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 if (val < 0) 1400 ret = -EINVAL; 1401 else 1402 WRITE_ONCE(sk->sk_ll_usec, val); 1403 break; 1404 case SO_PREFER_BUSY_POLL: 1405 if (valbool && !sockopt_capable(CAP_NET_ADMIN)) 1406 ret = -EPERM; 1407 else 1408 WRITE_ONCE(sk->sk_prefer_busy_poll, valbool); 1409 break; 1410 case SO_BUSY_POLL_BUDGET: 1411 if (val > READ_ONCE(sk->sk_busy_poll_budget) && !sockopt_capable(CAP_NET_ADMIN)) { 1412 ret = -EPERM; 1413 } else { 1414 if (val < 0 || val > U16_MAX) 1415 ret = -EINVAL; 1416 else 1417 WRITE_ONCE(sk->sk_busy_poll_budget, val); 1418 } 1419 break; 1420 #endif 1421 1422 case SO_MAX_PACING_RATE: 1423 { 1424 unsigned long ulval = (val == ~0U) ? ~0UL : (unsigned int)val; 1425 1426 if (sizeof(ulval) != sizeof(val) && 1427 optlen >= sizeof(ulval) && 1428 copy_from_sockptr(&ulval, optval, sizeof(ulval))) { 1429 ret = -EFAULT; 1430 break; 1431 } 1432 if (ulval != ~0UL) 1433 cmpxchg(&sk->sk_pacing_status, 1434 SK_PACING_NONE, 1435 SK_PACING_NEEDED); 1436 sk->sk_max_pacing_rate = ulval; 1437 sk->sk_pacing_rate = min(sk->sk_pacing_rate, ulval); 1438 break; 1439 } 1440 case SO_INCOMING_CPU: 1441 reuseport_update_incoming_cpu(sk, val); 1442 break; 1443 1444 case SO_CNX_ADVICE: 1445 if (val == 1) 1446 dst_negative_advice(sk); 1447 break; 1448 1449 case SO_ZEROCOPY: 1450 if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) { 1451 if (!(sk_is_tcp(sk) || 1452 (sk->sk_type == SOCK_DGRAM && 1453 sk->sk_protocol == IPPROTO_UDP))) 1454 ret = -EOPNOTSUPP; 1455 } else if (sk->sk_family != PF_RDS) { 1456 ret = -EOPNOTSUPP; 1457 } 1458 if (!ret) { 1459 if (val < 0 || val > 1) 1460 ret = -EINVAL; 1461 else 1462 sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool); 1463 } 1464 break; 1465 1466 case SO_TXTIME: 1467 if (optlen != sizeof(struct sock_txtime)) { 1468 ret = -EINVAL; 1469 break; 1470 } else if (copy_from_sockptr(&sk_txtime, optval, 1471 sizeof(struct sock_txtime))) { 1472 ret = -EFAULT; 1473 break; 1474 } else if (sk_txtime.flags & ~SOF_TXTIME_FLAGS_MASK) { 1475 ret = -EINVAL; 1476 break; 1477 } 1478 /* CLOCK_MONOTONIC is only used by sch_fq, and this packet 1479 * scheduler has enough safe guards. 1480 */ 1481 if (sk_txtime.clockid != CLOCK_MONOTONIC && 1482 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) { 1483 ret = -EPERM; 1484 break; 1485 } 1486 sock_valbool_flag(sk, SOCK_TXTIME, true); 1487 sk->sk_clockid = sk_txtime.clockid; 1488 sk->sk_txtime_deadline_mode = 1489 !!(sk_txtime.flags & SOF_TXTIME_DEADLINE_MODE); 1490 sk->sk_txtime_report_errors = 1491 !!(sk_txtime.flags & SOF_TXTIME_REPORT_ERRORS); 1492 break; 1493 1494 case SO_BINDTOIFINDEX: 1495 ret = sock_bindtoindex_locked(sk, val); 1496 break; 1497 1498 case SO_BUF_LOCK: 1499 if (val & ~SOCK_BUF_LOCK_MASK) { 1500 ret = -EINVAL; 1501 break; 1502 } 1503 sk->sk_userlocks = val | (sk->sk_userlocks & 1504 ~SOCK_BUF_LOCK_MASK); 1505 break; 1506 1507 case SO_RESERVE_MEM: 1508 { 1509 int delta; 1510 1511 if (val < 0) { 1512 ret = -EINVAL; 1513 break; 1514 } 1515 1516 delta = val - sk->sk_reserved_mem; 1517 if (delta < 0) 1518 sock_release_reserved_memory(sk, -delta); 1519 else 1520 ret = sock_reserve_memory(sk, delta); 1521 break; 1522 } 1523 1524 case SO_TXREHASH: 1525 if (val < -1 || val > 1) { 1526 ret = -EINVAL; 1527 break; 1528 } 1529 if ((u8)val == SOCK_TXREHASH_DEFAULT) 1530 val = READ_ONCE(sock_net(sk)->core.sysctl_txrehash); 1531 /* Paired with READ_ONCE() in tcp_rtx_synack() */ 1532 WRITE_ONCE(sk->sk_txrehash, (u8)val); 1533 break; 1534 1535 default: 1536 ret = -ENOPROTOOPT; 1537 break; 1538 } 1539 sockopt_release_sock(sk); 1540 return ret; 1541 } 1542 1543 int sock_setsockopt(struct socket *sock, int level, int optname, 1544 sockptr_t optval, unsigned int optlen) 1545 { 1546 return sk_setsockopt(sock->sk, level, optname, 1547 optval, optlen); 1548 } 1549 EXPORT_SYMBOL(sock_setsockopt); 1550 1551 static const struct cred *sk_get_peer_cred(struct sock *sk) 1552 { 1553 const struct cred *cred; 1554 1555 spin_lock(&sk->sk_peer_lock); 1556 cred = get_cred(sk->sk_peer_cred); 1557 spin_unlock(&sk->sk_peer_lock); 1558 1559 return cred; 1560 } 1561 1562 static void cred_to_ucred(struct pid *pid, const struct cred *cred, 1563 struct ucred *ucred) 1564 { 1565 ucred->pid = pid_vnr(pid); 1566 ucred->uid = ucred->gid = -1; 1567 if (cred) { 1568 struct user_namespace *current_ns = current_user_ns(); 1569 1570 ucred->uid = from_kuid_munged(current_ns, cred->euid); 1571 ucred->gid = from_kgid_munged(current_ns, cred->egid); 1572 } 1573 } 1574 1575 static int groups_to_user(sockptr_t dst, const struct group_info *src) 1576 { 1577 struct user_namespace *user_ns = current_user_ns(); 1578 int i; 1579 1580 for (i = 0; i < src->ngroups; i++) { 1581 gid_t gid = from_kgid_munged(user_ns, src->gid[i]); 1582 1583 if (copy_to_sockptr_offset(dst, i * sizeof(gid), &gid, sizeof(gid))) 1584 return -EFAULT; 1585 } 1586 1587 return 0; 1588 } 1589 1590 int sk_getsockopt(struct sock *sk, int level, int optname, 1591 sockptr_t optval, sockptr_t optlen) 1592 { 1593 struct socket *sock = sk->sk_socket; 1594 1595 union { 1596 int val; 1597 u64 val64; 1598 unsigned long ulval; 1599 struct linger ling; 1600 struct old_timeval32 tm32; 1601 struct __kernel_old_timeval tm; 1602 struct __kernel_sock_timeval stm; 1603 struct sock_txtime txtime; 1604 struct so_timestamping timestamping; 1605 } v; 1606 1607 int lv = sizeof(int); 1608 int len; 1609 1610 if (copy_from_sockptr(&len, optlen, sizeof(int))) 1611 return -EFAULT; 1612 if (len < 0) 1613 return -EINVAL; 1614 1615 memset(&v, 0, sizeof(v)); 1616 1617 switch (optname) { 1618 case SO_DEBUG: 1619 v.val = sock_flag(sk, SOCK_DBG); 1620 break; 1621 1622 case SO_DONTROUTE: 1623 v.val = sock_flag(sk, SOCK_LOCALROUTE); 1624 break; 1625 1626 case SO_BROADCAST: 1627 v.val = sock_flag(sk, SOCK_BROADCAST); 1628 break; 1629 1630 case SO_SNDBUF: 1631 v.val = sk->sk_sndbuf; 1632 break; 1633 1634 case SO_RCVBUF: 1635 v.val = sk->sk_rcvbuf; 1636 break; 1637 1638 case SO_REUSEADDR: 1639 v.val = sk->sk_reuse; 1640 break; 1641 1642 case SO_REUSEPORT: 1643 v.val = sk->sk_reuseport; 1644 break; 1645 1646 case SO_KEEPALIVE: 1647 v.val = sock_flag(sk, SOCK_KEEPOPEN); 1648 break; 1649 1650 case SO_TYPE: 1651 v.val = sk->sk_type; 1652 break; 1653 1654 case SO_PROTOCOL: 1655 v.val = sk->sk_protocol; 1656 break; 1657 1658 case SO_DOMAIN: 1659 v.val = sk->sk_family; 1660 break; 1661 1662 case SO_ERROR: 1663 v.val = -sock_error(sk); 1664 if (v.val == 0) 1665 v.val = xchg(&sk->sk_err_soft, 0); 1666 break; 1667 1668 case SO_OOBINLINE: 1669 v.val = sock_flag(sk, SOCK_URGINLINE); 1670 break; 1671 1672 case SO_NO_CHECK: 1673 v.val = sk->sk_no_check_tx; 1674 break; 1675 1676 case SO_PRIORITY: 1677 v.val = sk->sk_priority; 1678 break; 1679 1680 case SO_LINGER: 1681 lv = sizeof(v.ling); 1682 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER); 1683 v.ling.l_linger = sk->sk_lingertime / HZ; 1684 break; 1685 1686 case SO_BSDCOMPAT: 1687 break; 1688 1689 case SO_TIMESTAMP_OLD: 1690 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && 1691 !sock_flag(sk, SOCK_TSTAMP_NEW) && 1692 !sock_flag(sk, SOCK_RCVTSTAMPNS); 1693 break; 1694 1695 case SO_TIMESTAMPNS_OLD: 1696 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && !sock_flag(sk, SOCK_TSTAMP_NEW); 1697 break; 1698 1699 case SO_TIMESTAMP_NEW: 1700 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && sock_flag(sk, SOCK_TSTAMP_NEW); 1701 break; 1702 1703 case SO_TIMESTAMPNS_NEW: 1704 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && sock_flag(sk, SOCK_TSTAMP_NEW); 1705 break; 1706 1707 case SO_TIMESTAMPING_OLD: 1708 lv = sizeof(v.timestamping); 1709 v.timestamping.flags = sk->sk_tsflags; 1710 v.timestamping.bind_phc = sk->sk_bind_phc; 1711 break; 1712 1713 case SO_RCVTIMEO_OLD: 1714 case SO_RCVTIMEO_NEW: 1715 lv = sock_get_timeout(sk->sk_rcvtimeo, &v, SO_RCVTIMEO_OLD == optname); 1716 break; 1717 1718 case SO_SNDTIMEO_OLD: 1719 case SO_SNDTIMEO_NEW: 1720 lv = sock_get_timeout(sk->sk_sndtimeo, &v, SO_SNDTIMEO_OLD == optname); 1721 break; 1722 1723 case SO_RCVLOWAT: 1724 v.val = sk->sk_rcvlowat; 1725 break; 1726 1727 case SO_SNDLOWAT: 1728 v.val = 1; 1729 break; 1730 1731 case SO_PASSCRED: 1732 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags); 1733 break; 1734 1735 case SO_PEERCRED: 1736 { 1737 struct ucred peercred; 1738 if (len > sizeof(peercred)) 1739 len = sizeof(peercred); 1740 1741 spin_lock(&sk->sk_peer_lock); 1742 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred); 1743 spin_unlock(&sk->sk_peer_lock); 1744 1745 if (copy_to_sockptr(optval, &peercred, len)) 1746 return -EFAULT; 1747 goto lenout; 1748 } 1749 1750 case SO_PEERGROUPS: 1751 { 1752 const struct cred *cred; 1753 int ret, n; 1754 1755 cred = sk_get_peer_cred(sk); 1756 if (!cred) 1757 return -ENODATA; 1758 1759 n = cred->group_info->ngroups; 1760 if (len < n * sizeof(gid_t)) { 1761 len = n * sizeof(gid_t); 1762 put_cred(cred); 1763 return copy_to_sockptr(optlen, &len, sizeof(int)) ? -EFAULT : -ERANGE; 1764 } 1765 len = n * sizeof(gid_t); 1766 1767 ret = groups_to_user(optval, cred->group_info); 1768 put_cred(cred); 1769 if (ret) 1770 return ret; 1771 goto lenout; 1772 } 1773 1774 case SO_PEERNAME: 1775 { 1776 char address[128]; 1777 1778 lv = sock->ops->getname(sock, (struct sockaddr *)address, 2); 1779 if (lv < 0) 1780 return -ENOTCONN; 1781 if (lv < len) 1782 return -EINVAL; 1783 if (copy_to_sockptr(optval, address, len)) 1784 return -EFAULT; 1785 goto lenout; 1786 } 1787 1788 /* Dubious BSD thing... Probably nobody even uses it, but 1789 * the UNIX standard wants it for whatever reason... -DaveM 1790 */ 1791 case SO_ACCEPTCONN: 1792 v.val = sk->sk_state == TCP_LISTEN; 1793 break; 1794 1795 case SO_PASSSEC: 1796 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags); 1797 break; 1798 1799 case SO_PEERSEC: 1800 return security_socket_getpeersec_stream(sock, 1801 optval, optlen, len); 1802 1803 case SO_MARK: 1804 v.val = sk->sk_mark; 1805 break; 1806 1807 case SO_RCVMARK: 1808 v.val = sock_flag(sk, SOCK_RCVMARK); 1809 break; 1810 1811 case SO_RXQ_OVFL: 1812 v.val = sock_flag(sk, SOCK_RXQ_OVFL); 1813 break; 1814 1815 case SO_WIFI_STATUS: 1816 v.val = sock_flag(sk, SOCK_WIFI_STATUS); 1817 break; 1818 1819 case SO_PEEK_OFF: 1820 if (!sock->ops->set_peek_off) 1821 return -EOPNOTSUPP; 1822 1823 v.val = sk->sk_peek_off; 1824 break; 1825 case SO_NOFCS: 1826 v.val = sock_flag(sk, SOCK_NOFCS); 1827 break; 1828 1829 case SO_BINDTODEVICE: 1830 return sock_getbindtodevice(sk, optval, optlen, len); 1831 1832 case SO_GET_FILTER: 1833 len = sk_get_filter(sk, optval, len); 1834 if (len < 0) 1835 return len; 1836 1837 goto lenout; 1838 1839 case SO_LOCK_FILTER: 1840 v.val = sock_flag(sk, SOCK_FILTER_LOCKED); 1841 break; 1842 1843 case SO_BPF_EXTENSIONS: 1844 v.val = bpf_tell_extensions(); 1845 break; 1846 1847 case SO_SELECT_ERR_QUEUE: 1848 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE); 1849 break; 1850 1851 #ifdef CONFIG_NET_RX_BUSY_POLL 1852 case SO_BUSY_POLL: 1853 v.val = sk->sk_ll_usec; 1854 break; 1855 case SO_PREFER_BUSY_POLL: 1856 v.val = READ_ONCE(sk->sk_prefer_busy_poll); 1857 break; 1858 #endif 1859 1860 case SO_MAX_PACING_RATE: 1861 if (sizeof(v.ulval) != sizeof(v.val) && len >= sizeof(v.ulval)) { 1862 lv = sizeof(v.ulval); 1863 v.ulval = sk->sk_max_pacing_rate; 1864 } else { 1865 /* 32bit version */ 1866 v.val = min_t(unsigned long, sk->sk_max_pacing_rate, ~0U); 1867 } 1868 break; 1869 1870 case SO_INCOMING_CPU: 1871 v.val = READ_ONCE(sk->sk_incoming_cpu); 1872 break; 1873 1874 case SO_MEMINFO: 1875 { 1876 u32 meminfo[SK_MEMINFO_VARS]; 1877 1878 sk_get_meminfo(sk, meminfo); 1879 1880 len = min_t(unsigned int, len, sizeof(meminfo)); 1881 if (copy_to_sockptr(optval, &meminfo, len)) 1882 return -EFAULT; 1883 1884 goto lenout; 1885 } 1886 1887 #ifdef CONFIG_NET_RX_BUSY_POLL 1888 case SO_INCOMING_NAPI_ID: 1889 v.val = READ_ONCE(sk->sk_napi_id); 1890 1891 /* aggregate non-NAPI IDs down to 0 */ 1892 if (v.val < MIN_NAPI_ID) 1893 v.val = 0; 1894 1895 break; 1896 #endif 1897 1898 case SO_COOKIE: 1899 lv = sizeof(u64); 1900 if (len < lv) 1901 return -EINVAL; 1902 v.val64 = sock_gen_cookie(sk); 1903 break; 1904 1905 case SO_ZEROCOPY: 1906 v.val = sock_flag(sk, SOCK_ZEROCOPY); 1907 break; 1908 1909 case SO_TXTIME: 1910 lv = sizeof(v.txtime); 1911 v.txtime.clockid = sk->sk_clockid; 1912 v.txtime.flags |= sk->sk_txtime_deadline_mode ? 1913 SOF_TXTIME_DEADLINE_MODE : 0; 1914 v.txtime.flags |= sk->sk_txtime_report_errors ? 1915 SOF_TXTIME_REPORT_ERRORS : 0; 1916 break; 1917 1918 case SO_BINDTOIFINDEX: 1919 v.val = READ_ONCE(sk->sk_bound_dev_if); 1920 break; 1921 1922 case SO_NETNS_COOKIE: 1923 lv = sizeof(u64); 1924 if (len != lv) 1925 return -EINVAL; 1926 v.val64 = sock_net(sk)->net_cookie; 1927 break; 1928 1929 case SO_BUF_LOCK: 1930 v.val = sk->sk_userlocks & SOCK_BUF_LOCK_MASK; 1931 break; 1932 1933 case SO_RESERVE_MEM: 1934 v.val = sk->sk_reserved_mem; 1935 break; 1936 1937 case SO_TXREHASH: 1938 v.val = sk->sk_txrehash; 1939 break; 1940 1941 default: 1942 /* We implement the SO_SNDLOWAT etc to not be settable 1943 * (1003.1g 7). 1944 */ 1945 return -ENOPROTOOPT; 1946 } 1947 1948 if (len > lv) 1949 len = lv; 1950 if (copy_to_sockptr(optval, &v, len)) 1951 return -EFAULT; 1952 lenout: 1953 if (copy_to_sockptr(optlen, &len, sizeof(int))) 1954 return -EFAULT; 1955 return 0; 1956 } 1957 1958 int sock_getsockopt(struct socket *sock, int level, int optname, 1959 char __user *optval, int __user *optlen) 1960 { 1961 return sk_getsockopt(sock->sk, level, optname, 1962 USER_SOCKPTR(optval), 1963 USER_SOCKPTR(optlen)); 1964 } 1965 1966 /* 1967 * Initialize an sk_lock. 1968 * 1969 * (We also register the sk_lock with the lock validator.) 1970 */ 1971 static inline void sock_lock_init(struct sock *sk) 1972 { 1973 if (sk->sk_kern_sock) 1974 sock_lock_init_class_and_name( 1975 sk, 1976 af_family_kern_slock_key_strings[sk->sk_family], 1977 af_family_kern_slock_keys + sk->sk_family, 1978 af_family_kern_key_strings[sk->sk_family], 1979 af_family_kern_keys + sk->sk_family); 1980 else 1981 sock_lock_init_class_and_name( 1982 sk, 1983 af_family_slock_key_strings[sk->sk_family], 1984 af_family_slock_keys + sk->sk_family, 1985 af_family_key_strings[sk->sk_family], 1986 af_family_keys + sk->sk_family); 1987 } 1988 1989 /* 1990 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet, 1991 * even temporarly, because of RCU lookups. sk_node should also be left as is. 1992 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end 1993 */ 1994 static void sock_copy(struct sock *nsk, const struct sock *osk) 1995 { 1996 const struct proto *prot = READ_ONCE(osk->sk_prot); 1997 #ifdef CONFIG_SECURITY_NETWORK 1998 void *sptr = nsk->sk_security; 1999 #endif 2000 2001 /* If we move sk_tx_queue_mapping out of the private section, 2002 * we must check if sk_tx_queue_clear() is called after 2003 * sock_copy() in sk_clone_lock(). 2004 */ 2005 BUILD_BUG_ON(offsetof(struct sock, sk_tx_queue_mapping) < 2006 offsetof(struct sock, sk_dontcopy_begin) || 2007 offsetof(struct sock, sk_tx_queue_mapping) >= 2008 offsetof(struct sock, sk_dontcopy_end)); 2009 2010 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin)); 2011 2012 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end, 2013 prot->obj_size - offsetof(struct sock, sk_dontcopy_end)); 2014 2015 #ifdef CONFIG_SECURITY_NETWORK 2016 nsk->sk_security = sptr; 2017 security_sk_clone(osk, nsk); 2018 #endif 2019 } 2020 2021 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority, 2022 int family) 2023 { 2024 struct sock *sk; 2025 struct kmem_cache *slab; 2026 2027 slab = prot->slab; 2028 if (slab != NULL) { 2029 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO); 2030 if (!sk) 2031 return sk; 2032 if (want_init_on_alloc(priority)) 2033 sk_prot_clear_nulls(sk, prot->obj_size); 2034 } else 2035 sk = kmalloc(prot->obj_size, priority); 2036 2037 if (sk != NULL) { 2038 if (security_sk_alloc(sk, family, priority)) 2039 goto out_free; 2040 2041 if (!try_module_get(prot->owner)) 2042 goto out_free_sec; 2043 } 2044 2045 return sk; 2046 2047 out_free_sec: 2048 security_sk_free(sk); 2049 out_free: 2050 if (slab != NULL) 2051 kmem_cache_free(slab, sk); 2052 else 2053 kfree(sk); 2054 return NULL; 2055 } 2056 2057 static void sk_prot_free(struct proto *prot, struct sock *sk) 2058 { 2059 struct kmem_cache *slab; 2060 struct module *owner; 2061 2062 owner = prot->owner; 2063 slab = prot->slab; 2064 2065 cgroup_sk_free(&sk->sk_cgrp_data); 2066 mem_cgroup_sk_free(sk); 2067 security_sk_free(sk); 2068 if (slab != NULL) 2069 kmem_cache_free(slab, sk); 2070 else 2071 kfree(sk); 2072 module_put(owner); 2073 } 2074 2075 /** 2076 * sk_alloc - All socket objects are allocated here 2077 * @net: the applicable net namespace 2078 * @family: protocol family 2079 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 2080 * @prot: struct proto associated with this new sock instance 2081 * @kern: is this to be a kernel socket? 2082 */ 2083 struct sock *sk_alloc(struct net *net, int family, gfp_t priority, 2084 struct proto *prot, int kern) 2085 { 2086 struct sock *sk; 2087 2088 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family); 2089 if (sk) { 2090 sk->sk_family = family; 2091 /* 2092 * See comment in struct sock definition to understand 2093 * why we need sk_prot_creator -acme 2094 */ 2095 sk->sk_prot = sk->sk_prot_creator = prot; 2096 sk->sk_kern_sock = kern; 2097 sock_lock_init(sk); 2098 sk->sk_net_refcnt = kern ? 0 : 1; 2099 if (likely(sk->sk_net_refcnt)) { 2100 get_net_track(net, &sk->ns_tracker, priority); 2101 sock_inuse_add(net, 1); 2102 } else { 2103 __netns_tracker_alloc(net, &sk->ns_tracker, 2104 false, priority); 2105 } 2106 2107 sock_net_set(sk, net); 2108 refcount_set(&sk->sk_wmem_alloc, 1); 2109 2110 mem_cgroup_sk_alloc(sk); 2111 cgroup_sk_alloc(&sk->sk_cgrp_data); 2112 sock_update_classid(&sk->sk_cgrp_data); 2113 sock_update_netprioidx(&sk->sk_cgrp_data); 2114 sk_tx_queue_clear(sk); 2115 } 2116 2117 return sk; 2118 } 2119 EXPORT_SYMBOL(sk_alloc); 2120 2121 /* Sockets having SOCK_RCU_FREE will call this function after one RCU 2122 * grace period. This is the case for UDP sockets and TCP listeners. 2123 */ 2124 static void __sk_destruct(struct rcu_head *head) 2125 { 2126 struct sock *sk = container_of(head, struct sock, sk_rcu); 2127 struct sk_filter *filter; 2128 2129 if (sk->sk_destruct) 2130 sk->sk_destruct(sk); 2131 2132 filter = rcu_dereference_check(sk->sk_filter, 2133 refcount_read(&sk->sk_wmem_alloc) == 0); 2134 if (filter) { 2135 sk_filter_uncharge(sk, filter); 2136 RCU_INIT_POINTER(sk->sk_filter, NULL); 2137 } 2138 2139 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP); 2140 2141 #ifdef CONFIG_BPF_SYSCALL 2142 bpf_sk_storage_free(sk); 2143 #endif 2144 2145 if (atomic_read(&sk->sk_omem_alloc)) 2146 pr_debug("%s: optmem leakage (%d bytes) detected\n", 2147 __func__, atomic_read(&sk->sk_omem_alloc)); 2148 2149 if (sk->sk_frag.page) { 2150 put_page(sk->sk_frag.page); 2151 sk->sk_frag.page = NULL; 2152 } 2153 2154 /* We do not need to acquire sk->sk_peer_lock, we are the last user. */ 2155 put_cred(sk->sk_peer_cred); 2156 put_pid(sk->sk_peer_pid); 2157 2158 if (likely(sk->sk_net_refcnt)) 2159 put_net_track(sock_net(sk), &sk->ns_tracker); 2160 else 2161 __netns_tracker_free(sock_net(sk), &sk->ns_tracker, false); 2162 2163 sk_prot_free(sk->sk_prot_creator, sk); 2164 } 2165 2166 void sk_destruct(struct sock *sk) 2167 { 2168 bool use_call_rcu = sock_flag(sk, SOCK_RCU_FREE); 2169 2170 if (rcu_access_pointer(sk->sk_reuseport_cb)) { 2171 reuseport_detach_sock(sk); 2172 use_call_rcu = true; 2173 } 2174 2175 if (use_call_rcu) 2176 call_rcu(&sk->sk_rcu, __sk_destruct); 2177 else 2178 __sk_destruct(&sk->sk_rcu); 2179 } 2180 2181 static void __sk_free(struct sock *sk) 2182 { 2183 if (likely(sk->sk_net_refcnt)) 2184 sock_inuse_add(sock_net(sk), -1); 2185 2186 if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk))) 2187 sock_diag_broadcast_destroy(sk); 2188 else 2189 sk_destruct(sk); 2190 } 2191 2192 void sk_free(struct sock *sk) 2193 { 2194 /* 2195 * We subtract one from sk_wmem_alloc and can know if 2196 * some packets are still in some tx queue. 2197 * If not null, sock_wfree() will call __sk_free(sk) later 2198 */ 2199 if (refcount_dec_and_test(&sk->sk_wmem_alloc)) 2200 __sk_free(sk); 2201 } 2202 EXPORT_SYMBOL(sk_free); 2203 2204 static void sk_init_common(struct sock *sk) 2205 { 2206 skb_queue_head_init(&sk->sk_receive_queue); 2207 skb_queue_head_init(&sk->sk_write_queue); 2208 skb_queue_head_init(&sk->sk_error_queue); 2209 2210 rwlock_init(&sk->sk_callback_lock); 2211 lockdep_set_class_and_name(&sk->sk_receive_queue.lock, 2212 af_rlock_keys + sk->sk_family, 2213 af_family_rlock_key_strings[sk->sk_family]); 2214 lockdep_set_class_and_name(&sk->sk_write_queue.lock, 2215 af_wlock_keys + sk->sk_family, 2216 af_family_wlock_key_strings[sk->sk_family]); 2217 lockdep_set_class_and_name(&sk->sk_error_queue.lock, 2218 af_elock_keys + sk->sk_family, 2219 af_family_elock_key_strings[sk->sk_family]); 2220 lockdep_set_class_and_name(&sk->sk_callback_lock, 2221 af_callback_keys + sk->sk_family, 2222 af_family_clock_key_strings[sk->sk_family]); 2223 } 2224 2225 /** 2226 * sk_clone_lock - clone a socket, and lock its clone 2227 * @sk: the socket to clone 2228 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 2229 * 2230 * Caller must unlock socket even in error path (bh_unlock_sock(newsk)) 2231 */ 2232 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority) 2233 { 2234 struct proto *prot = READ_ONCE(sk->sk_prot); 2235 struct sk_filter *filter; 2236 bool is_charged = true; 2237 struct sock *newsk; 2238 2239 newsk = sk_prot_alloc(prot, priority, sk->sk_family); 2240 if (!newsk) 2241 goto out; 2242 2243 sock_copy(newsk, sk); 2244 2245 newsk->sk_prot_creator = prot; 2246 2247 /* SANITY */ 2248 if (likely(newsk->sk_net_refcnt)) { 2249 get_net_track(sock_net(newsk), &newsk->ns_tracker, priority); 2250 sock_inuse_add(sock_net(newsk), 1); 2251 } else { 2252 /* Kernel sockets are not elevating the struct net refcount. 2253 * Instead, use a tracker to more easily detect if a layer 2254 * is not properly dismantling its kernel sockets at netns 2255 * destroy time. 2256 */ 2257 __netns_tracker_alloc(sock_net(newsk), &newsk->ns_tracker, 2258 false, priority); 2259 } 2260 sk_node_init(&newsk->sk_node); 2261 sock_lock_init(newsk); 2262 bh_lock_sock(newsk); 2263 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL; 2264 newsk->sk_backlog.len = 0; 2265 2266 atomic_set(&newsk->sk_rmem_alloc, 0); 2267 2268 /* sk_wmem_alloc set to one (see sk_free() and sock_wfree()) */ 2269 refcount_set(&newsk->sk_wmem_alloc, 1); 2270 2271 atomic_set(&newsk->sk_omem_alloc, 0); 2272 sk_init_common(newsk); 2273 2274 newsk->sk_dst_cache = NULL; 2275 newsk->sk_dst_pending_confirm = 0; 2276 newsk->sk_wmem_queued = 0; 2277 newsk->sk_forward_alloc = 0; 2278 newsk->sk_reserved_mem = 0; 2279 atomic_set(&newsk->sk_drops, 0); 2280 newsk->sk_send_head = NULL; 2281 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK; 2282 atomic_set(&newsk->sk_zckey, 0); 2283 2284 sock_reset_flag(newsk, SOCK_DONE); 2285 2286 /* sk->sk_memcg will be populated at accept() time */ 2287 newsk->sk_memcg = NULL; 2288 2289 cgroup_sk_clone(&newsk->sk_cgrp_data); 2290 2291 rcu_read_lock(); 2292 filter = rcu_dereference(sk->sk_filter); 2293 if (filter != NULL) 2294 /* though it's an empty new sock, the charging may fail 2295 * if sysctl_optmem_max was changed between creation of 2296 * original socket and cloning 2297 */ 2298 is_charged = sk_filter_charge(newsk, filter); 2299 RCU_INIT_POINTER(newsk->sk_filter, filter); 2300 rcu_read_unlock(); 2301 2302 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) { 2303 /* We need to make sure that we don't uncharge the new 2304 * socket if we couldn't charge it in the first place 2305 * as otherwise we uncharge the parent's filter. 2306 */ 2307 if (!is_charged) 2308 RCU_INIT_POINTER(newsk->sk_filter, NULL); 2309 sk_free_unlock_clone(newsk); 2310 newsk = NULL; 2311 goto out; 2312 } 2313 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL); 2314 2315 if (bpf_sk_storage_clone(sk, newsk)) { 2316 sk_free_unlock_clone(newsk); 2317 newsk = NULL; 2318 goto out; 2319 } 2320 2321 /* Clear sk_user_data if parent had the pointer tagged 2322 * as not suitable for copying when cloning. 2323 */ 2324 if (sk_user_data_is_nocopy(newsk)) 2325 newsk->sk_user_data = NULL; 2326 2327 newsk->sk_err = 0; 2328 newsk->sk_err_soft = 0; 2329 newsk->sk_priority = 0; 2330 newsk->sk_incoming_cpu = raw_smp_processor_id(); 2331 2332 /* Before updating sk_refcnt, we must commit prior changes to memory 2333 * (Documentation/RCU/rculist_nulls.rst for details) 2334 */ 2335 smp_wmb(); 2336 refcount_set(&newsk->sk_refcnt, 2); 2337 2338 sk_set_socket(newsk, NULL); 2339 sk_tx_queue_clear(newsk); 2340 RCU_INIT_POINTER(newsk->sk_wq, NULL); 2341 2342 if (newsk->sk_prot->sockets_allocated) 2343 sk_sockets_allocated_inc(newsk); 2344 2345 if (sock_needs_netstamp(sk) && newsk->sk_flags & SK_FLAGS_TIMESTAMP) 2346 net_enable_timestamp(); 2347 out: 2348 return newsk; 2349 } 2350 EXPORT_SYMBOL_GPL(sk_clone_lock); 2351 2352 void sk_free_unlock_clone(struct sock *sk) 2353 { 2354 /* It is still raw copy of parent, so invalidate 2355 * destructor and make plain sk_free() */ 2356 sk->sk_destruct = NULL; 2357 bh_unlock_sock(sk); 2358 sk_free(sk); 2359 } 2360 EXPORT_SYMBOL_GPL(sk_free_unlock_clone); 2361 2362 static u32 sk_dst_gso_max_size(struct sock *sk, struct dst_entry *dst) 2363 { 2364 bool is_ipv6 = false; 2365 u32 max_size; 2366 2367 #if IS_ENABLED(CONFIG_IPV6) 2368 is_ipv6 = (sk->sk_family == AF_INET6 && 2369 !ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr)); 2370 #endif 2371 /* pairs with the WRITE_ONCE() in netif_set_gso(_ipv4)_max_size() */ 2372 max_size = is_ipv6 ? READ_ONCE(dst->dev->gso_max_size) : 2373 READ_ONCE(dst->dev->gso_ipv4_max_size); 2374 if (max_size > GSO_LEGACY_MAX_SIZE && !sk_is_tcp(sk)) 2375 max_size = GSO_LEGACY_MAX_SIZE; 2376 2377 return max_size - (MAX_TCP_HEADER + 1); 2378 } 2379 2380 void sk_setup_caps(struct sock *sk, struct dst_entry *dst) 2381 { 2382 u32 max_segs = 1; 2383 2384 sk_dst_set(sk, dst); 2385 sk->sk_route_caps = dst->dev->features; 2386 if (sk_is_tcp(sk)) 2387 sk->sk_route_caps |= NETIF_F_GSO; 2388 if (sk->sk_route_caps & NETIF_F_GSO) 2389 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE; 2390 if (unlikely(sk->sk_gso_disabled)) 2391 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 2392 if (sk_can_gso(sk)) { 2393 if (dst->header_len && !xfrm_dst_offload_ok(dst)) { 2394 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 2395 } else { 2396 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM; 2397 sk->sk_gso_max_size = sk_dst_gso_max_size(sk, dst); 2398 /* pairs with the WRITE_ONCE() in netif_set_gso_max_segs() */ 2399 max_segs = max_t(u32, READ_ONCE(dst->dev->gso_max_segs), 1); 2400 } 2401 } 2402 sk->sk_gso_max_segs = max_segs; 2403 } 2404 EXPORT_SYMBOL_GPL(sk_setup_caps); 2405 2406 /* 2407 * Simple resource managers for sockets. 2408 */ 2409 2410 2411 /* 2412 * Write buffer destructor automatically called from kfree_skb. 2413 */ 2414 void sock_wfree(struct sk_buff *skb) 2415 { 2416 struct sock *sk = skb->sk; 2417 unsigned int len = skb->truesize; 2418 bool free; 2419 2420 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) { 2421 if (sock_flag(sk, SOCK_RCU_FREE) && 2422 sk->sk_write_space == sock_def_write_space) { 2423 rcu_read_lock(); 2424 free = refcount_sub_and_test(len, &sk->sk_wmem_alloc); 2425 sock_def_write_space_wfree(sk); 2426 rcu_read_unlock(); 2427 if (unlikely(free)) 2428 __sk_free(sk); 2429 return; 2430 } 2431 2432 /* 2433 * Keep a reference on sk_wmem_alloc, this will be released 2434 * after sk_write_space() call 2435 */ 2436 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc)); 2437 sk->sk_write_space(sk); 2438 len = 1; 2439 } 2440 /* 2441 * if sk_wmem_alloc reaches 0, we must finish what sk_free() 2442 * could not do because of in-flight packets 2443 */ 2444 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc)) 2445 __sk_free(sk); 2446 } 2447 EXPORT_SYMBOL(sock_wfree); 2448 2449 /* This variant of sock_wfree() is used by TCP, 2450 * since it sets SOCK_USE_WRITE_QUEUE. 2451 */ 2452 void __sock_wfree(struct sk_buff *skb) 2453 { 2454 struct sock *sk = skb->sk; 2455 2456 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc)) 2457 __sk_free(sk); 2458 } 2459 2460 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk) 2461 { 2462 skb_orphan(skb); 2463 skb->sk = sk; 2464 #ifdef CONFIG_INET 2465 if (unlikely(!sk_fullsock(sk))) { 2466 skb->destructor = sock_edemux; 2467 sock_hold(sk); 2468 return; 2469 } 2470 #endif 2471 skb->destructor = sock_wfree; 2472 skb_set_hash_from_sk(skb, sk); 2473 /* 2474 * We used to take a refcount on sk, but following operation 2475 * is enough to guarantee sk_free() wont free this sock until 2476 * all in-flight packets are completed 2477 */ 2478 refcount_add(skb->truesize, &sk->sk_wmem_alloc); 2479 } 2480 EXPORT_SYMBOL(skb_set_owner_w); 2481 2482 static bool can_skb_orphan_partial(const struct sk_buff *skb) 2483 { 2484 #ifdef CONFIG_TLS_DEVICE 2485 /* Drivers depend on in-order delivery for crypto offload, 2486 * partial orphan breaks out-of-order-OK logic. 2487 */ 2488 if (skb->decrypted) 2489 return false; 2490 #endif 2491 return (skb->destructor == sock_wfree || 2492 (IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree)); 2493 } 2494 2495 /* This helper is used by netem, as it can hold packets in its 2496 * delay queue. We want to allow the owner socket to send more 2497 * packets, as if they were already TX completed by a typical driver. 2498 * But we also want to keep skb->sk set because some packet schedulers 2499 * rely on it (sch_fq for example). 2500 */ 2501 void skb_orphan_partial(struct sk_buff *skb) 2502 { 2503 if (skb_is_tcp_pure_ack(skb)) 2504 return; 2505 2506 if (can_skb_orphan_partial(skb) && skb_set_owner_sk_safe(skb, skb->sk)) 2507 return; 2508 2509 skb_orphan(skb); 2510 } 2511 EXPORT_SYMBOL(skb_orphan_partial); 2512 2513 /* 2514 * Read buffer destructor automatically called from kfree_skb. 2515 */ 2516 void sock_rfree(struct sk_buff *skb) 2517 { 2518 struct sock *sk = skb->sk; 2519 unsigned int len = skb->truesize; 2520 2521 atomic_sub(len, &sk->sk_rmem_alloc); 2522 sk_mem_uncharge(sk, len); 2523 } 2524 EXPORT_SYMBOL(sock_rfree); 2525 2526 /* 2527 * Buffer destructor for skbs that are not used directly in read or write 2528 * path, e.g. for error handler skbs. Automatically called from kfree_skb. 2529 */ 2530 void sock_efree(struct sk_buff *skb) 2531 { 2532 sock_put(skb->sk); 2533 } 2534 EXPORT_SYMBOL(sock_efree); 2535 2536 /* Buffer destructor for prefetch/receive path where reference count may 2537 * not be held, e.g. for listen sockets. 2538 */ 2539 #ifdef CONFIG_INET 2540 void sock_pfree(struct sk_buff *skb) 2541 { 2542 if (sk_is_refcounted(skb->sk)) 2543 sock_gen_put(skb->sk); 2544 } 2545 EXPORT_SYMBOL(sock_pfree); 2546 #endif /* CONFIG_INET */ 2547 2548 kuid_t sock_i_uid(struct sock *sk) 2549 { 2550 kuid_t uid; 2551 2552 read_lock_bh(&sk->sk_callback_lock); 2553 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID; 2554 read_unlock_bh(&sk->sk_callback_lock); 2555 return uid; 2556 } 2557 EXPORT_SYMBOL(sock_i_uid); 2558 2559 unsigned long sock_i_ino(struct sock *sk) 2560 { 2561 unsigned long ino; 2562 2563 read_lock_bh(&sk->sk_callback_lock); 2564 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0; 2565 read_unlock_bh(&sk->sk_callback_lock); 2566 return ino; 2567 } 2568 EXPORT_SYMBOL(sock_i_ino); 2569 2570 /* 2571 * Allocate a skb from the socket's send buffer. 2572 */ 2573 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 2574 gfp_t priority) 2575 { 2576 if (force || 2577 refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) { 2578 struct sk_buff *skb = alloc_skb(size, priority); 2579 2580 if (skb) { 2581 skb_set_owner_w(skb, sk); 2582 return skb; 2583 } 2584 } 2585 return NULL; 2586 } 2587 EXPORT_SYMBOL(sock_wmalloc); 2588 2589 static void sock_ofree(struct sk_buff *skb) 2590 { 2591 struct sock *sk = skb->sk; 2592 2593 atomic_sub(skb->truesize, &sk->sk_omem_alloc); 2594 } 2595 2596 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size, 2597 gfp_t priority) 2598 { 2599 struct sk_buff *skb; 2600 2601 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */ 2602 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) > 2603 READ_ONCE(sysctl_optmem_max)) 2604 return NULL; 2605 2606 skb = alloc_skb(size, priority); 2607 if (!skb) 2608 return NULL; 2609 2610 atomic_add(skb->truesize, &sk->sk_omem_alloc); 2611 skb->sk = sk; 2612 skb->destructor = sock_ofree; 2613 return skb; 2614 } 2615 2616 /* 2617 * Allocate a memory block from the socket's option memory buffer. 2618 */ 2619 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority) 2620 { 2621 int optmem_max = READ_ONCE(sysctl_optmem_max); 2622 2623 if ((unsigned int)size <= optmem_max && 2624 atomic_read(&sk->sk_omem_alloc) + size < optmem_max) { 2625 void *mem; 2626 /* First do the add, to avoid the race if kmalloc 2627 * might sleep. 2628 */ 2629 atomic_add(size, &sk->sk_omem_alloc); 2630 mem = kmalloc(size, priority); 2631 if (mem) 2632 return mem; 2633 atomic_sub(size, &sk->sk_omem_alloc); 2634 } 2635 return NULL; 2636 } 2637 EXPORT_SYMBOL(sock_kmalloc); 2638 2639 /* Free an option memory block. Note, we actually want the inline 2640 * here as this allows gcc to detect the nullify and fold away the 2641 * condition entirely. 2642 */ 2643 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size, 2644 const bool nullify) 2645 { 2646 if (WARN_ON_ONCE(!mem)) 2647 return; 2648 if (nullify) 2649 kfree_sensitive(mem); 2650 else 2651 kfree(mem); 2652 atomic_sub(size, &sk->sk_omem_alloc); 2653 } 2654 2655 void sock_kfree_s(struct sock *sk, void *mem, int size) 2656 { 2657 __sock_kfree_s(sk, mem, size, false); 2658 } 2659 EXPORT_SYMBOL(sock_kfree_s); 2660 2661 void sock_kzfree_s(struct sock *sk, void *mem, int size) 2662 { 2663 __sock_kfree_s(sk, mem, size, true); 2664 } 2665 EXPORT_SYMBOL(sock_kzfree_s); 2666 2667 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock. 2668 I think, these locks should be removed for datagram sockets. 2669 */ 2670 static long sock_wait_for_wmem(struct sock *sk, long timeo) 2671 { 2672 DEFINE_WAIT(wait); 2673 2674 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk); 2675 for (;;) { 2676 if (!timeo) 2677 break; 2678 if (signal_pending(current)) 2679 break; 2680 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 2681 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 2682 if (refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) 2683 break; 2684 if (sk->sk_shutdown & SEND_SHUTDOWN) 2685 break; 2686 if (sk->sk_err) 2687 break; 2688 timeo = schedule_timeout(timeo); 2689 } 2690 finish_wait(sk_sleep(sk), &wait); 2691 return timeo; 2692 } 2693 2694 2695 /* 2696 * Generic send/receive buffer handlers 2697 */ 2698 2699 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 2700 unsigned long data_len, int noblock, 2701 int *errcode, int max_page_order) 2702 { 2703 struct sk_buff *skb; 2704 long timeo; 2705 int err; 2706 2707 timeo = sock_sndtimeo(sk, noblock); 2708 for (;;) { 2709 err = sock_error(sk); 2710 if (err != 0) 2711 goto failure; 2712 2713 err = -EPIPE; 2714 if (sk->sk_shutdown & SEND_SHUTDOWN) 2715 goto failure; 2716 2717 if (sk_wmem_alloc_get(sk) < READ_ONCE(sk->sk_sndbuf)) 2718 break; 2719 2720 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk); 2721 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 2722 err = -EAGAIN; 2723 if (!timeo) 2724 goto failure; 2725 if (signal_pending(current)) 2726 goto interrupted; 2727 timeo = sock_wait_for_wmem(sk, timeo); 2728 } 2729 skb = alloc_skb_with_frags(header_len, data_len, max_page_order, 2730 errcode, sk->sk_allocation); 2731 if (skb) 2732 skb_set_owner_w(skb, sk); 2733 return skb; 2734 2735 interrupted: 2736 err = sock_intr_errno(timeo); 2737 failure: 2738 *errcode = err; 2739 return NULL; 2740 } 2741 EXPORT_SYMBOL(sock_alloc_send_pskb); 2742 2743 int __sock_cmsg_send(struct sock *sk, struct cmsghdr *cmsg, 2744 struct sockcm_cookie *sockc) 2745 { 2746 u32 tsflags; 2747 2748 switch (cmsg->cmsg_type) { 2749 case SO_MARK: 2750 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) && 2751 !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) 2752 return -EPERM; 2753 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 2754 return -EINVAL; 2755 sockc->mark = *(u32 *)CMSG_DATA(cmsg); 2756 break; 2757 case SO_TIMESTAMPING_OLD: 2758 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 2759 return -EINVAL; 2760 2761 tsflags = *(u32 *)CMSG_DATA(cmsg); 2762 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK) 2763 return -EINVAL; 2764 2765 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK; 2766 sockc->tsflags |= tsflags; 2767 break; 2768 case SCM_TXTIME: 2769 if (!sock_flag(sk, SOCK_TXTIME)) 2770 return -EINVAL; 2771 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64))) 2772 return -EINVAL; 2773 sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg)); 2774 break; 2775 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */ 2776 case SCM_RIGHTS: 2777 case SCM_CREDENTIALS: 2778 break; 2779 default: 2780 return -EINVAL; 2781 } 2782 return 0; 2783 } 2784 EXPORT_SYMBOL(__sock_cmsg_send); 2785 2786 int sock_cmsg_send(struct sock *sk, struct msghdr *msg, 2787 struct sockcm_cookie *sockc) 2788 { 2789 struct cmsghdr *cmsg; 2790 int ret; 2791 2792 for_each_cmsghdr(cmsg, msg) { 2793 if (!CMSG_OK(msg, cmsg)) 2794 return -EINVAL; 2795 if (cmsg->cmsg_level != SOL_SOCKET) 2796 continue; 2797 ret = __sock_cmsg_send(sk, cmsg, sockc); 2798 if (ret) 2799 return ret; 2800 } 2801 return 0; 2802 } 2803 EXPORT_SYMBOL(sock_cmsg_send); 2804 2805 static void sk_enter_memory_pressure(struct sock *sk) 2806 { 2807 if (!sk->sk_prot->enter_memory_pressure) 2808 return; 2809 2810 sk->sk_prot->enter_memory_pressure(sk); 2811 } 2812 2813 static void sk_leave_memory_pressure(struct sock *sk) 2814 { 2815 if (sk->sk_prot->leave_memory_pressure) { 2816 INDIRECT_CALL_INET_1(sk->sk_prot->leave_memory_pressure, 2817 tcp_leave_memory_pressure, sk); 2818 } else { 2819 unsigned long *memory_pressure = sk->sk_prot->memory_pressure; 2820 2821 if (memory_pressure && READ_ONCE(*memory_pressure)) 2822 WRITE_ONCE(*memory_pressure, 0); 2823 } 2824 } 2825 2826 DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key); 2827 2828 /** 2829 * skb_page_frag_refill - check that a page_frag contains enough room 2830 * @sz: minimum size of the fragment we want to get 2831 * @pfrag: pointer to page_frag 2832 * @gfp: priority for memory allocation 2833 * 2834 * Note: While this allocator tries to use high order pages, there is 2835 * no guarantee that allocations succeed. Therefore, @sz MUST be 2836 * less or equal than PAGE_SIZE. 2837 */ 2838 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp) 2839 { 2840 if (pfrag->page) { 2841 if (page_ref_count(pfrag->page) == 1) { 2842 pfrag->offset = 0; 2843 return true; 2844 } 2845 if (pfrag->offset + sz <= pfrag->size) 2846 return true; 2847 put_page(pfrag->page); 2848 } 2849 2850 pfrag->offset = 0; 2851 if (SKB_FRAG_PAGE_ORDER && 2852 !static_branch_unlikely(&net_high_order_alloc_disable_key)) { 2853 /* Avoid direct reclaim but allow kswapd to wake */ 2854 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) | 2855 __GFP_COMP | __GFP_NOWARN | 2856 __GFP_NORETRY, 2857 SKB_FRAG_PAGE_ORDER); 2858 if (likely(pfrag->page)) { 2859 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER; 2860 return true; 2861 } 2862 } 2863 pfrag->page = alloc_page(gfp); 2864 if (likely(pfrag->page)) { 2865 pfrag->size = PAGE_SIZE; 2866 return true; 2867 } 2868 return false; 2869 } 2870 EXPORT_SYMBOL(skb_page_frag_refill); 2871 2872 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag) 2873 { 2874 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation))) 2875 return true; 2876 2877 sk_enter_memory_pressure(sk); 2878 sk_stream_moderate_sndbuf(sk); 2879 return false; 2880 } 2881 EXPORT_SYMBOL(sk_page_frag_refill); 2882 2883 void __lock_sock(struct sock *sk) 2884 __releases(&sk->sk_lock.slock) 2885 __acquires(&sk->sk_lock.slock) 2886 { 2887 DEFINE_WAIT(wait); 2888 2889 for (;;) { 2890 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait, 2891 TASK_UNINTERRUPTIBLE); 2892 spin_unlock_bh(&sk->sk_lock.slock); 2893 schedule(); 2894 spin_lock_bh(&sk->sk_lock.slock); 2895 if (!sock_owned_by_user(sk)) 2896 break; 2897 } 2898 finish_wait(&sk->sk_lock.wq, &wait); 2899 } 2900 2901 void __release_sock(struct sock *sk) 2902 __releases(&sk->sk_lock.slock) 2903 __acquires(&sk->sk_lock.slock) 2904 { 2905 struct sk_buff *skb, *next; 2906 2907 while ((skb = sk->sk_backlog.head) != NULL) { 2908 sk->sk_backlog.head = sk->sk_backlog.tail = NULL; 2909 2910 spin_unlock_bh(&sk->sk_lock.slock); 2911 2912 do { 2913 next = skb->next; 2914 prefetch(next); 2915 DEBUG_NET_WARN_ON_ONCE(skb_dst_is_noref(skb)); 2916 skb_mark_not_on_list(skb); 2917 sk_backlog_rcv(sk, skb); 2918 2919 cond_resched(); 2920 2921 skb = next; 2922 } while (skb != NULL); 2923 2924 spin_lock_bh(&sk->sk_lock.slock); 2925 } 2926 2927 /* 2928 * Doing the zeroing here guarantee we can not loop forever 2929 * while a wild producer attempts to flood us. 2930 */ 2931 sk->sk_backlog.len = 0; 2932 } 2933 2934 void __sk_flush_backlog(struct sock *sk) 2935 { 2936 spin_lock_bh(&sk->sk_lock.slock); 2937 __release_sock(sk); 2938 spin_unlock_bh(&sk->sk_lock.slock); 2939 } 2940 EXPORT_SYMBOL_GPL(__sk_flush_backlog); 2941 2942 /** 2943 * sk_wait_data - wait for data to arrive at sk_receive_queue 2944 * @sk: sock to wait on 2945 * @timeo: for how long 2946 * @skb: last skb seen on sk_receive_queue 2947 * 2948 * Now socket state including sk->sk_err is changed only under lock, 2949 * hence we may omit checks after joining wait queue. 2950 * We check receive queue before schedule() only as optimization; 2951 * it is very likely that release_sock() added new data. 2952 */ 2953 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb) 2954 { 2955 DEFINE_WAIT_FUNC(wait, woken_wake_function); 2956 int rc; 2957 2958 add_wait_queue(sk_sleep(sk), &wait); 2959 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk); 2960 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait); 2961 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk); 2962 remove_wait_queue(sk_sleep(sk), &wait); 2963 return rc; 2964 } 2965 EXPORT_SYMBOL(sk_wait_data); 2966 2967 /** 2968 * __sk_mem_raise_allocated - increase memory_allocated 2969 * @sk: socket 2970 * @size: memory size to allocate 2971 * @amt: pages to allocate 2972 * @kind: allocation type 2973 * 2974 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc 2975 */ 2976 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind) 2977 { 2978 bool memcg_charge = mem_cgroup_sockets_enabled && sk->sk_memcg; 2979 struct proto *prot = sk->sk_prot; 2980 bool charged = true; 2981 long allocated; 2982 2983 sk_memory_allocated_add(sk, amt); 2984 allocated = sk_memory_allocated(sk); 2985 if (memcg_charge && 2986 !(charged = mem_cgroup_charge_skmem(sk->sk_memcg, amt, 2987 gfp_memcg_charge()))) 2988 goto suppress_allocation; 2989 2990 /* Under limit. */ 2991 if (allocated <= sk_prot_mem_limits(sk, 0)) { 2992 sk_leave_memory_pressure(sk); 2993 return 1; 2994 } 2995 2996 /* Under pressure. */ 2997 if (allocated > sk_prot_mem_limits(sk, 1)) 2998 sk_enter_memory_pressure(sk); 2999 3000 /* Over hard limit. */ 3001 if (allocated > sk_prot_mem_limits(sk, 2)) 3002 goto suppress_allocation; 3003 3004 /* guarantee minimum buffer size under pressure */ 3005 if (kind == SK_MEM_RECV) { 3006 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot)) 3007 return 1; 3008 3009 } else { /* SK_MEM_SEND */ 3010 int wmem0 = sk_get_wmem0(sk, prot); 3011 3012 if (sk->sk_type == SOCK_STREAM) { 3013 if (sk->sk_wmem_queued < wmem0) 3014 return 1; 3015 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) { 3016 return 1; 3017 } 3018 } 3019 3020 if (sk_has_memory_pressure(sk)) { 3021 u64 alloc; 3022 3023 if (!sk_under_memory_pressure(sk)) 3024 return 1; 3025 alloc = sk_sockets_allocated_read_positive(sk); 3026 if (sk_prot_mem_limits(sk, 2) > alloc * 3027 sk_mem_pages(sk->sk_wmem_queued + 3028 atomic_read(&sk->sk_rmem_alloc) + 3029 sk->sk_forward_alloc)) 3030 return 1; 3031 } 3032 3033 suppress_allocation: 3034 3035 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) { 3036 sk_stream_moderate_sndbuf(sk); 3037 3038 /* Fail only if socket is _under_ its sndbuf. 3039 * In this case we cannot block, so that we have to fail. 3040 */ 3041 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf) { 3042 /* Force charge with __GFP_NOFAIL */ 3043 if (memcg_charge && !charged) { 3044 mem_cgroup_charge_skmem(sk->sk_memcg, amt, 3045 gfp_memcg_charge() | __GFP_NOFAIL); 3046 } 3047 return 1; 3048 } 3049 } 3050 3051 if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged)) 3052 trace_sock_exceed_buf_limit(sk, prot, allocated, kind); 3053 3054 sk_memory_allocated_sub(sk, amt); 3055 3056 if (memcg_charge && charged) 3057 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt); 3058 3059 return 0; 3060 } 3061 3062 /** 3063 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated 3064 * @sk: socket 3065 * @size: memory size to allocate 3066 * @kind: allocation type 3067 * 3068 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means 3069 * rmem allocation. This function assumes that protocols which have 3070 * memory_pressure use sk_wmem_queued as write buffer accounting. 3071 */ 3072 int __sk_mem_schedule(struct sock *sk, int size, int kind) 3073 { 3074 int ret, amt = sk_mem_pages(size); 3075 3076 sk->sk_forward_alloc += amt << PAGE_SHIFT; 3077 ret = __sk_mem_raise_allocated(sk, size, amt, kind); 3078 if (!ret) 3079 sk->sk_forward_alloc -= amt << PAGE_SHIFT; 3080 return ret; 3081 } 3082 EXPORT_SYMBOL(__sk_mem_schedule); 3083 3084 /** 3085 * __sk_mem_reduce_allocated - reclaim memory_allocated 3086 * @sk: socket 3087 * @amount: number of quanta 3088 * 3089 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc 3090 */ 3091 void __sk_mem_reduce_allocated(struct sock *sk, int amount) 3092 { 3093 sk_memory_allocated_sub(sk, amount); 3094 3095 if (mem_cgroup_sockets_enabled && sk->sk_memcg) 3096 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount); 3097 3098 if (sk_under_memory_pressure(sk) && 3099 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0))) 3100 sk_leave_memory_pressure(sk); 3101 } 3102 3103 /** 3104 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated 3105 * @sk: socket 3106 * @amount: number of bytes (rounded down to a PAGE_SIZE multiple) 3107 */ 3108 void __sk_mem_reclaim(struct sock *sk, int amount) 3109 { 3110 amount >>= PAGE_SHIFT; 3111 sk->sk_forward_alloc -= amount << PAGE_SHIFT; 3112 __sk_mem_reduce_allocated(sk, amount); 3113 } 3114 EXPORT_SYMBOL(__sk_mem_reclaim); 3115 3116 int sk_set_peek_off(struct sock *sk, int val) 3117 { 3118 sk->sk_peek_off = val; 3119 return 0; 3120 } 3121 EXPORT_SYMBOL_GPL(sk_set_peek_off); 3122 3123 /* 3124 * Set of default routines for initialising struct proto_ops when 3125 * the protocol does not support a particular function. In certain 3126 * cases where it makes no sense for a protocol to have a "do nothing" 3127 * function, some default processing is provided. 3128 */ 3129 3130 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len) 3131 { 3132 return -EOPNOTSUPP; 3133 } 3134 EXPORT_SYMBOL(sock_no_bind); 3135 3136 int sock_no_connect(struct socket *sock, struct sockaddr *saddr, 3137 int len, int flags) 3138 { 3139 return -EOPNOTSUPP; 3140 } 3141 EXPORT_SYMBOL(sock_no_connect); 3142 3143 int sock_no_socketpair(struct socket *sock1, struct socket *sock2) 3144 { 3145 return -EOPNOTSUPP; 3146 } 3147 EXPORT_SYMBOL(sock_no_socketpair); 3148 3149 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags, 3150 bool kern) 3151 { 3152 return -EOPNOTSUPP; 3153 } 3154 EXPORT_SYMBOL(sock_no_accept); 3155 3156 int sock_no_getname(struct socket *sock, struct sockaddr *saddr, 3157 int peer) 3158 { 3159 return -EOPNOTSUPP; 3160 } 3161 EXPORT_SYMBOL(sock_no_getname); 3162 3163 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 3164 { 3165 return -EOPNOTSUPP; 3166 } 3167 EXPORT_SYMBOL(sock_no_ioctl); 3168 3169 int sock_no_listen(struct socket *sock, int backlog) 3170 { 3171 return -EOPNOTSUPP; 3172 } 3173 EXPORT_SYMBOL(sock_no_listen); 3174 3175 int sock_no_shutdown(struct socket *sock, int how) 3176 { 3177 return -EOPNOTSUPP; 3178 } 3179 EXPORT_SYMBOL(sock_no_shutdown); 3180 3181 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len) 3182 { 3183 return -EOPNOTSUPP; 3184 } 3185 EXPORT_SYMBOL(sock_no_sendmsg); 3186 3187 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len) 3188 { 3189 return -EOPNOTSUPP; 3190 } 3191 EXPORT_SYMBOL(sock_no_sendmsg_locked); 3192 3193 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len, 3194 int flags) 3195 { 3196 return -EOPNOTSUPP; 3197 } 3198 EXPORT_SYMBOL(sock_no_recvmsg); 3199 3200 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma) 3201 { 3202 /* Mirror missing mmap method error code */ 3203 return -ENODEV; 3204 } 3205 EXPORT_SYMBOL(sock_no_mmap); 3206 3207 /* 3208 * When a file is received (via SCM_RIGHTS, etc), we must bump the 3209 * various sock-based usage counts. 3210 */ 3211 void __receive_sock(struct file *file) 3212 { 3213 struct socket *sock; 3214 3215 sock = sock_from_file(file); 3216 if (sock) { 3217 sock_update_netprioidx(&sock->sk->sk_cgrp_data); 3218 sock_update_classid(&sock->sk->sk_cgrp_data); 3219 } 3220 } 3221 3222 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags) 3223 { 3224 ssize_t res; 3225 struct msghdr msg = {.msg_flags = flags}; 3226 struct kvec iov; 3227 char *kaddr = kmap(page); 3228 iov.iov_base = kaddr + offset; 3229 iov.iov_len = size; 3230 res = kernel_sendmsg(sock, &msg, &iov, 1, size); 3231 kunmap(page); 3232 return res; 3233 } 3234 EXPORT_SYMBOL(sock_no_sendpage); 3235 3236 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page, 3237 int offset, size_t size, int flags) 3238 { 3239 ssize_t res; 3240 struct msghdr msg = {.msg_flags = flags}; 3241 struct kvec iov; 3242 char *kaddr = kmap(page); 3243 3244 iov.iov_base = kaddr + offset; 3245 iov.iov_len = size; 3246 res = kernel_sendmsg_locked(sk, &msg, &iov, 1, size); 3247 kunmap(page); 3248 return res; 3249 } 3250 EXPORT_SYMBOL(sock_no_sendpage_locked); 3251 3252 /* 3253 * Default Socket Callbacks 3254 */ 3255 3256 static void sock_def_wakeup(struct sock *sk) 3257 { 3258 struct socket_wq *wq; 3259 3260 rcu_read_lock(); 3261 wq = rcu_dereference(sk->sk_wq); 3262 if (skwq_has_sleeper(wq)) 3263 wake_up_interruptible_all(&wq->wait); 3264 rcu_read_unlock(); 3265 } 3266 3267 static void sock_def_error_report(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_poll(&wq->wait, EPOLLERR); 3275 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR); 3276 rcu_read_unlock(); 3277 } 3278 3279 void sock_def_readable(struct sock *sk) 3280 { 3281 struct socket_wq *wq; 3282 3283 trace_sk_data_ready(sk); 3284 3285 rcu_read_lock(); 3286 wq = rcu_dereference(sk->sk_wq); 3287 if (skwq_has_sleeper(wq)) 3288 wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI | 3289 EPOLLRDNORM | EPOLLRDBAND); 3290 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN); 3291 rcu_read_unlock(); 3292 } 3293 3294 static void sock_def_write_space(struct sock *sk) 3295 { 3296 struct socket_wq *wq; 3297 3298 rcu_read_lock(); 3299 3300 /* Do not wake up a writer until he can make "significant" 3301 * progress. --DaveM 3302 */ 3303 if (sock_writeable(sk)) { 3304 wq = rcu_dereference(sk->sk_wq); 3305 if (skwq_has_sleeper(wq)) 3306 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT | 3307 EPOLLWRNORM | EPOLLWRBAND); 3308 3309 /* Should agree with poll, otherwise some programs break */ 3310 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT); 3311 } 3312 3313 rcu_read_unlock(); 3314 } 3315 3316 /* An optimised version of sock_def_write_space(), should only be called 3317 * for SOCK_RCU_FREE sockets under RCU read section and after putting 3318 * ->sk_wmem_alloc. 3319 */ 3320 static void sock_def_write_space_wfree(struct sock *sk) 3321 { 3322 /* Do not wake up a writer until he can make "significant" 3323 * progress. --DaveM 3324 */ 3325 if (sock_writeable(sk)) { 3326 struct socket_wq *wq = rcu_dereference(sk->sk_wq); 3327 3328 /* rely on refcount_sub from sock_wfree() */ 3329 smp_mb__after_atomic(); 3330 if (wq && waitqueue_active(&wq->wait)) 3331 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT | 3332 EPOLLWRNORM | EPOLLWRBAND); 3333 3334 /* Should agree with poll, otherwise some programs break */ 3335 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT); 3336 } 3337 } 3338 3339 static void sock_def_destruct(struct sock *sk) 3340 { 3341 } 3342 3343 void sk_send_sigurg(struct sock *sk) 3344 { 3345 if (sk->sk_socket && sk->sk_socket->file) 3346 if (send_sigurg(&sk->sk_socket->file->f_owner)) 3347 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI); 3348 } 3349 EXPORT_SYMBOL(sk_send_sigurg); 3350 3351 void sk_reset_timer(struct sock *sk, struct timer_list* timer, 3352 unsigned long expires) 3353 { 3354 if (!mod_timer(timer, expires)) 3355 sock_hold(sk); 3356 } 3357 EXPORT_SYMBOL(sk_reset_timer); 3358 3359 void sk_stop_timer(struct sock *sk, struct timer_list* timer) 3360 { 3361 if (del_timer(timer)) 3362 __sock_put(sk); 3363 } 3364 EXPORT_SYMBOL(sk_stop_timer); 3365 3366 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer) 3367 { 3368 if (del_timer_sync(timer)) 3369 __sock_put(sk); 3370 } 3371 EXPORT_SYMBOL(sk_stop_timer_sync); 3372 3373 void sock_init_data_uid(struct socket *sock, struct sock *sk, kuid_t uid) 3374 { 3375 sk_init_common(sk); 3376 sk->sk_send_head = NULL; 3377 3378 timer_setup(&sk->sk_timer, NULL, 0); 3379 3380 sk->sk_allocation = GFP_KERNEL; 3381 sk->sk_rcvbuf = READ_ONCE(sysctl_rmem_default); 3382 sk->sk_sndbuf = READ_ONCE(sysctl_wmem_default); 3383 sk->sk_state = TCP_CLOSE; 3384 sk->sk_use_task_frag = true; 3385 sk_set_socket(sk, sock); 3386 3387 sock_set_flag(sk, SOCK_ZAPPED); 3388 3389 if (sock) { 3390 sk->sk_type = sock->type; 3391 RCU_INIT_POINTER(sk->sk_wq, &sock->wq); 3392 sock->sk = sk; 3393 } else { 3394 RCU_INIT_POINTER(sk->sk_wq, NULL); 3395 } 3396 sk->sk_uid = uid; 3397 3398 rwlock_init(&sk->sk_callback_lock); 3399 if (sk->sk_kern_sock) 3400 lockdep_set_class_and_name( 3401 &sk->sk_callback_lock, 3402 af_kern_callback_keys + sk->sk_family, 3403 af_family_kern_clock_key_strings[sk->sk_family]); 3404 else 3405 lockdep_set_class_and_name( 3406 &sk->sk_callback_lock, 3407 af_callback_keys + sk->sk_family, 3408 af_family_clock_key_strings[sk->sk_family]); 3409 3410 sk->sk_state_change = sock_def_wakeup; 3411 sk->sk_data_ready = sock_def_readable; 3412 sk->sk_write_space = sock_def_write_space; 3413 sk->sk_error_report = sock_def_error_report; 3414 sk->sk_destruct = sock_def_destruct; 3415 3416 sk->sk_frag.page = NULL; 3417 sk->sk_frag.offset = 0; 3418 sk->sk_peek_off = -1; 3419 3420 sk->sk_peer_pid = NULL; 3421 sk->sk_peer_cred = NULL; 3422 spin_lock_init(&sk->sk_peer_lock); 3423 3424 sk->sk_write_pending = 0; 3425 sk->sk_rcvlowat = 1; 3426 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; 3427 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT; 3428 3429 sk->sk_stamp = SK_DEFAULT_STAMP; 3430 #if BITS_PER_LONG==32 3431 seqlock_init(&sk->sk_stamp_seq); 3432 #endif 3433 atomic_set(&sk->sk_zckey, 0); 3434 3435 #ifdef CONFIG_NET_RX_BUSY_POLL 3436 sk->sk_napi_id = 0; 3437 sk->sk_ll_usec = READ_ONCE(sysctl_net_busy_read); 3438 #endif 3439 3440 sk->sk_max_pacing_rate = ~0UL; 3441 sk->sk_pacing_rate = ~0UL; 3442 WRITE_ONCE(sk->sk_pacing_shift, 10); 3443 sk->sk_incoming_cpu = -1; 3444 3445 sk_rx_queue_clear(sk); 3446 /* 3447 * Before updating sk_refcnt, we must commit prior changes to memory 3448 * (Documentation/RCU/rculist_nulls.rst for details) 3449 */ 3450 smp_wmb(); 3451 refcount_set(&sk->sk_refcnt, 1); 3452 atomic_set(&sk->sk_drops, 0); 3453 } 3454 EXPORT_SYMBOL(sock_init_data_uid); 3455 3456 void sock_init_data(struct socket *sock, struct sock *sk) 3457 { 3458 kuid_t uid = sock ? 3459 SOCK_INODE(sock)->i_uid : 3460 make_kuid(sock_net(sk)->user_ns, 0); 3461 3462 sock_init_data_uid(sock, sk, uid); 3463 } 3464 EXPORT_SYMBOL(sock_init_data); 3465 3466 void lock_sock_nested(struct sock *sk, int subclass) 3467 { 3468 /* The sk_lock has mutex_lock() semantics here. */ 3469 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_); 3470 3471 might_sleep(); 3472 spin_lock_bh(&sk->sk_lock.slock); 3473 if (sock_owned_by_user_nocheck(sk)) 3474 __lock_sock(sk); 3475 sk->sk_lock.owned = 1; 3476 spin_unlock_bh(&sk->sk_lock.slock); 3477 } 3478 EXPORT_SYMBOL(lock_sock_nested); 3479 3480 void release_sock(struct sock *sk) 3481 { 3482 spin_lock_bh(&sk->sk_lock.slock); 3483 if (sk->sk_backlog.tail) 3484 __release_sock(sk); 3485 3486 /* Warning : release_cb() might need to release sk ownership, 3487 * ie call sock_release_ownership(sk) before us. 3488 */ 3489 if (sk->sk_prot->release_cb) 3490 sk->sk_prot->release_cb(sk); 3491 3492 sock_release_ownership(sk); 3493 if (waitqueue_active(&sk->sk_lock.wq)) 3494 wake_up(&sk->sk_lock.wq); 3495 spin_unlock_bh(&sk->sk_lock.slock); 3496 } 3497 EXPORT_SYMBOL(release_sock); 3498 3499 bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock) 3500 { 3501 might_sleep(); 3502 spin_lock_bh(&sk->sk_lock.slock); 3503 3504 if (!sock_owned_by_user_nocheck(sk)) { 3505 /* 3506 * Fast path return with bottom halves disabled and 3507 * sock::sk_lock.slock held. 3508 * 3509 * The 'mutex' is not contended and holding 3510 * sock::sk_lock.slock prevents all other lockers to 3511 * proceed so the corresponding unlock_sock_fast() can 3512 * avoid the slow path of release_sock() completely and 3513 * just release slock. 3514 * 3515 * From a semantical POV this is equivalent to 'acquiring' 3516 * the 'mutex', hence the corresponding lockdep 3517 * mutex_release() has to happen in the fast path of 3518 * unlock_sock_fast(). 3519 */ 3520 return false; 3521 } 3522 3523 __lock_sock(sk); 3524 sk->sk_lock.owned = 1; 3525 __acquire(&sk->sk_lock.slock); 3526 spin_unlock_bh(&sk->sk_lock.slock); 3527 return true; 3528 } 3529 EXPORT_SYMBOL(__lock_sock_fast); 3530 3531 int sock_gettstamp(struct socket *sock, void __user *userstamp, 3532 bool timeval, bool time32) 3533 { 3534 struct sock *sk = sock->sk; 3535 struct timespec64 ts; 3536 3537 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 3538 ts = ktime_to_timespec64(sock_read_timestamp(sk)); 3539 if (ts.tv_sec == -1) 3540 return -ENOENT; 3541 if (ts.tv_sec == 0) { 3542 ktime_t kt = ktime_get_real(); 3543 sock_write_timestamp(sk, kt); 3544 ts = ktime_to_timespec64(kt); 3545 } 3546 3547 if (timeval) 3548 ts.tv_nsec /= 1000; 3549 3550 #ifdef CONFIG_COMPAT_32BIT_TIME 3551 if (time32) 3552 return put_old_timespec32(&ts, userstamp); 3553 #endif 3554 #ifdef CONFIG_SPARC64 3555 /* beware of padding in sparc64 timeval */ 3556 if (timeval && !in_compat_syscall()) { 3557 struct __kernel_old_timeval __user tv = { 3558 .tv_sec = ts.tv_sec, 3559 .tv_usec = ts.tv_nsec, 3560 }; 3561 if (copy_to_user(userstamp, &tv, sizeof(tv))) 3562 return -EFAULT; 3563 return 0; 3564 } 3565 #endif 3566 return put_timespec64(&ts, userstamp); 3567 } 3568 EXPORT_SYMBOL(sock_gettstamp); 3569 3570 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag) 3571 { 3572 if (!sock_flag(sk, flag)) { 3573 unsigned long previous_flags = sk->sk_flags; 3574 3575 sock_set_flag(sk, flag); 3576 /* 3577 * we just set one of the two flags which require net 3578 * time stamping, but time stamping might have been on 3579 * already because of the other one 3580 */ 3581 if (sock_needs_netstamp(sk) && 3582 !(previous_flags & SK_FLAGS_TIMESTAMP)) 3583 net_enable_timestamp(); 3584 } 3585 } 3586 3587 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, 3588 int level, int type) 3589 { 3590 struct sock_exterr_skb *serr; 3591 struct sk_buff *skb; 3592 int copied, err; 3593 3594 err = -EAGAIN; 3595 skb = sock_dequeue_err_skb(sk); 3596 if (skb == NULL) 3597 goto out; 3598 3599 copied = skb->len; 3600 if (copied > len) { 3601 msg->msg_flags |= MSG_TRUNC; 3602 copied = len; 3603 } 3604 err = skb_copy_datagram_msg(skb, 0, msg, copied); 3605 if (err) 3606 goto out_free_skb; 3607 3608 sock_recv_timestamp(msg, sk, skb); 3609 3610 serr = SKB_EXT_ERR(skb); 3611 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee); 3612 3613 msg->msg_flags |= MSG_ERRQUEUE; 3614 err = copied; 3615 3616 out_free_skb: 3617 kfree_skb(skb); 3618 out: 3619 return err; 3620 } 3621 EXPORT_SYMBOL(sock_recv_errqueue); 3622 3623 /* 3624 * Get a socket option on an socket. 3625 * 3626 * FIX: POSIX 1003.1g is very ambiguous here. It states that 3627 * asynchronous errors should be reported by getsockopt. We assume 3628 * this means if you specify SO_ERROR (otherwise whats the point of it). 3629 */ 3630 int sock_common_getsockopt(struct socket *sock, int level, int optname, 3631 char __user *optval, int __user *optlen) 3632 { 3633 struct sock *sk = sock->sk; 3634 3635 /* IPV6_ADDRFORM can change sk->sk_prot under us. */ 3636 return READ_ONCE(sk->sk_prot)->getsockopt(sk, level, optname, optval, optlen); 3637 } 3638 EXPORT_SYMBOL(sock_common_getsockopt); 3639 3640 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 3641 int flags) 3642 { 3643 struct sock *sk = sock->sk; 3644 int addr_len = 0; 3645 int err; 3646 3647 err = sk->sk_prot->recvmsg(sk, msg, size, flags, &addr_len); 3648 if (err >= 0) 3649 msg->msg_namelen = addr_len; 3650 return err; 3651 } 3652 EXPORT_SYMBOL(sock_common_recvmsg); 3653 3654 /* 3655 * Set socket options on an inet socket. 3656 */ 3657 int sock_common_setsockopt(struct socket *sock, int level, int optname, 3658 sockptr_t optval, unsigned int optlen) 3659 { 3660 struct sock *sk = sock->sk; 3661 3662 /* IPV6_ADDRFORM can change sk->sk_prot under us. */ 3663 return READ_ONCE(sk->sk_prot)->setsockopt(sk, level, optname, optval, optlen); 3664 } 3665 EXPORT_SYMBOL(sock_common_setsockopt); 3666 3667 void sk_common_release(struct sock *sk) 3668 { 3669 if (sk->sk_prot->destroy) 3670 sk->sk_prot->destroy(sk); 3671 3672 /* 3673 * Observation: when sk_common_release is called, processes have 3674 * no access to socket. But net still has. 3675 * Step one, detach it from networking: 3676 * 3677 * A. Remove from hash tables. 3678 */ 3679 3680 sk->sk_prot->unhash(sk); 3681 3682 /* 3683 * In this point socket cannot receive new packets, but it is possible 3684 * that some packets are in flight because some CPU runs receiver and 3685 * did hash table lookup before we unhashed socket. They will achieve 3686 * receive queue and will be purged by socket destructor. 3687 * 3688 * Also we still have packets pending on receive queue and probably, 3689 * our own packets waiting in device queues. sock_destroy will drain 3690 * receive queue, but transmitted packets will delay socket destruction 3691 * until the last reference will be released. 3692 */ 3693 3694 sock_orphan(sk); 3695 3696 xfrm_sk_free_policy(sk); 3697 3698 sock_put(sk); 3699 } 3700 EXPORT_SYMBOL(sk_common_release); 3701 3702 void sk_get_meminfo(const struct sock *sk, u32 *mem) 3703 { 3704 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS); 3705 3706 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk); 3707 mem[SK_MEMINFO_RCVBUF] = READ_ONCE(sk->sk_rcvbuf); 3708 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk); 3709 mem[SK_MEMINFO_SNDBUF] = READ_ONCE(sk->sk_sndbuf); 3710 mem[SK_MEMINFO_FWD_ALLOC] = sk->sk_forward_alloc; 3711 mem[SK_MEMINFO_WMEM_QUEUED] = READ_ONCE(sk->sk_wmem_queued); 3712 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc); 3713 mem[SK_MEMINFO_BACKLOG] = READ_ONCE(sk->sk_backlog.len); 3714 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops); 3715 } 3716 3717 #ifdef CONFIG_PROC_FS 3718 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR); 3719 3720 int sock_prot_inuse_get(struct net *net, struct proto *prot) 3721 { 3722 int cpu, idx = prot->inuse_idx; 3723 int res = 0; 3724 3725 for_each_possible_cpu(cpu) 3726 res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx]; 3727 3728 return res >= 0 ? res : 0; 3729 } 3730 EXPORT_SYMBOL_GPL(sock_prot_inuse_get); 3731 3732 int sock_inuse_get(struct net *net) 3733 { 3734 int cpu, res = 0; 3735 3736 for_each_possible_cpu(cpu) 3737 res += per_cpu_ptr(net->core.prot_inuse, cpu)->all; 3738 3739 return res; 3740 } 3741 3742 EXPORT_SYMBOL_GPL(sock_inuse_get); 3743 3744 static int __net_init sock_inuse_init_net(struct net *net) 3745 { 3746 net->core.prot_inuse = alloc_percpu(struct prot_inuse); 3747 if (net->core.prot_inuse == NULL) 3748 return -ENOMEM; 3749 return 0; 3750 } 3751 3752 static void __net_exit sock_inuse_exit_net(struct net *net) 3753 { 3754 free_percpu(net->core.prot_inuse); 3755 } 3756 3757 static struct pernet_operations net_inuse_ops = { 3758 .init = sock_inuse_init_net, 3759 .exit = sock_inuse_exit_net, 3760 }; 3761 3762 static __init int net_inuse_init(void) 3763 { 3764 if (register_pernet_subsys(&net_inuse_ops)) 3765 panic("Cannot initialize net inuse counters"); 3766 3767 return 0; 3768 } 3769 3770 core_initcall(net_inuse_init); 3771 3772 static int assign_proto_idx(struct proto *prot) 3773 { 3774 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR); 3775 3776 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) { 3777 pr_err("PROTO_INUSE_NR exhausted\n"); 3778 return -ENOSPC; 3779 } 3780 3781 set_bit(prot->inuse_idx, proto_inuse_idx); 3782 return 0; 3783 } 3784 3785 static void release_proto_idx(struct proto *prot) 3786 { 3787 if (prot->inuse_idx != PROTO_INUSE_NR - 1) 3788 clear_bit(prot->inuse_idx, proto_inuse_idx); 3789 } 3790 #else 3791 static inline int assign_proto_idx(struct proto *prot) 3792 { 3793 return 0; 3794 } 3795 3796 static inline void release_proto_idx(struct proto *prot) 3797 { 3798 } 3799 3800 #endif 3801 3802 static void tw_prot_cleanup(struct timewait_sock_ops *twsk_prot) 3803 { 3804 if (!twsk_prot) 3805 return; 3806 kfree(twsk_prot->twsk_slab_name); 3807 twsk_prot->twsk_slab_name = NULL; 3808 kmem_cache_destroy(twsk_prot->twsk_slab); 3809 twsk_prot->twsk_slab = NULL; 3810 } 3811 3812 static int tw_prot_init(const struct proto *prot) 3813 { 3814 struct timewait_sock_ops *twsk_prot = prot->twsk_prot; 3815 3816 if (!twsk_prot) 3817 return 0; 3818 3819 twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", 3820 prot->name); 3821 if (!twsk_prot->twsk_slab_name) 3822 return -ENOMEM; 3823 3824 twsk_prot->twsk_slab = 3825 kmem_cache_create(twsk_prot->twsk_slab_name, 3826 twsk_prot->twsk_obj_size, 0, 3827 SLAB_ACCOUNT | prot->slab_flags, 3828 NULL); 3829 if (!twsk_prot->twsk_slab) { 3830 pr_crit("%s: Can't create timewait sock SLAB cache!\n", 3831 prot->name); 3832 return -ENOMEM; 3833 } 3834 3835 return 0; 3836 } 3837 3838 static void req_prot_cleanup(struct request_sock_ops *rsk_prot) 3839 { 3840 if (!rsk_prot) 3841 return; 3842 kfree(rsk_prot->slab_name); 3843 rsk_prot->slab_name = NULL; 3844 kmem_cache_destroy(rsk_prot->slab); 3845 rsk_prot->slab = NULL; 3846 } 3847 3848 static int req_prot_init(const struct proto *prot) 3849 { 3850 struct request_sock_ops *rsk_prot = prot->rsk_prot; 3851 3852 if (!rsk_prot) 3853 return 0; 3854 3855 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", 3856 prot->name); 3857 if (!rsk_prot->slab_name) 3858 return -ENOMEM; 3859 3860 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name, 3861 rsk_prot->obj_size, 0, 3862 SLAB_ACCOUNT | prot->slab_flags, 3863 NULL); 3864 3865 if (!rsk_prot->slab) { 3866 pr_crit("%s: Can't create request sock SLAB cache!\n", 3867 prot->name); 3868 return -ENOMEM; 3869 } 3870 return 0; 3871 } 3872 3873 int proto_register(struct proto *prot, int alloc_slab) 3874 { 3875 int ret = -ENOBUFS; 3876 3877 if (prot->memory_allocated && !prot->sysctl_mem) { 3878 pr_err("%s: missing sysctl_mem\n", prot->name); 3879 return -EINVAL; 3880 } 3881 if (prot->memory_allocated && !prot->per_cpu_fw_alloc) { 3882 pr_err("%s: missing per_cpu_fw_alloc\n", prot->name); 3883 return -EINVAL; 3884 } 3885 if (alloc_slab) { 3886 prot->slab = kmem_cache_create_usercopy(prot->name, 3887 prot->obj_size, 0, 3888 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT | 3889 prot->slab_flags, 3890 prot->useroffset, prot->usersize, 3891 NULL); 3892 3893 if (prot->slab == NULL) { 3894 pr_crit("%s: Can't create sock SLAB cache!\n", 3895 prot->name); 3896 goto out; 3897 } 3898 3899 if (req_prot_init(prot)) 3900 goto out_free_request_sock_slab; 3901 3902 if (tw_prot_init(prot)) 3903 goto out_free_timewait_sock_slab; 3904 } 3905 3906 mutex_lock(&proto_list_mutex); 3907 ret = assign_proto_idx(prot); 3908 if (ret) { 3909 mutex_unlock(&proto_list_mutex); 3910 goto out_free_timewait_sock_slab; 3911 } 3912 list_add(&prot->node, &proto_list); 3913 mutex_unlock(&proto_list_mutex); 3914 return ret; 3915 3916 out_free_timewait_sock_slab: 3917 if (alloc_slab) 3918 tw_prot_cleanup(prot->twsk_prot); 3919 out_free_request_sock_slab: 3920 if (alloc_slab) { 3921 req_prot_cleanup(prot->rsk_prot); 3922 3923 kmem_cache_destroy(prot->slab); 3924 prot->slab = NULL; 3925 } 3926 out: 3927 return ret; 3928 } 3929 EXPORT_SYMBOL(proto_register); 3930 3931 void proto_unregister(struct proto *prot) 3932 { 3933 mutex_lock(&proto_list_mutex); 3934 release_proto_idx(prot); 3935 list_del(&prot->node); 3936 mutex_unlock(&proto_list_mutex); 3937 3938 kmem_cache_destroy(prot->slab); 3939 prot->slab = NULL; 3940 3941 req_prot_cleanup(prot->rsk_prot); 3942 tw_prot_cleanup(prot->twsk_prot); 3943 } 3944 EXPORT_SYMBOL(proto_unregister); 3945 3946 int sock_load_diag_module(int family, int protocol) 3947 { 3948 if (!protocol) { 3949 if (!sock_is_registered(family)) 3950 return -ENOENT; 3951 3952 return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK, 3953 NETLINK_SOCK_DIAG, family); 3954 } 3955 3956 #ifdef CONFIG_INET 3957 if (family == AF_INET && 3958 protocol != IPPROTO_RAW && 3959 protocol < MAX_INET_PROTOS && 3960 !rcu_access_pointer(inet_protos[protocol])) 3961 return -ENOENT; 3962 #endif 3963 3964 return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK, 3965 NETLINK_SOCK_DIAG, family, protocol); 3966 } 3967 EXPORT_SYMBOL(sock_load_diag_module); 3968 3969 #ifdef CONFIG_PROC_FS 3970 static void *proto_seq_start(struct seq_file *seq, loff_t *pos) 3971 __acquires(proto_list_mutex) 3972 { 3973 mutex_lock(&proto_list_mutex); 3974 return seq_list_start_head(&proto_list, *pos); 3975 } 3976 3977 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos) 3978 { 3979 return seq_list_next(v, &proto_list, pos); 3980 } 3981 3982 static void proto_seq_stop(struct seq_file *seq, void *v) 3983 __releases(proto_list_mutex) 3984 { 3985 mutex_unlock(&proto_list_mutex); 3986 } 3987 3988 static char proto_method_implemented(const void *method) 3989 { 3990 return method == NULL ? 'n' : 'y'; 3991 } 3992 static long sock_prot_memory_allocated(struct proto *proto) 3993 { 3994 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L; 3995 } 3996 3997 static const char *sock_prot_memory_pressure(struct proto *proto) 3998 { 3999 return proto->memory_pressure != NULL ? 4000 proto_memory_pressure(proto) ? "yes" : "no" : "NI"; 4001 } 4002 4003 static void proto_seq_printf(struct seq_file *seq, struct proto *proto) 4004 { 4005 4006 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s " 4007 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n", 4008 proto->name, 4009 proto->obj_size, 4010 sock_prot_inuse_get(seq_file_net(seq), proto), 4011 sock_prot_memory_allocated(proto), 4012 sock_prot_memory_pressure(proto), 4013 proto->max_header, 4014 proto->slab == NULL ? "no" : "yes", 4015 module_name(proto->owner), 4016 proto_method_implemented(proto->close), 4017 proto_method_implemented(proto->connect), 4018 proto_method_implemented(proto->disconnect), 4019 proto_method_implemented(proto->accept), 4020 proto_method_implemented(proto->ioctl), 4021 proto_method_implemented(proto->init), 4022 proto_method_implemented(proto->destroy), 4023 proto_method_implemented(proto->shutdown), 4024 proto_method_implemented(proto->setsockopt), 4025 proto_method_implemented(proto->getsockopt), 4026 proto_method_implemented(proto->sendmsg), 4027 proto_method_implemented(proto->recvmsg), 4028 proto_method_implemented(proto->sendpage), 4029 proto_method_implemented(proto->bind), 4030 proto_method_implemented(proto->backlog_rcv), 4031 proto_method_implemented(proto->hash), 4032 proto_method_implemented(proto->unhash), 4033 proto_method_implemented(proto->get_port), 4034 proto_method_implemented(proto->enter_memory_pressure)); 4035 } 4036 4037 static int proto_seq_show(struct seq_file *seq, void *v) 4038 { 4039 if (v == &proto_list) 4040 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s", 4041 "protocol", 4042 "size", 4043 "sockets", 4044 "memory", 4045 "press", 4046 "maxhdr", 4047 "slab", 4048 "module", 4049 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n"); 4050 else 4051 proto_seq_printf(seq, list_entry(v, struct proto, node)); 4052 return 0; 4053 } 4054 4055 static const struct seq_operations proto_seq_ops = { 4056 .start = proto_seq_start, 4057 .next = proto_seq_next, 4058 .stop = proto_seq_stop, 4059 .show = proto_seq_show, 4060 }; 4061 4062 static __net_init int proto_init_net(struct net *net) 4063 { 4064 if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops, 4065 sizeof(struct seq_net_private))) 4066 return -ENOMEM; 4067 4068 return 0; 4069 } 4070 4071 static __net_exit void proto_exit_net(struct net *net) 4072 { 4073 remove_proc_entry("protocols", net->proc_net); 4074 } 4075 4076 4077 static __net_initdata struct pernet_operations proto_net_ops = { 4078 .init = proto_init_net, 4079 .exit = proto_exit_net, 4080 }; 4081 4082 static int __init proto_init(void) 4083 { 4084 return register_pernet_subsys(&proto_net_ops); 4085 } 4086 4087 subsys_initcall(proto_init); 4088 4089 #endif /* PROC_FS */ 4090 4091 #ifdef CONFIG_NET_RX_BUSY_POLL 4092 bool sk_busy_loop_end(void *p, unsigned long start_time) 4093 { 4094 struct sock *sk = p; 4095 4096 return !skb_queue_empty_lockless(&sk->sk_receive_queue) || 4097 sk_busy_loop_timeout(sk, start_time); 4098 } 4099 EXPORT_SYMBOL(sk_busy_loop_end); 4100 #endif /* CONFIG_NET_RX_BUSY_POLL */ 4101 4102 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len) 4103 { 4104 if (!sk->sk_prot->bind_add) 4105 return -EOPNOTSUPP; 4106 return sk->sk_prot->bind_add(sk, addr, addr_len); 4107 } 4108 EXPORT_SYMBOL(sock_bind_add); 4109