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