1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * INET An implementation of the TCP/IP protocol suite for the LINUX 4 * operating system. INET is implemented using the BSD Socket 5 * interface as the means of communication with the user level. 6 * 7 * Generic socket support routines. Memory allocators, socket lock/release 8 * handler for protocols to use and generic option handler. 9 * 10 * Authors: Ross Biro 11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 12 * Florian La Roche, <flla@stud.uni-sb.de> 13 * Alan Cox, <A.Cox@swansea.ac.uk> 14 * 15 * Fixes: 16 * Alan Cox : Numerous verify_area() problems 17 * Alan Cox : Connecting on a connecting socket 18 * now returns an error for tcp. 19 * Alan Cox : sock->protocol is set correctly. 20 * and is not sometimes left as 0. 21 * Alan Cox : connect handles icmp errors on a 22 * connect properly. Unfortunately there 23 * is a restart syscall nasty there. I 24 * can't match BSD without hacking the C 25 * library. Ideas urgently sought! 26 * Alan Cox : Disallow bind() to addresses that are 27 * not ours - especially broadcast ones!! 28 * Alan Cox : Socket 1024 _IS_ ok for users. (fencepost) 29 * Alan Cox : sock_wfree/sock_rfree don't destroy sockets, 30 * instead they leave that for the DESTROY timer. 31 * Alan Cox : Clean up error flag in accept 32 * Alan Cox : TCP ack handling is buggy, the DESTROY timer 33 * was buggy. Put a remove_sock() in the handler 34 * for memory when we hit 0. Also altered the timer 35 * code. The ACK stuff can wait and needs major 36 * TCP layer surgery. 37 * Alan Cox : Fixed TCP ack bug, removed remove sock 38 * and fixed timer/inet_bh race. 39 * Alan Cox : Added zapped flag for TCP 40 * Alan Cox : Move kfree_skb into skbuff.c and tidied up surplus code 41 * Alan Cox : for new sk_buff allocations wmalloc/rmalloc now call alloc_skb 42 * Alan Cox : kfree_s calls now are kfree_skbmem so we can track skb resources 43 * Alan Cox : Supports socket option broadcast now as does udp. Packet and raw need fixing. 44 * Alan Cox : Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so... 45 * Rick Sladkey : Relaxed UDP rules for matching packets. 46 * C.E.Hawkins : IFF_PROMISC/SIOCGHWADDR support 47 * Pauline Middelink : identd support 48 * Alan Cox : Fixed connect() taking signals I think. 49 * Alan Cox : SO_LINGER supported 50 * Alan Cox : Error reporting fixes 51 * Anonymous : inet_create tidied up (sk->reuse setting) 52 * Alan Cox : inet sockets don't set sk->type! 53 * Alan Cox : Split socket option code 54 * Alan Cox : Callbacks 55 * Alan Cox : Nagle flag for Charles & Johannes stuff 56 * Alex : Removed restriction on inet fioctl 57 * Alan Cox : Splitting INET from NET core 58 * Alan Cox : Fixed bogus SO_TYPE handling in getsockopt() 59 * Adam Caldwell : Missing return in SO_DONTROUTE/SO_DEBUG code 60 * Alan Cox : Split IP from generic code 61 * Alan Cox : New kfree_skbmem() 62 * Alan Cox : Make SO_DEBUG superuser only. 63 * Alan Cox : Allow anyone to clear SO_DEBUG 64 * (compatibility fix) 65 * Alan Cox : Added optimistic memory grabbing for AF_UNIX throughput. 66 * Alan Cox : Allocator for a socket is settable. 67 * Alan Cox : SO_ERROR includes soft errors. 68 * Alan Cox : Allow NULL arguments on some SO_ opts 69 * Alan Cox : Generic socket allocation to make hooks 70 * easier (suggested by Craig Metz). 71 * Michael Pall : SO_ERROR returns positive errno again 72 * Steve Whitehouse: Added default destructor to free 73 * protocol private data. 74 * Steve Whitehouse: Added various other default routines 75 * common to several socket families. 76 * Chris Evans : Call suser() check last on F_SETOWN 77 * Jay Schulist : Added SO_ATTACH_FILTER and SO_DETACH_FILTER. 78 * Andi Kleen : Add sock_kmalloc()/sock_kfree_s() 79 * Andi Kleen : Fix write_space callback 80 * Chris Evans : Security fixes - signedness again 81 * Arnaldo C. Melo : cleanups, use skb_queue_purge 82 * 83 * To Fix: 84 */ 85 86 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 87 88 #include <asm/unaligned.h> 89 #include <linux/capability.h> 90 #include <linux/errno.h> 91 #include <linux/errqueue.h> 92 #include <linux/types.h> 93 #include <linux/socket.h> 94 #include <linux/in.h> 95 #include <linux/kernel.h> 96 #include <linux/module.h> 97 #include <linux/proc_fs.h> 98 #include <linux/seq_file.h> 99 #include <linux/sched.h> 100 #include <linux/sched/mm.h> 101 #include <linux/timer.h> 102 #include <linux/string.h> 103 #include <linux/sockios.h> 104 #include <linux/net.h> 105 #include <linux/mm.h> 106 #include <linux/slab.h> 107 #include <linux/interrupt.h> 108 #include <linux/poll.h> 109 #include <linux/tcp.h> 110 #include <linux/init.h> 111 #include <linux/highmem.h> 112 #include <linux/user_namespace.h> 113 #include <linux/static_key.h> 114 #include <linux/memcontrol.h> 115 #include <linux/prefetch.h> 116 #include <linux/compat.h> 117 #include <linux/mroute.h> 118 #include <linux/mroute6.h> 119 #include <linux/icmpv6.h> 120 121 #include <linux/uaccess.h> 122 123 #include <linux/netdevice.h> 124 #include <net/protocol.h> 125 #include <linux/skbuff.h> 126 #include <net/net_namespace.h> 127 #include <net/request_sock.h> 128 #include <net/sock.h> 129 #include <linux/net_tstamp.h> 130 #include <net/xfrm.h> 131 #include <linux/ipsec.h> 132 #include <net/cls_cgroup.h> 133 #include <net/netprio_cgroup.h> 134 #include <linux/sock_diag.h> 135 136 #include <linux/filter.h> 137 #include <net/sock_reuseport.h> 138 #include <net/bpf_sk_storage.h> 139 140 #include <trace/events/sock.h> 141 142 #include <net/tcp.h> 143 #include <net/busy_poll.h> 144 #include <net/phonet/phonet.h> 145 146 #include <linux/ethtool.h> 147 148 #include "dev.h" 149 150 static DEFINE_MUTEX(proto_list_mutex); 151 static LIST_HEAD(proto_list); 152 153 static void sock_def_write_space_wfree(struct sock *sk); 154 static void sock_def_write_space(struct sock *sk); 155 156 /** 157 * sk_ns_capable - General socket capability test 158 * @sk: Socket to use a capability on or through 159 * @user_ns: The user namespace of the capability to use 160 * @cap: The capability to use 161 * 162 * Test to see if the opener of the socket had when the socket was 163 * created and the current process has the capability @cap in the user 164 * namespace @user_ns. 165 */ 166 bool sk_ns_capable(const struct sock *sk, 167 struct user_namespace *user_ns, int cap) 168 { 169 return file_ns_capable(sk->sk_socket->file, user_ns, cap) && 170 ns_capable(user_ns, cap); 171 } 172 EXPORT_SYMBOL(sk_ns_capable); 173 174 /** 175 * sk_capable - Socket global capability test 176 * @sk: Socket to use a capability on or through 177 * @cap: The global capability to use 178 * 179 * Test to see if the opener of the socket had when the socket was 180 * created and the current process has the capability @cap in all user 181 * namespaces. 182 */ 183 bool sk_capable(const struct sock *sk, int cap) 184 { 185 return sk_ns_capable(sk, &init_user_ns, cap); 186 } 187 EXPORT_SYMBOL(sk_capable); 188 189 /** 190 * sk_net_capable - Network namespace socket capability test 191 * @sk: Socket to use a capability on or through 192 * @cap: The capability to use 193 * 194 * Test to see if the opener of the socket had when the socket was created 195 * and the current process has the capability @cap over the network namespace 196 * the socket is a member of. 197 */ 198 bool sk_net_capable(const struct sock *sk, int cap) 199 { 200 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap); 201 } 202 EXPORT_SYMBOL(sk_net_capable); 203 204 /* 205 * Each address family might have different locking rules, so we have 206 * one slock key per address family and separate keys for internal and 207 * userspace sockets. 208 */ 209 static struct lock_class_key af_family_keys[AF_MAX]; 210 static struct lock_class_key af_family_kern_keys[AF_MAX]; 211 static struct lock_class_key af_family_slock_keys[AF_MAX]; 212 static struct lock_class_key af_family_kern_slock_keys[AF_MAX]; 213 214 /* 215 * Make lock validator output more readable. (we pre-construct these 216 * strings build-time, so that runtime initialization of socket 217 * locks is fast): 218 */ 219 220 #define _sock_locks(x) \ 221 x "AF_UNSPEC", x "AF_UNIX" , x "AF_INET" , \ 222 x "AF_AX25" , x "AF_IPX" , x "AF_APPLETALK", \ 223 x "AF_NETROM", x "AF_BRIDGE" , x "AF_ATMPVC" , \ 224 x "AF_X25" , x "AF_INET6" , x "AF_ROSE" , \ 225 x "AF_DECnet", x "AF_NETBEUI" , x "AF_SECURITY" , \ 226 x "AF_KEY" , x "AF_NETLINK" , x "AF_PACKET" , \ 227 x "AF_ASH" , x "AF_ECONET" , x "AF_ATMSVC" , \ 228 x "AF_RDS" , x "AF_SNA" , x "AF_IRDA" , \ 229 x "AF_PPPOX" , x "AF_WANPIPE" , x "AF_LLC" , \ 230 x "27" , x "28" , x "AF_CAN" , \ 231 x "AF_TIPC" , x "AF_BLUETOOTH", x "IUCV" , \ 232 x "AF_RXRPC" , x "AF_ISDN" , x "AF_PHONET" , \ 233 x "AF_IEEE802154", x "AF_CAIF" , x "AF_ALG" , \ 234 x "AF_NFC" , x "AF_VSOCK" , x "AF_KCM" , \ 235 x "AF_QIPCRTR", x "AF_SMC" , x "AF_XDP" , \ 236 x "AF_MCTP" , \ 237 x "AF_MAX" 238 239 static const char *const af_family_key_strings[AF_MAX+1] = { 240 _sock_locks("sk_lock-") 241 }; 242 static const char *const af_family_slock_key_strings[AF_MAX+1] = { 243 _sock_locks("slock-") 244 }; 245 static const char *const af_family_clock_key_strings[AF_MAX+1] = { 246 _sock_locks("clock-") 247 }; 248 249 static const char *const af_family_kern_key_strings[AF_MAX+1] = { 250 _sock_locks("k-sk_lock-") 251 }; 252 static const char *const af_family_kern_slock_key_strings[AF_MAX+1] = { 253 _sock_locks("k-slock-") 254 }; 255 static const char *const af_family_kern_clock_key_strings[AF_MAX+1] = { 256 _sock_locks("k-clock-") 257 }; 258 static const char *const af_family_rlock_key_strings[AF_MAX+1] = { 259 _sock_locks("rlock-") 260 }; 261 static const char *const af_family_wlock_key_strings[AF_MAX+1] = { 262 _sock_locks("wlock-") 263 }; 264 static const char *const af_family_elock_key_strings[AF_MAX+1] = { 265 _sock_locks("elock-") 266 }; 267 268 /* 269 * sk_callback_lock and sk queues locking rules are per-address-family, 270 * so split the lock classes by using a per-AF key: 271 */ 272 static struct lock_class_key af_callback_keys[AF_MAX]; 273 static struct lock_class_key af_rlock_keys[AF_MAX]; 274 static struct lock_class_key af_wlock_keys[AF_MAX]; 275 static struct lock_class_key af_elock_keys[AF_MAX]; 276 static struct lock_class_key af_kern_callback_keys[AF_MAX]; 277 278 /* Run time adjustable parameters. */ 279 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX; 280 EXPORT_SYMBOL(sysctl_wmem_max); 281 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX; 282 EXPORT_SYMBOL(sysctl_rmem_max); 283 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX; 284 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX; 285 286 /* Maximal space eaten by iovec or ancillary data plus some space */ 287 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512); 288 EXPORT_SYMBOL(sysctl_optmem_max); 289 290 int sysctl_tstamp_allow_data __read_mostly = 1; 291 292 DEFINE_STATIC_KEY_FALSE(memalloc_socks_key); 293 EXPORT_SYMBOL_GPL(memalloc_socks_key); 294 295 /** 296 * sk_set_memalloc - sets %SOCK_MEMALLOC 297 * @sk: socket to set it on 298 * 299 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves. 300 * It's the responsibility of the admin to adjust min_free_kbytes 301 * to meet the requirements 302 */ 303 void sk_set_memalloc(struct sock *sk) 304 { 305 sock_set_flag(sk, SOCK_MEMALLOC); 306 sk->sk_allocation |= __GFP_MEMALLOC; 307 static_branch_inc(&memalloc_socks_key); 308 } 309 EXPORT_SYMBOL_GPL(sk_set_memalloc); 310 311 void sk_clear_memalloc(struct sock *sk) 312 { 313 sock_reset_flag(sk, SOCK_MEMALLOC); 314 sk->sk_allocation &= ~__GFP_MEMALLOC; 315 static_branch_dec(&memalloc_socks_key); 316 317 /* 318 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward 319 * progress of swapping. SOCK_MEMALLOC may be cleared while 320 * it has rmem allocations due to the last swapfile being deactivated 321 * but there is a risk that the socket is unusable due to exceeding 322 * the rmem limits. Reclaim the reserves and obey rmem limits again. 323 */ 324 sk_mem_reclaim(sk); 325 } 326 EXPORT_SYMBOL_GPL(sk_clear_memalloc); 327 328 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb) 329 { 330 int ret; 331 unsigned int noreclaim_flag; 332 333 /* these should have been dropped before queueing */ 334 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC)); 335 336 noreclaim_flag = memalloc_noreclaim_save(); 337 ret = INDIRECT_CALL_INET(sk->sk_backlog_rcv, 338 tcp_v6_do_rcv, 339 tcp_v4_do_rcv, 340 sk, skb); 341 memalloc_noreclaim_restore(noreclaim_flag); 342 343 return ret; 344 } 345 EXPORT_SYMBOL(__sk_backlog_rcv); 346 347 void sk_error_report(struct sock *sk) 348 { 349 sk->sk_error_report(sk); 350 351 switch (sk->sk_family) { 352 case AF_INET: 353 fallthrough; 354 case AF_INET6: 355 trace_inet_sk_error_report(sk); 356 break; 357 default: 358 break; 359 } 360 } 361 EXPORT_SYMBOL(sk_error_report); 362 363 int sock_get_timeout(long timeo, void *optval, bool old_timeval) 364 { 365 struct __kernel_sock_timeval tv; 366 367 if (timeo == MAX_SCHEDULE_TIMEOUT) { 368 tv.tv_sec = 0; 369 tv.tv_usec = 0; 370 } else { 371 tv.tv_sec = timeo / HZ; 372 tv.tv_usec = ((timeo % HZ) * USEC_PER_SEC) / HZ; 373 } 374 375 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) { 376 struct old_timeval32 tv32 = { tv.tv_sec, tv.tv_usec }; 377 *(struct old_timeval32 *)optval = tv32; 378 return sizeof(tv32); 379 } 380 381 if (old_timeval) { 382 struct __kernel_old_timeval old_tv; 383 old_tv.tv_sec = tv.tv_sec; 384 old_tv.tv_usec = tv.tv_usec; 385 *(struct __kernel_old_timeval *)optval = old_tv; 386 return sizeof(old_tv); 387 } 388 389 *(struct __kernel_sock_timeval *)optval = tv; 390 return sizeof(tv); 391 } 392 EXPORT_SYMBOL(sock_get_timeout); 393 394 int sock_copy_user_timeval(struct __kernel_sock_timeval *tv, 395 sockptr_t optval, int optlen, bool old_timeval) 396 { 397 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) { 398 struct old_timeval32 tv32; 399 400 if (optlen < sizeof(tv32)) 401 return -EINVAL; 402 403 if (copy_from_sockptr(&tv32, optval, sizeof(tv32))) 404 return -EFAULT; 405 tv->tv_sec = tv32.tv_sec; 406 tv->tv_usec = tv32.tv_usec; 407 } else if (old_timeval) { 408 struct __kernel_old_timeval old_tv; 409 410 if (optlen < sizeof(old_tv)) 411 return -EINVAL; 412 if (copy_from_sockptr(&old_tv, optval, sizeof(old_tv))) 413 return -EFAULT; 414 tv->tv_sec = old_tv.tv_sec; 415 tv->tv_usec = old_tv.tv_usec; 416 } else { 417 if (optlen < sizeof(*tv)) 418 return -EINVAL; 419 if (copy_from_sockptr(tv, optval, sizeof(*tv))) 420 return -EFAULT; 421 } 422 423 return 0; 424 } 425 EXPORT_SYMBOL(sock_copy_user_timeval); 426 427 static int sock_set_timeout(long *timeo_p, sockptr_t optval, int optlen, 428 bool old_timeval) 429 { 430 struct __kernel_sock_timeval tv; 431 int err = sock_copy_user_timeval(&tv, optval, optlen, old_timeval); 432 long val; 433 434 if (err) 435 return err; 436 437 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC) 438 return -EDOM; 439 440 if (tv.tv_sec < 0) { 441 static int warned __read_mostly; 442 443 WRITE_ONCE(*timeo_p, 0); 444 if (warned < 10 && net_ratelimit()) { 445 warned++; 446 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n", 447 __func__, current->comm, task_pid_nr(current)); 448 } 449 return 0; 450 } 451 val = MAX_SCHEDULE_TIMEOUT; 452 if ((tv.tv_sec || tv.tv_usec) && 453 (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1))) 454 val = tv.tv_sec * HZ + DIV_ROUND_UP((unsigned long)tv.tv_usec, 455 USEC_PER_SEC / HZ); 456 WRITE_ONCE(*timeo_p, val); 457 return 0; 458 } 459 460 static bool sock_needs_netstamp(const struct sock *sk) 461 { 462 switch (sk->sk_family) { 463 case AF_UNSPEC: 464 case AF_UNIX: 465 return false; 466 default: 467 return true; 468 } 469 } 470 471 static void sock_disable_timestamp(struct sock *sk, unsigned long flags) 472 { 473 if (sk->sk_flags & flags) { 474 sk->sk_flags &= ~flags; 475 if (sock_needs_netstamp(sk) && 476 !(sk->sk_flags & SK_FLAGS_TIMESTAMP)) 477 net_disable_timestamp(); 478 } 479 } 480 481 482 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 483 { 484 unsigned long flags; 485 struct sk_buff_head *list = &sk->sk_receive_queue; 486 487 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) { 488 atomic_inc(&sk->sk_drops); 489 trace_sock_rcvqueue_full(sk, skb); 490 return -ENOMEM; 491 } 492 493 if (!sk_rmem_schedule(sk, skb, skb->truesize)) { 494 atomic_inc(&sk->sk_drops); 495 return -ENOBUFS; 496 } 497 498 skb->dev = NULL; 499 skb_set_owner_r(skb, sk); 500 501 /* we escape from rcu protected region, make sure we dont leak 502 * a norefcounted dst 503 */ 504 skb_dst_force(skb); 505 506 spin_lock_irqsave(&list->lock, flags); 507 sock_skb_set_dropcount(sk, skb); 508 __skb_queue_tail(list, skb); 509 spin_unlock_irqrestore(&list->lock, flags); 510 511 if (!sock_flag(sk, SOCK_DEAD)) 512 sk->sk_data_ready(sk); 513 return 0; 514 } 515 EXPORT_SYMBOL(__sock_queue_rcv_skb); 516 517 int sock_queue_rcv_skb_reason(struct sock *sk, struct sk_buff *skb, 518 enum skb_drop_reason *reason) 519 { 520 enum skb_drop_reason drop_reason; 521 int err; 522 523 err = sk_filter(sk, skb); 524 if (err) { 525 drop_reason = SKB_DROP_REASON_SOCKET_FILTER; 526 goto out; 527 } 528 err = __sock_queue_rcv_skb(sk, skb); 529 switch (err) { 530 case -ENOMEM: 531 drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF; 532 break; 533 case -ENOBUFS: 534 drop_reason = SKB_DROP_REASON_PROTO_MEM; 535 break; 536 default: 537 drop_reason = SKB_NOT_DROPPED_YET; 538 break; 539 } 540 out: 541 if (reason) 542 *reason = drop_reason; 543 return err; 544 } 545 EXPORT_SYMBOL(sock_queue_rcv_skb_reason); 546 547 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, 548 const int nested, unsigned int trim_cap, bool refcounted) 549 { 550 int rc = NET_RX_SUCCESS; 551 552 if (sk_filter_trim_cap(sk, skb, trim_cap)) 553 goto discard_and_relse; 554 555 skb->dev = NULL; 556 557 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) { 558 atomic_inc(&sk->sk_drops); 559 goto discard_and_relse; 560 } 561 if (nested) 562 bh_lock_sock_nested(sk); 563 else 564 bh_lock_sock(sk); 565 if (!sock_owned_by_user(sk)) { 566 /* 567 * trylock + unlock semantics: 568 */ 569 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_); 570 571 rc = sk_backlog_rcv(sk, skb); 572 573 mutex_release(&sk->sk_lock.dep_map, _RET_IP_); 574 } else if (sk_add_backlog(sk, skb, READ_ONCE(sk->sk_rcvbuf))) { 575 bh_unlock_sock(sk); 576 atomic_inc(&sk->sk_drops); 577 goto discard_and_relse; 578 } 579 580 bh_unlock_sock(sk); 581 out: 582 if (refcounted) 583 sock_put(sk); 584 return rc; 585 discard_and_relse: 586 kfree_skb(skb); 587 goto out; 588 } 589 EXPORT_SYMBOL(__sk_receive_skb); 590 591 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ip6_dst_check(struct dst_entry *, 592 u32)); 593 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ipv4_dst_check(struct dst_entry *, 594 u32)); 595 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie) 596 { 597 struct dst_entry *dst = __sk_dst_get(sk); 598 599 if (dst && dst->obsolete && 600 INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check, 601 dst, cookie) == NULL) { 602 sk_tx_queue_clear(sk); 603 sk->sk_dst_pending_confirm = 0; 604 RCU_INIT_POINTER(sk->sk_dst_cache, NULL); 605 dst_release(dst); 606 return NULL; 607 } 608 609 return dst; 610 } 611 EXPORT_SYMBOL(__sk_dst_check); 612 613 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie) 614 { 615 struct dst_entry *dst = sk_dst_get(sk); 616 617 if (dst && dst->obsolete && 618 INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check, 619 dst, cookie) == NULL) { 620 sk_dst_reset(sk); 621 dst_release(dst); 622 return NULL; 623 } 624 625 return dst; 626 } 627 EXPORT_SYMBOL(sk_dst_check); 628 629 static int sock_bindtoindex_locked(struct sock *sk, int ifindex) 630 { 631 int ret = -ENOPROTOOPT; 632 #ifdef CONFIG_NETDEVICES 633 struct net *net = sock_net(sk); 634 635 /* Sorry... */ 636 ret = -EPERM; 637 if (sk->sk_bound_dev_if && !ns_capable(net->user_ns, CAP_NET_RAW)) 638 goto out; 639 640 ret = -EINVAL; 641 if (ifindex < 0) 642 goto out; 643 644 /* Paired with all READ_ONCE() done locklessly. */ 645 WRITE_ONCE(sk->sk_bound_dev_if, ifindex); 646 647 if (sk->sk_prot->rehash) 648 sk->sk_prot->rehash(sk); 649 sk_dst_reset(sk); 650 651 ret = 0; 652 653 out: 654 #endif 655 656 return ret; 657 } 658 659 int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk) 660 { 661 int ret; 662 663 if (lock_sk) 664 lock_sock(sk); 665 ret = sock_bindtoindex_locked(sk, ifindex); 666 if (lock_sk) 667 release_sock(sk); 668 669 return ret; 670 } 671 EXPORT_SYMBOL(sock_bindtoindex); 672 673 static int sock_setbindtodevice(struct sock *sk, sockptr_t optval, int optlen) 674 { 675 int ret = -ENOPROTOOPT; 676 #ifdef CONFIG_NETDEVICES 677 struct net *net = sock_net(sk); 678 char devname[IFNAMSIZ]; 679 int index; 680 681 ret = -EINVAL; 682 if (optlen < 0) 683 goto out; 684 685 /* Bind this socket to a particular device like "eth0", 686 * as specified in the passed interface name. If the 687 * name is "" or the option length is zero the socket 688 * is not bound. 689 */ 690 if (optlen > IFNAMSIZ - 1) 691 optlen = IFNAMSIZ - 1; 692 memset(devname, 0, sizeof(devname)); 693 694 ret = -EFAULT; 695 if (copy_from_sockptr(devname, optval, optlen)) 696 goto out; 697 698 index = 0; 699 if (devname[0] != '\0') { 700 struct net_device *dev; 701 702 rcu_read_lock(); 703 dev = dev_get_by_name_rcu(net, devname); 704 if (dev) 705 index = dev->ifindex; 706 rcu_read_unlock(); 707 ret = -ENODEV; 708 if (!dev) 709 goto out; 710 } 711 712 sockopt_lock_sock(sk); 713 ret = sock_bindtoindex_locked(sk, index); 714 sockopt_release_sock(sk); 715 out: 716 #endif 717 718 return ret; 719 } 720 721 static int sock_getbindtodevice(struct sock *sk, sockptr_t optval, 722 sockptr_t optlen, int len) 723 { 724 int ret = -ENOPROTOOPT; 725 #ifdef CONFIG_NETDEVICES 726 int bound_dev_if = READ_ONCE(sk->sk_bound_dev_if); 727 struct net *net = sock_net(sk); 728 char devname[IFNAMSIZ]; 729 730 if (bound_dev_if == 0) { 731 len = 0; 732 goto zero; 733 } 734 735 ret = -EINVAL; 736 if (len < IFNAMSIZ) 737 goto out; 738 739 ret = netdev_get_name(net, devname, bound_dev_if); 740 if (ret) 741 goto out; 742 743 len = strlen(devname) + 1; 744 745 ret = -EFAULT; 746 if (copy_to_sockptr(optval, devname, len)) 747 goto out; 748 749 zero: 750 ret = -EFAULT; 751 if (copy_to_sockptr(optlen, &len, sizeof(int))) 752 goto out; 753 754 ret = 0; 755 756 out: 757 #endif 758 759 return ret; 760 } 761 762 bool sk_mc_loop(struct sock *sk) 763 { 764 if (dev_recursion_level()) 765 return false; 766 if (!sk) 767 return true; 768 switch (sk->sk_family) { 769 case AF_INET: 770 return inet_sk(sk)->mc_loop; 771 #if IS_ENABLED(CONFIG_IPV6) 772 case AF_INET6: 773 return inet6_sk(sk)->mc_loop; 774 #endif 775 } 776 WARN_ON_ONCE(1); 777 return true; 778 } 779 EXPORT_SYMBOL(sk_mc_loop); 780 781 void sock_set_reuseaddr(struct sock *sk) 782 { 783 lock_sock(sk); 784 sk->sk_reuse = SK_CAN_REUSE; 785 release_sock(sk); 786 } 787 EXPORT_SYMBOL(sock_set_reuseaddr); 788 789 void sock_set_reuseport(struct sock *sk) 790 { 791 lock_sock(sk); 792 sk->sk_reuseport = true; 793 release_sock(sk); 794 } 795 EXPORT_SYMBOL(sock_set_reuseport); 796 797 void sock_no_linger(struct sock *sk) 798 { 799 lock_sock(sk); 800 sk->sk_lingertime = 0; 801 sock_set_flag(sk, SOCK_LINGER); 802 release_sock(sk); 803 } 804 EXPORT_SYMBOL(sock_no_linger); 805 806 void sock_set_priority(struct sock *sk, u32 priority) 807 { 808 lock_sock(sk); 809 WRITE_ONCE(sk->sk_priority, priority); 810 release_sock(sk); 811 } 812 EXPORT_SYMBOL(sock_set_priority); 813 814 void sock_set_sndtimeo(struct sock *sk, s64 secs) 815 { 816 lock_sock(sk); 817 if (secs && secs < MAX_SCHEDULE_TIMEOUT / HZ - 1) 818 WRITE_ONCE(sk->sk_sndtimeo, secs * HZ); 819 else 820 WRITE_ONCE(sk->sk_sndtimeo, MAX_SCHEDULE_TIMEOUT); 821 release_sock(sk); 822 } 823 EXPORT_SYMBOL(sock_set_sndtimeo); 824 825 static void __sock_set_timestamps(struct sock *sk, bool val, bool new, bool ns) 826 { 827 if (val) { 828 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, new); 829 sock_valbool_flag(sk, SOCK_RCVTSTAMPNS, ns); 830 sock_set_flag(sk, SOCK_RCVTSTAMP); 831 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 832 } else { 833 sock_reset_flag(sk, SOCK_RCVTSTAMP); 834 sock_reset_flag(sk, SOCK_RCVTSTAMPNS); 835 } 836 } 837 838 void sock_enable_timestamps(struct sock *sk) 839 { 840 lock_sock(sk); 841 __sock_set_timestamps(sk, true, false, true); 842 release_sock(sk); 843 } 844 EXPORT_SYMBOL(sock_enable_timestamps); 845 846 void sock_set_timestamp(struct sock *sk, int optname, bool valbool) 847 { 848 switch (optname) { 849 case SO_TIMESTAMP_OLD: 850 __sock_set_timestamps(sk, valbool, false, false); 851 break; 852 case SO_TIMESTAMP_NEW: 853 __sock_set_timestamps(sk, valbool, true, false); 854 break; 855 case SO_TIMESTAMPNS_OLD: 856 __sock_set_timestamps(sk, valbool, false, true); 857 break; 858 case SO_TIMESTAMPNS_NEW: 859 __sock_set_timestamps(sk, valbool, true, true); 860 break; 861 } 862 } 863 864 static int sock_timestamping_bind_phc(struct sock *sk, int phc_index) 865 { 866 struct net *net = sock_net(sk); 867 struct net_device *dev = NULL; 868 bool match = false; 869 int *vclock_index; 870 int i, num; 871 872 if (sk->sk_bound_dev_if) 873 dev = dev_get_by_index(net, sk->sk_bound_dev_if); 874 875 if (!dev) { 876 pr_err("%s: sock not bind to device\n", __func__); 877 return -EOPNOTSUPP; 878 } 879 880 num = ethtool_get_phc_vclocks(dev, &vclock_index); 881 dev_put(dev); 882 883 for (i = 0; i < num; i++) { 884 if (*(vclock_index + i) == phc_index) { 885 match = true; 886 break; 887 } 888 } 889 890 if (num > 0) 891 kfree(vclock_index); 892 893 if (!match) 894 return -EINVAL; 895 896 sk->sk_bind_phc = phc_index; 897 898 return 0; 899 } 900 901 int sock_set_timestamping(struct sock *sk, int optname, 902 struct so_timestamping timestamping) 903 { 904 int val = timestamping.flags; 905 int ret; 906 907 if (val & ~SOF_TIMESTAMPING_MASK) 908 return -EINVAL; 909 910 if (val & SOF_TIMESTAMPING_OPT_ID_TCP && 911 !(val & SOF_TIMESTAMPING_OPT_ID)) 912 return -EINVAL; 913 914 if (val & SOF_TIMESTAMPING_OPT_ID && 915 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) { 916 if (sk_is_tcp(sk)) { 917 if ((1 << sk->sk_state) & 918 (TCPF_CLOSE | TCPF_LISTEN)) 919 return -EINVAL; 920 if (val & SOF_TIMESTAMPING_OPT_ID_TCP) 921 atomic_set(&sk->sk_tskey, tcp_sk(sk)->write_seq); 922 else 923 atomic_set(&sk->sk_tskey, tcp_sk(sk)->snd_una); 924 } else { 925 atomic_set(&sk->sk_tskey, 0); 926 } 927 } 928 929 if (val & SOF_TIMESTAMPING_OPT_STATS && 930 !(val & SOF_TIMESTAMPING_OPT_TSONLY)) 931 return -EINVAL; 932 933 if (val & SOF_TIMESTAMPING_BIND_PHC) { 934 ret = sock_timestamping_bind_phc(sk, timestamping.bind_phc); 935 if (ret) 936 return ret; 937 } 938 939 sk->sk_tsflags = val; 940 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, optname == SO_TIMESTAMPING_NEW); 941 942 if (val & SOF_TIMESTAMPING_RX_SOFTWARE) 943 sock_enable_timestamp(sk, 944 SOCK_TIMESTAMPING_RX_SOFTWARE); 945 else 946 sock_disable_timestamp(sk, 947 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE)); 948 return 0; 949 } 950 951 void sock_set_keepalive(struct sock *sk) 952 { 953 lock_sock(sk); 954 if (sk->sk_prot->keepalive) 955 sk->sk_prot->keepalive(sk, true); 956 sock_valbool_flag(sk, SOCK_KEEPOPEN, true); 957 release_sock(sk); 958 } 959 EXPORT_SYMBOL(sock_set_keepalive); 960 961 static void __sock_set_rcvbuf(struct sock *sk, int val) 962 { 963 /* Ensure val * 2 fits into an int, to prevent max_t() from treating it 964 * as a negative value. 965 */ 966 val = min_t(int, val, INT_MAX / 2); 967 sk->sk_userlocks |= SOCK_RCVBUF_LOCK; 968 969 /* We double it on the way in to account for "struct sk_buff" etc. 970 * overhead. Applications assume that the SO_RCVBUF setting they make 971 * will allow that much actual data to be received on that socket. 972 * 973 * Applications are unaware that "struct sk_buff" and other overheads 974 * allocate from the receive buffer during socket buffer allocation. 975 * 976 * And after considering the possible alternatives, returning the value 977 * we actually used in getsockopt is the most desirable behavior. 978 */ 979 WRITE_ONCE(sk->sk_rcvbuf, max_t(int, val * 2, SOCK_MIN_RCVBUF)); 980 } 981 982 void sock_set_rcvbuf(struct sock *sk, int val) 983 { 984 lock_sock(sk); 985 __sock_set_rcvbuf(sk, val); 986 release_sock(sk); 987 } 988 EXPORT_SYMBOL(sock_set_rcvbuf); 989 990 static void __sock_set_mark(struct sock *sk, u32 val) 991 { 992 if (val != sk->sk_mark) { 993 WRITE_ONCE(sk->sk_mark, val); 994 sk_dst_reset(sk); 995 } 996 } 997 998 void sock_set_mark(struct sock *sk, u32 val) 999 { 1000 lock_sock(sk); 1001 __sock_set_mark(sk, val); 1002 release_sock(sk); 1003 } 1004 EXPORT_SYMBOL(sock_set_mark); 1005 1006 static void sock_release_reserved_memory(struct sock *sk, int bytes) 1007 { 1008 /* Round down bytes to multiple of pages */ 1009 bytes = round_down(bytes, PAGE_SIZE); 1010 1011 WARN_ON(bytes > sk->sk_reserved_mem); 1012 WRITE_ONCE(sk->sk_reserved_mem, sk->sk_reserved_mem - bytes); 1013 sk_mem_reclaim(sk); 1014 } 1015 1016 static int sock_reserve_memory(struct sock *sk, int bytes) 1017 { 1018 long allocated; 1019 bool charged; 1020 int pages; 1021 1022 if (!mem_cgroup_sockets_enabled || !sk->sk_memcg || !sk_has_account(sk)) 1023 return -EOPNOTSUPP; 1024 1025 if (!bytes) 1026 return 0; 1027 1028 pages = sk_mem_pages(bytes); 1029 1030 /* pre-charge to memcg */ 1031 charged = mem_cgroup_charge_skmem(sk->sk_memcg, pages, 1032 GFP_KERNEL | __GFP_RETRY_MAYFAIL); 1033 if (!charged) 1034 return -ENOMEM; 1035 1036 /* pre-charge to forward_alloc */ 1037 sk_memory_allocated_add(sk, pages); 1038 allocated = sk_memory_allocated(sk); 1039 /* If the system goes into memory pressure with this 1040 * precharge, give up and return error. 1041 */ 1042 if (allocated > sk_prot_mem_limits(sk, 1)) { 1043 sk_memory_allocated_sub(sk, pages); 1044 mem_cgroup_uncharge_skmem(sk->sk_memcg, pages); 1045 return -ENOMEM; 1046 } 1047 sk->sk_forward_alloc += pages << PAGE_SHIFT; 1048 1049 WRITE_ONCE(sk->sk_reserved_mem, 1050 sk->sk_reserved_mem + (pages << PAGE_SHIFT)); 1051 1052 return 0; 1053 } 1054 1055 void sockopt_lock_sock(struct sock *sk) 1056 { 1057 /* When current->bpf_ctx is set, the setsockopt is called from 1058 * a bpf prog. bpf has ensured the sk lock has been 1059 * acquired before calling setsockopt(). 1060 */ 1061 if (has_current_bpf_ctx()) 1062 return; 1063 1064 lock_sock(sk); 1065 } 1066 EXPORT_SYMBOL(sockopt_lock_sock); 1067 1068 void sockopt_release_sock(struct sock *sk) 1069 { 1070 if (has_current_bpf_ctx()) 1071 return; 1072 1073 release_sock(sk); 1074 } 1075 EXPORT_SYMBOL(sockopt_release_sock); 1076 1077 bool sockopt_ns_capable(struct user_namespace *ns, int cap) 1078 { 1079 return has_current_bpf_ctx() || ns_capable(ns, cap); 1080 } 1081 EXPORT_SYMBOL(sockopt_ns_capable); 1082 1083 bool sockopt_capable(int cap) 1084 { 1085 return has_current_bpf_ctx() || capable(cap); 1086 } 1087 EXPORT_SYMBOL(sockopt_capable); 1088 1089 /* 1090 * This is meant for all protocols to use and covers goings on 1091 * at the socket level. Everything here is generic. 1092 */ 1093 1094 int sk_setsockopt(struct sock *sk, int level, int optname, 1095 sockptr_t optval, unsigned int optlen) 1096 { 1097 struct so_timestamping timestamping; 1098 struct socket *sock = sk->sk_socket; 1099 struct sock_txtime sk_txtime; 1100 int val; 1101 int valbool; 1102 struct linger ling; 1103 int ret = 0; 1104 1105 /* 1106 * Options without arguments 1107 */ 1108 1109 if (optname == SO_BINDTODEVICE) 1110 return sock_setbindtodevice(sk, optval, optlen); 1111 1112 if (optlen < sizeof(int)) 1113 return -EINVAL; 1114 1115 if (copy_from_sockptr(&val, optval, sizeof(val))) 1116 return -EFAULT; 1117 1118 valbool = val ? 1 : 0; 1119 1120 sockopt_lock_sock(sk); 1121 1122 switch (optname) { 1123 case SO_DEBUG: 1124 if (val && !sockopt_capable(CAP_NET_ADMIN)) 1125 ret = -EACCES; 1126 else 1127 sock_valbool_flag(sk, SOCK_DBG, valbool); 1128 break; 1129 case SO_REUSEADDR: 1130 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE); 1131 break; 1132 case SO_REUSEPORT: 1133 sk->sk_reuseport = valbool; 1134 break; 1135 case SO_TYPE: 1136 case SO_PROTOCOL: 1137 case SO_DOMAIN: 1138 case SO_ERROR: 1139 ret = -ENOPROTOOPT; 1140 break; 1141 case SO_DONTROUTE: 1142 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool); 1143 sk_dst_reset(sk); 1144 break; 1145 case SO_BROADCAST: 1146 sock_valbool_flag(sk, SOCK_BROADCAST, valbool); 1147 break; 1148 case SO_SNDBUF: 1149 /* Don't error on this BSD doesn't and if you think 1150 * about it this is right. Otherwise apps have to 1151 * play 'guess the biggest size' games. RCVBUF/SNDBUF 1152 * are treated in BSD as hints 1153 */ 1154 val = min_t(u32, val, READ_ONCE(sysctl_wmem_max)); 1155 set_sndbuf: 1156 /* Ensure val * 2 fits into an int, to prevent max_t() 1157 * from treating it as a negative value. 1158 */ 1159 val = min_t(int, val, INT_MAX / 2); 1160 sk->sk_userlocks |= SOCK_SNDBUF_LOCK; 1161 WRITE_ONCE(sk->sk_sndbuf, 1162 max_t(int, val * 2, SOCK_MIN_SNDBUF)); 1163 /* Wake up sending tasks if we upped the value. */ 1164 sk->sk_write_space(sk); 1165 break; 1166 1167 case SO_SNDBUFFORCE: 1168 if (!sockopt_capable(CAP_NET_ADMIN)) { 1169 ret = -EPERM; 1170 break; 1171 } 1172 1173 /* No negative values (to prevent underflow, as val will be 1174 * multiplied by 2). 1175 */ 1176 if (val < 0) 1177 val = 0; 1178 goto set_sndbuf; 1179 1180 case SO_RCVBUF: 1181 /* Don't error on this BSD doesn't and if you think 1182 * about it this is right. Otherwise apps have to 1183 * play 'guess the biggest size' games. RCVBUF/SNDBUF 1184 * are treated in BSD as hints 1185 */ 1186 __sock_set_rcvbuf(sk, min_t(u32, val, READ_ONCE(sysctl_rmem_max))); 1187 break; 1188 1189 case SO_RCVBUFFORCE: 1190 if (!sockopt_capable(CAP_NET_ADMIN)) { 1191 ret = -EPERM; 1192 break; 1193 } 1194 1195 /* No negative values (to prevent underflow, as val will be 1196 * multiplied by 2). 1197 */ 1198 __sock_set_rcvbuf(sk, max(val, 0)); 1199 break; 1200 1201 case SO_KEEPALIVE: 1202 if (sk->sk_prot->keepalive) 1203 sk->sk_prot->keepalive(sk, valbool); 1204 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool); 1205 break; 1206 1207 case SO_OOBINLINE: 1208 sock_valbool_flag(sk, SOCK_URGINLINE, valbool); 1209 break; 1210 1211 case SO_NO_CHECK: 1212 sk->sk_no_check_tx = valbool; 1213 break; 1214 1215 case SO_PRIORITY: 1216 if ((val >= 0 && val <= 6) || 1217 sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) || 1218 sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) 1219 WRITE_ONCE(sk->sk_priority, val); 1220 else 1221 ret = -EPERM; 1222 break; 1223 1224 case SO_LINGER: 1225 if (optlen < sizeof(ling)) { 1226 ret = -EINVAL; /* 1003.1g */ 1227 break; 1228 } 1229 if (copy_from_sockptr(&ling, optval, sizeof(ling))) { 1230 ret = -EFAULT; 1231 break; 1232 } 1233 if (!ling.l_onoff) 1234 sock_reset_flag(sk, SOCK_LINGER); 1235 else { 1236 #if (BITS_PER_LONG == 32) 1237 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ) 1238 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT; 1239 else 1240 #endif 1241 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ; 1242 sock_set_flag(sk, SOCK_LINGER); 1243 } 1244 break; 1245 1246 case SO_BSDCOMPAT: 1247 break; 1248 1249 case SO_PASSCRED: 1250 if (valbool) 1251 set_bit(SOCK_PASSCRED, &sock->flags); 1252 else 1253 clear_bit(SOCK_PASSCRED, &sock->flags); 1254 break; 1255 1256 case SO_PASSPIDFD: 1257 if (valbool) 1258 set_bit(SOCK_PASSPIDFD, &sock->flags); 1259 else 1260 clear_bit(SOCK_PASSPIDFD, &sock->flags); 1261 break; 1262 1263 case SO_TIMESTAMP_OLD: 1264 case SO_TIMESTAMP_NEW: 1265 case SO_TIMESTAMPNS_OLD: 1266 case SO_TIMESTAMPNS_NEW: 1267 sock_set_timestamp(sk, optname, valbool); 1268 break; 1269 1270 case SO_TIMESTAMPING_NEW: 1271 case SO_TIMESTAMPING_OLD: 1272 if (optlen == sizeof(timestamping)) { 1273 if (copy_from_sockptr(×tamping, optval, 1274 sizeof(timestamping))) { 1275 ret = -EFAULT; 1276 break; 1277 } 1278 } else { 1279 memset(×tamping, 0, sizeof(timestamping)); 1280 timestamping.flags = val; 1281 } 1282 ret = sock_set_timestamping(sk, optname, timestamping); 1283 break; 1284 1285 case SO_RCVLOWAT: 1286 if (val < 0) 1287 val = INT_MAX; 1288 if (sock && sock->ops->set_rcvlowat) 1289 ret = sock->ops->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 if (valbool) 1365 set_bit(SOCK_PASSSEC, &sock->flags); 1366 else 1367 clear_bit(SOCK_PASSSEC, &sock->flags); 1368 break; 1369 case SO_MARK: 1370 if (!sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) && 1371 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) { 1372 ret = -EPERM; 1373 break; 1374 } 1375 1376 __sock_set_mark(sk, val); 1377 break; 1378 case SO_RCVMARK: 1379 sock_valbool_flag(sk, SOCK_RCVMARK, valbool); 1380 break; 1381 1382 case SO_RXQ_OVFL: 1383 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool); 1384 break; 1385 1386 case SO_WIFI_STATUS: 1387 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool); 1388 break; 1389 1390 case SO_PEEK_OFF: 1391 if (sock->ops->set_peek_off) 1392 ret = sock->ops->set_peek_off(sk, val); 1393 else 1394 ret = -EOPNOTSUPP; 1395 break; 1396 1397 case SO_NOFCS: 1398 sock_valbool_flag(sk, SOCK_NOFCS, valbool); 1399 break; 1400 1401 case SO_SELECT_ERR_QUEUE: 1402 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool); 1403 break; 1404 1405 #ifdef CONFIG_NET_RX_BUSY_POLL 1406 case SO_BUSY_POLL: 1407 if (val < 0) 1408 ret = -EINVAL; 1409 else 1410 WRITE_ONCE(sk->sk_ll_usec, val); 1411 break; 1412 case SO_PREFER_BUSY_POLL: 1413 if (valbool && !sockopt_capable(CAP_NET_ADMIN)) 1414 ret = -EPERM; 1415 else 1416 WRITE_ONCE(sk->sk_prefer_busy_poll, valbool); 1417 break; 1418 case SO_BUSY_POLL_BUDGET: 1419 if (val > READ_ONCE(sk->sk_busy_poll_budget) && !sockopt_capable(CAP_NET_ADMIN)) { 1420 ret = -EPERM; 1421 } else { 1422 if (val < 0 || val > U16_MAX) 1423 ret = -EINVAL; 1424 else 1425 WRITE_ONCE(sk->sk_busy_poll_budget, val); 1426 } 1427 break; 1428 #endif 1429 1430 case SO_MAX_PACING_RATE: 1431 { 1432 unsigned long ulval = (val == ~0U) ? ~0UL : (unsigned int)val; 1433 1434 if (sizeof(ulval) != sizeof(val) && 1435 optlen >= sizeof(ulval) && 1436 copy_from_sockptr(&ulval, optval, sizeof(ulval))) { 1437 ret = -EFAULT; 1438 break; 1439 } 1440 if (ulval != ~0UL) 1441 cmpxchg(&sk->sk_pacing_status, 1442 SK_PACING_NONE, 1443 SK_PACING_NEEDED); 1444 /* Pairs with READ_ONCE() from sk_getsockopt() */ 1445 WRITE_ONCE(sk->sk_max_pacing_rate, ulval); 1446 sk->sk_pacing_rate = min(sk->sk_pacing_rate, ulval); 1447 break; 1448 } 1449 case SO_INCOMING_CPU: 1450 reuseport_update_incoming_cpu(sk, val); 1451 break; 1452 1453 case SO_CNX_ADVICE: 1454 if (val == 1) 1455 dst_negative_advice(sk); 1456 break; 1457 1458 case SO_ZEROCOPY: 1459 if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) { 1460 if (!(sk_is_tcp(sk) || 1461 (sk->sk_type == SOCK_DGRAM && 1462 sk->sk_protocol == IPPROTO_UDP))) 1463 ret = -EOPNOTSUPP; 1464 } else if (sk->sk_family != PF_RDS) { 1465 ret = -EOPNOTSUPP; 1466 } 1467 if (!ret) { 1468 if (val < 0 || val > 1) 1469 ret = -EINVAL; 1470 else 1471 sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool); 1472 } 1473 break; 1474 1475 case SO_TXTIME: 1476 if (optlen != sizeof(struct sock_txtime)) { 1477 ret = -EINVAL; 1478 break; 1479 } else if (copy_from_sockptr(&sk_txtime, optval, 1480 sizeof(struct sock_txtime))) { 1481 ret = -EFAULT; 1482 break; 1483 } else if (sk_txtime.flags & ~SOF_TXTIME_FLAGS_MASK) { 1484 ret = -EINVAL; 1485 break; 1486 } 1487 /* CLOCK_MONOTONIC is only used by sch_fq, and this packet 1488 * scheduler has enough safe guards. 1489 */ 1490 if (sk_txtime.clockid != CLOCK_MONOTONIC && 1491 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) { 1492 ret = -EPERM; 1493 break; 1494 } 1495 sock_valbool_flag(sk, SOCK_TXTIME, true); 1496 sk->sk_clockid = sk_txtime.clockid; 1497 sk->sk_txtime_deadline_mode = 1498 !!(sk_txtime.flags & SOF_TXTIME_DEADLINE_MODE); 1499 sk->sk_txtime_report_errors = 1500 !!(sk_txtime.flags & SOF_TXTIME_REPORT_ERRORS); 1501 break; 1502 1503 case SO_BINDTOIFINDEX: 1504 ret = sock_bindtoindex_locked(sk, val); 1505 break; 1506 1507 case SO_BUF_LOCK: 1508 if (val & ~SOCK_BUF_LOCK_MASK) { 1509 ret = -EINVAL; 1510 break; 1511 } 1512 sk->sk_userlocks = val | (sk->sk_userlocks & 1513 ~SOCK_BUF_LOCK_MASK); 1514 break; 1515 1516 case SO_RESERVE_MEM: 1517 { 1518 int delta; 1519 1520 if (val < 0) { 1521 ret = -EINVAL; 1522 break; 1523 } 1524 1525 delta = val - sk->sk_reserved_mem; 1526 if (delta < 0) 1527 sock_release_reserved_memory(sk, -delta); 1528 else 1529 ret = sock_reserve_memory(sk, delta); 1530 break; 1531 } 1532 1533 case SO_TXREHASH: 1534 if (val < -1 || val > 1) { 1535 ret = -EINVAL; 1536 break; 1537 } 1538 if ((u8)val == SOCK_TXREHASH_DEFAULT) 1539 val = READ_ONCE(sock_net(sk)->core.sysctl_txrehash); 1540 /* Paired with READ_ONCE() in tcp_rtx_synack() 1541 * and sk_getsockopt(). 1542 */ 1543 WRITE_ONCE(sk->sk_txrehash, (u8)val); 1544 break; 1545 1546 default: 1547 ret = -ENOPROTOOPT; 1548 break; 1549 } 1550 sockopt_release_sock(sk); 1551 return ret; 1552 } 1553 1554 int sock_setsockopt(struct socket *sock, int level, int optname, 1555 sockptr_t optval, unsigned int optlen) 1556 { 1557 return sk_setsockopt(sock->sk, level, optname, 1558 optval, optlen); 1559 } 1560 EXPORT_SYMBOL(sock_setsockopt); 1561 1562 static const struct cred *sk_get_peer_cred(struct sock *sk) 1563 { 1564 const struct cred *cred; 1565 1566 spin_lock(&sk->sk_peer_lock); 1567 cred = get_cred(sk->sk_peer_cred); 1568 spin_unlock(&sk->sk_peer_lock); 1569 1570 return cred; 1571 } 1572 1573 static void cred_to_ucred(struct pid *pid, const struct cred *cred, 1574 struct ucred *ucred) 1575 { 1576 ucred->pid = pid_vnr(pid); 1577 ucred->uid = ucred->gid = -1; 1578 if (cred) { 1579 struct user_namespace *current_ns = current_user_ns(); 1580 1581 ucred->uid = from_kuid_munged(current_ns, cred->euid); 1582 ucred->gid = from_kgid_munged(current_ns, cred->egid); 1583 } 1584 } 1585 1586 static int groups_to_user(sockptr_t dst, const struct group_info *src) 1587 { 1588 struct user_namespace *user_ns = current_user_ns(); 1589 int i; 1590 1591 for (i = 0; i < src->ngroups; i++) { 1592 gid_t gid = from_kgid_munged(user_ns, src->gid[i]); 1593 1594 if (copy_to_sockptr_offset(dst, i * sizeof(gid), &gid, sizeof(gid))) 1595 return -EFAULT; 1596 } 1597 1598 return 0; 1599 } 1600 1601 int sk_getsockopt(struct sock *sk, int level, int optname, 1602 sockptr_t optval, sockptr_t optlen) 1603 { 1604 struct socket *sock = sk->sk_socket; 1605 1606 union { 1607 int val; 1608 u64 val64; 1609 unsigned long ulval; 1610 struct linger ling; 1611 struct old_timeval32 tm32; 1612 struct __kernel_old_timeval tm; 1613 struct __kernel_sock_timeval stm; 1614 struct sock_txtime txtime; 1615 struct so_timestamping timestamping; 1616 } v; 1617 1618 int lv = sizeof(int); 1619 int len; 1620 1621 if (copy_from_sockptr(&len, optlen, sizeof(int))) 1622 return -EFAULT; 1623 if (len < 0) 1624 return -EINVAL; 1625 1626 memset(&v, 0, sizeof(v)); 1627 1628 switch (optname) { 1629 case SO_DEBUG: 1630 v.val = sock_flag(sk, SOCK_DBG); 1631 break; 1632 1633 case SO_DONTROUTE: 1634 v.val = sock_flag(sk, SOCK_LOCALROUTE); 1635 break; 1636 1637 case SO_BROADCAST: 1638 v.val = sock_flag(sk, SOCK_BROADCAST); 1639 break; 1640 1641 case SO_SNDBUF: 1642 v.val = READ_ONCE(sk->sk_sndbuf); 1643 break; 1644 1645 case SO_RCVBUF: 1646 v.val = READ_ONCE(sk->sk_rcvbuf); 1647 break; 1648 1649 case SO_REUSEADDR: 1650 v.val = sk->sk_reuse; 1651 break; 1652 1653 case SO_REUSEPORT: 1654 v.val = sk->sk_reuseport; 1655 break; 1656 1657 case SO_KEEPALIVE: 1658 v.val = sock_flag(sk, SOCK_KEEPOPEN); 1659 break; 1660 1661 case SO_TYPE: 1662 v.val = sk->sk_type; 1663 break; 1664 1665 case SO_PROTOCOL: 1666 v.val = sk->sk_protocol; 1667 break; 1668 1669 case SO_DOMAIN: 1670 v.val = sk->sk_family; 1671 break; 1672 1673 case SO_ERROR: 1674 v.val = -sock_error(sk); 1675 if (v.val == 0) 1676 v.val = xchg(&sk->sk_err_soft, 0); 1677 break; 1678 1679 case SO_OOBINLINE: 1680 v.val = sock_flag(sk, SOCK_URGINLINE); 1681 break; 1682 1683 case SO_NO_CHECK: 1684 v.val = sk->sk_no_check_tx; 1685 break; 1686 1687 case SO_PRIORITY: 1688 v.val = READ_ONCE(sk->sk_priority); 1689 break; 1690 1691 case SO_LINGER: 1692 lv = sizeof(v.ling); 1693 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER); 1694 v.ling.l_linger = sk->sk_lingertime / HZ; 1695 break; 1696 1697 case SO_BSDCOMPAT: 1698 break; 1699 1700 case SO_TIMESTAMP_OLD: 1701 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && 1702 !sock_flag(sk, SOCK_TSTAMP_NEW) && 1703 !sock_flag(sk, SOCK_RCVTSTAMPNS); 1704 break; 1705 1706 case SO_TIMESTAMPNS_OLD: 1707 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && !sock_flag(sk, SOCK_TSTAMP_NEW); 1708 break; 1709 1710 case SO_TIMESTAMP_NEW: 1711 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && sock_flag(sk, SOCK_TSTAMP_NEW); 1712 break; 1713 1714 case SO_TIMESTAMPNS_NEW: 1715 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && sock_flag(sk, SOCK_TSTAMP_NEW); 1716 break; 1717 1718 case SO_TIMESTAMPING_OLD: 1719 lv = sizeof(v.timestamping); 1720 v.timestamping.flags = sk->sk_tsflags; 1721 v.timestamping.bind_phc = sk->sk_bind_phc; 1722 break; 1723 1724 case SO_RCVTIMEO_OLD: 1725 case SO_RCVTIMEO_NEW: 1726 lv = sock_get_timeout(READ_ONCE(sk->sk_rcvtimeo), &v, 1727 SO_RCVTIMEO_OLD == optname); 1728 break; 1729 1730 case SO_SNDTIMEO_OLD: 1731 case SO_SNDTIMEO_NEW: 1732 lv = sock_get_timeout(READ_ONCE(sk->sk_sndtimeo), &v, 1733 SO_SNDTIMEO_OLD == optname); 1734 break; 1735 1736 case SO_RCVLOWAT: 1737 v.val = READ_ONCE(sk->sk_rcvlowat); 1738 break; 1739 1740 case SO_SNDLOWAT: 1741 v.val = 1; 1742 break; 1743 1744 case SO_PASSCRED: 1745 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags); 1746 break; 1747 1748 case SO_PASSPIDFD: 1749 v.val = !!test_bit(SOCK_PASSPIDFD, &sock->flags); 1750 break; 1751 1752 case SO_PEERCRED: 1753 { 1754 struct ucred peercred; 1755 if (len > sizeof(peercred)) 1756 len = sizeof(peercred); 1757 1758 spin_lock(&sk->sk_peer_lock); 1759 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred); 1760 spin_unlock(&sk->sk_peer_lock); 1761 1762 if (copy_to_sockptr(optval, &peercred, len)) 1763 return -EFAULT; 1764 goto lenout; 1765 } 1766 1767 case SO_PEERPIDFD: 1768 { 1769 struct pid *peer_pid; 1770 struct file *pidfd_file = NULL; 1771 int pidfd; 1772 1773 if (len > sizeof(pidfd)) 1774 len = sizeof(pidfd); 1775 1776 spin_lock(&sk->sk_peer_lock); 1777 peer_pid = get_pid(sk->sk_peer_pid); 1778 spin_unlock(&sk->sk_peer_lock); 1779 1780 if (!peer_pid) 1781 return -ESRCH; 1782 1783 pidfd = pidfd_prepare(peer_pid, 0, &pidfd_file); 1784 put_pid(peer_pid); 1785 if (pidfd < 0) 1786 return pidfd; 1787 1788 if (copy_to_sockptr(optval, &pidfd, len) || 1789 copy_to_sockptr(optlen, &len, sizeof(int))) { 1790 put_unused_fd(pidfd); 1791 fput(pidfd_file); 1792 1793 return -EFAULT; 1794 } 1795 1796 fd_install(pidfd, pidfd_file); 1797 return 0; 1798 } 1799 1800 case SO_PEERGROUPS: 1801 { 1802 const struct cred *cred; 1803 int ret, n; 1804 1805 cred = sk_get_peer_cred(sk); 1806 if (!cred) 1807 return -ENODATA; 1808 1809 n = cred->group_info->ngroups; 1810 if (len < n * sizeof(gid_t)) { 1811 len = n * sizeof(gid_t); 1812 put_cred(cred); 1813 return copy_to_sockptr(optlen, &len, sizeof(int)) ? -EFAULT : -ERANGE; 1814 } 1815 len = n * sizeof(gid_t); 1816 1817 ret = groups_to_user(optval, cred->group_info); 1818 put_cred(cred); 1819 if (ret) 1820 return ret; 1821 goto lenout; 1822 } 1823 1824 case SO_PEERNAME: 1825 { 1826 char address[128]; 1827 1828 lv = sock->ops->getname(sock, (struct sockaddr *)address, 2); 1829 if (lv < 0) 1830 return -ENOTCONN; 1831 if (lv < len) 1832 return -EINVAL; 1833 if (copy_to_sockptr(optval, address, len)) 1834 return -EFAULT; 1835 goto lenout; 1836 } 1837 1838 /* Dubious BSD thing... Probably nobody even uses it, but 1839 * the UNIX standard wants it for whatever reason... -DaveM 1840 */ 1841 case SO_ACCEPTCONN: 1842 v.val = sk->sk_state == TCP_LISTEN; 1843 break; 1844 1845 case SO_PASSSEC: 1846 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags); 1847 break; 1848 1849 case SO_PEERSEC: 1850 return security_socket_getpeersec_stream(sock, 1851 optval, optlen, len); 1852 1853 case SO_MARK: 1854 v.val = READ_ONCE(sk->sk_mark); 1855 break; 1856 1857 case SO_RCVMARK: 1858 v.val = sock_flag(sk, SOCK_RCVMARK); 1859 break; 1860 1861 case SO_RXQ_OVFL: 1862 v.val = sock_flag(sk, SOCK_RXQ_OVFL); 1863 break; 1864 1865 case SO_WIFI_STATUS: 1866 v.val = sock_flag(sk, SOCK_WIFI_STATUS); 1867 break; 1868 1869 case SO_PEEK_OFF: 1870 if (!sock->ops->set_peek_off) 1871 return -EOPNOTSUPP; 1872 1873 v.val = READ_ONCE(sk->sk_peek_off); 1874 break; 1875 case SO_NOFCS: 1876 v.val = sock_flag(sk, SOCK_NOFCS); 1877 break; 1878 1879 case SO_BINDTODEVICE: 1880 return sock_getbindtodevice(sk, optval, optlen, len); 1881 1882 case SO_GET_FILTER: 1883 len = sk_get_filter(sk, optval, len); 1884 if (len < 0) 1885 return len; 1886 1887 goto lenout; 1888 1889 case SO_LOCK_FILTER: 1890 v.val = sock_flag(sk, SOCK_FILTER_LOCKED); 1891 break; 1892 1893 case SO_BPF_EXTENSIONS: 1894 v.val = bpf_tell_extensions(); 1895 break; 1896 1897 case SO_SELECT_ERR_QUEUE: 1898 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE); 1899 break; 1900 1901 #ifdef CONFIG_NET_RX_BUSY_POLL 1902 case SO_BUSY_POLL: 1903 v.val = READ_ONCE(sk->sk_ll_usec); 1904 break; 1905 case SO_PREFER_BUSY_POLL: 1906 v.val = READ_ONCE(sk->sk_prefer_busy_poll); 1907 break; 1908 #endif 1909 1910 case SO_MAX_PACING_RATE: 1911 /* The READ_ONCE() pair with the WRITE_ONCE() in sk_setsockopt() */ 1912 if (sizeof(v.ulval) != sizeof(v.val) && len >= sizeof(v.ulval)) { 1913 lv = sizeof(v.ulval); 1914 v.ulval = READ_ONCE(sk->sk_max_pacing_rate); 1915 } else { 1916 /* 32bit version */ 1917 v.val = min_t(unsigned long, ~0U, 1918 READ_ONCE(sk->sk_max_pacing_rate)); 1919 } 1920 break; 1921 1922 case SO_INCOMING_CPU: 1923 v.val = READ_ONCE(sk->sk_incoming_cpu); 1924 break; 1925 1926 case SO_MEMINFO: 1927 { 1928 u32 meminfo[SK_MEMINFO_VARS]; 1929 1930 sk_get_meminfo(sk, meminfo); 1931 1932 len = min_t(unsigned int, len, sizeof(meminfo)); 1933 if (copy_to_sockptr(optval, &meminfo, len)) 1934 return -EFAULT; 1935 1936 goto lenout; 1937 } 1938 1939 #ifdef CONFIG_NET_RX_BUSY_POLL 1940 case SO_INCOMING_NAPI_ID: 1941 v.val = READ_ONCE(sk->sk_napi_id); 1942 1943 /* aggregate non-NAPI IDs down to 0 */ 1944 if (v.val < MIN_NAPI_ID) 1945 v.val = 0; 1946 1947 break; 1948 #endif 1949 1950 case SO_COOKIE: 1951 lv = sizeof(u64); 1952 if (len < lv) 1953 return -EINVAL; 1954 v.val64 = sock_gen_cookie(sk); 1955 break; 1956 1957 case SO_ZEROCOPY: 1958 v.val = sock_flag(sk, SOCK_ZEROCOPY); 1959 break; 1960 1961 case SO_TXTIME: 1962 lv = sizeof(v.txtime); 1963 v.txtime.clockid = sk->sk_clockid; 1964 v.txtime.flags |= sk->sk_txtime_deadline_mode ? 1965 SOF_TXTIME_DEADLINE_MODE : 0; 1966 v.txtime.flags |= sk->sk_txtime_report_errors ? 1967 SOF_TXTIME_REPORT_ERRORS : 0; 1968 break; 1969 1970 case SO_BINDTOIFINDEX: 1971 v.val = READ_ONCE(sk->sk_bound_dev_if); 1972 break; 1973 1974 case SO_NETNS_COOKIE: 1975 lv = sizeof(u64); 1976 if (len != lv) 1977 return -EINVAL; 1978 v.val64 = sock_net(sk)->net_cookie; 1979 break; 1980 1981 case SO_BUF_LOCK: 1982 v.val = sk->sk_userlocks & SOCK_BUF_LOCK_MASK; 1983 break; 1984 1985 case SO_RESERVE_MEM: 1986 v.val = READ_ONCE(sk->sk_reserved_mem); 1987 break; 1988 1989 case SO_TXREHASH: 1990 /* Paired with WRITE_ONCE() in sk_setsockopt() */ 1991 v.val = READ_ONCE(sk->sk_txrehash); 1992 break; 1993 1994 default: 1995 /* We implement the SO_SNDLOWAT etc to not be settable 1996 * (1003.1g 7). 1997 */ 1998 return -ENOPROTOOPT; 1999 } 2000 2001 if (len > lv) 2002 len = lv; 2003 if (copy_to_sockptr(optval, &v, len)) 2004 return -EFAULT; 2005 lenout: 2006 if (copy_to_sockptr(optlen, &len, sizeof(int))) 2007 return -EFAULT; 2008 return 0; 2009 } 2010 2011 int sock_getsockopt(struct socket *sock, int level, int optname, 2012 char __user *optval, int __user *optlen) 2013 { 2014 return sk_getsockopt(sock->sk, level, optname, 2015 USER_SOCKPTR(optval), 2016 USER_SOCKPTR(optlen)); 2017 } 2018 2019 /* 2020 * Initialize an sk_lock. 2021 * 2022 * (We also register the sk_lock with the lock validator.) 2023 */ 2024 static inline void sock_lock_init(struct sock *sk) 2025 { 2026 if (sk->sk_kern_sock) 2027 sock_lock_init_class_and_name( 2028 sk, 2029 af_family_kern_slock_key_strings[sk->sk_family], 2030 af_family_kern_slock_keys + sk->sk_family, 2031 af_family_kern_key_strings[sk->sk_family], 2032 af_family_kern_keys + sk->sk_family); 2033 else 2034 sock_lock_init_class_and_name( 2035 sk, 2036 af_family_slock_key_strings[sk->sk_family], 2037 af_family_slock_keys + sk->sk_family, 2038 af_family_key_strings[sk->sk_family], 2039 af_family_keys + sk->sk_family); 2040 } 2041 2042 /* 2043 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet, 2044 * even temporarly, because of RCU lookups. sk_node should also be left as is. 2045 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end 2046 */ 2047 static void sock_copy(struct sock *nsk, const struct sock *osk) 2048 { 2049 const struct proto *prot = READ_ONCE(osk->sk_prot); 2050 #ifdef CONFIG_SECURITY_NETWORK 2051 void *sptr = nsk->sk_security; 2052 #endif 2053 2054 /* If we move sk_tx_queue_mapping out of the private section, 2055 * we must check if sk_tx_queue_clear() is called after 2056 * sock_copy() in sk_clone_lock(). 2057 */ 2058 BUILD_BUG_ON(offsetof(struct sock, sk_tx_queue_mapping) < 2059 offsetof(struct sock, sk_dontcopy_begin) || 2060 offsetof(struct sock, sk_tx_queue_mapping) >= 2061 offsetof(struct sock, sk_dontcopy_end)); 2062 2063 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin)); 2064 2065 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end, 2066 prot->obj_size - offsetof(struct sock, sk_dontcopy_end)); 2067 2068 #ifdef CONFIG_SECURITY_NETWORK 2069 nsk->sk_security = sptr; 2070 security_sk_clone(osk, nsk); 2071 #endif 2072 } 2073 2074 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority, 2075 int family) 2076 { 2077 struct sock *sk; 2078 struct kmem_cache *slab; 2079 2080 slab = prot->slab; 2081 if (slab != NULL) { 2082 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO); 2083 if (!sk) 2084 return sk; 2085 if (want_init_on_alloc(priority)) 2086 sk_prot_clear_nulls(sk, prot->obj_size); 2087 } else 2088 sk = kmalloc(prot->obj_size, priority); 2089 2090 if (sk != NULL) { 2091 if (security_sk_alloc(sk, family, priority)) 2092 goto out_free; 2093 2094 if (!try_module_get(prot->owner)) 2095 goto out_free_sec; 2096 } 2097 2098 return sk; 2099 2100 out_free_sec: 2101 security_sk_free(sk); 2102 out_free: 2103 if (slab != NULL) 2104 kmem_cache_free(slab, sk); 2105 else 2106 kfree(sk); 2107 return NULL; 2108 } 2109 2110 static void sk_prot_free(struct proto *prot, struct sock *sk) 2111 { 2112 struct kmem_cache *slab; 2113 struct module *owner; 2114 2115 owner = prot->owner; 2116 slab = prot->slab; 2117 2118 cgroup_sk_free(&sk->sk_cgrp_data); 2119 mem_cgroup_sk_free(sk); 2120 security_sk_free(sk); 2121 if (slab != NULL) 2122 kmem_cache_free(slab, sk); 2123 else 2124 kfree(sk); 2125 module_put(owner); 2126 } 2127 2128 /** 2129 * sk_alloc - All socket objects are allocated here 2130 * @net: the applicable net namespace 2131 * @family: protocol family 2132 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 2133 * @prot: struct proto associated with this new sock instance 2134 * @kern: is this to be a kernel socket? 2135 */ 2136 struct sock *sk_alloc(struct net *net, int family, gfp_t priority, 2137 struct proto *prot, int kern) 2138 { 2139 struct sock *sk; 2140 2141 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family); 2142 if (sk) { 2143 sk->sk_family = family; 2144 /* 2145 * See comment in struct sock definition to understand 2146 * why we need sk_prot_creator -acme 2147 */ 2148 sk->sk_prot = sk->sk_prot_creator = prot; 2149 sk->sk_kern_sock = kern; 2150 sock_lock_init(sk); 2151 sk->sk_net_refcnt = kern ? 0 : 1; 2152 if (likely(sk->sk_net_refcnt)) { 2153 get_net_track(net, &sk->ns_tracker, priority); 2154 sock_inuse_add(net, 1); 2155 } else { 2156 __netns_tracker_alloc(net, &sk->ns_tracker, 2157 false, priority); 2158 } 2159 2160 sock_net_set(sk, net); 2161 refcount_set(&sk->sk_wmem_alloc, 1); 2162 2163 mem_cgroup_sk_alloc(sk); 2164 cgroup_sk_alloc(&sk->sk_cgrp_data); 2165 sock_update_classid(&sk->sk_cgrp_data); 2166 sock_update_netprioidx(&sk->sk_cgrp_data); 2167 sk_tx_queue_clear(sk); 2168 } 2169 2170 return sk; 2171 } 2172 EXPORT_SYMBOL(sk_alloc); 2173 2174 /* Sockets having SOCK_RCU_FREE will call this function after one RCU 2175 * grace period. This is the case for UDP sockets and TCP listeners. 2176 */ 2177 static void __sk_destruct(struct rcu_head *head) 2178 { 2179 struct sock *sk = container_of(head, struct sock, sk_rcu); 2180 struct sk_filter *filter; 2181 2182 if (sk->sk_destruct) 2183 sk->sk_destruct(sk); 2184 2185 filter = rcu_dereference_check(sk->sk_filter, 2186 refcount_read(&sk->sk_wmem_alloc) == 0); 2187 if (filter) { 2188 sk_filter_uncharge(sk, filter); 2189 RCU_INIT_POINTER(sk->sk_filter, NULL); 2190 } 2191 2192 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP); 2193 2194 #ifdef CONFIG_BPF_SYSCALL 2195 bpf_sk_storage_free(sk); 2196 #endif 2197 2198 if (atomic_read(&sk->sk_omem_alloc)) 2199 pr_debug("%s: optmem leakage (%d bytes) detected\n", 2200 __func__, atomic_read(&sk->sk_omem_alloc)); 2201 2202 if (sk->sk_frag.page) { 2203 put_page(sk->sk_frag.page); 2204 sk->sk_frag.page = NULL; 2205 } 2206 2207 /* We do not need to acquire sk->sk_peer_lock, we are the last user. */ 2208 put_cred(sk->sk_peer_cred); 2209 put_pid(sk->sk_peer_pid); 2210 2211 if (likely(sk->sk_net_refcnt)) 2212 put_net_track(sock_net(sk), &sk->ns_tracker); 2213 else 2214 __netns_tracker_free(sock_net(sk), &sk->ns_tracker, false); 2215 2216 sk_prot_free(sk->sk_prot_creator, sk); 2217 } 2218 2219 void sk_destruct(struct sock *sk) 2220 { 2221 bool use_call_rcu = sock_flag(sk, SOCK_RCU_FREE); 2222 2223 if (rcu_access_pointer(sk->sk_reuseport_cb)) { 2224 reuseport_detach_sock(sk); 2225 use_call_rcu = true; 2226 } 2227 2228 if (use_call_rcu) 2229 call_rcu(&sk->sk_rcu, __sk_destruct); 2230 else 2231 __sk_destruct(&sk->sk_rcu); 2232 } 2233 2234 static void __sk_free(struct sock *sk) 2235 { 2236 if (likely(sk->sk_net_refcnt)) 2237 sock_inuse_add(sock_net(sk), -1); 2238 2239 if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk))) 2240 sock_diag_broadcast_destroy(sk); 2241 else 2242 sk_destruct(sk); 2243 } 2244 2245 void sk_free(struct sock *sk) 2246 { 2247 /* 2248 * We subtract one from sk_wmem_alloc and can know if 2249 * some packets are still in some tx queue. 2250 * If not null, sock_wfree() will call __sk_free(sk) later 2251 */ 2252 if (refcount_dec_and_test(&sk->sk_wmem_alloc)) 2253 __sk_free(sk); 2254 } 2255 EXPORT_SYMBOL(sk_free); 2256 2257 static void sk_init_common(struct sock *sk) 2258 { 2259 skb_queue_head_init(&sk->sk_receive_queue); 2260 skb_queue_head_init(&sk->sk_write_queue); 2261 skb_queue_head_init(&sk->sk_error_queue); 2262 2263 rwlock_init(&sk->sk_callback_lock); 2264 lockdep_set_class_and_name(&sk->sk_receive_queue.lock, 2265 af_rlock_keys + sk->sk_family, 2266 af_family_rlock_key_strings[sk->sk_family]); 2267 lockdep_set_class_and_name(&sk->sk_write_queue.lock, 2268 af_wlock_keys + sk->sk_family, 2269 af_family_wlock_key_strings[sk->sk_family]); 2270 lockdep_set_class_and_name(&sk->sk_error_queue.lock, 2271 af_elock_keys + sk->sk_family, 2272 af_family_elock_key_strings[sk->sk_family]); 2273 lockdep_set_class_and_name(&sk->sk_callback_lock, 2274 af_callback_keys + sk->sk_family, 2275 af_family_clock_key_strings[sk->sk_family]); 2276 } 2277 2278 /** 2279 * sk_clone_lock - clone a socket, and lock its clone 2280 * @sk: the socket to clone 2281 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 2282 * 2283 * Caller must unlock socket even in error path (bh_unlock_sock(newsk)) 2284 */ 2285 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority) 2286 { 2287 struct proto *prot = READ_ONCE(sk->sk_prot); 2288 struct sk_filter *filter; 2289 bool is_charged = true; 2290 struct sock *newsk; 2291 2292 newsk = sk_prot_alloc(prot, priority, sk->sk_family); 2293 if (!newsk) 2294 goto out; 2295 2296 sock_copy(newsk, sk); 2297 2298 newsk->sk_prot_creator = prot; 2299 2300 /* SANITY */ 2301 if (likely(newsk->sk_net_refcnt)) { 2302 get_net_track(sock_net(newsk), &newsk->ns_tracker, priority); 2303 sock_inuse_add(sock_net(newsk), 1); 2304 } else { 2305 /* Kernel sockets are not elevating the struct net refcount. 2306 * Instead, use a tracker to more easily detect if a layer 2307 * is not properly dismantling its kernel sockets at netns 2308 * destroy time. 2309 */ 2310 __netns_tracker_alloc(sock_net(newsk), &newsk->ns_tracker, 2311 false, priority); 2312 } 2313 sk_node_init(&newsk->sk_node); 2314 sock_lock_init(newsk); 2315 bh_lock_sock(newsk); 2316 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL; 2317 newsk->sk_backlog.len = 0; 2318 2319 atomic_set(&newsk->sk_rmem_alloc, 0); 2320 2321 /* sk_wmem_alloc set to one (see sk_free() and sock_wfree()) */ 2322 refcount_set(&newsk->sk_wmem_alloc, 1); 2323 2324 atomic_set(&newsk->sk_omem_alloc, 0); 2325 sk_init_common(newsk); 2326 2327 newsk->sk_dst_cache = NULL; 2328 newsk->sk_dst_pending_confirm = 0; 2329 newsk->sk_wmem_queued = 0; 2330 newsk->sk_forward_alloc = 0; 2331 newsk->sk_reserved_mem = 0; 2332 atomic_set(&newsk->sk_drops, 0); 2333 newsk->sk_send_head = NULL; 2334 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK; 2335 atomic_set(&newsk->sk_zckey, 0); 2336 2337 sock_reset_flag(newsk, SOCK_DONE); 2338 2339 /* sk->sk_memcg will be populated at accept() time */ 2340 newsk->sk_memcg = NULL; 2341 2342 cgroup_sk_clone(&newsk->sk_cgrp_data); 2343 2344 rcu_read_lock(); 2345 filter = rcu_dereference(sk->sk_filter); 2346 if (filter != NULL) 2347 /* though it's an empty new sock, the charging may fail 2348 * if sysctl_optmem_max was changed between creation of 2349 * original socket and cloning 2350 */ 2351 is_charged = sk_filter_charge(newsk, filter); 2352 RCU_INIT_POINTER(newsk->sk_filter, filter); 2353 rcu_read_unlock(); 2354 2355 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) { 2356 /* We need to make sure that we don't uncharge the new 2357 * socket if we couldn't charge it in the first place 2358 * as otherwise we uncharge the parent's filter. 2359 */ 2360 if (!is_charged) 2361 RCU_INIT_POINTER(newsk->sk_filter, NULL); 2362 sk_free_unlock_clone(newsk); 2363 newsk = NULL; 2364 goto out; 2365 } 2366 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL); 2367 2368 if (bpf_sk_storage_clone(sk, newsk)) { 2369 sk_free_unlock_clone(newsk); 2370 newsk = NULL; 2371 goto out; 2372 } 2373 2374 /* Clear sk_user_data if parent had the pointer tagged 2375 * as not suitable for copying when cloning. 2376 */ 2377 if (sk_user_data_is_nocopy(newsk)) 2378 newsk->sk_user_data = NULL; 2379 2380 newsk->sk_err = 0; 2381 newsk->sk_err_soft = 0; 2382 newsk->sk_priority = 0; 2383 newsk->sk_incoming_cpu = raw_smp_processor_id(); 2384 2385 /* Before updating sk_refcnt, we must commit prior changes to memory 2386 * (Documentation/RCU/rculist_nulls.rst for details) 2387 */ 2388 smp_wmb(); 2389 refcount_set(&newsk->sk_refcnt, 2); 2390 2391 sk_set_socket(newsk, NULL); 2392 sk_tx_queue_clear(newsk); 2393 RCU_INIT_POINTER(newsk->sk_wq, NULL); 2394 2395 if (newsk->sk_prot->sockets_allocated) 2396 sk_sockets_allocated_inc(newsk); 2397 2398 if (sock_needs_netstamp(sk) && newsk->sk_flags & SK_FLAGS_TIMESTAMP) 2399 net_enable_timestamp(); 2400 out: 2401 return newsk; 2402 } 2403 EXPORT_SYMBOL_GPL(sk_clone_lock); 2404 2405 void sk_free_unlock_clone(struct sock *sk) 2406 { 2407 /* It is still raw copy of parent, so invalidate 2408 * destructor and make plain sk_free() */ 2409 sk->sk_destruct = NULL; 2410 bh_unlock_sock(sk); 2411 sk_free(sk); 2412 } 2413 EXPORT_SYMBOL_GPL(sk_free_unlock_clone); 2414 2415 static u32 sk_dst_gso_max_size(struct sock *sk, struct dst_entry *dst) 2416 { 2417 bool is_ipv6 = false; 2418 u32 max_size; 2419 2420 #if IS_ENABLED(CONFIG_IPV6) 2421 is_ipv6 = (sk->sk_family == AF_INET6 && 2422 !ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr)); 2423 #endif 2424 /* pairs with the WRITE_ONCE() in netif_set_gso(_ipv4)_max_size() */ 2425 max_size = is_ipv6 ? READ_ONCE(dst->dev->gso_max_size) : 2426 READ_ONCE(dst->dev->gso_ipv4_max_size); 2427 if (max_size > GSO_LEGACY_MAX_SIZE && !sk_is_tcp(sk)) 2428 max_size = GSO_LEGACY_MAX_SIZE; 2429 2430 return max_size - (MAX_TCP_HEADER + 1); 2431 } 2432 2433 void sk_setup_caps(struct sock *sk, struct dst_entry *dst) 2434 { 2435 u32 max_segs = 1; 2436 2437 sk->sk_route_caps = dst->dev->features; 2438 if (sk_is_tcp(sk)) 2439 sk->sk_route_caps |= NETIF_F_GSO; 2440 if (sk->sk_route_caps & NETIF_F_GSO) 2441 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE; 2442 if (unlikely(sk->sk_gso_disabled)) 2443 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 2444 if (sk_can_gso(sk)) { 2445 if (dst->header_len && !xfrm_dst_offload_ok(dst)) { 2446 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 2447 } else { 2448 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM; 2449 sk->sk_gso_max_size = sk_dst_gso_max_size(sk, dst); 2450 /* pairs with the WRITE_ONCE() in netif_set_gso_max_segs() */ 2451 max_segs = max_t(u32, READ_ONCE(dst->dev->gso_max_segs), 1); 2452 } 2453 } 2454 sk->sk_gso_max_segs = max_segs; 2455 sk_dst_set(sk, dst); 2456 } 2457 EXPORT_SYMBOL_GPL(sk_setup_caps); 2458 2459 /* 2460 * Simple resource managers for sockets. 2461 */ 2462 2463 2464 /* 2465 * Write buffer destructor automatically called from kfree_skb. 2466 */ 2467 void sock_wfree(struct sk_buff *skb) 2468 { 2469 struct sock *sk = skb->sk; 2470 unsigned int len = skb->truesize; 2471 bool free; 2472 2473 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) { 2474 if (sock_flag(sk, SOCK_RCU_FREE) && 2475 sk->sk_write_space == sock_def_write_space) { 2476 rcu_read_lock(); 2477 free = refcount_sub_and_test(len, &sk->sk_wmem_alloc); 2478 sock_def_write_space_wfree(sk); 2479 rcu_read_unlock(); 2480 if (unlikely(free)) 2481 __sk_free(sk); 2482 return; 2483 } 2484 2485 /* 2486 * Keep a reference on sk_wmem_alloc, this will be released 2487 * after sk_write_space() call 2488 */ 2489 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc)); 2490 sk->sk_write_space(sk); 2491 len = 1; 2492 } 2493 /* 2494 * if sk_wmem_alloc reaches 0, we must finish what sk_free() 2495 * could not do because of in-flight packets 2496 */ 2497 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc)) 2498 __sk_free(sk); 2499 } 2500 EXPORT_SYMBOL(sock_wfree); 2501 2502 /* This variant of sock_wfree() is used by TCP, 2503 * since it sets SOCK_USE_WRITE_QUEUE. 2504 */ 2505 void __sock_wfree(struct sk_buff *skb) 2506 { 2507 struct sock *sk = skb->sk; 2508 2509 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc)) 2510 __sk_free(sk); 2511 } 2512 2513 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk) 2514 { 2515 skb_orphan(skb); 2516 skb->sk = sk; 2517 #ifdef CONFIG_INET 2518 if (unlikely(!sk_fullsock(sk))) { 2519 skb->destructor = sock_edemux; 2520 sock_hold(sk); 2521 return; 2522 } 2523 #endif 2524 skb->destructor = sock_wfree; 2525 skb_set_hash_from_sk(skb, sk); 2526 /* 2527 * We used to take a refcount on sk, but following operation 2528 * is enough to guarantee sk_free() wont free this sock until 2529 * all in-flight packets are completed 2530 */ 2531 refcount_add(skb->truesize, &sk->sk_wmem_alloc); 2532 } 2533 EXPORT_SYMBOL(skb_set_owner_w); 2534 2535 static bool can_skb_orphan_partial(const struct sk_buff *skb) 2536 { 2537 #ifdef CONFIG_TLS_DEVICE 2538 /* Drivers depend on in-order delivery for crypto offload, 2539 * partial orphan breaks out-of-order-OK logic. 2540 */ 2541 if (skb->decrypted) 2542 return false; 2543 #endif 2544 return (skb->destructor == sock_wfree || 2545 (IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree)); 2546 } 2547 2548 /* This helper is used by netem, as it can hold packets in its 2549 * delay queue. We want to allow the owner socket to send more 2550 * packets, as if they were already TX completed by a typical driver. 2551 * But we also want to keep skb->sk set because some packet schedulers 2552 * rely on it (sch_fq for example). 2553 */ 2554 void skb_orphan_partial(struct sk_buff *skb) 2555 { 2556 if (skb_is_tcp_pure_ack(skb)) 2557 return; 2558 2559 if (can_skb_orphan_partial(skb) && skb_set_owner_sk_safe(skb, skb->sk)) 2560 return; 2561 2562 skb_orphan(skb); 2563 } 2564 EXPORT_SYMBOL(skb_orphan_partial); 2565 2566 /* 2567 * Read buffer destructor automatically called from kfree_skb. 2568 */ 2569 void sock_rfree(struct sk_buff *skb) 2570 { 2571 struct sock *sk = skb->sk; 2572 unsigned int len = skb->truesize; 2573 2574 atomic_sub(len, &sk->sk_rmem_alloc); 2575 sk_mem_uncharge(sk, len); 2576 } 2577 EXPORT_SYMBOL(sock_rfree); 2578 2579 /* 2580 * Buffer destructor for skbs that are not used directly in read or write 2581 * path, e.g. for error handler skbs. Automatically called from kfree_skb. 2582 */ 2583 void sock_efree(struct sk_buff *skb) 2584 { 2585 sock_put(skb->sk); 2586 } 2587 EXPORT_SYMBOL(sock_efree); 2588 2589 /* Buffer destructor for prefetch/receive path where reference count may 2590 * not be held, e.g. for listen sockets. 2591 */ 2592 #ifdef CONFIG_INET 2593 void sock_pfree(struct sk_buff *skb) 2594 { 2595 if (sk_is_refcounted(skb->sk)) 2596 sock_gen_put(skb->sk); 2597 } 2598 EXPORT_SYMBOL(sock_pfree); 2599 #endif /* CONFIG_INET */ 2600 2601 kuid_t sock_i_uid(struct sock *sk) 2602 { 2603 kuid_t uid; 2604 2605 read_lock_bh(&sk->sk_callback_lock); 2606 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID; 2607 read_unlock_bh(&sk->sk_callback_lock); 2608 return uid; 2609 } 2610 EXPORT_SYMBOL(sock_i_uid); 2611 2612 unsigned long __sock_i_ino(struct sock *sk) 2613 { 2614 unsigned long ino; 2615 2616 read_lock(&sk->sk_callback_lock); 2617 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0; 2618 read_unlock(&sk->sk_callback_lock); 2619 return ino; 2620 } 2621 EXPORT_SYMBOL(__sock_i_ino); 2622 2623 unsigned long sock_i_ino(struct sock *sk) 2624 { 2625 unsigned long ino; 2626 2627 local_bh_disable(); 2628 ino = __sock_i_ino(sk); 2629 local_bh_enable(); 2630 return ino; 2631 } 2632 EXPORT_SYMBOL(sock_i_ino); 2633 2634 /* 2635 * Allocate a skb from the socket's send buffer. 2636 */ 2637 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 2638 gfp_t priority) 2639 { 2640 if (force || 2641 refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) { 2642 struct sk_buff *skb = alloc_skb(size, priority); 2643 2644 if (skb) { 2645 skb_set_owner_w(skb, sk); 2646 return skb; 2647 } 2648 } 2649 return NULL; 2650 } 2651 EXPORT_SYMBOL(sock_wmalloc); 2652 2653 static void sock_ofree(struct sk_buff *skb) 2654 { 2655 struct sock *sk = skb->sk; 2656 2657 atomic_sub(skb->truesize, &sk->sk_omem_alloc); 2658 } 2659 2660 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size, 2661 gfp_t priority) 2662 { 2663 struct sk_buff *skb; 2664 2665 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */ 2666 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) > 2667 READ_ONCE(sysctl_optmem_max)) 2668 return NULL; 2669 2670 skb = alloc_skb(size, priority); 2671 if (!skb) 2672 return NULL; 2673 2674 atomic_add(skb->truesize, &sk->sk_omem_alloc); 2675 skb->sk = sk; 2676 skb->destructor = sock_ofree; 2677 return skb; 2678 } 2679 2680 /* 2681 * Allocate a memory block from the socket's option memory buffer. 2682 */ 2683 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority) 2684 { 2685 int optmem_max = READ_ONCE(sysctl_optmem_max); 2686 2687 if ((unsigned int)size <= optmem_max && 2688 atomic_read(&sk->sk_omem_alloc) + size < optmem_max) { 2689 void *mem; 2690 /* First do the add, to avoid the race if kmalloc 2691 * might sleep. 2692 */ 2693 atomic_add(size, &sk->sk_omem_alloc); 2694 mem = kmalloc(size, priority); 2695 if (mem) 2696 return mem; 2697 atomic_sub(size, &sk->sk_omem_alloc); 2698 } 2699 return NULL; 2700 } 2701 EXPORT_SYMBOL(sock_kmalloc); 2702 2703 /* Free an option memory block. Note, we actually want the inline 2704 * here as this allows gcc to detect the nullify and fold away the 2705 * condition entirely. 2706 */ 2707 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size, 2708 const bool nullify) 2709 { 2710 if (WARN_ON_ONCE(!mem)) 2711 return; 2712 if (nullify) 2713 kfree_sensitive(mem); 2714 else 2715 kfree(mem); 2716 atomic_sub(size, &sk->sk_omem_alloc); 2717 } 2718 2719 void sock_kfree_s(struct sock *sk, void *mem, int size) 2720 { 2721 __sock_kfree_s(sk, mem, size, false); 2722 } 2723 EXPORT_SYMBOL(sock_kfree_s); 2724 2725 void sock_kzfree_s(struct sock *sk, void *mem, int size) 2726 { 2727 __sock_kfree_s(sk, mem, size, true); 2728 } 2729 EXPORT_SYMBOL(sock_kzfree_s); 2730 2731 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock. 2732 I think, these locks should be removed for datagram sockets. 2733 */ 2734 static long sock_wait_for_wmem(struct sock *sk, long timeo) 2735 { 2736 DEFINE_WAIT(wait); 2737 2738 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk); 2739 for (;;) { 2740 if (!timeo) 2741 break; 2742 if (signal_pending(current)) 2743 break; 2744 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 2745 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 2746 if (refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) 2747 break; 2748 if (sk->sk_shutdown & SEND_SHUTDOWN) 2749 break; 2750 if (sk->sk_err) 2751 break; 2752 timeo = schedule_timeout(timeo); 2753 } 2754 finish_wait(sk_sleep(sk), &wait); 2755 return timeo; 2756 } 2757 2758 2759 /* 2760 * Generic send/receive buffer handlers 2761 */ 2762 2763 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 2764 unsigned long data_len, int noblock, 2765 int *errcode, int max_page_order) 2766 { 2767 struct sk_buff *skb; 2768 long timeo; 2769 int err; 2770 2771 timeo = sock_sndtimeo(sk, noblock); 2772 for (;;) { 2773 err = sock_error(sk); 2774 if (err != 0) 2775 goto failure; 2776 2777 err = -EPIPE; 2778 if (sk->sk_shutdown & SEND_SHUTDOWN) 2779 goto failure; 2780 2781 if (sk_wmem_alloc_get(sk) < READ_ONCE(sk->sk_sndbuf)) 2782 break; 2783 2784 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk); 2785 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 2786 err = -EAGAIN; 2787 if (!timeo) 2788 goto failure; 2789 if (signal_pending(current)) 2790 goto interrupted; 2791 timeo = sock_wait_for_wmem(sk, timeo); 2792 } 2793 skb = alloc_skb_with_frags(header_len, data_len, max_page_order, 2794 errcode, sk->sk_allocation); 2795 if (skb) 2796 skb_set_owner_w(skb, sk); 2797 return skb; 2798 2799 interrupted: 2800 err = sock_intr_errno(timeo); 2801 failure: 2802 *errcode = err; 2803 return NULL; 2804 } 2805 EXPORT_SYMBOL(sock_alloc_send_pskb); 2806 2807 int __sock_cmsg_send(struct sock *sk, struct cmsghdr *cmsg, 2808 struct sockcm_cookie *sockc) 2809 { 2810 u32 tsflags; 2811 2812 switch (cmsg->cmsg_type) { 2813 case SO_MARK: 2814 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) && 2815 !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) 2816 return -EPERM; 2817 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 2818 return -EINVAL; 2819 sockc->mark = *(u32 *)CMSG_DATA(cmsg); 2820 break; 2821 case SO_TIMESTAMPING_OLD: 2822 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 2823 return -EINVAL; 2824 2825 tsflags = *(u32 *)CMSG_DATA(cmsg); 2826 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK) 2827 return -EINVAL; 2828 2829 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK; 2830 sockc->tsflags |= tsflags; 2831 break; 2832 case SCM_TXTIME: 2833 if (!sock_flag(sk, SOCK_TXTIME)) 2834 return -EINVAL; 2835 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64))) 2836 return -EINVAL; 2837 sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg)); 2838 break; 2839 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */ 2840 case SCM_RIGHTS: 2841 case SCM_CREDENTIALS: 2842 break; 2843 default: 2844 return -EINVAL; 2845 } 2846 return 0; 2847 } 2848 EXPORT_SYMBOL(__sock_cmsg_send); 2849 2850 int sock_cmsg_send(struct sock *sk, struct msghdr *msg, 2851 struct sockcm_cookie *sockc) 2852 { 2853 struct cmsghdr *cmsg; 2854 int ret; 2855 2856 for_each_cmsghdr(cmsg, msg) { 2857 if (!CMSG_OK(msg, cmsg)) 2858 return -EINVAL; 2859 if (cmsg->cmsg_level != SOL_SOCKET) 2860 continue; 2861 ret = __sock_cmsg_send(sk, cmsg, sockc); 2862 if (ret) 2863 return ret; 2864 } 2865 return 0; 2866 } 2867 EXPORT_SYMBOL(sock_cmsg_send); 2868 2869 static void sk_enter_memory_pressure(struct sock *sk) 2870 { 2871 if (!sk->sk_prot->enter_memory_pressure) 2872 return; 2873 2874 sk->sk_prot->enter_memory_pressure(sk); 2875 } 2876 2877 static void sk_leave_memory_pressure(struct sock *sk) 2878 { 2879 if (sk->sk_prot->leave_memory_pressure) { 2880 INDIRECT_CALL_INET_1(sk->sk_prot->leave_memory_pressure, 2881 tcp_leave_memory_pressure, sk); 2882 } else { 2883 unsigned long *memory_pressure = sk->sk_prot->memory_pressure; 2884 2885 if (memory_pressure && READ_ONCE(*memory_pressure)) 2886 WRITE_ONCE(*memory_pressure, 0); 2887 } 2888 } 2889 2890 DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key); 2891 2892 /** 2893 * skb_page_frag_refill - check that a page_frag contains enough room 2894 * @sz: minimum size of the fragment we want to get 2895 * @pfrag: pointer to page_frag 2896 * @gfp: priority for memory allocation 2897 * 2898 * Note: While this allocator tries to use high order pages, there is 2899 * no guarantee that allocations succeed. Therefore, @sz MUST be 2900 * less or equal than PAGE_SIZE. 2901 */ 2902 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp) 2903 { 2904 if (pfrag->page) { 2905 if (page_ref_count(pfrag->page) == 1) { 2906 pfrag->offset = 0; 2907 return true; 2908 } 2909 if (pfrag->offset + sz <= pfrag->size) 2910 return true; 2911 put_page(pfrag->page); 2912 } 2913 2914 pfrag->offset = 0; 2915 if (SKB_FRAG_PAGE_ORDER && 2916 !static_branch_unlikely(&net_high_order_alloc_disable_key)) { 2917 /* Avoid direct reclaim but allow kswapd to wake */ 2918 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) | 2919 __GFP_COMP | __GFP_NOWARN | 2920 __GFP_NORETRY, 2921 SKB_FRAG_PAGE_ORDER); 2922 if (likely(pfrag->page)) { 2923 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER; 2924 return true; 2925 } 2926 } 2927 pfrag->page = alloc_page(gfp); 2928 if (likely(pfrag->page)) { 2929 pfrag->size = PAGE_SIZE; 2930 return true; 2931 } 2932 return false; 2933 } 2934 EXPORT_SYMBOL(skb_page_frag_refill); 2935 2936 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag) 2937 { 2938 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation))) 2939 return true; 2940 2941 sk_enter_memory_pressure(sk); 2942 sk_stream_moderate_sndbuf(sk); 2943 return false; 2944 } 2945 EXPORT_SYMBOL(sk_page_frag_refill); 2946 2947 void __lock_sock(struct sock *sk) 2948 __releases(&sk->sk_lock.slock) 2949 __acquires(&sk->sk_lock.slock) 2950 { 2951 DEFINE_WAIT(wait); 2952 2953 for (;;) { 2954 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait, 2955 TASK_UNINTERRUPTIBLE); 2956 spin_unlock_bh(&sk->sk_lock.slock); 2957 schedule(); 2958 spin_lock_bh(&sk->sk_lock.slock); 2959 if (!sock_owned_by_user(sk)) 2960 break; 2961 } 2962 finish_wait(&sk->sk_lock.wq, &wait); 2963 } 2964 2965 void __release_sock(struct sock *sk) 2966 __releases(&sk->sk_lock.slock) 2967 __acquires(&sk->sk_lock.slock) 2968 { 2969 struct sk_buff *skb, *next; 2970 2971 while ((skb = sk->sk_backlog.head) != NULL) { 2972 sk->sk_backlog.head = sk->sk_backlog.tail = NULL; 2973 2974 spin_unlock_bh(&sk->sk_lock.slock); 2975 2976 do { 2977 next = skb->next; 2978 prefetch(next); 2979 DEBUG_NET_WARN_ON_ONCE(skb_dst_is_noref(skb)); 2980 skb_mark_not_on_list(skb); 2981 sk_backlog_rcv(sk, skb); 2982 2983 cond_resched(); 2984 2985 skb = next; 2986 } while (skb != NULL); 2987 2988 spin_lock_bh(&sk->sk_lock.slock); 2989 } 2990 2991 /* 2992 * Doing the zeroing here guarantee we can not loop forever 2993 * while a wild producer attempts to flood us. 2994 */ 2995 sk->sk_backlog.len = 0; 2996 } 2997 2998 void __sk_flush_backlog(struct sock *sk) 2999 { 3000 spin_lock_bh(&sk->sk_lock.slock); 3001 __release_sock(sk); 3002 spin_unlock_bh(&sk->sk_lock.slock); 3003 } 3004 EXPORT_SYMBOL_GPL(__sk_flush_backlog); 3005 3006 /** 3007 * sk_wait_data - wait for data to arrive at sk_receive_queue 3008 * @sk: sock to wait on 3009 * @timeo: for how long 3010 * @skb: last skb seen on sk_receive_queue 3011 * 3012 * Now socket state including sk->sk_err is changed only under lock, 3013 * hence we may omit checks after joining wait queue. 3014 * We check receive queue before schedule() only as optimization; 3015 * it is very likely that release_sock() added new data. 3016 */ 3017 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb) 3018 { 3019 DEFINE_WAIT_FUNC(wait, woken_wake_function); 3020 int rc; 3021 3022 add_wait_queue(sk_sleep(sk), &wait); 3023 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk); 3024 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait); 3025 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk); 3026 remove_wait_queue(sk_sleep(sk), &wait); 3027 return rc; 3028 } 3029 EXPORT_SYMBOL(sk_wait_data); 3030 3031 /** 3032 * __sk_mem_raise_allocated - increase memory_allocated 3033 * @sk: socket 3034 * @size: memory size to allocate 3035 * @amt: pages to allocate 3036 * @kind: allocation type 3037 * 3038 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc 3039 */ 3040 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind) 3041 { 3042 bool memcg_charge = mem_cgroup_sockets_enabled && sk->sk_memcg; 3043 struct proto *prot = sk->sk_prot; 3044 bool charged = true; 3045 long allocated; 3046 3047 sk_memory_allocated_add(sk, amt); 3048 allocated = sk_memory_allocated(sk); 3049 if (memcg_charge && 3050 !(charged = mem_cgroup_charge_skmem(sk->sk_memcg, amt, 3051 gfp_memcg_charge()))) 3052 goto suppress_allocation; 3053 3054 /* Under limit. */ 3055 if (allocated <= sk_prot_mem_limits(sk, 0)) { 3056 sk_leave_memory_pressure(sk); 3057 return 1; 3058 } 3059 3060 /* Under pressure. */ 3061 if (allocated > sk_prot_mem_limits(sk, 1)) 3062 sk_enter_memory_pressure(sk); 3063 3064 /* Over hard limit. */ 3065 if (allocated > sk_prot_mem_limits(sk, 2)) 3066 goto suppress_allocation; 3067 3068 /* guarantee minimum buffer size under pressure */ 3069 if (kind == SK_MEM_RECV) { 3070 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot)) 3071 return 1; 3072 3073 } else { /* SK_MEM_SEND */ 3074 int wmem0 = sk_get_wmem0(sk, prot); 3075 3076 if (sk->sk_type == SOCK_STREAM) { 3077 if (sk->sk_wmem_queued < wmem0) 3078 return 1; 3079 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) { 3080 return 1; 3081 } 3082 } 3083 3084 if (sk_has_memory_pressure(sk)) { 3085 u64 alloc; 3086 3087 if (!sk_under_memory_pressure(sk)) 3088 return 1; 3089 alloc = sk_sockets_allocated_read_positive(sk); 3090 if (sk_prot_mem_limits(sk, 2) > alloc * 3091 sk_mem_pages(sk->sk_wmem_queued + 3092 atomic_read(&sk->sk_rmem_alloc) + 3093 sk->sk_forward_alloc)) 3094 return 1; 3095 } 3096 3097 suppress_allocation: 3098 3099 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) { 3100 sk_stream_moderate_sndbuf(sk); 3101 3102 /* Fail only if socket is _under_ its sndbuf. 3103 * In this case we cannot block, so that we have to fail. 3104 */ 3105 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf) { 3106 /* Force charge with __GFP_NOFAIL */ 3107 if (memcg_charge && !charged) { 3108 mem_cgroup_charge_skmem(sk->sk_memcg, amt, 3109 gfp_memcg_charge() | __GFP_NOFAIL); 3110 } 3111 return 1; 3112 } 3113 } 3114 3115 if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged)) 3116 trace_sock_exceed_buf_limit(sk, prot, allocated, kind); 3117 3118 sk_memory_allocated_sub(sk, amt); 3119 3120 if (memcg_charge && charged) 3121 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt); 3122 3123 return 0; 3124 } 3125 3126 /** 3127 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated 3128 * @sk: socket 3129 * @size: memory size to allocate 3130 * @kind: allocation type 3131 * 3132 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means 3133 * rmem allocation. This function assumes that protocols which have 3134 * memory_pressure use sk_wmem_queued as write buffer accounting. 3135 */ 3136 int __sk_mem_schedule(struct sock *sk, int size, int kind) 3137 { 3138 int ret, amt = sk_mem_pages(size); 3139 3140 sk->sk_forward_alloc += amt << PAGE_SHIFT; 3141 ret = __sk_mem_raise_allocated(sk, size, amt, kind); 3142 if (!ret) 3143 sk->sk_forward_alloc -= amt << PAGE_SHIFT; 3144 return ret; 3145 } 3146 EXPORT_SYMBOL(__sk_mem_schedule); 3147 3148 /** 3149 * __sk_mem_reduce_allocated - reclaim memory_allocated 3150 * @sk: socket 3151 * @amount: number of quanta 3152 * 3153 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc 3154 */ 3155 void __sk_mem_reduce_allocated(struct sock *sk, int amount) 3156 { 3157 sk_memory_allocated_sub(sk, amount); 3158 3159 if (mem_cgroup_sockets_enabled && sk->sk_memcg) 3160 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount); 3161 3162 if (sk_under_memory_pressure(sk) && 3163 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0))) 3164 sk_leave_memory_pressure(sk); 3165 } 3166 3167 /** 3168 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated 3169 * @sk: socket 3170 * @amount: number of bytes (rounded down to a PAGE_SIZE multiple) 3171 */ 3172 void __sk_mem_reclaim(struct sock *sk, int amount) 3173 { 3174 amount >>= PAGE_SHIFT; 3175 sk->sk_forward_alloc -= amount << PAGE_SHIFT; 3176 __sk_mem_reduce_allocated(sk, amount); 3177 } 3178 EXPORT_SYMBOL(__sk_mem_reclaim); 3179 3180 int sk_set_peek_off(struct sock *sk, int val) 3181 { 3182 WRITE_ONCE(sk->sk_peek_off, val); 3183 return 0; 3184 } 3185 EXPORT_SYMBOL_GPL(sk_set_peek_off); 3186 3187 /* 3188 * Set of default routines for initialising struct proto_ops when 3189 * the protocol does not support a particular function. In certain 3190 * cases where it makes no sense for a protocol to have a "do nothing" 3191 * function, some default processing is provided. 3192 */ 3193 3194 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len) 3195 { 3196 return -EOPNOTSUPP; 3197 } 3198 EXPORT_SYMBOL(sock_no_bind); 3199 3200 int sock_no_connect(struct socket *sock, struct sockaddr *saddr, 3201 int len, int flags) 3202 { 3203 return -EOPNOTSUPP; 3204 } 3205 EXPORT_SYMBOL(sock_no_connect); 3206 3207 int sock_no_socketpair(struct socket *sock1, struct socket *sock2) 3208 { 3209 return -EOPNOTSUPP; 3210 } 3211 EXPORT_SYMBOL(sock_no_socketpair); 3212 3213 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags, 3214 bool kern) 3215 { 3216 return -EOPNOTSUPP; 3217 } 3218 EXPORT_SYMBOL(sock_no_accept); 3219 3220 int sock_no_getname(struct socket *sock, struct sockaddr *saddr, 3221 int peer) 3222 { 3223 return -EOPNOTSUPP; 3224 } 3225 EXPORT_SYMBOL(sock_no_getname); 3226 3227 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 3228 { 3229 return -EOPNOTSUPP; 3230 } 3231 EXPORT_SYMBOL(sock_no_ioctl); 3232 3233 int sock_no_listen(struct socket *sock, int backlog) 3234 { 3235 return -EOPNOTSUPP; 3236 } 3237 EXPORT_SYMBOL(sock_no_listen); 3238 3239 int sock_no_shutdown(struct socket *sock, int how) 3240 { 3241 return -EOPNOTSUPP; 3242 } 3243 EXPORT_SYMBOL(sock_no_shutdown); 3244 3245 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len) 3246 { 3247 return -EOPNOTSUPP; 3248 } 3249 EXPORT_SYMBOL(sock_no_sendmsg); 3250 3251 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len) 3252 { 3253 return -EOPNOTSUPP; 3254 } 3255 EXPORT_SYMBOL(sock_no_sendmsg_locked); 3256 3257 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len, 3258 int flags) 3259 { 3260 return -EOPNOTSUPP; 3261 } 3262 EXPORT_SYMBOL(sock_no_recvmsg); 3263 3264 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma) 3265 { 3266 /* Mirror missing mmap method error code */ 3267 return -ENODEV; 3268 } 3269 EXPORT_SYMBOL(sock_no_mmap); 3270 3271 /* 3272 * When a file is received (via SCM_RIGHTS, etc), we must bump the 3273 * various sock-based usage counts. 3274 */ 3275 void __receive_sock(struct file *file) 3276 { 3277 struct socket *sock; 3278 3279 sock = sock_from_file(file); 3280 if (sock) { 3281 sock_update_netprioidx(&sock->sk->sk_cgrp_data); 3282 sock_update_classid(&sock->sk->sk_cgrp_data); 3283 } 3284 } 3285 3286 /* 3287 * Default Socket Callbacks 3288 */ 3289 3290 static void sock_def_wakeup(struct sock *sk) 3291 { 3292 struct socket_wq *wq; 3293 3294 rcu_read_lock(); 3295 wq = rcu_dereference(sk->sk_wq); 3296 if (skwq_has_sleeper(wq)) 3297 wake_up_interruptible_all(&wq->wait); 3298 rcu_read_unlock(); 3299 } 3300 3301 static void sock_def_error_report(struct sock *sk) 3302 { 3303 struct socket_wq *wq; 3304 3305 rcu_read_lock(); 3306 wq = rcu_dereference(sk->sk_wq); 3307 if (skwq_has_sleeper(wq)) 3308 wake_up_interruptible_poll(&wq->wait, EPOLLERR); 3309 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR); 3310 rcu_read_unlock(); 3311 } 3312 3313 void sock_def_readable(struct sock *sk) 3314 { 3315 struct socket_wq *wq; 3316 3317 trace_sk_data_ready(sk); 3318 3319 rcu_read_lock(); 3320 wq = rcu_dereference(sk->sk_wq); 3321 if (skwq_has_sleeper(wq)) 3322 wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI | 3323 EPOLLRDNORM | EPOLLRDBAND); 3324 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN); 3325 rcu_read_unlock(); 3326 } 3327 3328 static void sock_def_write_space(struct sock *sk) 3329 { 3330 struct socket_wq *wq; 3331 3332 rcu_read_lock(); 3333 3334 /* Do not wake up a writer until he can make "significant" 3335 * progress. --DaveM 3336 */ 3337 if (sock_writeable(sk)) { 3338 wq = rcu_dereference(sk->sk_wq); 3339 if (skwq_has_sleeper(wq)) 3340 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT | 3341 EPOLLWRNORM | EPOLLWRBAND); 3342 3343 /* Should agree with poll, otherwise some programs break */ 3344 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT); 3345 } 3346 3347 rcu_read_unlock(); 3348 } 3349 3350 /* An optimised version of sock_def_write_space(), should only be called 3351 * for SOCK_RCU_FREE sockets under RCU read section and after putting 3352 * ->sk_wmem_alloc. 3353 */ 3354 static void sock_def_write_space_wfree(struct sock *sk) 3355 { 3356 /* Do not wake up a writer until he can make "significant" 3357 * progress. --DaveM 3358 */ 3359 if (sock_writeable(sk)) { 3360 struct socket_wq *wq = rcu_dereference(sk->sk_wq); 3361 3362 /* rely on refcount_sub from sock_wfree() */ 3363 smp_mb__after_atomic(); 3364 if (wq && waitqueue_active(&wq->wait)) 3365 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT | 3366 EPOLLWRNORM | EPOLLWRBAND); 3367 3368 /* Should agree with poll, otherwise some programs break */ 3369 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT); 3370 } 3371 } 3372 3373 static void sock_def_destruct(struct sock *sk) 3374 { 3375 } 3376 3377 void sk_send_sigurg(struct sock *sk) 3378 { 3379 if (sk->sk_socket && sk->sk_socket->file) 3380 if (send_sigurg(&sk->sk_socket->file->f_owner)) 3381 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI); 3382 } 3383 EXPORT_SYMBOL(sk_send_sigurg); 3384 3385 void sk_reset_timer(struct sock *sk, struct timer_list* timer, 3386 unsigned long expires) 3387 { 3388 if (!mod_timer(timer, expires)) 3389 sock_hold(sk); 3390 } 3391 EXPORT_SYMBOL(sk_reset_timer); 3392 3393 void sk_stop_timer(struct sock *sk, struct timer_list* timer) 3394 { 3395 if (del_timer(timer)) 3396 __sock_put(sk); 3397 } 3398 EXPORT_SYMBOL(sk_stop_timer); 3399 3400 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer) 3401 { 3402 if (del_timer_sync(timer)) 3403 __sock_put(sk); 3404 } 3405 EXPORT_SYMBOL(sk_stop_timer_sync); 3406 3407 void sock_init_data_uid(struct socket *sock, struct sock *sk, kuid_t uid) 3408 { 3409 sk_init_common(sk); 3410 sk->sk_send_head = NULL; 3411 3412 timer_setup(&sk->sk_timer, NULL, 0); 3413 3414 sk->sk_allocation = GFP_KERNEL; 3415 sk->sk_rcvbuf = READ_ONCE(sysctl_rmem_default); 3416 sk->sk_sndbuf = READ_ONCE(sysctl_wmem_default); 3417 sk->sk_state = TCP_CLOSE; 3418 sk->sk_use_task_frag = true; 3419 sk_set_socket(sk, sock); 3420 3421 sock_set_flag(sk, SOCK_ZAPPED); 3422 3423 if (sock) { 3424 sk->sk_type = sock->type; 3425 RCU_INIT_POINTER(sk->sk_wq, &sock->wq); 3426 sock->sk = sk; 3427 } else { 3428 RCU_INIT_POINTER(sk->sk_wq, NULL); 3429 } 3430 sk->sk_uid = uid; 3431 3432 rwlock_init(&sk->sk_callback_lock); 3433 if (sk->sk_kern_sock) 3434 lockdep_set_class_and_name( 3435 &sk->sk_callback_lock, 3436 af_kern_callback_keys + sk->sk_family, 3437 af_family_kern_clock_key_strings[sk->sk_family]); 3438 else 3439 lockdep_set_class_and_name( 3440 &sk->sk_callback_lock, 3441 af_callback_keys + sk->sk_family, 3442 af_family_clock_key_strings[sk->sk_family]); 3443 3444 sk->sk_state_change = sock_def_wakeup; 3445 sk->sk_data_ready = sock_def_readable; 3446 sk->sk_write_space = sock_def_write_space; 3447 sk->sk_error_report = sock_def_error_report; 3448 sk->sk_destruct = sock_def_destruct; 3449 3450 sk->sk_frag.page = NULL; 3451 sk->sk_frag.offset = 0; 3452 sk->sk_peek_off = -1; 3453 3454 sk->sk_peer_pid = NULL; 3455 sk->sk_peer_cred = NULL; 3456 spin_lock_init(&sk->sk_peer_lock); 3457 3458 sk->sk_write_pending = 0; 3459 sk->sk_rcvlowat = 1; 3460 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; 3461 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT; 3462 3463 sk->sk_stamp = SK_DEFAULT_STAMP; 3464 #if BITS_PER_LONG==32 3465 seqlock_init(&sk->sk_stamp_seq); 3466 #endif 3467 atomic_set(&sk->sk_zckey, 0); 3468 3469 #ifdef CONFIG_NET_RX_BUSY_POLL 3470 sk->sk_napi_id = 0; 3471 sk->sk_ll_usec = READ_ONCE(sysctl_net_busy_read); 3472 #endif 3473 3474 sk->sk_max_pacing_rate = ~0UL; 3475 sk->sk_pacing_rate = ~0UL; 3476 WRITE_ONCE(sk->sk_pacing_shift, 10); 3477 sk->sk_incoming_cpu = -1; 3478 3479 sk_rx_queue_clear(sk); 3480 /* 3481 * Before updating sk_refcnt, we must commit prior changes to memory 3482 * (Documentation/RCU/rculist_nulls.rst for details) 3483 */ 3484 smp_wmb(); 3485 refcount_set(&sk->sk_refcnt, 1); 3486 atomic_set(&sk->sk_drops, 0); 3487 } 3488 EXPORT_SYMBOL(sock_init_data_uid); 3489 3490 void sock_init_data(struct socket *sock, struct sock *sk) 3491 { 3492 kuid_t uid = sock ? 3493 SOCK_INODE(sock)->i_uid : 3494 make_kuid(sock_net(sk)->user_ns, 0); 3495 3496 sock_init_data_uid(sock, sk, uid); 3497 } 3498 EXPORT_SYMBOL(sock_init_data); 3499 3500 void lock_sock_nested(struct sock *sk, int subclass) 3501 { 3502 /* The sk_lock has mutex_lock() semantics here. */ 3503 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_); 3504 3505 might_sleep(); 3506 spin_lock_bh(&sk->sk_lock.slock); 3507 if (sock_owned_by_user_nocheck(sk)) 3508 __lock_sock(sk); 3509 sk->sk_lock.owned = 1; 3510 spin_unlock_bh(&sk->sk_lock.slock); 3511 } 3512 EXPORT_SYMBOL(lock_sock_nested); 3513 3514 void release_sock(struct sock *sk) 3515 { 3516 spin_lock_bh(&sk->sk_lock.slock); 3517 if (sk->sk_backlog.tail) 3518 __release_sock(sk); 3519 3520 /* Warning : release_cb() might need to release sk ownership, 3521 * ie call sock_release_ownership(sk) before us. 3522 */ 3523 if (sk->sk_prot->release_cb) 3524 sk->sk_prot->release_cb(sk); 3525 3526 sock_release_ownership(sk); 3527 if (waitqueue_active(&sk->sk_lock.wq)) 3528 wake_up(&sk->sk_lock.wq); 3529 spin_unlock_bh(&sk->sk_lock.slock); 3530 } 3531 EXPORT_SYMBOL(release_sock); 3532 3533 bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock) 3534 { 3535 might_sleep(); 3536 spin_lock_bh(&sk->sk_lock.slock); 3537 3538 if (!sock_owned_by_user_nocheck(sk)) { 3539 /* 3540 * Fast path return with bottom halves disabled and 3541 * sock::sk_lock.slock held. 3542 * 3543 * The 'mutex' is not contended and holding 3544 * sock::sk_lock.slock prevents all other lockers to 3545 * proceed so the corresponding unlock_sock_fast() can 3546 * avoid the slow path of release_sock() completely and 3547 * just release slock. 3548 * 3549 * From a semantical POV this is equivalent to 'acquiring' 3550 * the 'mutex', hence the corresponding lockdep 3551 * mutex_release() has to happen in the fast path of 3552 * unlock_sock_fast(). 3553 */ 3554 return false; 3555 } 3556 3557 __lock_sock(sk); 3558 sk->sk_lock.owned = 1; 3559 __acquire(&sk->sk_lock.slock); 3560 spin_unlock_bh(&sk->sk_lock.slock); 3561 return true; 3562 } 3563 EXPORT_SYMBOL(__lock_sock_fast); 3564 3565 int sock_gettstamp(struct socket *sock, void __user *userstamp, 3566 bool timeval, bool time32) 3567 { 3568 struct sock *sk = sock->sk; 3569 struct timespec64 ts; 3570 3571 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 3572 ts = ktime_to_timespec64(sock_read_timestamp(sk)); 3573 if (ts.tv_sec == -1) 3574 return -ENOENT; 3575 if (ts.tv_sec == 0) { 3576 ktime_t kt = ktime_get_real(); 3577 sock_write_timestamp(sk, kt); 3578 ts = ktime_to_timespec64(kt); 3579 } 3580 3581 if (timeval) 3582 ts.tv_nsec /= 1000; 3583 3584 #ifdef CONFIG_COMPAT_32BIT_TIME 3585 if (time32) 3586 return put_old_timespec32(&ts, userstamp); 3587 #endif 3588 #ifdef CONFIG_SPARC64 3589 /* beware of padding in sparc64 timeval */ 3590 if (timeval && !in_compat_syscall()) { 3591 struct __kernel_old_timeval __user tv = { 3592 .tv_sec = ts.tv_sec, 3593 .tv_usec = ts.tv_nsec, 3594 }; 3595 if (copy_to_user(userstamp, &tv, sizeof(tv))) 3596 return -EFAULT; 3597 return 0; 3598 } 3599 #endif 3600 return put_timespec64(&ts, userstamp); 3601 } 3602 EXPORT_SYMBOL(sock_gettstamp); 3603 3604 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag) 3605 { 3606 if (!sock_flag(sk, flag)) { 3607 unsigned long previous_flags = sk->sk_flags; 3608 3609 sock_set_flag(sk, flag); 3610 /* 3611 * we just set one of the two flags which require net 3612 * time stamping, but time stamping might have been on 3613 * already because of the other one 3614 */ 3615 if (sock_needs_netstamp(sk) && 3616 !(previous_flags & SK_FLAGS_TIMESTAMP)) 3617 net_enable_timestamp(); 3618 } 3619 } 3620 3621 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, 3622 int level, int type) 3623 { 3624 struct sock_exterr_skb *serr; 3625 struct sk_buff *skb; 3626 int copied, err; 3627 3628 err = -EAGAIN; 3629 skb = sock_dequeue_err_skb(sk); 3630 if (skb == NULL) 3631 goto out; 3632 3633 copied = skb->len; 3634 if (copied > len) { 3635 msg->msg_flags |= MSG_TRUNC; 3636 copied = len; 3637 } 3638 err = skb_copy_datagram_msg(skb, 0, msg, copied); 3639 if (err) 3640 goto out_free_skb; 3641 3642 sock_recv_timestamp(msg, sk, skb); 3643 3644 serr = SKB_EXT_ERR(skb); 3645 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee); 3646 3647 msg->msg_flags |= MSG_ERRQUEUE; 3648 err = copied; 3649 3650 out_free_skb: 3651 kfree_skb(skb); 3652 out: 3653 return err; 3654 } 3655 EXPORT_SYMBOL(sock_recv_errqueue); 3656 3657 /* 3658 * Get a socket option on an socket. 3659 * 3660 * FIX: POSIX 1003.1g is very ambiguous here. It states that 3661 * asynchronous errors should be reported by getsockopt. We assume 3662 * this means if you specify SO_ERROR (otherwise whats the point of it). 3663 */ 3664 int sock_common_getsockopt(struct socket *sock, int level, int optname, 3665 char __user *optval, int __user *optlen) 3666 { 3667 struct sock *sk = sock->sk; 3668 3669 /* IPV6_ADDRFORM can change sk->sk_prot under us. */ 3670 return READ_ONCE(sk->sk_prot)->getsockopt(sk, level, optname, optval, optlen); 3671 } 3672 EXPORT_SYMBOL(sock_common_getsockopt); 3673 3674 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 3675 int flags) 3676 { 3677 struct sock *sk = sock->sk; 3678 int addr_len = 0; 3679 int err; 3680 3681 err = sk->sk_prot->recvmsg(sk, msg, size, flags, &addr_len); 3682 if (err >= 0) 3683 msg->msg_namelen = addr_len; 3684 return err; 3685 } 3686 EXPORT_SYMBOL(sock_common_recvmsg); 3687 3688 /* 3689 * Set socket options on an inet socket. 3690 */ 3691 int sock_common_setsockopt(struct socket *sock, int level, int optname, 3692 sockptr_t optval, unsigned int optlen) 3693 { 3694 struct sock *sk = sock->sk; 3695 3696 /* IPV6_ADDRFORM can change sk->sk_prot under us. */ 3697 return READ_ONCE(sk->sk_prot)->setsockopt(sk, level, optname, optval, optlen); 3698 } 3699 EXPORT_SYMBOL(sock_common_setsockopt); 3700 3701 void sk_common_release(struct sock *sk) 3702 { 3703 if (sk->sk_prot->destroy) 3704 sk->sk_prot->destroy(sk); 3705 3706 /* 3707 * Observation: when sk_common_release is called, processes have 3708 * no access to socket. But net still has. 3709 * Step one, detach it from networking: 3710 * 3711 * A. Remove from hash tables. 3712 */ 3713 3714 sk->sk_prot->unhash(sk); 3715 3716 /* 3717 * In this point socket cannot receive new packets, but it is possible 3718 * that some packets are in flight because some CPU runs receiver and 3719 * did hash table lookup before we unhashed socket. They will achieve 3720 * receive queue and will be purged by socket destructor. 3721 * 3722 * Also we still have packets pending on receive queue and probably, 3723 * our own packets waiting in device queues. sock_destroy will drain 3724 * receive queue, but transmitted packets will delay socket destruction 3725 * until the last reference will be released. 3726 */ 3727 3728 sock_orphan(sk); 3729 3730 xfrm_sk_free_policy(sk); 3731 3732 sock_put(sk); 3733 } 3734 EXPORT_SYMBOL(sk_common_release); 3735 3736 void sk_get_meminfo(const struct sock *sk, u32 *mem) 3737 { 3738 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS); 3739 3740 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk); 3741 mem[SK_MEMINFO_RCVBUF] = READ_ONCE(sk->sk_rcvbuf); 3742 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk); 3743 mem[SK_MEMINFO_SNDBUF] = READ_ONCE(sk->sk_sndbuf); 3744 mem[SK_MEMINFO_FWD_ALLOC] = sk->sk_forward_alloc; 3745 mem[SK_MEMINFO_WMEM_QUEUED] = READ_ONCE(sk->sk_wmem_queued); 3746 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc); 3747 mem[SK_MEMINFO_BACKLOG] = READ_ONCE(sk->sk_backlog.len); 3748 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops); 3749 } 3750 3751 #ifdef CONFIG_PROC_FS 3752 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR); 3753 3754 int sock_prot_inuse_get(struct net *net, struct proto *prot) 3755 { 3756 int cpu, idx = prot->inuse_idx; 3757 int res = 0; 3758 3759 for_each_possible_cpu(cpu) 3760 res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx]; 3761 3762 return res >= 0 ? res : 0; 3763 } 3764 EXPORT_SYMBOL_GPL(sock_prot_inuse_get); 3765 3766 int sock_inuse_get(struct net *net) 3767 { 3768 int cpu, res = 0; 3769 3770 for_each_possible_cpu(cpu) 3771 res += per_cpu_ptr(net->core.prot_inuse, cpu)->all; 3772 3773 return res; 3774 } 3775 3776 EXPORT_SYMBOL_GPL(sock_inuse_get); 3777 3778 static int __net_init sock_inuse_init_net(struct net *net) 3779 { 3780 net->core.prot_inuse = alloc_percpu(struct prot_inuse); 3781 if (net->core.prot_inuse == NULL) 3782 return -ENOMEM; 3783 return 0; 3784 } 3785 3786 static void __net_exit sock_inuse_exit_net(struct net *net) 3787 { 3788 free_percpu(net->core.prot_inuse); 3789 } 3790 3791 static struct pernet_operations net_inuse_ops = { 3792 .init = sock_inuse_init_net, 3793 .exit = sock_inuse_exit_net, 3794 }; 3795 3796 static __init int net_inuse_init(void) 3797 { 3798 if (register_pernet_subsys(&net_inuse_ops)) 3799 panic("Cannot initialize net inuse counters"); 3800 3801 return 0; 3802 } 3803 3804 core_initcall(net_inuse_init); 3805 3806 static int assign_proto_idx(struct proto *prot) 3807 { 3808 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR); 3809 3810 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) { 3811 pr_err("PROTO_INUSE_NR exhausted\n"); 3812 return -ENOSPC; 3813 } 3814 3815 set_bit(prot->inuse_idx, proto_inuse_idx); 3816 return 0; 3817 } 3818 3819 static void release_proto_idx(struct proto *prot) 3820 { 3821 if (prot->inuse_idx != PROTO_INUSE_NR - 1) 3822 clear_bit(prot->inuse_idx, proto_inuse_idx); 3823 } 3824 #else 3825 static inline int assign_proto_idx(struct proto *prot) 3826 { 3827 return 0; 3828 } 3829 3830 static inline void release_proto_idx(struct proto *prot) 3831 { 3832 } 3833 3834 #endif 3835 3836 static void tw_prot_cleanup(struct timewait_sock_ops *twsk_prot) 3837 { 3838 if (!twsk_prot) 3839 return; 3840 kfree(twsk_prot->twsk_slab_name); 3841 twsk_prot->twsk_slab_name = NULL; 3842 kmem_cache_destroy(twsk_prot->twsk_slab); 3843 twsk_prot->twsk_slab = NULL; 3844 } 3845 3846 static int tw_prot_init(const struct proto *prot) 3847 { 3848 struct timewait_sock_ops *twsk_prot = prot->twsk_prot; 3849 3850 if (!twsk_prot) 3851 return 0; 3852 3853 twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", 3854 prot->name); 3855 if (!twsk_prot->twsk_slab_name) 3856 return -ENOMEM; 3857 3858 twsk_prot->twsk_slab = 3859 kmem_cache_create(twsk_prot->twsk_slab_name, 3860 twsk_prot->twsk_obj_size, 0, 3861 SLAB_ACCOUNT | prot->slab_flags, 3862 NULL); 3863 if (!twsk_prot->twsk_slab) { 3864 pr_crit("%s: Can't create timewait sock SLAB cache!\n", 3865 prot->name); 3866 return -ENOMEM; 3867 } 3868 3869 return 0; 3870 } 3871 3872 static void req_prot_cleanup(struct request_sock_ops *rsk_prot) 3873 { 3874 if (!rsk_prot) 3875 return; 3876 kfree(rsk_prot->slab_name); 3877 rsk_prot->slab_name = NULL; 3878 kmem_cache_destroy(rsk_prot->slab); 3879 rsk_prot->slab = NULL; 3880 } 3881 3882 static int req_prot_init(const struct proto *prot) 3883 { 3884 struct request_sock_ops *rsk_prot = prot->rsk_prot; 3885 3886 if (!rsk_prot) 3887 return 0; 3888 3889 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", 3890 prot->name); 3891 if (!rsk_prot->slab_name) 3892 return -ENOMEM; 3893 3894 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name, 3895 rsk_prot->obj_size, 0, 3896 SLAB_ACCOUNT | prot->slab_flags, 3897 NULL); 3898 3899 if (!rsk_prot->slab) { 3900 pr_crit("%s: Can't create request sock SLAB cache!\n", 3901 prot->name); 3902 return -ENOMEM; 3903 } 3904 return 0; 3905 } 3906 3907 int proto_register(struct proto *prot, int alloc_slab) 3908 { 3909 int ret = -ENOBUFS; 3910 3911 if (prot->memory_allocated && !prot->sysctl_mem) { 3912 pr_err("%s: missing sysctl_mem\n", prot->name); 3913 return -EINVAL; 3914 } 3915 if (prot->memory_allocated && !prot->per_cpu_fw_alloc) { 3916 pr_err("%s: missing per_cpu_fw_alloc\n", prot->name); 3917 return -EINVAL; 3918 } 3919 if (alloc_slab) { 3920 prot->slab = kmem_cache_create_usercopy(prot->name, 3921 prot->obj_size, 0, 3922 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT | 3923 prot->slab_flags, 3924 prot->useroffset, prot->usersize, 3925 NULL); 3926 3927 if (prot->slab == NULL) { 3928 pr_crit("%s: Can't create sock SLAB cache!\n", 3929 prot->name); 3930 goto out; 3931 } 3932 3933 if (req_prot_init(prot)) 3934 goto out_free_request_sock_slab; 3935 3936 if (tw_prot_init(prot)) 3937 goto out_free_timewait_sock_slab; 3938 } 3939 3940 mutex_lock(&proto_list_mutex); 3941 ret = assign_proto_idx(prot); 3942 if (ret) { 3943 mutex_unlock(&proto_list_mutex); 3944 goto out_free_timewait_sock_slab; 3945 } 3946 list_add(&prot->node, &proto_list); 3947 mutex_unlock(&proto_list_mutex); 3948 return ret; 3949 3950 out_free_timewait_sock_slab: 3951 if (alloc_slab) 3952 tw_prot_cleanup(prot->twsk_prot); 3953 out_free_request_sock_slab: 3954 if (alloc_slab) { 3955 req_prot_cleanup(prot->rsk_prot); 3956 3957 kmem_cache_destroy(prot->slab); 3958 prot->slab = NULL; 3959 } 3960 out: 3961 return ret; 3962 } 3963 EXPORT_SYMBOL(proto_register); 3964 3965 void proto_unregister(struct proto *prot) 3966 { 3967 mutex_lock(&proto_list_mutex); 3968 release_proto_idx(prot); 3969 list_del(&prot->node); 3970 mutex_unlock(&proto_list_mutex); 3971 3972 kmem_cache_destroy(prot->slab); 3973 prot->slab = NULL; 3974 3975 req_prot_cleanup(prot->rsk_prot); 3976 tw_prot_cleanup(prot->twsk_prot); 3977 } 3978 EXPORT_SYMBOL(proto_unregister); 3979 3980 int sock_load_diag_module(int family, int protocol) 3981 { 3982 if (!protocol) { 3983 if (!sock_is_registered(family)) 3984 return -ENOENT; 3985 3986 return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK, 3987 NETLINK_SOCK_DIAG, family); 3988 } 3989 3990 #ifdef CONFIG_INET 3991 if (family == AF_INET && 3992 protocol != IPPROTO_RAW && 3993 protocol < MAX_INET_PROTOS && 3994 !rcu_access_pointer(inet_protos[protocol])) 3995 return -ENOENT; 3996 #endif 3997 3998 return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK, 3999 NETLINK_SOCK_DIAG, family, protocol); 4000 } 4001 EXPORT_SYMBOL(sock_load_diag_module); 4002 4003 #ifdef CONFIG_PROC_FS 4004 static void *proto_seq_start(struct seq_file *seq, loff_t *pos) 4005 __acquires(proto_list_mutex) 4006 { 4007 mutex_lock(&proto_list_mutex); 4008 return seq_list_start_head(&proto_list, *pos); 4009 } 4010 4011 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos) 4012 { 4013 return seq_list_next(v, &proto_list, pos); 4014 } 4015 4016 static void proto_seq_stop(struct seq_file *seq, void *v) 4017 __releases(proto_list_mutex) 4018 { 4019 mutex_unlock(&proto_list_mutex); 4020 } 4021 4022 static char proto_method_implemented(const void *method) 4023 { 4024 return method == NULL ? 'n' : 'y'; 4025 } 4026 static long sock_prot_memory_allocated(struct proto *proto) 4027 { 4028 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L; 4029 } 4030 4031 static const char *sock_prot_memory_pressure(struct proto *proto) 4032 { 4033 return proto->memory_pressure != NULL ? 4034 proto_memory_pressure(proto) ? "yes" : "no" : "NI"; 4035 } 4036 4037 static void proto_seq_printf(struct seq_file *seq, struct proto *proto) 4038 { 4039 4040 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s " 4041 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n", 4042 proto->name, 4043 proto->obj_size, 4044 sock_prot_inuse_get(seq_file_net(seq), proto), 4045 sock_prot_memory_allocated(proto), 4046 sock_prot_memory_pressure(proto), 4047 proto->max_header, 4048 proto->slab == NULL ? "no" : "yes", 4049 module_name(proto->owner), 4050 proto_method_implemented(proto->close), 4051 proto_method_implemented(proto->connect), 4052 proto_method_implemented(proto->disconnect), 4053 proto_method_implemented(proto->accept), 4054 proto_method_implemented(proto->ioctl), 4055 proto_method_implemented(proto->init), 4056 proto_method_implemented(proto->destroy), 4057 proto_method_implemented(proto->shutdown), 4058 proto_method_implemented(proto->setsockopt), 4059 proto_method_implemented(proto->getsockopt), 4060 proto_method_implemented(proto->sendmsg), 4061 proto_method_implemented(proto->recvmsg), 4062 proto_method_implemented(proto->bind), 4063 proto_method_implemented(proto->backlog_rcv), 4064 proto_method_implemented(proto->hash), 4065 proto_method_implemented(proto->unhash), 4066 proto_method_implemented(proto->get_port), 4067 proto_method_implemented(proto->enter_memory_pressure)); 4068 } 4069 4070 static int proto_seq_show(struct seq_file *seq, void *v) 4071 { 4072 if (v == &proto_list) 4073 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s", 4074 "protocol", 4075 "size", 4076 "sockets", 4077 "memory", 4078 "press", 4079 "maxhdr", 4080 "slab", 4081 "module", 4082 "cl co di ac io in de sh ss gs se re bi br ha uh gp em\n"); 4083 else 4084 proto_seq_printf(seq, list_entry(v, struct proto, node)); 4085 return 0; 4086 } 4087 4088 static const struct seq_operations proto_seq_ops = { 4089 .start = proto_seq_start, 4090 .next = proto_seq_next, 4091 .stop = proto_seq_stop, 4092 .show = proto_seq_show, 4093 }; 4094 4095 static __net_init int proto_init_net(struct net *net) 4096 { 4097 if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops, 4098 sizeof(struct seq_net_private))) 4099 return -ENOMEM; 4100 4101 return 0; 4102 } 4103 4104 static __net_exit void proto_exit_net(struct net *net) 4105 { 4106 remove_proc_entry("protocols", net->proc_net); 4107 } 4108 4109 4110 static __net_initdata struct pernet_operations proto_net_ops = { 4111 .init = proto_init_net, 4112 .exit = proto_exit_net, 4113 }; 4114 4115 static int __init proto_init(void) 4116 { 4117 return register_pernet_subsys(&proto_net_ops); 4118 } 4119 4120 subsys_initcall(proto_init); 4121 4122 #endif /* PROC_FS */ 4123 4124 #ifdef CONFIG_NET_RX_BUSY_POLL 4125 bool sk_busy_loop_end(void *p, unsigned long start_time) 4126 { 4127 struct sock *sk = p; 4128 4129 return !skb_queue_empty_lockless(&sk->sk_receive_queue) || 4130 sk_busy_loop_timeout(sk, start_time); 4131 } 4132 EXPORT_SYMBOL(sk_busy_loop_end); 4133 #endif /* CONFIG_NET_RX_BUSY_POLL */ 4134 4135 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len) 4136 { 4137 if (!sk->sk_prot->bind_add) 4138 return -EOPNOTSUPP; 4139 return sk->sk_prot->bind_add(sk, addr, addr_len); 4140 } 4141 EXPORT_SYMBOL(sock_bind_add); 4142 4143 /* Copy 'size' bytes from userspace and return `size` back to userspace */ 4144 int sock_ioctl_inout(struct sock *sk, unsigned int cmd, 4145 void __user *arg, void *karg, size_t size) 4146 { 4147 int ret; 4148 4149 if (copy_from_user(karg, arg, size)) 4150 return -EFAULT; 4151 4152 ret = READ_ONCE(sk->sk_prot)->ioctl(sk, cmd, karg); 4153 if (ret) 4154 return ret; 4155 4156 if (copy_to_user(arg, karg, size)) 4157 return -EFAULT; 4158 4159 return 0; 4160 } 4161 EXPORT_SYMBOL(sock_ioctl_inout); 4162 4163 /* This is the most common ioctl prep function, where the result (4 bytes) is 4164 * copied back to userspace if the ioctl() returns successfully. No input is 4165 * copied from userspace as input argument. 4166 */ 4167 static int sock_ioctl_out(struct sock *sk, unsigned int cmd, void __user *arg) 4168 { 4169 int ret, karg = 0; 4170 4171 ret = READ_ONCE(sk->sk_prot)->ioctl(sk, cmd, &karg); 4172 if (ret) 4173 return ret; 4174 4175 return put_user(karg, (int __user *)arg); 4176 } 4177 4178 /* A wrapper around sock ioctls, which copies the data from userspace 4179 * (depending on the protocol/ioctl), and copies back the result to userspace. 4180 * The main motivation for this function is to pass kernel memory to the 4181 * protocol ioctl callbacks, instead of userspace memory. 4182 */ 4183 int sk_ioctl(struct sock *sk, unsigned int cmd, void __user *arg) 4184 { 4185 int rc = 1; 4186 4187 if (sk->sk_type == SOCK_RAW && sk->sk_family == AF_INET) 4188 rc = ipmr_sk_ioctl(sk, cmd, arg); 4189 else if (sk->sk_type == SOCK_RAW && sk->sk_family == AF_INET6) 4190 rc = ip6mr_sk_ioctl(sk, cmd, arg); 4191 else if (sk_is_phonet(sk)) 4192 rc = phonet_sk_ioctl(sk, cmd, arg); 4193 4194 /* If ioctl was processed, returns its value */ 4195 if (rc <= 0) 4196 return rc; 4197 4198 /* Otherwise call the default handler */ 4199 return sock_ioctl_out(sk, cmd, arg); 4200 } 4201 EXPORT_SYMBOL(sk_ioctl); 4202