 745 ret = -EFAULT; 746 if (copy_to_sockptr(optval, devname, len)) 747 goto out; 748 749 zero: 750 ret = -EFAULT; 751 if (copy_to_sockptr(optlen, &len, sizeof(int))) 752 goto out; 753 754 ret = 0; 755 756 out: 757 #endif 758 759 return ret; 760 } 761 762 bool sk_mc_loop(struct sock *sk) 763 { 764 if (dev_recursion_level()) 765 return false; 766 if (!sk) 767 return true; 768 switch (sk->sk_family) { 769 case AF_INET: 770 return inet_sk(sk)->mc_loop; 771 #if IS_ENABLED(CONFIG_IPV6) 772 case AF_INET6: 773 return inet6_sk(sk)->mc_loop; 774 #endif 775 } 776 WARN_ON_ONCE(1); 777 return true; 778 } 779 EXPORT_SYMBOL(sk_mc_loop); 780 781 void sock_set_reuseaddr(struct sock *sk) 782 { 783 lock_sock(sk); 784 sk->sk_reuse = SK_CAN_REUSE; 785 release_sock(sk); 786 } 787 EXPORT_SYMBOL(sock_set_reuseaddr); 788 789 void sock_set_reuseport(struct sock *sk) 790 { 791 lock_sock(sk); 792 sk->sk_reuseport = true; 793 release_sock(sk); 794 } 795 EXPORT_SYMBOL(sock_set_reuseport); 796 797 void sock_no_linger(struct sock *sk) 798 { 799 lock_sock(sk); 800 sk->sk_lingertime = 0; 801 sock_set_flag(sk, SOCK_LINGER); 802 release_sock(sk); 803 } 804 EXPORT_SYMBOL(sock_no_linger); 805 806 void sock_set_priority(struct sock *sk, u32 priority) 807 { 808 lock_sock(sk); 809 WRITE_ONCE(sk->sk_priority, priority); 810 release_sock(sk); 811 } 812 EXPORT_SYMBOL(sock_set_priority); 813 814 void sock_set_sndtimeo(struct sock *sk, s64 secs) 815 { 816 lock_sock(sk); 817 if (secs && secs < MAX_SCHEDULE_TIMEOUT / HZ - 1) 818 WRITE_ONCE(sk->sk_sndtimeo, secs * HZ); 819 else 820 WRITE_ONCE(sk->sk_sndtimeo, MAX_SCHEDULE_TIMEOUT); 821 release_sock(sk); 822 } 823 EXPORT_SYMBOL(sock_set_sndtimeo); 824 825 static void __sock_set_timestamps(struct sock *sk, bool val, bool new, bool ns) 826 { 827 if (val) { 828 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, new); 829 sock_valbool_flag(sk, SOCK_RCVTSTAMPNS, ns); 830 sock_set_flag(sk, SOCK_RCVTSTAMP); 831 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 832 } else { 833 sock_reset_flag(sk, SOCK_RCVTSTAMP); 834 sock_reset_flag(sk, SOCK_RCVTSTAMPNS); 835 } 836 } 837 838 void sock_enable_timestamps(struct sock *sk) 839 { 840 lock_sock(sk); 841 __sock_set_timestamps(sk, true, false, true); 842 release_sock(sk); 843 } 844 EXPORT_SYMBOL(sock_enable_timestamps); 845 846 void sock_set_timestamp(struct sock *sk, int optname, bool valbool) 847 { 848 switch (optname) { 849 case SO_TIMESTAMP_OLD: 850 __sock_set_timestamps(sk, valbool, false, false); 851 break; 852 case SO_TIMESTAMP_NEW: 853 __sock_set_timestamps(sk, valbool, true, false); 854 break; 855 case SO_TIMESTAMPNS_OLD: 856 __sock_set_timestamps(sk, valbool, false, true); 857 break; 858 case SO_TIMESTAMPNS_NEW: 859 __sock_set_timestamps(sk, valbool, true, true); 860 break; 861 } 862 } 863 864 static int sock_timestamping_bind_phc(struct sock *sk, int phc_index) 865 { 866 struct net *net = sock_net(sk); 867 struct net_device *dev = NULL; 868 bool match = false; 869 int *vclock_index; 870 int i, num; 871 872 if (sk->sk_bound_dev_if) 873 dev = dev_get_by_index(net, sk->sk_bound_dev_if); 874 875 if (!dev) { 876 pr_err("%s: sock not bind to device\n", __func__); 877 return -EOPNOTSUPP; 878 } 879 880 num = ethtool_get_phc_vclocks(dev, &vclock_index); 881 dev_put(dev); 882 883 for (i = 0; i < num; i++) { 884 if (*(vclock_index + i) == phc_index) { 885 match = true; 886 break; 887 } 888 } 889 890 if (num > 0) 891 kfree(vclock_index); 892 893 if (!match) 894 return -EINVAL; 895 896 sk->sk_bind_phc = phc_index; 897 898 return 0; 899 } 900 901 int sock_set_timestamping(struct sock *sk, int optname, 902 struct so_timestamping timestamping) 903 { 904 int val = timestamping.flags; 905 int ret; 906 907 if (val & ~SOF_TIMESTAMPING_MASK) 908 return -EINVAL; 909 910 if (val & SOF_TIMESTAMPING_OPT_ID_TCP && 911 !(val & SOF_TIMESTAMPING_OPT_ID)) 912 return -EINVAL; 913 914 if (val & SOF_TIMESTAMPING_OPT_ID && 915 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) { 916 if (sk_is_tcp(sk)) { 917 if ((1 << sk->sk_state) & 918 (TCPF_CLOSE | TCPF_LISTEN)) 919 return -EINVAL; 920 if (val & SOF_TIMESTAMPING_OPT_ID_TCP) 921 atomic_set(&sk->sk_tskey, tcp_sk(sk)->write_seq); 922 else 923 atomic_set(&sk->sk_tskey, tcp_sk(sk)->snd_una); 924 } else { 925 atomic_set(&sk->sk_tskey, 0); 926 } 927 } 928 929 if (val & SOF_TIMESTAMPING_OPT_STATS && 930 !(val & SOF_TIMESTAMPING_OPT_TSONLY)) 931 return -EINVAL; 932 933 if (val & SOF_TIMESTAMPING_BIND_PHC) { 934 ret = sock_timestamping_bind_phc(sk, timestamping.bind_phc); 935 if (ret) 936 return ret; 937 } 938 939 sk->sk_tsflags = val; 940 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, optname == SO_TIMESTAMPING_NEW); 941 942 if (val & SOF_TIMESTAMPING_RX_SOFTWARE) 943 sock_enable_timestamp(sk, 944 SOCK_TIMESTAMPING_RX_SOFTWARE); 945 else 946 sock_disable_timestamp(sk, 947 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE)); 948 return 0; 949 } 950 951 void sock_set_keepalive(struct sock *sk) 952 { 953 lock_sock(sk); 954 if (sk->sk_prot->keepalive) 955 sk->sk_prot->keepalive(sk, true); 956 sock_valbool_flag(sk, SOCK_KEEPOPEN, true); 957 release_sock(sk); 958 } 959 EXPORT_SYMBOL(sock_set_keepalive); 960 961 static void __sock_set_rcvbuf(struct sock *sk, int val) 962 { 963 /* Ensure val * 2 fits into an int, to prevent max_t() from treating it 964 * as a negative value. 965 */ 966 val = min_t(int, val, INT_MAX / 2); 967 sk->sk_userlocks |= SOCK_RCVBUF_LOCK; 968 969 /* We double it on the way in to account for "struct sk_buff" etc. 970 * overhead. Applications assume that the SO_RCVBUF setting they make 971 * will allow that much actual data to be received on that socket. 972 * 973 * Applications are unaware that "struct sk_buff" and other overheads 974 * allocate from the receive buffer during socket buffer allocation. 975 * 976 * And after considering the possible alternatives, returning the value 977 * we actually used in getsockopt is the most desirable behavior. 978 */ 979 WRITE_ONCE(sk->sk_rcvbuf, max_t(int, val * 2, SOCK_MIN_RCVBUF)); 980 } 981 982 void sock_set_rcvbuf(struct sock *sk, int val) 983 { 984 lock_sock(sk); 985 __sock_set_rcvbuf(sk, val); 986 release_sock(sk); 987 } 988 EXPORT_SYMBOL(sock_set_rcvbuf); 989 990 static void __sock_set_mark(struct sock *sk, u32 val) 991 { 992 if (val != sk->sk_mark) { 993 WRITE_ONCE(sk->sk_mark, val); 994 sk_dst_reset(sk); 995 } 996 } 997 998 void sock_set_mark(struct sock *sk, u32 val) 999 { 1000 lock_sock(sk); 1001 __sock_set_mark(sk, val); 1002 release_sock(sk); 1003 } 1004 EXPORT_SYMBOL(sock_set_mark); 1005 1006 static void sock_release_reserved_memory(struct sock *sk, int bytes) 1007 { 1008 /* Round down bytes to multiple of pages */ 1009 bytes = round_down(bytes, PAGE_SIZE); 1010 1011 WARN_ON(bytes > sk->sk_reserved_mem); 1012 WRITE_ONCE(sk->sk_reserved_mem, sk->sk_reserved_mem - bytes); 1013 sk_mem_reclaim(sk); 1014 } 1015 1016 static int sock_reserve_memory(struct sock *sk, int bytes) 1017 { 1018 long allocated; 1019 bool charged; 1020 int pages; 1021 1022 if (!mem_cgroup_sockets_enabled || !sk->sk_memcg || !sk_has_account(sk)) 1023 return -EOPNOTSUPP; 1024 1025 if (!bytes) 1026 return 0; 1027 1028 pages = sk_mem_pages(bytes); 1029 1030 /* pre-charge to memcg */ 1031 charged = mem_cgroup_charge_skmem(sk->sk_memcg, pages, 1032 GFP_KERNEL | __GFP_RETRY_MAYFAIL); 1033 if (!charged) 1034 return -ENOMEM; 1035 1036 /* pre-charge to forward_alloc */ 1037 sk_memory_allocated_add(sk, pages); 1038 allocated = sk_memory_allocated(sk); 1039 /* If the system goes into memory pressure with this 1040 * precharge, give up and return error. 1041 */ 1042 if (allocated > sk_prot_mem_limits(sk, 1)) { 1043 sk_memory_allocated_sub(sk, pages); 1044 mem_cgroup_uncharge_skmem(sk->sk_memcg, pages); 1045 return -ENOMEM; 1046 } 1047 sk->sk_forward_alloc += pages << PAGE_SHIFT; 1048 1049 WRITE_ONCE(sk->sk_reserved_mem, 1050 sk->sk_reserved_mem + (pages << PAGE_SHIFT)); 1051 1052 return 0; 1053 } 1054 1055 void sockopt_lock_sock(struct sock *sk) 1056 { 1057 /* When current->bpf_ctx is set, the setsockopt is called from 1058 * a bpf prog. bpf has ensured the sk lock has been 1059 * acquired before calling setsockopt(). 1060 */ 1061 if (has_current_bpf_ctx()) 1062 return; 1063 1064 lock_sock(sk); 1065 } 1066 EXPORT_SYMBOL(sockopt_lock_sock); 1067 1068 void sockopt_release_sock(struct sock *sk) 1069 { 1070 if (has_current_bpf_ctx()) 1071 return; 1072 1073 release_sock(sk); 1074 } 1075 EXPORT_SYMBOL(sockopt_release_sock); 1076 1077 bool sockopt_ns_capable(struct user_namespace *ns, int cap) 1078 { 1079 return has_current_bpf_ctx() || ns_capable(ns, cap); 1080 } 1081 EXPORT_SYMBOL(sockopt_ns_capable); 1082 1083 bool sockopt_capable(int cap) 1084 { 1085 return has_current_bpf_ctx() || capable(cap); 1086 } 1087 EXPORT_SYMBOL(sockopt_capable); 1088 1089 /* 1090 * This is meant for all protocols to use and covers goings on 1091 * at the socket level. Everything here is generic. 1092 */ 1093 1094 int sk_setsockopt(struct sock *sk, int level, int optname, 1095 sockptr_t optval, unsigned int optlen) 1096 { 1097 struct so_timestamping timestamping; 1098 struct socket *sock = sk->sk_socket; 1099 struct sock_txtime sk_txtime; 1100 int val; 1101 int valbool; 1102 struct linger ling; 1103 int ret = 0; 1104 1105 /* 1106 * Options without arguments 1107 */ 1108 1109 if (optname == SO_BINDTODEVICE) 1110 return sock_setbindtodevice(sk, optval, optlen); 1111 1112 if (optlen < sizeof(int)) 1113 return -EINVAL; 1114 1115 if (copy_from_sockptr(&val, optval, sizeof(val))) 1116 return -EFAULT; 1117 1118 valbool = val ? 1 : 0; 1119 1120 sockopt_lock_sock(sk); 1121 1122 switch (optname) { 1123 case SO_DEBUG: 1124 if (val && !sockopt_capable(CAP_NET_ADMIN)) 1125 ret = -EACCES; 1126 else 1127 sock_valbool_flag(sk, SOCK_DBG, valbool); 1128 break; 1129 case SO_REUSEADDR: 1130 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE); 1131 break; 1132 case SO_REUSEPORT: 1133 sk->sk_reuseport = valbool; 1134 break; 1135 case SO_TYPE: 1136 case SO_PROTOCOL: 1137 case SO_DOMAIN: 1138 case SO_ERROR: 1139 ret = -ENOPROTOOPT; 1140 break; 1141 case SO_DONTROUTE: 1142 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool); 1143 sk_dst_reset(sk); 1144 break; 1145 case SO_BROADCAST: 1146 sock_valbool_flag(sk, SOCK_BROADCAST, valbool); 1147 break; 1148 case SO_SNDBUF: 1149 /* Don't error on this BSD doesn't and if you think 1150 * about it this is right. Otherwise apps have to 1151 * play 'guess the biggest size' games. RCVBUF/SNDBUF 1152 * are treated in BSD as hints 1153 */ 1154 val = min_t(u32, val, READ_ONCE(sysctl_wmem_max)); 1155 set_sndbuf: 1156 /* Ensure val * 2 fits into an int, to prevent max_t() 1157 * from treating it as a negative value. 1158 */ 1159 val = min_t(int, val, INT_MAX / 2); 1160 sk->sk_userlocks |= SOCK_SNDBUF_LOCK; 1161 WRITE_ONCE(sk->sk_sndbuf, 1162 max_t(int, val * 2, SOCK_MIN_SNDBUF)); 1163 /* Wake up sending tasks if we upped the value. */ 1164 sk->sk_write_space(sk); 1165 break; 1166 1167 case SO_SNDBUFFORCE: 1168 if (!sockopt_capable(CAP_NET_ADMIN)) { 1169 ret = -EPERM; 1170 break; 1171 } 1172 1173 /* No negative values (to prevent underflow, as val will be 1174 * multiplied by 2). 1175 */ 1176 if (val < 0) 1177 val = 0; 1178 goto set_sndbuf; 1179 1180 case SO_RCVBUF: 1181 /* Don't error on this BSD doesn't and if you think 1182 * about it this is right. Otherwise apps have to 1183 * play 'guess the biggest size' games. RCVBUF/SNDBUF 1184 * are treated in BSD as hints 1185 */ 1186 __sock_set_rcvbuf(sk, min_t(u32, val, READ_ONCE(sysctl_rmem_max))); 1187 break; 1188 1189 case SO_RCVBUFFORCE: 1190 if (!sockopt_capable(CAP_NET_ADMIN)) { 1191 ret = -EPERM; 1192 break; 1193 } 1194 1195 /* No negative values (to prevent underflow, as val will be 1196 * multiplied by 2). 1197 */ 1198 __sock_set_rcvbuf(sk, max(val, 0)); 1199 break; 1200 1201 case SO_KEEPALIVE: 1202 if (sk->sk_prot->keepalive) 1203 sk->sk_prot->keepalive(sk, valbool); 1204 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool); 1205 break; 1206 1207 case SO_OOBINLINE: 1208 sock_valbool_flag(sk, SOCK_URGINLINE, valbool); 1209 break; 1210 1211 case SO_NO_CHECK: 1212 sk->sk_no_check_tx = valbool; 1213 break; 1214 1215 case SO_PRIORITY: 1216 if ((val >= 0 && val <= 6) || 1217 sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) || 1218 sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) 1219 WRITE_ONCE(sk->sk_priority, val); 1220 else 1221 ret = -EPERM; 1222 break; 1223 1224 case SO_LINGER: 1225 if (optlen < sizeof(ling)) { 1226 ret = -EINVAL; /* 1003.1g */ 1227 break; 1228 } 1229 if (copy_from_sockptr(&ling, optval, sizeof(ling))) { 1230 ret = -EFAULT; 1231 break; 1232 } 1233 if (!ling.l_onoff) 1234 sock_reset_flag(sk, SOCK_LINGER); 1235 else { 1236 #if (BITS_PER_LONG == 32) 1237 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ) 1238 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT; 1239 else 1240 #endif 1241 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ; 1242 sock_set_flag(sk, SOCK_LINGER); 1243 } 1244 break; 1245 1246 case SO_BSDCOMPAT: 1247 break; 1248 1249 case SO_PASSCRED: 1250 assign_bit(SOCK_PASSCRED, &sock->flags, valbool); 1251 break; 1252 1253 case SO_PASSPIDFD: 1254 assign_bit(SOCK_PASSPIDFD, &sock->flags, valbool); 1255 break; 1256 1257 case SO_TIMESTAMP_OLD: 1258 case SO_TIMESTAMP_NEW: 1259 case SO_TIMESTAMPNS_OLD: 1260 case SO_TIMESTAMPNS_NEW: 1261 sock_set_timestamp(sk, optname, valbool); 1262 break; 1263 1264 case SO_TIMESTAMPING_NEW: 1265 case SO_TIMESTAMPING_OLD: 1266 if (optlen == sizeof(timestamping)) { 1267 if (copy_from_sockptr(×tamping, optval, 1268 sizeof(timestamping))) { 1269 ret = -EFAULT; 1270 break; 1271 } 1272 } else { 1273 memset(×tamping, 0, sizeof(timestamping)); 1274 timestamping.flags = val; 1275 } 1276 ret = sock_set_timestamping(sk, optname, timestamping); 1277 break; 1278 1279 case SO_RCVLOWAT: 1280 if (val < 0) 1281 val = INT_MAX; 1282 if (sock && sock->ops->set_rcvlowat) 1283 ret = sock->ops->set_rcvlowat(sk, val); 1284 else 1285 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1); 1286 break; 1287 1288 case SO_RCVTIMEO_OLD: 1289 case SO_RCVTIMEO_NEW: 1290 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, 1291 optlen, optname == SO_RCVTIMEO_OLD); 1292 break; 1293 1294 case SO_SNDTIMEO_OLD: 1295 case SO_SNDTIMEO_NEW: 1296 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, 1297 optlen, optname == SO_SNDTIMEO_OLD); 1298 break; 1299 1300 case SO_ATTACH_FILTER: { 1301 struct sock_fprog fprog; 1302 1303 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen); 1304 if (!ret) 1305 ret = sk_attach_filter(&fprog, sk); 1306 break; 1307 } 1308 case SO_ATTACH_BPF: 1309 ret = -EINVAL; 1310 if (optlen == sizeof(u32)) { 1311 u32 ufd; 1312 1313 ret = -EFAULT; 1314 if (copy_from_sockptr(&ufd, optval, sizeof(ufd))) 1315 break; 1316 1317 ret = sk_attach_bpf(ufd, sk); 1318 } 1319 break; 1320 1321 case SO_ATTACH_REUSEPORT_CBPF: { 1322 struct sock_fprog fprog; 1323 1324 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen); 1325 if (!ret) 1326 ret = sk_reuseport_attach_filter(&fprog, sk); 1327 break; 1328 } 1329 case SO_ATTACH_REUSEPORT_EBPF: 1330 ret = -EINVAL; 1331 if (optlen == sizeof(u32)) { 1332 u32 ufd; 1333 1334 ret = -EFAULT; 1335 if (copy_from_sockptr(&ufd, optval, sizeof(ufd))) 1336 break; 1337 1338 ret = sk_reuseport_attach_bpf(ufd, sk); 1339 } 1340 break; 1341 1342 case SO_DETACH_REUSEPORT_BPF: 1343 ret = reuseport_detach_prog(sk); 1344 break; 1345 1346 case SO_DETACH_FILTER: 1347 ret = sk_detach_filter(sk); 1348 break; 1349 1350 case SO_LOCK_FILTER: 1351 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool) 1352 ret = -EPERM; 1353 else 1354 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool); 1355 break; 1356 1357 case SO_PASSSEC: 1358 assign_bit(SOCK_PASSSEC, &sock->flags, valbool); 1359 break; 1360 case SO_MARK: 1361 if (!sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) && 1362 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) { 1363 ret = -EPERM; 1364 break; 1365 } 1366 1367 __sock_set_mark(sk, val); 1368 break; 1369 case SO_RCVMARK: 1370 sock_valbool_flag(sk, SOCK_RCVMARK, valbool); 1371 break; 1372 1373 case SO_RXQ_OVFL: 1374 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool); 1375 break; 1376 1377 case SO_WIFI_STATUS: 1378 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool); 1379 break; 1380 1381 case SO_PEEK_OFF: 1382 if (sock->ops->set_peek_off) 1383 ret = sock->ops->set_peek_off(sk, val); 1384 else 1385 ret = -EOPNOTSUPP; 1386 break; 1387 1388 case SO_NOFCS: 1389 sock_valbool_flag(sk, SOCK_NOFCS, valbool); 1390 break; 1391 1392 case SO_SELECT_ERR_QUEUE: 1393 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool); 1394 break; 1395 1396 #ifdef CONFIG_NET_RX_BUSY_POLL 1397 case SO_BUSY_POLL: 1398 if (val < 0) 1399 ret = -EINVAL; 1400 else 1401 WRITE_ONCE(sk->sk_ll_usec, val); 1402 break; 1403 case SO_PREFER_BUSY_POLL: 1404 if (valbool && !sockopt_capable(CAP_NET_ADMIN)) 1405 ret = -EPERM; 1406 else 1407 WRITE_ONCE(sk->sk_prefer_busy_poll, valbool); 1408 break; 1409 case SO_BUSY_POLL_BUDGET: 1410 if (val > READ_ONCE(sk->sk_busy_poll_budget) && !sockopt_capable(CAP_NET_ADMIN)) { 1411 ret = -EPERM; 1412 } else { 1413 if (val < 0 || val > U16_MAX) 1414 ret = -EINVAL; 1415 else 1416 WRITE_ONCE(sk->sk_busy_poll_budget, val); 1417 } 1418 break; 1419 #endif 1420 1421 case SO_MAX_PACING_RATE: 1422 { 1423 unsigned long ulval = (val == ~0U) ? ~0UL : (unsigned int)val; 1424 1425 if (sizeof(ulval) != sizeof(val) && 1426 optlen >= sizeof(ulval) && 1427 copy_from_sockptr(&ulval, optval, sizeof(ulval))) { 1428 ret = -EFAULT; 1429 break; 1430 } 1431 if (ulval != ~0UL) 1432 cmpxchg(&sk->sk_pacing_status, 1433 SK_PACING_NONE, 1434 SK_PACING_NEEDED); 1435 /* Pairs with READ_ONCE() from sk_getsockopt() */ 1436 WRITE_ONCE(sk->sk_max_pacing_rate, ulval); 1437 sk->sk_pacing_rate = min(sk->sk_pacing_rate, ulval); 1438 break; 1439 } 1440 case SO_INCOMING_CPU: 1441 reuseport_update_incoming_cpu(sk, val); 1442 break; 1443 1444 case SO_CNX_ADVICE: 1445 if (val == 1) 1446 dst_negative_advice(sk); 1447 break; 1448 1449 case SO_ZEROCOPY: 1450 if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) { 1451 if (!(sk_is_tcp(sk) || 1452 (sk->sk_type == SOCK_DGRAM && 1453 sk->sk_protocol == IPPROTO_UDP))) 1454 ret = -EOPNOTSUPP; 1455 } else if (sk->sk_family != PF_RDS) { 1456 ret = -EOPNOTSUPP; 1457 } 1458 if (!ret) { 1459 if (val < 0 || val > 1) 1460 ret = -EINVAL; 1461 else 1462 sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool); 1463 } 1464 break; 1465 1466 case SO_TXTIME: 1467 if (optlen != sizeof(struct sock_txtime)) { 1468 ret = -EINVAL; 1469 break; 1470 } else if (copy_from_sockptr(&sk_txtime, optval, 1471 sizeof(struct sock_txtime))) { 1472 ret = -EFAULT; 1473 break; 1474 } else if (sk_txtime.flags & ~SOF_TXTIME_FLAGS_MASK) { 1475 ret = -EINVAL; 1476 break; 1477 } 1478 /* CLOCK_MONOTONIC is only used by sch_fq, and this packet 1479 * scheduler has enough safe guards. 1480 */ 1481 if (sk_txtime.clockid != CLOCK_MONOTONIC && 1482 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) { 1483 ret = -EPERM; 1484 break; 1485 } 1486 sock_valbool_flag(sk, SOCK_TXTIME, true); 1487 sk->sk_clockid = sk_txtime.clockid; 1488 sk->sk_txtime_deadline_mode = 1489 !!(sk_txtime.flags & SOF_TXTIME_DEADLINE_MODE); 1490 sk->sk_txtime_report_errors = 1491 !!(sk_txtime.flags & SOF_TXTIME_REPORT_ERRORS); 1492 break; 1493 1494 case SO_BINDTOIFINDEX: 1495 ret = sock_bindtoindex_locked(sk, val); 1496 break; 1497 1498 case SO_BUF_LOCK: 1499 if (val & ~SOCK_BUF_LOCK_MASK) { 1500 ret = -EINVAL; 1501 break; 1502 } 1503 sk->sk_userlocks = val | (sk->sk_userlocks & 1504 ~SOCK_BUF_LOCK_MASK); 1505 break; 1506 1507 case SO_RESERVE_MEM: 1508 { 1509 int delta; 1510 1511 if (val < 0) { 1512 ret = -EINVAL; 1513 break; 1514 } 1515 1516 delta = val - sk->sk_reserved_mem; 1517 if (delta < 0) 1518 sock_release_reserved_memory(sk, -delta); 1519 else 1520 ret = sock_reserve_memory(sk, delta); 1521 break; 1522 } 1523 1524 case SO_TXREHASH: 1525 if (val < -1 || val > 1) { 1526 ret = -EINVAL; 1527 break; 1528 } 1529 if ((u8)val == SOCK_TXREHASH_DEFAULT) 1530 val = READ_ONCE(sock_net(sk)->core.sysctl_txrehash); 1531 /* Paired with READ_ONCE() in tcp_rtx_synack() 1532 * and sk_getsockopt(). 1533 */ 1534 WRITE_ONCE(sk->sk_txrehash, (u8)val); 1535 break; 1536 1537 default: 1538 ret = -ENOPROTOOPT; 1539 break; 1540 } 1541 sockopt_release_sock(sk); 1542 return ret; 1543 } 1544 1545 int sock_setsockopt(struct socket *sock, int level, int optname, 1546 sockptr_t optval, unsigned int optlen) 1547 { 1548 return sk_setsockopt(sock->sk, level, optname, 1549 optval, optlen); 1550 } 1551 EXPORT_SYMBOL(sock_setsockopt); 1552 1553 static const struct cred *sk_get_peer_cred(struct sock *sk) 1554 { 1555 const struct cred *cred; 1556 1557 spin_lock(&sk->sk_peer_lock); 1558 cred = get_cred(sk->sk_peer_cred); 1559 spin_unlock(&sk->sk_peer_lock); 1560 1561 return cred; 1562 } 1563 1564 static void cred_to_ucred(struct pid *pid, const struct cred *cred, 1565 struct ucred *ucred) 1566 { 1567 ucred->pid = pid_vnr(pid); 1568 ucred->uid = ucred->gid = -1; 1569 if (cred) { 1570 struct user_namespace *current_ns = current_user_ns(); 1571 1572 ucred->uid = from_kuid_munged(current_ns, cred->euid); 1573 ucred->gid = from_kgid_munged(current_ns, cred->egid); 1574 } 1575 } 1576 1577 static int groups_to_user(sockptr_t dst, const struct group_info *src) 1578 { 1579 struct user_namespace *user_ns = current_user_ns(); 1580 int i; 1581 1582 for (i = 0; i < src->ngroups; i++) { 1583 gid_t gid = from_kgid_munged(user_ns, src->gid[i]); 1584 1585 if (copy_to_sockptr_offset(dst, i * sizeof(gid), &gid, sizeof(gid))) 1586 return -EFAULT; 1587 } 1588 1589 return 0; 1590 } 1591 1592 int sk_getsockopt(struct sock *sk, int level, int optname, 1593 sockptr_t optval, sockptr_t optlen) 1594 { 1595 struct socket *sock = sk->sk_socket; 1596 1597 union { 1598 int val; 1599 u64 val64; 1600 unsigned long ulval; 1601 struct linger ling; 1602 struct old_timeval32 tm32; 1603 struct __kernel_old_timeval tm; 1604 struct __kernel_sock_timeval stm; 1605 struct sock_txtime txtime; 1606 struct so_timestamping timestamping; 1607 } v; 1608 1609 int lv = sizeof(int); 1610 int len; 1611 1612 if (copy_from_sockptr(&len, optlen, sizeof(int))) 1613 return -EFAULT; 1614 if (len < 0) 1615 return -EINVAL; 1616 1617 memset(&v, 0, sizeof(v)); 1618 1619 switch (optname) { 1620 case SO_DEBUG: 1621 v.val = sock_flag(sk, SOCK_DBG); 1622 break; 1623 1624 case SO_DONTROUTE: 1625 v.val = sock_flag(sk, SOCK_LOCALROUTE); 1626 break; 1627 1628 case SO_BROADCAST: 1629 v.val = sock_flag(sk, SOCK_BROADCAST); 1630 break; 1631 1632 case SO_SNDBUF: 1633 v.val = READ_ONCE(sk->sk_sndbuf); 1634 break; 1635 1636 case SO_RCVBUF: 1637 v.val = READ_ONCE(sk->sk_rcvbuf); 1638 break; 1639 1640 case SO_REUSEADDR: 1641 v.val = sk->sk_reuse; 1642 break; 1643 1644 case SO_REUSEPORT: 1645 v.val = sk->sk_reuseport; 1646 break; 1647 1648 case SO_KEEPALIVE: 1649 v.val = sock_flag(sk, SOCK_KEEPOPEN); 1650 break; 1651 1652 case SO_TYPE: 1653 v.val = sk->sk_type; 1654 break; 1655 1656 case SO_PROTOCOL: 1657 v.val = sk->sk_protocol; 1658 break; 1659 1660 case SO_DOMAIN: 1661 v.val = sk->sk_family; 1662 break; 1663 1664 case SO_ERROR: 1665 v.val = -sock_error(sk); 1666 if (v.val == 0) 1667 v.val = xchg(&sk->sk_err_soft, 0); 1668 break; 1669 1670 case SO_OOBINLINE: 1671 v.val = sock_flag(sk, SOCK_URGINLINE); 1672 break; 1673 1674 case SO_NO_CHECK: 1675 v.val = sk->sk_no_check_tx; 1676 break; 1677 1678 case SO_PRIORITY: 1679 v.val = READ_ONCE(sk->sk_priority); 1680 break; 1681 1682 case SO_LINGER: 1683 lv = sizeof(v.ling); 1684 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER); 1685 v.ling.l_linger = sk->sk_lingertime / HZ; 1686 break; 1687 1688 case SO_BSDCOMPAT: 1689 break; 1690 1691 case SO_TIMESTAMP_OLD: 1692 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && 1693 !sock_flag(sk, SOCK_TSTAMP_NEW) && 1694 !sock_flag(sk, SOCK_RCVTSTAMPNS); 1695 break; 1696 1697 case SO_TIMESTAMPNS_OLD: 1698 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && !sock_flag(sk, SOCK_TSTAMP_NEW); 1699 break; 1700 1701 case SO_TIMESTAMP_NEW: 1702 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && sock_flag(sk, SOCK_TSTAMP_NEW); 1703 break; 1704 1705 case SO_TIMESTAMPNS_NEW: 1706 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && sock_flag(sk, SOCK_TSTAMP_NEW); 1707 break; 1708 1709 case SO_TIMESTAMPING_OLD: 1710 lv = sizeof(v.timestamping); 1711 v.timestamping.flags = sk->sk_tsflags; 1712 v.timestamping.bind_phc = sk->sk_bind_phc; 1713 break; 1714 1715 case SO_RCVTIMEO_OLD: 1716 case SO_RCVTIMEO_NEW: 1717 lv = sock_get_timeout(READ_ONCE(sk->sk_rcvtimeo), &v, 1718 SO_RCVTIMEO_OLD == optname); 1719 break; 1720 1721 case SO_SNDTIMEO_OLD: 1722 case SO_SNDTIMEO_NEW: 1723 lv = sock_get_timeout(READ_ONCE(sk->sk_sndtimeo), &v, 1724 SO_SNDTIMEO_OLD == optname); 1725 break; 1726 1727 case SO_RCVLOWAT: 1728 v.val = READ_ONCE(sk->sk_rcvlowat); 1729 break; 1730 1731 case SO_SNDLOWAT: 1732 v.val = 1; 1733 break; 1734 1735 case SO_PASSCRED: 1736 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags); 1737 break; 1738 1739 case SO_PASSPIDFD: 1740 v.val = !!test_bit(SOCK_PASSPIDFD, &sock->flags); 1741 break; 1742 1743 case SO_PEERCRED: 1744 { 1745 struct ucred peercred; 1746 if (len > sizeof(peercred)) 1747 len = sizeof(peercred); 1748 1749 spin_lock(&sk->sk_peer_lock); 1750 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred); 1751 spin_unlock(&sk->sk_peer_lock); 1752 1753 if (copy_to_sockptr(optval, &peercred, len)) 1754 return -EFAULT; 1755 goto lenout; 1756 } 1757 1758 case SO_PEERPIDFD: 1759 { 1760 struct pid *peer_pid; 1761 struct file *pidfd_file = NULL; 1762 int pidfd; 1763 1764 if (len > sizeof(pidfd)) 1765 len = sizeof(pidfd); 1766 1767 spin_lock(&sk->sk_peer_lock); 1768 peer_pid = get_pid(sk->sk_peer_pid); 1769 spin_unlock(&sk->sk_peer_lock); 1770 1771 if (!peer_pid) 1772 return -ESRCH; 1773 1774 pidfd = pidfd_prepare(peer_pid, 0, &pidfd_file); 1775 put_pid(peer_pid); 1776 if (pidfd < 0) 1777 return pidfd; 1778 1779 if (copy_to_sockptr(optval, &pidfd, len) || 1780 copy_to_sockptr(optlen, &len, sizeof(int))) { 1781 put_unused_fd(pidfd); 1782 fput(pidfd_file); 1783 1784 return -EFAULT; 1785 } 1786 1787 fd_install(pidfd, pidfd_file); 1788 return 0; 1789 } 1790 1791 case SO_PEERGROUPS: 1792 { 1793 const struct cred *cred; 1794 int ret, n; 1795 1796 cred = sk_get_peer_cred(sk); 1797 if (!cred) 1798 return -ENODATA; 1799 1800 n = cred->group_info->ngroups; 1801 if (len < n * sizeof(gid_t)) { 1802 len = n * sizeof(gid_t); 1803 put_cred(cred); 1804 return copy_to_sockptr(optlen, &len, sizeof(int)) ? -EFAULT : -ERANGE; 1805 } 1806 len = n * sizeof(gid_t); 1807 1808 ret = groups_to_user(optval, cred->group_info); 1809 put_cred(cred); 1810 if (ret) 1811 return ret; 1812 goto lenout; 1813 } 1814 1815 case SO_PEERNAME: 1816 { 1817 struct sockaddr_storage address; 1818 1819 lv = sock->ops->getname(sock, (struct sockaddr *)&address, 2); 1820 if (lv < 0) 1821 return -ENOTCONN; 1822 if (lv < len) 1823 return -EINVAL; 1824 if (copy_to_sockptr(optval, &address, len)) 1825 return -EFAULT; 1826 goto lenout; 1827 } 1828 1829 /* Dubious BSD thing... Probably nobody even uses it, but 1830 * the UNIX standard wants it for whatever reason... -DaveM 1831 */ 1832 case SO_ACCEPTCONN: 1833 v.val = sk->sk_state == TCP_LISTEN; 1834 break; 1835 1836 case SO_PASSSEC: 1837 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags); 1838 break; 1839 1840 case SO_PEERSEC: 1841 return security_socket_getpeersec_stream(sock, 1842 optval, optlen, len); 1843 1844 case SO_MARK: 1845 v.val = READ_ONCE(sk->sk_mark); 1846 break; 1847 1848 case SO_RCVMARK: 1849 v.val = sock_flag(sk, SOCK_RCVMARK); 1850 break; 1851 1852 case SO_RXQ_OVFL: 1853 v.val = sock_flag(sk, SOCK_RXQ_OVFL); 1854 break; 1855 1856 case SO_WIFI_STATUS: 1857 v.val = sock_flag(sk, SOCK_WIFI_STATUS); 1858 break; 1859 1860 case SO_PEEK_OFF: 1861 if (!sock->ops->set_peek_off) 1862 return -EOPNOTSUPP; 1863 1864 v.val = READ_ONCE(sk->sk_peek_off); 1865 break; 1866 case SO_NOFCS: 1867 v.val = sock_flag(sk, SOCK_NOFCS); 1868 break; 1869 1870 case SO_BINDTODEVICE: 1871 return sock_getbindtodevice(sk, optval, optlen, len); 1872 1873 case SO_GET_FILTER: 1874 len = sk_get_filter(sk, optval, len); 1875 if (len < 0) 1876 return len; 1877 1878 goto lenout; 1879 1880 case SO_LOCK_FILTER: 1881 v.val = sock_flag(sk, SOCK_FILTER_LOCKED); 1882 break; 1883 1884 case SO_BPF_EXTENSIONS: 1885 v.val = bpf_tell_extensions(); 1886 break; 1887 1888 case SO_SELECT_ERR_QUEUE: 1889 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE); 1890 break; 1891 1892 #ifdef CONFIG_NET_RX_BUSY_POLL 1893 case SO_BUSY_POLL: 1894 v.val = READ_ONCE(sk->sk_ll_usec); 1895 break; 1896 case SO_PREFER_BUSY_POLL: 1897 v.val = READ_ONCE(sk->sk_prefer_busy_poll); 1898 break; 1899 #endif 1900 1901 case SO_MAX_PACING_RATE: 1902 /* The READ_ONCE() pair with the WRITE_ONCE() in sk_setsockopt() */ 1903 if (sizeof(v.ulval) != sizeof(v.val) && len >= sizeof(v.ulval)) { 1904 lv = sizeof(v.ulval); 1905 v.ulval = READ_ONCE(sk->sk_max_pacing_rate); 1906 } else { 1907 /* 32bit version */ 1908 v.val = min_t(unsigned long, ~0U, 1909 READ_ONCE(sk->sk_max_pacing_rate)); 1910 } 1911 break; 1912 1913 case SO_INCOMING_CPU: 1914 v.val = READ_ONCE(sk->sk_incoming_cpu); 1915 break; 1916 1917 case SO_MEMINFO: 1918 { 1919 u32 meminfo[SK_MEMINFO_VARS]; 1920 1921 sk_get_meminfo(sk, meminfo); 1922 1923 len = min_t(unsigned int, len, sizeof(meminfo)); 1924 if (copy_to_sockptr(optval, &meminfo, len)) 1925 return -EFAULT; 1926 1927 goto lenout; 1928 } 1929 1930 #ifdef CONFIG_NET_RX_BUSY_POLL 1931 case SO_INCOMING_NAPI_ID: 1932 v.val = READ_ONCE(sk->sk_napi_id); 1933 1934 /* aggregate non-NAPI IDs down to 0 */ 1935 if (v.val < MIN_NAPI_ID) 1936 v.val = 0; 1937 1938 break; 1939 #endif 1940 1941 case SO_COOKIE: 1942 lv = sizeof(u64); 1943 if (len < lv) 1944 return -EINVAL; 1945 v.val64 = sock_gen_cookie(sk); 1946 break; 1947 1948 case SO_ZEROCOPY: 1949 v.val = sock_flag(sk, SOCK_ZEROCOPY); 1950 break; 1951 1952 case SO_TXTIME: 1953 lv = sizeof(v.txtime); 1954 v.txtime.clockid = sk->sk_clockid; 1955 v.txtime.flags |= sk->sk_txtime_deadline_mode ? 1956 SOF_TXTIME_DEADLINE_MODE : 0; 1957 v.txtime.flags |= sk->sk_txtime_report_errors ? 1958 SOF_TXTIME_REPORT_ERRORS : 0; 1959 break; 1960 1961 case SO_BINDTOIFINDEX: 1962 v.val = READ_ONCE(sk->sk_bound_dev_if); 1963 break; 1964 1965 case SO_NETNS_COOKIE: 1966 lv = sizeof(u64); 1967 if (len != lv) 1968 return -EINVAL; 1969 v.val64 = sock_net(sk)->net_cookie; 1970 break; 1971 1972 case SO_BUF_LOCK: 1973 v.val = sk->sk_userlocks & SOCK_BUF_LOCK_MASK; 1974 break; 1975 1976 case SO_RESERVE_MEM: 1977 v.val = READ_ONCE(sk->sk_reserved_mem); 1978 break; 1979 1980 case SO_TXREHASH: 1981 /* Paired with WRITE_ONCE() in sk_setsockopt() */ 1982 v.val = READ_ONCE(sk->sk_txrehash); 1983 break; 1984 1985 default: 1986 /* We implement the SO_SNDLOWAT etc to not be settable 1987 * (1003.1g 7). 1988 */ 1989 return -ENOPROTOOPT; 1990 } 1991 1992 if (len > lv) 1993 len = lv; 1994 if (copy_to_sockptr(optval, &v, len)) 1995 return -EFAULT; 1996 lenout: 1997 if (copy_to_sockptr(optlen, &len, sizeof(int))) 1998 return -EFAULT; 1999 return 0; 2000 } 2001 2002 int sock_getsockopt(struct socket *sock, int level, int optname, 2003 char __user *optval, int __user *optlen) 2004 { 2005 return sk_getsockopt(sock->sk, level, optname, 2006 USER_SOCKPTR(optval), 2007 USER_SOCKPTR(optlen)); 2008 } 2009 2010 /* 2011 * Initialize an sk_lock. 2012 * 2013 * (We also register the sk_lock with the lock validator.) 2014 */ 2015 static inline void sock_lock_init(struct sock *sk) 2016 { 2017 if (sk->sk_kern_sock) 2018 sock_lock_init_class_and_name( 2019 sk, 2020 af_family_kern_slock_key_strings[sk->sk_family], 2021 af_family_kern_slock_keys + sk->sk_family, 2022 af_family_kern_key_strings[sk->sk_family], 2023 af_family_kern_keys + sk->sk_family); 2024 else 2025 sock_lock_init_class_and_name( 2026 sk, 2027 af_family_slock_key_strings[sk->sk_family], 2028 af_family_slock_keys + sk->sk_family, 2029 af_family_key_strings[sk->sk_family], 2030 af_family_keys + sk->sk_family); 2031 } 2032 2033 /* 2034 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet, 2035 * even temporarly, because of RCU lookups. sk_node should also be left as is. 2036 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end 2037 */ 2038 static void sock_copy(struct sock *nsk, const struct sock *osk) 2039 { 2040 const struct proto *prot = READ_ONCE(osk->sk_prot); 2041 #ifdef CONFIG_SECURITY_NETWORK 2042 void *sptr = nsk->sk_security; 2043 #endif 2044 2045 /* If we move sk_tx_queue_mapping out of the private section, 2046 * we must check if sk_tx_queue_clear() is called after 2047 * sock_copy() in sk_clone_lock(). 2048 */ 2049 BUILD_BUG_ON(offsetof(struct sock, sk_tx_queue_mapping) < 2050 offsetof(struct sock, sk_dontcopy_begin) || 2051 offsetof(struct sock, sk_tx_queue_mapping) >= 2052 offsetof(struct sock, sk_dontcopy_end)); 2053 2054 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin)); 2055 2056 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end, 2057 prot->obj_size - offsetof(struct sock, sk_dontcopy_end)); 2058 2059 #ifdef CONFIG_SECURITY_NETWORK 2060 nsk->sk_security = sptr; 2061 security_sk_clone(osk, nsk); 2062 #endif 2063 } 2064 2065 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority, 2066 int family) 2067 { 2068 struct sock *sk; 2069 struct kmem_cache *slab; 2070 2071 slab = prot->slab; 2072 if (slab != NULL) { 2073 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO); 2074 if (!sk) 2075 return sk; 2076 if (want_init_on_alloc(priority)) 2077 sk_prot_clear_nulls(sk, prot->obj_size); 2078 } else 2079 sk = kmalloc(prot->obj_size, priority); 2080 2081 if (sk != NULL) { 2082 if (security_sk_alloc(sk, family, priority)) 2083 goto out_free; 2084 2085 if (!try_module_get(prot->owner)) 2086 goto out_free_sec; 2087 } 2088 2089 return sk; 2090 2091 out_free_sec: 2092 security_sk_free(sk); 2093 out_free: 2094 if (slab != NULL) 2095 kmem_cache_free(slab, sk); 2096 else 2097 kfree(sk); 2098 return NULL; 2099 } 2100 2101 static void sk_prot_free(struct proto *prot, struct sock *sk) 2102 { 2103 struct kmem_cache *slab; 2104 struct module *owner; 2105 2106 owner = prot->owner; 2107 slab = prot->slab; 2108 2109 cgroup_sk_free(&sk->sk_cgrp_data); 2110 mem_cgroup_sk_free(sk); 2111 security_sk_free(sk); 2112 if (slab != NULL) 2113 kmem_cache_free(slab, sk); 2114 else 2115 kfree(sk); 2116 module_put(owner); 2117 } 2118 2119 /** 2120 * sk_alloc - All socket objects are allocated here 2121 * @net: the applicable net namespace 2122 * @family: protocol family 2123 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 2124 * @prot: struct proto associated with this new sock instance 2125 * @kern: is this to be a kernel socket? 2126 */ 2127 struct sock *sk_alloc(struct net *net, int family, gfp_t priority, 2128 struct proto *prot, int kern) 2129 { 2130 struct sock *sk; 2131 2132 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family); 2133 if (sk) { 2134 sk->sk_family = family; 2135 /* 2136 * See comment in struct sock definition to understand 2137 * why we need sk_prot_creator -acme 2138 */ 2139 sk->sk_prot = sk->sk_prot_creator = prot; 2140 sk->sk_kern_sock = kern; 2141 sock_lock_init(sk); 2142 sk->sk_net_refcnt = kern ? 0 : 1; 2143 if (likely(sk->sk_net_refcnt)) { 2144 get_net_track(net, &sk->ns_tracker, priority); 2145 sock_inuse_add(net, 1); 2146 } else { 2147 __netns_tracker_alloc(net, &sk->ns_tracker, 2148 false, priority); 2149 } 2150 2151 sock_net_set(sk, net); 2152 refcount_set(&sk->sk_wmem_alloc, 1); 2153 2154 mem_cgroup_sk_alloc(sk); 2155 cgroup_sk_alloc(&sk->sk_cgrp_data); 2156 sock_update_classid(&sk->sk_cgrp_data); 2157 sock_update_netprioidx(&sk->sk_cgrp_data); 2158 sk_tx_queue_clear(sk); 2159 } 2160 2161 return sk; 2162 } 2163 EXPORT_SYMBOL(sk_alloc); 2164 2165 /* Sockets having SOCK_RCU_FREE will call this function after one RCU 2166 * grace period. This is the case for UDP sockets and TCP listeners. 2167 */ 2168 static void __sk_destruct(struct rcu_head *head) 2169 { 2170 struct sock *sk = container_of(head, struct sock, sk_rcu); 2171 struct sk_filter *filter; 2172 2173 if (sk->sk_destruct) 2174 sk->sk_destruct(sk); 2175 2176 filter = rcu_dereference_check(sk->sk_filter, 2177 refcount_read(&sk->sk_wmem_alloc) == 0); 2178 if (filter) { 2179 sk_filter_uncharge(sk, filter); 2180 RCU_INIT_POINTER(sk->sk_filter, NULL); 2181 } 2182 2183 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP); 2184 2185 #ifdef CONFIG_BPF_SYSCALL 2186 bpf_sk_storage_free(sk); 2187 #endif 2188 2189 if (atomic_read(&sk->sk_omem_alloc)) 2190 pr_debug("%s: optmem leakage (%d bytes) detected\n", 2191 __func__, atomic_read(&sk->sk_omem_alloc)); 2192 2193 if (sk->sk_frag.page) { 2194 put_page(sk->sk_frag.page); 2195 sk->sk_frag.page = NULL; 2196 } 2197 2198 /* We do not need to acquire sk->sk_peer_lock, we are the last user. */ 2199 put_cred(sk->sk_peer_cred); 2200 put_pid(sk->sk_peer_pid); 2201 2202 if (likely(sk->sk_net_refcnt)) 2203 put_net_track(sock_net(sk), &sk->ns_tracker); 2204 else 2205 __netns_tracker_free(sock_net(sk), &sk->ns_tracker, false); 2206 2207 sk_prot_free(sk->sk_prot_creator, sk); 2208 } 2209 2210 void sk_destruct(struct sock *sk) 2211 { 2212 bool use_call_rcu = sock_flag(sk, SOCK_RCU_FREE); 2213 2214 if (rcu_access_pointer(sk->sk_reuseport_cb)) { 2215 reuseport_detach_sock(sk); 2216 use_call_rcu = true; 2217 } 2218 2219 if (use_call_rcu) 2220 call_rcu(&sk->sk_rcu, __sk_destruct); 2221 else 2222 __sk_destruct(&sk->sk_rcu); 2223 } 2224 2225 static void __sk_free(struct sock *sk) 2226 { 2227 if (likely(sk->sk_net_refcnt)) 2228 sock_inuse_add(sock_net(sk), -1); 2229 2230 if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk))) 2231 sock_diag_broadcast_destroy(sk); 2232 else 2233 sk_destruct(sk); 2234 } 2235 2236 void sk_free(struct sock *sk) 2237 { 2238 /* 2239 * We subtract one from sk_wmem_alloc and can know if 2240 * some packets are still in some tx queue. 2241 * If not null, sock_wfree() will call __sk_free(sk) later 2242 */ 2243 if (refcount_dec_and_test(&sk->sk_wmem_alloc)) 2244 __sk_free(sk); 2245 } 2246 EXPORT_SYMBOL(sk_free); 2247 2248 static void sk_init_common(struct sock *sk) 2249 { 2250 skb_queue_head_init(&sk->sk_receive_queue); 2251 skb_queue_head_init(&sk->sk_write_queue); 2252 skb_queue_head_init(&sk->sk_error_queue); 2253 2254 rwlock_init(&sk->sk_callback_lock); 2255 lockdep_set_class_and_name(&sk->sk_receive_queue.lock, 2256 af_rlock_keys + sk->sk_family, 2257 af_family_rlock_key_strings[sk->sk_family]); 2258 lockdep_set_class_and_name(&sk->sk_write_queue.lock, 2259 af_wlock_keys + sk->sk_family, 2260 af_family_wlock_key_strings[sk->sk_family]); 2261 lockdep_set_class_and_name(&sk->sk_error_queue.lock, 2262 af_elock_keys + sk->sk_family, 2263 af_family_elock_key_strings[sk->sk_family]); 2264 lockdep_set_class_and_name(&sk->sk_callback_lock, 2265 af_callback_keys + sk->sk_family, 2266 af_family_clock_key_strings[sk->sk_family]); 2267 } 2268 2269 /** 2270 * sk_clone_lock - clone a socket, and lock its clone 2271 * @sk: the socket to clone 2272 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 2273 * 2274 * Caller must unlock socket even in error path (bh_unlock_sock(newsk)) 2275 */ 2276 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority) 2277 { 2278 struct proto *prot = READ_ONCE(sk->sk_prot); 2279 struct sk_filter *filter; 2280 bool is_charged = true; 2281 struct sock *newsk; 2282 2283 newsk = sk_prot_alloc(prot, priority, sk->sk_family); 2284 if (!newsk) 2285 goto out; 2286 2287 sock_copy(newsk, sk); 2288 2289 newsk->sk_prot_creator = prot; 2290 2291 /* SANITY */ 2292 if (likely(newsk->sk_net_refcnt)) { 2293 get_net_track(sock_net(newsk), &newsk->ns_tracker, priority); 2294 sock_inuse_add(sock_net(newsk), 1); 2295 } else { 2296 /* Kernel sockets are not elevating the struct net refcount. 2297 * Instead, use a tracker to more easily detect if a layer 2298 * is not properly dismantling its kernel sockets at netns 2299 * destroy time. 2300 */ 2301 __netns_tracker_alloc(sock_net(newsk), &newsk->ns_tracker, 2302 false, priority); 2303 } 2304 sk_node_init(&newsk->sk_node); 2305 sock_lock_init(newsk); 2306 bh_lock_sock(newsk); 2307 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL; 2308 newsk->sk_backlog.len = 0; 2309 2310 atomic_set(&newsk->sk_rmem_alloc, 0); 2311 2312 /* sk_wmem_alloc set to one (see sk_free() and sock_wfree()) */ 2313 refcount_set(&newsk->sk_wmem_alloc, 1); 2314 2315 atomic_set(&newsk->sk_omem_alloc, 0); 2316 sk_init_common(newsk); 2317 2318 newsk->sk_dst_cache = NULL; 2319 newsk->sk_dst_pending_confirm = 0; 2320 newsk->sk_wmem_queued = 0; 2321 newsk->sk_forward_alloc = 0; 2322 newsk->sk_reserved_mem = 0; 2323 atomic_set(&newsk->sk_drops, 0); 2324 newsk->sk_send_head = NULL; 2325 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK; 2326 atomic_set(&newsk->sk_zckey, 0); 2327 2328 sock_reset_flag(newsk, SOCK_DONE); 2329 2330 /* sk->sk_memcg will be populated at accept() time */ 2331 newsk->sk_memcg = NULL; 2332 2333 cgroup_sk_clone(&newsk->sk_cgrp_data); 2334 2335 rcu_read_lock(); 2336 filter = rcu_dereference(sk->sk_filter); 2337 if (filter != NULL) 2338 /* though it's an empty new sock, the charging may fail 2339 * if sysctl_optmem_max was changed between creation of 2340 * original socket and cloning 2341 */ 2342 is_charged = sk_filter_charge(newsk, filter); 2343 RCU_INIT_POINTER(newsk->sk_filter, filter); 2344 rcu_read_unlock(); 2345 2346 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) { 2347 /* We need to make sure that we don't uncharge the new 2348 * socket if we couldn't charge it in the first place 2349 * as otherwise we uncharge the parent's filter. 2350 */ 2351 if (!is_charged) 2352 RCU_INIT_POINTER(newsk->sk_filter, NULL); 2353 sk_free_unlock_clone(newsk); 2354 newsk = NULL; 2355 goto out; 2356 } 2357 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL); 2358 2359 if (bpf_sk_storage_clone(sk, newsk)) { 2360 sk_free_unlock_clone(newsk); 2361 newsk = NULL; 2362 goto out; 2363 } 2364 2365 /* Clear sk_user_data if parent had the pointer tagged 2366 * as not suitable for copying when cloning. 2367 */ 2368 if (sk_user_data_is_nocopy(newsk)) 2369 newsk->sk_user_data = NULL; 2370 2371 newsk->sk_err = 0; 2372 newsk->sk_err_soft = 0; 2373 newsk->sk_priority = 0; 2374 newsk->sk_incoming_cpu = raw_smp_processor_id(); 2375 2376 /* Before updating sk_refcnt, we must commit prior changes to memory 2377 * (Documentation/RCU/rculist_nulls.rst for details) 2378 */ 2379 smp_wmb(); 2380 refcount_set(&newsk->sk_refcnt, 2); 2381 2382 sk_set_socket(newsk, NULL); 2383 sk_tx_queue_clear(newsk); 2384 RCU_INIT_POINTER(newsk->sk_wq, NULL); 2385 2386 if (newsk->sk_prot->sockets_allocated) 2387 sk_sockets_allocated_inc(newsk); 2388 2389 if (sock_needs_netstamp(sk) && newsk->sk_flags & SK_FLAGS_TIMESTAMP) 2390 net_enable_timestamp(); 2391 out: 2392 return newsk; 2393 } 2394 EXPORT_SYMBOL_GPL(sk_clone_lock); 2395 2396 void sk_free_unlock_clone(struct sock *sk) 2397 { 2398 /* It is still raw copy of parent, so invalidate 2399 * destructor and make plain sk_free() */ 2400 sk->sk_destruct = NULL; 2401 bh_unlock_sock(sk); 2402 sk_free(sk); 2403 } 2404 EXPORT_SYMBOL_GPL(sk_free_unlock_clone); 2405 2406 static u32 sk_dst_gso_max_size(struct sock *sk, struct dst_entry *dst) 2407 { 2408 bool is_ipv6 = false; 2409 u32 max_size; 2410 2411 #if IS_ENABLED(CONFIG_IPV6) 2412 is_ipv6 = (sk->sk_family == AF_INET6 && 2413 !ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr)); 2414 #endif 2415 /* pairs with the WRITE_ONCE() in netif_set_gso(_ipv4)_max_size() */ 2416 max_size = is_ipv6 ? READ_ONCE(dst->dev->gso_max_size) : 2417 READ_ONCE(dst->dev->gso_ipv4_max_size); 2418 if (max_size > GSO_LEGACY_MAX_SIZE && !sk_is_tcp(sk)) 2419 max_size = GSO_LEGACY_MAX_SIZE; 2420 2421 return max_size - (MAX_TCP_HEADER + 1); 2422 } 2423 2424 void sk_setup_caps(struct sock *sk, struct dst_entry *dst) 2425 { 2426 u32 max_segs = 1; 2427 2428 sk->sk_route_caps = dst->dev->features; 2429 if (sk_is_tcp(sk)) 2430 sk->sk_route_caps |= NETIF_F_GSO; 2431 if (sk->sk_route_caps & NETIF_F_GSO) 2432 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE; 2433 if (unlikely(sk->sk_gso_disabled)) 2434 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 2435 if (sk_can_gso(sk)) { 2436 if (dst->header_len && !xfrm_dst_offload_ok(dst)) { 2437 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 2438 } else { 2439 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM; 2440 sk->sk_gso_max_size = sk_dst_gso_max_size(sk, dst); 2441 /* pairs with the WRITE_ONCE() in netif_set_gso_max_segs() */ 2442 max_segs = max_t(u32, READ_ONCE(dst->dev->gso_max_segs), 1); 2443 } 2444 } 2445 sk->sk_gso_max_segs = max_segs; 2446 sk_dst_set(sk, dst); 2447 } 2448 EXPORT_SYMBOL_GPL(sk_setup_caps); 2449 2450 /* 2451 * Simple resource managers for sockets. 2452 */ 2453 2454 2455 /* 2456 * Write buffer destructor automatically called from kfree_skb. 2457 */ 2458 void sock_wfree(struct sk_buff *skb) 2459 { 2460 struct sock *sk = skb->sk; 2461 unsigned int len = skb->truesize; 2462 bool free; 2463 2464 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) { 2465 if (sock_flag(sk, SOCK_RCU_FREE) && 2466 sk->sk_write_space == sock_def_write_space) { 2467 rcu_read_lock(); 2468 free = refcount_sub_and_test(len, &sk->sk_wmem_alloc); 2469 sock_def_write_space_wfree(sk); 2470 rcu_read_unlock(); 2471 if (unlikely(free)) 2472 __sk_free(sk); 2473 return; 2474 } 2475 2476 /* 2477 * Keep a reference on sk_wmem_alloc, this will be released 2478 * after sk_write_space() call 2479 */ 2480 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc)); 2481 sk->sk_write_space(sk); 2482 len = 1; 2483 } 2484 /* 2485 * if sk_wmem_alloc reaches 0, we must finish what sk_free() 2486 * could not do because of in-flight packets 2487 */ 2488 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc)) 2489 __sk_free(sk); 2490 } 2491 EXPORT_SYMBOL(sock_wfree); 2492 2493 /* This variant of sock_wfree() is used by TCP, 2494 * since it sets SOCK_USE_WRITE_QUEUE. 2495 */ 2496 void __sock_wfree(struct sk_buff *skb) 2497 { 2498 struct sock *sk = skb->sk; 2499 2500 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc)) 2501 __sk_free(sk); 2502 } 2503 2504 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk) 2505 { 2506 skb_orphan(skb); 2507 skb->sk = sk; 2508 #ifdef CONFIG_INET 2509 if (unlikely(!sk_fullsock(sk))) { 2510 skb->destructor = sock_edemux; 2511 sock_hold(sk); 2512 return; 2513 } 2514 #endif 2515 skb->destructor = sock_wfree; 2516 skb_set_hash_from_sk(skb, sk); 2517 /* 2518 * We used to take a refcount on sk, but following operation 2519 * is enough to guarantee sk_free() wont free this sock until 2520 * all in-flight packets are completed 2521 */ 2522 refcount_add(skb->truesize, &sk->sk_wmem_alloc); 2523 } 2524 EXPORT_SYMBOL(skb_set_owner_w); 2525 2526 static bool can_skb_orphan_partial(const struct sk_buff *skb) 2527 { 2528 #ifdef CONFIG_TLS_DEVICE 2529 /* Drivers depend on in-order delivery for crypto offload, 2530 * partial orphan breaks out-of-order-OK logic. 2531 */ 2532 if (skb->decrypted) 2533 return false; 2534 #endif 2535 return (skb->destructor == sock_wfree || 2536 (IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree)); 2537 } 2538 2539 /* This helper is used by netem, as it can hold packets in its 2540 * delay queue. We want to allow the owner socket to send more 2541 * packets, as if they were already TX completed by a typical driver. 2542 * But we also want to keep skb->sk set because some packet schedulers 2543 * rely on it (sch_fq for example). 2544 */ 2545 void skb_orphan_partial(struct sk_buff *skb) 2546 { 2547 if (skb_is_tcp_pure_ack(skb)) 2548 return; 2549 2550 if (can_skb_orphan_partial(skb) && skb_set_owner_sk_safe(skb, skb->sk)) 2551 return; 2552 2553 skb_orphan(skb); 2554 } 2555 EXPORT_SYMBOL(skb_orphan_partial); 2556 2557 /* 2558 * Read buffer destructor automatically called from kfree_skb. 2559 */ 2560 void sock_rfree(struct sk_buff *skb) 2561 { 2562 struct sock *sk = skb->sk; 2563 unsigned int len = skb->truesize; 2564 2565 atomic_sub(len, &sk->sk_rmem_alloc); 2566 sk_mem_uncharge(sk, len); 2567 } 2568 EXPORT_SYMBOL(sock_rfree); 2569 2570 /* 2571 * Buffer destructor for skbs that are not used directly in read or write 2572 * path, e.g. for error handler skbs. Automatically called from kfree_skb. 2573 */ 2574 void sock_efree(struct sk_buff *skb) 2575 { 2576 sock_put(skb->sk); 2577 } 2578 EXPORT_SYMBOL(sock_efree); 2579 2580 /* Buffer destructor for prefetch/receive path where reference count may 2581 * not be held, e.g. for listen sockets. 2582 */ 2583 #ifdef CONFIG_INET 2584 void sock_pfree(struct sk_buff *skb) 2585 { 2586 if (sk_is_refcounted(skb->sk)) 2587 sock_gen_put(skb->sk); 2588 } 2589 EXPORT_SYMBOL(sock_pfree); 2590 #endif /* CONFIG_INET */ 2591 2592 kuid_t sock_i_uid(struct sock *sk) 2593 { 2594 kuid_t uid; 2595 2596 read_lock_bh(&sk->sk_callback_lock); 2597 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID; 2598 read_unlock_bh(&sk->sk_callback_lock); 2599 return uid; 2600 } 2601 EXPORT_SYMBOL(sock_i_uid); 2602 2603 unsigned long __sock_i_ino(struct sock *sk) 2604 { 2605 unsigned long ino; 2606 2607 read_lock(&sk->sk_callback_lock); 2608 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0; 2609 read_unlock(&sk->sk_callback_lock); 2610 return ino; 2611 } 2612 EXPORT_SYMBOL(__sock_i_ino); 2613 2614 unsigned long sock_i_ino(struct sock *sk) 2615 { 2616 unsigned long ino; 2617 2618 local_bh_disable(); 2619 ino = __sock_i_ino(sk); 2620 local_bh_enable(); 2621 return ino; 2622 } 2623 EXPORT_SYMBOL(sock_i_ino); 2624 2625 /* 2626 * Allocate a skb from the socket's send buffer. 2627 */ 2628 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 2629 gfp_t priority) 2630 { 2631 if (force || 2632 refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) { 2633 struct sk_buff *skb = alloc_skb(size, priority); 2634 2635 if (skb) { 2636 skb_set_owner_w(skb, sk); 2637 return skb; 2638 } 2639 } 2640 return NULL; 2641 } 2642 EXPORT_SYMBOL(sock_wmalloc); 2643 2644 static void sock_ofree(struct sk_buff *skb) 2645 { 2646 struct sock *sk = skb->sk; 2647 2648 atomic_sub(skb->truesize, &sk->sk_omem_alloc); 2649 } 2650 2651 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size, 2652 gfp_t priority) 2653 { 2654 struct sk_buff *skb; 2655 2656 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */ 2657 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) > 2658 READ_ONCE(sysctl_optmem_max)) 2659 return NULL; 2660 2661 skb = alloc_skb(size, priority); 2662 if (!skb) 2663 return NULL; 2664 2665 atomic_add(skb->truesize, &sk->sk_omem_alloc); 2666 skb->sk = sk; 2667 skb->destructor = sock_ofree; 2668 return skb; 2669 } 2670 2671 /* 2672 * Allocate a memory block from the socket's option memory buffer. 2673 */ 2674 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority) 2675 { 2676 int optmem_max = READ_ONCE(sysctl_optmem_max); 2677 2678 if ((unsigned int)size <= optmem_max && 2679 atomic_read(&sk->sk_omem_alloc) + size < optmem_max) { 2680 void *mem; 2681 /* First do the add, to avoid the race if kmalloc 2682 * might sleep. 2683 */ 2684 atomic_add(size, &sk->sk_omem_alloc); 2685 mem = kmalloc(size, priority); 2686 if (mem) 2687 return mem; 2688 atomic_sub(size, &sk->sk_omem_alloc); 2689 } 2690 return NULL; 2691 } 2692 EXPORT_SYMBOL(sock_kmalloc); 2693 2694 /* Free an option memory block. Note, we actually want the inline 2695 * here as this allows gcc to detect the nullify and fold away the 2696 * condition entirely. 2697 */ 2698 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size, 2699 const bool nullify) 2700 { 2701 if (WARN_ON_ONCE(!mem)) 2702 return; 2703 if (nullify) 2704 kfree_sensitive(mem); 2705 else 2706 kfree(mem); 2707 atomic_sub(size, &sk->sk_omem_alloc); 2708 } 2709 2710 void sock_kfree_s(struct sock *sk, void *mem, int size) 2711 { 2712 __sock_kfree_s(sk, mem, size, false); 2713 } 2714 EXPORT_SYMBOL(sock_kfree_s); 2715 2716 void sock_kzfree_s(struct sock *sk, void *mem, int size) 2717 { 2718 __sock_kfree_s(sk, mem, size, true); 2719 } 2720 EXPORT_SYMBOL(sock_kzfree_s); 2721 2722 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock. 2723 I think, these locks should be removed for datagram sockets. 2724 */ 2725 static long sock_wait_for_wmem(struct sock *sk, long timeo) 2726 { 2727 DEFINE_WAIT(wait); 2728 2729 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk); 2730 for (;;) { 2731 if (!timeo) 2732 break; 2733 if (signal_pending(current)) 2734 break; 2735 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 2736 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 2737 if (refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) 2738 break; 2739 if (sk->sk_shutdown & SEND_SHUTDOWN) 2740 break; 2741 if (sk->sk_err) 2742 break; 2743 timeo = schedule_timeout(timeo); 2744 } 2745 finish_wait(sk_sleep(sk), &wait); 2746 return timeo; 2747 } 2748 2749 2750 /* 2751 * Generic send/receive buffer handlers 2752 */ 2753 2754 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 2755 unsigned long data_len, int noblock, 2756 int *errcode, int max_page_order) 2757 { 2758 struct sk_buff *skb; 2759 long timeo; 2760 int err; 2761 2762 timeo = sock_sndtimeo(sk, noblock); 2763 for (;;) { 2764 err = sock_error(sk); 2765 if (err != 0) 2766 goto failure; 2767 2768 err = -EPIPE; 2769 if (sk->sk_shutdown & SEND_SHUTDOWN) 2770 goto failure; 2771 2772 if (sk_wmem_alloc_get(sk) < READ_ONCE(sk->sk_sndbuf)) 2773 break; 2774 2775 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk); 2776 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 2777 err = -EAGAIN; 2778 if (!timeo) 2779 goto failure; 2780 if (signal_pending(current)) 2781 goto interrupted; 2782 timeo = sock_wait_for_wmem(sk, timeo); 2783 } 2784 skb = alloc_skb_with_frags(header_len, data_len, max_page_order, 2785 errcode, sk->sk_allocation); 2786 if (skb) 2787 skb_set_owner_w(skb, sk); 2788 return skb; 2789 2790 interrupted: 2791 err = sock_intr_errno(timeo); 2792 failure: 2793 *errcode = err; 2794 return NULL; 2795 } 2796 EXPORT_SYMBOL(sock_alloc_send_pskb); 2797 2798 int __sock_cmsg_send(struct sock *sk, struct cmsghdr *cmsg, 2799 struct sockcm_cookie *sockc) 2800 { 2801 u32 tsflags; 2802 2803 switch (cmsg->cmsg_type) { 2804 case SO_MARK: 2805 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) && 2806 !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) 2807 return -EPERM; 2808 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 2809 return -EINVAL; 2810 sockc->mark = *(u32 *)CMSG_DATA(cmsg); 2811 break; 2812 case SO_TIMESTAMPING_OLD: 2813 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 2814 return -EINVAL; 2815 2816 tsflags = *(u32 *)CMSG_DATA(cmsg); 2817 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK) 2818 return -EINVAL; 2819 2820 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK; 2821 sockc->tsflags |= tsflags; 2822 break; 2823 case SCM_TXTIME: 2824 if (!sock_flag(sk, SOCK_TXTIME)) 2825 return -EINVAL; 2826 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64))) 2827 return -EINVAL; 2828 sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg)); 2829 break; 2830 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */ 2831 case SCM_RIGHTS: 2832 case SCM_CREDENTIALS: 2833 break; 2834 default: 2835 return -EINVAL; 2836 } 2837 return 0; 2838 } 2839 EXPORT_SYMBOL(__sock_cmsg_send); 2840 2841 int sock_cmsg_send(struct sock *sk, struct msghdr *msg, 2842 struct sockcm_cookie *sockc) 2843 { 2844 struct cmsghdr *cmsg; 2845 int ret; 2846 2847 for_each_cmsghdr(cmsg, msg) { 2848 if (!CMSG_OK(msg, cmsg)) 2849 return -EINVAL; 2850 if (cmsg->cmsg_level != SOL_SOCKET) 2851 continue; 2852 ret = __sock_cmsg_send(sk, cmsg, sockc); 2853 if (ret) 2854 return ret; 2855 } 2856 return 0; 2857 } 2858 EXPORT_SYMBOL(sock_cmsg_send); 2859 2860 static void sk_enter_memory_pressure(struct sock *sk) 2861 { 2862 if (!sk->sk_prot->enter_memory_pressure) 2863 return; 2864 2865 sk->sk_prot->enter_memory_pressure(sk); 2866 } 2867 2868 static void sk_leave_memory_pressure(struct sock *sk) 2869 { 2870 if (sk->sk_prot->leave_memory_pressure) { 2871 INDIRECT_CALL_INET_1(sk->sk_prot->leave_memory_pressure, 2872 tcp_leave_memory_pressure, sk); 2873 } else { 2874 unsigned long *memory_pressure = sk->sk_prot->memory_pressure; 2875 2876 if (memory_pressure && READ_ONCE(*memory_pressure)) 2877 WRITE_ONCE(*memory_pressure, 0); 2878 } 2879 } 2880 2881 DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key); 2882 2883 /** 2884 * skb_page_frag_refill - check that a page_frag contains enough room 2885 * @sz: minimum size of the fragment we want to get 2886 * @pfrag: pointer to page_frag 2887 * @gfp: priority for memory allocation 2888 * 2889 * Note: While this allocator tries to use high order pages, there is 2890 * no guarantee that allocations succeed. Therefore, @sz MUST be 2891 * less or equal than PAGE_SIZE. 2892 */ 2893 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp) 2894 { 2895 if (pfrag->page) { 2896 if (page_ref_count(pfrag->page) == 1) { 2897 pfrag->offset = 0; 2898 return true; 2899 } 2900 if (pfrag->offset + sz <= pfrag->size) 2901 return true; 2902 put_page(pfrag->page); 2903 } 2904 2905 pfrag->offset = 0; 2906 if (SKB_FRAG_PAGE_ORDER && 2907 !static_branch_unlikely(&net_high_order_alloc_disable_key)) { 2908 /* Avoid direct reclaim but allow kswapd to wake */ 2909 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) | 2910 __GFP_COMP | __GFP_NOWARN | 2911 __GFP_NORETRY, 2912 SKB_FRAG_PAGE_ORDER); 2913 if (likely(pfrag->page)) { 2914 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER; 2915 return true; 2916 } 2917 } 2918 pfrag->page = alloc_page(gfp); 2919 if (likely(pfrag->page)) { 2920 pfrag->size = PAGE_SIZE; 2921 return true; 2922 } 2923 return false; 2924 } 2925 EXPORT_SYMBOL(skb_page_frag_refill); 2926 2927 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag) 2928 { 2929 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation))) 2930 return true; 2931 2932 sk_enter_memory_pressure(sk); 2933 sk_stream_moderate_sndbuf(sk); 2934 return false; 2935 } 2936 EXPORT_SYMBOL(sk_page_frag_refill); 2937 2938 void __lock_sock(struct sock *sk) 2939 __releases(&sk->sk_lock.slock) 2940 __acquires(&sk->sk_lock.slock) 2941 { 2942 DEFINE_WAIT(wait); 2943 2944 for (;;) { 2945 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait, 2946 TASK_UNINTERRUPTIBLE); 2947 spin_unlock_bh(&sk->sk_lock.slock); 2948 schedule(); 2949 spin_lock_bh(&sk->sk_lock.slock); 2950 if (!sock_owned_by_user(sk)) 2951 break; 2952 } 2953 finish_wait(&sk->sk_lock.wq, &wait); 2954 } 2955 2956 void __release_sock(struct sock *sk) 2957 __releases(&sk->sk_lock.slock) 2958 __acquires(&sk->sk_lock.slock) 2959 { 2960 struct sk_buff *skb, *next; 2961 2962 while ((skb = sk->sk_backlog.head) != NULL) { 2963 sk->sk_backlog.head = sk->sk_backlog.tail = NULL; 2964 2965 spin_unlock_bh(&sk->sk_lock.slock); 2966 2967 do { 2968 next = skb->next; 2969 prefetch(next); 2970 DEBUG_NET_WARN_ON_ONCE(skb_dst_is_noref(skb)); 2971 skb_mark_not_on_list(skb); 2972 sk_backlog_rcv(sk, skb); 2973 2974 cond_resched(); 2975 2976 skb = next; 2977 } while (skb != NULL); 2978 2979 spin_lock_bh(&sk->sk_lock.slock); 2980 } 2981 2982 /* 2983 * Doing the zeroing here guarantee we can not loop forever 2984 * while a wild producer attempts to flood us. 2985 */ 2986 sk->sk_backlog.len = 0; 2987 } 2988 2989 void __sk_flush_backlog(struct sock *sk) 2990 { 2991 spin_lock_bh(&sk->sk_lock.slock); 2992 __release_sock(sk); 2993 spin_unlock_bh(&sk->sk_lock.slock); 2994 } 2995 EXPORT_SYMBOL_GPL(__sk_flush_backlog); 2996 2997 /** 2998 * sk_wait_data - wait for data to arrive at sk_receive_queue 2999 * @sk: sock to wait on 3000 * @timeo: for how long 3001 * @skb: last skb seen on sk_receive_queue 3002 * 3003 * Now socket state including sk->sk_err is changed only under lock, 3004 * hence we may omit checks after joining wait queue. 3005 * We check receive queue before schedule() only as optimization; 3006 * it is very likely that release_sock() added new data. 3007 */ 3008 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb) 3009 { 3010 DEFINE_WAIT_FUNC(wait, woken_wake_function); 3011 int rc; 3012 3013 add_wait_queue(sk_sleep(sk), &wait); 3014 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk); 3015 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait); 3016 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk); 3017 remove_wait_queue(sk_sleep(sk), &wait); 3018 return rc; 3019 } 3020 EXPORT_SYMBOL(sk_wait_data); 3021 3022 /** 3023 * __sk_mem_raise_allocated - increase memory_allocated 3024 * @sk: socket 3025 * @size: memory size to allocate 3026 * @amt: pages to allocate 3027 * @kind: allocation type 3028 * 3029 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc 3030 */ 3031 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind) 3032 { 3033 bool memcg_charge = mem_cgroup_sockets_enabled && sk->sk_memcg; 3034 struct proto *prot = sk->sk_prot; 3035 bool charged = true; 3036 long allocated; 3037 3038 sk_memory_allocated_add(sk, amt); 3039 allocated = sk_memory_allocated(sk); 3040 if (memcg_charge && 3041 !(charged = mem_cgroup_charge_skmem(sk->sk_memcg, amt, 3042 gfp_memcg_charge()))) 3043 goto suppress_allocation; 3044 3045 /* Under limit. */ 3046 if (allocated <= sk_prot_mem_limits(sk, 0)) { 3047 sk_leave_memory_pressure(sk); 3048 return 1; 3049 } 3050 3051 /* Under pressure. */ 3052 if (allocated > sk_prot_mem_limits(sk, 1)) 3053 sk_enter_memory_pressure(sk); 3054 3055 /* Over hard limit. */ 3056 if (allocated > sk_prot_mem_limits(sk, 2)) 3057 goto suppress_allocation; 3058 3059 /* guarantee minimum buffer size under pressure */ 3060 if (kind == SK_MEM_RECV) { 3061 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot)) 3062 return 1; 3063 3064 } else { /* SK_MEM_SEND */ 3065 int wmem0 = sk_get_wmem0(sk, prot); 3066 3067 if (sk->sk_type == SOCK_STREAM) { 3068 if (sk->sk_wmem_queued < wmem0) 3069 return 1; 3070 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) { 3071 return 1; 3072 } 3073 } 3074 3075 if (sk_has_memory_pressure(sk)) { 3076 u64 alloc; 3077 3078 if (!sk_under_memory_pressure(sk)) 3079 return 1; 3080 alloc = sk_sockets_allocated_read_positive(sk); 3081 if (sk_prot_mem_limits(sk, 2) > alloc * 3082 sk_mem_pages(sk->sk_wmem_queued + 3083 atomic_read(&sk->sk_rmem_alloc) + 3084 sk->sk_forward_alloc)) 3085 return 1; 3086 } 3087 3088 suppress_allocation: 3089 3090 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) { 3091 sk_stream_moderate_sndbuf(sk); 3092 3093 /* Fail only if socket is _under_ its sndbuf. 3094 * In this case we cannot block, so that we have to fail. 3095 */ 3096 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf) { 3097 /* Force charge with __GFP_NOFAIL */ 3098 if (memcg_charge && !charged) { 3099 mem_cgroup_charge_skmem(sk->sk_memcg, amt, 3100 gfp_memcg_charge() | __GFP_NOFAIL); 3101 } 3102 return 1; 3103 } 3104 } 3105 3106 if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged)) 3107 trace_sock_exceed_buf_limit(sk, prot, allocated, kind); 3108 3109 sk_memory_allocated_sub(sk, amt); 3110 3111 if (memcg_charge && charged) 3112 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt); 3113 3114 return 0; 3115 } 3116 3117 /** 3118 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated 3119 * @sk: socket 3120 * @size: memory size to allocate 3121 * @kind: allocation type 3122 * 3123 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means 3124 * rmem allocation. This function assumes that protocols which have 3125 * memory_pressure use sk_wmem_queued as write buffer accounting. 3126 */ 3127 int __sk_mem_schedule(struct sock *sk, int size, int kind) 3128 { 3129 int ret, amt = sk_mem_pages(size); 3130 3131 sk->sk_forward_alloc += amt << PAGE_SHIFT; 3132 ret = __sk_mem_raise_allocated(sk, size, amt, kind); 3133 if (!ret) 3134 sk->sk_forward_alloc -= amt << PAGE_SHIFT; 3135 return ret; 3136 } 3137 EXPORT_SYMBOL(__sk_mem_schedule); 3138 3139 /** 3140 * __sk_mem_reduce_allocated - reclaim memory_allocated 3141 * @sk: socket 3142 * @amount: number of quanta 3143 * 3144 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc 3145 */ 3146 void __sk_mem_reduce_allocated(struct sock *sk, int amount) 3147 { 3148 sk_memory_allocated_sub(sk, amount); 3149 3150 if (mem_cgroup_sockets_enabled && sk->sk_memcg) 3151 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount); 3152 3153 if (sk_under_memory_pressure(sk) && 3154 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0))) 3155 sk_leave_memory_pressure(sk); 3156 } 3157 3158 /** 3159 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated 3160 * @sk: socket 3161 * @amount: number of bytes (rounded down to a PAGE_SIZE multiple) 3162 */ 3163 void __sk_mem_reclaim(struct sock *sk, int amount) 3164 { 3165 amount >>= PAGE_SHIFT; 3166 sk->sk_forward_alloc -= amount << PAGE_SHIFT; 3167 __sk_mem_reduce_allocated(sk, amount); 3168 } 3169 EXPORT_SYMBOL(__sk_mem_reclaim); 3170 3171 int sk_set_peek_off(struct sock *sk, int val) 3172 { 3173 WRITE_ONCE(sk->sk_peek_off, val); 3174 return 0; 3175 } 3176 EXPORT_SYMBOL_GPL(sk_set_peek_off); 3177 3178 /* 3179 * Set of default routines for initialising struct proto_ops when 3180 * the protocol does not support a particular function. In certain 3181 * cases where it makes no sense for a protocol to have a "do nothing" 3182 * function, some default processing is provided. 3183 */ 3184 3185 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len) 3186 { 3187 return -EOPNOTSUPP; 3188 } 3189 EXPORT_SYMBOL(sock_no_bind); 3190 3191 int sock_no_connect(struct socket *sock, struct sockaddr *saddr, 3192 int len, int flags) 3193 { 3194 return -EOPNOTSUPP; 3195 } 3196 EXPORT_SYMBOL(sock_no_connect); 3197 3198 int sock_no_socketpair(struct socket *sock1, struct socket *sock2) 3199 { 3200 return -EOPNOTSUPP; 3201 } 3202 EXPORT_SYMBOL(sock_no_socketpair); 3203 3204 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags, 3205 bool kern) 3206 { 3207 return -EOPNOTSUPP; 3208 } 3209 EXPORT_SYMBOL(sock_no_accept); 3210 3211 int sock_no_getname(struct socket *sock, struct sockaddr *saddr, 3212 int peer) 3213 { 3214 return -EOPNOTSUPP; 3215 } 3216 EXPORT_SYMBOL(sock_no_getname); 3217 3218 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 3219 { 3220 return -EOPNOTSUPP; 3221 } 3222 EXPORT_SYMBOL(sock_no_ioctl); 3223 3224 int sock_no_listen(struct socket *sock, int backlog) 3225 { 3226 return -EOPNOTSUPP; 3227 } 3228 EXPORT_SYMBOL(sock_no_listen); 3229 3230 int sock_no_shutdown(struct socket *sock, int how) 3231 { 3232 return -EOPNOTSUPP; 3233 } 3234 EXPORT_SYMBOL(sock_no_shutdown); 3235 3236 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len) 3237 { 3238 return -EOPNOTSUPP; 3239 } 3240 EXPORT_SYMBOL(sock_no_sendmsg); 3241 3242 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len) 3243 { 3244 return -EOPNOTSUPP; 3245 } 3246 EXPORT_SYMBOL(sock_no_sendmsg_locked); 3247 3248 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len, 3249 int flags) 3250 { 3251 return -EOPNOTSUPP; 3252 } 3253 EXPORT_SYMBOL(sock_no_recvmsg); 3254 3255 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma) 3256 { 3257 /* Mirror missing mmap method error code */ 3258 return -ENODEV; 3259 } 3260 EXPORT_SYMBOL(sock_no_mmap); 3261 3262 /* 3263 * When a file is received (via SCM_RIGHTS, etc), we must bump the 3264 * various sock-based usage counts. 3265 */ 3266 void __receive_sock(struct file *file) 3267 { 3268 struct socket *sock; 3269 3270 sock = sock_from_file(file); 3271 if (sock) { 3272 sock_update_netprioidx(&sock->sk->sk_cgrp_data); 3273 sock_update_classid(&sock->sk->sk_cgrp_data); 3274 } 3275 } 3276 3277 /* 3278 * Default Socket Callbacks 3279 */ 3280 3281 static void sock_def_wakeup(struct sock *sk) 3282 { 3283 struct socket_wq *wq; 3284 3285 rcu_read_lock(); 3286 wq = rcu_dereference(sk->sk_wq); 3287 if (skwq_has_sleeper(wq)) 3288 wake_up_interruptible_all(&wq->wait); 3289 rcu_read_unlock(); 3290 } 3291 3292 static void sock_def_error_report(struct sock *sk) 3293 { 3294 struct socket_wq *wq; 3295 3296 rcu_read_lock(); 3297 wq = rcu_dereference(sk->sk_wq); 3298 if (skwq_has_sleeper(wq)) 3299 wake_up_interruptible_poll(&wq->wait, EPOLLERR); 3300 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR); 3301 rcu_read_unlock(); 3302 } 3303 3304 void sock_def_readable(struct sock *sk) 3305 { 3306 struct socket_wq *wq; 3307 3308 trace_sk_data_ready(sk); 3309 3310 rcu_read_lock(); 3311 wq = rcu_dereference(sk->sk_wq); 3312 if (skwq_has_sleeper(wq)) 3313 wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI | 3314 EPOLLRDNORM | EPOLLRDBAND); 3315 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN); 3316 rcu_read_unlock(); 3317 } 3318 3319 static void sock_def_write_space(struct sock *sk) 3320 { 3321 struct socket_wq *wq; 3322 3323 rcu_read_lock(); 3324 3325 /* Do not wake up a writer until he can make "significant" 3326 * progress. --DaveM 3327 */ 3328 if (sock_writeable(sk)) { 3329 wq = rcu_dereference(sk->sk_wq); 3330 if (skwq_has_sleeper(wq)) 3331 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT | 3332 EPOLLWRNORM | EPOLLWRBAND); 3333 3334 /* Should agree with poll, otherwise some programs break */ 3335 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT); 3336 } 3337 3338 rcu_read_unlock(); 3339 } 3340 3341 /* An optimised version of sock_def_write_space(), should only be called 3342 * for SOCK_RCU_FREE sockets under RCU read section and after putting 3343 * ->sk_wmem_alloc. 3344 */ 3345 static void sock_def_write_space_wfree(struct sock *sk) 3346 { 3347 /* Do not wake up a writer until he can make "significant" 3348 * progress. --DaveM 3349 */ 3350 if (sock_writeable(sk)) { 3351 struct socket_wq *wq = rcu_dereference(sk->sk_wq); 3352 3353 /* rely on refcount_sub from sock_wfree() */ 3354 smp_mb__after_atomic(); 3355 if (wq && waitqueue_active(&wq->wait)) 3356 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT | 3357 EPOLLWRNORM | EPOLLWRBAND); 3358 3359 /* Should agree with poll, otherwise some programs break */ 3360 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT); 3361 } 3362 } 3363 3364 static void sock_def_destruct(struct sock *sk) 3365 { 3366 } 3367 3368 void sk_send_sigurg(struct sock *sk) 3369 { 3370 if (sk->sk_socket && sk->sk_socket->file) 3371 if (send_sigurg(&sk->sk_socket->file->f_owner)) 3372 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI); 3373 } 3374 EXPORT_SYMBOL(sk_send_sigurg); 3375 3376 void sk_reset_timer(struct sock *sk, struct timer_list* timer, 3377 unsigned long expires) 3378 { 3379 if (!mod_timer(timer, expires)) 3380 sock_hold(sk); 3381 } 3382 EXPORT_SYMBOL(sk_reset_timer); 3383 3384 void sk_stop_timer(struct sock *sk, struct timer_list* timer) 3385 { 3386 if (del_timer(timer)) 3387 __sock_put(sk); 3388 } 3389 EXPORT_SYMBOL(sk_stop_timer); 3390 3391 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer) 3392 { 3393 if (del_timer_sync(timer)) 3394 __sock_put(sk); 3395 } 3396 EXPORT_SYMBOL(sk_stop_timer_sync); 3397 3398 void sock_init_data_uid(struct socket *sock, struct sock *sk, kuid_t uid) 3399 { 3400 sk_init_common(sk); 3401 sk->sk_send_head = NULL; 3402 3403 timer_setup(&sk->sk_timer, NULL, 0); 3404 3405 sk->sk_allocation = GFP_KERNEL; 3406 sk->sk_rcvbuf = READ_ONCE(sysctl_rmem_default); 3407 sk->sk_sndbuf = READ_ONCE(sysctl_wmem_default); 3408 sk->sk_state = TCP_CLOSE; 3409 sk->sk_use_task_frag = true; 3410 sk_set_socket(sk, sock); 3411 3412 sock_set_flag(sk, SOCK_ZAPPED); 3413 3414 if (sock) { 3415 sk->sk_type = sock->type; 3416 RCU_INIT_POINTER(sk->sk_wq, &sock->wq); 3417 sock->sk = sk; 3418 } else { 3419 RCU_INIT_POINTER(sk->sk_wq, NULL); 3420 } 3421 sk->sk_uid = uid; 3422 3423 rwlock_init(&sk->sk_callback_lock); 3424 if (sk->sk_kern_sock) 3425 lockdep_set_class_and_name( 3426 &sk->sk_callback_lock, 3427 af_kern_callback_keys + sk->sk_family, 3428 af_family_kern_clock_key_strings[sk->sk_family]); 3429 else 3430 lockdep_set_class_and_name( 3431 &sk->sk_callback_lock, 3432 af_callback_keys + sk->sk_family, 3433 af_family_clock_key_strings[sk->sk_family]); 3434 3435 sk->sk_state_change = sock_def_wakeup; 3436 sk->sk_data_ready = sock_def_readable; 3437 sk->sk_write_space = sock_def_write_space; 3438 sk->sk_error_report = sock_def_error_report; 3439 sk->sk_destruct = sock_def_destruct; 3440 3441 sk->sk_frag.page = NULL; 3442 sk->sk_frag.offset = 0; 3443 sk->sk_peek_off = -1; 3444 3445 sk->sk_peer_pid = NULL; 3446 sk->sk_peer_cred = NULL; 3447 spin_lock_init(&sk->sk_peer_lock); 3448 3449 sk->sk_write_pending = 0; 3450 sk->sk_rcvlowat = 1; 3451 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; 3452 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT; 3453 3454 sk->sk_stamp = SK_DEFAULT_STAMP; 3455 #if BITS_PER_LONG==32 3456 seqlock_init(&sk->sk_stamp_seq); 3457 #endif 3458 atomic_set(&sk->sk_zckey, 0); 3459 3460 #ifdef CONFIG_NET_RX_BUSY_POLL 3461 sk->sk_napi_id = 0; 3462 sk->sk_ll_usec = READ_ONCE(sysctl_net_busy_read); 3463 #endif 3464 3465 sk->sk_max_pacing_rate = ~0UL; 3466 sk->sk_pacing_rate = ~0UL; 3467 WRITE_ONCE(sk->sk_pacing_shift, 10); 3468 sk->sk_incoming_cpu = -1; 3469 3470 sk_rx_queue_clear(sk); 3471 /* 3472 * Before updating sk_refcnt, we must commit prior changes to memory 3473 * (Documentation/RCU/rculist_nulls.rst for details) 3474 */ 3475 smp_wmb(); 3476 refcount_set(&sk->sk_refcnt, 1); 3477 atomic_set(&sk->sk_drops, 0); 3478 } 3479 EXPORT_SYMBOL(sock_init_data_uid); 3480 3481 void sock_init_data(struct socket *sock, struct sock *sk) 3482 { 3483 kuid_t uid = sock ? 3484 SOCK_INODE(sock)->i_uid : 3485 make_kuid(sock_net(sk)->user_ns, 0); 3486 3487 sock_init_data_uid(sock, sk, uid); 3488 } 3489 EXPORT_SYMBOL(sock_init_data); 3490 3491 void lock_sock_nested(struct sock *sk, int subclass) 3492 { 3493 /* The sk_lock has mutex_lock() semantics here. */ 3494 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_); 3495 3496 might_sleep(); 3497 spin_lock_bh(&sk->sk_lock.slock); 3498 if (sock_owned_by_user_nocheck(sk)) 3499 __lock_sock(sk); 3500 sk->sk_lock.owned = 1; 3501 spin_unlock_bh(&sk->sk_lock.slock); 3502 } 3503 EXPORT_SYMBOL(lock_sock_nested); 3504 3505 void release_sock(struct sock *sk) 3506 { 3507 spin_lock_bh(&sk->sk_lock.slock); 3508 if (sk->sk_backlog.tail) 3509 __release_sock(sk); 3510 3511 /* Warning : release_cb() might need to release sk ownership, 3512 * ie call sock_release_ownership(sk) before us. 3513 */ 3514 if (sk->sk_prot->release_cb) 3515 sk->sk_prot->release_cb(sk); 3516 3517 sock_release_ownership(sk); 3518 if (waitqueue_active(&sk->sk_lock.wq)) 3519 wake_up(&sk->sk_lock.wq); 3520 spin_unlock_bh(&sk->sk_lock.slock); 3521 } 3522 EXPORT_SYMBOL(release_sock); 3523 3524 bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock) 3525 { 3526 might_sleep(); 3527 spin_lock_bh(&sk->sk_lock.slock); 3528 3529 if (!sock_owned_by_user_nocheck(sk)) { 3530 /* 3531 * Fast path return with bottom halves disabled and 3532 * sock::sk_lock.slock held. 3533 * 3534 * The 'mutex' is not contended and holding 3535 * sock::sk_lock.slock prevents all other lockers to 3536 * proceed so the corresponding unlock_sock_fast() can 3537 * avoid the slow path of release_sock() completely and 3538 * just release slock. 3539 * 3540 * From a semantical POV this is equivalent to 'acquiring' 3541 * the 'mutex', hence the corresponding lockdep 3542 * mutex_release() has to happen in the fast path of 3543 * unlock_sock_fast(). 3544 */ 3545 return false; 3546 } 3547 3548 __lock_sock(sk); 3549 sk->sk_lock.owned = 1; 3550 __acquire(&sk->sk_lock.slock); 3551 spin_unlock_bh(&sk->sk_lock.slock); 3552 return true; 3553 } 3554 EXPORT_SYMBOL(__lock_sock_fast); 3555 3556 int sock_gettstamp(struct socket *sock, void __user *userstamp, 3557 bool timeval, bool time32) 3558 { 3559 struct sock *sk = sock->sk; 3560 struct timespec64 ts; 3561 3562 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 3563 ts = ktime_to_timespec64(sock_read_timestamp(sk)); 3564 if (ts.tv_sec == -1) 3565 return -ENOENT; 3566 if (ts.tv_sec == 0) { 3567 ktime_t kt = ktime_get_real(); 3568 sock_write_timestamp(sk, kt); 3569 ts = ktime_to_timespec64(kt); 3570 } 3571 3572 if (timeval) 3573 ts.tv_nsec /= 1000; 3574 3575 #ifdef CONFIG_COMPAT_32BIT_TIME 3576 if (time32) 3577 return put_old_timespec32(&ts, userstamp); 3578 #endif 3579 #ifdef CONFIG_SPARC64 3580 /* beware of padding in sparc64 timeval */ 3581 if (timeval && !in_compat_syscall()) { 3582 struct __kernel_old_timeval __user tv = { 3583 .tv_sec = ts.tv_sec, 3584 .tv_usec = ts.tv_nsec, 3585 }; 3586 if (copy_to_user(userstamp, &tv, sizeof(tv))) 3587 return -EFAULT; 3588 return 0; 3589 } 3590 #endif 3591 return put_timespec64(&ts, userstamp); 3592 } 3593 EXPORT_SYMBOL(sock_gettstamp); 3594 3595 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag) 3596 { 3597 if (!sock_flag(sk, flag)) { 3598 unsigned long previous_flags = sk->sk_flags; 3599 3600 sock_set_flag(sk, flag); 3601 /* 3602 * we just set one of the two flags which require net 3603 * time stamping, but time stamping might have been on 3604 * already because of the other one 3605 */ 3606 if (sock_needs_netstamp(sk) && 3607 !(previous_flags & SK_FLAGS_TIMESTAMP)) 3608 net_enable_timestamp(); 3609 } 3610 } 3611 3612 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, 3613 int level, int type) 3614 { 3615 struct sock_exterr_skb *serr; 3616 struct sk_buff *skb; 3617 int copied, err; 3618 3619 err = -EAGAIN; 3620 skb = sock_dequeue_err_skb(sk); 3621 if (skb == NULL) 3622 goto out; 3623 3624 copied = skb->len; 3625 if (copied > len) { 3626 msg->msg_flags |= MSG_TRUNC; 3627 copied = len; 3628 } 3629 err = skb_copy_datagram_msg(skb, 0, msg, copied); 3630 if (err) 3631 goto out_free_skb; 3632 3633 sock_recv_timestamp(msg, sk, skb); 3634 3635 serr = SKB_EXT_ERR(skb); 3636 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee); 3637 3638 msg->msg_flags |= MSG_ERRQUEUE; 3639 err = copied; 3640 3641 out_free_skb: 3642 kfree_skb(skb); 3643 out: 3644 return err; 3645 } 3646 EXPORT_SYMBOL(sock_recv_errqueue); 3647 3648 /* 3649 * Get a socket option on an socket. 3650 * 3651 * FIX: POSIX 1003.1g is very ambiguous here. It states that 3652 * asynchronous errors should be reported by getsockopt. We assume 3653 * this means if you specify SO_ERROR (otherwise whats the point of it). 3654 */ 3655 int sock_common_getsockopt(struct socket *sock, int level, int optname, 3656 char __user *optval, int __user *optlen) 3657 { 3658 struct sock *sk = sock->sk; 3659 3660 /* IPV6_ADDRFORM can change sk->sk_prot under us. */ 3661 return READ_ONCE(sk->sk_prot)->getsockopt(sk, level, optname, optval, optlen); 3662 } 3663 EXPORT_SYMBOL(sock_common_getsockopt); 3664 3665 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 3666 int flags) 3667 { 3668 struct sock *sk = sock->sk; 3669 int addr_len = 0; 3670 int err; 3671 3672 err = sk->sk_prot->recvmsg(sk, msg, size, flags, &addr_len); 3673 if (err >= 0) 3674 msg->msg_namelen = addr_len; 3675 return err; 3676 } 3677 EXPORT_SYMBOL(sock_common_recvmsg); 3678 3679 /* 3680 * Set socket options on an inet socket. 3681 */ 3682 int sock_common_setsockopt(struct socket *sock, int level, int optname, 3683 sockptr_t optval, unsigned int optlen) 3684 { 3685 struct sock *sk = sock->sk; 3686 3687 /* IPV6_ADDRFORM can change sk->sk_prot under us. */ 3688 return READ_ONCE(sk->sk_prot)->setsockopt(sk, level, optname, optval, optlen); 3689 } 3690 EXPORT_SYMBOL(sock_common_setsockopt); 3691 3692 void sk_common_release(struct sock *sk) 3693 { 3694 if (sk->sk_prot->destroy) 3695 sk->sk_prot->destroy(sk); 3696 3697 /* 3698 * Observation: when sk_common_release is called, processes have 3699 * no access to socket. But net still has. 3700 * Step one, detach it from networking: 3701 * 3702 * A. Remove from hash tables. 3703 */ 3704 3705 sk->sk_prot->unhash(sk); 3706 3707 /* 3708 * In this point socket cannot receive new packets, but it is possible 3709 * that some packets are in flight because some CPU runs receiver and 3710 * did hash table lookup before we unhashed socket. They will achieve 3711 * receive queue and will be purged by socket destructor. 3712 * 3713 * Also we still have packets pending on receive queue and probably, 3714 * our own packets waiting in device queues. sock_destroy will drain 3715 * receive queue, but transmitted packets will delay socket destruction 3716 * until the last reference will be released. 3717 */ 3718 3719 sock_orphan(sk); 3720 3721 xfrm_sk_free_policy(sk); 3722 3723 sock_put(sk); 3724 } 3725 EXPORT_SYMBOL(sk_common_release); 3726 3727 void sk_get_meminfo(const struct sock *sk, u32 *mem) 3728 { 3729 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS); 3730 3731 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk); 3732 mem[SK_MEMINFO_RCVBUF] = READ_ONCE(sk->sk_rcvbuf); 3733 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk); 3734 mem[SK_MEMINFO_SNDBUF] = READ_ONCE(sk->sk_sndbuf); 3735 mem[SK_MEMINFO_FWD_ALLOC] = sk->sk_forward_alloc; 3736 mem[SK_MEMINFO_WMEM_QUEUED] = READ_ONCE(sk->sk_wmem_queued); 3737 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc); 3738 mem[SK_MEMINFO_BACKLOG] = READ_ONCE(sk->sk_backlog.len); 3739 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops); 3740 } 3741 3742 #ifdef CONFIG_PROC_FS 3743 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR); 3744 3745 int sock_prot_inuse_get(struct net *net, struct proto *prot) 3746 { 3747 int cpu, idx = prot->inuse_idx; 3748 int res = 0; 3749 3750 for_each_possible_cpu(cpu) 3751 res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx]; 3752 3753 return res >= 0 ? res : 0; 3754 } 3755 EXPORT_SYMBOL_GPL(sock_prot_inuse_get); 3756 3757 int sock_inuse_get(struct net *net) 3758 { 3759 int cpu, res = 0; 3760 3761 for_each_possible_cpu(cpu) 3762 res += per_cpu_ptr(net->core.prot_inuse, cpu)->all; 3763 3764 return res; 3765 } 3766 3767 EXPORT_SYMBOL_GPL(sock_inuse_get); 3768 3769 static int __net_init sock_inuse_init_net(struct net *net) 3770 { 3771 net->core.prot_inuse = alloc_percpu(struct prot_inuse); 3772 if (net->core.prot_inuse == NULL) 3773 return -ENOMEM; 3774 return 0; 3775 } 3776 3777 static void __net_exit sock_inuse_exit_net(struct net *net) 3778 { 3779 free_percpu(net->core.prot_inuse); 3780 } 3781 3782 static struct pernet_operations net_inuse_ops = { 3783 .init = sock_inuse_init_net, 3784 .exit = sock_inuse_exit_net, 3785 }; 3786 3787 static __init int net_inuse_init(void) 3788 { 3789 if (register_pernet_subsys(&net_inuse_ops)) 3790 panic("Cannot initialize net inuse counters"); 3791 3792 return 0; 3793 } 3794 3795 core_initcall(net_inuse_init); 3796 3797 static int assign_proto_idx(struct proto *prot) 3798 { 3799 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR); 3800 3801 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) { 3802 pr_err("PROTO_INUSE_NR exhausted\n"); 3803 return -ENOSPC; 3804 } 3805 3806 set_bit(prot->inuse_idx, proto_inuse_idx); 3807 return 0; 3808 } 3809 3810 static void release_proto_idx(struct proto *prot) 3811 { 3812 if (prot->inuse_idx != PROTO_INUSE_NR - 1) 3813 clear_bit(prot->inuse_idx, proto_inuse_idx); 3814 } 3815 #else 3816 static inline int assign_proto_idx(struct proto *prot) 3817 { 3818 return 0; 3819 } 3820 3821 static inline void release_proto_idx(struct proto *prot) 3822 { 3823 } 3824 3825 #endif 3826 3827 static void tw_prot_cleanup(struct timewait_sock_ops *twsk_prot) 3828 { 3829 if (!twsk_prot) 3830 return; 3831 kfree(twsk_prot->twsk_slab_name); 3832 twsk_prot->twsk_slab_name = NULL; 3833 kmem_cache_destroy(twsk_prot->twsk_slab); 3834 twsk_prot->twsk_slab = NULL; 3835 } 3836 3837 static int tw_prot_init(const struct proto *prot) 3838 { 3839 struct timewait_sock_ops *twsk_prot = prot->twsk_prot; 3840 3841 if (!twsk_prot) 3842 return 0; 3843 3844 twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", 3845 prot->name); 3846 if (!twsk_prot->twsk_slab_name) 3847 return -ENOMEM; 3848 3849 twsk_prot->twsk_slab = 3850 kmem_cache_create(twsk_prot->twsk_slab_name, 3851 twsk_prot->twsk_obj_size, 0, 3852 SLAB_ACCOUNT | prot->slab_flags, 3853 NULL); 3854 if (!twsk_prot->twsk_slab) { 3855 pr_crit("%s: Can't create timewait sock SLAB cache!\n", 3856 prot->name); 3857 return -ENOMEM; 3858 } 3859 3860 return 0; 3861 } 3862 3863 static void req_prot_cleanup(struct request_sock_ops *rsk_prot) 3864 { 3865 if (!rsk_prot) 3866 return; 3867 kfree(rsk_prot->slab_name); 3868 rsk_prot->slab_name = NULL; 3869 kmem_cache_destroy(rsk_prot->slab); 3870 rsk_prot->slab = NULL; 3871 } 3872 3873 static int req_prot_init(const struct proto *prot) 3874 { 3875 struct request_sock_ops *rsk_prot = prot->rsk_prot; 3876 3877 if (!rsk_prot) 3878 return 0; 3879 3880 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", 3881 prot->name); 3882 if (!rsk_prot->slab_name) 3883 return -ENOMEM; 3884 3885 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name, 3886 rsk_prot->obj_size, 0, 3887 SLAB_ACCOUNT | prot->slab_flags, 3888 NULL); 3889 3890 if (!rsk_prot->slab) { 3891 pr_crit("%s: Can't create request sock SLAB cache!\n", 3892 prot->name); 3893 return -ENOMEM; 3894 } 3895 return 0; 3896 } 3897 3898 int proto_register(struct proto *prot, int alloc_slab) 3899 { 3900 int ret = -ENOBUFS; 3901 3902 if (prot->memory_allocated && !prot->sysctl_mem) { 3903 pr_err("%s: missing sysctl_mem\n", prot->name); 3904 return -EINVAL; 3905 } 3906 if (prot->memory_allocated && !prot->per_cpu_fw_alloc) { 3907 pr_err("%s: missing per_cpu_fw_alloc\n", prot->name); 3908 return -EINVAL; 3909 } 3910 if (alloc_slab) { 3911 prot->slab = kmem_cache_create_usercopy(prot->name, 3912 prot->obj_size, 0, 3913 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT | 3914 prot->slab_flags, 3915 prot->useroffset, prot->usersize, 3916 NULL); 3917 3918 if (prot->slab == NULL) { 3919 pr_crit("%s: Can't create sock SLAB cache!\n", 3920 prot->name); 3921 goto out; 3922 } 3923 3924 if (req_prot_init(prot)) 3925 goto out_free_request_sock_slab; 3926 3927 if (tw_prot_init(prot)) 3928 goto out_free_timewait_sock_slab; 3929 } 3930 3931 mutex_lock(&proto_list_mutex); 3932 ret = assign_proto_idx(prot); 3933 if (ret) { 3934 mutex_unlock(&proto_list_mutex); 3935 goto out_free_timewait_sock_slab; 3936 } 3937 list_add(&prot->node, &proto_list); 3938 mutex_unlock(&proto_list_mutex); 3939 return ret; 3940 3941 out_free_timewait_sock_slab: 3942 if (alloc_slab) 3943 tw_prot_cleanup(prot->twsk_prot); 3944 out_free_request_sock_slab: 3945 if (alloc_slab) { 3946 req_prot_cleanup(prot->rsk_prot); 3947 3948 kmem_cache_destroy(prot->slab); 3949 prot->slab = NULL; 3950 } 3951 out: 3952 return ret; 3953 } 3954 EXPORT_SYMBOL(proto_register); 3955 3956 void proto_unregister(struct proto *prot) 3957 { 3958 mutex_lock(&proto_list_mutex); 3959 release_proto_idx(prot); 3960 list_del(&prot->node); 3961 mutex_unlock(&proto_list_mutex); 3962 3963 kmem_cache_destroy(prot->slab); 3964 prot->slab = NULL; 3965 3966 req_prot_cleanup(prot->rsk_prot); 3967 tw_prot_cleanup(prot->twsk_prot); 3968 } 3969 EXPORT_SYMBOL(proto_unregister); 3970 3971 int sock_load_diag_module(int family, int protocol) 3972 { 3973 if (!protocol) { 3974 if (!sock_is_registered(family)) 3975 return -ENOENT; 3976 3977 return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK, 3978 NETLINK_SOCK_DIAG, family); 3979 } 3980 3981 #ifdef CONFIG_INET 3982 if (family == AF_INET && 3983 protocol != IPPROTO_RAW && 3984 protocol < MAX_INET_PROTOS && 3985 !rcu_access_pointer(inet_protos[protocol])) 3986 return -ENOENT; 3987 #endif 3988 3989 return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK, 3990 NETLINK_SOCK_DIAG, family, protocol); 3991 } 3992 EXPORT_SYMBOL(sock_load_diag_module); 3993 3994 #ifdef CONFIG_PROC_FS 3995 static void *proto_seq_start(struct seq_file *seq, loff_t *pos) 3996 __acquires(proto_list_mutex) 3997 { 3998 mutex_lock(&proto_list_mutex); 3999 return seq_list_start_head(&proto_list, *pos); 4000 } 4001 4002 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos) 4003 { 4004 return seq_list_next(v, &proto_list, pos); 4005 } 4006 4007 static void proto_seq_stop(struct seq_file *seq, void *v) 4008 __releases(proto_list_mutex) 4009 { 4010 mutex_unlock(&proto_list_mutex); 4011 } 4012 4013 static char proto_method_implemented(const void *method) 4014 { 4015 return method == NULL ? 'n' : 'y'; 4016 } 4017 static long sock_prot_memory_allocated(struct proto *proto) 4018 { 4019 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L; 4020 } 4021 4022 static const char *sock_prot_memory_pressure(struct proto *proto) 4023 { 4024 return proto->memory_pressure != NULL ? 4025 proto_memory_pressure(proto) ? "yes" : "no" : "NI"; 4026 } 4027 4028 static void proto_seq_printf(struct seq_file *seq, struct proto *proto) 4029 { 4030 4031 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s " 4032 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n", 4033 proto->name, 4034 proto->obj_size, 4035 sock_prot_inuse_get(seq_file_net(seq), proto), 4036 sock_prot_memory_allocated(proto), 4037 sock_prot_memory_pressure(proto), 4038 proto->max_header, 4039 proto->slab == NULL ? "no" : "yes", 4040 module_name(proto->owner), 4041 proto_method_implemented(proto->close), 4042 proto_method_implemented(proto->connect), 4043 proto_method_implemented(proto->disconnect), 4044 proto_method_implemented(proto->accept), 4045 proto_method_implemented(proto->ioctl), 4046 proto_method_implemented(proto->init), 4047 proto_method_implemented(proto->destroy), 4048 proto_method_implemented(proto->shutdown), 4049 proto_method_implemented(proto->setsockopt), 4050 proto_method_implemented(proto->getsockopt), 4051 proto_method_implemented(proto->sendmsg), 4052 proto_method_implemented(proto->recvmsg), 4053 proto_method_implemented(proto->bind), 4054 proto_method_implemented(proto->backlog_rcv), 4055 proto_method_implemented(proto->hash), 4056 proto_method_implemented(proto->unhash), 4057 proto_method_implemented(proto->get_port), 4058 proto_method_implemented(proto->enter_memory_pressure)); 4059 } 4060 4061 static int proto_seq_show(struct seq_file *seq, void *v) 4062 { 4063 if (v == &proto_list) 4064 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s", 4065 "protocol", 4066 "size", 4067 "sockets", 4068 "memory", 4069 "press", 4070 "maxhdr", 4071 "slab", 4072 "module", 4073 "cl co di ac io in de sh ss gs se re bi br ha uh gp em\n"); 4074 else 4075 proto_seq_printf(seq, list_entry(v, struct proto, node)); 4076 return 0; 4077 } 4078 4079 static const struct seq_operations proto_seq_ops = { 4080 .start = proto_seq_start, 4081 .next = proto_seq_next, 4082 .stop = proto_seq_stop, 4083 .show = proto_seq_show, 4084 }; 4085 4086 static __net_init int proto_init_net(struct net *net) 4087 { 4088 if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops, 4089 sizeof(struct seq_net_private))) 4090 return -ENOMEM; 4091 4092 return 0; 4093 } 4094 4095 static __net_exit void proto_exit_net(struct net *net) 4096 { 4097 remove_proc_entry("protocols", net->proc_net); 4098 } 4099 4100 4101 static __net_initdata struct pernet_operations proto_net_ops = { 4102 .init = proto_init_net, 4103 .exit = proto_exit_net, 4104 }; 4105 4106 static int __init proto_init(void) 4107 { 4108 return register_pernet_subsys(&proto_net_ops); 4109 } 4110 4111 subsys_initcall(proto_init); 4112 4113 #endif /* PROC_FS */ 4114 4115 #ifdef CONFIG_NET_RX_BUSY_POLL 4116 bool sk_busy_loop_end(void *p, unsigned long start_time) 4117 { 4118 struct sock *sk = p; 4119 4120 return !skb_queue_empty_lockless(&sk->sk_receive_queue) || 4121 sk_busy_loop_timeout(sk, start_time); 4122 } 4123 EXPORT_SYMBOL(sk_busy_loop_end); 4124 #endif /* CONFIG_NET_RX_BUSY_POLL */ 4125 4126 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len) 4127 { 4128 if (!sk->sk_prot->bind_add) 4129 return -EOPNOTSUPP; 4130 return sk->sk_prot->bind_add(sk, addr, addr_len); 4131 } 4132 EXPORT_SYMBOL(sock_bind_add); 4133 4134 /* Copy 'size' bytes from userspace and return `size` back to userspace */ 4135 int sock_ioctl_inout(struct sock *sk, unsigned int cmd, 4136 void __user *arg, void *karg, size_t size) 4137 { 4138 int ret; 4139 4140 if (copy_from_user(karg, arg, size)) 4141 return -EFAULT; 4142 4143 ret = READ_ONCE(sk->sk_prot)->ioctl(sk, cmd, karg); 4144 if (ret) 4145 return ret; 4146 4147 if (copy_to_user(arg, karg, size)) 4148 return -EFAULT; 4149 4150 return 0; 4151 } 4152 EXPORT_SYMBOL(sock_ioctl_inout); 4153 4154 /* This is the most common ioctl prep function, where the result (4 bytes) is 4155 * copied back to userspace if the ioctl() returns successfully. No input is 4156 * copied from userspace as input argument. 4157 */ 4158 static int sock_ioctl_out(struct sock *sk, unsigned int cmd, void __user *arg) 4159 { 4160 int ret, karg = 0; 4161 4162 ret = READ_ONCE(sk->sk_prot)->ioctl(sk, cmd, &karg); 4163 if (ret) 4164 return ret; 4165 4166 return put_user(karg, (int __user *)arg); 4167 } 4168 4169 /* A wrapper around sock ioctls, which copies the data from userspace 4170 * (depending on the protocol/ioctl), and copies back the result to userspace. 4171 * The main motivation for this function is to pass kernel memory to the 4172 * protocol ioctl callbacks, instead of userspace memory. 4173 */ 4174 int sk_ioctl(struct sock *sk, unsigned int cmd, void __user *arg) 4175 { 4176 int rc = 1; 4177 4178 if (sk->sk_type == SOCK_RAW && sk->sk_family == AF_INET) 4179 rc = ipmr_sk_ioctl(sk, cmd, arg); 4180 else if (sk->sk_type == SOCK_RAW && sk->sk_family == AF_INET6) 4181 rc = ip6mr_sk_ioctl(sk, cmd, arg); 4182 else if (sk_is_phonet(sk)) 4183 rc = phonet_sk_ioctl(sk, cmd, arg); 4184 4185 /* If ioctl was processed, returns its value */ 4186 if (rc <= 0) 4187 return rc; 4188 4189 /* Otherwise call the default handler */ 4190 return sock_ioctl_out(sk, cmd, arg); 4191 } 4192 EXPORT_SYMBOL(sk_ioctl); 4193