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 /* Paired with READ_ONCE() in tcp_rtx_synack() */ 1535 WRITE_ONCE(sk->sk_txrehash, (u8)val); 1536 break; 1537 1538 default: 1539 ret = -ENOPROTOOPT; 1540 break; 1541 } 1542 sockopt_release_sock(sk); 1543 return ret; 1544 } 1545 1546 int sock_setsockopt(struct socket *sock, int level, int optname, 1547 sockptr_t optval, unsigned int optlen) 1548 { 1549 return sk_setsockopt(sock->sk, level, optname, 1550 optval, optlen); 1551 } 1552 EXPORT_SYMBOL(sock_setsockopt); 1553 1554 static const struct cred *sk_get_peer_cred(struct sock *sk) 1555 { 1556 const struct cred *cred; 1557 1558 spin_lock(&sk->sk_peer_lock); 1559 cred = get_cred(sk->sk_peer_cred); 1560 spin_unlock(&sk->sk_peer_lock); 1561 1562 return cred; 1563 } 1564 1565 static void cred_to_ucred(struct pid *pid, const struct cred *cred, 1566 struct ucred *ucred) 1567 { 1568 ucred->pid = pid_vnr(pid); 1569 ucred->uid = ucred->gid = -1; 1570 if (cred) { 1571 struct user_namespace *current_ns = current_user_ns(); 1572 1573 ucred->uid = from_kuid_munged(current_ns, cred->euid); 1574 ucred->gid = from_kgid_munged(current_ns, cred->egid); 1575 } 1576 } 1577 1578 static int groups_to_user(sockptr_t dst, const struct group_info *src) 1579 { 1580 struct user_namespace *user_ns = current_user_ns(); 1581 int i; 1582 1583 for (i = 0; i < src->ngroups; i++) { 1584 gid_t gid = from_kgid_munged(user_ns, src->gid[i]); 1585 1586 if (copy_to_sockptr_offset(dst, i * sizeof(gid), &gid, sizeof(gid))) 1587 return -EFAULT; 1588 } 1589 1590 return 0; 1591 } 1592 1593 int sk_getsockopt(struct sock *sk, int level, int optname, 1594 sockptr_t optval, sockptr_t optlen) 1595 { 1596 struct socket *sock = sk->sk_socket; 1597 1598 union { 1599 int val; 1600 u64 val64; 1601 unsigned long ulval; 1602 struct linger ling; 1603 struct old_timeval32 tm32; 1604 struct __kernel_old_timeval tm; 1605 struct __kernel_sock_timeval stm; 1606 struct sock_txtime txtime; 1607 struct so_timestamping timestamping; 1608 } v; 1609 1610 int lv = sizeof(int); 1611 int len; 1612 1613 if (copy_from_sockptr(&len, optlen, sizeof(int))) 1614 return -EFAULT; 1615 if (len < 0) 1616 return -EINVAL; 1617 1618 memset(&v, 0, sizeof(v)); 1619 1620 switch (optname) { 1621 case SO_DEBUG: 1622 v.val = sock_flag(sk, SOCK_DBG); 1623 break; 1624 1625 case SO_DONTROUTE: 1626 v.val = sock_flag(sk, SOCK_LOCALROUTE); 1627 break; 1628 1629 case SO_BROADCAST: 1630 v.val = sock_flag(sk, SOCK_BROADCAST); 1631 break; 1632 1633 case SO_SNDBUF: 1634 v.val = sk->sk_sndbuf; 1635 break; 1636 1637 case SO_RCVBUF: 1638 v.val = sk->sk_rcvbuf; 1639 break; 1640 1641 case SO_REUSEADDR: 1642 v.val = sk->sk_reuse; 1643 break; 1644 1645 case SO_REUSEPORT: 1646 v.val = sk->sk_reuseport; 1647 break; 1648 1649 case SO_KEEPALIVE: 1650 v.val = sock_flag(sk, SOCK_KEEPOPEN); 1651 break; 1652 1653 case SO_TYPE: 1654 v.val = sk->sk_type; 1655 break; 1656 1657 case SO_PROTOCOL: 1658 v.val = sk->sk_protocol; 1659 break; 1660 1661 case SO_DOMAIN: 1662 v.val = sk->sk_family; 1663 break; 1664 1665 case SO_ERROR: 1666 v.val = -sock_error(sk); 1667 if (v.val == 0) 1668 v.val = xchg(&sk->sk_err_soft, 0); 1669 break; 1670 1671 case SO_OOBINLINE: 1672 v.val = sock_flag(sk, SOCK_URGINLINE); 1673 break; 1674 1675 case SO_NO_CHECK: 1676 v.val = sk->sk_no_check_tx; 1677 break; 1678 1679 case SO_PRIORITY: 1680 v.val = sk->sk_priority; 1681 break; 1682 1683 case SO_LINGER: 1684 lv = sizeof(v.ling); 1685 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER); 1686 v.ling.l_linger = sk->sk_lingertime / HZ; 1687 break; 1688 1689 case SO_BSDCOMPAT: 1690 break; 1691 1692 case SO_TIMESTAMP_OLD: 1693 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && 1694 !sock_flag(sk, SOCK_TSTAMP_NEW) && 1695 !sock_flag(sk, SOCK_RCVTSTAMPNS); 1696 break; 1697 1698 case SO_TIMESTAMPNS_OLD: 1699 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && !sock_flag(sk, SOCK_TSTAMP_NEW); 1700 break; 1701 1702 case SO_TIMESTAMP_NEW: 1703 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && sock_flag(sk, SOCK_TSTAMP_NEW); 1704 break; 1705 1706 case SO_TIMESTAMPNS_NEW: 1707 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && sock_flag(sk, SOCK_TSTAMP_NEW); 1708 break; 1709 1710 case SO_TIMESTAMPING_OLD: 1711 lv = sizeof(v.timestamping); 1712 v.timestamping.flags = sk->sk_tsflags; 1713 v.timestamping.bind_phc = sk->sk_bind_phc; 1714 break; 1715 1716 case SO_RCVTIMEO_OLD: 1717 case SO_RCVTIMEO_NEW: 1718 lv = sock_get_timeout(sk->sk_rcvtimeo, &v, SO_RCVTIMEO_OLD == optname); 1719 break; 1720 1721 case SO_SNDTIMEO_OLD: 1722 case SO_SNDTIMEO_NEW: 1723 lv = sock_get_timeout(sk->sk_sndtimeo, &v, SO_SNDTIMEO_OLD == optname); 1724 break; 1725 1726 case SO_RCVLOWAT: 1727 v.val = sk->sk_rcvlowat; 1728 break; 1729 1730 case SO_SNDLOWAT: 1731 v.val = 1; 1732 break; 1733 1734 case SO_PASSCRED: 1735 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags); 1736 break; 1737 1738 case SO_PEERCRED: 1739 { 1740 struct ucred peercred; 1741 if (len > sizeof(peercred)) 1742 len = sizeof(peercred); 1743 1744 spin_lock(&sk->sk_peer_lock); 1745 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred); 1746 spin_unlock(&sk->sk_peer_lock); 1747 1748 if (copy_to_sockptr(optval, &peercred, len)) 1749 return -EFAULT; 1750 goto lenout; 1751 } 1752 1753 case SO_PEERGROUPS: 1754 { 1755 const struct cred *cred; 1756 int ret, n; 1757 1758 cred = sk_get_peer_cred(sk); 1759 if (!cred) 1760 return -ENODATA; 1761 1762 n = cred->group_info->ngroups; 1763 if (len < n * sizeof(gid_t)) { 1764 len = n * sizeof(gid_t); 1765 put_cred(cred); 1766 return copy_to_sockptr(optlen, &len, sizeof(int)) ? -EFAULT : -ERANGE; 1767 } 1768 len = n * sizeof(gid_t); 1769 1770 ret = groups_to_user(optval, cred->group_info); 1771 put_cred(cred); 1772 if (ret) 1773 return ret; 1774 goto lenout; 1775 } 1776 1777 case SO_PEERNAME: 1778 { 1779 char address[128]; 1780 1781 lv = sock->ops->getname(sock, (struct sockaddr *)address, 2); 1782 if (lv < 0) 1783 return -ENOTCONN; 1784 if (lv < len) 1785 return -EINVAL; 1786 if (copy_to_sockptr(optval, address, len)) 1787 return -EFAULT; 1788 goto lenout; 1789 } 1790 1791 /* Dubious BSD thing... Probably nobody even uses it, but 1792 * the UNIX standard wants it for whatever reason... -DaveM 1793 */ 1794 case SO_ACCEPTCONN: 1795 v.val = sk->sk_state == TCP_LISTEN; 1796 break; 1797 1798 case SO_PASSSEC: 1799 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags); 1800 break; 1801 1802 case SO_PEERSEC: 1803 return security_socket_getpeersec_stream(sock, 1804 optval, optlen, len); 1805 1806 case SO_MARK: 1807 v.val = sk->sk_mark; 1808 break; 1809 1810 case SO_RCVMARK: 1811 v.val = sock_flag(sk, SOCK_RCVMARK); 1812 break; 1813 1814 case SO_RXQ_OVFL: 1815 v.val = sock_flag(sk, SOCK_RXQ_OVFL); 1816 break; 1817 1818 case SO_WIFI_STATUS: 1819 v.val = sock_flag(sk, SOCK_WIFI_STATUS); 1820 break; 1821 1822 case SO_PEEK_OFF: 1823 if (!sock->ops->set_peek_off) 1824 return -EOPNOTSUPP; 1825 1826 v.val = sk->sk_peek_off; 1827 break; 1828 case SO_NOFCS: 1829 v.val = sock_flag(sk, SOCK_NOFCS); 1830 break; 1831 1832 case SO_BINDTODEVICE: 1833 return sock_getbindtodevice(sk, optval, optlen, len); 1834 1835 case SO_GET_FILTER: 1836 len = sk_get_filter(sk, optval, len); 1837 if (len < 0) 1838 return len; 1839 1840 goto lenout; 1841 1842 case SO_LOCK_FILTER: 1843 v.val = sock_flag(sk, SOCK_FILTER_LOCKED); 1844 break; 1845 1846 case SO_BPF_EXTENSIONS: 1847 v.val = bpf_tell_extensions(); 1848 break; 1849 1850 case SO_SELECT_ERR_QUEUE: 1851 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE); 1852 break; 1853 1854 #ifdef CONFIG_NET_RX_BUSY_POLL 1855 case SO_BUSY_POLL: 1856 v.val = sk->sk_ll_usec; 1857 break; 1858 case SO_PREFER_BUSY_POLL: 1859 v.val = READ_ONCE(sk->sk_prefer_busy_poll); 1860 break; 1861 #endif 1862 1863 case SO_MAX_PACING_RATE: 1864 if (sizeof(v.ulval) != sizeof(v.val) && len >= sizeof(v.ulval)) { 1865 lv = sizeof(v.ulval); 1866 v.ulval = sk->sk_max_pacing_rate; 1867 } else { 1868 /* 32bit version */ 1869 v.val = min_t(unsigned long, sk->sk_max_pacing_rate, ~0U); 1870 } 1871 break; 1872 1873 case SO_INCOMING_CPU: 1874 v.val = READ_ONCE(sk->sk_incoming_cpu); 1875 break; 1876 1877 case SO_MEMINFO: 1878 { 1879 u32 meminfo[SK_MEMINFO_VARS]; 1880 1881 sk_get_meminfo(sk, meminfo); 1882 1883 len = min_t(unsigned int, len, sizeof(meminfo)); 1884 if (copy_to_sockptr(optval, &meminfo, len)) 1885 return -EFAULT; 1886 1887 goto lenout; 1888 } 1889 1890 #ifdef CONFIG_NET_RX_BUSY_POLL 1891 case SO_INCOMING_NAPI_ID: 1892 v.val = READ_ONCE(sk->sk_napi_id); 1893 1894 /* aggregate non-NAPI IDs down to 0 */ 1895 if (v.val < MIN_NAPI_ID) 1896 v.val = 0; 1897 1898 break; 1899 #endif 1900 1901 case SO_COOKIE: 1902 lv = sizeof(u64); 1903 if (len < lv) 1904 return -EINVAL; 1905 v.val64 = sock_gen_cookie(sk); 1906 break; 1907 1908 case SO_ZEROCOPY: 1909 v.val = sock_flag(sk, SOCK_ZEROCOPY); 1910 break; 1911 1912 case SO_TXTIME: 1913 lv = sizeof(v.txtime); 1914 v.txtime.clockid = sk->sk_clockid; 1915 v.txtime.flags |= sk->sk_txtime_deadline_mode ? 1916 SOF_TXTIME_DEADLINE_MODE : 0; 1917 v.txtime.flags |= sk->sk_txtime_report_errors ? 1918 SOF_TXTIME_REPORT_ERRORS : 0; 1919 break; 1920 1921 case SO_BINDTOIFINDEX: 1922 v.val = READ_ONCE(sk->sk_bound_dev_if); 1923 break; 1924 1925 case SO_NETNS_COOKIE: 1926 lv = sizeof(u64); 1927 if (len != lv) 1928 return -EINVAL; 1929 v.val64 = sock_net(sk)->net_cookie; 1930 break; 1931 1932 case SO_BUF_LOCK: 1933 v.val = sk->sk_userlocks & SOCK_BUF_LOCK_MASK; 1934 break; 1935 1936 case SO_RESERVE_MEM: 1937 v.val = sk->sk_reserved_mem; 1938 break; 1939 1940 case SO_TXREHASH: 1941 v.val = sk->sk_txrehash; 1942 break; 1943 1944 default: 1945 /* We implement the SO_SNDLOWAT etc to not be settable 1946 * (1003.1g 7). 1947 */ 1948 return -ENOPROTOOPT; 1949 } 1950 1951 if (len > lv) 1952 len = lv; 1953 if (copy_to_sockptr(optval, &v, len)) 1954 return -EFAULT; 1955 lenout: 1956 if (copy_to_sockptr(optlen, &len, sizeof(int))) 1957 return -EFAULT; 1958 return 0; 1959 } 1960 1961 int sock_getsockopt(struct socket *sock, int level, int optname, 1962 char __user *optval, int __user *optlen) 1963 { 1964 return sk_getsockopt(sock->sk, level, optname, 1965 USER_SOCKPTR(optval), 1966 USER_SOCKPTR(optlen)); 1967 } 1968 1969 /* 1970 * Initialize an sk_lock. 1971 * 1972 * (We also register the sk_lock with the lock validator.) 1973 */ 1974 static inline void sock_lock_init(struct sock *sk) 1975 { 1976 if (sk->sk_kern_sock) 1977 sock_lock_init_class_and_name( 1978 sk, 1979 af_family_kern_slock_key_strings[sk->sk_family], 1980 af_family_kern_slock_keys + sk->sk_family, 1981 af_family_kern_key_strings[sk->sk_family], 1982 af_family_kern_keys + sk->sk_family); 1983 else 1984 sock_lock_init_class_and_name( 1985 sk, 1986 af_family_slock_key_strings[sk->sk_family], 1987 af_family_slock_keys + sk->sk_family, 1988 af_family_key_strings[sk->sk_family], 1989 af_family_keys + sk->sk_family); 1990 } 1991 1992 /* 1993 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet, 1994 * even temporarly, because of RCU lookups. sk_node should also be left as is. 1995 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end 1996 */ 1997 static void sock_copy(struct sock *nsk, const struct sock *osk) 1998 { 1999 const struct proto *prot = READ_ONCE(osk->sk_prot); 2000 #ifdef CONFIG_SECURITY_NETWORK 2001 void *sptr = nsk->sk_security; 2002 #endif 2003 2004 /* If we move sk_tx_queue_mapping out of the private section, 2005 * we must check if sk_tx_queue_clear() is called after 2006 * sock_copy() in sk_clone_lock(). 2007 */ 2008 BUILD_BUG_ON(offsetof(struct sock, sk_tx_queue_mapping) < 2009 offsetof(struct sock, sk_dontcopy_begin) || 2010 offsetof(struct sock, sk_tx_queue_mapping) >= 2011 offsetof(struct sock, sk_dontcopy_end)); 2012 2013 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin)); 2014 2015 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end, 2016 prot->obj_size - offsetof(struct sock, sk_dontcopy_end)); 2017 2018 #ifdef CONFIG_SECURITY_NETWORK 2019 nsk->sk_security = sptr; 2020 security_sk_clone(osk, nsk); 2021 #endif 2022 } 2023 2024 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority, 2025 int family) 2026 { 2027 struct sock *sk; 2028 struct kmem_cache *slab; 2029 2030 slab = prot->slab; 2031 if (slab != NULL) { 2032 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO); 2033 if (!sk) 2034 return sk; 2035 if (want_init_on_alloc(priority)) 2036 sk_prot_clear_nulls(sk, prot->obj_size); 2037 } else 2038 sk = kmalloc(prot->obj_size, priority); 2039 2040 if (sk != NULL) { 2041 if (security_sk_alloc(sk, family, priority)) 2042 goto out_free; 2043 2044 if (!try_module_get(prot->owner)) 2045 goto out_free_sec; 2046 } 2047 2048 return sk; 2049 2050 out_free_sec: 2051 security_sk_free(sk); 2052 out_free: 2053 if (slab != NULL) 2054 kmem_cache_free(slab, sk); 2055 else 2056 kfree(sk); 2057 return NULL; 2058 } 2059 2060 static void sk_prot_free(struct proto *prot, struct sock *sk) 2061 { 2062 struct kmem_cache *slab; 2063 struct module *owner; 2064 2065 owner = prot->owner; 2066 slab = prot->slab; 2067 2068 cgroup_sk_free(&sk->sk_cgrp_data); 2069 mem_cgroup_sk_free(sk); 2070 security_sk_free(sk); 2071 if (slab != NULL) 2072 kmem_cache_free(slab, sk); 2073 else 2074 kfree(sk); 2075 module_put(owner); 2076 } 2077 2078 /** 2079 * sk_alloc - All socket objects are allocated here 2080 * @net: the applicable net namespace 2081 * @family: protocol family 2082 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 2083 * @prot: struct proto associated with this new sock instance 2084 * @kern: is this to be a kernel socket? 2085 */ 2086 struct sock *sk_alloc(struct net *net, int family, gfp_t priority, 2087 struct proto *prot, int kern) 2088 { 2089 struct sock *sk; 2090 2091 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family); 2092 if (sk) { 2093 sk->sk_family = family; 2094 /* 2095 * See comment in struct sock definition to understand 2096 * why we need sk_prot_creator -acme 2097 */ 2098 sk->sk_prot = sk->sk_prot_creator = prot; 2099 sk->sk_kern_sock = kern; 2100 sock_lock_init(sk); 2101 sk->sk_net_refcnt = kern ? 0 : 1; 2102 if (likely(sk->sk_net_refcnt)) { 2103 get_net_track(net, &sk->ns_tracker, priority); 2104 sock_inuse_add(net, 1); 2105 } else { 2106 __netns_tracker_alloc(net, &sk->ns_tracker, 2107 false, priority); 2108 } 2109 2110 sock_net_set(sk, net); 2111 refcount_set(&sk->sk_wmem_alloc, 1); 2112 2113 mem_cgroup_sk_alloc(sk); 2114 cgroup_sk_alloc(&sk->sk_cgrp_data); 2115 sock_update_classid(&sk->sk_cgrp_data); 2116 sock_update_netprioidx(&sk->sk_cgrp_data); 2117 sk_tx_queue_clear(sk); 2118 } 2119 2120 return sk; 2121 } 2122 EXPORT_SYMBOL(sk_alloc); 2123 2124 /* Sockets having SOCK_RCU_FREE will call this function after one RCU 2125 * grace period. This is the case for UDP sockets and TCP listeners. 2126 */ 2127 static void __sk_destruct(struct rcu_head *head) 2128 { 2129 struct sock *sk = container_of(head, struct sock, sk_rcu); 2130 struct sk_filter *filter; 2131 2132 if (sk->sk_destruct) 2133 sk->sk_destruct(sk); 2134 2135 filter = rcu_dereference_check(sk->sk_filter, 2136 refcount_read(&sk->sk_wmem_alloc) == 0); 2137 if (filter) { 2138 sk_filter_uncharge(sk, filter); 2139 RCU_INIT_POINTER(sk->sk_filter, NULL); 2140 } 2141 2142 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP); 2143 2144 #ifdef CONFIG_BPF_SYSCALL 2145 bpf_sk_storage_free(sk); 2146 #endif 2147 2148 if (atomic_read(&sk->sk_omem_alloc)) 2149 pr_debug("%s: optmem leakage (%d bytes) detected\n", 2150 __func__, atomic_read(&sk->sk_omem_alloc)); 2151 2152 if (sk->sk_frag.page) { 2153 put_page(sk->sk_frag.page); 2154 sk->sk_frag.page = NULL; 2155 } 2156 2157 /* We do not need to acquire sk->sk_peer_lock, we are the last user. */ 2158 put_cred(sk->sk_peer_cred); 2159 put_pid(sk->sk_peer_pid); 2160 2161 if (likely(sk->sk_net_refcnt)) 2162 put_net_track(sock_net(sk), &sk->ns_tracker); 2163 else 2164 __netns_tracker_free(sock_net(sk), &sk->ns_tracker, false); 2165 2166 sk_prot_free(sk->sk_prot_creator, sk); 2167 } 2168 2169 void sk_destruct(struct sock *sk) 2170 { 2171 bool use_call_rcu = sock_flag(sk, SOCK_RCU_FREE); 2172 2173 if (rcu_access_pointer(sk->sk_reuseport_cb)) { 2174 reuseport_detach_sock(sk); 2175 use_call_rcu = true; 2176 } 2177 2178 if (use_call_rcu) 2179 call_rcu(&sk->sk_rcu, __sk_destruct); 2180 else 2181 __sk_destruct(&sk->sk_rcu); 2182 } 2183 2184 static void __sk_free(struct sock *sk) 2185 { 2186 if (likely(sk->sk_net_refcnt)) 2187 sock_inuse_add(sock_net(sk), -1); 2188 2189 if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk))) 2190 sock_diag_broadcast_destroy(sk); 2191 else 2192 sk_destruct(sk); 2193 } 2194 2195 void sk_free(struct sock *sk) 2196 { 2197 /* 2198 * We subtract one from sk_wmem_alloc and can know if 2199 * some packets are still in some tx queue. 2200 * If not null, sock_wfree() will call __sk_free(sk) later 2201 */ 2202 if (refcount_dec_and_test(&sk->sk_wmem_alloc)) 2203 __sk_free(sk); 2204 } 2205 EXPORT_SYMBOL(sk_free); 2206 2207 static void sk_init_common(struct sock *sk) 2208 { 2209 skb_queue_head_init(&sk->sk_receive_queue); 2210 skb_queue_head_init(&sk->sk_write_queue); 2211 skb_queue_head_init(&sk->sk_error_queue); 2212 2213 rwlock_init(&sk->sk_callback_lock); 2214 lockdep_set_class_and_name(&sk->sk_receive_queue.lock, 2215 af_rlock_keys + sk->sk_family, 2216 af_family_rlock_key_strings[sk->sk_family]); 2217 lockdep_set_class_and_name(&sk->sk_write_queue.lock, 2218 af_wlock_keys + sk->sk_family, 2219 af_family_wlock_key_strings[sk->sk_family]); 2220 lockdep_set_class_and_name(&sk->sk_error_queue.lock, 2221 af_elock_keys + sk->sk_family, 2222 af_family_elock_key_strings[sk->sk_family]); 2223 lockdep_set_class_and_name(&sk->sk_callback_lock, 2224 af_callback_keys + sk->sk_family, 2225 af_family_clock_key_strings[sk->sk_family]); 2226 } 2227 2228 /** 2229 * sk_clone_lock - clone a socket, and lock its clone 2230 * @sk: the socket to clone 2231 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 2232 * 2233 * Caller must unlock socket even in error path (bh_unlock_sock(newsk)) 2234 */ 2235 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority) 2236 { 2237 struct proto *prot = READ_ONCE(sk->sk_prot); 2238 struct sk_filter *filter; 2239 bool is_charged = true; 2240 struct sock *newsk; 2241 2242 newsk = sk_prot_alloc(prot, priority, sk->sk_family); 2243 if (!newsk) 2244 goto out; 2245 2246 sock_copy(newsk, sk); 2247 2248 newsk->sk_prot_creator = prot; 2249 2250 /* SANITY */ 2251 if (likely(newsk->sk_net_refcnt)) { 2252 get_net_track(sock_net(newsk), &newsk->ns_tracker, priority); 2253 sock_inuse_add(sock_net(newsk), 1); 2254 } else { 2255 /* Kernel sockets are not elevating the struct net refcount. 2256 * Instead, use a tracker to more easily detect if a layer 2257 * is not properly dismantling its kernel sockets at netns 2258 * destroy time. 2259 */ 2260 __netns_tracker_alloc(sock_net(newsk), &newsk->ns_tracker, 2261 false, priority); 2262 } 2263 sk_node_init(&newsk->sk_node); 2264 sock_lock_init(newsk); 2265 bh_lock_sock(newsk); 2266 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL; 2267 newsk->sk_backlog.len = 0; 2268 2269 atomic_set(&newsk->sk_rmem_alloc, 0); 2270 2271 /* sk_wmem_alloc set to one (see sk_free() and sock_wfree()) */ 2272 refcount_set(&newsk->sk_wmem_alloc, 1); 2273 2274 atomic_set(&newsk->sk_omem_alloc, 0); 2275 sk_init_common(newsk); 2276 2277 newsk->sk_dst_cache = NULL; 2278 newsk->sk_dst_pending_confirm = 0; 2279 newsk->sk_wmem_queued = 0; 2280 newsk->sk_forward_alloc = 0; 2281 newsk->sk_reserved_mem = 0; 2282 atomic_set(&newsk->sk_drops, 0); 2283 newsk->sk_send_head = NULL; 2284 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK; 2285 atomic_set(&newsk->sk_zckey, 0); 2286 2287 sock_reset_flag(newsk, SOCK_DONE); 2288 2289 /* sk->sk_memcg will be populated at accept() time */ 2290 newsk->sk_memcg = NULL; 2291 2292 cgroup_sk_clone(&newsk->sk_cgrp_data); 2293 2294 rcu_read_lock(); 2295 filter = rcu_dereference(sk->sk_filter); 2296 if (filter != NULL) 2297 /* though it's an empty new sock, the charging may fail 2298 * if sysctl_optmem_max was changed between creation of 2299 * original socket and cloning 2300 */ 2301 is_charged = sk_filter_charge(newsk, filter); 2302 RCU_INIT_POINTER(newsk->sk_filter, filter); 2303 rcu_read_unlock(); 2304 2305 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) { 2306 /* We need to make sure that we don't uncharge the new 2307 * socket if we couldn't charge it in the first place 2308 * as otherwise we uncharge the parent's filter. 2309 */ 2310 if (!is_charged) 2311 RCU_INIT_POINTER(newsk->sk_filter, NULL); 2312 sk_free_unlock_clone(newsk); 2313 newsk = NULL; 2314 goto out; 2315 } 2316 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL); 2317 2318 if (bpf_sk_storage_clone(sk, newsk)) { 2319 sk_free_unlock_clone(newsk); 2320 newsk = NULL; 2321 goto out; 2322 } 2323 2324 /* Clear sk_user_data if parent had the pointer tagged 2325 * as not suitable for copying when cloning. 2326 */ 2327 if (sk_user_data_is_nocopy(newsk)) 2328 newsk->sk_user_data = NULL; 2329 2330 newsk->sk_err = 0; 2331 newsk->sk_err_soft = 0; 2332 newsk->sk_priority = 0; 2333 newsk->sk_incoming_cpu = raw_smp_processor_id(); 2334 2335 /* Before updating sk_refcnt, we must commit prior changes to memory 2336 * (Documentation/RCU/rculist_nulls.rst for details) 2337 */ 2338 smp_wmb(); 2339 refcount_set(&newsk->sk_refcnt, 2); 2340 2341 /* Increment the counter in the same struct proto as the master 2342 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that 2343 * is the same as sk->sk_prot->socks, as this field was copied 2344 * with memcpy). 2345 * 2346 * This _changes_ the previous behaviour, where 2347 * tcp_create_openreq_child always was incrementing the 2348 * equivalent to tcp_prot->socks (inet_sock_nr), so this have 2349 * to be taken into account in all callers. -acme 2350 */ 2351 sk_refcnt_debug_inc(newsk); 2352 sk_set_socket(newsk, NULL); 2353 sk_tx_queue_clear(newsk); 2354 RCU_INIT_POINTER(newsk->sk_wq, NULL); 2355 2356 if (newsk->sk_prot->sockets_allocated) 2357 sk_sockets_allocated_inc(newsk); 2358 2359 if (sock_needs_netstamp(sk) && newsk->sk_flags & SK_FLAGS_TIMESTAMP) 2360 net_enable_timestamp(); 2361 out: 2362 return newsk; 2363 } 2364 EXPORT_SYMBOL_GPL(sk_clone_lock); 2365 2366 void sk_free_unlock_clone(struct sock *sk) 2367 { 2368 /* It is still raw copy of parent, so invalidate 2369 * destructor and make plain sk_free() */ 2370 sk->sk_destruct = NULL; 2371 bh_unlock_sock(sk); 2372 sk_free(sk); 2373 } 2374 EXPORT_SYMBOL_GPL(sk_free_unlock_clone); 2375 2376 static void sk_trim_gso_size(struct sock *sk) 2377 { 2378 if (sk->sk_gso_max_size <= GSO_LEGACY_MAX_SIZE) 2379 return; 2380 #if IS_ENABLED(CONFIG_IPV6) 2381 if (sk->sk_family == AF_INET6 && 2382 sk_is_tcp(sk) && 2383 !ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr)) 2384 return; 2385 #endif 2386 sk->sk_gso_max_size = GSO_LEGACY_MAX_SIZE; 2387 } 2388 2389 void sk_setup_caps(struct sock *sk, struct dst_entry *dst) 2390 { 2391 u32 max_segs = 1; 2392 2393 sk_dst_set(sk, dst); 2394 sk->sk_route_caps = dst->dev->features; 2395 if (sk_is_tcp(sk)) 2396 sk->sk_route_caps |= NETIF_F_GSO; 2397 if (sk->sk_route_caps & NETIF_F_GSO) 2398 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE; 2399 if (unlikely(sk->sk_gso_disabled)) 2400 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 2401 if (sk_can_gso(sk)) { 2402 if (dst->header_len && !xfrm_dst_offload_ok(dst)) { 2403 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 2404 } else { 2405 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM; 2406 /* pairs with the WRITE_ONCE() in netif_set_gso_max_size() */ 2407 sk->sk_gso_max_size = READ_ONCE(dst->dev->gso_max_size); 2408 sk_trim_gso_size(sk); 2409 sk->sk_gso_max_size -= (MAX_TCP_HEADER + 1); 2410 /* pairs with the WRITE_ONCE() in netif_set_gso_max_segs() */ 2411 max_segs = max_t(u32, READ_ONCE(dst->dev->gso_max_segs), 1); 2412 } 2413 } 2414 sk->sk_gso_max_segs = max_segs; 2415 } 2416 EXPORT_SYMBOL_GPL(sk_setup_caps); 2417 2418 /* 2419 * Simple resource managers for sockets. 2420 */ 2421 2422 2423 /* 2424 * Write buffer destructor automatically called from kfree_skb. 2425 */ 2426 void sock_wfree(struct sk_buff *skb) 2427 { 2428 struct sock *sk = skb->sk; 2429 unsigned int len = skb->truesize; 2430 bool free; 2431 2432 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) { 2433 if (sock_flag(sk, SOCK_RCU_FREE) && 2434 sk->sk_write_space == sock_def_write_space) { 2435 rcu_read_lock(); 2436 free = refcount_sub_and_test(len, &sk->sk_wmem_alloc); 2437 sock_def_write_space_wfree(sk); 2438 rcu_read_unlock(); 2439 if (unlikely(free)) 2440 __sk_free(sk); 2441 return; 2442 } 2443 2444 /* 2445 * Keep a reference on sk_wmem_alloc, this will be released 2446 * after sk_write_space() call 2447 */ 2448 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc)); 2449 sk->sk_write_space(sk); 2450 len = 1; 2451 } 2452 /* 2453 * if sk_wmem_alloc reaches 0, we must finish what sk_free() 2454 * could not do because of in-flight packets 2455 */ 2456 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc)) 2457 __sk_free(sk); 2458 } 2459 EXPORT_SYMBOL(sock_wfree); 2460 2461 /* This variant of sock_wfree() is used by TCP, 2462 * since it sets SOCK_USE_WRITE_QUEUE. 2463 */ 2464 void __sock_wfree(struct sk_buff *skb) 2465 { 2466 struct sock *sk = skb->sk; 2467 2468 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc)) 2469 __sk_free(sk); 2470 } 2471 2472 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk) 2473 { 2474 skb_orphan(skb); 2475 skb->sk = sk; 2476 #ifdef CONFIG_INET 2477 if (unlikely(!sk_fullsock(sk))) { 2478 skb->destructor = sock_edemux; 2479 sock_hold(sk); 2480 return; 2481 } 2482 #endif 2483 skb->destructor = sock_wfree; 2484 skb_set_hash_from_sk(skb, sk); 2485 /* 2486 * We used to take a refcount on sk, but following operation 2487 * is enough to guarantee sk_free() wont free this sock until 2488 * all in-flight packets are completed 2489 */ 2490 refcount_add(skb->truesize, &sk->sk_wmem_alloc); 2491 } 2492 EXPORT_SYMBOL(skb_set_owner_w); 2493 2494 static bool can_skb_orphan_partial(const struct sk_buff *skb) 2495 { 2496 #ifdef CONFIG_TLS_DEVICE 2497 /* Drivers depend on in-order delivery for crypto offload, 2498 * partial orphan breaks out-of-order-OK logic. 2499 */ 2500 if (skb->decrypted) 2501 return false; 2502 #endif 2503 return (skb->destructor == sock_wfree || 2504 (IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree)); 2505 } 2506 2507 /* This helper is used by netem, as it can hold packets in its 2508 * delay queue. We want to allow the owner socket to send more 2509 * packets, as if they were already TX completed by a typical driver. 2510 * But we also want to keep skb->sk set because some packet schedulers 2511 * rely on it (sch_fq for example). 2512 */ 2513 void skb_orphan_partial(struct sk_buff *skb) 2514 { 2515 if (skb_is_tcp_pure_ack(skb)) 2516 return; 2517 2518 if (can_skb_orphan_partial(skb) && skb_set_owner_sk_safe(skb, skb->sk)) 2519 return; 2520 2521 skb_orphan(skb); 2522 } 2523 EXPORT_SYMBOL(skb_orphan_partial); 2524 2525 /* 2526 * Read buffer destructor automatically called from kfree_skb. 2527 */ 2528 void sock_rfree(struct sk_buff *skb) 2529 { 2530 struct sock *sk = skb->sk; 2531 unsigned int len = skb->truesize; 2532 2533 atomic_sub(len, &sk->sk_rmem_alloc); 2534 sk_mem_uncharge(sk, len); 2535 } 2536 EXPORT_SYMBOL(sock_rfree); 2537 2538 /* 2539 * Buffer destructor for skbs that are not used directly in read or write 2540 * path, e.g. for error handler skbs. Automatically called from kfree_skb. 2541 */ 2542 void sock_efree(struct sk_buff *skb) 2543 { 2544 sock_put(skb->sk); 2545 } 2546 EXPORT_SYMBOL(sock_efree); 2547 2548 /* Buffer destructor for prefetch/receive path where reference count may 2549 * not be held, e.g. for listen sockets. 2550 */ 2551 #ifdef CONFIG_INET 2552 void sock_pfree(struct sk_buff *skb) 2553 { 2554 if (sk_is_refcounted(skb->sk)) 2555 sock_gen_put(skb->sk); 2556 } 2557 EXPORT_SYMBOL(sock_pfree); 2558 #endif /* CONFIG_INET */ 2559 2560 kuid_t sock_i_uid(struct sock *sk) 2561 { 2562 kuid_t uid; 2563 2564 read_lock_bh(&sk->sk_callback_lock); 2565 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID; 2566 read_unlock_bh(&sk->sk_callback_lock); 2567 return uid; 2568 } 2569 EXPORT_SYMBOL(sock_i_uid); 2570 2571 unsigned long sock_i_ino(struct sock *sk) 2572 { 2573 unsigned long ino; 2574 2575 read_lock_bh(&sk->sk_callback_lock); 2576 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0; 2577 read_unlock_bh(&sk->sk_callback_lock); 2578 return ino; 2579 } 2580 EXPORT_SYMBOL(sock_i_ino); 2581 2582 /* 2583 * Allocate a skb from the socket's send buffer. 2584 */ 2585 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 2586 gfp_t priority) 2587 { 2588 if (force || 2589 refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) { 2590 struct sk_buff *skb = alloc_skb(size, priority); 2591 2592 if (skb) { 2593 skb_set_owner_w(skb, sk); 2594 return skb; 2595 } 2596 } 2597 return NULL; 2598 } 2599 EXPORT_SYMBOL(sock_wmalloc); 2600 2601 static void sock_ofree(struct sk_buff *skb) 2602 { 2603 struct sock *sk = skb->sk; 2604 2605 atomic_sub(skb->truesize, &sk->sk_omem_alloc); 2606 } 2607 2608 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size, 2609 gfp_t priority) 2610 { 2611 struct sk_buff *skb; 2612 2613 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */ 2614 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) > 2615 READ_ONCE(sysctl_optmem_max)) 2616 return NULL; 2617 2618 skb = alloc_skb(size, priority); 2619 if (!skb) 2620 return NULL; 2621 2622 atomic_add(skb->truesize, &sk->sk_omem_alloc); 2623 skb->sk = sk; 2624 skb->destructor = sock_ofree; 2625 return skb; 2626 } 2627 2628 /* 2629 * Allocate a memory block from the socket's option memory buffer. 2630 */ 2631 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority) 2632 { 2633 int optmem_max = READ_ONCE(sysctl_optmem_max); 2634 2635 if ((unsigned int)size <= optmem_max && 2636 atomic_read(&sk->sk_omem_alloc) + size < optmem_max) { 2637 void *mem; 2638 /* First do the add, to avoid the race if kmalloc 2639 * might sleep. 2640 */ 2641 atomic_add(size, &sk->sk_omem_alloc); 2642 mem = kmalloc(size, priority); 2643 if (mem) 2644 return mem; 2645 atomic_sub(size, &sk->sk_omem_alloc); 2646 } 2647 return NULL; 2648 } 2649 EXPORT_SYMBOL(sock_kmalloc); 2650 2651 /* Free an option memory block. Note, we actually want the inline 2652 * here as this allows gcc to detect the nullify and fold away the 2653 * condition entirely. 2654 */ 2655 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size, 2656 const bool nullify) 2657 { 2658 if (WARN_ON_ONCE(!mem)) 2659 return; 2660 if (nullify) 2661 kfree_sensitive(mem); 2662 else 2663 kfree(mem); 2664 atomic_sub(size, &sk->sk_omem_alloc); 2665 } 2666 2667 void sock_kfree_s(struct sock *sk, void *mem, int size) 2668 { 2669 __sock_kfree_s(sk, mem, size, false); 2670 } 2671 EXPORT_SYMBOL(sock_kfree_s); 2672 2673 void sock_kzfree_s(struct sock *sk, void *mem, int size) 2674 { 2675 __sock_kfree_s(sk, mem, size, true); 2676 } 2677 EXPORT_SYMBOL(sock_kzfree_s); 2678 2679 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock. 2680 I think, these locks should be removed for datagram sockets. 2681 */ 2682 static long sock_wait_for_wmem(struct sock *sk, long timeo) 2683 { 2684 DEFINE_WAIT(wait); 2685 2686 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk); 2687 for (;;) { 2688 if (!timeo) 2689 break; 2690 if (signal_pending(current)) 2691 break; 2692 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 2693 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 2694 if (refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) 2695 break; 2696 if (sk->sk_shutdown & SEND_SHUTDOWN) 2697 break; 2698 if (sk->sk_err) 2699 break; 2700 timeo = schedule_timeout(timeo); 2701 } 2702 finish_wait(sk_sleep(sk), &wait); 2703 return timeo; 2704 } 2705 2706 2707 /* 2708 * Generic send/receive buffer handlers 2709 */ 2710 2711 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 2712 unsigned long data_len, int noblock, 2713 int *errcode, int max_page_order) 2714 { 2715 struct sk_buff *skb; 2716 long timeo; 2717 int err; 2718 2719 timeo = sock_sndtimeo(sk, noblock); 2720 for (;;) { 2721 err = sock_error(sk); 2722 if (err != 0) 2723 goto failure; 2724 2725 err = -EPIPE; 2726 if (sk->sk_shutdown & SEND_SHUTDOWN) 2727 goto failure; 2728 2729 if (sk_wmem_alloc_get(sk) < READ_ONCE(sk->sk_sndbuf)) 2730 break; 2731 2732 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk); 2733 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 2734 err = -EAGAIN; 2735 if (!timeo) 2736 goto failure; 2737 if (signal_pending(current)) 2738 goto interrupted; 2739 timeo = sock_wait_for_wmem(sk, timeo); 2740 } 2741 skb = alloc_skb_with_frags(header_len, data_len, max_page_order, 2742 errcode, sk->sk_allocation); 2743 if (skb) 2744 skb_set_owner_w(skb, sk); 2745 return skb; 2746 2747 interrupted: 2748 err = sock_intr_errno(timeo); 2749 failure: 2750 *errcode = err; 2751 return NULL; 2752 } 2753 EXPORT_SYMBOL(sock_alloc_send_pskb); 2754 2755 int __sock_cmsg_send(struct sock *sk, struct cmsghdr *cmsg, 2756 struct sockcm_cookie *sockc) 2757 { 2758 u32 tsflags; 2759 2760 switch (cmsg->cmsg_type) { 2761 case SO_MARK: 2762 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) && 2763 !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) 2764 return -EPERM; 2765 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 2766 return -EINVAL; 2767 sockc->mark = *(u32 *)CMSG_DATA(cmsg); 2768 break; 2769 case SO_TIMESTAMPING_OLD: 2770 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 2771 return -EINVAL; 2772 2773 tsflags = *(u32 *)CMSG_DATA(cmsg); 2774 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK) 2775 return -EINVAL; 2776 2777 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK; 2778 sockc->tsflags |= tsflags; 2779 break; 2780 case SCM_TXTIME: 2781 if (!sock_flag(sk, SOCK_TXTIME)) 2782 return -EINVAL; 2783 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64))) 2784 return -EINVAL; 2785 sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg)); 2786 break; 2787 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */ 2788 case SCM_RIGHTS: 2789 case SCM_CREDENTIALS: 2790 break; 2791 default: 2792 return -EINVAL; 2793 } 2794 return 0; 2795 } 2796 EXPORT_SYMBOL(__sock_cmsg_send); 2797 2798 int sock_cmsg_send(struct sock *sk, struct msghdr *msg, 2799 struct sockcm_cookie *sockc) 2800 { 2801 struct cmsghdr *cmsg; 2802 int ret; 2803 2804 for_each_cmsghdr(cmsg, msg) { 2805 if (!CMSG_OK(msg, cmsg)) 2806 return -EINVAL; 2807 if (cmsg->cmsg_level != SOL_SOCKET) 2808 continue; 2809 ret = __sock_cmsg_send(sk, cmsg, sockc); 2810 if (ret) 2811 return ret; 2812 } 2813 return 0; 2814 } 2815 EXPORT_SYMBOL(sock_cmsg_send); 2816 2817 static void sk_enter_memory_pressure(struct sock *sk) 2818 { 2819 if (!sk->sk_prot->enter_memory_pressure) 2820 return; 2821 2822 sk->sk_prot->enter_memory_pressure(sk); 2823 } 2824 2825 static void sk_leave_memory_pressure(struct sock *sk) 2826 { 2827 if (sk->sk_prot->leave_memory_pressure) { 2828 sk->sk_prot->leave_memory_pressure(sk); 2829 } else { 2830 unsigned long *memory_pressure = sk->sk_prot->memory_pressure; 2831 2832 if (memory_pressure && READ_ONCE(*memory_pressure)) 2833 WRITE_ONCE(*memory_pressure, 0); 2834 } 2835 } 2836 2837 DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key); 2838 2839 /** 2840 * skb_page_frag_refill - check that a page_frag contains enough room 2841 * @sz: minimum size of the fragment we want to get 2842 * @pfrag: pointer to page_frag 2843 * @gfp: priority for memory allocation 2844 * 2845 * Note: While this allocator tries to use high order pages, there is 2846 * no guarantee that allocations succeed. Therefore, @sz MUST be 2847 * less or equal than PAGE_SIZE. 2848 */ 2849 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp) 2850 { 2851 if (pfrag->page) { 2852 if (page_ref_count(pfrag->page) == 1) { 2853 pfrag->offset = 0; 2854 return true; 2855 } 2856 if (pfrag->offset + sz <= pfrag->size) 2857 return true; 2858 put_page(pfrag->page); 2859 } 2860 2861 pfrag->offset = 0; 2862 if (SKB_FRAG_PAGE_ORDER && 2863 !static_branch_unlikely(&net_high_order_alloc_disable_key)) { 2864 /* Avoid direct reclaim but allow kswapd to wake */ 2865 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) | 2866 __GFP_COMP | __GFP_NOWARN | 2867 __GFP_NORETRY, 2868 SKB_FRAG_PAGE_ORDER); 2869 if (likely(pfrag->page)) { 2870 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER; 2871 return true; 2872 } 2873 } 2874 pfrag->page = alloc_page(gfp); 2875 if (likely(pfrag->page)) { 2876 pfrag->size = PAGE_SIZE; 2877 return true; 2878 } 2879 return false; 2880 } 2881 EXPORT_SYMBOL(skb_page_frag_refill); 2882 2883 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag) 2884 { 2885 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation))) 2886 return true; 2887 2888 sk_enter_memory_pressure(sk); 2889 sk_stream_moderate_sndbuf(sk); 2890 return false; 2891 } 2892 EXPORT_SYMBOL(sk_page_frag_refill); 2893 2894 void __lock_sock(struct sock *sk) 2895 __releases(&sk->sk_lock.slock) 2896 __acquires(&sk->sk_lock.slock) 2897 { 2898 DEFINE_WAIT(wait); 2899 2900 for (;;) { 2901 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait, 2902 TASK_UNINTERRUPTIBLE); 2903 spin_unlock_bh(&sk->sk_lock.slock); 2904 schedule(); 2905 spin_lock_bh(&sk->sk_lock.slock); 2906 if (!sock_owned_by_user(sk)) 2907 break; 2908 } 2909 finish_wait(&sk->sk_lock.wq, &wait); 2910 } 2911 2912 void __release_sock(struct sock *sk) 2913 __releases(&sk->sk_lock.slock) 2914 __acquires(&sk->sk_lock.slock) 2915 { 2916 struct sk_buff *skb, *next; 2917 2918 while ((skb = sk->sk_backlog.head) != NULL) { 2919 sk->sk_backlog.head = sk->sk_backlog.tail = NULL; 2920 2921 spin_unlock_bh(&sk->sk_lock.slock); 2922 2923 do { 2924 next = skb->next; 2925 prefetch(next); 2926 DEBUG_NET_WARN_ON_ONCE(skb_dst_is_noref(skb)); 2927 skb_mark_not_on_list(skb); 2928 sk_backlog_rcv(sk, skb); 2929 2930 cond_resched(); 2931 2932 skb = next; 2933 } while (skb != NULL); 2934 2935 spin_lock_bh(&sk->sk_lock.slock); 2936 } 2937 2938 /* 2939 * Doing the zeroing here guarantee we can not loop forever 2940 * while a wild producer attempts to flood us. 2941 */ 2942 sk->sk_backlog.len = 0; 2943 } 2944 2945 void __sk_flush_backlog(struct sock *sk) 2946 { 2947 spin_lock_bh(&sk->sk_lock.slock); 2948 __release_sock(sk); 2949 spin_unlock_bh(&sk->sk_lock.slock); 2950 } 2951 EXPORT_SYMBOL_GPL(__sk_flush_backlog); 2952 2953 /** 2954 * sk_wait_data - wait for data to arrive at sk_receive_queue 2955 * @sk: sock to wait on 2956 * @timeo: for how long 2957 * @skb: last skb seen on sk_receive_queue 2958 * 2959 * Now socket state including sk->sk_err is changed only under lock, 2960 * hence we may omit checks after joining wait queue. 2961 * We check receive queue before schedule() only as optimization; 2962 * it is very likely that release_sock() added new data. 2963 */ 2964 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb) 2965 { 2966 DEFINE_WAIT_FUNC(wait, woken_wake_function); 2967 int rc; 2968 2969 add_wait_queue(sk_sleep(sk), &wait); 2970 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk); 2971 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait); 2972 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk); 2973 remove_wait_queue(sk_sleep(sk), &wait); 2974 return rc; 2975 } 2976 EXPORT_SYMBOL(sk_wait_data); 2977 2978 /** 2979 * __sk_mem_raise_allocated - increase memory_allocated 2980 * @sk: socket 2981 * @size: memory size to allocate 2982 * @amt: pages to allocate 2983 * @kind: allocation type 2984 * 2985 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc 2986 */ 2987 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind) 2988 { 2989 bool memcg_charge = mem_cgroup_sockets_enabled && sk->sk_memcg; 2990 struct proto *prot = sk->sk_prot; 2991 bool charged = true; 2992 long allocated; 2993 2994 sk_memory_allocated_add(sk, amt); 2995 allocated = sk_memory_allocated(sk); 2996 if (memcg_charge && 2997 !(charged = mem_cgroup_charge_skmem(sk->sk_memcg, amt, 2998 gfp_memcg_charge()))) 2999 goto suppress_allocation; 3000 3001 /* Under limit. */ 3002 if (allocated <= sk_prot_mem_limits(sk, 0)) { 3003 sk_leave_memory_pressure(sk); 3004 return 1; 3005 } 3006 3007 /* Under pressure. */ 3008 if (allocated > sk_prot_mem_limits(sk, 1)) 3009 sk_enter_memory_pressure(sk); 3010 3011 /* Over hard limit. */ 3012 if (allocated > sk_prot_mem_limits(sk, 2)) 3013 goto suppress_allocation; 3014 3015 /* guarantee minimum buffer size under pressure */ 3016 if (kind == SK_MEM_RECV) { 3017 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot)) 3018 return 1; 3019 3020 } else { /* SK_MEM_SEND */ 3021 int wmem0 = sk_get_wmem0(sk, prot); 3022 3023 if (sk->sk_type == SOCK_STREAM) { 3024 if (sk->sk_wmem_queued < wmem0) 3025 return 1; 3026 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) { 3027 return 1; 3028 } 3029 } 3030 3031 if (sk_has_memory_pressure(sk)) { 3032 u64 alloc; 3033 3034 if (!sk_under_memory_pressure(sk)) 3035 return 1; 3036 alloc = sk_sockets_allocated_read_positive(sk); 3037 if (sk_prot_mem_limits(sk, 2) > alloc * 3038 sk_mem_pages(sk->sk_wmem_queued + 3039 atomic_read(&sk->sk_rmem_alloc) + 3040 sk->sk_forward_alloc)) 3041 return 1; 3042 } 3043 3044 suppress_allocation: 3045 3046 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) { 3047 sk_stream_moderate_sndbuf(sk); 3048 3049 /* Fail only if socket is _under_ its sndbuf. 3050 * In this case we cannot block, so that we have to fail. 3051 */ 3052 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf) { 3053 /* Force charge with __GFP_NOFAIL */ 3054 if (memcg_charge && !charged) { 3055 mem_cgroup_charge_skmem(sk->sk_memcg, amt, 3056 gfp_memcg_charge() | __GFP_NOFAIL); 3057 } 3058 return 1; 3059 } 3060 } 3061 3062 if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged)) 3063 trace_sock_exceed_buf_limit(sk, prot, allocated, kind); 3064 3065 sk_memory_allocated_sub(sk, amt); 3066 3067 if (memcg_charge && charged) 3068 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt); 3069 3070 return 0; 3071 } 3072 3073 /** 3074 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated 3075 * @sk: socket 3076 * @size: memory size to allocate 3077 * @kind: allocation type 3078 * 3079 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means 3080 * rmem allocation. This function assumes that protocols which have 3081 * memory_pressure use sk_wmem_queued as write buffer accounting. 3082 */ 3083 int __sk_mem_schedule(struct sock *sk, int size, int kind) 3084 { 3085 int ret, amt = sk_mem_pages(size); 3086 3087 sk->sk_forward_alloc += amt << PAGE_SHIFT; 3088 ret = __sk_mem_raise_allocated(sk, size, amt, kind); 3089 if (!ret) 3090 sk->sk_forward_alloc -= amt << PAGE_SHIFT; 3091 return ret; 3092 } 3093 EXPORT_SYMBOL(__sk_mem_schedule); 3094 3095 /** 3096 * __sk_mem_reduce_allocated - reclaim memory_allocated 3097 * @sk: socket 3098 * @amount: number of quanta 3099 * 3100 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc 3101 */ 3102 void __sk_mem_reduce_allocated(struct sock *sk, int amount) 3103 { 3104 sk_memory_allocated_sub(sk, amount); 3105 3106 if (mem_cgroup_sockets_enabled && sk->sk_memcg) 3107 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount); 3108 3109 if (sk_under_memory_pressure(sk) && 3110 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0))) 3111 sk_leave_memory_pressure(sk); 3112 } 3113 3114 /** 3115 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated 3116 * @sk: socket 3117 * @amount: number of bytes (rounded down to a PAGE_SIZE multiple) 3118 */ 3119 void __sk_mem_reclaim(struct sock *sk, int amount) 3120 { 3121 amount >>= PAGE_SHIFT; 3122 sk->sk_forward_alloc -= amount << PAGE_SHIFT; 3123 __sk_mem_reduce_allocated(sk, amount); 3124 } 3125 EXPORT_SYMBOL(__sk_mem_reclaim); 3126 3127 int sk_set_peek_off(struct sock *sk, int val) 3128 { 3129 sk->sk_peek_off = val; 3130 return 0; 3131 } 3132 EXPORT_SYMBOL_GPL(sk_set_peek_off); 3133 3134 /* 3135 * Set of default routines for initialising struct proto_ops when 3136 * the protocol does not support a particular function. In certain 3137 * cases where it makes no sense for a protocol to have a "do nothing" 3138 * function, some default processing is provided. 3139 */ 3140 3141 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len) 3142 { 3143 return -EOPNOTSUPP; 3144 } 3145 EXPORT_SYMBOL(sock_no_bind); 3146 3147 int sock_no_connect(struct socket *sock, struct sockaddr *saddr, 3148 int len, int flags) 3149 { 3150 return -EOPNOTSUPP; 3151 } 3152 EXPORT_SYMBOL(sock_no_connect); 3153 3154 int sock_no_socketpair(struct socket *sock1, struct socket *sock2) 3155 { 3156 return -EOPNOTSUPP; 3157 } 3158 EXPORT_SYMBOL(sock_no_socketpair); 3159 3160 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags, 3161 bool kern) 3162 { 3163 return -EOPNOTSUPP; 3164 } 3165 EXPORT_SYMBOL(sock_no_accept); 3166 3167 int sock_no_getname(struct socket *sock, struct sockaddr *saddr, 3168 int peer) 3169 { 3170 return -EOPNOTSUPP; 3171 } 3172 EXPORT_SYMBOL(sock_no_getname); 3173 3174 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 3175 { 3176 return -EOPNOTSUPP; 3177 } 3178 EXPORT_SYMBOL(sock_no_ioctl); 3179 3180 int sock_no_listen(struct socket *sock, int backlog) 3181 { 3182 return -EOPNOTSUPP; 3183 } 3184 EXPORT_SYMBOL(sock_no_listen); 3185 3186 int sock_no_shutdown(struct socket *sock, int how) 3187 { 3188 return -EOPNOTSUPP; 3189 } 3190 EXPORT_SYMBOL(sock_no_shutdown); 3191 3192 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len) 3193 { 3194 return -EOPNOTSUPP; 3195 } 3196 EXPORT_SYMBOL(sock_no_sendmsg); 3197 3198 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len) 3199 { 3200 return -EOPNOTSUPP; 3201 } 3202 EXPORT_SYMBOL(sock_no_sendmsg_locked); 3203 3204 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len, 3205 int flags) 3206 { 3207 return -EOPNOTSUPP; 3208 } 3209 EXPORT_SYMBOL(sock_no_recvmsg); 3210 3211 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma) 3212 { 3213 /* Mirror missing mmap method error code */ 3214 return -ENODEV; 3215 } 3216 EXPORT_SYMBOL(sock_no_mmap); 3217 3218 /* 3219 * When a file is received (via SCM_RIGHTS, etc), we must bump the 3220 * various sock-based usage counts. 3221 */ 3222 void __receive_sock(struct file *file) 3223 { 3224 struct socket *sock; 3225 3226 sock = sock_from_file(file); 3227 if (sock) { 3228 sock_update_netprioidx(&sock->sk->sk_cgrp_data); 3229 sock_update_classid(&sock->sk->sk_cgrp_data); 3230 } 3231 } 3232 3233 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags) 3234 { 3235 ssize_t res; 3236 struct msghdr msg = {.msg_flags = flags}; 3237 struct kvec iov; 3238 char *kaddr = kmap(page); 3239 iov.iov_base = kaddr + offset; 3240 iov.iov_len = size; 3241 res = kernel_sendmsg(sock, &msg, &iov, 1, size); 3242 kunmap(page); 3243 return res; 3244 } 3245 EXPORT_SYMBOL(sock_no_sendpage); 3246 3247 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page, 3248 int offset, size_t size, int flags) 3249 { 3250 ssize_t res; 3251 struct msghdr msg = {.msg_flags = flags}; 3252 struct kvec iov; 3253 char *kaddr = kmap(page); 3254 3255 iov.iov_base = kaddr + offset; 3256 iov.iov_len = size; 3257 res = kernel_sendmsg_locked(sk, &msg, &iov, 1, size); 3258 kunmap(page); 3259 return res; 3260 } 3261 EXPORT_SYMBOL(sock_no_sendpage_locked); 3262 3263 /* 3264 * Default Socket Callbacks 3265 */ 3266 3267 static void sock_def_wakeup(struct sock *sk) 3268 { 3269 struct socket_wq *wq; 3270 3271 rcu_read_lock(); 3272 wq = rcu_dereference(sk->sk_wq); 3273 if (skwq_has_sleeper(wq)) 3274 wake_up_interruptible_all(&wq->wait); 3275 rcu_read_unlock(); 3276 } 3277 3278 static void sock_def_error_report(struct sock *sk) 3279 { 3280 struct socket_wq *wq; 3281 3282 rcu_read_lock(); 3283 wq = rcu_dereference(sk->sk_wq); 3284 if (skwq_has_sleeper(wq)) 3285 wake_up_interruptible_poll(&wq->wait, EPOLLERR); 3286 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR); 3287 rcu_read_unlock(); 3288 } 3289 3290 void sock_def_readable(struct sock *sk) 3291 { 3292 struct socket_wq *wq; 3293 3294 rcu_read_lock(); 3295 wq = rcu_dereference(sk->sk_wq); 3296 if (skwq_has_sleeper(wq)) 3297 wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI | 3298 EPOLLRDNORM | EPOLLRDBAND); 3299 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN); 3300 rcu_read_unlock(); 3301 } 3302 3303 static void sock_def_write_space(struct sock *sk) 3304 { 3305 struct socket_wq *wq; 3306 3307 rcu_read_lock(); 3308 3309 /* Do not wake up a writer until he can make "significant" 3310 * progress. --DaveM 3311 */ 3312 if (sock_writeable(sk)) { 3313 wq = rcu_dereference(sk->sk_wq); 3314 if (skwq_has_sleeper(wq)) 3315 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT | 3316 EPOLLWRNORM | EPOLLWRBAND); 3317 3318 /* Should agree with poll, otherwise some programs break */ 3319 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT); 3320 } 3321 3322 rcu_read_unlock(); 3323 } 3324 3325 /* An optimised version of sock_def_write_space(), should only be called 3326 * for SOCK_RCU_FREE sockets under RCU read section and after putting 3327 * ->sk_wmem_alloc. 3328 */ 3329 static void sock_def_write_space_wfree(struct sock *sk) 3330 { 3331 /* Do not wake up a writer until he can make "significant" 3332 * progress. --DaveM 3333 */ 3334 if (sock_writeable(sk)) { 3335 struct socket_wq *wq = rcu_dereference(sk->sk_wq); 3336 3337 /* rely on refcount_sub from sock_wfree() */ 3338 smp_mb__after_atomic(); 3339 if (wq && waitqueue_active(&wq->wait)) 3340 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT | 3341 EPOLLWRNORM | EPOLLWRBAND); 3342 3343 /* Should agree with poll, otherwise some programs break */ 3344 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT); 3345 } 3346 } 3347 3348 static void sock_def_destruct(struct sock *sk) 3349 { 3350 } 3351 3352 void sk_send_sigurg(struct sock *sk) 3353 { 3354 if (sk->sk_socket && sk->sk_socket->file) 3355 if (send_sigurg(&sk->sk_socket->file->f_owner)) 3356 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI); 3357 } 3358 EXPORT_SYMBOL(sk_send_sigurg); 3359 3360 void sk_reset_timer(struct sock *sk, struct timer_list* timer, 3361 unsigned long expires) 3362 { 3363 if (!mod_timer(timer, expires)) 3364 sock_hold(sk); 3365 } 3366 EXPORT_SYMBOL(sk_reset_timer); 3367 3368 void sk_stop_timer(struct sock *sk, struct timer_list* timer) 3369 { 3370 if (del_timer(timer)) 3371 __sock_put(sk); 3372 } 3373 EXPORT_SYMBOL(sk_stop_timer); 3374 3375 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer) 3376 { 3377 if (del_timer_sync(timer)) 3378 __sock_put(sk); 3379 } 3380 EXPORT_SYMBOL(sk_stop_timer_sync); 3381 3382 void sock_init_data(struct socket *sock, struct sock *sk) 3383 { 3384 sk_init_common(sk); 3385 sk->sk_send_head = NULL; 3386 3387 timer_setup(&sk->sk_timer, NULL, 0); 3388 3389 sk->sk_allocation = GFP_KERNEL; 3390 sk->sk_rcvbuf = READ_ONCE(sysctl_rmem_default); 3391 sk->sk_sndbuf = READ_ONCE(sysctl_wmem_default); 3392 sk->sk_state = TCP_CLOSE; 3393 sk->sk_use_task_frag = true; 3394 sk_set_socket(sk, sock); 3395 3396 sock_set_flag(sk, SOCK_ZAPPED); 3397 3398 if (sock) { 3399 sk->sk_type = sock->type; 3400 RCU_INIT_POINTER(sk->sk_wq, &sock->wq); 3401 sock->sk = sk; 3402 sk->sk_uid = SOCK_INODE(sock)->i_uid; 3403 } else { 3404 RCU_INIT_POINTER(sk->sk_wq, NULL); 3405 sk->sk_uid = make_kuid(sock_net(sk)->user_ns, 0); 3406 } 3407 3408 rwlock_init(&sk->sk_callback_lock); 3409 if (sk->sk_kern_sock) 3410 lockdep_set_class_and_name( 3411 &sk->sk_callback_lock, 3412 af_kern_callback_keys + sk->sk_family, 3413 af_family_kern_clock_key_strings[sk->sk_family]); 3414 else 3415 lockdep_set_class_and_name( 3416 &sk->sk_callback_lock, 3417 af_callback_keys + sk->sk_family, 3418 af_family_clock_key_strings[sk->sk_family]); 3419 3420 sk->sk_state_change = sock_def_wakeup; 3421 sk->sk_data_ready = sock_def_readable; 3422 sk->sk_write_space = sock_def_write_space; 3423 sk->sk_error_report = sock_def_error_report; 3424 sk->sk_destruct = sock_def_destruct; 3425 3426 sk->sk_frag.page = NULL; 3427 sk->sk_frag.offset = 0; 3428 sk->sk_peek_off = -1; 3429 3430 sk->sk_peer_pid = NULL; 3431 sk->sk_peer_cred = NULL; 3432 spin_lock_init(&sk->sk_peer_lock); 3433 3434 sk->sk_write_pending = 0; 3435 sk->sk_rcvlowat = 1; 3436 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; 3437 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT; 3438 3439 sk->sk_stamp = SK_DEFAULT_STAMP; 3440 #if BITS_PER_LONG==32 3441 seqlock_init(&sk->sk_stamp_seq); 3442 #endif 3443 atomic_set(&sk->sk_zckey, 0); 3444 3445 #ifdef CONFIG_NET_RX_BUSY_POLL 3446 sk->sk_napi_id = 0; 3447 sk->sk_ll_usec = READ_ONCE(sysctl_net_busy_read); 3448 #endif 3449 3450 sk->sk_max_pacing_rate = ~0UL; 3451 sk->sk_pacing_rate = ~0UL; 3452 WRITE_ONCE(sk->sk_pacing_shift, 10); 3453 sk->sk_incoming_cpu = -1; 3454 sk->sk_txrehash = SOCK_TXREHASH_DEFAULT; 3455 3456 sk_rx_queue_clear(sk); 3457 /* 3458 * Before updating sk_refcnt, we must commit prior changes to memory 3459 * (Documentation/RCU/rculist_nulls.rst for details) 3460 */ 3461 smp_wmb(); 3462 refcount_set(&sk->sk_refcnt, 1); 3463 atomic_set(&sk->sk_drops, 0); 3464 } 3465 EXPORT_SYMBOL(sock_init_data); 3466 3467 void lock_sock_nested(struct sock *sk, int subclass) 3468 { 3469 /* The sk_lock has mutex_lock() semantics here. */ 3470 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_); 3471 3472 might_sleep(); 3473 spin_lock_bh(&sk->sk_lock.slock); 3474 if (sock_owned_by_user_nocheck(sk)) 3475 __lock_sock(sk); 3476 sk->sk_lock.owned = 1; 3477 spin_unlock_bh(&sk->sk_lock.slock); 3478 } 3479 EXPORT_SYMBOL(lock_sock_nested); 3480 3481 void release_sock(struct sock *sk) 3482 { 3483 spin_lock_bh(&sk->sk_lock.slock); 3484 if (sk->sk_backlog.tail) 3485 __release_sock(sk); 3486 3487 /* Warning : release_cb() might need to release sk ownership, 3488 * ie call sock_release_ownership(sk) before us. 3489 */ 3490 if (sk->sk_prot->release_cb) 3491 sk->sk_prot->release_cb(sk); 3492 3493 sock_release_ownership(sk); 3494 if (waitqueue_active(&sk->sk_lock.wq)) 3495 wake_up(&sk->sk_lock.wq); 3496 spin_unlock_bh(&sk->sk_lock.slock); 3497 } 3498 EXPORT_SYMBOL(release_sock); 3499 3500 bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock) 3501 { 3502 might_sleep(); 3503 spin_lock_bh(&sk->sk_lock.slock); 3504 3505 if (!sock_owned_by_user_nocheck(sk)) { 3506 /* 3507 * Fast path return with bottom halves disabled and 3508 * sock::sk_lock.slock held. 3509 * 3510 * The 'mutex' is not contended and holding 3511 * sock::sk_lock.slock prevents all other lockers to 3512 * proceed so the corresponding unlock_sock_fast() can 3513 * avoid the slow path of release_sock() completely and 3514 * just release slock. 3515 * 3516 * From a semantical POV this is equivalent to 'acquiring' 3517 * the 'mutex', hence the corresponding lockdep 3518 * mutex_release() has to happen in the fast path of 3519 * unlock_sock_fast(). 3520 */ 3521 return false; 3522 } 3523 3524 __lock_sock(sk); 3525 sk->sk_lock.owned = 1; 3526 __acquire(&sk->sk_lock.slock); 3527 spin_unlock_bh(&sk->sk_lock.slock); 3528 return true; 3529 } 3530 EXPORT_SYMBOL(__lock_sock_fast); 3531 3532 int sock_gettstamp(struct socket *sock, void __user *userstamp, 3533 bool timeval, bool time32) 3534 { 3535 struct sock *sk = sock->sk; 3536 struct timespec64 ts; 3537 3538 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 3539 ts = ktime_to_timespec64(sock_read_timestamp(sk)); 3540 if (ts.tv_sec == -1) 3541 return -ENOENT; 3542 if (ts.tv_sec == 0) { 3543 ktime_t kt = ktime_get_real(); 3544 sock_write_timestamp(sk, kt); 3545 ts = ktime_to_timespec64(kt); 3546 } 3547 3548 if (timeval) 3549 ts.tv_nsec /= 1000; 3550 3551 #ifdef CONFIG_COMPAT_32BIT_TIME 3552 if (time32) 3553 return put_old_timespec32(&ts, userstamp); 3554 #endif 3555 #ifdef CONFIG_SPARC64 3556 /* beware of padding in sparc64 timeval */ 3557 if (timeval && !in_compat_syscall()) { 3558 struct __kernel_old_timeval __user tv = { 3559 .tv_sec = ts.tv_sec, 3560 .tv_usec = ts.tv_nsec, 3561 }; 3562 if (copy_to_user(userstamp, &tv, sizeof(tv))) 3563 return -EFAULT; 3564 return 0; 3565 } 3566 #endif 3567 return put_timespec64(&ts, userstamp); 3568 } 3569 EXPORT_SYMBOL(sock_gettstamp); 3570 3571 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag) 3572 { 3573 if (!sock_flag(sk, flag)) { 3574 unsigned long previous_flags = sk->sk_flags; 3575 3576 sock_set_flag(sk, flag); 3577 /* 3578 * we just set one of the two flags which require net 3579 * time stamping, but time stamping might have been on 3580 * already because of the other one 3581 */ 3582 if (sock_needs_netstamp(sk) && 3583 !(previous_flags & SK_FLAGS_TIMESTAMP)) 3584 net_enable_timestamp(); 3585 } 3586 } 3587 3588 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, 3589 int level, int type) 3590 { 3591 struct sock_exterr_skb *serr; 3592 struct sk_buff *skb; 3593 int copied, err; 3594 3595 err = -EAGAIN; 3596 skb = sock_dequeue_err_skb(sk); 3597 if (skb == NULL) 3598 goto out; 3599 3600 copied = skb->len; 3601 if (copied > len) { 3602 msg->msg_flags |= MSG_TRUNC; 3603 copied = len; 3604 } 3605 err = skb_copy_datagram_msg(skb, 0, msg, copied); 3606 if (err) 3607 goto out_free_skb; 3608 3609 sock_recv_timestamp(msg, sk, skb); 3610 3611 serr = SKB_EXT_ERR(skb); 3612 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee); 3613 3614 msg->msg_flags |= MSG_ERRQUEUE; 3615 err = copied; 3616 3617 out_free_skb: 3618 kfree_skb(skb); 3619 out: 3620 return err; 3621 } 3622 EXPORT_SYMBOL(sock_recv_errqueue); 3623 3624 /* 3625 * Get a socket option on an socket. 3626 * 3627 * FIX: POSIX 1003.1g is very ambiguous here. It states that 3628 * asynchronous errors should be reported by getsockopt. We assume 3629 * this means if you specify SO_ERROR (otherwise whats the point of it). 3630 */ 3631 int sock_common_getsockopt(struct socket *sock, int level, int optname, 3632 char __user *optval, int __user *optlen) 3633 { 3634 struct sock *sk = sock->sk; 3635 3636 /* IPV6_ADDRFORM can change sk->sk_prot under us. */ 3637 return READ_ONCE(sk->sk_prot)->getsockopt(sk, level, optname, optval, optlen); 3638 } 3639 EXPORT_SYMBOL(sock_common_getsockopt); 3640 3641 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 3642 int flags) 3643 { 3644 struct sock *sk = sock->sk; 3645 int addr_len = 0; 3646 int err; 3647 3648 err = sk->sk_prot->recvmsg(sk, msg, size, flags, &addr_len); 3649 if (err >= 0) 3650 msg->msg_namelen = addr_len; 3651 return err; 3652 } 3653 EXPORT_SYMBOL(sock_common_recvmsg); 3654 3655 /* 3656 * Set socket options on an inet socket. 3657 */ 3658 int sock_common_setsockopt(struct socket *sock, int level, int optname, 3659 sockptr_t optval, unsigned int optlen) 3660 { 3661 struct sock *sk = sock->sk; 3662 3663 /* IPV6_ADDRFORM can change sk->sk_prot under us. */ 3664 return READ_ONCE(sk->sk_prot)->setsockopt(sk, level, optname, optval, optlen); 3665 } 3666 EXPORT_SYMBOL(sock_common_setsockopt); 3667 3668 void sk_common_release(struct sock *sk) 3669 { 3670 if (sk->sk_prot->destroy) 3671 sk->sk_prot->destroy(sk); 3672 3673 /* 3674 * Observation: when sk_common_release is called, processes have 3675 * no access to socket. But net still has. 3676 * Step one, detach it from networking: 3677 * 3678 * A. Remove from hash tables. 3679 */ 3680 3681 sk->sk_prot->unhash(sk); 3682 3683 /* 3684 * In this point socket cannot receive new packets, but it is possible 3685 * that some packets are in flight because some CPU runs receiver and 3686 * did hash table lookup before we unhashed socket. They will achieve 3687 * receive queue and will be purged by socket destructor. 3688 * 3689 * Also we still have packets pending on receive queue and probably, 3690 * our own packets waiting in device queues. sock_destroy will drain 3691 * receive queue, but transmitted packets will delay socket destruction 3692 * until the last reference will be released. 3693 */ 3694 3695 sock_orphan(sk); 3696 3697 xfrm_sk_free_policy(sk); 3698 3699 sk_refcnt_debug_release(sk); 3700 3701 sock_put(sk); 3702 } 3703 EXPORT_SYMBOL(sk_common_release); 3704 3705 void sk_get_meminfo(const struct sock *sk, u32 *mem) 3706 { 3707 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS); 3708 3709 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk); 3710 mem[SK_MEMINFO_RCVBUF] = READ_ONCE(sk->sk_rcvbuf); 3711 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk); 3712 mem[SK_MEMINFO_SNDBUF] = READ_ONCE(sk->sk_sndbuf); 3713 mem[SK_MEMINFO_FWD_ALLOC] = sk->sk_forward_alloc; 3714 mem[SK_MEMINFO_WMEM_QUEUED] = READ_ONCE(sk->sk_wmem_queued); 3715 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc); 3716 mem[SK_MEMINFO_BACKLOG] = READ_ONCE(sk->sk_backlog.len); 3717 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops); 3718 } 3719 3720 #ifdef CONFIG_PROC_FS 3721 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR); 3722 3723 int sock_prot_inuse_get(struct net *net, struct proto *prot) 3724 { 3725 int cpu, idx = prot->inuse_idx; 3726 int res = 0; 3727 3728 for_each_possible_cpu(cpu) 3729 res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx]; 3730 3731 return res >= 0 ? res : 0; 3732 } 3733 EXPORT_SYMBOL_GPL(sock_prot_inuse_get); 3734 3735 int sock_inuse_get(struct net *net) 3736 { 3737 int cpu, res = 0; 3738 3739 for_each_possible_cpu(cpu) 3740 res += per_cpu_ptr(net->core.prot_inuse, cpu)->all; 3741 3742 return res; 3743 } 3744 3745 EXPORT_SYMBOL_GPL(sock_inuse_get); 3746 3747 static int __net_init sock_inuse_init_net(struct net *net) 3748 { 3749 net->core.prot_inuse = alloc_percpu(struct prot_inuse); 3750 if (net->core.prot_inuse == NULL) 3751 return -ENOMEM; 3752 return 0; 3753 } 3754 3755 static void __net_exit sock_inuse_exit_net(struct net *net) 3756 { 3757 free_percpu(net->core.prot_inuse); 3758 } 3759 3760 static struct pernet_operations net_inuse_ops = { 3761 .init = sock_inuse_init_net, 3762 .exit = sock_inuse_exit_net, 3763 }; 3764 3765 static __init int net_inuse_init(void) 3766 { 3767 if (register_pernet_subsys(&net_inuse_ops)) 3768 panic("Cannot initialize net inuse counters"); 3769 3770 return 0; 3771 } 3772 3773 core_initcall(net_inuse_init); 3774 3775 static int assign_proto_idx(struct proto *prot) 3776 { 3777 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR); 3778 3779 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) { 3780 pr_err("PROTO_INUSE_NR exhausted\n"); 3781 return -ENOSPC; 3782 } 3783 3784 set_bit(prot->inuse_idx, proto_inuse_idx); 3785 return 0; 3786 } 3787 3788 static void release_proto_idx(struct proto *prot) 3789 { 3790 if (prot->inuse_idx != PROTO_INUSE_NR - 1) 3791 clear_bit(prot->inuse_idx, proto_inuse_idx); 3792 } 3793 #else 3794 static inline int assign_proto_idx(struct proto *prot) 3795 { 3796 return 0; 3797 } 3798 3799 static inline void release_proto_idx(struct proto *prot) 3800 { 3801 } 3802 3803 #endif 3804 3805 static void tw_prot_cleanup(struct timewait_sock_ops *twsk_prot) 3806 { 3807 if (!twsk_prot) 3808 return; 3809 kfree(twsk_prot->twsk_slab_name); 3810 twsk_prot->twsk_slab_name = NULL; 3811 kmem_cache_destroy(twsk_prot->twsk_slab); 3812 twsk_prot->twsk_slab = NULL; 3813 } 3814 3815 static int tw_prot_init(const struct proto *prot) 3816 { 3817 struct timewait_sock_ops *twsk_prot = prot->twsk_prot; 3818 3819 if (!twsk_prot) 3820 return 0; 3821 3822 twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", 3823 prot->name); 3824 if (!twsk_prot->twsk_slab_name) 3825 return -ENOMEM; 3826 3827 twsk_prot->twsk_slab = 3828 kmem_cache_create(twsk_prot->twsk_slab_name, 3829 twsk_prot->twsk_obj_size, 0, 3830 SLAB_ACCOUNT | prot->slab_flags, 3831 NULL); 3832 if (!twsk_prot->twsk_slab) { 3833 pr_crit("%s: Can't create timewait sock SLAB cache!\n", 3834 prot->name); 3835 return -ENOMEM; 3836 } 3837 3838 return 0; 3839 } 3840 3841 static void req_prot_cleanup(struct request_sock_ops *rsk_prot) 3842 { 3843 if (!rsk_prot) 3844 return; 3845 kfree(rsk_prot->slab_name); 3846 rsk_prot->slab_name = NULL; 3847 kmem_cache_destroy(rsk_prot->slab); 3848 rsk_prot->slab = NULL; 3849 } 3850 3851 static int req_prot_init(const struct proto *prot) 3852 { 3853 struct request_sock_ops *rsk_prot = prot->rsk_prot; 3854 3855 if (!rsk_prot) 3856 return 0; 3857 3858 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", 3859 prot->name); 3860 if (!rsk_prot->slab_name) 3861 return -ENOMEM; 3862 3863 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name, 3864 rsk_prot->obj_size, 0, 3865 SLAB_ACCOUNT | prot->slab_flags, 3866 NULL); 3867 3868 if (!rsk_prot->slab) { 3869 pr_crit("%s: Can't create request sock SLAB cache!\n", 3870 prot->name); 3871 return -ENOMEM; 3872 } 3873 return 0; 3874 } 3875 3876 int proto_register(struct proto *prot, int alloc_slab) 3877 { 3878 int ret = -ENOBUFS; 3879 3880 if (prot->memory_allocated && !prot->sysctl_mem) { 3881 pr_err("%s: missing sysctl_mem\n", prot->name); 3882 return -EINVAL; 3883 } 3884 if (prot->memory_allocated && !prot->per_cpu_fw_alloc) { 3885 pr_err("%s: missing per_cpu_fw_alloc\n", prot->name); 3886 return -EINVAL; 3887 } 3888 if (alloc_slab) { 3889 prot->slab = kmem_cache_create_usercopy(prot->name, 3890 prot->obj_size, 0, 3891 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT | 3892 prot->slab_flags, 3893 prot->useroffset, prot->usersize, 3894 NULL); 3895 3896 if (prot->slab == NULL) { 3897 pr_crit("%s: Can't create sock SLAB cache!\n", 3898 prot->name); 3899 goto out; 3900 } 3901 3902 if (req_prot_init(prot)) 3903 goto out_free_request_sock_slab; 3904 3905 if (tw_prot_init(prot)) 3906 goto out_free_timewait_sock_slab; 3907 } 3908 3909 mutex_lock(&proto_list_mutex); 3910 ret = assign_proto_idx(prot); 3911 if (ret) { 3912 mutex_unlock(&proto_list_mutex); 3913 goto out_free_timewait_sock_slab; 3914 } 3915 list_add(&prot->node, &proto_list); 3916 mutex_unlock(&proto_list_mutex); 3917 return ret; 3918 3919 out_free_timewait_sock_slab: 3920 if (alloc_slab) 3921 tw_prot_cleanup(prot->twsk_prot); 3922 out_free_request_sock_slab: 3923 if (alloc_slab) { 3924 req_prot_cleanup(prot->rsk_prot); 3925 3926 kmem_cache_destroy(prot->slab); 3927 prot->slab = NULL; 3928 } 3929 out: 3930 return ret; 3931 } 3932 EXPORT_SYMBOL(proto_register); 3933 3934 void proto_unregister(struct proto *prot) 3935 { 3936 mutex_lock(&proto_list_mutex); 3937 release_proto_idx(prot); 3938 list_del(&prot->node); 3939 mutex_unlock(&proto_list_mutex); 3940 3941 kmem_cache_destroy(prot->slab); 3942 prot->slab = NULL; 3943 3944 req_prot_cleanup(prot->rsk_prot); 3945 tw_prot_cleanup(prot->twsk_prot); 3946 } 3947 EXPORT_SYMBOL(proto_unregister); 3948 3949 int sock_load_diag_module(int family, int protocol) 3950 { 3951 if (!protocol) { 3952 if (!sock_is_registered(family)) 3953 return -ENOENT; 3954 3955 return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK, 3956 NETLINK_SOCK_DIAG, family); 3957 } 3958 3959 #ifdef CONFIG_INET 3960 if (family == AF_INET && 3961 protocol != IPPROTO_RAW && 3962 protocol < MAX_INET_PROTOS && 3963 !rcu_access_pointer(inet_protos[protocol])) 3964 return -ENOENT; 3965 #endif 3966 3967 return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK, 3968 NETLINK_SOCK_DIAG, family, protocol); 3969 } 3970 EXPORT_SYMBOL(sock_load_diag_module); 3971 3972 #ifdef CONFIG_PROC_FS 3973 static void *proto_seq_start(struct seq_file *seq, loff_t *pos) 3974 __acquires(proto_list_mutex) 3975 { 3976 mutex_lock(&proto_list_mutex); 3977 return seq_list_start_head(&proto_list, *pos); 3978 } 3979 3980 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos) 3981 { 3982 return seq_list_next(v, &proto_list, pos); 3983 } 3984 3985 static void proto_seq_stop(struct seq_file *seq, void *v) 3986 __releases(proto_list_mutex) 3987 { 3988 mutex_unlock(&proto_list_mutex); 3989 } 3990 3991 static char proto_method_implemented(const void *method) 3992 { 3993 return method == NULL ? 'n' : 'y'; 3994 } 3995 static long sock_prot_memory_allocated(struct proto *proto) 3996 { 3997 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L; 3998 } 3999 4000 static const char *sock_prot_memory_pressure(struct proto *proto) 4001 { 4002 return proto->memory_pressure != NULL ? 4003 proto_memory_pressure(proto) ? "yes" : "no" : "NI"; 4004 } 4005 4006 static void proto_seq_printf(struct seq_file *seq, struct proto *proto) 4007 { 4008 4009 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s " 4010 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n", 4011 proto->name, 4012 proto->obj_size, 4013 sock_prot_inuse_get(seq_file_net(seq), proto), 4014 sock_prot_memory_allocated(proto), 4015 sock_prot_memory_pressure(proto), 4016 proto->max_header, 4017 proto->slab == NULL ? "no" : "yes", 4018 module_name(proto->owner), 4019 proto_method_implemented(proto->close), 4020 proto_method_implemented(proto->connect), 4021 proto_method_implemented(proto->disconnect), 4022 proto_method_implemented(proto->accept), 4023 proto_method_implemented(proto->ioctl), 4024 proto_method_implemented(proto->init), 4025 proto_method_implemented(proto->destroy), 4026 proto_method_implemented(proto->shutdown), 4027 proto_method_implemented(proto->setsockopt), 4028 proto_method_implemented(proto->getsockopt), 4029 proto_method_implemented(proto->sendmsg), 4030 proto_method_implemented(proto->recvmsg), 4031 proto_method_implemented(proto->sendpage), 4032 proto_method_implemented(proto->bind), 4033 proto_method_implemented(proto->backlog_rcv), 4034 proto_method_implemented(proto->hash), 4035 proto_method_implemented(proto->unhash), 4036 proto_method_implemented(proto->get_port), 4037 proto_method_implemented(proto->enter_memory_pressure)); 4038 } 4039 4040 static int proto_seq_show(struct seq_file *seq, void *v) 4041 { 4042 if (v == &proto_list) 4043 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s", 4044 "protocol", 4045 "size", 4046 "sockets", 4047 "memory", 4048 "press", 4049 "maxhdr", 4050 "slab", 4051 "module", 4052 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n"); 4053 else 4054 proto_seq_printf(seq, list_entry(v, struct proto, node)); 4055 return 0; 4056 } 4057 4058 static const struct seq_operations proto_seq_ops = { 4059 .start = proto_seq_start, 4060 .next = proto_seq_next, 4061 .stop = proto_seq_stop, 4062 .show = proto_seq_show, 4063 }; 4064 4065 static __net_init int proto_init_net(struct net *net) 4066 { 4067 if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops, 4068 sizeof(struct seq_net_private))) 4069 return -ENOMEM; 4070 4071 return 0; 4072 } 4073 4074 static __net_exit void proto_exit_net(struct net *net) 4075 { 4076 remove_proc_entry("protocols", net->proc_net); 4077 } 4078 4079 4080 static __net_initdata struct pernet_operations proto_net_ops = { 4081 .init = proto_init_net, 4082 .exit = proto_exit_net, 4083 }; 4084 4085 static int __init proto_init(void) 4086 { 4087 return register_pernet_subsys(&proto_net_ops); 4088 } 4089 4090 subsys_initcall(proto_init); 4091 4092 #endif /* PROC_FS */ 4093 4094 #ifdef CONFIG_NET_RX_BUSY_POLL 4095 bool sk_busy_loop_end(void *p, unsigned long start_time) 4096 { 4097 struct sock *sk = p; 4098 4099 return !skb_queue_empty_lockless(&sk->sk_receive_queue) || 4100 sk_busy_loop_timeout(sk, start_time); 4101 } 4102 EXPORT_SYMBOL(sk_busy_loop_end); 4103 #endif /* CONFIG_NET_RX_BUSY_POLL */ 4104 4105 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len) 4106 { 4107 if (!sk->sk_prot->bind_add) 4108 return -EOPNOTSUPP; 4109 return sk->sk_prot->bind_add(sk, addr, addr_len); 4110 } 4111 EXPORT_SYMBOL(sock_bind_add); 4112