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